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         /* OK to pass negative loff_t, we can catch out-of-range */
 775         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 776         return 0;
 777 }
 778
 779 static ssize_t mem_read(struct file * file, char __user * buf,
 780                         size_t count, loff_t *ppos)
 781 {
 782         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 783         char *page;
 784         unsigned long src = *ppos;
 785         int ret = -ESRCH;
 786         struct mm_struct *mm;
 787
 788         if (!task)
 789                 goto out_no_task;
 790
 791         if (check_mem_permission(task))
 792                 goto out;
 793
 794         ret = -ENOMEM;
 795         page = (char *)__get_free_page(GFP_TEMPORARY);
 796         if (!page)
 797                 goto out;
 798
 799         ret = 0;
 800
 801         mm = get_task_mm(task);
 802         if (!mm)
 803                 goto out_free;
 804
 805         ret = -EIO;
 806
 807         if (file->private_data != (void*)((long)current->self_exec_id))
 808                 goto out_put;
 809
 810         ret = 0;
 811
 812         while (count > 0) {
 813                 int this_len, retval;
 814
 815                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 816                 retval = access_process_vm(task, src, page, this_len, 0);
 817                 if (!retval || check_mem_permission(task)) {
 818                         if (!ret)
 819                                 ret = -EIO;
 820                         break;
 821                 }
 822
 823                 if (copy_to_user(buf, page, retval)) {
 824                         ret = -EFAULT;
 825                         break;
 826                 }
 827
 828                 ret += retval;
 829                 src += retval;
 830                 buf += retval;
 831                 count -= retval;
 832         }
 833         *ppos = src;
 834
 835 out_put:
 836         mmput(mm);
 837 out_free:
 838         free_page((unsigned long) page);
 839 out:
 840         put_task_struct(task);
 841 out_no_task:
 842         return ret;
 843 }
 844
 845 #define mem_write NULL
 846
 847 #ifndef mem_write
 848 /* This is a security hazard */
 849 static ssize_t mem_write(struct file * file, const char __user *buf,
 850                          size_t count, loff_t *ppos)
 851 {
 852         int copied;
 853         char *page;
 854         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 855         unsigned long dst = *ppos;
 856
 857         copied = -ESRCH;
 858         if (!task)
 859                 goto out_no_task;
 860
 861         if (check_mem_permission(task))
 862                 goto out;
 863
 864         copied = -ENOMEM;
 865         page = (char *)__get_free_page(GFP_TEMPORARY);
 866         if (!page)
 867                 goto out;
 868
 869         copied = 0;
 870         while (count > 0) {
 871                 int this_len, retval;
 872
 873                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 874                 if (copy_from_user(page, buf, this_len)) {
 875                         copied = -EFAULT;
 876                         break;
 877                 }
 878                 retval = access_process_vm(task, dst, page, this_len, 1);
 879                 if (!retval) {
 880                         if (!copied)
 881                                 copied = -EIO;
 882                         break;
 883                 }
 884                 copied += retval;
 885                 buf += retval;
 886                 dst += retval;
 887                 count -= retval;
 888         }
 889         *ppos = dst;
 890         free_page((unsigned long) page);
 891 out:
 892         put_task_struct(task);
 893 out_no_task:
 894         return copied;
 895 }
 896 #endif
 897
 898 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 899 {
 900         switch (orig) {
 901         case 0:
 902                 file->f_pos = offset;
 903                 break;
 904         case 1:
 905                 file->f_pos += offset;
 906                 break;
 907         default:
 908                 return -EINVAL;
 909         }
 910         force_successful_syscall_return();
 911         return file->f_pos;
 912 }
 913
 914 static const struct file_operations proc_mem_operations = {
 915         .llseek         = mem_lseek,
 916         .read           = mem_read,
 917         .write          = mem_write,
 918         .open           = mem_open,
 919 };
 920
 921 static ssize_t environ_read(struct file *file, char __user *buf,
 922                         size_t count, loff_t *ppos)
 923 {
 924         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 925         char *page;
 926         unsigned long src = *ppos;
 927         int ret = -ESRCH;
 928         struct mm_struct *mm;
 929
 930         if (!task)
 931                 goto out_no_task;
 932
 933         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 934                 goto out;
 935
 936         ret = -ENOMEM;
 937         page = (char *)__get_free_page(GFP_TEMPORARY);
 938         if (!page)
 939                 goto out;
 940
 941         ret = 0;
 942
 943         mm = get_task_mm(task);
 944         if (!mm)
 945                 goto out_free;
 946
 947         while (count > 0) {
 948                 int this_len, retval, max_len;
 949
 950                 this_len = mm->env_end - (mm->env_start + src);
 951
 952                 if (this_len <= 0)
 953                         break;
 954
 955                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 956                 this_len = (this_len > max_len) ? max_len : this_len;
 957
 958                 retval = access_process_vm(task, (mm->env_start + src),
 959                         page, this_len, 0);
 960
 961                 if (retval <= 0) {
 962                         ret = retval;
 963                         break;
 964                 }
 965
 966                 if (copy_to_user(buf, page, retval)) {
 967                         ret = -EFAULT;
 968                         break;
 969                 }
 970
 971                 ret += retval;
 972                 src += retval;
 973                 buf += retval;
 974                 count -= retval;
 975         }
 976         *ppos = src;
 977
 978         mmput(mm);
 979 out_free:
 980         free_page((unsigned long) page);
 981 out:
 982         put_task_struct(task);
 983 out_no_task:
 984         return ret;
 985 }
 986
 987 static const struct file_operations proc_environ_operations = {
 988         .read           = environ_read,
 989         .llseek         = generic_file_llseek,
 990 };
 991
 992 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 993                                 size_t count, loff_t *ppos)
 994 {
 995         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 996         char buffer[PROC_NUMBUF];
 997         size_t len;
 998         int oom_adjust = OOM_DISABLE;
 999         unsigned long flags;
1000
1001         if (!task)
1002                 return -ESRCH;
1003
1004         if (lock_task_sighand(task, &flags)) {
1005                 oom_adjust = task->signal->oom_adj;
1006                 unlock_task_sighand(task, &flags);
1007         }
1008
1009         put_task_struct(task);
1010
1011         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1012
1013         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1014 }
1015
1016 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1017                                 size_t count, loff_t *ppos)
1018 {
1019         struct task_struct *task;
1020         char buffer[PROC_NUMBUF];
1021         long oom_adjust;
1022         unsigned long flags;
1023         int err;
1024
1025         memset(buffer, 0, sizeof(buffer));
1026         if (count > sizeof(buffer) - 1)
1027                 count = sizeof(buffer) - 1;
1028         if (copy_from_user(buffer, buf, count))
1029                 return -EFAULT;
1030
1031         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1032         if (err)
1033                 return -EINVAL;
1034         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1035              oom_adjust != OOM_DISABLE)
1036                 return -EINVAL;
1037
1038         task = get_proc_task(file->f_path.dentry->d_inode);
1039         if (!task)
1040                 return -ESRCH;
1041         if (!lock_task_sighand(task, &flags)) {
1042                 put_task_struct(task);
1043                 return -ESRCH;
1044         }
1045
1046         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1047                 unlock_task_sighand(task, &flags);
1048                 put_task_struct(task);
1049                 return -EACCES;
1050         }
1051
1052         /*
1053          * Warn that /proc/pid/oom_adj is deprecated, see
1054          * Documentation/feature-removal-schedule.txt.
1055          */
1056         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1057                         "please use /proc/%d/oom_score_adj instead.\n",
1058                         current->comm, task_pid_nr(current),
1059                         task_pid_nr(task), task_pid_nr(task));
1060         task->signal->oom_adj = oom_adjust;
1061         /*
1062          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1063          * value is always attainable.
1064          */
1065         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1066                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1067         else
1068                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1069                                                                 -OOM_DISABLE;
1070         unlock_task_sighand(task, &flags);
1071         put_task_struct(task);
1072
1073         return count;
1074 }
1075
1076 static const struct file_operations proc_oom_adjust_operations = {
1077         .read           = oom_adjust_read,
1078         .write          = oom_adjust_write,
1079         .llseek         = generic_file_llseek,
1080 };
1081
1082 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1083                                         size_t count, loff_t *ppos)
1084 {
1085         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1086         char buffer[PROC_NUMBUF];
1087         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1088         unsigned long flags;
1089         size_t len;
1090
1091         if (!task)
1092                 return -ESRCH;
1093         if (lock_task_sighand(task, &flags)) {
1094                 oom_score_adj = task->signal->oom_score_adj;
1095                 unlock_task_sighand(task, &flags);
1096         }
1097         put_task_struct(task);
1098         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1099         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1100 }
1101
1102 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1103                                         size_t count, loff_t *ppos)
1104 {
1105         struct task_struct *task;
1106         char buffer[PROC_NUMBUF];
1107         unsigned long flags;
1108         long oom_score_adj;
1109         int err;
1110
1111         memset(buffer, 0, sizeof(buffer));
1112         if (count > sizeof(buffer) - 1)
1113                 count = sizeof(buffer) - 1;
1114         if (copy_from_user(buffer, buf, count))
1115                 return -EFAULT;
1116
1117         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1118         if (err)
1119                 return -EINVAL;
1120         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1121                         oom_score_adj > OOM_SCORE_ADJ_MAX)
1122                 return -EINVAL;
1123
1124         task = get_proc_task(file->f_path.dentry->d_inode);
1125         if (!task)
1126                 return -ESRCH;
1127         if (!lock_task_sighand(task, &flags)) {
1128                 put_task_struct(task);
1129                 return -ESRCH;
1130         }
1131         if (oom_score_adj < task->signal->oom_score_adj &&
1132                         !capable(CAP_SYS_RESOURCE)) {
1133                 unlock_task_sighand(task, &flags);
1134                 put_task_struct(task);
1135                 return -EACCES;
1136         }
1137
1138         task->signal->oom_score_adj = oom_score_adj;
1139         /*
1140          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1141          * always attainable.
