fs/proc/base.c

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