fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/user_namespace.h>
  85 #include <linux/fs_struct.h>
  86 #include <linux/slab.h>
  87 #include <linux/flex_array.h>
  88 #ifdef CONFIG_HARDWALL
  89 #include <asm/hardwall.h>
  90 #endif
  91 #include <trace/events/oom.h>
  92 #include "internal.h"
  93
  94 /* NOTE:
  95  *      Implementing inode permission operations in /proc is almost
  96  *      certainly an error.  Permission checks need to happen during
  97  *      each system call not at open time.  The reason is that most of
  98  *      what we wish to check for permissions in /proc varies at runtime.
  99  *
 100  *      The classic example of a problem is opening file descriptors
 101  *      in /proc for a task before it execs a suid executable.
 102  */
 103
 104 struct pid_entry {
 105         char *name;
 106         int len;
 107         umode_t mode;
 108         const struct inode_operations *iop;
 109         const struct file_operations *fop;
 110         union proc_op op;
 111 };
 112
 113 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 114         .name = (NAME),                                 \
 115         .len  = sizeof(NAME) - 1,                       \
 116         .mode = MODE,                                   \
 117         .iop  = IOP,                                    \
 118         .fop  = FOP,                                    \
 119         .op   = OP,                                     \
 120 }
 121
 122 #define DIR(NAME, MODE, iops, fops)     \
 123         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 124 #define LNK(NAME, get_link)                                     \
 125         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 126                 &proc_pid_link_inode_operations, NULL,          \
 127                 { .proc_get_link = get_link } )
 128 #define REG(NAME, MODE, fops)                           \
 129         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 130 #define INF(NAME, MODE, read)                           \
 131         NOD(NAME, (S_IFREG|(MODE)),                     \
 132                 NULL, &proc_info_file_operations,       \
 133                 { .proc_read = read } )
 134 #define ONE(NAME, MODE, show)                           \
 135         NOD(NAME, (S_IFREG|(MODE)),                     \
 136                 NULL, &proc_single_file_operations,     \
 137                 { .proc_show = show } )
 138
 139 static int proc_fd_permission(struct inode *inode, int mask);
 140
 141 /*
 142  * Count the number of hardlinks for the pid_entry table, excluding the .
 143  * and .. links.
 144  */
 145 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 146         unsigned int n)
 147 {
 148         unsigned int i;
 149         unsigned int count;
 150
 151         count = 0;
 152         for (i = 0; i < n; ++i) {
 153                 if (S_ISDIR(entries[i].mode))
 154                         ++count;
 155         }
 156
 157         return count;
 158 }
 159
 160 static int get_task_root(struct task_struct *task, struct path *root)
 161 {
 162         int result = -ENOENT;
 163
 164         task_lock(task);
 165         if (task->fs) {
 166                 get_fs_root(task->fs, root);
 167                 result = 0;
 168         }
 169         task_unlock(task);
 170         return result;
 171 }
 172
 173 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 174 {
 175         struct task_struct *task = get_proc_task(dentry->d_inode);
 176         int result = -ENOENT;
 177
 178         if (task) {
 179                 task_lock(task);
 180                 if (task->fs) {
 181                         get_fs_pwd(task->fs, path);
 182                         result = 0;
 183                 }
 184                 task_unlock(task);
 185                 put_task_struct(task);
 186         }
 187         return result;
 188 }
 189
 190 static int proc_root_link(struct dentry *dentry, struct path *path)
 191 {
 192         struct task_struct *task = get_proc_task(dentry->d_inode);
 193         int result = -ENOENT;
 194
 195         if (task) {
 196                 result = get_task_root(task, path);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 203 {
 204         int res = 0;
 205         unsigned int len;
 206         struct mm_struct *mm = get_task_mm(task);
 207         if (!mm)
 208                 goto out;
 209         if (!mm->arg_end)
 210                 goto out_mm;    /* Shh! No looking before we're done */
 211
 212         len = mm->arg_end - mm->arg_start;
 213
 214         if (len > PAGE_SIZE)
 215                 len = PAGE_SIZE;
 216
 217         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 218
 219         // If the nul at the end of args has been overwritten, then
 220         // assume application is using setproctitle(3).
 221         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 222                 len = strnlen(buffer, res);
 223                 if (len < res) {
 224                     res = len;
 225                 } else {
 226                         len = mm->env_end - mm->env_start;
 227                         if (len > PAGE_SIZE - res)
 228                                 len = PAGE_SIZE - res;
 229                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 230                         res = strnlen(buffer, res);
 231                 }
 232         }
 233 out_mm:
 234         mmput(mm);
 235 out:
 236         return res;
 237 }
 238
 239 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 240 {
 241         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 242         int res = PTR_ERR(mm);
 243         if (mm && !IS_ERR(mm)) {
 244                 unsigned int nwords = 0;
 245                 do {
 246                         nwords += 2;
 247                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 248                 res = nwords * sizeof(mm->saved_auxv[0]);
 249                 if (res > PAGE_SIZE)
 250                         res = PAGE_SIZE;
 251                 memcpy(buffer, mm->saved_auxv, res);
 252                 mmput(mm);
 253         }
 254         return res;
 255 }
 256
 257
 258 #ifdef CONFIG_KALLSYMS
 259 /*
 260  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 261  * Returns the resolved symbol.  If that fails, simply return the address.
 262  */
 263 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 264 {
 265         unsigned long wchan;
 266         char symname[KSYM_NAME_LEN];
 267
 268         wchan = get_wchan(task);
 269
 270         if (lookup_symbol_name(wchan, symname) < 0)
 271                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 272                         return 0;
 273                 else
 274                         return sprintf(buffer, "%lu", wchan);
 275         else
 276                 return sprintf(buffer, "%s", symname);
 277 }
 278 #endif /* CONFIG_KALLSYMS */
 279
 280 static int lock_trace(struct task_struct *task)
 281 {
 282         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 283         if (err)
 284                 return err;
 285         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 286                 mutex_unlock(&task->signal->cred_guard_mutex);
 287                 return -EPERM;
 288         }
 289         return 0;
 290 }
 291
 292 static void unlock_trace(struct task_struct *task)
 293 {
 294         mutex_unlock(&task->signal->cred_guard_mutex);
 295 }
 296
 297 #ifdef CONFIG_STACKTRACE
 298
 299 #define MAX_STACK_TRACE_DEPTH   64
 300
 301 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 302                           struct pid *pid, struct task_struct *task)
 303 {
 304         struct stack_trace trace;
 305         unsigned long *entries;
 306         int err;
 307         int i;
 308
 309         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 310         if (!entries)
 311                 return -ENOMEM;
 312
 313         trace.nr_entries        = 0;
 314         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 315         trace.entries           = entries;
 316         trace.skip              = 0;
 317
 318         err = lock_trace(task);
 319         if (!err) {
 320                 save_stack_trace_tsk(task, &trace);
 321
 322                 for (i = 0; i < trace.nr_entries; i++) {
 323                         seq_printf(m, "[<%pK>] %pS\n",
 324                                    (void *)entries[i], (void *)entries[i]);
 325                 }
 326                 unlock_trace(task);
 327         }
 328         kfree(entries);
 329
 330         return err;
 331 }
 332 #endif
 333
 334 #ifdef CONFIG_SCHEDSTATS
 335 /*
 336  * Provides /proc/PID/schedstat
 337  */
 338 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 339 {
 340         return sprintf(buffer, "%llu %llu %lu\n",
 341                         (unsigned long long)task->se.sum_exec_runtime,
 342                         (unsigned long long)task->sched_info.run_delay,
 343                         task->sched_info.pcount);
 344 }
 345 #endif
 346
 347 #ifdef CONFIG_LATENCYTOP
 348 static int lstats_show_proc(struct seq_file *m, void *v)
 349 {
 350         int i;
 351         struct inode *inode = m->private;
 352         struct task_struct *task = get_proc_task(inode);
 353
 354         if (!task)
 355                 return -ESRCH;
 356         seq_puts(m, "Latency Top version : v0.1\n");
 357         for (i = 0; i < 32; i++) {
 358                 struct latency_record *lr = &task->latency_record[i];
 359                 if (lr->backtrace[0]) {
 360                         int q;
 361                         seq_printf(m, "%i %li %li",
 362                                    lr->count, lr->time, lr->max);
 363                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 364                                 unsigned long bt = lr->backtrace[q];
 365                                 if (!bt)
 366                                         break;
 367                                 if (bt == ULONG_MAX)
 368                                         break;
 369                                 seq_printf(m, " %ps", (void *)bt);
 370                         }
 371                         seq_putc(m, '\n');
 372                 }
 373
 374         }
 375         put_task_struct(task);
 376         return 0;
 377 }
 378
 379 static int lstats_open(struct inode *inode, struct file *file)
 380 {
 381         return single_open(file, lstats_show_proc, inode);
 382 }
 383
 384 static ssize_t lstats_write(struct file *file, const char __user *buf,
 385                             size_t count, loff_t *offs)
 386 {
 387         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 388
 389         if (!task)
 390                 return -ESRCH;
 391         clear_all_latency_tracing(task);
 392         put_task_struct(task);
 393
 394         return count;
 395 }
 396
 397 static const struct file_operations proc_lstats_operations = {
 398         .open           = lstats_open,
 399         .read           = seq_read,
 400         .write          = lstats_write,
 401         .llseek         = seq_lseek,
 402         .release        = single_release,
 403 };
 404
 405 #endif
 406
 407 static int proc_oom_score(struct task_struct *task, char *buffer)
 408 {
 409         unsigned long totalpages = totalram_pages + total_swap_pages;
 410         unsigned long points = 0;
 411
 412         read_lock(&tasklist_lock);
 413         if (pid_alive(task))
 414                 points = oom_badness(task, NULL, NULL, totalpages) *
 415                                                 1000 / totalpages;
 416         read_unlock(&tasklist_lock);
 417         return sprintf(buffer, "%lu\n", points);
 418 }
 419
 420 struct limit_names {
 421         char *name;
 422         char *unit;
 423 };
 424
 425 static const struct limit_names lnames[RLIM_NLIMITS] = {
 426         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 427         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 428         [RLIMIT_DATA] = {"Max data size", "bytes"},
 429         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 430         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 431         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 432         [RLIMIT_NPROC] = {"Max processes", "processes"},
 433         [RLIMIT_NOFILE] = {"Max open files", "files"},
 434         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 435         [RLIMIT_AS] = {"Max address space", "bytes"},
 436         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 437         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 438         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 439         [RLIMIT_NICE] = {"Max nice priority", NULL},
 440         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 441         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 442 };
 443
 444 /* Display limits for a process */
 445 static int proc_pid_limits(struct task_struct *task, char *buffer)
 446 {
 447         unsigned int i;
 448         int count = 0;
 449         unsigned long flags;
 450         char *bufptr = buffer;
 451
 452         struct rlimit rlim[RLIM_NLIMITS];
 453
 454         if (!lock_task_sighand(task, &flags))
 455                 return 0;
 456         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 457         unlock_task_sighand(task, &flags);
 458
 459         /*
 460          * print the file header
 461          */
 462         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 463                         "Limit", "Soft Limit", "Hard Limit", "Units");
 464
 465         for (i = 0; i < RLIM_NLIMITS; i++) {
 466                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 467                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 468                                          lnames[i].name, "unlimited");
 469                 else
 470                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 471                                          lnames[i].name, rlim[i].rlim_cur);
 472
 473                 if (rlim[i].rlim_max == RLIM_INFINITY)
 474                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 475                 else
 476                         count += sprintf(&bufptr[count], "%-20lu ",
 477                                          rlim[i].rlim_max);
 478
 479                 if (lnames[i].unit)
 480                         count += sprintf(&bufptr[count], "%-10s\n",
 481                                          lnames[i].unit);
 482                 else
 483                         count += sprintf(&bufptr[count], "\n");
 484         }
 485
 486         return count;
 487 }
 488
 489 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 490 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 491 {
 492         long nr;
 493         unsigned long args[6], sp, pc;
 494         int res = lock_trace(task);
 495         if (res)
 496                 return res;
 497
 498         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 499                 res = sprintf(buffer, "running\n");
 500         else if (nr < 0)
 501                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 502         else
 503                 res = sprintf(buffer,
 504                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 505                        nr,
 506                        args[0], args[1], args[2], args[3], args[4], args[5],
 507                        sp, pc);
 508         unlock_trace(task);
 509         return res;
 510 }
 511 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 512
 513 /************************************************************************/
 514 /*                       Here the fs part begins                        */
 515 /************************************************************************/
 516
 517 /* permission checks */
 518 static int proc_fd_access_allowed(struct inode *inode)
 519 {
 520         struct task_struct *task;
 521         int allowed = 0;
 522         /* Allow access to a task's file descriptors if it is us or we
 523          * may use ptrace attach to the process and find out that
 524          * information.
 525          */
 526         task = get_proc_task(inode);
 527         if (task) {
 528                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 529                 put_task_struct(task);
 530         }
 531         return allowed;
 532 }
 533
 534 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 535 {
 536         int error;
 537         struct inode *inode = dentry->d_inode;
 538
 539         if (attr->ia_valid & ATTR_MODE)
 540                 return -EPERM;
 541
 542         error = inode_change_ok(inode, attr);
 543         if (error)
 544                 return error;
 545
 546         if ((attr->ia_valid & ATTR_SIZE) &&
 547             attr->ia_size != i_size_read(inode)) {
 548                 error = vmtruncate(inode, attr->ia_size);
 549                 if (error)
 550                         return error;
 551         }
 552
 553         setattr_copy(inode, attr);
 554         mark_inode_dirty(inode);
 555         return 0;
 556 }
 557
 558 /*
 559  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 560  * or euid/egid (for hide_pid_min=2)?