1142          */
1143         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1144                 task->signal->oom_adj = OOM_DISABLE;
1145         else
1146                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1147                                                         OOM_SCORE_ADJ_MAX;
1148         unlock_task_sighand(task, &flags);
1149         put_task_struct(task);
1150         return count;
1151 }
1152
1153 static const struct file_operations proc_oom_score_adj_operations = {
1154         .read           = oom_score_adj_read,
1155         .write          = oom_score_adj_write,
1156         .llseek         = default_llseek,
1157 };
1158
1159 #ifdef CONFIG_AUDITSYSCALL
1160 #define TMPBUFLEN 21
1161 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1162                                   size_t count, loff_t *ppos)
1163 {
1164         struct inode * inode = file->f_path.dentry->d_inode;
1165         struct task_struct *task = get_proc_task(inode);
1166         ssize_t length;
1167         char tmpbuf[TMPBUFLEN];
1168
1169         if (!task)
1170                 return -ESRCH;
1171         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1172                                 audit_get_loginuid(task));
1173         put_task_struct(task);
1174         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1175 }
1176
1177 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1178                                    size_t count, loff_t *ppos)
1179 {
1180         struct inode * inode = file->f_path.dentry->d_inode;
1181         char *page, *tmp;
1182         ssize_t length;
1183         uid_t loginuid;
1184
1185         if (!capable(CAP_AUDIT_CONTROL))
1186                 return -EPERM;
1187
1188         rcu_read_lock();
1189         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1190                 rcu_read_unlock();
1191                 return -EPERM;
1192         }
1193         rcu_read_unlock();
1194
1195         if (count >= PAGE_SIZE)
1196                 count = PAGE_SIZE - 1;
1197
1198         if (*ppos != 0) {
1199                 /* No partial writes. */
1200                 return -EINVAL;
1201         }
1202         page = (char*)__get_free_page(GFP_TEMPORARY);
1203         if (!page)
1204                 return -ENOMEM;
1205         length = -EFAULT;
1206         if (copy_from_user(page, buf, count))
1207                 goto out_free_page;
1208
1209         page[count] = '\0';
1210         loginuid = simple_strtoul(page, &tmp, 10);
1211         if (tmp == page) {
1212                 length = -EINVAL;
1213                 goto out_free_page;
1214
1215         }
1216         length = audit_set_loginuid(current, loginuid);
1217         if (likely(length == 0))
1218                 length = count;
1219
1220 out_free_page:
1221         free_page((unsigned long) page);
1222         return length;
1223 }
1224
1225 static const struct file_operations proc_loginuid_operations = {
1226         .read           = proc_loginuid_read,
1227         .write          = proc_loginuid_write,
1228         .llseek         = generic_file_llseek,
1229 };
1230
1231 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1232                                   size_t count, loff_t *ppos)
1233 {
1234         struct inode * inode = file->f_path.dentry->d_inode;
1235         struct task_struct *task = get_proc_task(inode);
1236         ssize_t length;
1237         char tmpbuf[TMPBUFLEN];
1238
1239         if (!task)
1240                 return -ESRCH;
1241         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1242                                 audit_get_sessionid(task));
1243         put_task_struct(task);
1244         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1245 }
1246
1247 static const struct file_operations proc_sessionid_operations = {
1248         .read           = proc_sessionid_read,
1249         .llseek         = generic_file_llseek,
1250 };
1251 #endif
1252
1253 #ifdef CONFIG_FAULT_INJECTION
1254 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1255                                       size_t count, loff_t *ppos)
1256 {
1257         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1258         char buffer[PROC_NUMBUF];
1259         size_t len;
1260         int make_it_fail;
1261
1262         if (!task)
1263                 return -ESRCH;
1264         make_it_fail = task->make_it_fail;
1265         put_task_struct(task);
1266
1267         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1268
1269         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1270 }
1271
1272 static ssize_t proc_fault_inject_write(struct file * file,
1273                         const char __user * buf, size_t count, loff_t *ppos)
1274 {
1275         struct task_struct *task;
1276         char buffer[PROC_NUMBUF], *end;
1277         int make_it_fail;
1278
1279         if (!capable(CAP_SYS_RESOURCE))
1280                 return -EPERM;
1281         memset(buffer, 0, sizeof(buffer));
1282         if (count > sizeof(buffer) - 1)
1283                 count = sizeof(buffer) - 1;
1284         if (copy_from_user(buffer, buf, count))
1285                 return -EFAULT;
1286         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1287         if (*end)
1288                 return -EINVAL;
1289         task = get_proc_task(file->f_dentry->d_inode);
1290         if (!task)
1291                 return -ESRCH;
1292         task->make_it_fail = make_it_fail;
1293         put_task_struct(task);
1294
1295         return count;
1296 }
1297
1298 static const struct file_operations proc_fault_inject_operations = {
1299         .read           = proc_fault_inject_read,
1300         .write          = proc_fault_inject_write,
1301         .llseek         = generic_file_llseek,
1302 };
1303 #endif
1304
1305
1306 #ifdef CONFIG_SCHED_DEBUG
1307 /*
1308  * Print out various scheduling related per-task fields:
1309  */
1310 static int sched_show(struct seq_file *m, void *v)
1311 {
1312         struct inode *inode = m->private;
1313         struct task_struct *p;
1314
1315         p = get_proc_task(inode);
1316         if (!p)
1317                 return -ESRCH;
1318         proc_sched_show_task(p, m);
1319
1320         put_task_struct(p);
1321
1322         return 0;
1323 }
1324
1325 static ssize_t
1326 sched_write(struct file *file, const char __user *buf,
1327             size_t count, loff_t *offset)
1328 {
1329         struct inode *inode = file->f_path.dentry->d_inode;
1330         struct task_struct *p;
1331
1332         p = get_proc_task(inode);
1333         if (!p)
1334                 return -ESRCH;
1335         proc_sched_set_task(p);
1336
1337         put_task_struct(p);
1338
1339         return count;
1340 }
1341
1342 static int sched_open(struct inode *inode, struct file *filp)
1343 {
1344         int ret;
1345
1346         ret = single_open(filp, sched_show, NULL);
1347         if (!ret) {
1348                 struct seq_file *m = filp->private_data;
1349
1350                 m->private = inode;
1351         }
1352         return ret;
1353 }
1354
1355 static const struct file_operations proc_pid_sched_operations = {
1356         .open           = sched_open,
1357         .read           = seq_read,
1358         .write          = sched_write,
1359         .llseek         = seq_lseek,
1360         .release        = single_release,
1361 };
1362
1363 #endif
1364
1365 static ssize_t comm_write(struct file *file, const char __user *buf,
1366                                 size_t count, loff_t *offset)
1367 {
1368         struct inode *inode = file->f_path.dentry->d_inode;
1369         struct task_struct *p;
1370         char buffer[TASK_COMM_LEN];
1371
1372         memset(buffer, 0, sizeof(buffer));
1373         if (count > sizeof(buffer) - 1)
1374                 count = sizeof(buffer) - 1;
1375         if (copy_from_user(buffer, buf, count))
1376                 return -EFAULT;
1377
1378         p = get_proc_task(inode);
1379         if (!p)
1380                 return -ESRCH;
1381
1382         if (same_thread_group(current, p))
1383                 set_task_comm(p, buffer);
1384         else
1385                 count = -EINVAL;
1386
1387         put_task_struct(p);
1388
1389         return count;
1390 }
1391
1392 static int comm_show(struct seq_file *m, void *v)
1393 {
1394         struct inode *inode = m->private;
1395         struct task_struct *p;
1396
1397         p = get_proc_task(inode);
1398         if (!p)
1399                 return -ESRCH;
1400
1401         task_lock(p);
1402         seq_printf(m, "%s\n", p->comm);
1403         task_unlock(p);
1404
1405         put_task_struct(p);
1406
1407         return 0;
1408 }
1409
1410 static int comm_open(struct inode *inode, struct file *filp)
1411 {
1412         int ret;
1413
1414         ret = single_open(filp, comm_show, NULL);
1415         if (!ret) {
1416                 struct seq_file *m = filp->private_data;
1417
1418                 m->private = inode;
1419         }
1420         return ret;
1421 }
1422
1423 static const struct file_operations proc_pid_set_comm_operations = {
1424         .open           = comm_open,
1425         .read           = seq_read,
1426         .write          = comm_write,
1427         .llseek         = seq_lseek,
1428         .release        = single_release,
1429 };
1430
1431 /*
1432  * We added or removed a vma mapping the executable. The vmas are only mapped
1433  * during exec and are not mapped with the mmap system call.
