fs/proc/base.c

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