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