fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #include "internal.h"
  87
  88 /* NOTE:
  89  *      Implementing inode permission operations in /proc is almost
  90  *      certainly an error.  Permission checks need to happen during
  91  *      each system call not at open time.  The reason is that most of
  92  *      what we wish to check for permissions in /proc varies at runtime.
  93  *
  94  *      The classic example of a problem is opening file descriptors
  95  *      in /proc for a task before it execs a suid executable.
  96  */
  97
  98 struct pid_entry {
  99         char *name;
 100         int len;
 101         mode_t mode;
 102         const struct inode_operations *iop;
 103         const struct file_operations *fop;
 104         union proc_op op;
 105 };
 106
 107 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 108         .name = (NAME),                                 \
 109         .len  = sizeof(NAME) - 1,                       \
 110         .mode = MODE,                                   \
 111         .iop  = IOP,                                    \
 112         .fop  = FOP,                                    \
 113         .op   = OP,                                     \
 114 }
 115
 116 #define DIR(NAME, MODE, iops, fops)     \
 117         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 118 #define LNK(NAME, get_link)                                     \
 119         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 120                 &proc_pid_link_inode_operations, NULL,          \
 121                 { .proc_get_link = get_link } )
 122 #define REG(NAME, MODE, fops)                           \
 123         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 124 #define INF(NAME, MODE, read)                           \
 125         NOD(NAME, (S_IFREG|(MODE)),                     \
 126                 NULL, &proc_info_file_operations,       \
 127                 { .proc_read = read } )
 128 #define ONE(NAME, MODE, show)                           \
 129         NOD(NAME, (S_IFREG|(MODE)),                     \
 130                 NULL, &proc_single_file_operations,     \
 131                 { .proc_show = show } )
 132
 133 /*
 134  * Count the number of hardlinks for the pid_entry table, excluding the .
 135  * and .. links.
 136  */
 137 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 138         unsigned int n)
 139 {
 140         unsigned int i;
 141         unsigned int count;
 142
 143         count = 0;
 144         for (i = 0; i < n; ++i) {
 145                 if (S_ISDIR(entries[i].mode))
 146                         ++count;
 147         }
 148
 149         return count;
 150 }
 151
 152 static int get_task_root(struct task_struct *task, struct path *root)
 153 {
 154         int result = -ENOENT;
 155
 156         task_lock(task);
 157         if (task->fs) {
 158                 get_fs_root(task->fs, root);
 159                 result = 0;
 160         }
 161         task_unlock(task);
 162         return result;
 163 }
 164
 165 static int proc_cwd_link(struct inode *inode, struct path *path)
 166 {
 167         struct task_struct *task = get_proc_task(inode);
 168         int result = -ENOENT;
 169
 170         if (task) {
 171                 task_lock(task);
 172                 if (task->fs) {
 173                         get_fs_pwd(task->fs, path);
 174                         result = 0;
 175                 }
 176                 task_unlock(task);
 177                 put_task_struct(task);
 178         }
 179         return result;
 180 }
 181
 182 static int proc_root_link(struct inode *inode, struct path *path)
 183 {
 184         struct task_struct *task = get_proc_task(inode);
 185         int result = -ENOENT;
 186
 187         if (task) {
 188                 result = get_task_root(task, path);
 189                 put_task_struct(task);
 190         }
 191         return result;
 192 }
 193
 194 /*
 195  * Return zero if current may access user memory in @task, -error if not.
 196  */
 197 static int check_mem_permission(struct task_struct *task)
 198 {
 199         /*
 200          * A task can always look at itself, in case it chooses
 201          * to use system calls instead of load instructions.
 202          */
 203         if (task == current)
 204                 return 0;
 205
 206         /*
 207          * If current is actively ptrace'ing, and would also be
 208          * permitted to freshly attach with ptrace now, permit it.
 209          */
 210         if (task_is_stopped_or_traced(task)) {
 211                 int match;
 212                 rcu_read_lock();
 213                 match = (tracehook_tracer_task(task) == current);
 214                 rcu_read_unlock();
 215                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 216                         return 0;
 217         }
 218
 219         /*
 220          * Noone else is allowed.
 221          */
 222         return -EPERM;
 223 }
 224
 225 struct mm_struct *mm_for_maps(struct task_struct *task)
 226 {
 227         struct mm_struct *mm;
 228
 229         if (mutex_lock_killable(&task->cred_guard_mutex))
 230                 return NULL;
 231
 232         mm = get_task_mm(task);
 233         if (mm && mm != current->mm &&
 234                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 235                 mmput(mm);
 236                 mm = NULL;
 237         }
 238         mutex_unlock(&task->cred_guard_mutex);
 239
 240         return mm;
 241 }
 242
 243 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 244 {
 245         int res = 0;
 246         unsigned int len;
 247         struct mm_struct *mm = get_task_mm(task);
 248         if (!mm)
 249                 goto out;
 250         if (!mm->arg_end)
 251                 goto out_mm;    /* Shh! No looking before we're done */
 252
 253         len = mm->arg_end - mm->arg_start;
 254
 255         if (len > PAGE_SIZE)
 256                 len = PAGE_SIZE;
 257
 258         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 259
 260         // If the nul at the end of args has been overwritten, then
 261         // assume application is using setproctitle(3).
 262         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 263                 len = strnlen(buffer, res);
 264                 if (len < res) {
 265                     res = len;
 266                 } else {
 267                         len = mm->env_end - mm->env_start;
 268                         if (len > PAGE_SIZE - res)
 269                                 len = PAGE_SIZE - res;
 270                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 271                         res = strnlen(buffer, res);
 272                 }
 273         }
 274 out_mm:
 275         mmput(mm);
 276 out:
 277         return res;
 278 }
 279
 280 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 281 {
 282         int res = 0;
 283         struct mm_struct *mm = get_task_mm(task);
 284         if (mm) {
 285                 unsigned int nwords = 0;
 286                 do {
 287                         nwords += 2;
 288                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 289                 res = nwords * sizeof(mm->saved_auxv[0]);
 290                 if (res > PAGE_SIZE)
 291                         res = PAGE_SIZE;
 292                 memcpy(buffer, mm->saved_auxv, res);
 293                 mmput(mm);
 294         }
 295         return res;
 296 }
 297
 298
 299 #ifdef CONFIG_KALLSYMS
 300 /*
 301  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 302  * Returns the resolved symbol.  If that fails, simply return the address.
 303  */
 304 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 305 {
 306         unsigned long wchan;
 307         char symname[KSYM_NAME_LEN];
 308
 309         wchan = get_wchan(task);
 310
 311         if (lookup_symbol_name(wchan, symname) < 0)
 312                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 313                         return 0;
 314                 else
 315                         return sprintf(buffer, "%lu", wchan);
 316         else
 317                 return sprintf(buffer, "%s", symname);
 318 }
 319 #endif /* CONFIG_KALLSYMS */
 320
 321 #ifdef CONFIG_STACKTRACE
 322
 323 #define MAX_STACK_TRACE_DEPTH   64
 324
 325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 326                           struct pid *pid, struct task_struct *task)
 327 {
 328         struct stack_trace trace;
 329         unsigned long *entries;
 330         int i;
 331
 332         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 333         if (!entries)
 334                 return -ENOMEM;
 335
 336         trace.nr_entries        = 0;
 337         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 338         trace.entries           = entries;
 339         trace.skip              = 0;
 340         save_stack_trace_tsk(task, &trace);
 341
 342         for (i = 0; i < trace.nr_entries; i++) {
 343                 seq_printf(m, "[<%p>] %pS\n",
 344                            (void *)entries[i], (void *)entries[i]);
 345         }
 346         kfree(entries);
 347
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_SCHEDSTATS
 353 /*
 354  * Provides /proc/PID/schedstat
 355  */
 356 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 357 {
 358         return sprintf(buffer, "%llu %llu %lu\n",
 359                         (unsigned long long)task->se.sum_exec_runtime,
 360                         (unsigned long long)task->sched_info.run_delay,
 361                         task->sched_info.pcount);
 362 }
 363 #endif
 364
 365 #ifdef CONFIG_LATENCYTOP
 366 static int lstats_show_proc(struct seq_file *m, void *v)
 367 {
 368         int i;
 369         struct inode *inode = m->private;
 370         struct task_struct *task = get_proc_task(inode);
 371
 372         if (!task)
 373                 return -ESRCH;
 374         seq_puts(m, "Latency Top version : v0.1\n");
 375         for (i = 0; i < 32; i++) {
 376                 if (task->latency_record[i].backtrace[0]) {
 377                         int q;
 378                         seq_printf(m, "%i %li %li ",
 379                                 task->latency_record[i].count,
 380                                 task->latency_record[i].time,
 381                                 task->latency_record[i].max);
 382                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 383                                 char sym[KSYM_SYMBOL_LEN];
 384                                 char *c;
 385                                 if (!task->latency_record[i].backtrace[q])
 386                                         break;
 387                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 388                                         break;
 389                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 390                                 c = strchr(sym, '+');
 391                                 if (c)
 392                                         *c = 0;
 393                                 seq_printf(m, "%s ", sym);
 394                         }
 395                         seq_printf(m, "\n");
 396                 }
 397
 398         }
 399         put_task_struct(task);
 400         return 0;
 401 }
 402
 403 static int lstats_open(struct inode *inode, struct file *file)
 404 {
 405         return single_open(file, lstats_show_proc, inode);
 406 }
 407
 408 static ssize_t lstats_write(struct file *file, const char __user *buf,
 409                             size_t count, loff_t *offs)
 410 {
 411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 412
 413         if (!task)
 414                 return -ESRCH;
 415         clear_all_latency_tracing(task);
 416         put_task_struct(task);
 417
 418         return count;
 419 }
 420
 421 static const struct file_operations proc_lstats_operations = {
 422         .open           = lstats_open,
 423         .read           = seq_read,
 424         .write          = lstats_write,
 425         .llseek         = seq_lseek,
 426         .release        = single_release,
 427 };
 428
 429 #endif
 430
 431 static int proc_oom_score(struct task_struct *task, char *buffer)
 432 {
 433         unsigned long points = 0;
 434
 435         read_lock(&tasklist_lock);
 436         if (pid_alive(task))
 437                 points = oom_badness(task, NULL, NULL,
 438                                         totalram_pages + total_swap_pages);
 439         read_unlock(&tasklist_lock);
 440         return sprintf(buffer, "%lu\n", points);
 441 }
 442
 443 struct limit_names {
 444         char *name;
 445         char *unit;
 446 };
 447
 448 static const struct limit_names lnames[RLIM_NLIMITS] = {
 449         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 450         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 451         [RLIMIT_DATA] = {"Max data size", "bytes"},
 452         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 453         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 454         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 455         [RLIMIT_NPROC] = {"Max processes", "processes"},
 456         [RLIMIT_NOFILE] = {"Max open files", "files"},
 457         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 458         [RLIMIT_AS] = {"Max address space", "bytes"},
 459         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 460         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 461         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 462         [RLIMIT_NICE] = {"Max nice priority", NULL},
 463         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 464         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 465 };
 466
 467 /* Display limits for a process */
 468 static int proc_pid_limits(struct task_struct *task, char *buffer)
 469 {
 470         unsigned int i;
 471         int count = 0;
 472         unsigned long flags;
 473         char *bufptr = buffer;
 474
 475         struct rlimit rlim[RLIM_NLIMITS];
 476
 477         if (!lock_task_sighand(task, &flags))
 478                 return 0;
 479         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 480         unlock_task_sighand(task, &flags);
 481
 482         /*
 483          * print the file header
 484          */
 485         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 486                         "Limit", "Soft Limit", "Hard Limit", "Units");
 487
 488         for (i = 0; i < RLIM_NLIMITS; i++) {
 489                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 490                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 491                                          lnames[i].name, "unlimited");
 492                 else
 493                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 494                                          lnames[i].name, rlim[i].rlim_cur);
 495
 496                 if (rlim[i].rlim_max == RLIM_INFINITY)
 497                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 498                 else
 499                         count += sprintf(&bufptr[count], "%-20lu ",
 500                                          rlim[i].rlim_max);
 501
 502                 if (lnames[i].unit)
 503                         count += sprintf(&bufptr[count], "%-10s\n",
 504                                          lnames[i].unit);
 505                 else
 506                         count += sprintf(&bufptr[count], "\n");
 507         }
 508
 509         return count;
 510 }
 511
 512 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 513 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 514 {
 515         long nr;
 516         unsigned long args[6], sp, pc;
 517
 518         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 519                 return sprintf(buffer, "running\n");
 520
 521         if (nr < 0)
 522                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 523
 524         return sprintf(buffer,
 525                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 526                        nr,
 527                        args[0], args[1], args[2], args[3], args[4], args[5],
 528                        sp, pc);
 529 }
 530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 531
 532 /************************************************************************/
 533 /*                       Here the fs part begins                        */
 534 /************************************************************************/
 535
 536 /* permission checks */
 537 static int proc_fd_access_allowed(struct inode *inode)
 538 {
 539         struct task_struct *task;
 540         int allowed = 0;
 541         /* Allow access to a task's file descriptors if it is us or we
 542          * may use ptrace attach to the process and find out that
 543          * information.