 561  */
 562 static bool has_pid_permissions(struct pid_namespace *pid,
 563                                  struct task_struct *task,
 564                                  int hide_pid_min)
 565 {
 566         if (pid->hide_pid < hide_pid_min)
 567                 return true;
 568         if (in_group_p(pid->pid_gid))
 569                 return true;
 570         return ptrace_may_access(task, PTRACE_MODE_READ);
 571 }
 572
 573
 574 static int proc_pid_permission(struct inode *inode, int mask)
 575 {
 576         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 577         struct task_struct *task;
 578         bool has_perms;
 579
 580         task = get_proc_task(inode);
 581         if (!task)
 582                 return -ESRCH;
 583         has_perms = has_pid_permissions(pid, task, 1);
 584         put_task_struct(task);
 585
 586         if (!has_perms) {
 587                 if (pid->hide_pid == 2) {
 588                         /*
 589                          * Let's make getdents(), stat(), and open()
 590                          * consistent with each other.  If a process
 591                          * may not stat() a file, it shouldn't be seen
 592                          * in procfs at all.
 593                          */
 594                         return -ENOENT;
 595                 }
 596
 597                 return -EPERM;
 598         }
 599         return generic_permission(inode, mask);
 600 }
 601
 602
 603
 604 static const struct inode_operations proc_def_inode_operations = {
 605         .setattr        = proc_setattr,
 606 };
 607
 608 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 609
 610 static ssize_t proc_info_read(struct file * file, char __user * buf,
 611                           size_t count, loff_t *ppos)
 612 {
 613         struct inode * inode = file->f_path.dentry->d_inode;
 614         unsigned long page;
 615         ssize_t length;
 616         struct task_struct *task = get_proc_task(inode);
 617
 618         length = -ESRCH;
 619         if (!task)
 620                 goto out_no_task;
 621
 622         if (count > PROC_BLOCK_SIZE)
 623                 count = PROC_BLOCK_SIZE;
 624
 625         length = -ENOMEM;
 626         if (!(page = __get_free_page(GFP_TEMPORARY)))
 627                 goto out;
 628
 629         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 630
 631         if (length >= 0)
 632                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 633         free_page(page);
 634 out:
 635         put_task_struct(task);
 636 out_no_task:
 637         return length;
 638 }
 639
 640 static const struct file_operations proc_info_file_operations = {
 641         .read           = proc_info_read,
 642         .llseek         = generic_file_llseek,
 643 };
 644
 645 static int proc_single_show(struct seq_file *m, void *v)
 646 {
 647         struct inode *inode = m->private;
 648         struct pid_namespace *ns;
 649         struct pid *pid;
 650         struct task_struct *task;
 651         int ret;
 652
 653         ns = inode->i_sb->s_fs_info;
 654         pid = proc_pid(inode);
 655         task = get_pid_task(pid, PIDTYPE_PID);
 656         if (!task)
 657                 return -ESRCH;
 658
 659         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 660
 661         put_task_struct(task);
 662         return ret;
 663 }
 664
 665 static int proc_single_open(struct inode *inode, struct file *filp)
 666 {
 667         return single_open(filp, proc_single_show, inode);
 668 }
 669
 670 static const struct file_operations proc_single_file_operations = {
 671         .open           = proc_single_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = single_release,
 675 };
 676
 677 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 678 {
 679         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 680         struct mm_struct *mm;
 681
 682         if (!task)
 683                 return -ESRCH;
 684
 685         mm = mm_access(task, mode);
 686         put_task_struct(task);
 687
 688         if (IS_ERR(mm))
 689                 return PTR_ERR(mm);
 690
 691         if (mm) {
 692                 /* ensure this mm_struct can't be freed */
 693                 atomic_inc(&mm->mm_count);
 694                 /* but do not pin its memory */
 695                 mmput(mm);
 696         }
 697
 698         file->private_data = mm;
 699
 700         return 0;
 701 }
 702
 703 static int mem_open(struct inode *inode, struct file *file)
 704 {
 705         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 706
 707         /* OK to pass negative loff_t, we can catch out-of-range */
 708         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 709
 710         return ret;
 711 }
 712
 713 static ssize_t mem_rw(struct file *file, char __user *buf,
 714                         size_t count, loff_t *ppos, int write)
 715 {
 716         struct mm_struct *mm = file->private_data;
 717         unsigned long addr = *ppos;
 718         ssize_t copied;
 719         char *page;
 720
 721         if (!mm)
 722                 return 0;
 723
 724         page = (char *)__get_free_page(GFP_TEMPORARY);
 725         if (!page)
 726                 return -ENOMEM;
 727
 728         copied = 0;
 729         if (!atomic_inc_not_zero(&mm->mm_users))
 730                 goto free;
 731
 732         while (count > 0) {
 733                 int this_len = min_t(int, count, PAGE_SIZE);
 734
 735                 if (write && copy_from_user(page, buf, this_len)) {
 736                         copied = -EFAULT;
 737                         break;
 738                 }
 739
 740                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 741                 if (!this_len) {
 742                         if (!copied)
 743                                 copied = -EIO;
 744                         break;
 745                 }
 746
 747                 if (!write && copy_to_user(buf, page, this_len)) {
 748                         copied = -EFAULT;
 749                         break;
 750                 }
 751
 752                 buf += this_len;
 753                 addr += this_len;
 754                 copied += this_len;
 755                 count -= this_len;
 756         }
 757         *ppos = addr;
 758
 759         mmput(mm);
 760 free:
 761         free_page((unsigned long) page);
 762         return copied;
 763 }
 764
 765 static ssize_t mem_read(struct file *file, char __user *buf,
 766                         size_t count, loff_t *ppos)
 767 {
 768         return mem_rw(file, buf, count, ppos, 0);
 769 }
 770
 771 static ssize_t mem_write(struct file *file, const char __user *buf,
 772                          size_t count, loff_t *ppos)
 773 {
 774         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 775 }
 776
 777 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 778 {
 779         switch (orig) {
 780         case 0:
 781                 file->f_pos = offset;
 782                 break;
 783         case 1:
 784                 file->f_pos += offset;
 785                 break;
 786         default:
 787                 return -EINVAL;
 788         }
 789         force_successful_syscall_return();
 790         return file->f_pos;
 791 }
 792
 793 static int mem_release(struct inode *inode, struct file *file)
 794 {
 795         struct mm_struct *mm = file->private_data;
 796         if (mm)
 797                 mmdrop(mm);
 798         return 0;
 799 }
 800
 801 static const struct file_operations proc_mem_operations = {
 802         .llseek         = mem_lseek,
 803         .read           = mem_read,
 804         .write          = mem_write,
 805         .open           = mem_open,
 806         .release        = mem_release,
 807 };
 808
 809 static int environ_open(struct inode *inode, struct file *file)
 810 {
 811         return __mem_open(inode, file, PTRACE_MODE_READ);
 812 }
 813
 814 static ssize_t environ_read(struct file *file, char __user *buf,
 815                         size_t count, loff_t *ppos)
 816 {
 817         char *page;
 818         unsigned long src = *ppos;
 819         int ret = 0;
 820         struct mm_struct *mm = file->private_data;
 821
 822         if (!mm)
 823                 return 0;
 824
 825         page = (char *)__get_free_page(GFP_TEMPORARY);
 826         if (!page)
 827                 return -ENOMEM;
 828
 829         ret = 0;
 830         if (!atomic_inc_not_zero(&mm->mm_users))
 831                 goto free;
 832         while (count > 0) {
 833                 size_t this_len, max_len;
 834                 int retval;
 835
 836                 if (src >= (mm->env_end - mm->env_start))
 837                         break;
 838
 839                 this_len = mm->env_end - (mm->env_start + src);
 840
 841                 max_len = min_t(size_t, PAGE_SIZE, count);
 842                 this_len = min(max_len, this_len);
 843
 844                 retval = access_remote_vm(mm, (mm->env_start + src),
 845                         page, this_len, 0);
 846
 847                 if (retval <= 0) {
 848                         ret = retval;
 849                         break;
 850                 }
 851
 852                 if (copy_to_user(buf, page, retval)) {
 853                         ret = -EFAULT;
 854                         break;
 855                 }
 856
 857                 ret += retval;
 858                 src += retval;
 859                 buf += retval;
 860                 count -= retval;
 861         }
 862         *ppos = src;
 863         mmput(mm);
 864
 865 free:
 866         free_page((unsigned long) page);
 867         return ret;
 868 }
 869
 870 static const struct file_operations proc_environ_operations = {
 871         .open           = environ_open,
 872         .read           = environ_read,
 873         .llseek         = generic_file_llseek,
 874         .release        = mem_release,
 875 };
 876
 877 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 878                                 size_t count, loff_t *ppos)
 879 {
 880         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 881         char buffer[PROC_NUMBUF];
 882         size_t len;
 883         int oom_adjust = OOM_DISABLE;
 884         unsigned long flags;
 885
 886         if (!task)
 887                 return -ESRCH;
 888
 889         if (lock_task_sighand(task, &flags)) {
 890                 oom_adjust = task->signal->oom_adj;
 891                 unlock_task_sighand(task, &flags);
 892         }
 893
 894         put_task_struct(task);
 895
 896         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 897
 898         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 899 }
 900
 901 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 902                                 size_t count, loff_t *ppos)
 903 {
 904         struct task_struct *task;
 905         char buffer[PROC_NUMBUF];
 906         int oom_adjust;
 907         unsigned long flags;
 908         int err;
 909
 910         memset(buffer, 0, sizeof(buffer));
 911         if (count > sizeof(buffer) - 1)
 912                 count = sizeof(buffer) - 1;
 913         if (copy_from_user(buffer, buf, count)) {
 914                 err = -EFAULT;
 915                 goto out;
 916         }
 917
 918         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 919         if (err)
 920                 goto out;
 921         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 922              oom_adjust != OOM_DISABLE) {
 923                 err = -EINVAL;
 924                 goto out;
 925         }
 926
 927         task = get_proc_task(file->f_path.dentry->d_inode);
 928         if (!task) {
 929                 err = -ESRCH;
 930                 goto out;
 931         }
 932
 933         task_lock(task);
 934         if (!task->mm) {
 935                 err = -EINVAL;
 936                 goto err_task_lock;
 937         }
 938
 939         if (!lock_task_sighand(task, &flags)) {
 940                 err = -ESRCH;
 941                 goto err_task_lock;
 942         }
 943
 944         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 945                 err = -EACCES;
 946                 goto err_sighand;
 947         }
 948
 949         /*
 950          * Warn that /proc/pid/oom_adj is deprecated, see
 951          * Documentation/feature-removal-schedule.txt.
 952          */
 953         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 954                   current->comm, task_pid_nr(current), task_pid_nr(task),
 955                   task_pid_nr(task));
 956         task->signal->oom_adj = oom_adjust;
 957         /*
 958          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 959          * value is always attainable.
 960          */
 961         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 962                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 963         else
 964                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 965                                                                 -OOM_DISABLE;
 966         trace_oom_score_adj_update(task);
 967 err_sighand:
 968         unlock_task_sighand(task, &flags);
 969 err_task_lock:
 970         task_unlock(task);
 971         put_task_struct(task);
 972 out:
 973         return err < 0 ? err : count;
 974 }
 975
 976 static const struct file_operations proc_oom_adjust_operations = {
 977         .read           = oom_adjust_read,
 978         .write          = oom_adjust_write,
 979         .llseek         = generic_file_llseek,
 980 };
 981
 982 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
 983                                         size_t count, loff_t *ppos)
 984 {
 985         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 986         char buffer[PROC_NUMBUF];
 987         int oom_score_adj = OOM_SCORE_ADJ_MIN;
 988         unsigned long flags;
 989         size_t len;
 990
 991         if (!task)
 992                 return -ESRCH;
 993         if (lock_task_sighand(task, &flags)) {
 994                 oom_score_adj = task->signal->oom_score_adj;
 995                 unlock_task_sighand(task, &flags);
 996         }
 997         put_task_struct(task);
 998         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
 999         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1000 }
1001
1002 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1003                                         size_t count, loff_t *ppos)
1004 {
1005         struct task_struct *task;
1006         char buffer[PROC_NUMBUF];
1007         unsigned long flags;
1008         int oom_score_adj;
1009         int err;
1010
1011         memset(buffer, 0, sizeof(buffer));
1012         if (count > sizeof(buffer) - 1)
1013                 count = sizeof(buffer) - 1;
1014         if (copy_from_user(buffer, buf, count)) {
1015                 err = -EFAULT;
1016                 goto out;
1017         }
1018
1019         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1020         if (err)
1021                 goto out;
1022         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1023                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1024                 err = -EINVAL;
1025                 goto out;
1026         }
1027
1028         task = get_proc_task(file->f_path.dentry->d_inode);
1029         if (!task) {
1030                 err = -ESRCH;
1031                 goto out;
1032         }
1033
1034         task_lock(task);
1035         if (!task->mm) {
1036                 err = -EINVAL;
1037                 goto err_task_lock;
1038         }
1039
1040         if (!lock_task_sighand(task, &flags)) {
1041                 err = -ESRCH;
1042                 goto err_task_lock;
1043         }
1044
1045         if (oom_score_adj < task->signal->oom_score_adj_min &&
1046                         !capable(CAP_SYS_RESOURCE)) {
1047                 err = -EACCES;
1048                 goto err_sighand;
1049         }
1050
1051         task->signal->oom_score_adj = oom_score_adj;
1052         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1053                 task->signal->oom_score_adj_min = oom_score_adj;
1054         trace_oom_score_adj_update(task);
1055         /*
1056          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1057          * always attainable.
