fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93
  94 /* NOTE:
  95  *      Implementing inode permission operations in /proc is almost
  96  *      certainly an error.  Permission checks need to happen during
  97  *      each system call not at open time.  The reason is that most of
  98  *      what we wish to check for permissions in /proc varies at runtime.
  99  *
 100  *      The classic example of a problem is opening file descriptors
 101  *      in /proc for a task before it execs a suid executable.
 102  */
 103
 104 struct pid_entry {
 105         char *name;
 106         int len;
 107         umode_t mode;
 108         const struct inode_operations *iop;
 109         const struct file_operations *fop;
 110         union proc_op op;
 111 };
 112
 113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 114         .name = (NAME),                                 \
 115         .len  = sizeof(NAME) - 1,                       \
 116         .mode = MODE,                                   \
 117         .iop  = IOP,                                    \
 118         .fop  = FOP,                                    \
 119         .op   = OP,                                     \
 120 }
 121
 122 #define DIR(NAME, MODE, iops, fops)     \
 123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 124 #define LNK(NAME, get_link)                                     \
 125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 126                 &proc_pid_link_inode_operations, NULL,          \
 127                 { .proc_get_link = get_link } )
 128 #define REG(NAME, MODE, fops)                           \
 129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 130 #define INF(NAME, MODE, read)                           \
 131         NOD(NAME, (S_IFREG|(MODE)),                     \
 132                 NULL, &proc_info_file_operations,       \
 133                 { .proc_read = read } )
 134 #define ONE(NAME, MODE, show)                           \
 135         NOD(NAME, (S_IFREG|(MODE)),                     \
 136                 NULL, &proc_single_file_operations,     \
 137                 { .proc_show = show } )
 138
 139 static int proc_fd_permission(struct inode *inode, int mask);
 140
 141 /*
 142  * Count the number of hardlinks for the pid_entry table, excluding the .
 143  * and .. links.
 144  */
 145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 146         unsigned int n)
 147 {
 148         unsigned int i;
 149         unsigned int count;
 150
 151         count = 0;
 152         for (i = 0; i < n; ++i) {
 153                 if (S_ISDIR(entries[i].mode))
 154                         ++count;
 155         }
 156
 157         return count;
 158 }
 159
 160 static int get_task_root(struct task_struct *task, struct path *root)
 161 {
 162         int result = -ENOENT;
 163
 164         task_lock(task);
 165         if (task->fs) {
 166                 get_fs_root(task->fs, root);
 167                 result = 0;
 168         }
 169         task_unlock(task);
 170         return result;
 171 }
 172
 173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 174 {
 175         struct task_struct *task = get_proc_task(dentry->d_inode);
 176         int result = -ENOENT;
 177
 178         if (task) {
 179                 task_lock(task);
 180                 if (task->fs) {
 181                         get_fs_pwd(task->fs, path);
 182                         result = 0;
 183                 }
 184                 task_unlock(task);
 185                 put_task_struct(task);
 186         }
 187         return result;
 188 }
 189
 190 static int proc_root_link(struct dentry *dentry, struct path *path)
 191 {
 192         struct task_struct *task = get_proc_task(dentry->d_inode);
 193         int result = -ENOENT;
 194
 195         if (task) {
 196                 result = get_task_root(task, path);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 203 {
 204         int res = 0;
 205         unsigned int len;
 206         struct mm_struct *mm = get_task_mm(task);
 207         if (!mm)
 208                 goto out;
 209         if (!mm->arg_end)
 210                 goto out_mm;    /* Shh! No looking before we're done */
 211
 212         len = mm->arg_end - mm->arg_start;
 213
 214         if (len > PAGE_SIZE)
 215                 len = PAGE_SIZE;
 216
 217         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 218
 219         // If the nul at the end of args has been overwritten, then
 220         // assume application is using setproctitle(3).
 221         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 222                 len = strnlen(buffer, res);
 223                 if (len < res) {
 224                     res = len;
 225                 } else {
 226                         len = mm->env_end - mm->env_start;
 227                         if (len > PAGE_SIZE - res)
 228                                 len = PAGE_SIZE - res;
 229                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 230                         res = strnlen(buffer, res);
 231                 }
 232         }
 233 out_mm:
 234         mmput(mm);
 235 out:
 236         return res;
 237 }
 238
 239 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 240 {
 241         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 242         int res = PTR_ERR(mm);
 243         if (mm && !IS_ERR(mm)) {
 244                 unsigned int nwords = 0;
 245                 do {
 246                         nwords += 2;
 247                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 248                 res = nwords * sizeof(mm->saved_auxv[0]);
 249                 if (res > PAGE_SIZE)
 250                         res = PAGE_SIZE;
 251                 memcpy(buffer, mm->saved_auxv, res);
 252                 mmput(mm);
 253         }
 254         return res;
 255 }
 256
 257
 258 #ifdef CONFIG_KALLSYMS
 259 /*
 260  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 261  * Returns the resolved symbol.  If that fails, simply return the address.
 262  */
 263 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 264 {
 265         unsigned long wchan;
 266         char symname[KSYM_NAME_LEN];
 267
 268         wchan = get_wchan(task);
 269
 270         if (lookup_symbol_name(wchan, symname) < 0)
 271                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 272                         return 0;
 273                 else
 274                         return sprintf(buffer, "%lu", wchan);
 275         else
 276                 return sprintf(buffer, "%s", symname);
 277 }
 278 #endif /* CONFIG_KALLSYMS */
 279
 280 static int lock_trace(struct task_struct *task)
 281 {
 282         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 283         if (err)
 284                 return err;
 285         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 286                 mutex_unlock(&task->signal->cred_guard_mutex);
 287                 return -EPERM;
 288         }
 289         return 0;
 290 }
 291
 292 static void unlock_trace(struct task_struct *task)
 293 {
 294         mutex_unlock(&task->signal->cred_guard_mutex);
 295 }
 296
 297 #ifdef CONFIG_STACKTRACE
 298
 299 #define MAX_STACK_TRACE_DEPTH   64
 300
 301 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 302                           struct pid *pid, struct task_struct *task)
 303 {
 304         struct stack_trace trace;
 305         unsigned long *entries;
 306         int err;
 307         int i;
 308
 309         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 310         if (!entries)
 311                 return -ENOMEM;
 312
 313         trace.nr_entries        = 0;
 314         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 315         trace.entries           = entries;
 316         trace.skip              = 0;
 317
 318         err = lock_trace(task);
 319         if (!err) {
 320                 save_stack_trace_tsk(task, &trace);
 321
 322                 for (i = 0; i < trace.nr_entries; i++) {
 323                         seq_printf(m, "[<%pK>] %pS\n",
 324                                    (void *)entries[i], (void *)entries[i]);
 325                 }
 326                 unlock_trace(task);
 327         }
 328         kfree(entries);
 329
 330         return err;
 331 }
 332 #endif
 333
 334 #ifdef CONFIG_SCHEDSTATS
 335 /*
 336  * Provides /proc/PID/schedstat
 337  */
 338 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 339 {
 340         return sprintf(buffer, "%llu %llu %lu\n",
 341                         (unsigned long long)task->se.sum_exec_runtime,
 342                         (unsigned long long)task->sched_info.run_delay,
 343                         task->sched_info.pcount);
 344 }
 345 #endif
 346
 347 #ifdef CONFIG_LATENCYTOP
 348 static int lstats_show_proc(struct seq_file *m, void *v)
 349 {
 350         int i;
 351         struct inode *inode = m->private;
 352         struct task_struct *task = get_proc_task(inode);
 353
 354         if (!task)
 355                 return -ESRCH;
 356         seq_puts(m, "Latency Top version : v0.1\n");
 357         for (i = 0; i < 32; i++) {
 358                 struct latency_record *lr = &task->latency_record[i];
 359                 if (lr->backtrace[0]) {
 360                         int q;
 361                         seq_printf(m, "%i %li %li",
 362                                    lr->count, lr->time, lr->max);
 363                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 364                                 unsigned long bt = lr->backtrace[q];
 365                                 if (!bt)
 366                                         break;
 367                                 if (bt == ULONG_MAX)
 368                                         break;
 369                                 seq_printf(m, " %ps", (void *)bt);
 370                         }
 371                         seq_putc(m, '\n');
 372                 }
 373
 374         }
 375         put_task_struct(task);
 376         return 0;
 377 }
 378
 379 static int lstats_open(struct inode *inode, struct file *file)
 380 {
 381         return single_open(file, lstats_show_proc, inode);
 382 }
 383
 384 static ssize_t lstats_write(struct file *file, const char __user *buf,
 385                             size_t count, loff_t *offs)
 386 {
 387         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 388
 389         if (!task)
 390                 return -ESRCH;
 391         clear_all_latency_tracing(task);
 392         put_task_struct(task);
 393
 394         return count;
 395 }
 396
 397 static const struct file_operations proc_lstats_operations = {
 398         .open           = lstats_open,
 399         .read           = seq_read,
 400         .write          = lstats_write,
 401         .llseek         = seq_lseek,
 402         .release        = single_release,
 403 };
 404
 405 #endif
 406
 407 static int proc_oom_score(struct task_struct *task, char *buffer)
 408 {
 409         unsigned long totalpages = totalram_pages + total_swap_pages;
 410         unsigned long points = 0;
 411
 412         read_lock(&tasklist_lock);
 413         if (pid_alive(task))
 414                 points = oom_badness(task, NULL, NULL, totalpages) *
 415                                                 1000 / totalpages;
 416         read_unlock(&tasklist_lock);
 417         return sprintf(buffer, "%lu\n", points);
 418 }
 419
 420 struct limit_names {
 421         char *name;
 422         char *unit;
 423 };
 424
 425 static const struct limit_names lnames[RLIM_NLIMITS] = {
 426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 428         [RLIMIT_DATA] = {"Max data size", "bytes"},
 429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 432         [RLIMIT_NPROC] = {"Max processes", "processes"},
 433         [RLIMIT_NOFILE] = {"Max open files", "files"},
 434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 435         [RLIMIT_AS] = {"Max address space", "bytes"},
 436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 439         [RLIMIT_NICE] = {"Max nice priority", NULL},
 440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 442 };
 443
 444 /* Display limits for a process */
 445 static int proc_pid_limits(struct task_struct *task, char *buffer)
 446 {
 447         unsigned int i;
 448         int count = 0;
 449         unsigned long flags;
 450         char *bufptr = buffer;
 451
 452         struct rlimit rlim[RLIM_NLIMITS];
 453
 454         if (!lock_task_sighand(task, &flags))
 455                 return 0;
 456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 457         unlock_task_sighand(task, &flags);
 458
 459         /*
 460          * print the file header
 461          */
 462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 463                         "Limit", "Soft Limit", "Hard Limit", "Units");
 464
 465         for (i = 0; i < RLIM_NLIMITS; i++) {
 466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 468                                          lnames[i].name, "unlimited");
 469                 else
 470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 471                                          lnames[i].name, rlim[i].rlim_cur);
 472
 473                 if (rlim[i].rlim_max == RLIM_INFINITY)
 474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 475                 else
 476                         count += sprintf(&bufptr[count], "%-20lu ",
 477                                          rlim[i].rlim_max);
 478
 479                 if (lnames[i].unit)
 480                         count += sprintf(&bufptr[count], "%-10s\n",
 481                                          lnames[i].unit);
 482                 else
 483                         count += sprintf(&bufptr[count], "\n");
 484         }
 485
 486         return count;
 487 }
 488
 489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 491 {
 492         long nr;
 493         unsigned long args[6], sp, pc;
 494         int res = lock_trace(task);
 495         if (res)
 496                 return res;
 497
 498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 499                 res = sprintf(buffer, "running\n");
 500         else if (nr < 0)
 501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 502         else
 503                 res = sprintf(buffer,
 504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 505                        nr,
 506                        args[0], args[1], args[2], args[3], args[4], args[5],
 507                        sp, pc);
 508         unlock_trace(task);
 509         return res;
 510 }
 511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 512
 513 /************************************************************************/
 514 /*                       Here the fs part begins                        */
 515 /************************************************************************/
 516
 517 /* permission checks */
 518 static int proc_fd_access_allowed(struct inode *inode)
 519 {
 520         struct task_struct *task;
 521         int allowed = 0;
 522         /* Allow access to a task's file descriptors if it is us or we
 523          * may use ptrace attach to the process and find out that
 524          * information.
 525          */
 526         task = get_proc_task(inode);
 527         if (task) {
 528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 529                 put_task_struct(task);
 530         }
 531         return allowed;
 532 }
 533
 534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 535 {
 536         int error;
 537         struct inode *inode = dentry->d_inode;
 538
 539         if (attr->ia_valid & ATTR_MODE)
 540                 return -EPERM;
 541
 542         error = inode_change_ok(inode, attr);
 543         if (error)
 544                 return error;
 545
 546         if ((attr->ia_valid & ATTR_SIZE) &&
 547             attr->ia_size != i_size_read(inode)) {
 548                 error = vmtruncate(inode, attr->ia_size);
 549                 if (error)
 550                         return error;
 551         }
 552
 553         setattr_copy(inode, attr);
 554         mark_inode_dirty(inode);
 555         return 0;
 556 }
 557
 558 /*
 559  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 560  * or euid/egid (for hide_pid_min=2)?