 544          */
 545         task = get_proc_task(inode);
 546         if (task) {
 547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 548                 put_task_struct(task);
 549         }
 550         return allowed;
 551 }
 552
 553 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 554 {
 555         int error;
 556         struct inode *inode = dentry->d_inode;
 557
 558         if (attr->ia_valid & ATTR_MODE)
 559                 return -EPERM;
 560
 561         error = inode_change_ok(inode, attr);
 562         if (error)
 563                 return error;
 564
 565         if ((attr->ia_valid & ATTR_SIZE) &&
 566             attr->ia_size != i_size_read(inode)) {
 567                 error = vmtruncate(inode, attr->ia_size);
 568                 if (error)
 569                         return error;
 570         }
 571
 572         setattr_copy(inode, attr);
 573         mark_inode_dirty(inode);
 574         return 0;
 575 }
 576
 577 static const struct inode_operations proc_def_inode_operations = {
 578         .setattr        = proc_setattr,
 579 };
 580
 581 static int mounts_open_common(struct inode *inode, struct file *file,
 582                               const struct seq_operations *op)
 583 {
 584         struct task_struct *task = get_proc_task(inode);
 585         struct nsproxy *nsp;
 586         struct mnt_namespace *ns = NULL;
 587         struct path root;
 588         struct proc_mounts *p;
 589         int ret = -EINVAL;
 590
 591         if (task) {
 592                 rcu_read_lock();
 593                 nsp = task_nsproxy(task);
 594                 if (nsp) {
 595                         ns = nsp->mnt_ns;
 596                         if (ns)
 597                                 get_mnt_ns(ns);
 598                 }
 599                 rcu_read_unlock();
 600                 if (ns && get_task_root(task, &root) == 0)
 601                         ret = 0;
 602                 put_task_struct(task);
 603         }
 604
 605         if (!ns)
 606                 goto err;
 607         if (ret)
 608                 goto err_put_ns;
 609
 610         ret = -ENOMEM;
 611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 612         if (!p)
 613                 goto err_put_path;
 614
 615         file->private_data = &p->m;
 616         ret = seq_open(file, op);
 617         if (ret)
 618                 goto err_free;
 619
 620         p->m.private = p;
 621         p->ns = ns;
 622         p->root = root;
 623         p->event = ns->event;
 624
 625         return 0;
 626
 627  err_free:
 628         kfree(p);
 629  err_put_path:
 630         path_put(&root);
 631  err_put_ns:
 632         put_mnt_ns(ns);
 633  err:
 634         return ret;
 635 }
 636
 637 static int mounts_release(struct inode *inode, struct file *file)
 638 {
 639         struct proc_mounts *p = file->private_data;
 640         path_put(&p->root);
 641         put_mnt_ns(p->ns);
 642         return seq_release(inode, file);
 643 }
 644
 645 static unsigned mounts_poll(struct file *file, poll_table *wait)
 646 {
 647         struct proc_mounts *p = file->private_data;
 648         unsigned res = POLLIN | POLLRDNORM;
 649
 650         poll_wait(file, &p->ns->poll, wait);
 651         if (mnt_had_events(p))
 652                 res |= POLLERR | POLLPRI;
 653
 654         return res;
 655 }
 656
 657 static int mounts_open(struct inode *inode, struct file *file)
 658 {
 659         return mounts_open_common(inode, file, &mounts_op);
 660 }
 661
 662 static const struct file_operations proc_mounts_operations = {
 663         .open           = mounts_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = mounts_release,
 667         .poll           = mounts_poll,
 668 };
 669
 670 static int mountinfo_open(struct inode *inode, struct file *file)
 671 {
 672         return mounts_open_common(inode, file, &mountinfo_op);
 673 }
 674
 675 static const struct file_operations proc_mountinfo_operations = {
 676         .open           = mountinfo_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = mounts_release,
 680         .poll           = mounts_poll,
 681 };
 682
 683 static int mountstats_open(struct inode *inode, struct file *file)
 684 {
 685         return mounts_open_common(inode, file, &mountstats_op);
 686 }
 687
 688 static const struct file_operations proc_mountstats_operations = {
 689         .open           = mountstats_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = mounts_release,
 693 };
 694
 695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 696
 697 static ssize_t proc_info_read(struct file * file, char __user * buf,
 698                           size_t count, loff_t *ppos)
 699 {
 700         struct inode * inode = file->f_path.dentry->d_inode;
 701         unsigned long page;
 702         ssize_t length;
 703         struct task_struct *task = get_proc_task(inode);
 704
 705         length = -ESRCH;
 706         if (!task)
 707                 goto out_no_task;
 708
 709         if (count > PROC_BLOCK_SIZE)
 710                 count = PROC_BLOCK_SIZE;
 711
 712         length = -ENOMEM;
 713         if (!(page = __get_free_page(GFP_TEMPORARY)))
 714                 goto out;
 715
 716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 717
 718         if (length >= 0)
 719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 720         free_page(page);
 721 out:
 722         put_task_struct(task);
 723 out_no_task:
 724         return length;
 725 }
 726
 727 static const struct file_operations proc_info_file_operations = {
 728         .read           = proc_info_read,
 729         .llseek         = generic_file_llseek,
 730 };
 731
 732 static int proc_single_show(struct seq_file *m, void *v)
 733 {
 734         struct inode *inode = m->private;
 735         struct pid_namespace *ns;
 736         struct pid *pid;
 737         struct task_struct *task;
 738         int ret;
 739
 740         ns = inode->i_sb->s_fs_info;
 741         pid = proc_pid(inode);
 742         task = get_pid_task(pid, PIDTYPE_PID);
 743         if (!task)
 744                 return -ESRCH;
 745
 746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 747
 748         put_task_struct(task);
 749         return ret;
 750 }
 751
 752 static int proc_single_open(struct inode *inode, struct file *filp)
 753 {
 754         int ret;
 755         ret = single_open(filp, proc_single_show, NULL);
 756         if (!ret) {
 757                 struct seq_file *m = filp->private_data;
 758
 759                 m->private = inode;
 760         }
 761         return ret;
 762 }
 763
 764 static const struct file_operations proc_single_file_operations = {
 765         .open           = proc_single_open,
 766         .read           = seq_read,
 767         .llseek         = seq_lseek,
 768         .release        = single_release,
 769 };
 770
 771 static int mem_open(struct inode* inode, struct file* file)
 772 {
 773         file->private_data = (void*)((long)current->self_exec_id);
 774         return 0;
 775 }
 776
 777 static ssize_t mem_read(struct file * file, char __user * buf,
 778                         size_t count, loff_t *ppos)
 779 {
 780         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 781         char *page;
 782         unsigned long src = *ppos;
 783         int ret = -ESRCH;
 784         struct mm_struct *mm;
 785
 786         if (!task)
 787                 goto out_no_task;
 788
 789         if (check_mem_permission(task))
 790                 goto out;
 791
 792         ret = -ENOMEM;
 793         page = (char *)__get_free_page(GFP_TEMPORARY);
 794         if (!page)
 795                 goto out;
 796
 797         ret = 0;
 798
 799         mm = get_task_mm(task);
 800         if (!mm)
 801                 goto out_free;
 802
 803         ret = -EIO;
 804
 805         if (file->private_data != (void*)((long)current->self_exec_id))
 806                 goto out_put;
 807
 808         ret = 0;
 809
 810         while (count > 0) {
 811                 int this_len, retval;
 812
 813                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 814                 retval = access_process_vm(task, src, page, this_len, 0);
 815                 if (!retval || check_mem_permission(task)) {
 816                         if (!ret)
 817                                 ret = -EIO;
 818                         break;
 819                 }
 820
 821                 if (copy_to_user(buf, page, retval)) {
 822                         ret = -EFAULT;
 823                         break;
 824                 }
 825
 826                 ret += retval;
 827                 src += retval;
 828                 buf += retval;
 829                 count -= retval;
 830         }
 831         *ppos = src;
 832
 833 out_put:
 834         mmput(mm);
 835 out_free:
 836         free_page((unsigned long) page);
 837 out:
 838         put_task_struct(task);
 839 out_no_task:
 840         return ret;
 841 }
 842
 843 #define mem_write NULL
 844
 845 #ifndef mem_write
 846 /* This is a security hazard */
 847 static ssize_t mem_write(struct file * file, const char __user *buf,
 848                          size_t count, loff_t *ppos)
 849 {
 850         int copied;
 851         char *page;
 852         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 853         unsigned long dst = *ppos;
 854
 855         copied = -ESRCH;
 856         if (!task)
 857                 goto out_no_task;
 858
 859         if (check_mem_permission(task))
 860                 goto out;
 861
 862         copied = -ENOMEM;
 863         page = (char *)__get_free_page(GFP_TEMPORARY);
 864         if (!page)
 865                 goto out;
 866
 867         copied = 0;
 868         while (count > 0) {
 869                 int this_len, retval;
 870
 871                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 872                 if (copy_from_user(page, buf, this_len)) {
 873                         copied = -EFAULT;
 874                         break;
 875                 }
 876                 retval = access_process_vm(task, dst, page, this_len, 1);
 877                 if (!retval) {
 878                         if (!copied)
 879                                 copied = -EIO;
 880                         break;
 881                 }
 882                 copied += retval;
 883                 buf += retval;
 884                 dst += retval;
 885                 count -= retval;
 886         }
 887         *ppos = dst;
 888         free_page((unsigned long) page);
 889 out:
 890         put_task_struct(task);
 891 out_no_task:
 892         return copied;
 893 }
 894 #endif
 895
 896 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 897 {
 898         switch (orig) {
 899         case 0:
 900                 file->f_pos = offset;
 901                 break;
 902         case 1:
 903                 file->f_pos += offset;
 904                 break;
 905         default:
 906                 return -EINVAL;
 907         }
 908         force_successful_syscall_return();
 909         return file->f_pos;
 910 }
 911
 912 static const struct file_operations proc_mem_operations = {
 913         .llseek         = mem_lseek,
 914         .read           = mem_read,
 915         .write          = mem_write,
 916         .open           = mem_open,
 917 };
 918
 919 static ssize_t environ_read(struct file *file, char __user *buf,
 920                         size_t count, loff_t *ppos)
 921 {
 922         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 923         char *page;
 924         unsigned long src = *ppos;
 925         int ret = -ESRCH;
 926         struct mm_struct *mm;
 927
 928         if (!task)
 929                 goto out_no_task;
 930
 931         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 932                 goto out;
 933
 934         ret = -ENOMEM;
 935         page = (char *)__get_free_page(GFP_TEMPORARY);
 936         if (!page)
 937                 goto out;
 938
 939         ret = 0;
 940
 941         mm = get_task_mm(task);
 942         if (!mm)
 943                 goto out_free;
 944
 945         while (count > 0) {
 946                 int this_len, retval, max_len;
 947
 948                 this_len = mm->env_end - (mm->env_start + src);
 949
 950                 if (this_len <= 0)
 951                         break;
 952
 953                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 954                 this_len = (this_len > max_len) ? max_len : this_len;
 955
 956                 retval = access_process_vm(task, (mm->env_start + src),
 957                         page, this_len, 0);
 958
 959                 if (retval <= 0) {
 960                         ret = retval;
 961                         break;
 962                 }
 963
 964                 if (copy_to_user(buf, page, retval)) {
 965                         ret = -EFAULT;
 966                         break;
 967                 }
 968
 969                 ret += retval;
 970                 src += retval;
 971                 buf += retval;
 972                 count -= retval;
 973         }
 974         *ppos = src;
 975
 976         mmput(mm);
 977 out_free:
 978         free_page((unsigned long) page);
 979 out:
 980         put_task_struct(task);
 981 out_no_task:
 982         return ret;
 983 }
 984
 985 static const struct file_operations proc_environ_operations = {
 986         .read           = environ_read,
 987         .llseek         = generic_file_llseek,
 988 };
 989
 990 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 991                                 size_t count, loff_t *ppos)
 992 {
 993         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 994         char buffer[PROC_NUMBUF];
 995         size_t len;
 996         int oom_adjust = OOM_DISABLE;
 997         unsigned long flags;
 998
 999         if (!task)
1000                 return -ESRCH;
1001
1002         if (lock_task_sighand(task, &flags)) {
1003                 oom_adjust = task->signal->oom_adj;
1004                 unlock_task_sighand(task, &flags);
1005         }
1006
1007         put_task_struct(task);
1008
1009         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1010
1011         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1012 }
1013
1014 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1015                                 size_t count, loff_t *ppos)
1016 {
1017         struct task_struct *task;
1018         char buffer[PROC_NUMBUF];
1019         long oom_adjust;
1020         unsigned long flags;
1021         int err;
1022
1023         memset(buffer, 0, sizeof(buffer));
1024         if (count > sizeof(buffer) - 1)
1025                 count = sizeof(buffer) - 1;
1026         if (copy_from_user(buffer, buf, count)) {
1027                 err = -EFAULT;
1028                 goto out;
1029         }
1030
1031         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1032         if (err)
1033                 goto out;
1034         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1035              oom_adjust != OOM_DISABLE) {
1036                 err = -EINVAL;
1037                 goto out;
1038         }
1039
1040         task = get_proc_task(file->f_path.dentry->d_inode);
1041         if (!task) {
1042                 err = -ESRCH;
1043                 goto out;
1044         }
1045
1046         task_lock(task);
1047         if (!task->mm) {
1048                 err = -EINVAL;
1049                 goto err_task_lock;
1050         }
1051
1052         if (!lock_task_sighand(task, &flags)) {
1053                 err = -ESRCH;
1054                 goto err_task_lock;
1055         }
1056
1057         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1058                 err = -EACCES;
1059                 goto err_sighand;
1060         }
1061
1062         if (oom_adjust != task->signal->oom_adj) {
1063                 if (oom_adjust == OOM_DISABLE)
1064                         atomic_inc(&task->mm->oom_disable_count);
1065                 if (task->signal->oom_adj == OOM_DISABLE)
1066                         atomic_dec(&task->mm->oom_disable_count);
1067         }
1068
1069         /*
1070          * Warn that /proc/pid/oom_adj is deprecated, see
1071          * Documentation/feature-removal-schedule.txt.