1434  * Callers must hold down_write() on the mm's mmap_sem for these
1435  */
1436 void added_exe_file_vma(struct mm_struct *mm)
1437 {
1438         mm->num_exe_file_vmas++;
1439 }
1440
1441 void removed_exe_file_vma(struct mm_struct *mm)
1442 {
1443         mm->num_exe_file_vmas--;
1444         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1445                 fput(mm->exe_file);
1446                 mm->exe_file = NULL;
1447         }
1448
1449 }
1450
1451 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1452 {
1453         if (new_exe_file)
1454                 get_file(new_exe_file);
1455         if (mm->exe_file)
1456                 fput(mm->exe_file);
1457         mm->exe_file = new_exe_file;
1458         mm->num_exe_file_vmas = 0;
1459 }
1460
1461 struct file *get_mm_exe_file(struct mm_struct *mm)
1462 {
1463         struct file *exe_file;
1464
1465         /* We need mmap_sem to protect against races with removal of
1466          * VM_EXECUTABLE vmas */
1467         down_read(&mm->mmap_sem);
1468         exe_file = mm->exe_file;
1469         if (exe_file)
1470                 get_file(exe_file);
1471         up_read(&mm->mmap_sem);
1472         return exe_file;
1473 }
1474
1475 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1476 {
1477         /* It's safe to write the exe_file pointer without exe_file_lock because
1478          * this is called during fork when the task is not yet in /proc */
1479         newmm->exe_file = get_mm_exe_file(oldmm);
1480 }
1481
1482 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1483 {
1484         struct task_struct *task;
1485         struct mm_struct *mm;
1486         struct file *exe_file;
1487
1488         task = get_proc_task(inode);
1489         if (!task)
1490                 return -ENOENT;
1491         mm = get_task_mm(task);
1492         put_task_struct(task);
1493         if (!mm)
1494                 return -ENOENT;
1495         exe_file = get_mm_exe_file(mm);
1496         mmput(mm);
1497         if (exe_file) {
1498                 *exe_path = exe_file->f_path;
1499                 path_get(&exe_file->f_path);
1500                 fput(exe_file);
1501                 return 0;
1502         } else
1503                 return -ENOENT;
1504 }
1505
1506 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1507 {
1508         struct inode *inode = dentry->d_inode;
1509         int error = -EACCES;
1510
1511         /* We don't need a base pointer in the /proc filesystem */
1512         path_put(&nd->path);
1513
1514         /* Are we allowed to snoop on the tasks file descriptors? */
1515         if (!proc_fd_access_allowed(inode))
1516                 goto out;
1517
1518         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1519 out:
1520         return ERR_PTR(error);
1521 }
1522
1523 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1524 {
1525         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1526         char *pathname;
1527         int len;
1528
1529         if (!tmp)
1530                 return -ENOMEM;
1531
1532         pathname = d_path_with_unreachable(path, tmp, PAGE_SIZE);
1533         len = PTR_ERR(pathname);
1534         if (IS_ERR(pathname))
1535                 goto out;
1536         len = tmp + PAGE_SIZE - 1 - pathname;
1537
1538         if (len > buflen)
1539                 len = buflen;
1540         if (copy_to_user(buffer, pathname, len))
1541                 len = -EFAULT;
1542  out:
1543         free_page((unsigned long)tmp);
1544         return len;
1545 }
1546
1547 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1548 {
1549         int error = -EACCES;
1550         struct inode *inode = dentry->d_inode;
1551         struct path path;
1552
1553         /* Are we allowed to snoop on the tasks file descriptors? */
1554         if (!proc_fd_access_allowed(inode))
1555                 goto out;
1556
1557         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1558         if (error)
1559                 goto out;
1560
1561         error = do_proc_readlink(&path, buffer, buflen);
1562         path_put(&path);
1563 out:
1564         return error;
1565 }
1566
1567 static const struct inode_operations proc_pid_link_inode_operations = {
1568         .readlink       = proc_pid_readlink,
1569         .follow_link    = proc_pid_follow_link,
1570         .setattr        = proc_setattr,
1571 };
1572
1573
1574 /* building an inode */
1575
1576 static int task_dumpable(struct task_struct *task)
1577 {
1578         int dumpable = 0;
1579         struct mm_struct *mm;
1580
1581         task_lock(task);
1582         mm = task->mm;
1583         if (mm)
1584                 dumpable = get_dumpable(mm);
1585         task_unlock(task);
1586         if(dumpable == 1)
1587                 return 1;
1588         return 0;
1589 }
1590
1591
1592 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1593 {
1594         struct inode * inode;
1595         struct proc_inode *ei;
1596         const struct cred *cred;
1597
1598         /* We need a new inode */
1599
1600         inode = new_inode(sb);
1601         if (!inode)
1602                 goto out;
1603
1604         /* Common stuff */
1605         ei = PROC_I(inode);
1606         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1607         inode->i_op = &proc_def_inode_operations;
1608
1609         /*
1610          * grab the reference to task.
1611          */
1612         ei->pid = get_task_pid(task, PIDTYPE_PID);
1613         if (!ei->pid)
1614                 goto out_unlock;
1615
1616         if (task_dumpable(task)) {
1617                 rcu_read_lock();
1618                 cred = __task_cred(task);
1619                 inode->i_uid = cred->euid;
1620                 inode->i_gid = cred->egid;
1621                 rcu_read_unlock();
1622         }
1623         security_task_to_inode(task, inode);
1624
1625 out:
1626         return inode;
1627
1628 out_unlock:
1629         iput(inode);
1630         return NULL;
1631 }
1632
1633 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1634 {
1635         struct inode *inode = dentry->d_inode;
1636         struct task_struct *task;
1637         const struct cred *cred;
1638
1639         generic_fillattr(inode, stat);
1640
1641         rcu_read_lock();
1642         stat->uid = 0;
1643         stat->gid = 0;
1644         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1645         if (task) {
1646                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1647                     task_dumpable(task)) {
1648                         cred = __task_cred(task);
1649                         stat->uid = cred->euid;
1650                         stat->gid = cred->egid;
1651                 }
1652         }
1653         rcu_read_unlock();
1654         return 0;
1655 }
1656
1657 /* dentry stuff */
1658
1659 /*
1660  *      Exceptional case: normally we are not allowed to unhash a busy
1661  * directory. In this case, however, we can do it - no aliasing problems
1662  * due to the way we treat inodes.
1663  *
1664  * Rewrite the inode's ownerships here because the owning task may have
1665  * performed a setuid(), etc.
1666  *
1667  * Before the /proc/pid/status file was created the only way to read
1668  * the effective uid of a /process was to stat /proc/pid.  Reading
1669  * /proc/pid/status is slow enough that procps and other packages
1670  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1671  * made this apply to all per process world readable and executable
1672  * directories.
1673  */
1674 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1675 {
1676         struct inode *inode = dentry->d_inode;
1677         struct task_struct *task = get_proc_task(inode);
1678         const struct cred *cred;
1679
1680         if (task) {
1681                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1682                     task_dumpable(task)) {
1683                         rcu_read_lock();
1684                         cred = __task_cred(task);
1685                         inode->i_uid = cred->euid;
1686                         inode->i_gid = cred->egid;
1687                         rcu_read_unlock();
1688                 } else {
1689                         inode->i_uid = 0;
1690                         inode->i_gid = 0;
1691                 }
1692                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1693                 security_task_to_inode(task, inode);
1694                 put_task_struct(task);
1695                 return 1;
1696         }
1697         d_drop(dentry);
1698         return 0;
1699 }
1700
1701 static int pid_delete_dentry(struct dentry * dentry)
1702 {
1703         /* Is the task we represent dead?
1704          * If so, then don't put the dentry on the lru list,
1705          * kill it immediately.
1706          */
1707         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1708 }
1709
1710 static const struct dentry_operations pid_dentry_operations =
1711 {
1712         .d_revalidate   = pid_revalidate,
1713         .d_delete       = pid_delete_dentry,
1714 };
1715
1716 /* Lookups */
1717
1718 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1719                                 struct task_struct *, const void *);
1720
1721 /*
1722  * Fill a directory entry.
1723  *
1724  * If possible create the dcache entry and derive our inode number and
1725  * file type from dcache entry.
1726  *
1727  * Since all of the proc inode numbers are dynamically generated, the inode
1728  * numbers do not exist until the inode is cache.  This means creating the
1729  * the dcache entry in readdir is necessary to keep the inode numbers
1730  * reported by readdir in sync with the inode numbers reported
1731  * by stat.
1732  */
1733 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1734         char *name, int len,
1735         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1736 {
1737         struct dentry *child, *dir = filp->f_path.dentry;
1738         struct inode *inode;
1739         struct qstr qname;
1740         ino_t ino = 0;
1741         unsigned type = DT_UNKNOWN;
1742
1743         qname.name = name;
1744         qname.len  = len;
1745         qname.hash = full_name_hash(name, len);
1746
1747         child = d_lookup(dir, &qname);
1748         if (!child) {
1749                 struct dentry *new;
1750                 new = d_alloc(dir, &qname);
1751                 if (new) {
1752                         child = instantiate(dir->d_inode, new, task, ptr);
1753                         if (child)
1754                                 dput(new);
1755                         else
1756                                 child = new;
1757                 }
1758         }
1759         if (!child || IS_ERR(child) || !child->d_inode)
1760                 goto end_instantiate;
1761         inode = child->d_inode;
1762         if (inode) {
1763                 ino = inode->i_ino;
1764                 type = inode->i_mode >> 12;
1765         }
1766         dput(child);
1767 end_instantiate:
1768         if (!ino)
1769                 ino = find_inode_number(dir, &qname);
1770         if (!ino)
1771                 ino = 1;
1772         return filldir(dirent, name, len, filp->f_pos, ino, type);
1773 }
1774
1775 static unsigned name_to_int(struct dentry *dentry)
1776 {
1777         const char *name = dentry->d_name.name;
1778         int len = dentry->d_name.len;
1779         unsigned n = 0;
1780
1781         if (len > 1 && *name == '0')
1782                 goto out;
1783         while (len-- > 0) {
1784                 unsigned c = *name++ - '0';
1785                 if (c > 9)
1786                         goto out;
1787                 if (n >= (~0U-9)/10)
1788                         goto out;
1789                 n *= 10;
1790                 n += c;
1791         }
1792         return n;
1793 out:
1794         return ~0U;
1795 }
1796
1797 #define PROC_FDINFO_MAX 64
1798
1799 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1800 {
1801         struct task_struct *task = get_proc_task(inode);
1802         struct files_struct *files = NULL;
1803         struct file *file;
1804         int fd = proc_fd(inode);
1805
1806         if (task) {
1807                 files = get_files_struct(task);
1808                 put_task_struct(task);
1809         }
1810         if (files) {
1811                 /*
1812                  * We are not taking a ref to the file structure, so we must
1813                  * hold ->file_lock.