 561  */
 562 static bool has_pid_permissions(struct pid_namespace *pid,
 563                                  struct task_struct *task,
 564                                  int hide_pid_min)
 565 {
 566         if (pid->hide_pid < hide_pid_min)
 567                 return true;
 568         if (in_group_p(pid->pid_gid))
 569                 return true;
 570         return ptrace_may_access(task, PTRACE_MODE_READ);
 571 }
 572
 573
 574 static int proc_pid_permission(struct inode *inode, int mask)
 575 {
 576         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 577         struct task_struct *task;
 578         bool has_perms;
 579
 580         task = get_proc_task(inode);
 581         if (!task)
 582                 return -ESRCH;
 583         has_perms = has_pid_permissions(pid, task, 1);
 584         put_task_struct(task);
 585
 586         if (!has_perms) {
 587                 if (pid->hide_pid == 2) {
 588                         /*
 589                          * Let's make getdents(), stat(), and open()
 590                          * consistent with each other.  If a process
 591                          * may not stat() a file, it shouldn't be seen
 592                          * in procfs at all.
 593                          */
 594                         return -ENOENT;
 595                 }
 596
 597                 return -EPERM;
 598         }
 599         return generic_permission(inode, mask);
 600 }
 601
 602
 603
 604 static const struct inode_operations proc_def_inode_operations = {
 605         .setattr        = proc_setattr,
 606 };
 607
 608 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 609
 610 static ssize_t proc_info_read(struct file * file, char __user * buf,
 611                           size_t count, loff_t *ppos)
 612 {
 613         struct inode * inode = file->f_path.dentry->d_inode;
 614         unsigned long page;
 615         ssize_t length;
 616         struct task_struct *task = get_proc_task(inode);
 617
 618         length = -ESRCH;
 619         if (!task)
 620                 goto out_no_task;
 621
 622         if (count > PROC_BLOCK_SIZE)
 623                 count = PROC_BLOCK_SIZE;
 624
 625         length = -ENOMEM;
 626         if (!(page = __get_free_page(GFP_TEMPORARY)))
 627                 goto out;
 628
 629         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 630
 631         if (length >= 0)
 632                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 633         free_page(page);
 634 out:
 635         put_task_struct(task);
 636 out_no_task:
 637         return length;
 638 }
 639
 640 static const struct file_operations proc_info_file_operations = {
 641         .read           = proc_info_read,
 642         .llseek         = generic_file_llseek,
 643 };
 644
 645 static int proc_single_show(struct seq_file *m, void *v)
 646 {
 647         struct inode *inode = m->private;
 648         struct pid_namespace *ns;
 649         struct pid *pid;
 650         struct task_struct *task;
 651         int ret;
 652
 653         ns = inode->i_sb->s_fs_info;
 654         pid = proc_pid(inode);
 655         task = get_pid_task(pid, PIDTYPE_PID);
 656         if (!task)
 657                 return -ESRCH;
 658
 659         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 660
 661         put_task_struct(task);
 662         return ret;
 663 }
 664
 665 static int proc_single_open(struct inode *inode, struct file *filp)
 666 {
 667         return single_open(filp, proc_single_show, inode);
 668 }
 669
 670 static const struct file_operations proc_single_file_operations = {
 671         .open           = proc_single_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = single_release,
 675 };
 676
 677 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 678 {
 679         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 680         struct mm_struct *mm;
 681
 682         if (!task)
 683                 return -ESRCH;
 684
 685         mm = mm_access(task, mode);
 686         put_task_struct(task);
 687
 688         if (IS_ERR(mm))
 689                 return PTR_ERR(mm);
 690
 691         if (mm) {
 692                 /* ensure this mm_struct can't be freed */
 693                 atomic_inc(&mm->mm_count);
 694                 /* but do not pin its memory */
 695                 mmput(mm);
 696         }
 697
 698         /* OK to pass negative loff_t, we can catch out-of-range */
 699         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 700         file->private_data = mm;
 701
 702         return 0;
 703 }
 704
 705 static int mem_open(struct inode *inode, struct file *file)
 706 {
 707         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
 708 }
 709
 710 static ssize_t mem_rw(struct file *file, char __user *buf,
 711                         size_t count, loff_t *ppos, int write)
 712 {
 713         struct mm_struct *mm = file->private_data;
 714         unsigned long addr = *ppos;
 715         ssize_t copied;
 716         char *page;
 717
 718         if (!mm)
 719                 return 0;
 720
 721         page = (char *)__get_free_page(GFP_TEMPORARY);
 722         if (!page)
 723                 return -ENOMEM;
 724
 725         copied = 0;
 726         if (!atomic_inc_not_zero(&mm->mm_users))
 727                 goto free;
 728
 729         while (count > 0) {
 730                 int this_len = min_t(int, count, PAGE_SIZE);
 731
 732                 if (write && copy_from_user(page, buf, this_len)) {
 733                         copied = -EFAULT;
 734                         break;
 735                 }
 736
 737                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 738                 if (!this_len) {
 739                         if (!copied)
 740                                 copied = -EIO;
 741                         break;
 742                 }
 743
 744                 if (!write && copy_to_user(buf, page, this_len)) {
 745                         copied = -EFAULT;
 746                         break;
 747                 }
 748
 749                 buf += this_len;
 750                 addr += this_len;
 751                 copied += this_len;
 752                 count -= this_len;
 753         }
 754         *ppos = addr;
 755
 756         mmput(mm);
 757 free:
 758         free_page((unsigned long) page);
 759         return copied;
 760 }
 761
 762 static ssize_t mem_read(struct file *file, char __user *buf,
 763                         size_t count, loff_t *ppos)
 764 {
 765         return mem_rw(file, buf, count, ppos, 0);
 766 }
 767
 768 static ssize_t mem_write(struct file *file, const char __user *buf,
 769                          size_t count, loff_t *ppos)
 770 {
 771         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 772 }
 773
 774 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 775 {
 776         switch (orig) {
 777         case 0:
 778                 file->f_pos = offset;
 779                 break;
 780         case 1:
 781                 file->f_pos += offset;
 782                 break;
 783         default:
 784                 return -EINVAL;
 785         }
 786         force_successful_syscall_return();
 787         return file->f_pos;
 788 }
 789
 790 static int mem_release(struct inode *inode, struct file *file)
 791 {
 792         struct mm_struct *mm = file->private_data;
 793         if (mm)
 794                 mmdrop(mm);
 795         return 0;
 796 }
 797
 798 static const struct file_operations proc_mem_operations = {
 799         .llseek         = mem_lseek,
 800         .read           = mem_read,
 801         .write          = mem_write,
 802         .open           = mem_open,
 803         .release        = mem_release,
 804 };
 805
 806 static int environ_open(struct inode *inode, struct file *file)
 807 {
 808         return __mem_open(inode, file, PTRACE_MODE_READ);
 809 }
 810
 811 static ssize_t environ_read(struct file *file, char __user *buf,
 812                         size_t count, loff_t *ppos)
 813 {
 814         char *page;
 815         unsigned long src = *ppos;
 816         int ret = 0;
 817         struct mm_struct *mm = file->private_data;
 818
 819         if (!mm)
 820                 return 0;
 821
 822         page = (char *)__get_free_page(GFP_TEMPORARY);
 823         if (!page)
 824                 return -ENOMEM;
 825
 826         ret = 0;
 827         if (!atomic_inc_not_zero(&mm->mm_users))
 828                 goto free;
 829         while (count > 0) {
 830                 int this_len, retval, max_len;
 831
 832                 this_len = mm->env_end - (mm->env_start + src);
 833
 834                 if (this_len <= 0)
 835                         break;
 836
 837                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 838                 this_len = (this_len > max_len) ? max_len : this_len;
 839
 840                 retval = access_remote_vm(mm, (mm->env_start + src),
 841                         page, this_len, 0);
 842
 843                 if (retval <= 0) {
 844                         ret = retval;
 845                         break;
 846                 }
 847
 848                 if (copy_to_user(buf, page, retval)) {
 849                         ret = -EFAULT;
 850                         break;
 851                 }
 852
 853                 ret += retval;
 854                 src += retval;
 855                 buf += retval;
 856                 count -= retval;
 857         }
 858         *ppos = src;
 859         mmput(mm);
 860
 861 free:
 862         free_page((unsigned long) page);
 863         return ret;
 864 }
 865
 866 static const struct file_operations proc_environ_operations = {
 867         .open           = environ_open,
 868         .read           = environ_read,
 869         .llseek         = generic_file_llseek,
 870         .release        = mem_release,
 871 };
 872
 873 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 874                                 size_t count, loff_t *ppos)
 875 {
 876         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 877         char buffer[PROC_NUMBUF];
 878         size_t len;
 879         int oom_adjust = OOM_DISABLE;
 880         unsigned long flags;
 881
 882         if (!task)
 883                 return -ESRCH;
 884
 885         if (lock_task_sighand(task, &flags)) {
 886                 oom_adjust = task->signal->oom_adj;
 887                 unlock_task_sighand(task, &flags);
 888         }
 889
 890         put_task_struct(task);
 891
 892         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 893
 894         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 895 }
 896
 897 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 898                                 size_t count, loff_t *ppos)
 899 {
 900         struct task_struct *task;
 901         char buffer[PROC_NUMBUF];
 902         int oom_adjust;
 903         unsigned long flags;
 904         int err;
 905
 906         memset(buffer, 0, sizeof(buffer));
 907         if (count > sizeof(buffer) - 1)
 908                 count = sizeof(buffer) - 1;
 909         if (copy_from_user(buffer, buf, count)) {
 910                 err = -EFAULT;
 911                 goto out;
 912         }
 913
 914         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 915         if (err)
 916                 goto out;
 917         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 918              oom_adjust != OOM_DISABLE) {
 919                 err = -EINVAL;
 920                 goto out;
 921         }
 922
 923         task = get_proc_task(file->f_path.dentry->d_inode);
 924         if (!task) {
 925                 err = -ESRCH;
 926                 goto out;
 927         }
 928
 929         task_lock(task);
 930         if (!task->mm) {
 931                 err = -EINVAL;
 932                 goto err_task_lock;
 933         }
 934
 935         if (!lock_task_sighand(task, &flags)) {
 936                 err = -ESRCH;
 937                 goto err_task_lock;
 938         }
 939
 940         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 941                 err = -EACCES;
 942                 goto err_sighand;
 943         }
 944
 945         /*
 946          * Warn that /proc/pid/oom_adj is deprecated, see
 947          * Documentation/feature-removal-schedule.txt.
 948          */
 949         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 950                   current->comm, task_pid_nr(current), task_pid_nr(task),
 951                   task_pid_nr(task));
 952         task->signal->oom_adj = oom_adjust;
 953         /*
 954          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 955          * value is always attainable.
 956          */
 957         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 958                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 959         else
 960                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 961                                                                 -OOM_DISABLE;
 962         trace_oom_score_adj_update(task);
 963 err_sighand:
 964         unlock_task_sighand(task, &flags);
 965 err_task_lock:
 966         task_unlock(task);
 967         put_task_struct(task);
 968 out:
 969         return err < 0 ? err : count;
 970 }
 971
 972 static const struct file_operations proc_oom_adjust_operations = {
 973         .read           = oom_adjust_read,
 974         .write          = oom_adjust_write,
 975         .llseek         = generic_file_llseek,
 976 };
 977
 978 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 979                                         size_t count, loff_t *ppos)
 980 {
 981         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 982         char buffer[PROC_NUMBUF];
 983         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 984         unsigned long flags;
 985         size_t len;
 986
 987         if (!task)
 988                 return -ESRCH;
 989         if (lock_task_sighand(task, &flags)) {
 990                 oom_score_adj = task->signal->oom_score_adj;
 991                 unlock_task_sighand(task, &flags);
 992         }
 993         put_task_struct(task);
 994         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 995         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 996 }
 997
 998 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
 999                                         size_t count, loff_t *ppos)
1000 {
1001         struct task_struct *task;
1002         char buffer[PROC_NUMBUF];
1003         unsigned long flags;
1004         int oom_score_adj;
1005         int err;
1006
1007         memset(buffer, 0, sizeof(buffer));
1008         if (count > sizeof(buffer) - 1)
1009                 count = sizeof(buffer) - 1;
1010         if (copy_from_user(buffer, buf, count)) {
1011                 err = -EFAULT;
1012                 goto out;
1013         }
1014
1015         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1016         if (err)
1017                 goto out;
1018         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1019                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1020                 err = -EINVAL;
1021                 goto out;
1022         }
1023
1024         task = get_proc_task(file->f_path.dentry->d_inode);
1025         if (!task) {
1026                 err = -ESRCH;
1027                 goto out;
1028         }
1029
1030         task_lock(task);
1031         if (!task->mm) {
1032                 err = -EINVAL;
1033                 goto err_task_lock;
1034         }
1035
1036         if (!lock_task_sighand(task, &flags)) {
1037                 err = -ESRCH;
1038                 goto err_task_lock;
1039         }
1040
1041         if (oom_score_adj < task->signal->oom_score_adj_min &&
1042                         !capable(CAP_SYS_RESOURCE)) {
1043                 err = -EACCES;
1044                 goto err_sighand;
1045         }
1046
1047         task->signal->oom_score_adj = oom_score_adj;
1048         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1049                 task->signal->oom_score_adj_min = oom_score_adj;
1050         trace_oom_score_adj_update(task);
1051         /*
1052          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1053          * always attainable.