1058          */
1059         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1060                 task->signal->oom_adj = OOM_DISABLE;
1061         else
1062                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1063                                                         OOM_SCORE_ADJ_MAX;
1064 err_sighand:
1065         unlock_task_sighand(task, &flags);
1066 err_task_lock:
1067         task_unlock(task);
1068         put_task_struct(task);
1069 out:
1070         return err < 0 ? err : count;
1071 }
1072
1073 static const struct file_operations proc_oom_score_adj_operations = {
1074         .read           = oom_score_adj_read,
1075         .write          = oom_score_adj_write,
1076         .llseek         = default_llseek,
1077 };
1078
1079 #ifdef CONFIG_AUDITSYSCALL
1080 #define TMPBUFLEN 21
1081 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1082                                   size_t count, loff_t *ppos)
1083 {
1084         struct inode * inode = file->f_path.dentry->d_inode;
1085         struct task_struct *task = get_proc_task(inode);
1086         ssize_t length;
1087         char tmpbuf[TMPBUFLEN];
1088
1089         if (!task)
1090                 return -ESRCH;
1091         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1092                                 audit_get_loginuid(task));
1093         put_task_struct(task);
1094         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1095 }
1096
1097 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1098                                    size_t count, loff_t *ppos)
1099 {
1100         struct inode * inode = file->f_path.dentry->d_inode;
1101         char *page, *tmp;
1102         ssize_t length;
1103         uid_t loginuid;
1104
1105         rcu_read_lock();
1106         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1107                 rcu_read_unlock();
1108                 return -EPERM;
1109         }
1110         rcu_read_unlock();
1111
1112         if (count >= PAGE_SIZE)
1113                 count = PAGE_SIZE - 1;
1114
1115         if (*ppos != 0) {
1116                 /* No partial writes. */
1117                 return -EINVAL;
1118         }
1119         page = (char*)__get_free_page(GFP_TEMPORARY);
1120         if (!page)
1121                 return -ENOMEM;
1122         length = -EFAULT;
1123         if (copy_from_user(page, buf, count))
1124                 goto out_free_page;
1125
1126         page[count] = '\0';
1127         loginuid = simple_strtoul(page, &tmp, 10);
1128         if (tmp == page) {
1129                 length = -EINVAL;
1130                 goto out_free_page;
1131
1132         }
1133         length = audit_set_loginuid(loginuid);
1134         if (likely(length == 0))
1135                 length = count;
1136
1137 out_free_page:
1138         free_page((unsigned long) page);
1139         return length;
1140 }
1141
1142 static const struct file_operations proc_loginuid_operations = {
1143         .read           = proc_loginuid_read,
1144         .write          = proc_loginuid_write,
1145         .llseek         = generic_file_llseek,
1146 };
1147
1148 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1149                                   size_t count, loff_t *ppos)
1150 {
1151         struct inode * inode = file->f_path.dentry->d_inode;
1152         struct task_struct *task = get_proc_task(inode);
1153         ssize_t length;
1154         char tmpbuf[TMPBUFLEN];
1155
1156         if (!task)
1157                 return -ESRCH;
1158         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1159                                 audit_get_sessionid(task));
1160         put_task_struct(task);
1161         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1162 }
1163
1164 static const struct file_operations proc_sessionid_operations = {
1165         .read           = proc_sessionid_read,
1166         .llseek         = generic_file_llseek,
1167 };
1168 #endif
1169
1170 #ifdef CONFIG_FAULT_INJECTION
1171 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1172                                       size_t count, loff_t *ppos)
1173 {
1174         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1175         char buffer[PROC_NUMBUF];
1176         size_t len;
1177         int make_it_fail;
1178
1179         if (!task)
1180                 return -ESRCH;
1181         make_it_fail = task->make_it_fail;
1182         put_task_struct(task);
1183
1184         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1185
1186         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1187 }
1188
1189 static ssize_t proc_fault_inject_write(struct file * file,
1190                         const char __user * buf, size_t count, loff_t *ppos)
1191 {
1192         struct task_struct *task;
1193         char buffer[PROC_NUMBUF], *end;
1194         int make_it_fail;
1195
1196         if (!capable(CAP_SYS_RESOURCE))
1197                 return -EPERM;
1198         memset(buffer, 0, sizeof(buffer));
1199         if (count > sizeof(buffer) - 1)
1200                 count = sizeof(buffer) - 1;
1201         if (copy_from_user(buffer, buf, count))
1202                 return -EFAULT;
1203         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1204         if (*end)
1205                 return -EINVAL;
1206         task = get_proc_task(file->f_dentry->d_inode);
1207         if (!task)
1208                 return -ESRCH;
1209         task->make_it_fail = make_it_fail;
1210         put_task_struct(task);
1211
1212         return count;
1213 }
1214
1215 static const struct file_operations proc_fault_inject_operations = {
1216         .read           = proc_fault_inject_read,
1217         .write          = proc_fault_inject_write,
1218         .llseek         = generic_file_llseek,
1219 };
1220 #endif
1221
1222
1223 #ifdef CONFIG_SCHED_DEBUG
1224 /*
1225  * Print out various scheduling related per-task fields:
1226  */
1227 static int sched_show(struct seq_file *m, void *v)
1228 {
1229         struct inode *inode = m->private;
1230         struct task_struct *p;
1231
1232         p = get_proc_task(inode);
1233         if (!p)
1234                 return -ESRCH;
1235         proc_sched_show_task(p, m);
1236
1237         put_task_struct(p);
1238
1239         return 0;
1240 }
1241
1242 static ssize_t
1243 sched_write(struct file *file, const char __user *buf,
1244             size_t count, loff_t *offset)
1245 {
1246         struct inode *inode = file->f_path.dentry->d_inode;
1247         struct task_struct *p;
1248
1249         p = get_proc_task(inode);
1250         if (!p)
1251                 return -ESRCH;
1252         proc_sched_set_task(p);
1253
1254         put_task_struct(p);
1255
1256         return count;
1257 }
1258
1259 static int sched_open(struct inode *inode, struct file *filp)
1260 {
1261         return single_open(filp, sched_show, inode);
1262 }
1263
1264 static const struct file_operations proc_pid_sched_operations = {
1265         .open           = sched_open,
1266         .read           = seq_read,
1267         .write          = sched_write,
1268         .llseek         = seq_lseek,
1269         .release        = single_release,
1270 };
1271
1272 #endif
1273
1274 #ifdef CONFIG_SCHED_AUTOGROUP
1275 /*
1276  * Print out autogroup related information:
1277  */
1278 static int sched_autogroup_show(struct seq_file *m, void *v)
1279 {
1280         struct inode *inode = m->private;
1281         struct task_struct *p;
1282
1283         p = get_proc_task(inode);
1284         if (!p)
1285                 return -ESRCH;
1286         proc_sched_autogroup_show_task(p, m);
1287
1288         put_task_struct(p);
1289
1290         return 0;
1291 }
1292
1293 static ssize_t
1294 sched_autogroup_write(struct file *file, const char __user *buf,
1295             size_t count, loff_t *offset)
1296 {
1297         struct inode *inode = file->f_path.dentry->d_inode;
1298         struct task_struct *p;
1299         char buffer[PROC_NUMBUF];
1300         int nice;
1301         int err;
1302
1303         memset(buffer, 0, sizeof(buffer));
1304         if (count > sizeof(buffer) - 1)
1305                 count = sizeof(buffer) - 1;
1306         if (copy_from_user(buffer, buf, count))
1307                 return -EFAULT;
1308
1309         err = kstrtoint(strstrip(buffer), 0, &nice);
1310         if (err < 0)
1311                 return err;
1312
1313         p = get_proc_task(inode);
1314         if (!p)
1315                 return -ESRCH;
1316
1317         err = proc_sched_autogroup_set_nice(p, nice);
1318         if (err)
1319                 count = err;
1320
1321         put_task_struct(p);
1322
1323         return count;
1324 }
1325
1326 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1327 {
1328         int ret;
1329
1330         ret = single_open(filp, sched_autogroup_show, NULL);
1331         if (!ret) {
1332                 struct seq_file *m = filp->private_data;
1333
1334                 m->private = inode;
1335         }
1336         return ret;
1337 }
1338
1339 static const struct file_operations proc_pid_sched_autogroup_operations = {
1340         .open           = sched_autogroup_open,
1341         .read           = seq_read,
1342         .write          = sched_autogroup_write,
1343         .llseek         = seq_lseek,
1344         .release        = single_release,
1345 };
1346
1347 #endif /* CONFIG_SCHED_AUTOGROUP */
1348
1349 static ssize_t comm_write(struct file *file, const char __user *buf,
1350                                 size_t count, loff_t *offset)
1351 {
1352         struct inode *inode = file->f_path.dentry->d_inode;
1353         struct task_struct *p;
1354         char buffer[TASK_COMM_LEN];
1355
1356         memset(buffer, 0, sizeof(buffer));
1357         if (count > sizeof(buffer) - 1)
1358                 count = sizeof(buffer) - 1;
1359         if (copy_from_user(buffer, buf, count))
1360                 return -EFAULT;
1361
1362         p = get_proc_task(inode);
1363         if (!p)
1364                 return -ESRCH;
1365
1366         if (same_thread_group(current, p))
1367                 set_task_comm(p, buffer);
1368         else
1369                 count = -EINVAL;
1370
1371         put_task_struct(p);
1372
1373         return count;
1374 }
1375
1376 static int comm_show(struct seq_file *m, void *v)
1377 {
1378         struct inode *inode = m->private;
1379         struct task_struct *p;
1380
1381         p = get_proc_task(inode);
1382         if (!p)
1383                 return -ESRCH;
1384
1385         task_lock(p);
1386         seq_printf(m, "%s\n", p->comm);
1387         task_unlock(p);
1388
1389         put_task_struct(p);
1390
1391         return 0;
1392 }
1393
1394 static int comm_open(struct inode *inode, struct file *filp)
1395 {
1396         return single_open(filp, comm_show, inode);
1397 }
1398
1399 static const struct file_operations proc_pid_set_comm_operations = {
1400         .open           = comm_open,
1401         .read           = seq_read,
1402         .write          = comm_write,
1403         .llseek         = seq_lseek,
1404         .release        = single_release,
1405 };
1406
1407 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1408 {
1409         struct task_struct *task;
1410         struct mm_struct *mm;
1411         struct file *exe_file;
1412
1413         task = get_proc_task(dentry->d_inode);
1414         if (!task)
1415                 return -ENOENT;
1416         mm = get_task_mm(task);
1417         put_task_struct(task);
1418         if (!mm)
1419                 return -ENOENT;
1420         exe_file = get_mm_exe_file(mm);
1421         mmput(mm);
1422         if (exe_file) {
1423                 *exe_path = exe_file->f_path;
1424                 path_get(&exe_file->f_path);
1425                 fput(exe_file);
1426                 return 0;
1427         } else
1428                 return -ENOENT;
1429 }
1430
1431 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1432 {
1433         struct inode *inode = dentry->d_inode;
1434         struct path path;
1435         int error = -EACCES;
1436
1437         /* Are we allowed to snoop on the tasks file descriptors? */
1438         if (!proc_fd_access_allowed(inode))
1439                 goto out;
1440
1441         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1442         if (error)
1443                 goto out;
1444
1445         nd_jump_link(nd, &path);
1446         return NULL;
1447 out:
1448         return ERR_PTR(error);
1449 }
1450
1451 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1452 {
1453         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1454         char *pathname;
1455         int len;
1456
1457         if (!tmp)
1458                 return -ENOMEM;
1459
1460         pathname = d_path(path, tmp, PAGE_SIZE);
1461         len = PTR_ERR(pathname);
1462         if (IS_ERR(pathname))
1463                 goto out;
1464         len = tmp + PAGE_SIZE - 1 - pathname;
1465
1466         if (len > buflen)
1467                 len = buflen;
1468         if (copy_to_user(buffer, pathname, len))
1469                 len = -EFAULT;
1470  out:
1471         free_page((unsigned long)tmp);
1472         return len;
1473 }
1474
1475 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1476 {
1477         int error = -EACCES;
1478         struct inode *inode = dentry->d_inode;
1479         struct path path;
1480
1481         /* Are we allowed to snoop on the tasks file descriptors? */
1482         if (!proc_fd_access_allowed(inode))
1483                 goto out;
1484
1485         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1486         if (error)
1487                 goto out;
1488
1489         error = do_proc_readlink(&path, buffer, buflen);
1490         path_put(&path);
1491 out:
1492         return error;
1493 }
1494
1495 static const struct inode_operations proc_pid_link_inode_operations = {
1496         .readlink       = proc_pid_readlink,
1497         .follow_link    = proc_pid_follow_link,
1498         .setattr        = proc_setattr,
1499 };
1500
1501
1502 /* building an inode */
1503
1504 static int task_dumpable(struct task_struct *task)
1505 {
1506         int dumpable = 0;
1507         struct mm_struct *mm;
1508
1509         task_lock(task);
1510         mm = task->mm;
1511         if (mm)
1512                 dumpable = get_dumpable(mm);
1513         task_unlock(task);
1514         if(dumpable == 1)
1515                 return 1;
1516         return 0;
1517 }
1518
1519 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1520 {
1521         struct inode * inode;
1522         struct proc_inode *ei;
1523         const struct cred *cred;
1524
1525         /* We need a new inode */
1526
1527         inode = new_inode(sb);
1528         if (!inode)
1529                 goto out;
1530
1531         /* Common stuff */
1532         ei = PROC_I(inode);
1533         inode->i_ino = get_next_ino();
1534         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1535         inode->i_op = &proc_def_inode_operations;
1536
1537         /*
1538          * grab the reference to task.