1814                  */
1815                 spin_lock(&files->file_lock);
1816                 file = fcheck_files(files, fd);
1817                 if (file) {
1818                         if (path) {
1819                                 *path = file->f_path;
1820                                 path_get(&file->f_path);
1821                         }
1822                         if (info)
1823                                 snprintf(info, PROC_FDINFO_MAX,
1824                                          "pos:\t%lli\n"
1825                                          "flags:\t0%o\n",
1826                                          (long long) file->f_pos,
1827                                          file->f_flags);
1828                         spin_unlock(&files->file_lock);
1829                         put_files_struct(files);
1830                         return 0;
1831                 }
1832                 spin_unlock(&files->file_lock);
1833                 put_files_struct(files);
1834         }
1835         return -ENOENT;
1836 }
1837
1838 static int proc_fd_link(struct inode *inode, struct path *path)
1839 {
1840         return proc_fd_info(inode, path, NULL);
1841 }
1842
1843 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1844 {
1845         struct inode *inode = dentry->d_inode;
1846         struct task_struct *task = get_proc_task(inode);
1847         int fd = proc_fd(inode);
1848         struct files_struct *files;
1849         const struct cred *cred;
1850
1851         if (task) {
1852                 files = get_files_struct(task);
1853                 if (files) {
1854                         rcu_read_lock();
1855                         if (fcheck_files(files, fd)) {
1856                                 rcu_read_unlock();
1857                                 put_files_struct(files);
1858                                 if (task_dumpable(task)) {
1859                                         rcu_read_lock();
1860                                         cred = __task_cred(task);
1861                                         inode->i_uid = cred->euid;
1862                                         inode->i_gid = cred->egid;
1863                                         rcu_read_unlock();
1864                                 } else {
1865                                         inode->i_uid = 0;
1866                                         inode->i_gid = 0;
1867                                 }
1868                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1869                                 security_task_to_inode(task, inode);
1870                                 put_task_struct(task);
1871                                 return 1;
1872                         }
1873                         rcu_read_unlock();
1874                         put_files_struct(files);
1875                 }
1876                 put_task_struct(task);
1877         }
1878         d_drop(dentry);
1879         return 0;
1880 }
1881
1882 static const struct dentry_operations tid_fd_dentry_operations =
1883 {
1884         .d_revalidate   = tid_fd_revalidate,
1885         .d_delete       = pid_delete_dentry,
1886 };
1887
1888 static struct dentry *proc_fd_instantiate(struct inode *dir,
1889         struct dentry *dentry, struct task_struct *task, const void *ptr)
1890 {
1891         unsigned fd = *(const unsigned *)ptr;
1892         struct file *file;
1893         struct files_struct *files;
1894         struct inode *inode;
1895         struct proc_inode *ei;
1896         struct dentry *error = ERR_PTR(-ENOENT);
1897
1898         inode = proc_pid_make_inode(dir->i_sb, task);
1899         if (!inode)
1900                 goto out;
1901         ei = PROC_I(inode);
1902         ei->fd = fd;
1903         files = get_files_struct(task);
1904         if (!files)
1905                 goto out_iput;
1906         inode->i_mode = S_IFLNK;
1907
1908         /*
1909          * We are not taking a ref to the file structure, so we must
1910          * hold ->file_lock.
1911          */
1912         spin_lock(&files->file_lock);
1913         file = fcheck_files(files, fd);
1914         if (!file)
1915                 goto out_unlock;
1916         if (file->f_mode & FMODE_READ)
1917                 inode->i_mode |= S_IRUSR | S_IXUSR;
1918         if (file->f_mode & FMODE_WRITE)
1919                 inode->i_mode |= S_IWUSR | S_IXUSR;
1920         spin_unlock(&files->file_lock);
1921         put_files_struct(files);
1922
1923         inode->i_op = &proc_pid_link_inode_operations;
1924         inode->i_size = 64;
1925         ei->op.proc_get_link = proc_fd_link;
1926         dentry->d_op = &tid_fd_dentry_operations;
1927         d_add(dentry, inode);
1928         /* Close the race of the process dying before we return the dentry */
1929         if (tid_fd_revalidate(dentry, NULL))
1930                 error = NULL;
1931
1932  out:
1933         return error;
1934 out_unlock:
1935         spin_unlock(&files->file_lock);
1936         put_files_struct(files);
1937 out_iput:
1938         iput(inode);
1939         goto out;
1940 }
1941
1942 static struct dentry *proc_lookupfd_common(struct inode *dir,
1943                                            struct dentry *dentry,
1944                                            instantiate_t instantiate)
1945 {
1946         struct task_struct *task = get_proc_task(dir);
1947         unsigned fd = name_to_int(dentry);
1948         struct dentry *result = ERR_PTR(-ENOENT);
1949
1950         if (!task)
1951                 goto out_no_task;
1952         if (fd == ~0U)
1953                 goto out;
1954
1955         result = instantiate(dir, dentry, task, &fd);
1956 out:
1957         put_task_struct(task);
1958 out_no_task:
1959         return result;
1960 }
1961
1962 static int proc_readfd_common(struct file * filp, void * dirent,
1963                               filldir_t filldir, instantiate_t instantiate)
1964 {
1965         struct dentry *dentry = filp->f_path.dentry;
1966         struct inode *inode = dentry->d_inode;
1967         struct task_struct *p = get_proc_task(inode);
1968         unsigned int fd, ino;
1969         int retval;
1970         struct files_struct * files;
1971
1972         retval = -ENOENT;
1973         if (!p)
1974                 goto out_no_task;
1975         retval = 0;
1976
1977         fd = filp->f_pos;
1978         switch (fd) {
1979                 case 0:
1980                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1981                                 goto out;
1982                         filp->f_pos++;
1983                 case 1:
1984                         ino = parent_ino(dentry);
1985                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1986                                 goto out;
1987                         filp->f_pos++;
1988                 default:
1989                         files = get_files_struct(p);
1990                         if (!files)
1991                                 goto out;
1992                         rcu_read_lock();
1993                         for (fd = filp->f_pos-2;
1994                              fd < files_fdtable(files)->max_fds;
1995                              fd++, filp->f_pos++) {
1996                                 char name[PROC_NUMBUF];
1997                                 int len;
1998
1999                                 if (!fcheck_files(files, fd))
2000                                         continue;
2001                                 rcu_read_unlock();
2002
2003                                 len = snprintf(name, sizeof(name), "%d", fd);
2004                                 if (proc_fill_cache(filp, dirent, filldir,
2005                                                     name, len, instantiate,
2006                                                     p, &fd) < 0) {
2007                                         rcu_read_lock();
2008                                         break;
2009                                 }
2010                                 rcu_read_lock();
2011                         }
2012                         rcu_read_unlock();
2013                         put_files_struct(files);
2014         }
2015 out:
2016         put_task_struct(p);
2017 out_no_task:
2018         return retval;
2019 }
2020
2021 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2022                                     struct nameidata *nd)
2023 {
2024         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2025 }
2026
2027 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2028 {
2029         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2030 }
2031
2032 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2033                                       size_t len, loff_t *ppos)
2034 {
2035         char tmp[PROC_FDINFO_MAX];
2036         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2037         if (!err)
2038                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2039         return err;
2040 }
2041
2042 static const struct file_operations proc_fdinfo_file_operations = {
2043         .open           = nonseekable_open,
2044         .read           = proc_fdinfo_read,
2045         .llseek         = no_llseek,
2046 };
2047
2048 static const struct file_operations proc_fd_operations = {
2049         .read           = generic_read_dir,
2050         .readdir        = proc_readfd,
2051         .llseek         = default_llseek,
2052 };
2053
2054 /*
2055  * /proc/pid/fd needs a special permission handler so that a process can still
2056  * access /proc/self/fd after it has executed a setuid().
2057  */
2058 static int proc_fd_permission(struct inode *inode, int mask)
2059 {
2060         int rv;
2061
2062         rv = generic_permission(inode, mask, NULL);
2063         if (rv == 0)
2064                 return 0;
2065         if (task_pid(current) == proc_pid(inode))
2066                 rv = 0;
2067         return rv;
2068 }
2069
2070 /*
2071  * proc directories can do almost nothing..