1072          */
1073         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1074                         "please use /proc/%d/oom_score_adj instead.\n",
1075                         current->comm, task_pid_nr(current),
1076                         task_pid_nr(task), task_pid_nr(task));
1077         task->signal->oom_adj = oom_adjust;
1078         /*
1079          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1080          * value is always attainable.
1081          */
1082         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1083                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1084         else
1085                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1086                                                                 -OOM_DISABLE;
1087 err_sighand:
1088         unlock_task_sighand(task, &flags);
1089 err_task_lock:
1090         task_unlock(task);
1091         put_task_struct(task);
1092 out:
1093         return err < 0 ? err : count;
1094 }
1095
1096 static const struct file_operations proc_oom_adjust_operations = {
1097         .read           = oom_adjust_read,
1098         .write          = oom_adjust_write,
1099         .llseek         = generic_file_llseek,
1100 };
1101
1102 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1103                                         size_t count, loff_t *ppos)
1104 {
1105         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1106         char buffer[PROC_NUMBUF];
1107         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1108         unsigned long flags;
1109         size_t len;
1110
1111         if (!task)
1112                 return -ESRCH;
1113         if (lock_task_sighand(task, &flags)) {
1114                 oom_score_adj = task->signal->oom_score_adj;
1115                 unlock_task_sighand(task, &flags);
1116         }
1117         put_task_struct(task);
1118         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1119         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1120 }
1121
1122 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1123                                         size_t count, loff_t *ppos)
1124 {
1125         struct task_struct *task;
1126         char buffer[PROC_NUMBUF];
1127         unsigned long flags;
1128         long oom_score_adj;
1129         int err;
1130
1131         memset(buffer, 0, sizeof(buffer));
1132         if (count > sizeof(buffer) - 1)
1133                 count = sizeof(buffer) - 1;
1134         if (copy_from_user(buffer, buf, count)) {
1135                 err = -EFAULT;
1136                 goto out;
1137         }
1138
1139         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1140         if (err)
1141                 goto out;
1142         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1143                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1144                 err = -EINVAL;
1145                 goto out;
1146         }
1147
1148         task = get_proc_task(file->f_path.dentry->d_inode);
1149         if (!task) {
1150                 err = -ESRCH;
1151                 goto out;
1152         }
1153
1154         task_lock(task);
1155         if (!task->mm) {
1156                 err = -EINVAL;
1157                 goto err_task_lock;
1158         }
1159
1160         if (!lock_task_sighand(task, &flags)) {
1161                 err = -ESRCH;
1162                 goto err_task_lock;
1163         }
1164
1165         if (oom_score_adj < task->signal->oom_score_adj &&
1166                         !capable(CAP_SYS_RESOURCE)) {
1167                 err = -EACCES;
1168                 goto err_sighand;
1169         }
1170
1171         if (oom_score_adj != task->signal->oom_score_adj) {
1172                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1173                         atomic_inc(&task->mm->oom_disable_count);
1174                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1175                         atomic_dec(&task->mm->oom_disable_count);
1176         }
1177         task->signal->oom_score_adj = oom_score_adj;
1178         /*
1179          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1180          * always attainable.
1181          */
1182         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1183                 task->signal->oom_adj = OOM_DISABLE;
1184         else
1185                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1186                                                         OOM_SCORE_ADJ_MAX;
1187 err_sighand:
1188         unlock_task_sighand(task, &flags);
1189 err_task_lock:
1190         task_unlock(task);
1191         put_task_struct(task);
1192 out:
1193         return err < 0 ? err : count;
1194 }
1195
1196 static const struct file_operations proc_oom_score_adj_operations = {
1197         .read           = oom_score_adj_read,
1198         .write          = oom_score_adj_write,
1199         .llseek         = default_llseek,
1200 };
1201
1202 #ifdef CONFIG_AUDITSYSCALL
1203 #define TMPBUFLEN 21
1204 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1205                                   size_t count, loff_t *ppos)
1206 {
1207         struct inode * inode = file->f_path.dentry->d_inode;
1208         struct task_struct *task = get_proc_task(inode);
1209         ssize_t length;
1210         char tmpbuf[TMPBUFLEN];
1211
1212         if (!task)
1213                 return -ESRCH;
1214         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1215                                 audit_get_loginuid(task));
1216         put_task_struct(task);
1217         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1218 }
1219
1220 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1221                                    size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file->f_path.dentry->d_inode;
1224         char *page, *tmp;
1225         ssize_t length;
1226         uid_t loginuid;
1227
1228         if (!capable(CAP_AUDIT_CONTROL))
1229                 return -EPERM;
1230
1231         rcu_read_lock();
1232         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1233                 rcu_read_unlock();
1234                 return -EPERM;
1235         }
1236         rcu_read_unlock();
1237
1238         if (count >= PAGE_SIZE)
1239                 count = PAGE_SIZE - 1;
1240
1241         if (*ppos != 0) {
1242                 /* No partial writes. */
1243                 return -EINVAL;
1244         }
1245         page = (char*)__get_free_page(GFP_TEMPORARY);
1246         if (!page)
1247                 return -ENOMEM;
1248         length = -EFAULT;
1249         if (copy_from_user(page, buf, count))
1250                 goto out_free_page;
1251
1252         page[count] = '\0';
1253         loginuid = simple_strtoul(page, &tmp, 10);
1254         if (tmp == page) {
1255                 length = -EINVAL;
1256                 goto out_free_page;
1257
1258         }
1259         length = audit_set_loginuid(current, loginuid);
1260         if (likely(length == 0))
1261                 length = count;
1262
1263 out_free_page:
1264         free_page((unsigned long) page);
1265         return length;
1266 }
1267
1268 static const struct file_operations proc_loginuid_operations = {
1269         .read           = proc_loginuid_read,
1270         .write          = proc_loginuid_write,
1271         .llseek         = generic_file_llseek,
1272 };
1273
1274 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1275                                   size_t count, loff_t *ppos)
1276 {
1277         struct inode * inode = file->f_path.dentry->d_inode;
1278         struct task_struct *task = get_proc_task(inode);
1279         ssize_t length;
1280         char tmpbuf[TMPBUFLEN];
1281
1282         if (!task)
1283                 return -ESRCH;
1284         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1285                                 audit_get_sessionid(task));
1286         put_task_struct(task);
1287         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1288 }
1289
1290 static const struct file_operations proc_sessionid_operations = {
1291         .read           = proc_sessionid_read,
1292         .llseek         = generic_file_llseek,
1293 };
1294 #endif
1295
1296 #ifdef CONFIG_FAULT_INJECTION
1297 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1298                                       size_t count, loff_t *ppos)
1299 {
1300         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1301         char buffer[PROC_NUMBUF];
1302         size_t len;
1303         int make_it_fail;
1304
1305         if (!task)
1306                 return -ESRCH;
1307         make_it_fail = task->make_it_fail;
1308         put_task_struct(task);
1309
1310         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1311
1312         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1313 }
1314
1315 static ssize_t proc_fault_inject_write(struct file * file,
1316                         const char __user * buf, size_t count, loff_t *ppos)
1317 {
1318         struct task_struct *task;
1319         char buffer[PROC_NUMBUF], *end;
1320         int make_it_fail;
1321
1322         if (!capable(CAP_SYS_RESOURCE))
1323                 return -EPERM;
1324         memset(buffer, 0, sizeof(buffer));
1325         if (count > sizeof(buffer) - 1)
1326                 count = sizeof(buffer) - 1;
1327         if (copy_from_user(buffer, buf, count))
1328                 return -EFAULT;
1329         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1330         if (*end)
1331                 return -EINVAL;
1332         task = get_proc_task(file->f_dentry->d_inode);
1333         if (!task)
1334                 return -ESRCH;
1335         task->make_it_fail = make_it_fail;
1336         put_task_struct(task);
1337
1338         return count;
1339 }
1340
1341 static const struct file_operations proc_fault_inject_operations = {
1342         .read           = proc_fault_inject_read,
1343         .write          = proc_fault_inject_write,
1344         .llseek         = generic_file_llseek,
1345 };
1346 #endif
1347
1348
1349 #ifdef CONFIG_SCHED_DEBUG
1350 /*
1351  * Print out various scheduling related per-task fields:
1352  */
1353 static int sched_show(struct seq_file *m, void *v)
1354 {
1355         struct inode *inode = m->private;
1356         struct task_struct *p;
1357
1358         p = get_proc_task(inode);
1359         if (!p)
1360                 return -ESRCH;
1361         proc_sched_show_task(p, m);
1362
1363         put_task_struct(p);
1364
1365         return 0;
1366 }
1367
1368 static ssize_t
1369 sched_write(struct file *file, const char __user *buf,
1370             size_t count, loff_t *offset)
1371 {
1372         struct inode *inode = file->f_path.dentry->d_inode;
1373         struct task_struct *p;
1374
1375         p = get_proc_task(inode);
1376         if (!p)
1377                 return -ESRCH;
1378         proc_sched_set_task(p);
1379
1380         put_task_struct(p);
1381
1382         return count;
1383 }
1384
1385 static int sched_open(struct inode *inode, struct file *filp)
1386 {
1387         int ret;
1388
1389         ret = single_open(filp, sched_show, NULL);
1390         if (!ret) {
1391                 struct seq_file *m = filp->private_data;
1392
1393                 m->private = inode;
1394         }
1395         return ret;
1396 }
1397
1398 static const struct file_operations proc_pid_sched_operations = {
1399         .open           = sched_open,
1400         .read           = seq_read,
1401         .write          = sched_write,
1402         .llseek         = seq_lseek,
1403         .release        = single_release,
1404 };
1405
1406 #endif
1407
1408 static ssize_t comm_write(struct file *file, const char __user *buf,
1409                                 size_t count, loff_t *offset)
1410 {
1411         struct inode *inode = file->f_path.dentry->d_inode;
1412         struct task_struct *p;
1413         char buffer[TASK_COMM_LEN];
1414
1415         memset(buffer, 0, sizeof(buffer));
1416         if (count > sizeof(buffer) - 1)
1417                 count = sizeof(buffer) - 1;
1418         if (copy_from_user(buffer, buf, count))
1419                 return -EFAULT;
1420
1421         p = get_proc_task(inode);
1422         if (!p)
1423                 return -ESRCH;
1424
1425         if (same_thread_group(current, p))
1426                 set_task_comm(p, buffer);
1427         else
1428                 count = -EINVAL;
1429
1430         put_task_struct(p);
1431
1432         return count;
1433 }
1434
1435 static int comm_show(struct seq_file *m, void *v)
1436 {
1437         struct inode *inode = m->private;
1438         struct task_struct *p;
1439
1440         p = get_proc_task(inode);
1441         if (!p)
1442                 return -ESRCH;
1443
1444         task_lock(p);
1445         seq_printf(m, "%s\n", p->comm);
1446         task_unlock(p);
1447
1448         put_task_struct(p);
1449
1450         return 0;
1451 }
1452
1453 static int comm_open(struct inode *inode, struct file *filp)
1454 {
1455         int ret;
1456
1457         ret = single_open(filp, comm_show, NULL);
1458         if (!ret) {
1459                 struct seq_file *m = filp->private_data;
1460
1461                 m->private = inode;
1462         }
1463         return ret;
1464 }
1465
1466 static const struct file_operations proc_pid_set_comm_operations = {
1467         .open           = comm_open,
1468         .read           = seq_read,
1469         .write          = comm_write,
1470         .llseek         = seq_lseek,
1471         .release        = single_release,
1472 };
1473
1474 /*
1475  * We added or removed a vma mapping the executable. The vmas are only mapped
1476  * during exec and are not mapped with the mmap system call.
