fs/proc/base.c

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