1054          */
1055         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1056                 task->signal->oom_adj = OOM_DISABLE;
1057         else
1058                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1059                                                         OOM_SCORE_ADJ_MAX;
1060 err_sighand:
1061         unlock_task_sighand(task, &flags);
1062 err_task_lock:
1063         task_unlock(task);
1064         put_task_struct(task);
1065 out:
1066         return err < 0 ? err : count;
1067 }
1068
1069 static const struct file_operations proc_oom_score_adj_operations = {
1070         .read           = oom_score_adj_read,
1071         .write          = oom_score_adj_write,
1072         .llseek         = default_llseek,
1073 };
1074
1075 #ifdef CONFIG_AUDITSYSCALL
1076 #define TMPBUFLEN 21
1077 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1078                                   size_t count, loff_t *ppos)
1079 {
1080         struct inode * inode = file->f_path.dentry->d_inode;
1081         struct task_struct *task = get_proc_task(inode);
1082         ssize_t length;
1083         char tmpbuf[TMPBUFLEN];
1084
1085         if (!task)
1086                 return -ESRCH;
1087         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1088                                 audit_get_loginuid(task));
1089         put_task_struct(task);
1090         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1091 }
1092
1093 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1094                                    size_t count, loff_t *ppos)
1095 {
1096         struct inode * inode = file->f_path.dentry->d_inode;
1097         char *page, *tmp;
1098         ssize_t length;
1099         uid_t loginuid;
1100
1101         rcu_read_lock();
1102         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1103                 rcu_read_unlock();
1104                 return -EPERM;
1105         }
1106         rcu_read_unlock();
1107
1108         if (count >= PAGE_SIZE)
1109                 count = PAGE_SIZE - 1;
1110
1111         if (*ppos != 0) {
1112                 /* No partial writes. */
1113                 return -EINVAL;
1114         }
1115         page = (char*)__get_free_page(GFP_TEMPORARY);
1116         if (!page)
1117                 return -ENOMEM;
1118         length = -EFAULT;
1119         if (copy_from_user(page, buf, count))
1120                 goto out_free_page;
1121
1122         page[count] = '\0';
1123         loginuid = simple_strtoul(page, &tmp, 10);
1124         if (tmp == page) {
1125                 length = -EINVAL;
1126                 goto out_free_page;
1127
1128         }
1129         length = audit_set_loginuid(loginuid);
1130         if (likely(length == 0))
1131                 length = count;
1132
1133 out_free_page:
1134         free_page((unsigned long) page);
1135         return length;
1136 }
1137
1138 static const struct file_operations proc_loginuid_operations = {
1139         .read           = proc_loginuid_read,
1140         .write          = proc_loginuid_write,
1141         .llseek         = generic_file_llseek,
1142 };
1143
1144 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1145                                   size_t count, loff_t *ppos)
1146 {
1147         struct inode * inode = file->f_path.dentry->d_inode;
1148         struct task_struct *task = get_proc_task(inode);
1149         ssize_t length;
1150         char tmpbuf[TMPBUFLEN];
1151
1152         if (!task)
1153                 return -ESRCH;
1154         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1155                                 audit_get_sessionid(task));
1156         put_task_struct(task);
1157         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1158 }
1159
1160 static const struct file_operations proc_sessionid_operations = {
1161         .read           = proc_sessionid_read,
1162         .llseek         = generic_file_llseek,
1163 };
1164 #endif
1165
1166 #ifdef CONFIG_FAULT_INJECTION
1167 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1168                                       size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1171         char buffer[PROC_NUMBUF];
1172         size_t len;
1173         int make_it_fail;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         make_it_fail = task->make_it_fail;
1178         put_task_struct(task);
1179
1180         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1181
1182         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1183 }
1184
1185 static ssize_t proc_fault_inject_write(struct file * file,
1186                         const char __user * buf, size_t count, loff_t *ppos)
1187 {
1188         struct task_struct *task;
1189         char buffer[PROC_NUMBUF], *end;
1190         int make_it_fail;
1191
1192         if (!capable(CAP_SYS_RESOURCE))
1193                 return -EPERM;
1194         memset(buffer, 0, sizeof(buffer));
1195         if (count > sizeof(buffer) - 1)
1196                 count = sizeof(buffer) - 1;
1197         if (copy_from_user(buffer, buf, count))
1198                 return -EFAULT;
1199         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1200         if (*end)
1201                 return -EINVAL;
1202         task = get_proc_task(file->f_dentry->d_inode);
1203         if (!task)
1204                 return -ESRCH;
1205         task->make_it_fail = make_it_fail;
1206         put_task_struct(task);
1207
1208         return count;
1209 }
1210
1211 static const struct file_operations proc_fault_inject_operations = {
1212         .read           = proc_fault_inject_read,
1213         .write          = proc_fault_inject_write,
1214         .llseek         = generic_file_llseek,
1215 };
1216 #endif
1217
1218
1219 #ifdef CONFIG_SCHED_DEBUG
1220 /*
1221  * Print out various scheduling related per-task fields:
1222  */
1223 static int sched_show(struct seq_file *m, void *v)
1224 {
1225         struct inode *inode = m->private;
1226         struct task_struct *p;
1227
1228         p = get_proc_task(inode);
1229         if (!p)
1230                 return -ESRCH;
1231         proc_sched_show_task(p, m);
1232
1233         put_task_struct(p);
1234
1235         return 0;
1236 }
1237
1238 static ssize_t
1239 sched_write(struct file *file, const char __user *buf,
1240             size_t count, loff_t *offset)
1241 {
1242         struct inode *inode = file->f_path.dentry->d_inode;
1243         struct task_struct *p;
1244
1245         p = get_proc_task(inode);
1246         if (!p)
1247                 return -ESRCH;
1248         proc_sched_set_task(p);
1249
1250         put_task_struct(p);
1251
1252         return count;
1253 }
1254
1255 static int sched_open(struct inode *inode, struct file *filp)
1256 {
1257         return single_open(filp, sched_show, inode);
1258 }
1259
1260 static const struct file_operations proc_pid_sched_operations = {
1261         .open           = sched_open,
1262         .read           = seq_read,
1263         .write          = sched_write,
1264         .llseek         = seq_lseek,
1265         .release        = single_release,
1266 };
1267
1268 #endif
1269
1270 #ifdef CONFIG_SCHED_AUTOGROUP
1271 /*
1272  * Print out autogroup related information:
1273  */
1274 static int sched_autogroup_show(struct seq_file *m, void *v)
1275 {
1276         struct inode *inode = m->private;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_autogroup_show_task(p, m);
1283
1284         put_task_struct(p);
1285
1286         return 0;
1287 }
1288
1289 static ssize_t
1290 sched_autogroup_write(struct file *file, const char __user *buf,
1291             size_t count, loff_t *offset)
1292 {
1293         struct inode *inode = file->f_path.dentry->d_inode;
1294         struct task_struct *p;
1295         char buffer[PROC_NUMBUF];
1296         int nice;
1297         int err;
1298
1299         memset(buffer, 0, sizeof(buffer));
1300         if (count > sizeof(buffer) - 1)
1301                 count = sizeof(buffer) - 1;
1302         if (copy_from_user(buffer, buf, count))
1303                 return -EFAULT;
1304
1305         err = kstrtoint(strstrip(buffer), 0, &nice);
1306         if (err < 0)
1307                 return err;
1308
1309         p = get_proc_task(inode);
1310         if (!p)
1311                 return -ESRCH;
1312
1313         err = proc_sched_autogroup_set_nice(p, nice);
1314         if (err)
1315                 count = err;
1316
1317         put_task_struct(p);
1318
1319         return count;
1320 }
1321
1322 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1323 {
1324         int ret;
1325
1326         ret = single_open(filp, sched_autogroup_show, NULL);
1327         if (!ret) {
1328                 struct seq_file *m = filp->private_data;
1329
1330                 m->private = inode;
1331         }
1332         return ret;
1333 }
1334
1335 static const struct file_operations proc_pid_sched_autogroup_operations = {
1336         .open           = sched_autogroup_open,
1337         .read           = seq_read,
1338         .write          = sched_autogroup_write,
1339         .llseek         = seq_lseek,
1340         .release        = single_release,
1341 };
1342
1343 #endif /* CONFIG_SCHED_AUTOGROUP */
1344
1345 static ssize_t comm_write(struct file *file, const char __user *buf,
1346                                 size_t count, loff_t *offset)
1347 {
1348         struct inode *inode = file->f_path.dentry->d_inode;
1349         struct task_struct *p;
1350         char buffer[TASK_COMM_LEN];
1351
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361
1362         if (same_thread_group(current, p))
1363                 set_task_comm(p, buffer);
1364         else
1365                 count = -EINVAL;
1366
1367         put_task_struct(p);
1368
1369         return count;
1370 }
1371
1372 static int comm_show(struct seq_file *m, void *v)
1373 {
1374         struct inode *inode = m->private;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380
1381         task_lock(p);
1382         seq_printf(m, "%s\n", p->comm);
1383         task_unlock(p);
1384
1385         put_task_struct(p);
1386
1387         return 0;
1388 }
1389
1390 static int comm_open(struct inode *inode, struct file *filp)
1391 {
1392         return single_open(filp, comm_show, inode);
1393 }
1394
1395 static const struct file_operations proc_pid_set_comm_operations = {
1396         .open           = comm_open,
1397         .read           = seq_read,
1398         .write          = comm_write,
1399         .llseek         = seq_lseek,
1400         .release        = single_release,
1401 };
1402
1403 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1404 {
1405         struct task_struct *task;
1406         struct mm_struct *mm;
1407         struct file *exe_file;
1408
1409         task = get_proc_task(dentry->d_inode);
1410         if (!task)
1411                 return -ENOENT;
1412         mm = get_task_mm(task);
1413         put_task_struct(task);
1414         if (!mm)
1415                 return -ENOENT;
1416         exe_file = get_mm_exe_file(mm);
1417         mmput(mm);
1418         if (exe_file) {
1419                 *exe_path = exe_file->f_path;
1420                 path_get(&exe_file->f_path);
1421                 fput(exe_file);
1422                 return 0;
1423         } else
1424                 return -ENOENT;
1425 }
1426
1427 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1428 {
1429         struct inode *inode = dentry->d_inode;
1430         int error = -EACCES;
1431
1432         /* We don't need a base pointer in the /proc filesystem */
1433         path_put(&nd->path);
1434
1435         /* Are we allowed to snoop on the tasks file descriptors? */
1436         if (!proc_fd_access_allowed(inode))
1437                 goto out;
1438
1439         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1440 out:
1441         return ERR_PTR(error);
1442 }
1443
1444 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1445 {
1446         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1447         char *pathname;
1448         int len;
1449
1450         if (!tmp)
1451                 return -ENOMEM;
1452
1453         pathname = d_path(path, tmp, PAGE_SIZE);
1454         len = PTR_ERR(pathname);
1455         if (IS_ERR(pathname))
1456                 goto out;
1457         len = tmp + PAGE_SIZE - 1 - pathname;
1458
1459         if (len > buflen)
1460                 len = buflen;
1461         if (copy_to_user(buffer, pathname, len))
1462                 len = -EFAULT;
1463  out:
1464         free_page((unsigned long)tmp);
1465         return len;
1466 }
1467
1468 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1469 {
1470         int error = -EACCES;
1471         struct inode *inode = dentry->d_inode;
1472         struct path path;
1473
1474         /* Are we allowed to snoop on the tasks file descriptors? */
1475         if (!proc_fd_access_allowed(inode))
1476                 goto out;
1477
1478         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1479         if (error)
1480                 goto out;
1481
1482         error = do_proc_readlink(&path, buffer, buflen);
1483         path_put(&path);
1484 out:
1485         return error;
1486 }
1487
1488 static const struct inode_operations proc_pid_link_inode_operations = {
1489         .readlink       = proc_pid_readlink,
1490         .follow_link    = proc_pid_follow_link,
1491         .setattr        = proc_setattr,
1492 };
1493
1494
1495 /* building an inode */
1496
1497 static int task_dumpable(struct task_struct *task)
1498 {
1499         int dumpable = 0;
1500         struct mm_struct *mm;
1501
1502         task_lock(task);
1503         mm = task->mm;
1504         if (mm)
1505                 dumpable = get_dumpable(mm);
1506         task_unlock(task);
1507         if(dumpable == 1)
1508                 return 1;
1509         return 0;
1510 }
1511
1512 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1513 {
1514         struct inode * inode;
1515         struct proc_inode *ei;
1516         const struct cred *cred;
1517
1518         /* We need a new inode */
1519
1520         inode = new_inode(sb);
1521         if (!inode)
1522                 goto out;
1523
1524         /* Common stuff */
1525         ei = PROC_I(inode);
1526         inode->i_ino = get_next_ino();
1527         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528         inode->i_op = &proc_def_inode_operations;
1529
1530         /*
1531          * grab the reference to task.