2072  */
2073 static const struct inode_operations proc_fd_inode_operations = {
2074         .lookup         = proc_lookupfd,
2075         .permission     = proc_fd_permission,
2076         .setattr        = proc_setattr,
2077 };
2078
2079 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2080         struct dentry *dentry, struct task_struct *task, const void *ptr)
2081 {
2082         unsigned fd = *(unsigned *)ptr;
2083         struct inode *inode;
2084         struct proc_inode *ei;
2085         struct dentry *error = ERR_PTR(-ENOENT);
2086
2087         inode = proc_pid_make_inode(dir->i_sb, task);
2088         if (!inode)
2089                 goto out;
2090         ei = PROC_I(inode);
2091         ei->fd = fd;
2092         inode->i_mode = S_IFREG | S_IRUSR;
2093         inode->i_fop = &proc_fdinfo_file_operations;
2094         dentry->d_op = &tid_fd_dentry_operations;
2095         d_add(dentry, inode);
2096         /* Close the race of the process dying before we return the dentry */
2097         if (tid_fd_revalidate(dentry, NULL))
2098                 error = NULL;
2099
2100  out:
2101         return error;
2102 }
2103
2104 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2105                                         struct dentry *dentry,
2106                                         struct nameidata *nd)
2107 {
2108         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2109 }
2110
2111 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2112 {
2113         return proc_readfd_common(filp, dirent, filldir,
2114                                   proc_fdinfo_instantiate);
2115 }
2116
2117 static const struct file_operations proc_fdinfo_operations = {
2118         .read           = generic_read_dir,
2119         .readdir        = proc_readfdinfo,
2120         .llseek         = default_llseek,
2121 };
2122
2123 /*
2124  * proc directories can do almost nothing..
2125  */
2126 static const struct inode_operations proc_fdinfo_inode_operations = {
2127         .lookup         = proc_lookupfdinfo,
2128         .setattr        = proc_setattr,
2129 };
2130
2131
2132 static struct dentry *proc_pident_instantiate(struct inode *dir,
2133         struct dentry *dentry, struct task_struct *task, const void *ptr)
2134 {
2135         const struct pid_entry *p = ptr;
2136         struct inode *inode;
2137         struct proc_inode *ei;
2138         struct dentry *error = ERR_PTR(-ENOENT);
2139
2140         inode = proc_pid_make_inode(dir->i_sb, task);
2141         if (!inode)
2142                 goto out;
2143
2144         ei = PROC_I(inode);
2145         inode->i_mode = p->mode;
2146         if (S_ISDIR(inode->i_mode))
2147                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2148         if (p->iop)
2149                 inode->i_op = p->iop;
2150         if (p->fop)
2151                 inode->i_fop = p->fop;
2152         ei->op = p->op;
2153         dentry->d_op = &pid_dentry_operations;
2154         d_add(dentry, inode);
2155         /* Close the race of the process dying before we return the dentry */
2156         if (pid_revalidate(dentry, NULL))
2157                 error = NULL;
2158 out:
2159         return error;
2160 }
2161
2162 static struct dentry *proc_pident_lookup(struct inode *dir,
2163                                          struct dentry *dentry,
2164                                          const struct pid_entry *ents,
2165                                          unsigned int nents)
2166 {
2167         struct dentry *error;
2168         struct task_struct *task = get_proc_task(dir);
2169         const struct pid_entry *p, *last;
2170
2171         error = ERR_PTR(-ENOENT);
2172
2173         if (!task)
2174                 goto out_no_task;
2175
2176         /*
2177          * Yes, it does not scale. And it should not. Don't add
2178          * new entries into /proc/<tgid>/ without very good reasons.
2179          */
2180         last = &ents[nents - 1];
2181         for (p = ents; p <= last; p++) {
2182                 if (p->len != dentry->d_name.len)
2183                         continue;
2184                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2185                         break;
2186         }
2187         if (p > last)
2188                 goto out;
2189
2190         error = proc_pident_instantiate(dir, dentry, task, p);
2191 out:
2192         put_task_struct(task);
2193 out_no_task:
2194         return error;
2195 }
2196
2197 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2198         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2199 {
2200         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2201                                 proc_pident_instantiate, task, p);
2202 }
2203
2204 static int proc_pident_readdir(struct file *filp,
2205                 void *dirent, filldir_t filldir,
2206                 const struct pid_entry *ents, unsigned int nents)
2207 {
2208         int i;
2209         struct dentry *dentry = filp->f_path.dentry;
2210         struct inode *inode = dentry->d_inode;
2211         struct task_struct *task = get_proc_task(inode);
2212         const struct pid_entry *p, *last;
2213         ino_t ino;
2214         int ret;
2215
2216         ret = -ENOENT;
2217         if (!task)
2218                 goto out_no_task;
2219
2220         ret = 0;
2221         i = filp->f_pos;
2222         switch (i) {
2223         case 0:
2224                 ino = inode->i_ino;
2225                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2226                         goto out;
2227                 i++;
2228                 filp->f_pos++;
2229                 /* fall through */
2230         case 1:
2231                 ino = parent_ino(dentry);
2232                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2233                         goto out;
2234                 i++;
2235                 filp->f_pos++;
2236                 /* fall through */
2237         default:
2238                 i -= 2;
2239                 if (i >= nents) {
2240                         ret = 1;
2241                         goto out;
2242                 }
2243                 p = ents + i;
2244                 last = &ents[nents - 1];
2245                 while (p <= last) {
2246                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2247                                 goto out;
2248                         filp->f_pos++;
2249                         p++;
2250                 }
2251         }
2252
2253         ret = 1;
2254 out:
2255         put_task_struct(task);
2256 out_no_task:
2257         return ret;
2258 }
2259
2260 #ifdef CONFIG_SECURITY
2261 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2262                                   size_t count, loff_t *ppos)
2263 {
2264         struct inode * inode = file->f_path.dentry->d_inode;
2265         char *p = NULL;
2266         ssize_t length;
2267         struct task_struct *task = get_proc_task(inode);
2268
2269         if (!task)
2270                 return -ESRCH;
2271
2272         length = security_getprocattr(task,
2273                                       (char*)file->f_path.dentry->d_name.name,
2274                                       &p);
2275         put_task_struct(task);
2276         if (length > 0)
2277                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2278         kfree(p);
2279         return length;
2280 }
2281
2282 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2283                                    size_t count, loff_t *ppos)
2284 {
2285         struct inode * inode = file->f_path.dentry->d_inode;
2286         char *page;
2287         ssize_t length;
2288         struct task_struct *task = get_proc_task(inode);
2289
2290         length = -ESRCH;
2291         if (!task)
2292                 goto out_no_task;
2293         if (count > PAGE_SIZE)
2294                 count = PAGE_SIZE;
2295
2296         /* No partial writes. */
2297         length = -EINVAL;
2298         if (*ppos != 0)
2299                 goto out;
2300
2301         length = -ENOMEM;
2302         page = (char*)__get_free_page(GFP_TEMPORARY);
2303         if (!page)
2304                 goto out;
2305
2306         length = -EFAULT;
2307         if (copy_from_user(page, buf, count))
2308                 goto out_free;
2309
2310         /* Guard against adverse ptrace interaction */
2311         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2312         if (length < 0)
2313                 goto out_free;
2314
2315         length = security_setprocattr(task,
2316                                       (char*)file->f_path.dentry->d_name.name,
2317                                       (void*)page, count);
2318         mutex_unlock(&task->cred_guard_mutex);
2319 out_free:
2320         free_page((unsigned long) page);
2321 out:
2322         put_task_struct(task);
2323 out_no_task:
2324         return length;
2325 }
2326
2327 static const struct file_operations proc_pid_attr_operations = {
2328         .read           = proc_pid_attr_read,
2329         .write          = proc_pid_attr_write,
2330         .llseek         = generic_file_llseek,
2331 };
2332
2333 static const struct pid_entry attr_dir_stuff[] = {
2334         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2335         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2336         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2337         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2338         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2339         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2340 };
2341
2342 static int proc_attr_dir_readdir(struct file * filp,
2343                              void * dirent, filldir_t filldir)
2344 {
2345         return proc_pident_readdir(filp,dirent,filldir,
2346                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2347 }
2348
2349 static const struct file_operations proc_attr_dir_operations = {
2350         .read           = generic_read_dir,
2351         .readdir        = proc_attr_dir_readdir,
2352         .llseek         = default_llseek,
2353 };
2354
2355 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2356                                 struct dentry *dentry, struct nameidata *nd)
2357 {
2358         return proc_pident_lookup(dir, dentry,
2359                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2360 }
2361
2362 static const struct inode_operations proc_attr_dir_inode_operations = {
2363         .lookup         = proc_attr_dir_lookup,
2364         .getattr        = pid_getattr,
2365         .setattr        = proc_setattr,
2366 };
2367
2368 #endif
2369
2370 #ifdef CONFIG_ELF_CORE
2371 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2372                                          size_t count, loff_t *ppos)
2373 {
2374         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2375         struct mm_struct *mm;
2376         char buffer[PROC_NUMBUF];
2377         size_t len;
2378         int ret;
2379
2380         if (!task)
2381                 return -ESRCH;
2382
2383         ret = 0;
2384         mm = get_task_mm(task);
2385         if (mm) {
2386                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2387                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2388                                 MMF_DUMP_FILTER_SHIFT));
2389                 mmput(mm);
2390                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2391         }
2392
2393         put_task_struct(task);
2394
2395         return ret;
2396 }
2397
2398 static ssize_t proc_coredump_filter_write(struct file *file,
2399                                           const char __user *buf,
2400                                           size_t count,
2401                                           loff_t *ppos)
2402 {
2403         struct task_struct *task;
2404         struct mm_struct *mm;
2405         char buffer[PROC_NUMBUF], *end;
2406         unsigned int val;
2407         int ret;
2408         int i;
2409         unsigned long mask;
2410
2411         ret = -EFAULT;
2412         memset(buffer, 0, sizeof(buffer));
2413         if (count > sizeof(buffer) - 1)
2414                 count = sizeof(buffer) - 1;
2415         if (copy_from_user(buffer, buf, count))
2416                 goto out_no_task;
2417
2418         ret = -EINVAL;
2419         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2420         if (*end == '\n')
2421                 end++;
2422         if (end - buffer == 0)
2423                 goto out_no_task;
2424
2425         ret = -ESRCH;
2426         task = get_proc_task(file->f_dentry->d_inode);
2427         if (!task)
2428                 goto out_no_task;
2429
2430         ret = end - buffer;
2431         mm = get_task_mm(task);
2432         if (!mm)
2433                 goto out_no_mm;
2434
2435         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2436                 if (val & mask)
2437                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2438                 else
2439                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2440         }
2441
2442         mmput(mm);
2443  out_no_mm:
2444         put_task_struct(task);
2445  out_no_task:
2446         return ret;
2447 }
2448
2449 static const struct file_operations proc_coredump_filter_operations = {
2450         .read           = proc_coredump_filter_read,
2451         .write          = proc_coredump_filter_write,
2452         .llseek         = generic_file_llseek,
2453 };
2454 #endif
2455
2456 /*
2457  * /proc/self:
2458  */
2459 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2460                               int buflen)
2461 {
2462         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2463         pid_t tgid = task_tgid_nr_ns(current, ns);
2464         char tmp[PROC_NUMBUF];
2465         if (!tgid)
2466                 return -ENOENT;
2467         sprintf(tmp, "%d", tgid);
2468         return vfs_readlink(dentry,buffer,buflen,tmp);
2469 }
2470
2471 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2472 {
2473         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2474         pid_t tgid = task_tgid_nr_ns(current, ns);
2475         char *name = ERR_PTR(-ENOENT);
2476         if (tgid) {
2477                 name = __getname();
2478                 if (!name)
2479                         name = ERR_PTR(-ENOMEM);
2480                 else
2481                         sprintf(name, "%d", tgid);
2482         }
2483         nd_set_link(nd, name);
2484         return NULL;
2485 }
2486
2487 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2488                                 void *cookie)
2489 {
2490         char *s = nd_get_link(nd);
2491         if (!IS_ERR(s))
2492                 __putname(s);
2493 }
2494
2495 static const struct inode_operations proc_self_inode_operations = {
2496         .readlink       = proc_self_readlink,
2497         .follow_link    = proc_self_follow_link,
2498         .put_link       = proc_self_put_link,
2499 };
2500
2501 /*
2502  * proc base
2503  *
2504  * These are the directory entries in the root directory of /proc
2505  * that properly belong to the /proc filesystem, as they describe
2506  * describe something that is process related.
