fs/proc/base.c

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