1477  * Callers must hold down_write() on the mm's mmap_sem for these
1478  */
1479 void added_exe_file_vma(struct mm_struct *mm)
1480 {
1481         mm->num_exe_file_vmas++;
1482 }
1483
1484 void removed_exe_file_vma(struct mm_struct *mm)
1485 {
1486         mm->num_exe_file_vmas--;
1487         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1488                 fput(mm->exe_file);
1489                 mm->exe_file = NULL;
1490         }
1491
1492 }
1493
1494 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1495 {
1496         if (new_exe_file)
1497                 get_file(new_exe_file);
1498         if (mm->exe_file)
1499                 fput(mm->exe_file);
1500         mm->exe_file = new_exe_file;
1501         mm->num_exe_file_vmas = 0;
1502 }
1503
1504 struct file *get_mm_exe_file(struct mm_struct *mm)
1505 {
1506         struct file *exe_file;
1507
1508         /* We need mmap_sem to protect against races with removal of
1509          * VM_EXECUTABLE vmas */
1510         down_read(&mm->mmap_sem);
1511         exe_file = mm->exe_file;
1512         if (exe_file)
1513                 get_file(exe_file);
1514         up_read(&mm->mmap_sem);
1515         return exe_file;
1516 }
1517
1518 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1519 {
1520         /* It's safe to write the exe_file pointer without exe_file_lock because
1521          * this is called during fork when the task is not yet in /proc */
1522         newmm->exe_file = get_mm_exe_file(oldmm);
1523 }
1524
1525 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1526 {
1527         struct task_struct *task;
1528         struct mm_struct *mm;
1529         struct file *exe_file;
1530
1531         task = get_proc_task(inode);
1532         if (!task)
1533                 return -ENOENT;
1534         mm = get_task_mm(task);
1535         put_task_struct(task);
1536         if (!mm)
1537                 return -ENOENT;
1538         exe_file = get_mm_exe_file(mm);
1539         mmput(mm);
1540         if (exe_file) {
1541                 *exe_path = exe_file->f_path;
1542                 path_get(&exe_file->f_path);
1543                 fput(exe_file);
1544                 return 0;
1545         } else
1546                 return -ENOENT;
1547 }
1548
1549 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1550 {
1551         struct inode *inode = dentry->d_inode;
1552         int error = -EACCES;
1553
1554         /* We don't need a base pointer in the /proc filesystem */
1555         path_put(&nd->path);
1556
1557         /* Are we allowed to snoop on the tasks file descriptors? */
1558         if (!proc_fd_access_allowed(inode))
1559                 goto out;
1560
1561         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1562 out:
1563         return ERR_PTR(error);
1564 }
1565
1566 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1567 {
1568         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1569         char *pathname;
1570         int len;
1571
1572         if (!tmp)
1573                 return -ENOMEM;
1574
1575         pathname = d_path_with_unreachable(path, tmp, PAGE_SIZE);
1576         len = PTR_ERR(pathname);
1577         if (IS_ERR(pathname))
1578                 goto out;
1579         len = tmp + PAGE_SIZE - 1 - pathname;
1580
1581         if (len > buflen)
1582                 len = buflen;
1583         if (copy_to_user(buffer, pathname, len))
1584                 len = -EFAULT;
1585  out:
1586         free_page((unsigned long)tmp);
1587         return len;
1588 }
1589
1590 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1591 {
1592         int error = -EACCES;
1593         struct inode *inode = dentry->d_inode;
1594         struct path path;
1595
1596         /* Are we allowed to snoop on the tasks file descriptors? */
1597         if (!proc_fd_access_allowed(inode))
1598                 goto out;
1599
1600         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1601         if (error)
1602                 goto out;
1603
1604         error = do_proc_readlink(&path, buffer, buflen);
1605         path_put(&path);
1606 out:
1607         return error;
1608 }
1609
1610 static const struct inode_operations proc_pid_link_inode_operations = {
1611         .readlink       = proc_pid_readlink,
1612         .follow_link    = proc_pid_follow_link,
1613         .setattr        = proc_setattr,
1614 };
1615
1616
1617 /* building an inode */
1618
1619 static int task_dumpable(struct task_struct *task)
1620 {
1621         int dumpable = 0;
1622         struct mm_struct *mm;
1623
1624         task_lock(task);
1625         mm = task->mm;
1626         if (mm)
1627                 dumpable = get_dumpable(mm);
1628         task_unlock(task);
1629         if(dumpable == 1)
1630                 return 1;
1631         return 0;
1632 }
1633
1634
1635 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1636 {
1637         struct inode * inode;
1638         struct proc_inode *ei;
1639         const struct cred *cred;
1640
1641         /* We need a new inode */
1642
1643         inode = new_inode(sb);
1644         if (!inode)
1645                 goto out;
1646
1647         /* Common stuff */
1648         ei = PROC_I(inode);
1649         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1650         inode->i_op = &proc_def_inode_operations;
1651
1652         /*
1653          * grab the reference to task.
1654          */
1655         ei->pid = get_task_pid(task, PIDTYPE_PID);
1656         if (!ei->pid)
1657                 goto out_unlock;
1658
1659         if (task_dumpable(task)) {
1660                 rcu_read_lock();
1661                 cred = __task_cred(task);
1662                 inode->i_uid = cred->euid;
1663                 inode->i_gid = cred->egid;
1664                 rcu_read_unlock();
1665         }
1666         security_task_to_inode(task, inode);
1667
1668 out:
1669         return inode;
1670
1671 out_unlock:
1672         iput(inode);
1673         return NULL;
1674 }
1675
1676 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1677 {
1678         struct inode *inode = dentry->d_inode;
1679         struct task_struct *task;
1680         const struct cred *cred;
1681
1682         generic_fillattr(inode, stat);
1683
1684         rcu_read_lock();
1685         stat->uid = 0;
1686         stat->gid = 0;
1687         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1688         if (task) {
1689                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1690                     task_dumpable(task)) {
1691                         cred = __task_cred(task);
1692                         stat->uid = cred->euid;
1693                         stat->gid = cred->egid;
1694                 }
1695         }
1696         rcu_read_unlock();
1697         return 0;
1698 }
1699
1700 /* dentry stuff */
1701
1702 /*
1703  *      Exceptional case: normally we are not allowed to unhash a busy
1704  * directory. In this case, however, we can do it - no aliasing problems
1705  * due to the way we treat inodes.
1706  *
1707  * Rewrite the inode's ownerships here because the owning task may have
1708  * performed a setuid(), etc.
1709  *
1710  * Before the /proc/pid/status file was created the only way to read
1711  * the effective uid of a /process was to stat /proc/pid.  Reading
1712  * /proc/pid/status is slow enough that procps and other packages
1713  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1714  * made this apply to all per process world readable and executable
1715  * directories.
1716  */
1717 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1718 {
1719         struct inode *inode = dentry->d_inode;
1720         struct task_struct *task = get_proc_task(inode);
1721         const struct cred *cred;
1722
1723         if (task) {
1724                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1725                     task_dumpable(task)) {
1726                         rcu_read_lock();
1727                         cred = __task_cred(task);
1728                         inode->i_uid = cred->euid;
1729                         inode->i_gid = cred->egid;
1730                         rcu_read_unlock();
1731                 } else {
1732                         inode->i_uid = 0;
1733                         inode->i_gid = 0;
1734                 }
1735                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1736                 security_task_to_inode(task, inode);
1737                 put_task_struct(task);
1738                 return 1;
1739         }
1740         d_drop(dentry);
1741         return 0;
1742 }
1743
1744 static int pid_delete_dentry(struct dentry * dentry)
1745 {
1746         /* Is the task we represent dead?
1747          * If so, then don't put the dentry on the lru list,
1748          * kill it immediately.
1749          */
1750         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1751 }
1752
1753 static const struct dentry_operations pid_dentry_operations =
1754 {
1755         .d_revalidate   = pid_revalidate,
1756         .d_delete       = pid_delete_dentry,
1757 };
1758
1759 /* Lookups */
1760
1761 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1762                                 struct task_struct *, const void *);
1763
1764 /*
1765  * Fill a directory entry.
1766  *
1767  * If possible create the dcache entry and derive our inode number and
1768  * file type from dcache entry.
1769  *
1770  * Since all of the proc inode numbers are dynamically generated, the inode
1771  * numbers do not exist until the inode is cache.  This means creating the
1772  * the dcache entry in readdir is necessary to keep the inode numbers
1773  * reported by readdir in sync with the inode numbers reported
1774  * by stat.