2507  */
2508 static const struct pid_entry proc_base_stuff[] = {
2509         NOD("self", S_IFLNK|S_IRWXUGO,
2510                 &proc_self_inode_operations, NULL, {}),
2511 };
2512
2513 /*
2514  *      Exceptional case: normally we are not allowed to unhash a busy
2515  * directory. In this case, however, we can do it - no aliasing problems
2516  * due to the way we treat inodes.
2517  */
2518 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2519 {
2520         struct inode *inode = dentry->d_inode;
2521         struct task_struct *task = get_proc_task(inode);
2522         if (task) {
2523                 put_task_struct(task);
2524                 return 1;
2525         }
2526         d_drop(dentry);
2527         return 0;
2528 }
2529
2530 static const struct dentry_operations proc_base_dentry_operations =
2531 {
2532         .d_revalidate   = proc_base_revalidate,
2533         .d_delete       = pid_delete_dentry,
2534 };
2535
2536 static struct dentry *proc_base_instantiate(struct inode *dir,
2537         struct dentry *dentry, struct task_struct *task, const void *ptr)
2538 {
2539         const struct pid_entry *p = ptr;
2540         struct inode *inode;
2541         struct proc_inode *ei;
2542         struct dentry *error;
2543
2544         /* Allocate the inode */
2545         error = ERR_PTR(-ENOMEM);
2546         inode = new_inode(dir->i_sb);
2547         if (!inode)
2548                 goto out;
2549
2550         /* Initialize the inode */
2551         ei = PROC_I(inode);
2552         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2553
2554         /*
2555          * grab the reference to the task.
2556          */
2557         ei->pid = get_task_pid(task, PIDTYPE_PID);
2558         if (!ei->pid)
2559                 goto out_iput;
2560
2561         inode->i_mode = p->mode;
2562         if (S_ISDIR(inode->i_mode))
2563                 inode->i_nlink = 2;
2564         if (S_ISLNK(inode->i_mode))
2565                 inode->i_size = 64;
2566         if (p->iop)
2567                 inode->i_op = p->iop;
2568         if (p->fop)
2569                 inode->i_fop = p->fop;
2570         ei->op = p->op;
2571         dentry->d_op = &proc_base_dentry_operations;
2572         d_add(dentry, inode);
2573         error = NULL;
2574 out:
2575         return error;
2576 out_iput:
2577         iput(inode);
2578         goto out;
2579 }
2580
2581 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2582 {
2583         struct dentry *error;
2584         struct task_struct *task = get_proc_task(dir);
2585         const struct pid_entry *p, *last;
2586
2587         error = ERR_PTR(-ENOENT);
2588
2589         if (!task)
2590                 goto out_no_task;
2591
2592         /* Lookup the directory entry */
2593         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2594         for (p = proc_base_stuff; p <= last; p++) {
2595                 if (p->len != dentry->d_name.len)
2596                         continue;
2597                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2598                         break;
2599         }
2600         if (p > last)
2601                 goto out;
2602
2603         error = proc_base_instantiate(dir, dentry, task, p);
2604
2605 out:
2606         put_task_struct(task);
2607 out_no_task:
2608         return error;
2609 }
2610
2611 static int proc_base_fill_cache(struct file *filp, void *dirent,
2612         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2613 {
2614         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2615                                 proc_base_instantiate, task, p);
2616 }
2617
2618 #ifdef CONFIG_TASK_IO_ACCOUNTING
2619 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2620 {
2621         struct task_io_accounting acct = task->ioac;
2622         unsigned long flags;
2623
2624         if (whole && lock_task_sighand(task, &flags)) {
2625                 struct task_struct *t = task;
2626
2627                 task_io_accounting_add(&acct, &task->signal->ioac);
2628                 while_each_thread(task, t)
2629                         task_io_accounting_add(&acct, &t->ioac);
2630
2631                 unlock_task_sighand(task, &flags);
2632         }
2633         return sprintf(buffer,
2634                         "rchar: %llu\n"
2635                         "wchar: %llu\n"
2636                         "syscr: %llu\n"
2637                         "syscw: %llu\n"
2638                         "read_bytes: %llu\n"
2639                         "write_bytes: %llu\n"
2640                         "cancelled_write_bytes: %llu\n",
2641                         (unsigned long long)acct.rchar,
2642                         (unsigned long long)acct.wchar,
2643                         (unsigned long long)acct.syscr,
2644                         (unsigned long long)acct.syscw,
2645                         (unsigned long long)acct.read_bytes,
2646                         (unsigned long long)acct.write_bytes,
2647                         (unsigned long long)acct.cancelled_write_bytes);
2648 }
2649
2650 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2651 {
2652         return do_io_accounting(task, buffer, 0);
2653 }
2654
2655 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2656 {
2657         return do_io_accounting(task, buffer, 1);
2658 }
2659 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2660
2661 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2662                                 struct pid *pid, struct task_struct *task)
2663 {
2664         seq_printf(m, "%08x\n", task->personality);
2665         return 0;
2666 }
2667
2668 /*
2669  * Thread groups
2670  */
2671 static const struct file_operations proc_task_operations;
2672 static const struct inode_operations proc_task_inode_operations;
2673
2674 static const struct pid_entry tgid_base_stuff[] = {
2675         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2676         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2677         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2678 #ifdef CONFIG_NET
2679         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2680 #endif
2681         REG("environ",    S_IRUSR, proc_environ_operations),
2682         INF("auxv",       S_IRUSR, proc_pid_auxv),
2683         ONE("status",     S_IRUGO, proc_pid_status),
2684         ONE("personality", S_IRUSR, proc_pid_personality),
2685         INF("limits",     S_IRUGO, proc_pid_limits),
2686 #ifdef CONFIG_SCHED_DEBUG
2687         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2688 #endif
2689         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2690 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2691         INF("syscall",    S_IRUSR, proc_pid_syscall),
2692 #endif
2693         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2694         ONE("stat",       S_IRUGO, proc_tgid_stat),
2695         ONE("statm",      S_IRUGO, proc_pid_statm),
2696         REG("maps",       S_IRUGO, proc_maps_operations),
2697 #ifdef CONFIG_NUMA
2698         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2699 #endif
2700         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2701         LNK("cwd",        proc_cwd_link),
2702         LNK("root",       proc_root_link),
2703         LNK("exe",        proc_exe_link),
2704         REG("mounts",     S_IRUGO, proc_mounts_operations),
2705         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2706         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2707 #ifdef CONFIG_PROC_PAGE_MONITOR
2708         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2709         REG("smaps",      S_IRUGO, proc_smaps_operations),
2710         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2711 #endif
2712 #ifdef CONFIG_SECURITY
2713         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2714 #endif
2715 #ifdef CONFIG_KALLSYMS
2716         INF("wchan",      S_IRUGO, proc_pid_wchan),
2717 #endif
2718 #ifdef CONFIG_STACKTRACE
2719         ONE("stack",      S_IRUSR, proc_pid_stack),
2720 #endif
2721 #ifdef CONFIG_SCHEDSTATS
2722         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2723 #endif
2724 #ifdef CONFIG_LATENCYTOP
2725         REG("latency",  S_IRUGO, proc_lstats_operations),
2726 #endif
2727 #ifdef CONFIG_PROC_PID_CPUSET
2728         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2729 #endif
2730 #ifdef CONFIG_CGROUPS
2731         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2732 #endif
2733         INF("oom_score",  S_IRUGO, proc_oom_score),
2734         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2735         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2736 #ifdef CONFIG_AUDITSYSCALL
2737         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2738         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2739 #endif
2740 #ifdef CONFIG_FAULT_INJECTION
2741         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2742 #endif
2743 #ifdef CONFIG_ELF_CORE
2744         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2745 #endif
2746 #ifdef CONFIG_TASK_IO_ACCOUNTING
2747         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2748 #endif
2749 };
2750
2751 static int proc_tgid_base_readdir(struct file * filp,
2752                              void * dirent, filldir_t filldir)
2753 {
2754         return proc_pident_readdir(filp,dirent,filldir,
2755                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2756 }
2757
2758 static const struct file_operations proc_tgid_base_operations = {
2759         .read           = generic_read_dir,
2760         .readdir        = proc_tgid_base_readdir,
2761         .llseek         = default_llseek,
2762 };
2763
2764 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2765         return proc_pident_lookup(dir, dentry,
2766                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2767 }
2768
2769 static const struct inode_operations proc_tgid_base_inode_operations = {
2770         .lookup         = proc_tgid_base_lookup,
2771         .getattr        = pid_getattr,
2772         .setattr        = proc_setattr,
2773 };
2774
2775 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2776 {
2777         struct dentry *dentry, *leader, *dir;
2778         char buf[PROC_NUMBUF];
2779         struct qstr name;
2780
2781         name.name = buf;
2782         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2783         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2784         if (dentry) {
2785                 shrink_dcache_parent(dentry);
2786                 d_drop(dentry);
2787                 dput(dentry);
2788         }
2789
2790         name.name = buf;
2791         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2792         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2793         if (!leader)
2794                 goto out;
2795
2796         name.name = "task";
2797         name.len = strlen(name.name);
2798         dir = d_hash_and_lookup(leader, &name);
2799         if (!dir)
2800                 goto out_put_leader;
2801
2802         name.name = buf;
2803         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2804         dentry = d_hash_and_lookup(dir, &name);
2805         if (dentry) {
2806                 shrink_dcache_parent(dentry);
2807                 d_drop(dentry);
2808                 dput(dentry);
2809         }
2810
2811         dput(dir);
2812 out_put_leader:
2813         dput(leader);
2814 out:
2815         return;
2816 }
2817
2818 /**
2819  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2820  * @task: task that should be flushed.