1539          */
1540         ei->pid = get_task_pid(task, PIDTYPE_PID);
1541         if (!ei->pid)
1542                 goto out_unlock;
1543
1544         if (task_dumpable(task)) {
1545                 rcu_read_lock();
1546                 cred = __task_cred(task);
1547                 inode->i_uid = cred->euid;
1548                 inode->i_gid = cred->egid;
1549                 rcu_read_unlock();
1550         }
1551         security_task_to_inode(task, inode);
1552
1553 out:
1554         return inode;
1555
1556 out_unlock:
1557         iput(inode);
1558         return NULL;
1559 }
1560
1561 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1562 {
1563         struct inode *inode = dentry->d_inode;
1564         struct task_struct *task;
1565         const struct cred *cred;
1566         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1567
1568         generic_fillattr(inode, stat);
1569
1570         rcu_read_lock();
1571         stat->uid = GLOBAL_ROOT_UID;
1572         stat->gid = GLOBAL_ROOT_GID;
1573         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1574         if (task) {
1575                 if (!has_pid_permissions(pid, task, 2)) {
1576                         rcu_read_unlock();
1577                         /*
1578                          * This doesn't prevent learning whether PID exists,
1579                          * it only makes getattr() consistent with readdir().
1580                          */
1581                         return -ENOENT;
1582                 }
1583                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1584                     task_dumpable(task)) {
1585                         cred = __task_cred(task);
1586                         stat->uid = cred->euid;
1587                         stat->gid = cred->egid;
1588                 }
1589         }
1590         rcu_read_unlock();
1591         return 0;
1592 }
1593
1594 /* dentry stuff */
1595
1596 /*
1597  *      Exceptional case: normally we are not allowed to unhash a busy
1598  * directory. In this case, however, we can do it - no aliasing problems
1599  * due to the way we treat inodes.
1600  *
1601  * Rewrite the inode's ownerships here because the owning task may have
1602  * performed a setuid(), etc.
1603  *
1604  * Before the /proc/pid/status file was created the only way to read
1605  * the effective uid of a /process was to stat /proc/pid.  Reading
1606  * /proc/pid/status is slow enough that procps and other packages
1607  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1608  * made this apply to all per process world readable and executable
1609  * directories.
1610  */
1611 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1612 {
1613         struct inode *inode;
1614         struct task_struct *task;
1615         const struct cred *cred;
1616
1617         if (flags & LOOKUP_RCU)
1618                 return -ECHILD;
1619
1620         inode = dentry->d_inode;
1621         task = get_proc_task(inode);
1622
1623         if (task) {
1624                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1625                     task_dumpable(task)) {
1626                         rcu_read_lock();
1627                         cred = __task_cred(task);
1628                         inode->i_uid = cred->euid;
1629                         inode->i_gid = cred->egid;
1630                         rcu_read_unlock();
1631                 } else {
1632                         inode->i_uid = GLOBAL_ROOT_UID;
1633                         inode->i_gid = GLOBAL_ROOT_GID;
1634                 }
1635                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1636                 security_task_to_inode(task, inode);
1637                 put_task_struct(task);
1638                 return 1;
1639         }
1640         d_drop(dentry);
1641         return 0;
1642 }
1643
1644 static int pid_delete_dentry(const struct dentry * dentry)
1645 {
1646         /* Is the task we represent dead?
1647          * If so, then don't put the dentry on the lru list,
1648          * kill it immediately.
1649          */
1650         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1651 }
1652
1653 const struct dentry_operations pid_dentry_operations =
1654 {
1655         .d_revalidate   = pid_revalidate,
1656         .d_delete       = pid_delete_dentry,
1657 };
1658
1659 /* Lookups */
1660
1661 /*
1662  * Fill a directory entry.
1663  *
1664  * If possible create the dcache entry and derive our inode number and
1665  * file type from dcache entry.
1666  *
1667  * Since all of the proc inode numbers are dynamically generated, the inode
1668  * numbers do not exist until the inode is cache.  This means creating the
1669  * the dcache entry in readdir is necessary to keep the inode numbers
1670  * reported by readdir in sync with the inode numbers reported
1671  * by stat.
1672  */
1673 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1674         const char *name, int len,
1675         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1676 {
1677         struct dentry *child, *dir = filp->f_path.dentry;
1678         struct inode *inode;
1679         struct qstr qname;
1680         ino_t ino = 0;
1681         unsigned type = DT_UNKNOWN;
1682
1683         qname.name = name;
1684         qname.len  = len;
1685         qname.hash = full_name_hash(name, len);
1686
1687         child = d_lookup(dir, &qname);
1688         if (!child) {
1689                 struct dentry *new;
1690                 new = d_alloc(dir, &qname);
1691                 if (new) {
1692                         child = instantiate(dir->d_inode, new, task, ptr);
1693                         if (child)
1694                                 dput(new);
1695                         else
1696                                 child = new;
1697                 }
1698         }
1699         if (!child || IS_ERR(child) || !child->d_inode)
1700                 goto end_instantiate;
1701         inode = child->d_inode;
1702         if (inode) {
1703                 ino = inode->i_ino;
1704                 type = inode->i_mode >> 12;
1705         }
1706         dput(child);
1707 end_instantiate:
1708         if (!ino)
1709                 ino = find_inode_number(dir, &qname);
1710         if (!ino)
1711                 ino = 1;
1712         return filldir(dirent, name, len, filp->f_pos, ino, type);
1713 }
1714
1715 static unsigned name_to_int(struct dentry *dentry)
1716 {
1717         const char *name = dentry->d_name.name;
1718         int len = dentry->d_name.len;
1719         unsigned n = 0;
1720
1721         if (len > 1 && *name == '0')
1722                 goto out;
1723         while (len-- > 0) {
1724                 unsigned c = *name++ - '0';
1725                 if (c > 9)
1726                         goto out;
1727                 if (n >= (~0U-9)/10)
1728                         goto out;
1729                 n *= 10;
1730                 n += c;
1731         }
1732         return n;
1733 out:
1734         return ~0U;
1735 }
1736
1737 #define PROC_FDINFO_MAX 64
1738
1739 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1740 {
1741         struct task_struct *task = get_proc_task(inode);
1742         struct files_struct *files = NULL;
1743         struct file *file;
1744         int fd = proc_fd(inode);
1745
1746         if (task) {
1747                 files = get_files_struct(task);
1748                 put_task_struct(task);
1749         }
1750         if (files) {
1751                 /*
1752                  * We are not taking a ref to the file structure, so we must
1753                  * hold ->file_lock.
1754                  */
1755                 spin_lock(&files->file_lock);
1756                 file = fcheck_files(files, fd);
1757                 if (file) {
1758                         unsigned int f_flags;
1759                         struct fdtable *fdt;
1760
1761                         fdt = files_fdtable(files);
1762                         f_flags = file->f_flags & ~O_CLOEXEC;
1763                         if (close_on_exec(fd, fdt))
1764                                 f_flags |= O_CLOEXEC;
1765
1766                         if (path) {
1767                                 *path = file->f_path;
1768                                 path_get(&file->f_path);
1769                         }
1770                         if (info)
1771                                 snprintf(info, PROC_FDINFO_MAX,
1772                                          "pos:\t%lli\n"
1773                                          "flags:\t0%o\n",
1774                                          (long long) file->f_pos,
1775                                          f_flags);
1776                         spin_unlock(&files->file_lock);
1777                         put_files_struct(files);
1778                         return 0;
1779                 }
1780                 spin_unlock(&files->file_lock);
1781                 put_files_struct(files);
1782         }
1783         return -ENOENT;
1784 }
1785
1786 static int proc_fd_link(struct dentry *dentry, struct path *path)
1787 {
1788         return proc_fd_info(dentry->d_inode, path, NULL);
1789 }
1790
1791 static int tid_fd_revalidate(struct dentry *dentry, unsigned int flags)
1792 {
1793         struct inode *inode;
1794         struct task_struct *task;
1795         int fd;
1796         struct files_struct *files;
1797         const struct cred *cred;
1798
1799         if (flags & LOOKUP_RCU)
1800                 return -ECHILD;
1801
1802         inode = dentry->d_inode;
1803         task = get_proc_task(inode);
1804         fd = proc_fd(inode);
1805
1806         if (task) {
1807                 files = get_files_struct(task);
1808                 if (files) {
1809                         struct file *file;
1810                         rcu_read_lock();
1811                         file = fcheck_files(files, fd);
1812                         if (file) {
1813                                 unsigned f_mode = file->f_mode;
1814
1815                                 rcu_read_unlock();
1816                                 put_files_struct(files);
1817
1818                                 if (task_dumpable(task)) {
1819                                         rcu_read_lock();
1820                                         cred = __task_cred(task);
1821                                         inode->i_uid = cred->euid;
1822                                         inode->i_gid = cred->egid;
1823                                         rcu_read_unlock();
1824                                 } else {
1825                                         inode->i_uid = GLOBAL_ROOT_UID;
1826                                         inode->i_gid = GLOBAL_ROOT_GID;
1827                                 }
1828
1829                                 if (S_ISLNK(inode->i_mode)) {
1830                                         unsigned i_mode = S_IFLNK;
1831                                         if (f_mode & FMODE_READ)
1832                                                 i_mode |= S_IRUSR | S_IXUSR;
1833                                         if (f_mode & FMODE_WRITE)
1834                                                 i_mode |= S_IWUSR | S_IXUSR;
1835                                         inode->i_mode = i_mode;
1836                                 }
1837
1838                                 security_task_to_inode(task, inode);
1839                                 put_task_struct(task);
1840                                 return 1;
1841                         }
1842                         rcu_read_unlock();
1843                         put_files_struct(files);
1844                 }
1845                 put_task_struct(task);
1846         }
1847         d_drop(dentry);
1848         return 0;
1849 }
1850
1851 static const struct dentry_operations tid_fd_dentry_operations =
1852 {
1853         .d_revalidate   = tid_fd_revalidate,
1854         .d_delete       = pid_delete_dentry,
1855 };
1856
1857 static struct dentry *proc_fd_instantiate(struct inode *dir,
1858         struct dentry *dentry, struct task_struct *task, const void *ptr)
1859 {
1860         unsigned fd = (unsigned long)ptr;
1861         struct inode *inode;
1862         struct proc_inode *ei;
1863         struct dentry *error = ERR_PTR(-ENOENT);
1864
1865         inode = proc_pid_make_inode(dir->i_sb, task);
1866         if (!inode)
1867                 goto out;
1868         ei = PROC_I(inode);
1869         ei->fd = fd;
1870
1871         inode->i_mode = S_IFLNK;
1872         inode->i_op = &proc_pid_link_inode_operations;
1873         inode->i_size = 64;
1874         ei->op.proc_get_link = proc_fd_link;
1875         d_set_d_op(dentry, &tid_fd_dentry_operations);
1876         d_add(dentry, inode);
1877         /* Close the race of the process dying before we return the dentry */
1878         if (tid_fd_revalidate(dentry, 0))
1879                 error = NULL;
1880
1881  out:
1882         return error;
1883 }
1884
1885 static struct dentry *proc_lookupfd_common(struct inode *dir,
1886                                            struct dentry *dentry,
1887                                            instantiate_t instantiate)
1888 {
1889         struct task_struct *task = get_proc_task(dir);
1890         unsigned fd = name_to_int(dentry);
1891         struct dentry *result = ERR_PTR(-ENOENT);
1892
1893         if (!task)
1894                 goto out_no_task;
1895         if (fd == ~0U)
1896                 goto out;
1897
1898         result = instantiate(dir, dentry, task, (void *)(unsigned long)fd);
1899 out:
1900         put_task_struct(task);
1901 out_no_task:
1902         return result;
1903 }
1904
1905 static int proc_readfd_common(struct file * filp, void * dirent,
1906                               filldir_t filldir, instantiate_t instantiate)
1907 {
1908         struct dentry *dentry = filp->f_path.dentry;
1909         struct inode *inode = dentry->d_inode;
1910         struct task_struct *p = get_proc_task(inode);
1911         unsigned int fd, ino;
1912         int retval;
1913         struct files_struct * files;
1914
1915         retval = -ENOENT;
1916         if (!p)
1917                 goto out_no_task;
1918         retval = 0;
1919
1920         fd = filp->f_pos;
1921         switch (fd) {
1922                 case 0:
1923                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1924                                 goto out;
1925                         filp->f_pos++;
1926                 case 1:
1927                         ino = parent_ino(dentry);
1928                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1929                                 goto out;
1930                         filp->f_pos++;
1931                 default:
1932                         files = get_files_struct(p);
1933                         if (!files)
1934                                 goto out;
1935                         rcu_read_lock();
1936                         for (fd = filp->f_pos-2;
1937                              fd < files_fdtable(files)->max_fds;
1938                              fd++, filp->f_pos++) {
1939                                 char name[PROC_NUMBUF];
1940                                 int len;
1941                                 int rv;
1942
1943                                 if (!fcheck_files(files, fd))
1944                                         continue;
1945                                 rcu_read_unlock();
1946
1947                                 len = snprintf(name, sizeof(name), "%d", fd);
1948                                 rv = proc_fill_cache(filp, dirent, filldir,
1949                                                      name, len, instantiate, p,
1950                                                      (void *)(unsigned long)fd);
1951                                 if (rv < 0)
1952                                         goto out_fd_loop;
1953                                 rcu_read_lock();
1954                         }
1955                         rcu_read_unlock();
1956 out_fd_loop:
1957                         put_files_struct(files);
1958         }
1959 out:
1960         put_task_struct(p);
1961 out_no_task:
1962         return retval;
1963 }
1964
1965 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1966                                     unsigned int flags)
1967 {
1968         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1969 }
1970
1971 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1972 {
1973         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1974 }
1975
1976 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1977                                       size_t len, loff_t *ppos)
1978 {
1979         char tmp[PROC_FDINFO_MAX];
1980         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1981         if (!err)
1982                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1983         return err;
1984 }
1985
1986 static const struct file_operations proc_fdinfo_file_operations = {
1987         .open           = nonseekable_open,
1988         .read           = proc_fdinfo_read,
1989         .llseek         = no_llseek,
1990 };
1991
1992 static const struct file_operations proc_fd_operations = {
1993         .read           = generic_read_dir,
1994         .readdir        = proc_readfd,
1995         .llseek         = default_llseek,
1996 };
1997
1998 #ifdef CONFIG_CHECKPOINT_RESTORE
1999
2000 /*
2001  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2002  * which represent vma start and end addresses.