1532          */
1533         ei->pid = get_task_pid(task, PIDTYPE_PID);
1534         if (!ei->pid)
1535                 goto out_unlock;
1536
1537         if (task_dumpable(task)) {
1538                 rcu_read_lock();
1539                 cred = __task_cred(task);
1540                 inode->i_uid = cred->euid;
1541                 inode->i_gid = cred->egid;
1542                 rcu_read_unlock();
1543         }
1544         security_task_to_inode(task, inode);
1545
1546 out:
1547         return inode;
1548
1549 out_unlock:
1550         iput(inode);
1551         return NULL;
1552 }
1553
1554 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1555 {
1556         struct inode *inode = dentry->d_inode;
1557         struct task_struct *task;
1558         const struct cred *cred;
1559         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1560
1561         generic_fillattr(inode, stat);
1562
1563         rcu_read_lock();
1564         stat->uid = GLOBAL_ROOT_UID;
1565         stat->gid = GLOBAL_ROOT_GID;
1566         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1567         if (task) {
1568                 if (!has_pid_permissions(pid, task, 2)) {
1569                         rcu_read_unlock();
1570                         /*
1571                          * This doesn't prevent learning whether PID exists,
1572                          * it only makes getattr() consistent with readdir().
1573                          */
1574                         return -ENOENT;
1575                 }
1576                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1577                     task_dumpable(task)) {
1578                         cred = __task_cred(task);
1579                         stat->uid = cred->euid;
1580                         stat->gid = cred->egid;
1581                 }
1582         }
1583         rcu_read_unlock();
1584         return 0;
1585 }
1586
1587 /* dentry stuff */
1588
1589 /*
1590  *      Exceptional case: normally we are not allowed to unhash a busy
1591  * directory. In this case, however, we can do it - no aliasing problems
1592  * due to the way we treat inodes.
1593  *
1594  * Rewrite the inode's ownerships here because the owning task may have
1595  * performed a setuid(), etc.
1596  *
1597  * Before the /proc/pid/status file was created the only way to read
1598  * the effective uid of a /process was to stat /proc/pid.  Reading
1599  * /proc/pid/status is slow enough that procps and other packages
1600  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1601  * made this apply to all per process world readable and executable
1602  * directories.
1603  */
1604 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1605 {
1606         struct inode *inode;
1607         struct task_struct *task;
1608         const struct cred *cred;
1609
1610         if (nd && nd->flags & LOOKUP_RCU)
1611                 return -ECHILD;
1612
1613         inode = dentry->d_inode;
1614         task = get_proc_task(inode);
1615
1616         if (task) {
1617                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1618                     task_dumpable(task)) {
1619                         rcu_read_lock();
1620                         cred = __task_cred(task);
1621                         inode->i_uid = cred->euid;
1622                         inode->i_gid = cred->egid;
1623                         rcu_read_unlock();
1624                 } else {
1625                         inode->i_uid = GLOBAL_ROOT_UID;
1626                         inode->i_gid = GLOBAL_ROOT_GID;
1627                 }
1628                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1629                 security_task_to_inode(task, inode);
1630                 put_task_struct(task);
1631                 return 1;
1632         }
1633         d_drop(dentry);
1634         return 0;
1635 }
1636
1637 static int pid_delete_dentry(const struct dentry * dentry)
1638 {
1639         /* Is the task we represent dead?
1640          * If so, then don't put the dentry on the lru list,
1641          * kill it immediately.
1642          */
1643         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1644 }
1645
1646 const struct dentry_operations pid_dentry_operations =
1647 {
1648         .d_revalidate   = pid_revalidate,
1649         .d_delete       = pid_delete_dentry,
1650 };
1651
1652 /* Lookups */
1653
1654 /*
1655  * Fill a directory entry.
1656  *
1657  * If possible create the dcache entry and derive our inode number and
1658  * file type from dcache entry.
1659  *
1660  * Since all of the proc inode numbers are dynamically generated, the inode
1661  * numbers do not exist until the inode is cache.  This means creating the
1662  * the dcache entry in readdir is necessary to keep the inode numbers
1663  * reported by readdir in sync with the inode numbers reported
1664  * by stat.
1665  */
1666 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1667         const char *name, int len,
1668         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1669 {
1670         struct dentry *child, *dir = filp->f_path.dentry;
1671         struct inode *inode;
1672         struct qstr qname;
1673         ino_t ino = 0;
1674         unsigned type = DT_UNKNOWN;
1675
1676         qname.name = name;
1677         qname.len  = len;
1678         qname.hash = full_name_hash(name, len);
1679
1680         child = d_lookup(dir, &qname);
1681         if (!child) {
1682                 struct dentry *new;
1683                 new = d_alloc(dir, &qname);
1684                 if (new) {
1685                         child = instantiate(dir->d_inode, new, task, ptr);
1686                         if (child)
1687                                 dput(new);
1688                         else
1689                                 child = new;
1690                 }
1691         }
1692         if (!child || IS_ERR(child) || !child->d_inode)
1693                 goto end_instantiate;
1694         inode = child->d_inode;
1695         if (inode) {
1696                 ino = inode->i_ino;
1697                 type = inode->i_mode >> 12;
1698         }
1699         dput(child);
1700 end_instantiate:
1701         if (!ino)
1702                 ino = find_inode_number(dir, &qname);
1703         if (!ino)
1704                 ino = 1;
1705         return filldir(dirent, name, len, filp->f_pos, ino, type);
1706 }
1707
1708 static unsigned name_to_int(struct dentry *dentry)
1709 {
1710         const char *name = dentry->d_name.name;
1711         int len = dentry->d_name.len;
1712         unsigned n = 0;
1713
1714         if (len > 1 && *name == '0')
1715                 goto out;
1716         while (len-- > 0) {
1717                 unsigned c = *name++ - '0';
1718                 if (c > 9)
1719                         goto out;
1720                 if (n >= (~0U-9)/10)
1721                         goto out;
1722                 n *= 10;
1723                 n += c;
1724         }
1725         return n;
1726 out:
1727         return ~0U;
1728 }
1729
1730 #define PROC_FDINFO_MAX 64
1731
1732 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1733 {
1734         struct task_struct *task = get_proc_task(inode);
1735         struct files_struct *files = NULL;
1736         struct file *file;
1737         int fd = proc_fd(inode);
1738
1739         if (task) {
1740                 files = get_files_struct(task);
1741                 put_task_struct(task);
1742         }
1743         if (files) {
1744                 /*
1745                  * We are not taking a ref to the file structure, so we must
1746                  * hold ->file_lock.
1747                  */
1748                 spin_lock(&files->file_lock);
1749                 file = fcheck_files(files, fd);
1750                 if (file) {
1751                         unsigned int f_flags;
1752                         struct fdtable *fdt;
1753
1754                         fdt = files_fdtable(files);
1755                         f_flags = file->f_flags & ~O_CLOEXEC;
1756                         if (close_on_exec(fd, fdt))
1757                                 f_flags |= O_CLOEXEC;
1758
1759                         if (path) {
1760                                 *path = file->f_path;
1761                                 path_get(&file->f_path);
1762                         }
1763                         if (info)
1764                                 snprintf(info, PROC_FDINFO_MAX,
1765                                          "pos:\t%lli\n"
1766                                          "flags:\t0%o\n",
1767                                          (long long) file->f_pos,
1768                                          f_flags);
1769                         spin_unlock(&files->file_lock);
1770                         put_files_struct(files);
1771                         return 0;
1772                 }
1773                 spin_unlock(&files->file_lock);
1774                 put_files_struct(files);
1775         }
1776         return -ENOENT;
1777 }
1778
1779 static int proc_fd_link(struct dentry *dentry, struct path *path)
1780 {
1781         return proc_fd_info(dentry->d_inode, path, NULL);
1782 }
1783
1784 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1785 {
1786         struct inode *inode;
1787         struct task_struct *task;
1788         int fd;
1789         struct files_struct *files;
1790         const struct cred *cred;
1791
1792         if (nd && nd->flags & LOOKUP_RCU)
1793                 return -ECHILD;
1794
1795         inode = dentry->d_inode;
1796         task = get_proc_task(inode);
1797         fd = proc_fd(inode);
1798
1799         if (task) {
1800                 files = get_files_struct(task);
1801                 if (files) {
1802                         struct file *file;
1803                         rcu_read_lock();
1804                         file = fcheck_files(files, fd);
1805                         if (file) {
1806                                 unsigned i_mode, f_mode = file->f_mode;
1807
1808                                 rcu_read_unlock();
1809                                 put_files_struct(files);
1810
1811                                 if (task_dumpable(task)) {
1812                                         rcu_read_lock();
1813                                         cred = __task_cred(task);
1814                                         inode->i_uid = cred->euid;
1815                                         inode->i_gid = cred->egid;
1816                                         rcu_read_unlock();
1817                                 } else {
1818                                         inode->i_uid = GLOBAL_ROOT_UID;
1819                                         inode->i_gid = GLOBAL_ROOT_GID;
1820                                 }
1821
1822                                 i_mode = S_IFLNK;
1823                                 if (f_mode & FMODE_READ)
1824                                         i_mode |= S_IRUSR | S_IXUSR;
1825                                 if (f_mode & FMODE_WRITE)
1826                                         i_mode |= S_IWUSR | S_IXUSR;
1827                                 inode->i_mode = i_mode;
1828
1829                                 security_task_to_inode(task, inode);
1830                                 put_task_struct(task);
1831                                 return 1;
1832                         }
1833                         rcu_read_unlock();
1834                         put_files_struct(files);
1835                 }
1836                 put_task_struct(task);
1837         }
1838         d_drop(dentry);
1839         return 0;
1840 }
1841
1842 static const struct dentry_operations tid_fd_dentry_operations =
1843 {
1844         .d_revalidate   = tid_fd_revalidate,
1845         .d_delete       = pid_delete_dentry,
1846 };
1847
1848 static struct dentry *proc_fd_instantiate(struct inode *dir,
1849         struct dentry *dentry, struct task_struct *task, const void *ptr)
1850 {
1851         unsigned fd = *(const unsigned *)ptr;
1852         struct inode *inode;
1853         struct proc_inode *ei;
1854         struct dentry *error = ERR_PTR(-ENOENT);
1855
1856         inode = proc_pid_make_inode(dir->i_sb, task);
1857         if (!inode)
1858                 goto out;
1859         ei = PROC_I(inode);
1860         ei->fd = fd;
1861
1862         inode->i_op = &proc_pid_link_inode_operations;
1863         inode->i_size = 64;
1864         ei->op.proc_get_link = proc_fd_link;
1865         d_set_d_op(dentry, &tid_fd_dentry_operations);
1866         d_add(dentry, inode);
1867         /* Close the race of the process dying before we return the dentry */
1868         if (tid_fd_revalidate(dentry, NULL))
1869                 error = NULL;
1870
1871  out:
1872         return error;
1873 }
1874
1875 static struct dentry *proc_lookupfd_common(struct inode *dir,
1876                                            struct dentry *dentry,
1877                                            instantiate_t instantiate)
1878 {
1879         struct task_struct *task = get_proc_task(dir);
1880         unsigned fd = name_to_int(dentry);
1881         struct dentry *result = ERR_PTR(-ENOENT);
1882
1883         if (!task)
1884                 goto out_no_task;
1885         if (fd == ~0U)
1886                 goto out;
1887
1888         result = instantiate(dir, dentry, task, &fd);
1889 out:
1890         put_task_struct(task);
1891 out_no_task:
1892         return result;
1893 }
1894
1895 static int proc_readfd_common(struct file * filp, void * dirent,
1896                               filldir_t filldir, instantiate_t instantiate)
1897 {
1898         struct dentry *dentry = filp->f_path.dentry;
1899         struct inode *inode = dentry->d_inode;
1900         struct task_struct *p = get_proc_task(inode);
1901         unsigned int fd, ino;
1902         int retval;
1903         struct files_struct * files;
1904
1905         retval = -ENOENT;
1906         if (!p)
1907                 goto out_no_task;
1908         retval = 0;
1909
1910         fd = filp->f_pos;
1911         switch (fd) {
1912                 case 0:
1913                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1914                                 goto out;
1915                         filp->f_pos++;
1916                 case 1:
1917                         ino = parent_ino(dentry);
1918                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1919                                 goto out;
1920                         filp->f_pos++;
1921                 default:
1922                         files = get_files_struct(p);
1923                         if (!files)
1924                                 goto out;
1925                         rcu_read_lock();
1926                         for (fd = filp->f_pos-2;
1927                              fd < files_fdtable(files)->max_fds;
1928                              fd++, filp->f_pos++) {
1929                                 char name[PROC_NUMBUF];
1930                                 int len;
1931
1932                                 if (!fcheck_files(files, fd))
1933                                         continue;
1934                                 rcu_read_unlock();
1935
1936                                 len = snprintf(name, sizeof(name), "%d", fd);
1937                                 if (proc_fill_cache(filp, dirent, filldir,
1938                                                     name, len, instantiate,
1939                                                     p, &fd) < 0) {
1940                                         rcu_read_lock();
1941                                         break;
1942                                 }
1943                                 rcu_read_lock();
1944                         }
1945                         rcu_read_unlock();
1946                         put_files_struct(files);
1947         }
1948 out:
1949         put_task_struct(p);
1950 out_no_task:
1951         return retval;
1952 }
1953
1954 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1955                                     struct nameidata *nd)
1956 {
1957         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1958 }
1959
1960 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1961 {
1962         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1963 }
1964
1965 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1966                                       size_t len, loff_t *ppos)
1967 {
1968         char tmp[PROC_FDINFO_MAX];
1969         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1970         if (!err)
1971                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1972         return err;
1973 }
1974
1975 static const struct file_operations proc_fdinfo_file_operations = {
1976         .open           = nonseekable_open,
1977         .read           = proc_fdinfo_read,
1978         .llseek         = no_llseek,
1979 };
1980
1981 static const struct file_operations proc_fd_operations = {
1982         .read           = generic_read_dir,
1983         .readdir        = proc_readfd,
1984         .llseek         = default_llseek,
1985 };
1986
1987 #ifdef CONFIG_CHECKPOINT_RESTORE
1988
1989 /*
1990  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1991  * which represent vma start and end addresses.