2821  *
2822  * When flushing dentries from proc, one needs to flush them from global
2823  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2824  * in. This call is supposed to do all of this job.
2825  *
2826  * Looks in the dcache for
2827  * /proc/@pid
2828  * /proc/@tgid/task/@pid
2829  * if either directory is present flushes it and all of it'ts children
2830  * from the dcache.
2831  *
2832  * It is safe and reasonable to cache /proc entries for a task until
2833  * that task exits.  After that they just clog up the dcache with
2834  * useless entries, possibly causing useful dcache entries to be
2835  * flushed instead.  This routine is proved to flush those useless
2836  * dcache entries at process exit time.
2837  *
2838  * NOTE: This routine is just an optimization so it does not guarantee
2839  *       that no dcache entries will exist at process exit time it
2840  *       just makes it very unlikely that any will persist.
2841  */
2842
2843 void proc_flush_task(struct task_struct *task)
2844 {
2845         int i;
2846         struct pid *pid, *tgid;
2847         struct upid *upid;
2848
2849         pid = task_pid(task);
2850         tgid = task_tgid(task);
2851
2852         for (i = 0; i <= pid->level; i++) {
2853                 upid = &pid->numbers[i];
2854                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2855                                         tgid->numbers[i].nr);
2856         }
2857
2858         upid = &pid->numbers[pid->level];
2859         if (upid->nr == 1)
2860                 pid_ns_release_proc(upid->ns);
2861 }
2862
2863 static struct dentry *proc_pid_instantiate(struct inode *dir,
2864                                            struct dentry * dentry,
2865                                            struct task_struct *task, const void *ptr)
2866 {
2867         struct dentry *error = ERR_PTR(-ENOENT);
2868         struct inode *inode;
2869
2870         inode = proc_pid_make_inode(dir->i_sb, task);
2871         if (!inode)
2872                 goto out;
2873
2874         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2875         inode->i_op = &proc_tgid_base_inode_operations;
2876         inode->i_fop = &proc_tgid_base_operations;
2877         inode->i_flags|=S_IMMUTABLE;
2878
2879         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2880                 ARRAY_SIZE(tgid_base_stuff));
2881
2882         dentry->d_op = &pid_dentry_operations;
2883
2884         d_add(dentry, inode);
2885         /* Close the race of the process dying before we return the dentry */
2886         if (pid_revalidate(dentry, NULL))
2887                 error = NULL;
2888 out:
2889         return error;
2890 }
2891
2892 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2893 {
2894         struct dentry *result;
2895         struct task_struct *task;
2896         unsigned tgid;
2897         struct pid_namespace *ns;
2898
2899         result = proc_base_lookup(dir, dentry);
2900         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2901                 goto out;
2902
2903         tgid = name_to_int(dentry);
2904         if (tgid == ~0U)
2905                 goto out;
2906
2907         ns = dentry->d_sb->s_fs_info;
2908         rcu_read_lock();
2909         task = find_task_by_pid_ns(tgid, ns);
2910         if (task)
2911                 get_task_struct(task);
2912         rcu_read_unlock();
2913         if (!task)
2914                 goto out;
2915
2916         result = proc_pid_instantiate(dir, dentry, task, NULL);
2917         put_task_struct(task);
2918 out:
2919         return result;
2920 }
2921
2922 /*
2923  * Find the first task with tgid >= tgid
2924  *
2925  */
2926 struct tgid_iter {
2927         unsigned int tgid;
2928         struct task_struct *task;
2929 };
2930 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2931 {
2932         struct pid *pid;
2933
2934         if (iter.task)
2935                 put_task_struct(iter.task);
2936         rcu_read_lock();
2937 retry:
2938         iter.task = NULL;
2939         pid = find_ge_pid(iter.tgid, ns);
2940         if (pid) {
2941                 iter.tgid = pid_nr_ns(pid, ns);
2942                 iter.task = pid_task(pid, PIDTYPE_PID);
2943                 /* What we to know is if the pid we have find is the
2944                  * pid of a thread_group_leader.  Testing for task
2945                  * being a thread_group_leader is the obvious thing
2946                  * todo but there is a window when it fails, due to
2947                  * the pid transfer logic in de_thread.
2948                  *
2949                  * So we perform the straight forward test of seeing
2950                  * if the pid we have found is the pid of a thread
2951                  * group leader, and don't worry if the task we have
2952                  * found doesn't happen to be a thread group leader.
2953                  * As we don't care in the case of readdir.