1775  */
1776 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1777         char *name, int len,
1778         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1779 {
1780         struct dentry *child, *dir = filp->f_path.dentry;
1781         struct inode *inode;
1782         struct qstr qname;
1783         ino_t ino = 0;
1784         unsigned type = DT_UNKNOWN;
1785
1786         qname.name = name;
1787         qname.len  = len;
1788         qname.hash = full_name_hash(name, len);
1789
1790         child = d_lookup(dir, &qname);
1791         if (!child) {
1792                 struct dentry *new;
1793                 new = d_alloc(dir, &qname);
1794                 if (new) {
1795                         child = instantiate(dir->d_inode, new, task, ptr);
1796                         if (child)
1797                                 dput(new);
1798                         else
1799                                 child = new;
1800                 }
1801         }
1802         if (!child || IS_ERR(child) || !child->d_inode)
1803                 goto end_instantiate;
1804         inode = child->d_inode;
1805         if (inode) {
1806                 ino = inode->i_ino;
1807                 type = inode->i_mode >> 12;
1808         }
1809         dput(child);
1810 end_instantiate:
1811         if (!ino)
1812                 ino = find_inode_number(dir, &qname);
1813         if (!ino)
1814                 ino = 1;
1815         return filldir(dirent, name, len, filp->f_pos, ino, type);
1816 }
1817
1818 static unsigned name_to_int(struct dentry *dentry)
1819 {
1820         const char *name = dentry->d_name.name;
1821         int len = dentry->d_name.len;
1822         unsigned n = 0;
1823
1824         if (len > 1 && *name == '0')
1825                 goto out;
1826         while (len-- > 0) {
1827                 unsigned c = *name++ - '0';
1828                 if (c > 9)
1829                         goto out;
1830                 if (n >= (~0U-9)/10)
1831                         goto out;
1832                 n *= 10;
1833                 n += c;
1834         }
1835         return n;
1836 out:
1837         return ~0U;
1838 }
1839
1840 #define PROC_FDINFO_MAX 64
1841
1842 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1843 {
1844         struct task_struct *task = get_proc_task(inode);
1845         struct files_struct *files = NULL;
1846         struct file *file;
1847         int fd = proc_fd(inode);
1848
1849         if (task) {
1850                 files = get_files_struct(task);
1851                 put_task_struct(task);
1852         }
1853         if (files) {
1854                 /*
1855                  * We are not taking a ref to the file structure, so we must
1856                  * hold ->file_lock.
1857                  */
1858                 spin_lock(&files->file_lock);
1859                 file = fcheck_files(files, fd);
1860                 if (file) {
1861                         if (path) {
1862                                 *path = file->f_path;
1863                                 path_get(&file->f_path);
1864                         }
1865                         if (info)
1866                                 snprintf(info, PROC_FDINFO_MAX,
1867                                          "pos:\t%lli\n"
1868                                          "flags:\t0%o\n",
1869                                          (long long) file->f_pos,
1870                                          file->f_flags);
1871                         spin_unlock(&files->file_lock);
1872                         put_files_struct(files);
1873                         return 0;
1874                 }
1875                 spin_unlock(&files->file_lock);
1876                 put_files_struct(files);
1877         }
1878         return -ENOENT;
1879 }
1880
1881 static int proc_fd_link(struct inode *inode, struct path *path)
1882 {
1883         return proc_fd_info(inode, path, NULL);
1884 }
1885
1886 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1887 {
1888         struct inode *inode = dentry->d_inode;
1889         struct task_struct *task = get_proc_task(inode);
1890         int fd = proc_fd(inode);
1891         struct files_struct *files;
1892         const struct cred *cred;
1893
1894         if (task) {
1895                 files = get_files_struct(task);
1896                 if (files) {
1897                         rcu_read_lock();
1898                         if (fcheck_files(files, fd)) {
1899                                 rcu_read_unlock();
1900                                 put_files_struct(files);
1901                                 if (task_dumpable(task)) {
1902                                         rcu_read_lock();
1903                                         cred = __task_cred(task);
1904                                         inode->i_uid = cred->euid;
1905                                         inode->i_gid = cred->egid;
1906                                         rcu_read_unlock();
1907                                 } else {
1908                                         inode->i_uid = 0;
1909                                         inode->i_gid = 0;
1910                                 }
1911                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1912                                 security_task_to_inode(task, inode);
1913                                 put_task_struct(task);
1914                                 return 1;
1915                         }
1916                         rcu_read_unlock();
1917                         put_files_struct(files);
1918                 }
1919                 put_task_struct(task);
1920         }
1921         d_drop(dentry);
1922         return 0;
1923 }
1924
1925 static const struct dentry_operations tid_fd_dentry_operations =
1926 {
1927         .d_revalidate   = tid_fd_revalidate,
1928         .d_delete       = pid_delete_dentry,
1929 };
1930
1931 static struct dentry *proc_fd_instantiate(struct inode *dir,
1932         struct dentry *dentry, struct task_struct *task, const void *ptr)
1933 {
1934         unsigned fd = *(const unsigned *)ptr;
1935         struct file *file;
1936         struct files_struct *files;
1937         struct inode *inode;
1938         struct proc_inode *ei;
1939         struct dentry *error = ERR_PTR(-ENOENT);
1940
1941         inode = proc_pid_make_inode(dir->i_sb, task);
1942         if (!inode)
1943                 goto out;
1944         ei = PROC_I(inode);
1945         ei->fd = fd;
1946         files = get_files_struct(task);
1947         if (!files)
1948                 goto out_iput;
1949         inode->i_mode = S_IFLNK;
1950
1951         /*
1952          * We are not taking a ref to the file structure, so we must
1953          * hold ->file_lock.
1954          */
1955         spin_lock(&files->file_lock);
1956         file = fcheck_files(files, fd);
1957         if (!file)
1958                 goto out_unlock;
1959         if (file->f_mode & FMODE_READ)
1960                 inode->i_mode |= S_IRUSR | S_IXUSR;
1961         if (file->f_mode & FMODE_WRITE)
1962                 inode->i_mode |= S_IWUSR | S_IXUSR;
1963         spin_unlock(&files->file_lock);
1964         put_files_struct(files);
1965
1966         inode->i_op = &proc_pid_link_inode_operations;
1967         inode->i_size = 64;
1968         ei->op.proc_get_link = proc_fd_link;
1969         dentry->d_op = &tid_fd_dentry_operations;
1970         d_add(dentry, inode);
1971         /* Close the race of the process dying before we return the dentry */
1972         if (tid_fd_revalidate(dentry, NULL))
1973                 error = NULL;
1974
1975  out:
1976         return error;
1977 out_unlock:
1978         spin_unlock(&files->file_lock);
1979         put_files_struct(files);
1980 out_iput:
1981         iput(inode);
1982         goto out;
1983 }
1984
1985 static struct dentry *proc_lookupfd_common(struct inode *dir,
1986                                            struct dentry *dentry,
1987                                            instantiate_t instantiate)
1988 {
1989         struct task_struct *task = get_proc_task(dir);
1990         unsigned fd = name_to_int(dentry);
1991         struct dentry *result = ERR_PTR(-ENOENT);
1992
1993         if (!task)
1994                 goto out_no_task;
1995         if (fd == ~0U)
1996                 goto out;
1997
1998         result = instantiate(dir, dentry, task, &fd);
1999 out:
2000         put_task_struct(task);
2001 out_no_task:
2002         return result;
2003 }
2004
2005 static int proc_readfd_common(struct file * filp, void * dirent,
2006                               filldir_t filldir, instantiate_t instantiate)
2007 {
2008         struct dentry *dentry = filp->f_path.dentry;
2009         struct inode *inode = dentry->d_inode;
2010         struct task_struct *p = get_proc_task(inode);
2011         unsigned int fd, ino;
2012         int retval;
2013         struct files_struct * files;
2014
2015         retval = -ENOENT;
2016         if (!p)
2017                 goto out_no_task;
2018         retval = 0;
2019
2020         fd = filp->f_pos;
2021         switch (fd) {
2022                 case 0:
2023                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2024                                 goto out;
2025                         filp->f_pos++;
2026                 case 1:
2027                         ino = parent_ino(dentry);
2028                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2029                                 goto out;
2030                         filp->f_pos++;
2031                 default:
2032                         files = get_files_struct(p);
2033                         if (!files)
2034                                 goto out;
2035                         rcu_read_lock();
2036                         for (fd = filp->f_pos-2;
2037                              fd < files_fdtable(files)->max_fds;
2038                              fd++, filp->f_pos++) {
2039                                 char name[PROC_NUMBUF];
2040                                 int len;
2041
2042                                 if (!fcheck_files(files, fd))
2043                                         continue;
2044                                 rcu_read_unlock();
2045
2046                                 len = snprintf(name, sizeof(name), "%d", fd);
2047                                 if (proc_fill_cache(filp, dirent, filldir,
2048                                                     name, len, instantiate,
2049                                                     p, &fd) < 0) {
2050                                         rcu_read_lock();
2051                                         break;
2052                                 }
2053                                 rcu_read_lock();
2054                         }
2055                         rcu_read_unlock();
2056                         put_files_struct(files);
2057         }
2058 out:
2059         put_task_struct(p);
2060 out_no_task:
2061         return retval;
2062 }
2063
2064 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2065                                     struct nameidata *nd)
2066 {
2067         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2068 }
2069
2070 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2071 {
2072         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2073 }
2074
2075 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2076                                       size_t len, loff_t *ppos)
2077 {
2078         char tmp[PROC_FDINFO_MAX];
2079         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2080         if (!err)
2081                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2082         return err;
2083 }
2084
2085 static const struct file_operations proc_fdinfo_file_operations = {
2086         .open           = nonseekable_open,
2087         .read           = proc_fdinfo_read,
2088         .llseek         = no_llseek,
2089 };
2090
2091 static const struct file_operations proc_fd_operations = {
2092         .read           = generic_read_dir,
2093         .readdir        = proc_readfd,
2094         .llseek         = default_llseek,
2095 };
2096
2097 /*
2098  * /proc/pid/fd needs a special permission handler so that a process can still
2099  * access /proc/self/fd after it has executed a setuid().
2100  */
2101 static int proc_fd_permission(struct inode *inode, int mask)
2102 {
2103         int rv;
2104
2105         rv = generic_permission(inode, mask, NULL);
2106         if (rv == 0)
2107                 return 0;
2108         if (task_pid(current) == proc_pid(inode))
2109                 rv = 0;
2110         return rv;
2111 }
2112
2113 /*
2114  * proc directories can do almost nothing..
2115  */
2116 static const struct inode_operations proc_fd_inode_operations = {
2117         .lookup         = proc_lookupfd,
2118         .permission     = proc_fd_permission,
2119         .setattr        = proc_setattr,
2120 };
2121
2122 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2123         struct dentry *dentry, struct task_struct *task, const void *ptr)
2124 {
2125         unsigned fd = *(unsigned *)ptr;
2126         struct inode *inode;
2127         struct proc_inode *ei;
2128         struct dentry *error = ERR_PTR(-ENOENT);
2129
2130         inode = proc_pid_make_inode(dir->i_sb, task);
2131         if (!inode)
2132                 goto out;
2133         ei = PROC_I(inode);
2134         ei->fd = fd;
2135         inode->i_mode = S_IFREG | S_IRUSR;
2136         inode->i_fop = &proc_fdinfo_file_operations;
2137         dentry->d_op = &tid_fd_dentry_operations;
2138         d_add(dentry, inode);
2139         /* Close the race of the process dying before we return the dentry */
2140         if (tid_fd_revalidate(dentry, NULL))
2141                 error = NULL;
2142
2143  out:
2144         return error;
2145 }
2146
2147 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2148                                         struct dentry *dentry,
2149                                         struct nameidata *nd)
2150 {
2151         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2152 }
2153
2154 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2155 {
2156         return proc_readfd_common(filp, dirent, filldir,
2157                                   proc_fdinfo_instantiate);
2158 }
2159
2160 static const struct file_operations proc_fdinfo_operations = {
2161         .read           = generic_read_dir,
2162         .readdir        = proc_readfdinfo,
2163         .llseek         = default_llseek,
2164 };
2165
2166 /*
2167  * proc directories can do almost nothing..