1992  */
1993 static int dname_to_vma_addr(struct dentry *dentry,
1994                              unsigned long *start, unsigned long *end)
1995 {
1996         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1997                 return -EINVAL;
1998
1999         return 0;
2000 }
2001
2002 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2003 {
2004         unsigned long vm_start, vm_end;
2005         bool exact_vma_exists = false;
2006         struct mm_struct *mm = NULL;
2007         struct task_struct *task;
2008         const struct cred *cred;
2009         struct inode *inode;
2010         int status = 0;
2011
2012         if (nd && nd->flags & LOOKUP_RCU)
2013                 return -ECHILD;
2014
2015         if (!capable(CAP_SYS_ADMIN)) {
2016                 status = -EACCES;
2017                 goto out_notask;
2018         }
2019
2020         inode = dentry->d_inode;
2021         task = get_proc_task(inode);
2022         if (!task)
2023                 goto out_notask;
2024
2025         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2026                 goto out;
2027
2028         mm = get_task_mm(task);
2029         if (!mm)
2030                 goto out;
2031
2032         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2033                 down_read(&mm->mmap_sem);
2034                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2035                 up_read(&mm->mmap_sem);
2036         }
2037
2038         mmput(mm);
2039
2040         if (exact_vma_exists) {
2041                 if (task_dumpable(task)) {
2042                         rcu_read_lock();
2043                         cred = __task_cred(task);
2044                         inode->i_uid = cred->euid;
2045                         inode->i_gid = cred->egid;
2046                         rcu_read_unlock();
2047                 } else {
2048                         inode->i_uid = GLOBAL_ROOT_UID;
2049                         inode->i_gid = GLOBAL_ROOT_GID;
2050                 }
2051                 security_task_to_inode(task, inode);
2052                 status = 1;
2053         }
2054
2055 out:
2056         put_task_struct(task);
2057
2058 out_notask:
2059         if (status <= 0)
2060                 d_drop(dentry);
2061
2062         return status;
2063 }
2064
2065 static const struct dentry_operations tid_map_files_dentry_operations = {
2066         .d_revalidate   = map_files_d_revalidate,
2067         .d_delete       = pid_delete_dentry,
2068 };
2069
2070 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2071 {
2072         unsigned long vm_start, vm_end;
2073         struct vm_area_struct *vma;
2074         struct task_struct *task;
2075         struct mm_struct *mm;
2076         int rc;
2077
2078         rc = -ENOENT;
2079         task = get_proc_task(dentry->d_inode);
2080         if (!task)
2081                 goto out;
2082
2083         mm = get_task_mm(task);
2084         put_task_struct(task);
2085         if (!mm)
2086                 goto out;
2087
2088         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2089         if (rc)
2090                 goto out_mmput;
2091
2092         down_read(&mm->mmap_sem);
2093         vma = find_exact_vma(mm, vm_start, vm_end);
2094         if (vma && vma->vm_file) {
2095                 *path = vma->vm_file->f_path;
2096                 path_get(path);
2097                 rc = 0;
2098         }
2099         up_read(&mm->mmap_sem);
2100
2101 out_mmput:
2102         mmput(mm);
2103 out:
2104         return rc;
2105 }
2106
2107 struct map_files_info {
2108         struct file     *file;
2109         unsigned long   len;
2110         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2111 };
2112
2113 static struct dentry *
2114 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2115                            struct task_struct *task, const void *ptr)
2116 {
2117         const struct file *file = ptr;
2118         struct proc_inode *ei;
2119         struct inode *inode;
2120
2121         if (!file)
2122                 return ERR_PTR(-ENOENT);
2123
2124         inode = proc_pid_make_inode(dir->i_sb, task);
2125         if (!inode)
2126                 return ERR_PTR(-ENOENT);
2127
2128         ei = PROC_I(inode);
2129         ei->op.proc_get_link = proc_map_files_get_link;
2130
2131         inode->i_op = &proc_pid_link_inode_operations;
2132         inode->i_size = 64;
2133         inode->i_mode = S_IFLNK;
2134
2135         if (file->f_mode & FMODE_READ)
2136                 inode->i_mode |= S_IRUSR;
2137         if (file->f_mode & FMODE_WRITE)
2138                 inode->i_mode |= S_IWUSR;
2139
2140         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2141         d_add(dentry, inode);
2142
2143         return NULL;
2144 }
2145
2146 static struct dentry *proc_map_files_lookup(struct inode *dir,
2147                 struct dentry *dentry, struct nameidata *nd)
2148 {
2149         unsigned long vm_start, vm_end;
2150         struct vm_area_struct *vma;
2151         struct task_struct *task;
2152         struct dentry *result;
2153         struct mm_struct *mm;
2154
2155         result = ERR_PTR(-EACCES);
2156         if (!capable(CAP_SYS_ADMIN))
2157                 goto out;
2158
2159         result = ERR_PTR(-ENOENT);
2160         task = get_proc_task(dir);
2161         if (!task)
2162                 goto out;
2163
2164         result = ERR_PTR(-EACCES);
2165         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2166                 goto out_put_task;
2167
2168         result = ERR_PTR(-ENOENT);
2169         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2170                 goto out_put_task;
2171
2172         mm = get_task_mm(task);
2173         if (!mm)
2174                 goto out_put_task;
2175
2176         down_read(&mm->mmap_sem);
2177         vma = find_exact_vma(mm, vm_start, vm_end);
2178         if (!vma)
2179                 goto out_no_vma;
2180
2181         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2182
2183 out_no_vma:
2184         up_read(&mm->mmap_sem);
2185         mmput(mm);
2186 out_put_task:
2187         put_task_struct(task);
2188 out:
2189         return result;
2190 }
2191
2192 static const struct inode_operations proc_map_files_inode_operations = {
2193         .lookup         = proc_map_files_lookup,
2194         .permission     = proc_fd_permission,
2195         .setattr        = proc_setattr,
2196 };
2197
2198 static int
2199 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2200 {
2201         struct dentry *dentry = filp->f_path.dentry;
2202         struct inode *inode = dentry->d_inode;
2203         struct vm_area_struct *vma;
2204         struct task_struct *task;
2205         struct mm_struct *mm;
2206         ino_t ino;
2207         int ret;
2208
2209         ret = -EACCES;
2210         if (!capable(CAP_SYS_ADMIN))
2211                 goto out;
2212
2213         ret = -ENOENT;
2214         task = get_proc_task(inode);
2215         if (!task)
2216                 goto out;
2217
2218         ret = -EACCES;
2219         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2220                 goto out_put_task;
2221
2222         ret = 0;
2223         switch (filp->f_pos) {
2224         case 0:
2225                 ino = inode->i_ino;
2226                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2227                         goto out_put_task;
2228                 filp->f_pos++;
2229         case 1:
2230                 ino = parent_ino(dentry);
2231                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2232                         goto out_put_task;
2233                 filp->f_pos++;
2234         default:
2235         {
2236                 unsigned long nr_files, pos, i;
2237                 struct flex_array *fa = NULL;
2238                 struct map_files_info info;
2239                 struct map_files_info *p;
2240
2241                 mm = get_task_mm(task);
2242                 if (!mm)
2243                         goto out_put_task;
2244                 down_read(&mm->mmap_sem);
2245
2246                 nr_files = 0;
2247
2248                 /*
2249                  * We need two passes here:
2250                  *
2251                  *  1) Collect vmas of mapped files with mmap_sem taken
2252                  *  2) Release mmap_sem and instantiate entries
2253                  *
2254                  * otherwise we get lockdep complained, since filldir()
2255                  * routine might require mmap_sem taken in might_fault().
2256                  */
2257
2258                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2259                         if (vma->vm_file && ++pos > filp->f_pos)
2260                                 nr_files++;
2261                 }
2262
2263                 if (nr_files) {
2264                         fa = flex_array_alloc(sizeof(info), nr_files,
2265                                                 GFP_KERNEL);
2266                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2267                                                         GFP_KERNEL)) {
2268                                 ret = -ENOMEM;
2269                                 if (fa)
2270                                         flex_array_free(fa);
2271                                 up_read(&mm->mmap_sem);
2272                                 mmput(mm);
2273                                 goto out_put_task;
2274                         }
2275                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2276                                         vma = vma->vm_next) {
2277                                 if (!vma->vm_file)
2278                                         continue;
2279                                 if (++pos <= filp->f_pos)
2280                                         continue;
2281
2282                                 get_file(vma->vm_file);
2283                                 info.file = vma->vm_file;
2284                                 info.len = snprintf(info.name,
2285                                                 sizeof(info.name), "%lx-%lx",
2286                                                 vma->vm_start, vma->vm_end);
2287                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2288                                         BUG();
2289                         }
2290                 }
2291                 up_read(&mm->mmap_sem);
2292
2293                 for (i = 0; i < nr_files; i++) {
2294                         p = flex_array_get(fa, i);
2295                         ret = proc_fill_cache(filp, dirent, filldir,
2296                                               p->name, p->len,
2297                                               proc_map_files_instantiate,
2298                                               task, p->file);
2299                         if (ret)
2300                                 break;
2301                         filp->f_pos++;
2302                         fput(p->file);
2303                 }
2304                 for (; i < nr_files; i++) {
2305                         /*
2306                          * In case of error don't forget
2307                          * to put rest of file refs.
2308                          */
2309                         p = flex_array_get(fa, i);
2310                         fput(p->file);
2311                 }
2312                 if (fa)
2313                         flex_array_free(fa);
2314                 mmput(mm);
2315         }
2316         }
2317
2318 out_put_task:
2319         put_task_struct(task);
2320 out:
2321         return ret;
2322 }
2323
2324 static const struct file_operations proc_map_files_operations = {
2325         .read           = generic_read_dir,
2326         .readdir        = proc_map_files_readdir,
2327         .llseek         = default_llseek,
2328 };
2329
2330 #endif /* CONFIG_CHECKPOINT_RESTORE */
2331
2332 /*
2333  * /proc/pid/fd needs a special permission handler so that a process can still
2334  * access /proc/self/fd after it has executed a setuid().
2335  */
2336 static int proc_fd_permission(struct inode *inode, int mask)
2337 {
2338         int rv = generic_permission(inode, mask);
2339         if (rv == 0)
2340                 return 0;
2341         if (task_pid(current) == proc_pid(inode))
2342                 rv = 0;
2343         return rv;
2344 }
2345
2346 /*
2347  * proc directories can do almost nothing..
2348  */
2349 static const struct inode_operations proc_fd_inode_operations = {
2350         .lookup         = proc_lookupfd,
2351         .permission     = proc_fd_permission,
2352         .setattr        = proc_setattr,
2353 };
2354
2355 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2356         struct dentry *dentry, struct task_struct *task, const void *ptr)
2357 {
2358         unsigned fd = *(unsigned *)ptr;
2359         struct inode *inode;
2360         struct proc_inode *ei;
2361         struct dentry *error = ERR_PTR(-ENOENT);
2362
2363         inode = proc_pid_make_inode(dir->i_sb, task);
2364         if (!inode)
2365                 goto out;
2366         ei = PROC_I(inode);
2367         ei->fd = fd;
2368         inode->i_mode = S_IFREG | S_IRUSR;
2369         inode->i_fop = &proc_fdinfo_file_operations;
2370         d_set_d_op(dentry, &tid_fd_dentry_operations);
2371         d_add(dentry, inode);
2372         /* Close the race of the process dying before we return the dentry */
2373         if (tid_fd_revalidate(dentry, NULL))
2374                 error = NULL;
2375
2376  out:
2377         return error;
2378 }
2379
2380 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2381                                         struct dentry *dentry,
2382                                         struct nameidata *nd)
2383 {
2384         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2385 }
2386
2387 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2388 {
2389         return proc_readfd_common(filp, dirent, filldir,
2390                                   proc_fdinfo_instantiate);
2391 }
2392
2393 static const struct file_operations proc_fdinfo_operations = {
2394         .read           = generic_read_dir,
2395         .readdir        = proc_readfdinfo,
2396         .llseek         = default_llseek,
2397 };
2398
2399 /*
2400  * proc directories can do almost nothing..