2003  */
2004 static int dname_to_vma_addr(struct dentry *dentry,
2005                              unsigned long *start, unsigned long *end)
2006 {
2007         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2008                 return -EINVAL;
2009
2010         return 0;
2011 }
2012
2013 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2014 {
2015         unsigned long vm_start, vm_end;
2016         bool exact_vma_exists = false;
2017         struct mm_struct *mm = NULL;
2018         struct task_struct *task;
2019         const struct cred *cred;
2020         struct inode *inode;
2021         int status = 0;
2022
2023         if (flags & LOOKUP_RCU)
2024                 return -ECHILD;
2025
2026         if (!capable(CAP_SYS_ADMIN)) {
2027                 status = -EACCES;
2028                 goto out_notask;
2029         }
2030
2031         inode = dentry->d_inode;
2032         task = get_proc_task(inode);
2033         if (!task)
2034                 goto out_notask;
2035
2036         mm = mm_access(task, PTRACE_MODE_READ);
2037         if (IS_ERR_OR_NULL(mm))
2038                 goto out;
2039
2040         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2041                 down_read(&mm->mmap_sem);
2042                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2043                 up_read(&mm->mmap_sem);
2044         }
2045
2046         mmput(mm);
2047
2048         if (exact_vma_exists) {
2049                 if (task_dumpable(task)) {
2050                         rcu_read_lock();
2051                         cred = __task_cred(task);
2052                         inode->i_uid = cred->euid;
2053                         inode->i_gid = cred->egid;
2054                         rcu_read_unlock();
2055                 } else {
2056                         inode->i_uid = GLOBAL_ROOT_UID;
2057                         inode->i_gid = GLOBAL_ROOT_GID;
2058                 }
2059                 security_task_to_inode(task, inode);
2060                 status = 1;
2061         }
2062
2063 out:
2064         put_task_struct(task);
2065
2066 out_notask:
2067         if (status <= 0)
2068                 d_drop(dentry);
2069
2070         return status;
2071 }
2072
2073 static const struct dentry_operations tid_map_files_dentry_operations = {
2074         .d_revalidate   = map_files_d_revalidate,
2075         .d_delete       = pid_delete_dentry,
2076 };
2077
2078 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2079 {
2080         unsigned long vm_start, vm_end;
2081         struct vm_area_struct *vma;
2082         struct task_struct *task;
2083         struct mm_struct *mm;
2084         int rc;
2085
2086         rc = -ENOENT;
2087         task = get_proc_task(dentry->d_inode);
2088         if (!task)
2089                 goto out;
2090
2091         mm = get_task_mm(task);
2092         put_task_struct(task);
2093         if (!mm)
2094                 goto out;
2095
2096         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2097         if (rc)
2098                 goto out_mmput;
2099
2100         down_read(&mm->mmap_sem);
2101         vma = find_exact_vma(mm, vm_start, vm_end);
2102         if (vma && vma->vm_file) {
2103                 *path = vma->vm_file->f_path;
2104                 path_get(path);
2105                 rc = 0;
2106         }
2107         up_read(&mm->mmap_sem);
2108
2109 out_mmput:
2110         mmput(mm);
2111 out:
2112         return rc;
2113 }
2114
2115 struct map_files_info {
2116         struct file     *file;
2117         unsigned long   len;
2118         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2119 };
2120
2121 static struct dentry *
2122 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2123                            struct task_struct *task, const void *ptr)
2124 {
2125         const struct file *file = ptr;
2126         struct proc_inode *ei;
2127         struct inode *inode;
2128
2129         if (!file)
2130                 return ERR_PTR(-ENOENT);
2131
2132         inode = proc_pid_make_inode(dir->i_sb, task);
2133         if (!inode)
2134                 return ERR_PTR(-ENOENT);
2135
2136         ei = PROC_I(inode);
2137         ei->op.proc_get_link = proc_map_files_get_link;
2138
2139         inode->i_op = &proc_pid_link_inode_operations;
2140         inode->i_size = 64;
2141         inode->i_mode = S_IFLNK;
2142
2143         if (file->f_mode & FMODE_READ)
2144                 inode->i_mode |= S_IRUSR;
2145         if (file->f_mode & FMODE_WRITE)
2146                 inode->i_mode |= S_IWUSR;
2147
2148         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2149         d_add(dentry, inode);
2150
2151         return NULL;
2152 }
2153
2154 static struct dentry *proc_map_files_lookup(struct inode *dir,
2155                 struct dentry *dentry, unsigned int flags)
2156 {
2157         unsigned long vm_start, vm_end;
2158         struct vm_area_struct *vma;
2159         struct task_struct *task;
2160         struct dentry *result;
2161         struct mm_struct *mm;
2162
2163         result = ERR_PTR(-EACCES);
2164         if (!capable(CAP_SYS_ADMIN))
2165                 goto out;
2166
2167         result = ERR_PTR(-ENOENT);
2168         task = get_proc_task(dir);
2169         if (!task)
2170                 goto out;
2171
2172         result = ERR_PTR(-EACCES);
2173         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2174                 goto out_put_task;
2175
2176         result = ERR_PTR(-ENOENT);
2177         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2178                 goto out_put_task;
2179
2180         mm = get_task_mm(task);
2181         if (!mm)
2182                 goto out_put_task;
2183
2184         down_read(&mm->mmap_sem);
2185         vma = find_exact_vma(mm, vm_start, vm_end);
2186         if (!vma)
2187                 goto out_no_vma;
2188
2189         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2190
2191 out_no_vma:
2192         up_read(&mm->mmap_sem);
2193         mmput(mm);
2194 out_put_task:
2195         put_task_struct(task);
2196 out:
2197         return result;
2198 }
2199
2200 static const struct inode_operations proc_map_files_inode_operations = {
2201         .lookup         = proc_map_files_lookup,
2202         .permission     = proc_fd_permission,
2203         .setattr        = proc_setattr,
2204 };
2205
2206 static int
2207 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2208 {
2209         struct dentry *dentry = filp->f_path.dentry;
2210         struct inode *inode = dentry->d_inode;
2211         struct vm_area_struct *vma;
2212         struct task_struct *task;
2213         struct mm_struct *mm;
2214         ino_t ino;
2215         int ret;
2216
2217         ret = -EACCES;
2218         if (!capable(CAP_SYS_ADMIN))
2219                 goto out;
2220
2221         ret = -ENOENT;
2222         task = get_proc_task(inode);
2223         if (!task)
2224                 goto out;
2225
2226         ret = -EACCES;
2227         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2228                 goto out_put_task;
2229
2230         ret = 0;
2231         switch (filp->f_pos) {
2232         case 0:
2233                 ino = inode->i_ino;
2234                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2235                         goto out_put_task;
2236                 filp->f_pos++;
2237         case 1:
2238                 ino = parent_ino(dentry);
2239                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2240                         goto out_put_task;
2241                 filp->f_pos++;
2242         default:
2243         {
2244                 unsigned long nr_files, pos, i;
2245                 struct flex_array *fa = NULL;
2246                 struct map_files_info info;
2247                 struct map_files_info *p;
2248
2249                 mm = get_task_mm(task);
2250                 if (!mm)
2251                         goto out_put_task;
2252                 down_read(&mm->mmap_sem);
2253
2254                 nr_files = 0;
2255
2256                 /*
2257                  * We need two passes here:
2258                  *
2259                  *  1) Collect vmas of mapped files with mmap_sem taken
2260                  *  2) Release mmap_sem and instantiate entries
2261                  *
2262                  * otherwise we get lockdep complained, since filldir()
2263                  * routine might require mmap_sem taken in might_fault().
2264                  */
2265
2266                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2267                         if (vma->vm_file && ++pos > filp->f_pos)
2268                                 nr_files++;
2269                 }
2270
2271                 if (nr_files) {
2272                         fa = flex_array_alloc(sizeof(info), nr_files,
2273                                                 GFP_KERNEL);
2274                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2275                                                         GFP_KERNEL)) {
2276                                 ret = -ENOMEM;
2277                                 if (fa)
2278                                         flex_array_free(fa);
2279                                 up_read(&mm->mmap_sem);
2280                                 mmput(mm);
2281                                 goto out_put_task;
2282                         }
2283                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2284                                         vma = vma->vm_next) {
2285                                 if (!vma->vm_file)
2286                                         continue;
2287                                 if (++pos <= filp->f_pos)
2288                                         continue;
2289
2290                                 get_file(vma->vm_file);
2291                                 info.file = vma->vm_file;
2292                                 info.len = snprintf(info.name,
2293                                                 sizeof(info.name), "%lx-%lx",
2294                                                 vma->vm_start, vma->vm_end);
2295                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2296                                         BUG();
2297                         }
2298                 }
2299                 up_read(&mm->mmap_sem);
2300
2301                 for (i = 0; i < nr_files; i++) {
2302                         p = flex_array_get(fa, i);
2303                         ret = proc_fill_cache(filp, dirent, filldir,
2304                                               p->name, p->len,
2305                                               proc_map_files_instantiate,
2306                                               task, p->file);
2307                         if (ret)
2308                                 break;
2309                         filp->f_pos++;
2310                         fput(p->file);
2311                 }
2312                 for (; i < nr_files; i++) {
2313                         /*
2314                          * In case of error don't forget
2315                          * to put rest of file refs.
2316                          */
2317                         p = flex_array_get(fa, i);
2318                         fput(p->file);
2319                 }
2320                 if (fa)
2321                         flex_array_free(fa);
2322                 mmput(mm);
2323         }
2324         }
2325
2326 out_put_task:
2327         put_task_struct(task);
2328 out:
2329         return ret;
2330 }
2331
2332 static const struct file_operations proc_map_files_operations = {
2333         .read           = generic_read_dir,
2334         .readdir        = proc_map_files_readdir,
2335         .llseek         = default_llseek,
2336 };
2337
2338 #endif /* CONFIG_CHECKPOINT_RESTORE */
2339
2340 /*
2341  * /proc/pid/fd needs a special permission handler so that a process can still
2342  * access /proc/self/fd after it has executed a setuid().
2343  */
2344 static int proc_fd_permission(struct inode *inode, int mask)
2345 {
2346         int rv = generic_permission(inode, mask);
2347         if (rv == 0)
2348                 return 0;
2349         if (task_pid(current) == proc_pid(inode))
2350                 rv = 0;
2351         return rv;
2352 }
2353
2354 /*
2355  * proc directories can do almost nothing..
2356  */
2357 static const struct inode_operations proc_fd_inode_operations = {
2358         .lookup         = proc_lookupfd,
2359         .permission     = proc_fd_permission,
2360         .setattr        = proc_setattr,
2361 };
2362
2363 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2364         struct dentry *dentry, struct task_struct *task, const void *ptr)
2365 {
2366         unsigned fd = (unsigned long)ptr;
2367         struct inode *inode;
2368         struct proc_inode *ei;
2369         struct dentry *error = ERR_PTR(-ENOENT);
2370
2371         inode = proc_pid_make_inode(dir->i_sb, task);
2372         if (!inode)
2373                 goto out;
2374         ei = PROC_I(inode);
2375         ei->fd = fd;
2376         inode->i_mode = S_IFREG | S_IRUSR;
2377         inode->i_fop = &proc_fdinfo_file_operations;
2378         d_set_d_op(dentry, &tid_fd_dentry_operations);
2379         d_add(dentry, inode);
2380         /* Close the race of the process dying before we return the dentry */
2381         if (tid_fd_revalidate(dentry, 0))
2382                 error = NULL;
2383
2384  out:
2385         return error;
2386 }
2387
2388 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2389                                         struct dentry *dentry,
2390                                         unsigned int flags)
2391 {
2392         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2393 }
2394
2395 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2396 {
2397         return proc_readfd_common(filp, dirent, filldir,
2398                                   proc_fdinfo_instantiate);
2399 }
2400
2401 static const struct file_operations proc_fdinfo_operations = {
2402         .read           = generic_read_dir,
2403         .readdir        = proc_readfdinfo,
2404         .llseek         = default_llseek,
2405 };
2406
2407 /*
2408  * proc directories can do almost nothing..
