fs/proc/base.c

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