2401  */
2402 static const struct inode_operations proc_fdinfo_inode_operations = {
2403         .lookup         = proc_lookupfdinfo,
2404         .setattr        = proc_setattr,
2405 };
2406
2407
2408 static struct dentry *proc_pident_instantiate(struct inode *dir,
2409         struct dentry *dentry, struct task_struct *task, const void *ptr)
2410 {
2411         const struct pid_entry *p = ptr;
2412         struct inode *inode;
2413         struct proc_inode *ei;
2414         struct dentry *error = ERR_PTR(-ENOENT);
2415
2416         inode = proc_pid_make_inode(dir->i_sb, task);
2417         if (!inode)
2418                 goto out;
2419
2420         ei = PROC_I(inode);
2421         inode->i_mode = p->mode;
2422         if (S_ISDIR(inode->i_mode))
2423                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2424         if (p->iop)
2425                 inode->i_op = p->iop;
2426         if (p->fop)
2427                 inode->i_fop = p->fop;
2428         ei->op = p->op;
2429         d_set_d_op(dentry, &pid_dentry_operations);
2430         d_add(dentry, inode);
2431         /* Close the race of the process dying before we return the dentry */
2432         if (pid_revalidate(dentry, NULL))
2433                 error = NULL;
2434 out:
2435         return error;
2436 }
2437
2438 static struct dentry *proc_pident_lookup(struct inode *dir,
2439                                          struct dentry *dentry,
2440                                          const struct pid_entry *ents,
2441                                          unsigned int nents)
2442 {
2443         struct dentry *error;
2444         struct task_struct *task = get_proc_task(dir);
2445         const struct pid_entry *p, *last;
2446
2447         error = ERR_PTR(-ENOENT);
2448
2449         if (!task)
2450                 goto out_no_task;
2451
2452         /*
2453          * Yes, it does not scale. And it should not. Don't add
2454          * new entries into /proc/<tgid>/ without very good reasons.
2455          */
2456         last = &ents[nents - 1];
2457         for (p = ents; p <= last; p++) {
2458                 if (p->len != dentry->d_name.len)
2459                         continue;
2460                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2461                         break;
2462         }
2463         if (p > last)
2464                 goto out;
2465
2466         error = proc_pident_instantiate(dir, dentry, task, p);
2467 out:
2468         put_task_struct(task);
2469 out_no_task:
2470         return error;
2471 }
2472
2473 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2474         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2475 {
2476         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2477                                 proc_pident_instantiate, task, p);
2478 }
2479
2480 static int proc_pident_readdir(struct file *filp,
2481                 void *dirent, filldir_t filldir,
2482                 const struct pid_entry *ents, unsigned int nents)
2483 {
2484         int i;
2485         struct dentry *dentry = filp->f_path.dentry;
2486         struct inode *inode = dentry->d_inode;
2487         struct task_struct *task = get_proc_task(inode);
2488         const struct pid_entry *p, *last;
2489         ino_t ino;
2490         int ret;
2491
2492         ret = -ENOENT;
2493         if (!task)
2494                 goto out_no_task;
2495
2496         ret = 0;
2497         i = filp->f_pos;
2498         switch (i) {
2499         case 0:
2500                 ino = inode->i_ino;
2501                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2502                         goto out;
2503                 i++;
2504                 filp->f_pos++;
2505                 /* fall through */
2506         case 1:
2507                 ino = parent_ino(dentry);
2508                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2509                         goto out;
2510                 i++;
2511                 filp->f_pos++;
2512                 /* fall through */
2513         default:
2514                 i -= 2;
2515                 if (i >= nents) {
2516                         ret = 1;
2517                         goto out;
2518                 }
2519                 p = ents + i;
2520                 last = &ents[nents - 1];
2521                 while (p <= last) {
2522                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2523                                 goto out;
2524                         filp->f_pos++;
2525                         p++;
2526                 }
2527         }
2528
2529         ret = 1;
2530 out:
2531         put_task_struct(task);
2532 out_no_task:
2533         return ret;
2534 }
2535
2536 #ifdef CONFIG_SECURITY
2537 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2538                                   size_t count, loff_t *ppos)
2539 {
2540         struct inode * inode = file->f_path.dentry->d_inode;
2541         char *p = NULL;
2542         ssize_t length;
2543         struct task_struct *task = get_proc_task(inode);
2544
2545         if (!task)
2546                 return -ESRCH;
2547
2548         length = security_getprocattr(task,
2549                                       (char*)file->f_path.dentry->d_name.name,
2550                                       &p);
2551         put_task_struct(task);
2552         if (length > 0)
2553                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2554         kfree(p);
2555         return length;
2556 }
2557
2558 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2559                                    size_t count, loff_t *ppos)
2560 {
2561         struct inode * inode = file->f_path.dentry->d_inode;
2562         char *page;
2563         ssize_t length;
2564         struct task_struct *task = get_proc_task(inode);
2565
2566         length = -ESRCH;
2567         if (!task)
2568                 goto out_no_task;
2569         if (count > PAGE_SIZE)
2570                 count = PAGE_SIZE;
2571
2572         /* No partial writes. */
2573         length = -EINVAL;
2574         if (*ppos != 0)
2575                 goto out;
2576
2577         length = -ENOMEM;
2578         page = (char*)__get_free_page(GFP_TEMPORARY);
2579         if (!page)
2580                 goto out;
2581
2582         length = -EFAULT;
2583         if (copy_from_user(page, buf, count))
2584                 goto out_free;
2585
2586         /* Guard against adverse ptrace interaction */
2587         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2588         if (length < 0)
2589                 goto out_free;
2590
2591         length = security_setprocattr(task,
2592                                       (char*)file->f_path.dentry->d_name.name,
2593                                       (void*)page, count);
2594         mutex_unlock(&task->signal->cred_guard_mutex);
2595 out_free:
2596         free_page((unsigned long) page);
2597 out:
2598         put_task_struct(task);
2599 out_no_task:
2600         return length;
2601 }
2602
2603 static const struct file_operations proc_pid_attr_operations = {
2604         .read           = proc_pid_attr_read,
2605         .write          = proc_pid_attr_write,
2606         .llseek         = generic_file_llseek,
2607 };
2608
2609 static const struct pid_entry attr_dir_stuff[] = {
2610         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2611         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2612         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2613         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2614         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2615         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2616 };
2617
2618 static int proc_attr_dir_readdir(struct file * filp,
2619                              void * dirent, filldir_t filldir)
2620 {
2621         return proc_pident_readdir(filp,dirent,filldir,
2622                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2623 }
2624
2625 static const struct file_operations proc_attr_dir_operations = {
2626         .read           = generic_read_dir,
2627         .readdir        = proc_attr_dir_readdir,
2628         .llseek         = default_llseek,
2629 };
2630
2631 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2632                                 struct dentry *dentry, struct nameidata *nd)
2633 {
2634         return proc_pident_lookup(dir, dentry,
2635                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2636 }
2637
2638 static const struct inode_operations proc_attr_dir_inode_operations = {
2639         .lookup         = proc_attr_dir_lookup,
2640         .getattr        = pid_getattr,
2641         .setattr        = proc_setattr,
2642 };
2643
2644 #endif
2645
2646 #ifdef CONFIG_ELF_CORE
2647 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2648                                          size_t count, loff_t *ppos)
2649 {
2650         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2651         struct mm_struct *mm;
2652         char buffer[PROC_NUMBUF];
2653         size_t len;
2654         int ret;
2655
2656         if (!task)
2657                 return -ESRCH;
2658
2659         ret = 0;
2660         mm = get_task_mm(task);
2661         if (mm) {
2662                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2663                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2664                                 MMF_DUMP_FILTER_SHIFT));
2665                 mmput(mm);
2666                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2667         }
2668
2669         put_task_struct(task);
2670
2671         return ret;
2672 }
2673
2674 static ssize_t proc_coredump_filter_write(struct file *file,
2675                                           const char __user *buf,
2676                                           size_t count,
2677                                           loff_t *ppos)
2678 {
2679         struct task_struct *task;
2680         struct mm_struct *mm;
2681         char buffer[PROC_NUMBUF], *end;
2682         unsigned int val;
2683         int ret;
2684         int i;
2685         unsigned long mask;
2686
2687         ret = -EFAULT;
2688         memset(buffer, 0, sizeof(buffer));
2689         if (count > sizeof(buffer) - 1)
2690                 count = sizeof(buffer) - 1;
2691         if (copy_from_user(buffer, buf, count))
2692                 goto out_no_task;
2693
2694         ret = -EINVAL;
2695         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2696         if (*end == '\n')
2697                 end++;
2698         if (end - buffer == 0)
2699                 goto out_no_task;
2700
2701         ret = -ESRCH;
2702         task = get_proc_task(file->f_dentry->d_inode);
2703         if (!task)
2704                 goto out_no_task;
2705
2706         ret = end - buffer;
2707         mm = get_task_mm(task);
2708         if (!mm)
2709                 goto out_no_mm;
2710
2711         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2712                 if (val & mask)
2713                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2714                 else
2715                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2716         }
2717
2718         mmput(mm);
2719  out_no_mm:
2720         put_task_struct(task);
2721  out_no_task:
2722         return ret;
2723 }
2724
2725 static const struct file_operations proc_coredump_filter_operations = {
2726         .read           = proc_coredump_filter_read,
2727         .write          = proc_coredump_filter_write,
2728         .llseek         = generic_file_llseek,
2729 };
2730 #endif
2731
2732 /*
2733  * /proc/self:
2734  */
2735 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2736                               int buflen)
2737 {
2738         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2739         pid_t tgid = task_tgid_nr_ns(current, ns);
2740         char tmp[PROC_NUMBUF];
2741         if (!tgid)
2742                 return -ENOENT;
2743         sprintf(tmp, "%d", tgid);
2744         return vfs_readlink(dentry,buffer,buflen,tmp);
2745 }
2746
2747 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2748 {
2749         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2750         pid_t tgid = task_tgid_nr_ns(current, ns);
2751         char *name = ERR_PTR(-ENOENT);
2752         if (tgid) {
2753                 name = __getname();
2754                 if (!name)
2755                         name = ERR_PTR(-ENOMEM);
2756                 else
2757                         sprintf(name, "%d", tgid);
2758         }
2759         nd_set_link(nd, name);
2760         return NULL;
2761 }
2762
2763 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2764                                 void *cookie)
2765 {
2766         char *s = nd_get_link(nd);
2767         if (!IS_ERR(s))
2768                 __putname(s);
2769 }
2770
2771 static const struct inode_operations proc_self_inode_operations = {
2772         .readlink       = proc_self_readlink,
2773         .follow_link    = proc_self_follow_link,
2774         .put_link       = proc_self_put_link,
2775 };
2776
2777 /*
2778  * proc base
2779  *
2780  * These are the directory entries in the root directory of /proc
2781  * that properly belong to the /proc filesystem, as they describe
2782  * describe something that is process related.
2783  */
2784 static const struct pid_entry proc_base_stuff[] = {
2785         NOD("self", S_IFLNK|S_IRWXUGO,
2786                 &proc_self_inode_operations, NULL, {}),
2787 };
2788
2789 static struct dentry *proc_base_instantiate(struct inode *dir,
2790         struct dentry *dentry, struct task_struct *task, const void *ptr)
2791 {
2792         const struct pid_entry *p = ptr;
2793         struct inode *inode;
2794         struct proc_inode *ei;
2795         struct dentry *error;
2796
2797         /* Allocate the inode */
2798         error = ERR_PTR(-ENOMEM);
2799         inode = new_inode(dir->i_sb);
2800         if (!inode)
2801                 goto out;
2802
2803         /* Initialize the inode */
2804         ei = PROC_I(inode);
2805         inode->i_ino = get_next_ino();
2806         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2807
2808         /*
2809          * grab the reference to the task.