2409  */
2410 static const struct inode_operations proc_fdinfo_inode_operations = {
2411         .lookup         = proc_lookupfdinfo,
2412         .setattr        = proc_setattr,
2413 };
2414
2415
2416 static struct dentry *proc_pident_instantiate(struct inode *dir,
2417         struct dentry *dentry, struct task_struct *task, const void *ptr)
2418 {
2419         const struct pid_entry *p = ptr;
2420         struct inode *inode;
2421         struct proc_inode *ei;
2422         struct dentry *error = ERR_PTR(-ENOENT);
2423
2424         inode = proc_pid_make_inode(dir->i_sb, task);
2425         if (!inode)
2426                 goto out;
2427
2428         ei = PROC_I(inode);
2429         inode->i_mode = p->mode;
2430         if (S_ISDIR(inode->i_mode))
2431                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2432         if (p->iop)
2433                 inode->i_op = p->iop;
2434         if (p->fop)
2435                 inode->i_fop = p->fop;
2436         ei->op = p->op;
2437         d_set_d_op(dentry, &pid_dentry_operations);
2438         d_add(dentry, inode);
2439         /* Close the race of the process dying before we return the dentry */
2440         if (pid_revalidate(dentry, 0))
2441                 error = NULL;
2442 out:
2443         return error;
2444 }
2445
2446 static struct dentry *proc_pident_lookup(struct inode *dir,
2447                                          struct dentry *dentry,
2448                                          const struct pid_entry *ents,
2449                                          unsigned int nents)
2450 {
2451         struct dentry *error;
2452         struct task_struct *task = get_proc_task(dir);
2453         const struct pid_entry *p, *last;
2454
2455         error = ERR_PTR(-ENOENT);
2456
2457         if (!task)
2458                 goto out_no_task;
2459
2460         /*
2461          * Yes, it does not scale. And it should not. Don't add
2462          * new entries into /proc/<tgid>/ without very good reasons.
2463          */
2464         last = &ents[nents - 1];
2465         for (p = ents; p <= last; p++) {
2466                 if (p->len != dentry->d_name.len)
2467                         continue;
2468                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2469                         break;
2470         }
2471         if (p > last)
2472                 goto out;
2473
2474         error = proc_pident_instantiate(dir, dentry, task, p);
2475 out:
2476         put_task_struct(task);
2477 out_no_task:
2478         return error;
2479 }
2480
2481 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2482         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2483 {
2484         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2485                                 proc_pident_instantiate, task, p);
2486 }
2487
2488 static int proc_pident_readdir(struct file *filp,
2489                 void *dirent, filldir_t filldir,
2490                 const struct pid_entry *ents, unsigned int nents)
2491 {
2492         int i;
2493         struct dentry *dentry = filp->f_path.dentry;
2494         struct inode *inode = dentry->d_inode;
2495         struct task_struct *task = get_proc_task(inode);
2496         const struct pid_entry *p, *last;
2497         ino_t ino;
2498         int ret;
2499
2500         ret = -ENOENT;
2501         if (!task)
2502                 goto out_no_task;
2503
2504         ret = 0;
2505         i = filp->f_pos;
2506         switch (i) {
2507         case 0:
2508                 ino = inode->i_ino;
2509                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2510                         goto out;
2511                 i++;
2512                 filp->f_pos++;
2513                 /* fall through */
2514         case 1:
2515                 ino = parent_ino(dentry);
2516                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2517                         goto out;
2518                 i++;
2519                 filp->f_pos++;
2520                 /* fall through */
2521         default:
2522                 i -= 2;
2523                 if (i >= nents) {
2524                         ret = 1;
2525                         goto out;
2526                 }
2527                 p = ents + i;
2528                 last = &ents[nents - 1];
2529                 while (p <= last) {
2530                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2531                                 goto out;
2532                         filp->f_pos++;
2533                         p++;
2534                 }
2535         }
2536
2537         ret = 1;
2538 out:
2539         put_task_struct(task);
2540 out_no_task:
2541         return ret;
2542 }
2543
2544 #ifdef CONFIG_SECURITY
2545 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2546                                   size_t count, loff_t *ppos)
2547 {
2548         struct inode * inode = file->f_path.dentry->d_inode;
2549         char *p = NULL;
2550         ssize_t length;
2551         struct task_struct *task = get_proc_task(inode);
2552
2553         if (!task)
2554                 return -ESRCH;
2555
2556         length = security_getprocattr(task,
2557                                       (char*)file->f_path.dentry->d_name.name,
2558                                       &p);
2559         put_task_struct(task);
2560         if (length > 0)
2561                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2562         kfree(p);
2563         return length;
2564 }
2565
2566 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2567                                    size_t count, loff_t *ppos)
2568 {
2569         struct inode * inode = file->f_path.dentry->d_inode;
2570         char *page;
2571         ssize_t length;
2572         struct task_struct *task = get_proc_task(inode);
2573
2574         length = -ESRCH;
2575         if (!task)
2576                 goto out_no_task;
2577         if (count > PAGE_SIZE)
2578                 count = PAGE_SIZE;
2579
2580         /* No partial writes. */
2581         length = -EINVAL;
2582         if (*ppos != 0)
2583                 goto out;
2584
2585         length = -ENOMEM;
2586         page = (char*)__get_free_page(GFP_TEMPORARY);
2587         if (!page)
2588                 goto out;
2589
2590         length = -EFAULT;
2591         if (copy_from_user(page, buf, count))
2592                 goto out_free;
2593
2594         /* Guard against adverse ptrace interaction */
2595         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2596         if (length < 0)
2597                 goto out_free;
2598
2599         length = security_setprocattr(task,
2600                                       (char*)file->f_path.dentry->d_name.name,
2601                                       (void*)page, count);
2602         mutex_unlock(&task->signal->cred_guard_mutex);
2603 out_free:
2604         free_page((unsigned long) page);
2605 out:
2606         put_task_struct(task);
2607 out_no_task:
2608         return length;
2609 }
2610
2611 static const struct file_operations proc_pid_attr_operations = {
2612         .read           = proc_pid_attr_read,
2613         .write          = proc_pid_attr_write,
2614         .llseek         = generic_file_llseek,
2615 };
2616
2617 static const struct pid_entry attr_dir_stuff[] = {
2618         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2619         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2620         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2621         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2622         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2623         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2624 };
2625
2626 static int proc_attr_dir_readdir(struct file * filp,
2627                              void * dirent, filldir_t filldir)
2628 {
2629         return proc_pident_readdir(filp,dirent,filldir,
2630                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2631 }
2632
2633 static const struct file_operations proc_attr_dir_operations = {
2634         .read           = generic_read_dir,
2635         .readdir        = proc_attr_dir_readdir,
2636         .llseek         = default_llseek,
2637 };
2638
2639 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2640                                 struct dentry *dentry, unsigned int flags)
2641 {
2642         return proc_pident_lookup(dir, dentry,
2643                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2644 }
2645
2646 static const struct inode_operations proc_attr_dir_inode_operations = {
2647         .lookup         = proc_attr_dir_lookup,
2648         .getattr        = pid_getattr,
2649         .setattr        = proc_setattr,
2650 };
2651
2652 #endif
2653
2654 #ifdef CONFIG_ELF_CORE
2655 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2656                                          size_t count, loff_t *ppos)
2657 {
2658         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2659         struct mm_struct *mm;
2660         char buffer[PROC_NUMBUF];
2661         size_t len;
2662         int ret;
2663
2664         if (!task)
2665                 return -ESRCH;
2666
2667         ret = 0;
2668         mm = get_task_mm(task);
2669         if (mm) {
2670                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2671                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2672                                 MMF_DUMP_FILTER_SHIFT));
2673                 mmput(mm);
2674                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2675         }
2676
2677         put_task_struct(task);
2678
2679         return ret;
2680 }
2681
2682 static ssize_t proc_coredump_filter_write(struct file *file,
2683                                           const char __user *buf,
2684                                           size_t count,
2685                                           loff_t *ppos)
2686 {
2687         struct task_struct *task;
2688         struct mm_struct *mm;
2689         char buffer[PROC_NUMBUF], *end;
2690         unsigned int val;
2691         int ret;
2692         int i;
2693         unsigned long mask;
2694
2695         ret = -EFAULT;
2696         memset(buffer, 0, sizeof(buffer));
2697         if (count > sizeof(buffer) - 1)
2698                 count = sizeof(buffer) - 1;
2699         if (copy_from_user(buffer, buf, count))
2700                 goto out_no_task;
2701
2702         ret = -EINVAL;
2703         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2704         if (*end == '\n')
2705                 end++;
2706         if (end - buffer == 0)
2707                 goto out_no_task;
2708
2709         ret = -ESRCH;
2710         task = get_proc_task(file->f_dentry->d_inode);
2711         if (!task)
2712                 goto out_no_task;
2713
2714         ret = end - buffer;
2715         mm = get_task_mm(task);
2716         if (!mm)
2717                 goto out_no_mm;
2718
2719         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2720                 if (val & mask)
2721                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2722                 else
2723                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2724         }
2725
2726         mmput(mm);
2727  out_no_mm:
2728         put_task_struct(task);
2729  out_no_task:
2730         return ret;
2731 }
2732
2733 static const struct file_operations proc_coredump_filter_operations = {
2734         .read           = proc_coredump_filter_read,
2735         .write          = proc_coredump_filter_write,
2736         .llseek         = generic_file_llseek,
2737 };
2738 #endif
2739
2740 /*
2741  * /proc/self:
2742  */
2743 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2744                               int buflen)
2745 {
2746         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2747         pid_t tgid = task_tgid_nr_ns(current, ns);
2748         char tmp[PROC_NUMBUF];
2749         if (!tgid)
2750                 return -ENOENT;
2751         sprintf(tmp, "%d", tgid);
2752         return vfs_readlink(dentry,buffer,buflen,tmp);
2753 }
2754
2755 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2756 {
2757         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2758         pid_t tgid = task_tgid_nr_ns(current, ns);
2759         char *name = ERR_PTR(-ENOENT);
2760         if (tgid) {
2761                 name = __getname();
2762                 if (!name)
2763                         name = ERR_PTR(-ENOMEM);
2764                 else
2765                         sprintf(name, "%d", tgid);
2766         }
2767         nd_set_link(nd, name);
2768         return NULL;
2769 }
2770
2771 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2772                                 void *cookie)
2773 {
2774         char *s = nd_get_link(nd);
2775         if (!IS_ERR(s))
2776                 __putname(s);
2777 }
2778
2779 static const struct inode_operations proc_self_inode_operations = {
2780         .readlink       = proc_self_readlink,
2781         .follow_link    = proc_self_follow_link,
2782         .put_link       = proc_self_put_link,
2783 };
2784
2785 /*
2786  * proc base
2787  *
2788  * These are the directory entries in the root directory of /proc
2789  * that properly belong to the /proc filesystem, as they describe
2790  * describe something that is process related.
2791  */
2792 static const struct pid_entry proc_base_stuff[] = {
2793         NOD("self", S_IFLNK|S_IRWXUGO,
2794                 &proc_self_inode_operations, NULL, {}),
2795 };
2796
2797 static struct dentry *proc_base_instantiate(struct inode *dir,
2798         struct dentry *dentry, struct task_struct *task, const void *ptr)
2799 {
2800         const struct pid_entry *p = ptr;
2801         struct inode *inode;
2802         struct proc_inode *ei;
2803         struct dentry *error;
2804
2805         /* Allocate the inode */
2806         error = ERR_PTR(-ENOMEM);
2807         inode = new_inode(dir->i_sb);
2808         if (!inode)
2809                 goto out;
2810
2811         /* Initialize the inode */
2812         ei = PROC_I(inode);
2813         inode->i_ino = get_next_ino();
2814         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2815
2816         /*
2817          * grab the reference to the task.