2168  */
2169 static const struct inode_operations proc_fdinfo_inode_operations = {
2170         .lookup         = proc_lookupfdinfo,
2171         .setattr        = proc_setattr,
2172 };
2173
2174
2175 static struct dentry *proc_pident_instantiate(struct inode *dir,
2176         struct dentry *dentry, struct task_struct *task, const void *ptr)
2177 {
2178         const struct pid_entry *p = ptr;
2179         struct inode *inode;
2180         struct proc_inode *ei;
2181         struct dentry *error = ERR_PTR(-ENOENT);
2182
2183         inode = proc_pid_make_inode(dir->i_sb, task);
2184         if (!inode)
2185                 goto out;
2186
2187         ei = PROC_I(inode);
2188         inode->i_mode = p->mode;
2189         if (S_ISDIR(inode->i_mode))
2190                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2191         if (p->iop)
2192                 inode->i_op = p->iop;
2193         if (p->fop)
2194                 inode->i_fop = p->fop;
2195         ei->op = p->op;
2196         dentry->d_op = &pid_dentry_operations;
2197         d_add(dentry, inode);
2198         /* Close the race of the process dying before we return the dentry */
2199         if (pid_revalidate(dentry, NULL))
2200                 error = NULL;
2201 out:
2202         return error;
2203 }
2204
2205 static struct dentry *proc_pident_lookup(struct inode *dir,
2206                                          struct dentry *dentry,
2207                                          const struct pid_entry *ents,
2208                                          unsigned int nents)
2209 {
2210         struct dentry *error;
2211         struct task_struct *task = get_proc_task(dir);
2212         const struct pid_entry *p, *last;
2213
2214         error = ERR_PTR(-ENOENT);
2215
2216         if (!task)
2217                 goto out_no_task;
2218
2219         /*
2220          * Yes, it does not scale. And it should not. Don't add
2221          * new entries into /proc/<tgid>/ without very good reasons.
2222          */
2223         last = &ents[nents - 1];
2224         for (p = ents; p <= last; p++) {
2225                 if (p->len != dentry->d_name.len)
2226                         continue;
2227                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2228                         break;
2229         }
2230         if (p > last)
2231                 goto out;
2232
2233         error = proc_pident_instantiate(dir, dentry, task, p);
2234 out:
2235         put_task_struct(task);
2236 out_no_task:
2237         return error;
2238 }
2239
2240 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2241         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2242 {
2243         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2244                                 proc_pident_instantiate, task, p);
2245 }
2246
2247 static int proc_pident_readdir(struct file *filp,
2248                 void *dirent, filldir_t filldir,
2249                 const struct pid_entry *ents, unsigned int nents)
2250 {
2251         int i;
2252         struct dentry *dentry = filp->f_path.dentry;
2253         struct inode *inode = dentry->d_inode;
2254         struct task_struct *task = get_proc_task(inode);
2255         const struct pid_entry *p, *last;
2256         ino_t ino;
2257         int ret;
2258
2259         ret = -ENOENT;
2260         if (!task)
2261                 goto out_no_task;
2262
2263         ret = 0;
2264         i = filp->f_pos;
2265         switch (i) {
2266         case 0:
2267                 ino = inode->i_ino;
2268                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2269                         goto out;
2270                 i++;
2271                 filp->f_pos++;
2272                 /* fall through */
2273         case 1:
2274                 ino = parent_ino(dentry);
2275                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2276                         goto out;
2277                 i++;
2278                 filp->f_pos++;
2279                 /* fall through */
2280         default:
2281                 i -= 2;
2282                 if (i >= nents) {
2283                         ret = 1;
2284                         goto out;
2285                 }
2286                 p = ents + i;
2287                 last = &ents[nents - 1];
2288                 while (p <= last) {
2289                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2290                                 goto out;
2291                         filp->f_pos++;
2292                         p++;
2293                 }
2294         }
2295
2296         ret = 1;
2297 out:
2298         put_task_struct(task);
2299 out_no_task:
2300         return ret;
2301 }
2302
2303 #ifdef CONFIG_SECURITY
2304 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2305                                   size_t count, loff_t *ppos)
2306 {
2307         struct inode * inode = file->f_path.dentry->d_inode;
2308         char *p = NULL;
2309         ssize_t length;
2310         struct task_struct *task = get_proc_task(inode);
2311
2312         if (!task)
2313                 return -ESRCH;
2314
2315         length = security_getprocattr(task,
2316                                       (char*)file->f_path.dentry->d_name.name,
2317                                       &p);
2318         put_task_struct(task);
2319         if (length > 0)
2320                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2321         kfree(p);
2322         return length;
2323 }
2324
2325 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2326                                    size_t count, loff_t *ppos)
2327 {
2328         struct inode * inode = file->f_path.dentry->d_inode;
2329         char *page;
2330         ssize_t length;
2331         struct task_struct *task = get_proc_task(inode);
2332
2333         length = -ESRCH;
2334         if (!task)
2335                 goto out_no_task;
2336         if (count > PAGE_SIZE)
2337                 count = PAGE_SIZE;
2338
2339         /* No partial writes. */
2340         length = -EINVAL;
2341         if (*ppos != 0)
2342                 goto out;
2343
2344         length = -ENOMEM;
2345         page = (char*)__get_free_page(GFP_TEMPORARY);
2346         if (!page)
2347                 goto out;
2348
2349         length = -EFAULT;
2350         if (copy_from_user(page, buf, count))
2351                 goto out_free;
2352
2353         /* Guard against adverse ptrace interaction */
2354         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2355         if (length < 0)
2356                 goto out_free;
2357
2358         length = security_setprocattr(task,
2359                                       (char*)file->f_path.dentry->d_name.name,
2360                                       (void*)page, count);
2361         mutex_unlock(&task->cred_guard_mutex);
2362 out_free:
2363         free_page((unsigned long) page);
2364 out:
2365         put_task_struct(task);
2366 out_no_task:
2367         return length;
2368 }
2369
2370 static const struct file_operations proc_pid_attr_operations = {
2371         .read           = proc_pid_attr_read,
2372         .write          = proc_pid_attr_write,
2373         .llseek         = generic_file_llseek,
2374 };
2375
2376 static const struct pid_entry attr_dir_stuff[] = {
2377         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2378         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2379         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2380         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2381         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2382         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2383 };
2384
2385 static int proc_attr_dir_readdir(struct file * filp,
2386                              void * dirent, filldir_t filldir)
2387 {
2388         return proc_pident_readdir(filp,dirent,filldir,
2389                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2390 }
2391
2392 static const struct file_operations proc_attr_dir_operations = {
2393         .read           = generic_read_dir,
2394         .readdir        = proc_attr_dir_readdir,
2395         .llseek         = default_llseek,
2396 };
2397
2398 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2399                                 struct dentry *dentry, struct nameidata *nd)
2400 {
2401         return proc_pident_lookup(dir, dentry,
2402                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2403 }
2404
2405 static const struct inode_operations proc_attr_dir_inode_operations = {
2406         .lookup         = proc_attr_dir_lookup,
2407         .getattr        = pid_getattr,
2408         .setattr        = proc_setattr,
2409 };
2410
2411 #endif
2412
2413 #ifdef CONFIG_ELF_CORE
2414 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2415                                          size_t count, loff_t *ppos)
2416 {
2417         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2418         struct mm_struct *mm;
2419         char buffer[PROC_NUMBUF];
2420         size_t len;
2421         int ret;
2422
2423         if (!task)
2424                 return -ESRCH;
2425
2426         ret = 0;
2427         mm = get_task_mm(task);
2428         if (mm) {
2429                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2430                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2431                                 MMF_DUMP_FILTER_SHIFT));
2432                 mmput(mm);
2433                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2434         }
2435
2436         put_task_struct(task);
2437
2438         return ret;
2439 }
2440
2441 static ssize_t proc_coredump_filter_write(struct file *file,
2442                                           const char __user *buf,
2443                                           size_t count,
2444                                           loff_t *ppos)
2445 {
2446         struct task_struct *task;
2447         struct mm_struct *mm;
2448         char buffer[PROC_NUMBUF], *end;
2449         unsigned int val;
2450         int ret;
2451         int i;
2452         unsigned long mask;
2453
2454         ret = -EFAULT;
2455         memset(buffer, 0, sizeof(buffer));
2456         if (count > sizeof(buffer) - 1)
2457                 count = sizeof(buffer) - 1;
2458         if (copy_from_user(buffer, buf, count))
2459                 goto out_no_task;
2460
2461         ret = -EINVAL;
2462         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2463         if (*end == '\n')
2464                 end++;
2465         if (end - buffer == 0)
2466                 goto out_no_task;
2467
2468         ret = -ESRCH;
2469         task = get_proc_task(file->f_dentry->d_inode);
2470         if (!task)
2471                 goto out_no_task;
2472
2473         ret = end - buffer;
2474         mm = get_task_mm(task);
2475         if (!mm)
2476                 goto out_no_mm;
2477
2478         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2479                 if (val & mask)
2480                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2481                 else
2482                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2483         }
2484
2485         mmput(mm);
2486  out_no_mm:
2487         put_task_struct(task);
2488  out_no_task:
2489         return ret;
2490 }
2491
2492 static const struct file_operations proc_coredump_filter_operations = {
2493         .read           = proc_coredump_filter_read,
2494         .write          = proc_coredump_filter_write,
2495         .llseek         = generic_file_llseek,
2496 };
2497 #endif
2498
2499 /*
2500  * /proc/self:
2501  */
2502 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2503                               int buflen)
2504 {
2505         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2506         pid_t tgid = task_tgid_nr_ns(current, ns);
2507         char tmp[PROC_NUMBUF];
2508         if (!tgid)
2509                 return -ENOENT;
2510         sprintf(tmp, "%d", tgid);
2511         return vfs_readlink(dentry,buffer,buflen,tmp);
2512 }
2513
2514 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2515 {
2516         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2517         pid_t tgid = task_tgid_nr_ns(current, ns);
2518         char *name = ERR_PTR(-ENOENT);
2519         if (tgid) {
2520                 name = __getname();
2521                 if (!name)
2522                         name = ERR_PTR(-ENOMEM);
2523                 else
2524                         sprintf(name, "%d", tgid);
2525         }
2526         nd_set_link(nd, name);
2527         return NULL;
2528 }
2529
2530 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2531                                 void *cookie)
2532 {
2533         char *s = nd_get_link(nd);
2534         if (!IS_ERR(s))
2535                 __putname(s);
2536 }
2537
2538 static const struct inode_operations proc_self_inode_operations = {
2539         .readlink       = proc_self_readlink,
2540         .follow_link    = proc_self_follow_link,
2541         .put_link       = proc_self_put_link,
2542 };
2543
2544 /*
2545  * proc base
2546  *
2547  * These are the directory entries in the root directory of /proc
2548  * that properly belong to the /proc filesystem, as they describe
2549  * describe something that is process related.
2550  */
2551 static const struct pid_entry proc_base_stuff[] = {
2552         NOD("self", S_IFLNK|S_IRWXUGO,
2553                 &proc_self_inode_operations, NULL, {}),
2554 };
2555
2556 /*
2557  *      Exceptional case: normally we are not allowed to unhash a busy
2558  * directory. In this case, however, we can do it - no aliasing problems
2559  * due to the way we treat inodes.
2560  */
2561 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2562 {
2563         struct inode *inode = dentry->d_inode;
2564         struct task_struct *task = get_proc_task(inode);
2565         if (task) {
2566                 put_task_struct(task);
2567                 return 1;
2568         }
2569         d_drop(dentry);
2570         return 0;
2571 }
2572
2573 static const struct dentry_operations proc_base_dentry_operations =
2574 {
2575         .d_revalidate   = proc_base_revalidate,
2576         .d_delete       = pid_delete_dentry,
2577 };
2578
2579 static struct dentry *proc_base_instantiate(struct inode *dir,
2580         struct dentry *dentry, struct task_struct *task, const void *ptr)
2581 {
2582         const struct pid_entry *p = ptr;
2583         struct inode *inode;
2584         struct proc_inode *ei;
2585         struct dentry *error;
2586
2587         /* Allocate the inode */
2588         error = ERR_PTR(-ENOMEM);
2589         inode = new_inode(dir->i_sb);
2590         if (!inode)
2591                 goto out;
2592
2593         /* Initialize the inode */
2594         ei = PROC_I(inode);
2595         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2596
2597         /*
2598          * grab the reference to the task.