2810          */
2811         ei->pid = get_task_pid(task, PIDTYPE_PID);
2812         if (!ei->pid)
2813                 goto out_iput;
2814
2815         inode->i_mode = p->mode;
2816         if (S_ISDIR(inode->i_mode))
2817                 set_nlink(inode, 2);
2818         if (S_ISLNK(inode->i_mode))
2819                 inode->i_size = 64;
2820         if (p->iop)
2821                 inode->i_op = p->iop;
2822         if (p->fop)
2823                 inode->i_fop = p->fop;
2824         ei->op = p->op;
2825         d_add(dentry, inode);
2826         error = NULL;
2827 out:
2828         return error;
2829 out_iput:
2830         iput(inode);
2831         goto out;
2832 }
2833
2834 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2835 {
2836         struct dentry *error;
2837         struct task_struct *task = get_proc_task(dir);
2838         const struct pid_entry *p, *last;
2839
2840         error = ERR_PTR(-ENOENT);
2841
2842         if (!task)
2843                 goto out_no_task;
2844
2845         /* Lookup the directory entry */
2846         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2847         for (p = proc_base_stuff; p <= last; p++) {
2848                 if (p->len != dentry->d_name.len)
2849                         continue;
2850                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2851                         break;
2852         }
2853         if (p > last)
2854                 goto out;
2855
2856         error = proc_base_instantiate(dir, dentry, task, p);
2857
2858 out:
2859         put_task_struct(task);
2860 out_no_task:
2861         return error;
2862 }
2863
2864 static int proc_base_fill_cache(struct file *filp, void *dirent,
2865         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2866 {
2867         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2868                                 proc_base_instantiate, task, p);
2869 }
2870
2871 #ifdef CONFIG_TASK_IO_ACCOUNTING
2872 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2873 {
2874         struct task_io_accounting acct = task->ioac;
2875         unsigned long flags;
2876         int result;
2877
2878         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2879         if (result)
2880                 return result;
2881
2882         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2883                 result = -EACCES;
2884                 goto out_unlock;
2885         }
2886
2887         if (whole && lock_task_sighand(task, &flags)) {
2888                 struct task_struct *t = task;
2889
2890                 task_io_accounting_add(&acct, &task->signal->ioac);
2891                 while_each_thread(task, t)
2892                         task_io_accounting_add(&acct, &t->ioac);
2893
2894                 unlock_task_sighand(task, &flags);
2895         }
2896         result = sprintf(buffer,
2897                         "rchar: %llu\n"
2898                         "wchar: %llu\n"
2899                         "syscr: %llu\n"
2900                         "syscw: %llu\n"
2901                         "read_bytes: %llu\n"
2902                         "write_bytes: %llu\n"
2903                         "cancelled_write_bytes: %llu\n",
2904                         (unsigned long long)acct.rchar,
2905                         (unsigned long long)acct.wchar,
2906                         (unsigned long long)acct.syscr,
2907                         (unsigned long long)acct.syscw,
2908                         (unsigned long long)acct.read_bytes,
2909                         (unsigned long long)acct.write_bytes,
2910                         (unsigned long long)acct.cancelled_write_bytes);
2911 out_unlock:
2912         mutex_unlock(&task->signal->cred_guard_mutex);
2913         return result;
2914 }
2915
2916 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2917 {
2918         return do_io_accounting(task, buffer, 0);
2919 }
2920
2921 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2922 {
2923         return do_io_accounting(task, buffer, 1);
2924 }
2925 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2926
2927 #ifdef CONFIG_USER_NS
2928 static int proc_id_map_open(struct inode *inode, struct file *file,
2929         struct seq_operations *seq_ops)
2930 {
2931         struct user_namespace *ns = NULL;
2932         struct task_struct *task;
2933         struct seq_file *seq;
2934         int ret = -EINVAL;
2935
2936         task = get_proc_task(inode);
2937         if (task) {
2938                 rcu_read_lock();
2939                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2940                 rcu_read_unlock();
2941                 put_task_struct(task);
2942         }
2943         if (!ns)
2944                 goto err;
2945
2946         ret = seq_open(file, seq_ops);
2947         if (ret)
2948                 goto err_put_ns;
2949
2950         seq = file->private_data;
2951         seq->private = ns;
2952
2953         return 0;
2954 err_put_ns:
2955         put_user_ns(ns);
2956 err:
2957         return ret;
2958 }
2959
2960 static int proc_id_map_release(struct inode *inode, struct file *file)
2961 {
2962         struct seq_file *seq = file->private_data;
2963         struct user_namespace *ns = seq->private;
2964         put_user_ns(ns);
2965         return seq_release(inode, file);
2966 }
2967
2968 static int proc_uid_map_open(struct inode *inode, struct file *file)
2969 {
2970         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2971 }
2972
2973 static int proc_gid_map_open(struct inode *inode, struct file *file)
2974 {
2975         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2976 }
2977
2978 static const struct file_operations proc_uid_map_operations = {
2979         .open           = proc_uid_map_open,
2980         .write          = proc_uid_map_write,
2981         .read           = seq_read,
2982         .llseek         = seq_lseek,
2983         .release        = proc_id_map_release,
2984 };
2985
2986 static const struct file_operations proc_gid_map_operations = {
2987         .open           = proc_gid_map_open,
2988         .write          = proc_gid_map_write,
2989         .read           = seq_read,
2990         .llseek         = seq_lseek,
2991         .release        = proc_id_map_release,
2992 };
2993 #endif /* CONFIG_USER_NS */
2994
2995 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2996                                 struct pid *pid, struct task_struct *task)
2997 {
2998         int err = lock_trace(task);
2999         if (!err) {
3000                 seq_printf(m, "%08x\n", task->personality);
3001                 unlock_trace(task);
3002         }
3003         return err;
3004 }
3005
3006 /*
3007  * Thread groups
3008  */
3009 static const struct file_operations proc_task_operations;
3010 static const struct inode_operations proc_task_inode_operations;
3011
3012 static const struct pid_entry tgid_base_stuff[] = {
3013         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3014         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3015 #ifdef CONFIG_CHECKPOINT_RESTORE
3016         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3017 #endif
3018         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3019         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3020 #ifdef CONFIG_NET
3021         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3022 #endif
3023         REG("environ",    S_IRUSR, proc_environ_operations),
3024         INF("auxv",       S_IRUSR, proc_pid_auxv),
3025         ONE("status",     S_IRUGO, proc_pid_status),
3026         ONE("personality", S_IRUGO, proc_pid_personality),
3027         INF("limits",     S_IRUGO, proc_pid_limits),
3028 #ifdef CONFIG_SCHED_DEBUG
3029         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3030 #endif
3031 #ifdef CONFIG_SCHED_AUTOGROUP
3032         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3033 #endif
3034         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3035 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3036         INF("syscall",    S_IRUGO, proc_pid_syscall),
3037 #endif
3038         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3039         ONE("stat",       S_IRUGO, proc_tgid_stat),
3040         ONE("statm",      S_IRUGO, proc_pid_statm),
3041         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3042 #ifdef CONFIG_NUMA
3043         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3044 #endif
3045         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3046         LNK("cwd",        proc_cwd_link),
3047         LNK("root",       proc_root_link),
3048         LNK("exe",        proc_exe_link),
3049         REG("mounts",     S_IRUGO, proc_mounts_operations),
3050         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3051         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3052 #ifdef CONFIG_PROC_PAGE_MONITOR
3053         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3054         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3055         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3056 #endif
3057 #ifdef CONFIG_SECURITY
3058         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3059 #endif
3060 #ifdef CONFIG_KALLSYMS
3061         INF("wchan",      S_IRUGO, proc_pid_wchan),
3062 #endif
3063 #ifdef CONFIG_STACKTRACE
3064         ONE("stack",      S_IRUGO, proc_pid_stack),
3065 #endif
3066 #ifdef CONFIG_SCHEDSTATS
3067         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3068 #endif
3069 #ifdef CONFIG_LATENCYTOP
3070         REG("latency",  S_IRUGO, proc_lstats_operations),
3071 #endif
3072 #ifdef CONFIG_PROC_PID_CPUSET
3073         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3074 #endif
3075 #ifdef CONFIG_CGROUPS
3076         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3077 #endif
3078         INF("oom_score",  S_IRUGO, proc_oom_score),
3079         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3080         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3081 #ifdef CONFIG_AUDITSYSCALL
3082         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3083         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3084 #endif
3085 #ifdef CONFIG_FAULT_INJECTION
3086         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3087 #endif
3088 #ifdef CONFIG_ELF_CORE
3089         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3090 #endif
3091 #ifdef CONFIG_TASK_IO_ACCOUNTING
3092         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3093 #endif
3094 #ifdef CONFIG_HARDWALL
3095         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3096 #endif
3097 #ifdef CONFIG_USER_NS
3098         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3099         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3100 #endif
3101 };
3102
3103 static int proc_tgid_base_readdir(struct file * filp,
3104                              void * dirent, filldir_t filldir)
3105 {
3106         return proc_pident_readdir(filp,dirent,filldir,
3107                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3108 }
3109
3110 static const struct file_operations proc_tgid_base_operations = {
3111         .read           = generic_read_dir,
3112         .readdir        = proc_tgid_base_readdir,
3113         .llseek         = default_llseek,
3114 };
3115
3116 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3117         return proc_pident_lookup(dir, dentry,
3118                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3119 }
3120
3121 static const struct inode_operations proc_tgid_base_inode_operations = {
3122         .lookup         = proc_tgid_base_lookup,
3123         .getattr        = pid_getattr,
3124         .setattr        = proc_setattr,
3125         .permission     = proc_pid_permission,
3126 };
3127
3128 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3129 {
3130         struct dentry *dentry, *leader, *dir;
3131         char buf[PROC_NUMBUF];
3132         struct qstr name;
3133
3134         name.name = buf;
3135         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3136         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3137         if (dentry) {
3138                 shrink_dcache_parent(dentry);
3139                 d_drop(dentry);
3140                 dput(dentry);
3141         }
3142
3143         name.name = buf;
3144         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3145         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3146         if (!leader)
3147                 goto out;
3148
3149         name.name = "task";
3150         name.len = strlen(name.name);
3151         dir = d_hash_and_lookup(leader, &name);
3152         if (!dir)
3153                 goto out_put_leader;
3154
3155         name.name = buf;
3156         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3157         dentry = d_hash_and_lookup(dir, &name);
3158         if (dentry) {
3159                 shrink_dcache_parent(dentry);
3160                 d_drop(dentry);
3161                 dput(dentry);
3162         }
3163
3164         dput(dir);
3165 out_put_leader:
3166         dput(leader);
3167 out:
3168         return;
3169 }
3170
3171 /**
3172  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3173  * @task: task that should be flushed.
3174  *
3175  * When flushing dentries from proc, one needs to flush them from global
3176  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3177  * in. This call is supposed to do all of this job.
3178  *
3179  * Looks in the dcache for
3180  * /proc/@pid
3181  * /proc/@tgid/task/@pid
3182  * if either directory is present flushes it and all of it'ts children
3183  * from the dcache.
3184  *
3185  * It is safe and reasonable to cache /proc entries for a task until
3186  * that task exits.  After that they just clog up the dcache with
3187  * useless entries, possibly causing useful dcache entries to be
3188  * flushed instead.  This routine is proved to flush those useless
3189  * dcache entries at process exit time.
3190  *
3191  * NOTE: This routine is just an optimization so it does not guarantee
3192  *       that no dcache entries will exist at process exit time it
3193  *       just makes it very unlikely that any will persist.
3194  */
3195
3196 void proc_flush_task(struct task_struct *task)
3197 {
3198         int i;
3199         struct pid *pid, *tgid;
3200         struct upid *upid;
3201
3202         pid = task_pid(task);
3203         tgid = task_tgid(task);
3204
3205         for (i = 0; i <= pid->level; i++) {
3206                 upid = &pid->numbers[i];
3207                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3208                                         tgid->numbers[i].nr);
3209         }
3210
3211         upid = &pid->numbers[pid->level];
3212         if (upid->nr == 1)
3213                 pid_ns_release_proc(upid->ns);
3214 }
3215
3216 static struct dentry *proc_pid_instantiate(struct inode *dir,
3217                                            struct dentry * dentry,
3218                                            struct task_struct *task, const void *ptr)
3219 {
3220         struct dentry *error = ERR_PTR(-ENOENT);
3221         struct inode *inode;
3222
3223         inode = proc_pid_make_inode(dir->i_sb, task);
3224         if (!inode)
3225                 goto out;
3226
3227         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3228         inode->i_op = &proc_tgid_base_inode_operations;
3229         inode->i_fop = &proc_tgid_base_operations;
3230         inode->i_flags|=S_IMMUTABLE;
3231
3232         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3233                                                   ARRAY_SIZE(tgid_base_stuff)));
3234
3235         d_set_d_op(dentry, &pid_dentry_operations);
3236
3237         d_add(dentry, inode);
3238         /* Close the race of the process dying before we return the dentry */
3239         if (pid_revalidate(dentry, NULL))
3240                 error = NULL;
3241 out:
3242         return error;
3243 }
3244
3245 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3246 {
3247         struct dentry *result;
3248         struct task_struct *task;
3249         unsigned tgid;
3250         struct pid_namespace *ns;
3251
3252         result = proc_base_lookup(dir, dentry);
3253         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3254                 goto out;
3255
3256         tgid = name_to_int(dentry);
3257         if (tgid == ~0U)
3258                 goto out;
3259
3260         ns = dentry->d_sb->s_fs_info;
3261         rcu_read_lock();
3262         task = find_task_by_pid_ns(tgid, ns);
3263         if (task)
3264                 get_task_struct(task);
3265         rcu_read_unlock();
3266         if (!task)
3267                 goto out;
3268
3269         result = proc_pid_instantiate(dir, dentry, task, NULL);
3270         put_task_struct(task);
3271 out:
3272         return result;
3273 }
3274
3275 /*
3276  * Find the first task with tgid >= tgid
3277  *
3278  */
3279 struct tgid_iter {
3280         unsigned int tgid;
3281         struct task_struct *task;
3282 };
3283 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3284 {
3285         struct pid *pid;
3286
3287         if (iter.task)
3288                 put_task_struct(iter.task);
3289         rcu_read_lock();
3290 retry:
3291         iter.task = NULL;
3292         pid = find_ge_pid(iter.tgid, ns);
3293         if (pid) {
3294                 iter.tgid = pid_nr_ns(pid, ns);
3295                 iter.task = pid_task(pid, PIDTYPE_PID);
3296                 /* What we to know is if the pid we have find is the
3297                  * pid of a thread_group_leader.  Testing for task
3298                  * being a thread_group_leader is the obvious thing
3299                  * todo but there is a window when it fails, due to
3300                  * the pid transfer logic in de_thread.