2818          */
2819         ei->pid = get_task_pid(task, PIDTYPE_PID);
2820         if (!ei->pid)
2821                 goto out_iput;
2822
2823         inode->i_mode = p->mode;
2824         if (S_ISDIR(inode->i_mode))
2825                 set_nlink(inode, 2);
2826         if (S_ISLNK(inode->i_mode))
2827                 inode->i_size = 64;
2828         if (p->iop)
2829                 inode->i_op = p->iop;
2830         if (p->fop)
2831                 inode->i_fop = p->fop;
2832         ei->op = p->op;
2833         d_add(dentry, inode);
2834         error = NULL;
2835 out:
2836         return error;
2837 out_iput:
2838         iput(inode);
2839         goto out;
2840 }
2841
2842 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2843 {
2844         struct dentry *error;
2845         struct task_struct *task = get_proc_task(dir);
2846         const struct pid_entry *p, *last;
2847
2848         error = ERR_PTR(-ENOENT);
2849
2850         if (!task)
2851                 goto out_no_task;
2852
2853         /* Lookup the directory entry */
2854         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2855         for (p = proc_base_stuff; p <= last; p++) {
2856                 if (p->len != dentry->d_name.len)
2857                         continue;
2858                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2859                         break;
2860         }
2861         if (p > last)
2862                 goto out;
2863
2864         error = proc_base_instantiate(dir, dentry, task, p);
2865
2866 out:
2867         put_task_struct(task);
2868 out_no_task:
2869         return error;
2870 }
2871
2872 static int proc_base_fill_cache(struct file *filp, void *dirent,
2873         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2874 {
2875         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2876                                 proc_base_instantiate, task, p);
2877 }
2878
2879 #ifdef CONFIG_TASK_IO_ACCOUNTING
2880 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2881 {
2882         struct task_io_accounting acct = task->ioac;
2883         unsigned long flags;
2884         int result;
2885
2886         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2887         if (result)
2888                 return result;
2889
2890         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2891                 result = -EACCES;
2892                 goto out_unlock;
2893         }
2894
2895         if (whole && lock_task_sighand(task, &flags)) {
2896                 struct task_struct *t = task;
2897
2898                 task_io_accounting_add(&acct, &task->signal->ioac);
2899                 while_each_thread(task, t)
2900                         task_io_accounting_add(&acct, &t->ioac);
2901
2902                 unlock_task_sighand(task, &flags);
2903         }
2904         result = sprintf(buffer,
2905                         "rchar: %llu\n"
2906                         "wchar: %llu\n"
2907                         "syscr: %llu\n"
2908                         "syscw: %llu\n"
2909                         "read_bytes: %llu\n"
2910                         "write_bytes: %llu\n"
2911                         "cancelled_write_bytes: %llu\n",
2912                         (unsigned long long)acct.rchar,
2913                         (unsigned long long)acct.wchar,
2914                         (unsigned long long)acct.syscr,
2915                         (unsigned long long)acct.syscw,
2916                         (unsigned long long)acct.read_bytes,
2917                         (unsigned long long)acct.write_bytes,
2918                         (unsigned long long)acct.cancelled_write_bytes);
2919 out_unlock:
2920         mutex_unlock(&task->signal->cred_guard_mutex);
2921         return result;
2922 }
2923
2924 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2925 {
2926         return do_io_accounting(task, buffer, 0);
2927 }
2928
2929 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2930 {
2931         return do_io_accounting(task, buffer, 1);
2932 }
2933 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2934
2935 #ifdef CONFIG_USER_NS
2936 static int proc_id_map_open(struct inode *inode, struct file *file,
2937         struct seq_operations *seq_ops)
2938 {
2939         struct user_namespace *ns = NULL;
2940         struct task_struct *task;
2941         struct seq_file *seq;
2942         int ret = -EINVAL;
2943
2944         task = get_proc_task(inode);
2945         if (task) {
2946                 rcu_read_lock();
2947                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2948                 rcu_read_unlock();
2949                 put_task_struct(task);
2950         }
2951         if (!ns)
2952                 goto err;
2953
2954         ret = seq_open(file, seq_ops);
2955         if (ret)
2956                 goto err_put_ns;
2957
2958         seq = file->private_data;
2959         seq->private = ns;
2960
2961         return 0;
2962 err_put_ns:
2963         put_user_ns(ns);
2964 err:
2965         return ret;
2966 }
2967
2968 static int proc_id_map_release(struct inode *inode, struct file *file)
2969 {
2970         struct seq_file *seq = file->private_data;
2971         struct user_namespace *ns = seq->private;
2972         put_user_ns(ns);
2973         return seq_release(inode, file);
2974 }
2975
2976 static int proc_uid_map_open(struct inode *inode, struct file *file)
2977 {
2978         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2979 }
2980
2981 static int proc_gid_map_open(struct inode *inode, struct file *file)
2982 {
2983         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2984 }
2985
2986 static const struct file_operations proc_uid_map_operations = {
2987         .open           = proc_uid_map_open,
2988         .write          = proc_uid_map_write,
2989         .read           = seq_read,
2990         .llseek         = seq_lseek,
2991         .release        = proc_id_map_release,
2992 };
2993
2994 static const struct file_operations proc_gid_map_operations = {
2995         .open           = proc_gid_map_open,
2996         .write          = proc_gid_map_write,
2997         .read           = seq_read,
2998         .llseek         = seq_lseek,
2999         .release        = proc_id_map_release,
3000 };
3001 #endif /* CONFIG_USER_NS */
3002
3003 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3004                                 struct pid *pid, struct task_struct *task)
3005 {
3006         int err = lock_trace(task);
3007         if (!err) {
3008                 seq_printf(m, "%08x\n", task->personality);
3009                 unlock_trace(task);
3010         }
3011         return err;
3012 }
3013
3014 /*
3015  * Thread groups
3016  */
3017 static const struct file_operations proc_task_operations;
3018 static const struct inode_operations proc_task_inode_operations;
3019
3020 static const struct pid_entry tgid_base_stuff[] = {
3021         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3022         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3023 #ifdef CONFIG_CHECKPOINT_RESTORE
3024         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3025 #endif
3026         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3027         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3028 #ifdef CONFIG_NET
3029         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3030 #endif
3031         REG("environ",    S_IRUSR, proc_environ_operations),
3032         INF("auxv",       S_IRUSR, proc_pid_auxv),
3033         ONE("status",     S_IRUGO, proc_pid_status),
3034         ONE("personality", S_IRUGO, proc_pid_personality),
3035         INF("limits",     S_IRUGO, proc_pid_limits),
3036 #ifdef CONFIG_SCHED_DEBUG
3037         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3038 #endif
3039 #ifdef CONFIG_SCHED_AUTOGROUP
3040         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3041 #endif
3042         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3043 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3044         INF("syscall",    S_IRUGO, proc_pid_syscall),
3045 #endif
3046         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3047         ONE("stat",       S_IRUGO, proc_tgid_stat),
3048         ONE("statm",      S_IRUGO, proc_pid_statm),
3049         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3050 #ifdef CONFIG_NUMA
3051         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3052 #endif
3053         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3054         LNK("cwd",        proc_cwd_link),
3055         LNK("root",       proc_root_link),
3056         LNK("exe",        proc_exe_link),
3057         REG("mounts",     S_IRUGO, proc_mounts_operations),
3058         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3059         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3060 #ifdef CONFIG_PROC_PAGE_MONITOR
3061         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3062         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3063         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3064 #endif
3065 #ifdef CONFIG_SECURITY
3066         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3067 #endif
3068 #ifdef CONFIG_KALLSYMS
3069         INF("wchan",      S_IRUGO, proc_pid_wchan),
3070 #endif
3071 #ifdef CONFIG_STACKTRACE
3072         ONE("stack",      S_IRUGO, proc_pid_stack),
3073 #endif
3074 #ifdef CONFIG_SCHEDSTATS
3075         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3076 #endif
3077 #ifdef CONFIG_LATENCYTOP
3078         REG("latency",  S_IRUGO, proc_lstats_operations),
3079 #endif
3080 #ifdef CONFIG_PROC_PID_CPUSET
3081         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3082 #endif
3083 #ifdef CONFIG_CGROUPS
3084         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3085 #endif
3086         INF("oom_score",  S_IRUGO, proc_oom_score),
3087         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3088         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3089 #ifdef CONFIG_AUDITSYSCALL
3090         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3091         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3092 #endif
3093 #ifdef CONFIG_FAULT_INJECTION
3094         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3095 #endif
3096 #ifdef CONFIG_ELF_CORE
3097         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3098 #endif
3099 #ifdef CONFIG_TASK_IO_ACCOUNTING
3100         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3101 #endif
3102 #ifdef CONFIG_HARDWALL
3103         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3104 #endif
3105 #ifdef CONFIG_USER_NS
3106         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3107         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3108 #endif
3109 };
3110
3111 static int proc_tgid_base_readdir(struct file * filp,
3112                              void * dirent, filldir_t filldir)
3113 {
3114         return proc_pident_readdir(filp,dirent,filldir,
3115                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3116 }
3117
3118 static const struct file_operations proc_tgid_base_operations = {
3119         .read           = generic_read_dir,
3120         .readdir        = proc_tgid_base_readdir,
3121         .llseek         = default_llseek,
3122 };
3123
3124 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3125 {
3126         return proc_pident_lookup(dir, dentry,
3127                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3128 }
3129
3130 static const struct inode_operations proc_tgid_base_inode_operations = {
3131         .lookup         = proc_tgid_base_lookup,
3132         .getattr        = pid_getattr,
3133         .setattr        = proc_setattr,
3134         .permission     = proc_pid_permission,
3135 };
3136
3137 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3138 {
3139         struct dentry *dentry, *leader, *dir;
3140         char buf[PROC_NUMBUF];
3141         struct qstr name;
3142
3143         name.name = buf;
3144         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3145         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3146         if (dentry) {
3147                 shrink_dcache_parent(dentry);
3148                 d_drop(dentry);
3149                 dput(dentry);
3150         }
3151
3152         name.name = buf;
3153         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3154         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3155         if (!leader)
3156                 goto out;
3157
3158         name.name = "task";
3159         name.len = strlen(name.name);
3160         dir = d_hash_and_lookup(leader, &name);
3161         if (!dir)
3162                 goto out_put_leader;
3163
3164         name.name = buf;
3165         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3166         dentry = d_hash_and_lookup(dir, &name);
3167         if (dentry) {
3168                 shrink_dcache_parent(dentry);
3169                 d_drop(dentry);
3170                 dput(dentry);
3171         }
3172
3173         dput(dir);
3174 out_put_leader:
3175         dput(leader);
3176 out:
3177         return;
3178 }
3179
3180 /**
3181  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3182  * @task: task that should be flushed.
3183  *
3184  * When flushing dentries from proc, one needs to flush them from global
3185  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3186  * in. This call is supposed to do all of this job.
3187  *
3188  * Looks in the dcache for
3189  * /proc/@pid
3190  * /proc/@tgid/task/@pid
3191  * if either directory is present flushes it and all of it'ts children
3192  * from the dcache.
3193  *
3194  * It is safe and reasonable to cache /proc entries for a task until
3195  * that task exits.  After that they just clog up the dcache with
3196  * useless entries, possibly causing useful dcache entries to be
3197  * flushed instead.  This routine is proved to flush those useless
3198  * dcache entries at process exit time.
3199  *
3200  * NOTE: This routine is just an optimization so it does not guarantee
3201  *       that no dcache entries will exist at process exit time it
3202  *       just makes it very unlikely that any will persist.
3203  */
3204
3205 void proc_flush_task(struct task_struct *task)
3206 {
3207         int i;
3208         struct pid *pid, *tgid;
3209         struct upid *upid;
3210
3211         pid = task_pid(task);
3212         tgid = task_tgid(task);
3213
3214         for (i = 0; i <= pid->level; i++) {
3215                 upid = &pid->numbers[i];
3216                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3217                                         tgid->numbers[i].nr);
3218         }
3219
3220         upid = &pid->numbers[pid->level];
3221         if (upid->nr == 1)
3222                 pid_ns_release_proc(upid->ns);
3223 }
3224
3225 static struct dentry *proc_pid_instantiate(struct inode *dir,
3226                                            struct dentry * dentry,
3227                                            struct task_struct *task, const void *ptr)
3228 {
3229         struct dentry *error = ERR_PTR(-ENOENT);
3230         struct inode *inode;
3231
3232         inode = proc_pid_make_inode(dir->i_sb, task);
3233         if (!inode)
3234                 goto out;
3235
3236         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3237         inode->i_op = &proc_tgid_base_inode_operations;
3238         inode->i_fop = &proc_tgid_base_operations;
3239         inode->i_flags|=S_IMMUTABLE;
3240
3241         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3242                                                   ARRAY_SIZE(tgid_base_stuff)));
3243
3244         d_set_d_op(dentry, &pid_dentry_operations);
3245
3246         d_add(dentry, inode);
3247         /* Close the race of the process dying before we return the dentry */
3248         if (pid_revalidate(dentry, 0))
3249                 error = NULL;
3250 out:
3251         return error;
3252 }
3253
3254 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3255 {
3256         struct dentry *result;
3257         struct task_struct *task;
3258         unsigned tgid;
3259         struct pid_namespace *ns;
3260
3261         result = proc_base_lookup(dir, dentry);
3262         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3263                 goto out;
3264
3265         tgid = name_to_int(dentry);
3266         if (tgid == ~0U)
3267                 goto out;
3268
3269         ns = dentry->d_sb->s_fs_info;
3270         rcu_read_lock();
3271         task = find_task_by_pid_ns(tgid, ns);
3272         if (task)
3273                 get_task_struct(task);
3274         rcu_read_unlock();
3275         if (!task)
3276                 goto out;
3277
3278         result = proc_pid_instantiate(dir, dentry, task, NULL);
3279         put_task_struct(task);
3280 out:
3281         return result;
3282 }
3283
3284 /*
3285  * Find the first task with tgid >= tgid
3286  *
3287  */
3288 struct tgid_iter {
3289         unsigned int tgid;
3290         struct task_struct *task;
3291 };
3292 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3293 {
3294         struct pid *pid;
3295
3296         if (iter.task)
3297                 put_task_struct(iter.task);
3298         rcu_read_lock();
3299 retry:
3300         iter.task = NULL;
3301         pid = find_ge_pid(iter.tgid, ns);
3302         if (pid) {
3303                 iter.tgid = pid_nr_ns(pid, ns);
3304                 iter.task = pid_task(pid, PIDTYPE_PID);
3305                 /* What we to know is if the pid we have find is the
3306                  * pid of a thread_group_leader.  Testing for task
3307                  * being a thread_group_leader is the obvious thing
3308                  * todo but there is a window when it fails, due to
3309                  * the pid transfer logic in de_thread.