2599          */
2600         ei->pid = get_task_pid(task, PIDTYPE_PID);
2601         if (!ei->pid)
2602                 goto out_iput;
2603
2604         inode->i_mode = p->mode;
2605         if (S_ISDIR(inode->i_mode))
2606                 inode->i_nlink = 2;
2607         if (S_ISLNK(inode->i_mode))
2608                 inode->i_size = 64;
2609         if (p->iop)
2610                 inode->i_op = p->iop;
2611         if (p->fop)
2612                 inode->i_fop = p->fop;
2613         ei->op = p->op;
2614         dentry->d_op = &proc_base_dentry_operations;
2615         d_add(dentry, inode);
2616         error = NULL;
2617 out:
2618         return error;
2619 out_iput:
2620         iput(inode);
2621         goto out;
2622 }
2623
2624 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2625 {
2626         struct dentry *error;
2627         struct task_struct *task = get_proc_task(dir);
2628         const struct pid_entry *p, *last;
2629
2630         error = ERR_PTR(-ENOENT);
2631
2632         if (!task)
2633                 goto out_no_task;
2634
2635         /* Lookup the directory entry */
2636         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2637         for (p = proc_base_stuff; p <= last; p++) {
2638                 if (p->len != dentry->d_name.len)
2639                         continue;
2640                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2641                         break;
2642         }
2643         if (p > last)
2644                 goto out;
2645
2646         error = proc_base_instantiate(dir, dentry, task, p);
2647
2648 out:
2649         put_task_struct(task);
2650 out_no_task:
2651         return error;
2652 }
2653
2654 static int proc_base_fill_cache(struct file *filp, void *dirent,
2655         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2656 {
2657         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2658                                 proc_base_instantiate, task, p);
2659 }
2660
2661 #ifdef CONFIG_TASK_IO_ACCOUNTING
2662 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2663 {
2664         struct task_io_accounting acct = task->ioac;
2665         unsigned long flags;
2666
2667         if (whole && lock_task_sighand(task, &flags)) {
2668                 struct task_struct *t = task;
2669
2670                 task_io_accounting_add(&acct, &task->signal->ioac);
2671                 while_each_thread(task, t)
2672                         task_io_accounting_add(&acct, &t->ioac);
2673
2674                 unlock_task_sighand(task, &flags);
2675         }
2676         return sprintf(buffer,
2677                         "rchar: %llu\n"
2678                         "wchar: %llu\n"
2679                         "syscr: %llu\n"
2680                         "syscw: %llu\n"
2681                         "read_bytes: %llu\n"
2682                         "write_bytes: %llu\n"
2683                         "cancelled_write_bytes: %llu\n",
2684                         (unsigned long long)acct.rchar,
2685                         (unsigned long long)acct.wchar,
2686                         (unsigned long long)acct.syscr,
2687                         (unsigned long long)acct.syscw,
2688                         (unsigned long long)acct.read_bytes,
2689                         (unsigned long long)acct.write_bytes,
2690                         (unsigned long long)acct.cancelled_write_bytes);
2691 }
2692
2693 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2694 {
2695         return do_io_accounting(task, buffer, 0);
2696 }
2697
2698 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2699 {
2700         return do_io_accounting(task, buffer, 1);
2701 }
2702 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2703
2704 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2705                                 struct pid *pid, struct task_struct *task)
2706 {
2707         seq_printf(m, "%08x\n", task->personality);
2708         return 0;
2709 }
2710
2711 /*
2712  * Thread groups
2713  */
2714 static const struct file_operations proc_task_operations;
2715 static const struct inode_operations proc_task_inode_operations;
2716
2717 static const struct pid_entry tgid_base_stuff[] = {
2718         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2719         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2720         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2721 #ifdef CONFIG_NET
2722         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2723 #endif
2724         REG("environ",    S_IRUSR, proc_environ_operations),
2725         INF("auxv",       S_IRUSR, proc_pid_auxv),
2726         ONE("status",     S_IRUGO, proc_pid_status),
2727         ONE("personality", S_IRUSR, proc_pid_personality),
2728         INF("limits",     S_IRUGO, proc_pid_limits),
2729 #ifdef CONFIG_SCHED_DEBUG
2730         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2731 #endif
2732         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2733 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2734         INF("syscall",    S_IRUSR, proc_pid_syscall),
2735 #endif
2736         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2737         ONE("stat",       S_IRUGO, proc_tgid_stat),
2738         ONE("statm",      S_IRUGO, proc_pid_statm),
2739         REG("maps",       S_IRUGO, proc_maps_operations),
2740 #ifdef CONFIG_NUMA
2741         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2742 #endif
2743         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2744         LNK("cwd",        proc_cwd_link),
2745         LNK("root",       proc_root_link),
2746         LNK("exe",        proc_exe_link),
2747         REG("mounts",     S_IRUGO, proc_mounts_operations),
2748         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2749         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2750 #ifdef CONFIG_PROC_PAGE_MONITOR
2751         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2752         REG("smaps",      S_IRUGO, proc_smaps_operations),
2753         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2754 #endif
2755 #ifdef CONFIG_SECURITY
2756         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2757 #endif
2758 #ifdef CONFIG_KALLSYMS
2759         INF("wchan",      S_IRUGO, proc_pid_wchan),
2760 #endif
2761 #ifdef CONFIG_STACKTRACE
2762         ONE("stack",      S_IRUSR, proc_pid_stack),
2763 #endif
2764 #ifdef CONFIG_SCHEDSTATS
2765         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2766 #endif
2767 #ifdef CONFIG_LATENCYTOP
2768         REG("latency",  S_IRUGO, proc_lstats_operations),
2769 #endif
2770 #ifdef CONFIG_PROC_PID_CPUSET
2771         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2772 #endif
2773 #ifdef CONFIG_CGROUPS
2774         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2775 #endif
2776         INF("oom_score",  S_IRUGO, proc_oom_score),
2777         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2778         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2779 #ifdef CONFIG_AUDITSYSCALL
2780         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2781         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2782 #endif
2783 #ifdef CONFIG_FAULT_INJECTION
2784         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2785 #endif
2786 #ifdef CONFIG_ELF_CORE
2787         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2788 #endif
2789 #ifdef CONFIG_TASK_IO_ACCOUNTING
2790         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2791 #endif
2792 };
2793
2794 static int proc_tgid_base_readdir(struct file * filp,
2795                              void * dirent, filldir_t filldir)
2796 {
2797         return proc_pident_readdir(filp,dirent,filldir,
2798                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2799 }
2800
2801 static const struct file_operations proc_tgid_base_operations = {
2802         .read           = generic_read_dir,
2803         .readdir        = proc_tgid_base_readdir,
2804         .llseek         = default_llseek,
2805 };
2806
2807 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2808         return proc_pident_lookup(dir, dentry,
2809                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2810 }
2811
2812 static const struct inode_operations proc_tgid_base_inode_operations = {
2813         .lookup         = proc_tgid_base_lookup,
2814         .getattr        = pid_getattr,
2815         .setattr        = proc_setattr,
2816 };
2817
2818 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2819 {
2820         struct dentry *dentry, *leader, *dir;
2821         char buf[PROC_NUMBUF];
2822         struct qstr name;
2823
2824         name.name = buf;
2825         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2826         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2827         if (dentry) {
2828                 shrink_dcache_parent(dentry);
2829                 d_drop(dentry);
2830                 dput(dentry);
2831         }
2832
2833         name.name = buf;
2834         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2835         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2836         if (!leader)
2837                 goto out;
2838
2839         name.name = "task";
2840         name.len = strlen(name.name);
2841         dir = d_hash_and_lookup(leader, &name);
2842         if (!dir)
2843                 goto out_put_leader;
2844
2845         name.name = buf;
2846         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2847         dentry = d_hash_and_lookup(dir, &name);
2848         if (dentry) {
2849                 shrink_dcache_parent(dentry);
2850                 d_drop(dentry);
2851                 dput(dentry);
2852         }
2853
2854         dput(dir);
2855 out_put_leader:
2856         dput(leader);
2857 out:
2858         return;
2859 }
2860
2861 /**
2862  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2863  * @task: task that should be flushed.
2864  *
2865  * When flushing dentries from proc, one needs to flush them from global
2866  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2867  * in. This call is supposed to do all of this job.
2868  *
2869  * Looks in the dcache for
2870  * /proc/@pid
2871  * /proc/@tgid/task/@pid
2872  * if either directory is present flushes it and all of it'ts children
2873  * from the dcache.
2874  *
2875  * It is safe and reasonable to cache /proc entries for a task until
2876  * that task exits.  After that they just clog up the dcache with
2877  * useless entries, possibly causing useful dcache entries to be
2878  * flushed instead.  This routine is proved to flush those useless
2879  * dcache entries at process exit time.
2880  *
2881  * NOTE: This routine is just an optimization so it does not guarantee
2882  *       that no dcache entries will exist at process exit time it
2883  *       just makes it very unlikely that any will persist.
2884  */
2885
2886 void proc_flush_task(struct task_struct *task)
2887 {
2888         int i;
2889         struct pid *pid, *tgid;
2890         struct upid *upid;
2891
2892         pid = task_pid(task);
2893         tgid = task_tgid(task);
2894
2895         for (i = 0; i <= pid->level; i++) {
2896                 upid = &pid->numbers[i];
2897                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2898                                         tgid->numbers[i].nr);
2899         }
2900
2901         upid = &pid->numbers[pid->level];
2902         if (upid->nr == 1)
2903                 pid_ns_release_proc(upid->ns);
2904 }
2905
2906 static struct dentry *proc_pid_instantiate(struct inode *dir,
2907                                            struct dentry * dentry,
2908                                            struct task_struct *task, const void *ptr)
2909 {
2910         struct dentry *error = ERR_PTR(-ENOENT);
2911         struct inode *inode;
2912
2913         inode = proc_pid_make_inode(dir->i_sb, task);
2914         if (!inode)
2915                 goto out;
2916
2917         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2918         inode->i_op = &proc_tgid_base_inode_operations;
2919         inode->i_fop = &proc_tgid_base_operations;
2920         inode->i_flags|=S_IMMUTABLE;
2921
2922         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2923                 ARRAY_SIZE(tgid_base_stuff));
2924
2925         dentry->d_op = &pid_dentry_operations;
2926
2927         d_add(dentry, inode);
2928         /* Close the race of the process dying before we return the dentry */
2929         if (pid_revalidate(dentry, NULL))
2930                 error = NULL;
2931 out:
2932         return error;
2933 }
2934
2935 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2936 {
2937         struct dentry *result;
2938         struct task_struct *task;
2939         unsigned tgid;
2940         struct pid_namespace *ns;
2941
2942         result = proc_base_lookup(dir, dentry);
2943         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2944                 goto out;
2945
2946         tgid = name_to_int(dentry);
2947         if (tgid == ~0U)
2948                 goto out;
2949
2950         ns = dentry->d_sb->s_fs_info;
2951         rcu_read_lock();
2952         task = find_task_by_pid_ns(tgid, ns);
2953         if (task)
2954                 get_task_struct(task);
2955         rcu_read_unlock();
2956         if (!task)
2957                 goto out;
2958
2959         result = proc_pid_instantiate(dir, dentry, task, NULL);
2960         put_task_struct(task);
2961 out:
2962         return result;
2963 }
2964
2965 /*
2966  * Find the first task with tgid >= tgid
2967  *
2968  */
2969 struct tgid_iter {
2970         unsigned int tgid;
2971         struct task_struct *task;
2972 };
2973 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2974 {
2975         struct pid *pid;
2976
2977         if (iter.task)
2978                 put_task_struct(iter.task);
2979         rcu_read_lock();
2980 retry:
2981         iter.task = NULL;
2982         pid = find_ge_pid(iter.tgid, ns);
2983         if (pid) {
2984                 iter.tgid = pid_nr_ns(pid, ns);
2985                 iter.task = pid_task(pid, PIDTYPE_PID);
2986                 /* What we to know is if the pid we have find is the
2987                  * pid of a thread_group_leader.  Testing for task
2988                  * being a thread_group_leader is the obvious thing
2989                  * todo but there is a window when it fails, due to
2990                  * the pid transfer logic in de_thread.
2991                  *
2992                  * So we perform the straight forward test of seeing
2993                  * if the pid we have found is the pid of a thread
2994                  * group leader, and don't worry if the task we have
2995                  * found doesn't happen to be a thread group leader.