2954                  */
2955                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2956                         iter.tgid += 1;
2957                         goto retry;
2958                 }
2959                 get_task_struct(iter.task);
2960         }
2961         rcu_read_unlock();
2962         return iter;
2963 }
2964
2965 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2966
2967 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2968         struct tgid_iter iter)
2969 {
2970         char name[PROC_NUMBUF];
2971         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2972         return proc_fill_cache(filp, dirent, filldir, name, len,
2973                                 proc_pid_instantiate, iter.task, NULL);
2974 }
2975
2976 /* for the /proc/ directory itself, after non-process stuff has been done */
2977 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2978 {
2979         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2980         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2981         struct tgid_iter iter;
2982         struct pid_namespace *ns;
2983
2984         if (!reaper)
2985                 goto out_no_task;
2986
2987         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2988                 const struct pid_entry *p = &proc_base_stuff[nr];
2989                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2990                         goto out;
2991         }
2992
2993         ns = filp->f_dentry->d_sb->s_fs_info;
2994         iter.task = NULL;
2995         iter.tgid = filp->f_pos - TGID_OFFSET;
2996         for (iter = next_tgid(ns, iter);
2997              iter.task;
2998              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2999                 filp->f_pos = iter.tgid + TGID_OFFSET;
3000                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3001                         put_task_struct(iter.task);
3002                         goto out;
3003                 }
3004         }
3005         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3006 out:
3007         put_task_struct(reaper);
3008 out_no_task:
3009         return 0;
3010 }
3011
3012 /*
3013  * Tasks
3014  */
3015 static const struct pid_entry tid_base_stuff[] = {
3016         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3017         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3018         REG("environ",   S_IRUSR, proc_environ_operations),
3019         INF("auxv",      S_IRUSR, proc_pid_auxv),
3020         ONE("status",    S_IRUGO, proc_pid_status),
3021         ONE("personality", S_IRUSR, proc_pid_personality),
3022         INF("limits",    S_IRUGO, proc_pid_limits),
3023 #ifdef CONFIG_SCHED_DEBUG
3024         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3025 #endif
3026         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3027 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3028         INF("syscall",   S_IRUSR, proc_pid_syscall),
3029 #endif
3030         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3031         ONE("stat",      S_IRUGO, proc_tid_stat),
3032         ONE("statm",     S_IRUGO, proc_pid_statm),
3033         REG("maps",      S_IRUGO, proc_maps_operations),
3034 #ifdef CONFIG_NUMA
3035         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3036 #endif
3037         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3038         LNK("cwd",       proc_cwd_link),
3039         LNK("root",      proc_root_link),
3040         LNK("exe",       proc_exe_link),
3041         REG("mounts",    S_IRUGO, proc_mounts_operations),
3042         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3043 #ifdef CONFIG_PROC_PAGE_MONITOR
3044         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3045         REG("smaps",     S_IRUGO, proc_smaps_operations),
3046         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3047 #endif
3048 #ifdef CONFIG_SECURITY
3049         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3050 #endif
3051 #ifdef CONFIG_KALLSYMS
3052         INF("wchan",     S_IRUGO, proc_pid_wchan),
3053 #endif
3054 #ifdef CONFIG_STACKTRACE
3055         ONE("stack",      S_IRUSR, proc_pid_stack),
3056 #endif
3057 #ifdef CONFIG_SCHEDSTATS
3058         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3059 #endif
3060 #ifdef CONFIG_LATENCYTOP
3061         REG("latency",  S_IRUGO, proc_lstats_operations),
3062 #endif
3063 #ifdef CONFIG_PROC_PID_CPUSET
3064         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3065 #endif
3066 #ifdef CONFIG_CGROUPS
3067         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3068 #endif
3069         INF("oom_score", S_IRUGO, proc_oom_score),
3070         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3071         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3072 #ifdef CONFIG_AUDITSYSCALL
3073         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3074         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3075 #endif
3076 #ifdef CONFIG_FAULT_INJECTION
3077         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3078 #endif
3079 #ifdef CONFIG_TASK_IO_ACCOUNTING
3080         INF("io",       S_IRUGO, proc_tid_io_accounting),
3081 #endif
3082 };
3083
3084 static int proc_tid_base_readdir(struct file * filp,
3085                              void * dirent, filldir_t filldir)
3086 {
3087         return proc_pident_readdir(filp,dirent,filldir,
3088                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3089 }
3090
3091 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3092         return proc_pident_lookup(dir, dentry,
3093                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3094 }
3095
3096 static const struct file_operations proc_tid_base_operations = {
3097         .read           = generic_read_dir,
3098         .readdir        = proc_tid_base_readdir,
3099         .llseek         = default_llseek,
3100 };
3101
3102 static const struct inode_operations proc_tid_base_inode_operations = {
3103         .lookup         = proc_tid_base_lookup,
3104         .getattr        = pid_getattr,
3105         .setattr        = proc_setattr,
3106 };
3107
3108 static struct dentry *proc_task_instantiate(struct inode *dir,
3109         struct dentry *dentry, struct task_struct *task, const void *ptr)
3110 {
3111         struct dentry *error = ERR_PTR(-ENOENT);
3112         struct inode *inode;
3113         inode = proc_pid_make_inode(dir->i_sb, task);
3114
3115         if (!inode)
3116                 goto out;
3117         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3118         inode->i_op = &proc_tid_base_inode_operations;
3119         inode->i_fop = &proc_tid_base_operations;
3120         inode->i_flags|=S_IMMUTABLE;
3121
3122         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3123                 ARRAY_SIZE(tid_base_stuff));
3124
3125         dentry->d_op = &pid_dentry_operations;
3126
3127         d_add(dentry, inode);
3128         /* Close the race of the process dying before we return the dentry */
3129         if (pid_revalidate(dentry, NULL))
3130                 error = NULL;
3131 out:
3132         return error;
3133 }
3134
3135 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3136 {
3137         struct dentry *result = ERR_PTR(-ENOENT);
3138         struct task_struct *task;
3139         struct task_struct *leader = get_proc_task(dir);
3140         unsigned tid;
3141         struct pid_namespace *ns;
3142
3143         if (!leader)
3144                 goto out_no_task;
3145
3146         tid = name_to_int(dentry);
3147         if (tid == ~0U)
3148                 goto out;
3149
3150         ns = dentry->d_sb->s_fs_info;
3151         rcu_read_lock();
3152         task = find_task_by_pid_ns(tid, ns);
3153         if (task)
3154                 get_task_struct(task);
3155         rcu_read_unlock();
3156         if (!task)
3157                 goto out;
3158         if (!same_thread_group(leader, task))
3159                 goto out_drop_task;
3160
3161         result = proc_task_instantiate(dir, dentry, task, NULL);
3162 out_drop_task:
3163         put_task_struct(task);
3164 out:
3165         put_task_struct(leader);
3166 out_no_task:
3167         return result;
3168 }
3169
3170 /*
3171  * Find the first tid of a thread group to return to user space.
3172  *
3173  * Usually this is just the thread group leader, but if the users
3174  * buffer was too small or there was a seek into the middle of the
3175  * directory we have more work todo.
3176  *
3177  * In the case of a short read we start with find_task_by_pid.
3178  *
3179  * In the case of a seek we start with the leader and walk nr
3180  * threads past it.
3181  */
3182 static struct task_struct *first_tid(struct task_struct *leader,
3183                 int tid, int nr, struct pid_namespace *ns)
3184 {
3185         struct task_struct *pos;
3186
3187         rcu_read_lock();
3188         /* Attempt to start with the pid of a thread */
3189         if (tid && (nr > 0)) {
3190                 pos = find_task_by_pid_ns(tid, ns);
3191                 if (pos && (pos->group_leader == leader))
3192                         goto found;
3193         }
3194
3195         /* If nr exceeds the number of threads there is nothing todo */
3196         pos = NULL;
3197         if (nr && nr >= get_nr_threads(leader))
3198                 goto out;
3199
3200         /* If we haven't found our starting place yet start
3201          * with the leader and walk nr threads forward.
3202          */
3203         for (pos = leader; nr > 0; --nr) {
3204                 pos = next_thread(pos);
3205                 if (pos == leader) {
3206                         pos = NULL;
3207                         goto out;
3208                 }
3209         }
3210 found:
3211         get_task_struct(pos);
3212 out:
3213         rcu_read_unlock();
3214         return pos;
3215 }
3216
3217 /*
3218  * Find the next thread in the thread list.
3219  * Return NULL if there is an error or no next thread.
3220  *
3221  * The reference to the input task_struct is released.
3222  */
3223 static struct task_struct *next_tid(struct task_struct *start)
3224 {
3225         struct task_struct *pos = NULL;
3226         rcu_read_lock();
3227         if (pid_alive(start)) {
3228                 pos = next_thread(start);
3229                 if (thread_group_leader(pos))
3230                         pos = NULL;
3231                 else
3232                         get_task_struct(pos);
3233         }
3234         rcu_read_unlock();
3235         put_task_struct(start);
3236         return pos;
3237 }
3238
3239 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3240         struct task_struct *task, int tid)
3241 {
3242         char name[PROC_NUMBUF];
3243         int len = snprintf(name, sizeof(name), "%d", tid);
3244         return proc_fill_cache(filp, dirent, filldir, name, len,
3245                                 proc_task_instantiate, task, NULL);
3246 }
3247
3248 /* for the /proc/TGID/task/ directories */
3249 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3250 {
3251         struct dentry *dentry = filp->f_path.dentry;
3252         struct inode *inode = dentry->d_inode;
3253         struct task_struct *leader = NULL;
3254         struct task_struct *task;
3255         int retval = -ENOENT;
3256         ino_t ino;
3257         int tid;
3258         struct pid_namespace *ns;
3259
3260         task = get_proc_task(inode);
3261         if (!task)
3262                 goto out_no_task;
3263         rcu_read_lock();
3264         if (pid_alive(task)) {
3265                 leader = task->group_leader;
3266                 get_task_struct(leader);
3267         }
3268         rcu_read_unlock();
3269         put_task_struct(task);
3270         if (!leader)
3271                 goto out_no_task;
3272         retval = 0;
3273
3274         switch ((unsigned long)filp->f_pos) {
3275         case 0:
3276                 ino = inode->i_ino;
3277                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3278                         goto out;
3279                 filp->f_pos++;
3280                 /* fall through */
3281         case 1:
3282                 ino = parent_ino(dentry);
3283                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3284                         goto out;
3285                 filp->f_pos++;
3286                 /* fall through */
3287         }
3288
3289         /* f_version caches the tgid value that the last readdir call couldn't
3290          * return. lseek aka telldir automagically resets f_version to 0.
3291          */
3292         ns = filp->f_dentry->d_sb->s_fs_info;
3293         tid = (int)filp->f_version;
3294         filp->f_version = 0;
3295         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3296              task;
3297              task = next_tid(task), filp->f_pos++) {
3298                 tid = task_pid_nr_ns(task, ns);
3299                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3300                         /* returning this tgid failed, save it as the first
3301                          * pid for the next readir call */
3302                         filp->f_version = (u64)tid;
3303                         put_task_struct(task);
3304                         break;
3305                 }
3306         }
3307 out:
3308         put_task_struct(leader);
3309 out_no_task:
3310         return retval;
3311 }
3312
3313 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3314 {
3315         struct inode *inode = dentry->d_inode;
3316         struct task_struct *p = get_proc_task(inode);
3317         generic_fillattr(inode, stat);
3318
3319         if (p) {
3320                 stat->nlink += get_nr_threads(p);
3321                 put_task_struct(p);
3322         }
3323
3324         return 0;
3325 }
3326
3327 static const struct inode_operations proc_task_inode_operations = {
3328         .lookup         = proc_task_lookup,
3329         .getattr        = proc_task_getattr,
3330         .setattr        = proc_setattr,
3331 };
3332
3333 static const struct file_operations proc_task_operations = {
3334         .read           = generic_read_dir,
3335         .readdir        = proc_task_readdir,
3336         .llseek         = default_llseek,
3337 };