fs/proc/base.c

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