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