3301                  *
3302                  * So we perform the straight forward test of seeing
3303                  * if the pid we have found is the pid of a thread
3304                  * group leader, and don't worry if the task we have
3305                  * found doesn't happen to be a thread group leader.
3306                  * As we don't care in the case of readdir.
3307                  */
3308                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3309                         iter.tgid += 1;
3310                         goto retry;
3311                 }
3312                 get_task_struct(iter.task);
3313         }
3314         rcu_read_unlock();
3315         return iter;
3316 }
3317
3318 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3319
3320 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3321         struct tgid_iter iter)
3322 {
3323         char name[PROC_NUMBUF];
3324         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3325         return proc_fill_cache(filp, dirent, filldir, name, len,
3326                                 proc_pid_instantiate, iter.task, NULL);
3327 }
3328
3329 static int fake_filldir(void *buf, const char *name, int namelen,
3330                         loff_t offset, u64 ino, unsigned d_type)
3331 {
3332         return 0;
3333 }
3334
3335 /* for the /proc/ directory itself, after non-process stuff has been done */
3336 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3337 {
3338         unsigned int nr;
3339         struct task_struct *reaper;
3340         struct tgid_iter iter;
3341         struct pid_namespace *ns;
3342         filldir_t __filldir;
3343
3344         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3345                 goto out_no_task;
3346         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3347
3348         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3349         if (!reaper)
3350                 goto out_no_task;
3351
3352         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3353                 const struct pid_entry *p = &proc_base_stuff[nr];
3354                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3355                         goto out;
3356         }
3357
3358         ns = filp->f_dentry->d_sb->s_fs_info;
3359         iter.task = NULL;
3360         iter.tgid = filp->f_pos - TGID_OFFSET;
3361         for (iter = next_tgid(ns, iter);
3362              iter.task;
3363              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3364                 if (has_pid_permissions(ns, iter.task, 2))
3365                         __filldir = filldir;
3366                 else
3367                         __filldir = fake_filldir;
3368
3369                 filp->f_pos = iter.tgid + TGID_OFFSET;
3370                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3371                         put_task_struct(iter.task);
3372                         goto out;
3373                 }
3374         }
3375         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3376 out:
3377         put_task_struct(reaper);
3378 out_no_task:
3379         return 0;
3380 }
3381
3382 /*
3383  * Tasks
3384  */
3385 static const struct pid_entry tid_base_stuff[] = {
3386         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3387         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3388         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3389         REG("environ",   S_IRUSR, proc_environ_operations),
3390         INF("auxv",      S_IRUSR, proc_pid_auxv),
3391         ONE("status",    S_IRUGO, proc_pid_status),
3392         ONE("personality", S_IRUGO, proc_pid_personality),
3393         INF("limits",    S_IRUGO, proc_pid_limits),
3394 #ifdef CONFIG_SCHED_DEBUG
3395         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3396 #endif
3397         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3398 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3399         INF("syscall",   S_IRUGO, proc_pid_syscall),
3400 #endif
3401         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3402         ONE("stat",      S_IRUGO, proc_tid_stat),
3403         ONE("statm",     S_IRUGO, proc_pid_statm),
3404         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3405 #ifdef CONFIG_NUMA
3406         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3407 #endif
3408         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3409         LNK("cwd",       proc_cwd_link),
3410         LNK("root",      proc_root_link),
3411         LNK("exe",       proc_exe_link),
3412         REG("mounts",    S_IRUGO, proc_mounts_operations),
3413         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3414 #ifdef CONFIG_PROC_PAGE_MONITOR
3415         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3416         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3417         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3418 #endif
3419 #ifdef CONFIG_SECURITY
3420         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3421 #endif
3422 #ifdef CONFIG_KALLSYMS
3423         INF("wchan",     S_IRUGO, proc_pid_wchan),
3424 #endif
3425 #ifdef CONFIG_STACKTRACE
3426         ONE("stack",      S_IRUGO, proc_pid_stack),
3427 #endif
3428 #ifdef CONFIG_SCHEDSTATS
3429         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3430 #endif
3431 #ifdef CONFIG_LATENCYTOP
3432         REG("latency",  S_IRUGO, proc_lstats_operations),
3433 #endif
3434 #ifdef CONFIG_PROC_PID_CPUSET
3435         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3436 #endif
3437 #ifdef CONFIG_CGROUPS
3438         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3439 #endif
3440         INF("oom_score", S_IRUGO, proc_oom_score),
3441         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3442         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3443 #ifdef CONFIG_AUDITSYSCALL
3444         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3445         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3446 #endif
3447 #ifdef CONFIG_FAULT_INJECTION
3448         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3449 #endif
3450 #ifdef CONFIG_TASK_IO_ACCOUNTING
3451         INF("io",       S_IRUSR, proc_tid_io_accounting),
3452 #endif
3453 #ifdef CONFIG_HARDWALL
3454         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3455 #endif
3456 #ifdef CONFIG_USER_NS
3457         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3458         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3459 #endif
3460 };
3461
3462 static int proc_tid_base_readdir(struct file * filp,
3463                              void * dirent, filldir_t filldir)
3464 {
3465         return proc_pident_readdir(filp,dirent,filldir,
3466                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3467 }
3468
3469 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3470         return proc_pident_lookup(dir, dentry,
3471                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3472 }
3473
3474 static const struct file_operations proc_tid_base_operations = {
3475         .read           = generic_read_dir,
3476         .readdir        = proc_tid_base_readdir,
3477         .llseek         = default_llseek,
3478 };
3479
3480 static const struct inode_operations proc_tid_base_inode_operations = {
3481         .lookup         = proc_tid_base_lookup,
3482         .getattr        = pid_getattr,
3483         .setattr        = proc_setattr,
3484 };
3485
3486 static struct dentry *proc_task_instantiate(struct inode *dir,
3487         struct dentry *dentry, struct task_struct *task, const void *ptr)
3488 {
3489         struct dentry *error = ERR_PTR(-ENOENT);
3490         struct inode *inode;
3491         inode = proc_pid_make_inode(dir->i_sb, task);
3492
3493         if (!inode)
3494                 goto out;
3495         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3496         inode->i_op = &proc_tid_base_inode_operations;
3497         inode->i_fop = &proc_tid_base_operations;
3498         inode->i_flags|=S_IMMUTABLE;
3499
3500         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3501                                                   ARRAY_SIZE(tid_base_stuff)));
3502
3503         d_set_d_op(dentry, &pid_dentry_operations);
3504
3505         d_add(dentry, inode);
3506         /* Close the race of the process dying before we return the dentry */
3507         if (pid_revalidate(dentry, NULL))
3508                 error = NULL;
3509 out:
3510         return error;
3511 }
3512
3513 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3514 {
3515         struct dentry *result = ERR_PTR(-ENOENT);
3516         struct task_struct *task;
3517         struct task_struct *leader = get_proc_task(dir);
3518         unsigned tid;
3519         struct pid_namespace *ns;
3520
3521         if (!leader)
3522                 goto out_no_task;
3523
3524         tid = name_to_int(dentry);
3525         if (tid == ~0U)
3526                 goto out;
3527
3528         ns = dentry->d_sb->s_fs_info;
3529         rcu_read_lock();
3530         task = find_task_by_pid_ns(tid, ns);
3531         if (task)
3532                 get_task_struct(task);
3533         rcu_read_unlock();
3534         if (!task)
3535                 goto out;
3536         if (!same_thread_group(leader, task))
3537                 goto out_drop_task;
3538
3539         result = proc_task_instantiate(dir, dentry, task, NULL);
3540 out_drop_task:
3541         put_task_struct(task);
3542 out:
3543         put_task_struct(leader);
3544 out_no_task:
3545         return result;
3546 }
3547
3548 /*
3549  * Find the first tid of a thread group to return to user space.
3550  *
3551  * Usually this is just the thread group leader, but if the users
3552  * buffer was too small or there was a seek into the middle of the
3553  * directory we have more work todo.
3554  *
3555  * In the case of a short read we start with find_task_by_pid.
3556  *
3557  * In the case of a seek we start with the leader and walk nr
3558  * threads past it.
3559  */
3560 static struct task_struct *first_tid(struct task_struct *leader,
3561                 int tid, int nr, struct pid_namespace *ns)
3562 {
3563         struct task_struct *pos;
3564
3565         rcu_read_lock();
3566         /* Attempt to start with the pid of a thread */
3567         if (tid && (nr > 0)) {
3568                 pos = find_task_by_pid_ns(tid, ns);
3569                 if (pos && (pos->group_leader == leader))
3570                         goto found;
3571         }
3572
3573         /* If nr exceeds the number of threads there is nothing todo */
3574         pos = NULL;
3575         if (nr && nr >= get_nr_threads(leader))
3576                 goto out;
3577
3578         /* If we haven't found our starting place yet start
3579          * with the leader and walk nr threads forward.
3580          */
3581         for (pos = leader; nr > 0; --nr) {
3582                 pos = next_thread(pos);
3583                 if (pos == leader) {
3584                         pos = NULL;
3585                         goto out;
3586                 }
3587         }
3588 found:
3589         get_task_struct(pos);
3590 out:
3591         rcu_read_unlock();
3592         return pos;
3593 }
3594
3595 /*
3596  * Find the next thread in the thread list.
3597  * Return NULL if there is an error or no next thread.
3598  *
3599  * The reference to the input task_struct is released.
3600  */
3601 static struct task_struct *next_tid(struct task_struct *start)
3602 {
3603         struct task_struct *pos = NULL;
3604         rcu_read_lock();
3605         if (pid_alive(start)) {
3606                 pos = next_thread(start);
3607                 if (thread_group_leader(pos))
3608                         pos = NULL;
3609                 else
3610                         get_task_struct(pos);
3611         }
3612         rcu_read_unlock();
3613         put_task_struct(start);
3614         return pos;
3615 }
3616
3617 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3618         struct task_struct *task, int tid)
3619 {
3620         char name[PROC_NUMBUF];
3621         int len = snprintf(name, sizeof(name), "%d", tid);
3622         return proc_fill_cache(filp, dirent, filldir, name, len,
3623                                 proc_task_instantiate, task, NULL);
3624 }
3625
3626 /* for the /proc/TGID/task/ directories */
3627 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3628 {
3629         struct dentry *dentry = filp->f_path.dentry;
3630         struct inode *inode = dentry->d_inode;
3631         struct task_struct *leader = NULL;
3632         struct task_struct *task;
3633         int retval = -ENOENT;
3634         ino_t ino;
3635         int tid;
3636         struct pid_namespace *ns;
3637
3638         task = get_proc_task(inode);
3639         if (!task)
3640                 goto out_no_task;
3641         rcu_read_lock();
3642         if (pid_alive(task)) {
3643                 leader = task->group_leader;
3644                 get_task_struct(leader);
3645         }
3646         rcu_read_unlock();
3647         put_task_struct(task);
3648         if (!leader)
3649                 goto out_no_task;
3650         retval = 0;
3651
3652         switch ((unsigned long)filp->f_pos) {
3653         case 0:
3654                 ino = inode->i_ino;
3655                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3656                         goto out;
3657                 filp->f_pos++;
3658                 /* fall through */
3659         case 1:
3660                 ino = parent_ino(dentry);
3661                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3662                         goto out;
3663                 filp->f_pos++;
3664                 /* fall through */
3665         }
3666
3667         /* f_version caches the tgid value that the last readdir call couldn't
3668          * return. lseek aka telldir automagically resets f_version to 0.
3669          */
3670         ns = filp->f_dentry->d_sb->s_fs_info;
3671         tid = (int)filp->f_version;
3672         filp->f_version = 0;
3673         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3674              task;
3675              task = next_tid(task), filp->f_pos++) {
3676                 tid = task_pid_nr_ns(task, ns);
3677                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3678                         /* returning this tgid failed, save it as the first
3679                          * pid for the next readir call */
3680                         filp->f_version = (u64)tid;
3681                         put_task_struct(task);
3682                         break;
3683                 }
3684         }
3685 out:
3686         put_task_struct(leader);
3687 out_no_task:
3688         return retval;
3689 }
3690
3691 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3692 {
3693         struct inode *inode = dentry->d_inode;
3694         struct task_struct *p = get_proc_task(inode);
3695         generic_fillattr(inode, stat);
3696
3697         if (p) {
3698                 stat->nlink += get_nr_threads(p);
3699                 put_task_struct(p);
3700         }
3701
3702         return 0;
3703 }
3704
3705 static const struct inode_operations proc_task_inode_operations = {
3706         .lookup         = proc_task_lookup,
3707         .getattr        = proc_task_getattr,
3708         .setattr        = proc_setattr,
3709         .permission     = proc_pid_permission,
3710 };
3711
3712 static const struct file_operations proc_task_operations = {
3713         .read           = generic_read_dir,
3714         .readdir        = proc_task_readdir,
3715         .llseek         = default_llseek,
3716 };