3310                  *
3311                  * So we perform the straight forward test of seeing
3312                  * if the pid we have found is the pid of a thread
3313                  * group leader, and don't worry if the task we have
3314                  * found doesn't happen to be a thread group leader.
3315                  * As we don't care in the case of readdir.
3316                  */
3317                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3318                         iter.tgid += 1;
3319                         goto retry;
3320                 }
3321                 get_task_struct(iter.task);
3322         }
3323         rcu_read_unlock();
3324         return iter;
3325 }
3326
3327 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3328
3329 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3330         struct tgid_iter iter)
3331 {
3332         char name[PROC_NUMBUF];
3333         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3334         return proc_fill_cache(filp, dirent, filldir, name, len,
3335                                 proc_pid_instantiate, iter.task, NULL);
3336 }
3337
3338 static int fake_filldir(void *buf, const char *name, int namelen,
3339                         loff_t offset, u64 ino, unsigned d_type)
3340 {
3341         return 0;
3342 }
3343
3344 /* for the /proc/ directory itself, after non-process stuff has been done */
3345 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3346 {
3347         unsigned int nr;
3348         struct task_struct *reaper;
3349         struct tgid_iter iter;
3350         struct pid_namespace *ns;
3351         filldir_t __filldir;
3352
3353         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3354                 goto out_no_task;
3355         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3356
3357         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3358         if (!reaper)
3359                 goto out_no_task;
3360
3361         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3362                 const struct pid_entry *p = &proc_base_stuff[nr];
3363                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3364                         goto out;
3365         }
3366
3367         ns = filp->f_dentry->d_sb->s_fs_info;
3368         iter.task = NULL;
3369         iter.tgid = filp->f_pos - TGID_OFFSET;
3370         for (iter = next_tgid(ns, iter);
3371              iter.task;
3372              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3373                 if (has_pid_permissions(ns, iter.task, 2))
3374                         __filldir = filldir;
3375                 else
3376                         __filldir = fake_filldir;
3377
3378                 filp->f_pos = iter.tgid + TGID_OFFSET;
3379                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3380                         put_task_struct(iter.task);
3381                         goto out;
3382                 }
3383         }
3384         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3385 out:
3386         put_task_struct(reaper);
3387 out_no_task:
3388         return 0;
3389 }
3390
3391 /*
3392  * Tasks
3393  */
3394 static const struct pid_entry tid_base_stuff[] = {
3395         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3396         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3397         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3398         REG("environ",   S_IRUSR, proc_environ_operations),
3399         INF("auxv",      S_IRUSR, proc_pid_auxv),
3400         ONE("status",    S_IRUGO, proc_pid_status),
3401         ONE("personality", S_IRUGO, proc_pid_personality),
3402         INF("limits",    S_IRUGO, proc_pid_limits),
3403 #ifdef CONFIG_SCHED_DEBUG
3404         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3405 #endif
3406         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3407 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3408         INF("syscall",   S_IRUGO, proc_pid_syscall),
3409 #endif
3410         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3411         ONE("stat",      S_IRUGO, proc_tid_stat),
3412         ONE("statm",     S_IRUGO, proc_pid_statm),
3413         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3414 #ifdef CONFIG_CHECKPOINT_RESTORE
3415         REG("children",  S_IRUGO, proc_tid_children_operations),
3416 #endif
3417 #ifdef CONFIG_NUMA
3418         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3419 #endif
3420         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3421         LNK("cwd",       proc_cwd_link),
3422         LNK("root",      proc_root_link),
3423         LNK("exe",       proc_exe_link),
3424         REG("mounts",    S_IRUGO, proc_mounts_operations),
3425         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3426 #ifdef CONFIG_PROC_PAGE_MONITOR
3427         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3428         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3429         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3430 #endif
3431 #ifdef CONFIG_SECURITY
3432         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3433 #endif
3434 #ifdef CONFIG_KALLSYMS
3435         INF("wchan",     S_IRUGO, proc_pid_wchan),
3436 #endif
3437 #ifdef CONFIG_STACKTRACE
3438         ONE("stack",      S_IRUGO, proc_pid_stack),
3439 #endif
3440 #ifdef CONFIG_SCHEDSTATS
3441         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3442 #endif
3443 #ifdef CONFIG_LATENCYTOP
3444         REG("latency",  S_IRUGO, proc_lstats_operations),
3445 #endif
3446 #ifdef CONFIG_PROC_PID_CPUSET
3447         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3448 #endif
3449 #ifdef CONFIG_CGROUPS
3450         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3451 #endif
3452         INF("oom_score", S_IRUGO, proc_oom_score),
3453         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3454         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3455 #ifdef CONFIG_AUDITSYSCALL
3456         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3457         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3458 #endif
3459 #ifdef CONFIG_FAULT_INJECTION
3460         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3461 #endif
3462 #ifdef CONFIG_TASK_IO_ACCOUNTING
3463         INF("io",       S_IRUSR, proc_tid_io_accounting),
3464 #endif
3465 #ifdef CONFIG_HARDWALL
3466         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3467 #endif
3468 #ifdef CONFIG_USER_NS
3469         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3470         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3471 #endif
3472 };
3473
3474 static int proc_tid_base_readdir(struct file * filp,
3475                              void * dirent, filldir_t filldir)
3476 {
3477         return proc_pident_readdir(filp,dirent,filldir,
3478                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3479 }
3480
3481 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3482 {
3483         return proc_pident_lookup(dir, dentry,
3484                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3485 }
3486
3487 static const struct file_operations proc_tid_base_operations = {
3488         .read           = generic_read_dir,
3489         .readdir        = proc_tid_base_readdir,
3490         .llseek         = default_llseek,
3491 };
3492
3493 static const struct inode_operations proc_tid_base_inode_operations = {
3494         .lookup         = proc_tid_base_lookup,
3495         .getattr        = pid_getattr,
3496         .setattr        = proc_setattr,
3497 };
3498
3499 static struct dentry *proc_task_instantiate(struct inode *dir,
3500         struct dentry *dentry, struct task_struct *task, const void *ptr)
3501 {
3502         struct dentry *error = ERR_PTR(-ENOENT);
3503         struct inode *inode;
3504         inode = proc_pid_make_inode(dir->i_sb, task);
3505
3506         if (!inode)
3507                 goto out;
3508         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3509         inode->i_op = &proc_tid_base_inode_operations;
3510         inode->i_fop = &proc_tid_base_operations;
3511         inode->i_flags|=S_IMMUTABLE;
3512
3513         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3514                                                   ARRAY_SIZE(tid_base_stuff)));
3515
3516         d_set_d_op(dentry, &pid_dentry_operations);
3517
3518         d_add(dentry, inode);
3519         /* Close the race of the process dying before we return the dentry */
3520         if (pid_revalidate(dentry, 0))
3521                 error = NULL;
3522 out:
3523         return error;
3524 }
3525
3526 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3527 {
3528         struct dentry *result = ERR_PTR(-ENOENT);
3529         struct task_struct *task;
3530         struct task_struct *leader = get_proc_task(dir);
3531         unsigned tid;
3532         struct pid_namespace *ns;
3533
3534         if (!leader)
3535                 goto out_no_task;
3536
3537         tid = name_to_int(dentry);
3538         if (tid == ~0U)
3539                 goto out;
3540
3541         ns = dentry->d_sb->s_fs_info;
3542         rcu_read_lock();
3543         task = find_task_by_pid_ns(tid, ns);
3544         if (task)
3545                 get_task_struct(task);
3546         rcu_read_unlock();
3547         if (!task)
3548                 goto out;
3549         if (!same_thread_group(leader, task))
3550                 goto out_drop_task;
3551
3552         result = proc_task_instantiate(dir, dentry, task, NULL);
3553 out_drop_task:
3554         put_task_struct(task);
3555 out:
3556         put_task_struct(leader);
3557 out_no_task:
3558         return result;
3559 }
3560
3561 /*
3562  * Find the first tid of a thread group to return to user space.
3563  *
3564  * Usually this is just the thread group leader, but if the users
3565  * buffer was too small or there was a seek into the middle of the
3566  * directory we have more work todo.
3567  *
3568  * In the case of a short read we start with find_task_by_pid.
3569  *
3570  * In the case of a seek we start with the leader and walk nr
3571  * threads past it.
3572  */
3573 static struct task_struct *first_tid(struct task_struct *leader,
3574                 int tid, int nr, struct pid_namespace *ns)
3575 {
3576         struct task_struct *pos;
3577
3578         rcu_read_lock();
3579         /* Attempt to start with the pid of a thread */
3580         if (tid && (nr > 0)) {
3581                 pos = find_task_by_pid_ns(tid, ns);
3582                 if (pos && (pos->group_leader == leader))
3583                         goto found;
3584         }
3585
3586         /* If nr exceeds the number of threads there is nothing todo */
3587         pos = NULL;
3588         if (nr && nr >= get_nr_threads(leader))
3589                 goto out;
3590
3591         /* If we haven't found our starting place yet start
3592          * with the leader and walk nr threads forward.
3593          */
3594         for (pos = leader; nr > 0; --nr) {
3595                 pos = next_thread(pos);
3596                 if (pos == leader) {
3597                         pos = NULL;
3598                         goto out;
3599                 }
3600         }
3601 found:
3602         get_task_struct(pos);
3603 out:
3604         rcu_read_unlock();
3605         return pos;
3606 }
3607
3608 /*
3609  * Find the next thread in the thread list.
3610  * Return NULL if there is an error or no next thread.
3611  *
3612  * The reference to the input task_struct is released.
3613  */
3614 static struct task_struct *next_tid(struct task_struct *start)
3615 {
3616         struct task_struct *pos = NULL;
3617         rcu_read_lock();
3618         if (pid_alive(start)) {
3619                 pos = next_thread(start);
3620                 if (thread_group_leader(pos))
3621                         pos = NULL;
3622                 else
3623                         get_task_struct(pos);
3624         }
3625         rcu_read_unlock();
3626         put_task_struct(start);
3627         return pos;
3628 }
3629
3630 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3631         struct task_struct *task, int tid)
3632 {
3633         char name[PROC_NUMBUF];
3634         int len = snprintf(name, sizeof(name), "%d", tid);
3635         return proc_fill_cache(filp, dirent, filldir, name, len,
3636                                 proc_task_instantiate, task, NULL);
3637 }
3638
3639 /* for the /proc/TGID/task/ directories */
3640 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3641 {
3642         struct dentry *dentry = filp->f_path.dentry;
3643         struct inode *inode = dentry->d_inode;
3644         struct task_struct *leader = NULL;
3645         struct task_struct *task;
3646         int retval = -ENOENT;
3647         ino_t ino;
3648         int tid;
3649         struct pid_namespace *ns;
3650
3651         task = get_proc_task(inode);
3652         if (!task)
3653                 goto out_no_task;
3654         rcu_read_lock();
3655         if (pid_alive(task)) {
3656                 leader = task->group_leader;
3657                 get_task_struct(leader);
3658         }
3659         rcu_read_unlock();
3660         put_task_struct(task);
3661         if (!leader)
3662                 goto out_no_task;
3663         retval = 0;
3664
3665         switch ((unsigned long)filp->f_pos) {
3666         case 0:
3667                 ino = inode->i_ino;
3668                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3669                         goto out;
3670                 filp->f_pos++;
3671                 /* fall through */
3672         case 1:
3673                 ino = parent_ino(dentry);
3674                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3675                         goto out;
3676                 filp->f_pos++;
3677                 /* fall through */
3678         }
3679
3680         /* f_version caches the tgid value that the last readdir call couldn't
3681          * return. lseek aka telldir automagically resets f_version to 0.
3682          */
3683         ns = filp->f_dentry->d_sb->s_fs_info;
3684         tid = (int)filp->f_version;
3685         filp->f_version = 0;
3686         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3687              task;
3688              task = next_tid(task), filp->f_pos++) {
3689                 tid = task_pid_nr_ns(task, ns);
3690                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3691                         /* returning this tgid failed, save it as the first
3692                          * pid for the next readir call */
3693                         filp->f_version = (u64)tid;
3694                         put_task_struct(task);
3695                         break;
3696                 }
3697         }
3698 out:
3699         put_task_struct(leader);
3700 out_no_task:
3701         return retval;
3702 }
3703
3704 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3705 {
3706         struct inode *inode = dentry->d_inode;
3707         struct task_struct *p = get_proc_task(inode);
3708         generic_fillattr(inode, stat);
3709
3710         if (p) {
3711                 stat->nlink += get_nr_threads(p);
3712                 put_task_struct(p);
3713         }
3714
3715         return 0;
3716 }
3717
3718 static const struct inode_operations proc_task_inode_operations = {
3719         .lookup         = proc_task_lookup,
3720         .getattr        = proc_task_getattr,
3721         .setattr        = proc_setattr,
3722         .permission     = proc_pid_permission,
3723 };
3724
3725 static const struct file_operations proc_task_operations = {
3726         .read           = generic_read_dir,
3727         .readdir        = proc_task_readdir,
3728         .llseek         = default_llseek,
3729 };