2996                  * As we don't care in the case of readdir.
2997                  */
2998                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2999                         iter.tgid += 1;
3000                         goto retry;
3001                 }
3002                 get_task_struct(iter.task);
3003         }
3004         rcu_read_unlock();
3005         return iter;
3006 }
3007
3008 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3009
3010 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3011         struct tgid_iter iter)
3012 {
3013         char name[PROC_NUMBUF];
3014         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3015         return proc_fill_cache(filp, dirent, filldir, name, len,
3016                                 proc_pid_instantiate, iter.task, NULL);
3017 }
3018
3019 /* for the /proc/ directory itself, after non-process stuff has been done */
3020 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3021 {
3022         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3023         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3024         struct tgid_iter iter;
3025         struct pid_namespace *ns;
3026
3027         if (!reaper)
3028                 goto out_no_task;
3029
3030         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3031                 const struct pid_entry *p = &proc_base_stuff[nr];
3032                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3033                         goto out;
3034         }
3035
3036         ns = filp->f_dentry->d_sb->s_fs_info;
3037         iter.task = NULL;
3038         iter.tgid = filp->f_pos - TGID_OFFSET;
3039         for (iter = next_tgid(ns, iter);
3040              iter.task;
3041              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3042                 filp->f_pos = iter.tgid + TGID_OFFSET;
3043                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3044                         put_task_struct(iter.task);
3045                         goto out;
3046                 }
3047         }
3048         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3049 out:
3050         put_task_struct(reaper);
3051 out_no_task:
3052         return 0;
3053 }
3054
3055 /*
3056  * Tasks
3057  */
3058 static const struct pid_entry tid_base_stuff[] = {
3059         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3060         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3061         REG("environ",   S_IRUSR, proc_environ_operations),
3062         INF("auxv",      S_IRUSR, proc_pid_auxv),
3063         ONE("status",    S_IRUGO, proc_pid_status),
3064         ONE("personality", S_IRUSR, proc_pid_personality),
3065         INF("limits",    S_IRUGO, proc_pid_limits),
3066 #ifdef CONFIG_SCHED_DEBUG
3067         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3068 #endif
3069         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3070 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3071         INF("syscall",   S_IRUSR, proc_pid_syscall),
3072 #endif
3073         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3074         ONE("stat",      S_IRUGO, proc_tid_stat),
3075         ONE("statm",     S_IRUGO, proc_pid_statm),
3076         REG("maps",      S_IRUGO, proc_maps_operations),
3077 #ifdef CONFIG_NUMA
3078         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3079 #endif
3080         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3081         LNK("cwd",       proc_cwd_link),
3082         LNK("root",      proc_root_link),
3083         LNK("exe",       proc_exe_link),
3084         REG("mounts",    S_IRUGO, proc_mounts_operations),
3085         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3086 #ifdef CONFIG_PROC_PAGE_MONITOR
3087         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3088         REG("smaps",     S_IRUGO, proc_smaps_operations),
3089         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3090 #endif
3091 #ifdef CONFIG_SECURITY
3092         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3093 #endif
3094 #ifdef CONFIG_KALLSYMS
3095         INF("wchan",     S_IRUGO, proc_pid_wchan),
3096 #endif
3097 #ifdef CONFIG_STACKTRACE
3098         ONE("stack",      S_IRUSR, proc_pid_stack),
3099 #endif
3100 #ifdef CONFIG_SCHEDSTATS
3101         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3102 #endif
3103 #ifdef CONFIG_LATENCYTOP
3104         REG("latency",  S_IRUGO, proc_lstats_operations),
3105 #endif
3106 #ifdef CONFIG_PROC_PID_CPUSET
3107         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3108 #endif
3109 #ifdef CONFIG_CGROUPS
3110         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3111 #endif
3112         INF("oom_score", S_IRUGO, proc_oom_score),
3113         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3114         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3115 #ifdef CONFIG_AUDITSYSCALL
3116         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3117         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3118 #endif
3119 #ifdef CONFIG_FAULT_INJECTION
3120         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3121 #endif
3122 #ifdef CONFIG_TASK_IO_ACCOUNTING
3123         INF("io",       S_IRUGO, proc_tid_io_accounting),
3124 #endif
3125 };
3126
3127 static int proc_tid_base_readdir(struct file * filp,
3128                              void * dirent, filldir_t filldir)
3129 {
3130         return proc_pident_readdir(filp,dirent,filldir,
3131                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3132 }
3133
3134 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3135         return proc_pident_lookup(dir, dentry,
3136                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3137 }
3138
3139 static const struct file_operations proc_tid_base_operations = {
3140         .read           = generic_read_dir,
3141         .readdir        = proc_tid_base_readdir,
3142         .llseek         = default_llseek,
3143 };
3144
3145 static const struct inode_operations proc_tid_base_inode_operations = {
3146         .lookup         = proc_tid_base_lookup,
3147         .getattr        = pid_getattr,
3148         .setattr        = proc_setattr,
3149 };
3150
3151 static struct dentry *proc_task_instantiate(struct inode *dir,
3152         struct dentry *dentry, struct task_struct *task, const void *ptr)
3153 {
3154         struct dentry *error = ERR_PTR(-ENOENT);
3155         struct inode *inode;
3156         inode = proc_pid_make_inode(dir->i_sb, task);
3157
3158         if (!inode)
3159                 goto out;
3160         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3161         inode->i_op = &proc_tid_base_inode_operations;
3162         inode->i_fop = &proc_tid_base_operations;
3163         inode->i_flags|=S_IMMUTABLE;
3164
3165         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3166                 ARRAY_SIZE(tid_base_stuff));
3167
3168         dentry->d_op = &pid_dentry_operations;
3169
3170         d_add(dentry, inode);
3171         /* Close the race of the process dying before we return the dentry */
3172         if (pid_revalidate(dentry, NULL))
3173                 error = NULL;
3174 out:
3175         return error;
3176 }
3177
3178 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3179 {
3180         struct dentry *result = ERR_PTR(-ENOENT);
3181         struct task_struct *task;
3182         struct task_struct *leader = get_proc_task(dir);
3183         unsigned tid;
3184         struct pid_namespace *ns;
3185
3186         if (!leader)
3187                 goto out_no_task;
3188
3189         tid = name_to_int(dentry);
3190         if (tid == ~0U)
3191                 goto out;
3192
3193         ns = dentry->d_sb->s_fs_info;
3194         rcu_read_lock();
3195         task = find_task_by_pid_ns(tid, ns);
3196         if (task)
3197                 get_task_struct(task);
3198         rcu_read_unlock();
3199         if (!task)
3200                 goto out;
3201         if (!same_thread_group(leader, task))
3202                 goto out_drop_task;
3203
3204         result = proc_task_instantiate(dir, dentry, task, NULL);
3205 out_drop_task:
3206         put_task_struct(task);
3207 out:
3208         put_task_struct(leader);
3209 out_no_task:
3210         return result;
3211 }
3212
3213 /*
3214  * Find the first tid of a thread group to return to user space.
3215  *
3216  * Usually this is just the thread group leader, but if the users
3217  * buffer was too small or there was a seek into the middle of the
3218  * directory we have more work todo.
3219  *
3220  * In the case of a short read we start with find_task_by_pid.
3221  *
3222  * In the case of a seek we start with the leader and walk nr
3223  * threads past it.
3224  */
3225 static struct task_struct *first_tid(struct task_struct *leader,
3226                 int tid, int nr, struct pid_namespace *ns)
3227 {
3228         struct task_struct *pos;
3229
3230         rcu_read_lock();
3231         /* Attempt to start with the pid of a thread */
3232         if (tid && (nr > 0)) {
3233                 pos = find_task_by_pid_ns(tid, ns);
3234                 if (pos && (pos->group_leader == leader))
3235                         goto found;
3236         }
3237
3238         /* If nr exceeds the number of threads there is nothing todo */
3239         pos = NULL;
3240         if (nr && nr >= get_nr_threads(leader))
3241                 goto out;
3242
3243         /* If we haven't found our starting place yet start
3244          * with the leader and walk nr threads forward.
3245          */
3246         for (pos = leader; nr > 0; --nr) {
3247                 pos = next_thread(pos);
3248                 if (pos == leader) {
3249                         pos = NULL;
3250                         goto out;
3251                 }
3252         }
3253 found:
3254         get_task_struct(pos);
3255 out:
3256         rcu_read_unlock();
3257         return pos;
3258 }
3259
3260 /*
3261  * Find the next thread in the thread list.
3262  * Return NULL if there is an error or no next thread.
3263  *
3264  * The reference to the input task_struct is released.
3265  */
3266 static struct task_struct *next_tid(struct task_struct *start)
3267 {
3268         struct task_struct *pos = NULL;
3269         rcu_read_lock();
3270         if (pid_alive(start)) {
3271                 pos = next_thread(start);
3272                 if (thread_group_leader(pos))
3273                         pos = NULL;
3274                 else
3275                         get_task_struct(pos);
3276         }
3277         rcu_read_unlock();
3278         put_task_struct(start);
3279         return pos;
3280 }
3281
3282 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3283         struct task_struct *task, int tid)
3284 {
3285         char name[PROC_NUMBUF];
3286         int len = snprintf(name, sizeof(name), "%d", tid);
3287         return proc_fill_cache(filp, dirent, filldir, name, len,
3288                                 proc_task_instantiate, task, NULL);
3289 }
3290
3291 /* for the /proc/TGID/task/ directories */
3292 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3293 {
3294         struct dentry *dentry = filp->f_path.dentry;
3295         struct inode *inode = dentry->d_inode;
3296         struct task_struct *leader = NULL;
3297         struct task_struct *task;
3298         int retval = -ENOENT;
3299         ino_t ino;
3300         int tid;
3301         struct pid_namespace *ns;
3302
3303         task = get_proc_task(inode);
3304         if (!task)
3305                 goto out_no_task;
3306         rcu_read_lock();
3307         if (pid_alive(task)) {
3308                 leader = task->group_leader;
3309                 get_task_struct(leader);
3310         }
3311         rcu_read_unlock();
3312         put_task_struct(task);
3313         if (!leader)
3314                 goto out_no_task;
3315         retval = 0;
3316
3317         switch ((unsigned long)filp->f_pos) {
3318         case 0:
3319                 ino = inode->i_ino;
3320                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3321                         goto out;
3322                 filp->f_pos++;
3323                 /* fall through */
3324         case 1:
3325                 ino = parent_ino(dentry);
3326                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3327                         goto out;
3328                 filp->f_pos++;
3329                 /* fall through */
3330         }
3331
3332         /* f_version caches the tgid value that the last readdir call couldn't
3333          * return. lseek aka telldir automagically resets f_version to 0.
3334          */
3335         ns = filp->f_dentry->d_sb->s_fs_info;
3336         tid = (int)filp->f_version;
3337         filp->f_version = 0;
3338         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3339              task;
3340              task = next_tid(task), filp->f_pos++) {
3341                 tid = task_pid_nr_ns(task, ns);
3342                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3343                         /* returning this tgid failed, save it as the first
3344                          * pid for the next readir call */
3345                         filp->f_version = (u64)tid;
3346                         put_task_struct(task);
3347                         break;
3348                 }
3349         }
3350 out:
3351         put_task_struct(leader);
3352 out_no_task:
3353         return retval;
3354 }
3355
3356 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3357 {
3358         struct inode *inode = dentry->d_inode;
3359         struct task_struct *p = get_proc_task(inode);
3360         generic_fillattr(inode, stat);
3361
3362         if (p) {
3363                 stat->nlink += get_nr_threads(p);
3364                 put_task_struct(p);
3365         }
3366
3367         return 0;
3368 }
3369
3370 static const struct inode_operations proc_task_inode_operations = {
3371         .lookup         = proc_task_lookup,
3372         .getattr        = proc_task_getattr,
3373         .setattr        = proc_setattr,
3374 };
3375
3376 static const struct file_operations proc_task_operations = {
3377         .read           = generic_read_dir,
3378         .readdir        = proc_task_readdir,
3379         .llseek         = default_llseek,
3380 };