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->signal->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->signal->cred_guard_mutex);
 239
 240         return mm;
 241 }
 242
 243 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 244 {
 245         int res = 0;
 246         unsigned int len;
 247         struct mm_struct *mm = get_task_mm(task);
 248         if (!mm)
 249                 goto out;
 250         if (!mm->arg_end)
 251                 goto out_mm;    /* Shh! No looking before we're done */
 252
 253         len = mm->arg_end - mm->arg_start;
 254
 255         if (len > PAGE_SIZE)
 256                 len = PAGE_SIZE;
 257
 258         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 259
 260         // If the nul at the end of args has been overwritten, then
 261         // assume application is using setproctitle(3).
 262         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 263                 len = strnlen(buffer, res);
 264                 if (len < res) {
 265                     res = len;
 266                 } else {
 267                         len = mm->env_end - mm->env_start;
 268                         if (len > PAGE_SIZE - res)
 269                                 len = PAGE_SIZE - res;
 270                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 271                         res = strnlen(buffer, res);
 272                 }
 273         }
 274 out_mm:
 275         mmput(mm);
 276 out:
 277         return res;
 278 }
 279
 280 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 281 {
 282         int res = 0;
 283         struct mm_struct *mm = get_task_mm(task);
 284         if (mm) {
 285                 unsigned int nwords = 0;
 286                 do {
 287                         nwords += 2;
 288                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 289                 res = nwords * sizeof(mm->saved_auxv[0]);
 290                 if (res > PAGE_SIZE)
 291                         res = PAGE_SIZE;
 292                 memcpy(buffer, mm->saved_auxv, res);
 293                 mmput(mm);
 294         }
 295         return res;
 296 }
 297
 298
 299 #ifdef CONFIG_KALLSYMS
 300 /*
 301  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 302  * Returns the resolved symbol.  If that fails, simply return the address.
 303  */
 304 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 305 {
 306         unsigned long wchan;
 307         char symname[KSYM_NAME_LEN];
 308
 309         wchan = get_wchan(task);
 310
 311         if (lookup_symbol_name(wchan, symname) < 0)
 312                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 313                         return 0;
 314                 else
 315                         return sprintf(buffer, "%lu", wchan);
 316         else
 317                 return sprintf(buffer, "%s", symname);
 318 }
 319 #endif /* CONFIG_KALLSYMS */
 320
 321 #ifdef CONFIG_STACKTRACE
 322
 323 #define MAX_STACK_TRACE_DEPTH   64
 324
 325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 326                           struct pid *pid, struct task_struct *task)
 327 {
 328         struct stack_trace trace;
 329         unsigned long *entries;
 330         int i;
 331
 332         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 333         if (!entries)
 334                 return -ENOMEM;
 335
 336         trace.nr_entries        = 0;
 337         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 338         trace.entries           = entries;
 339         trace.skip              = 0;
 340         save_stack_trace_tsk(task, &trace);
 341
 342         for (i = 0; i < trace.nr_entries; i++) {
 343                 seq_printf(m, "[<%p>] %pS\n",
 344                            (void *)entries[i], (void *)entries[i]);
 345         }
 346         kfree(entries);
 347
 348         return 0;
 349 }
 350 #endif
 351
 352 #ifdef CONFIG_SCHEDSTATS
 353 /*
 354  * Provides /proc/PID/schedstat
 355  */
 356 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 357 {
 358         return sprintf(buffer, "%llu %llu %lu\n",
 359                         (unsigned long long)task->se.sum_exec_runtime,
 360                         (unsigned long long)task->sched_info.run_delay,
 361                         task->sched_info.pcount);
 362 }
 363 #endif
 364
 365 #ifdef CONFIG_LATENCYTOP
 366 static int lstats_show_proc(struct seq_file *m, void *v)
 367 {
 368         int i;
 369         struct inode *inode = m->private;
 370         struct task_struct *task = get_proc_task(inode);
 371
 372         if (!task)
 373                 return -ESRCH;
 374         seq_puts(m, "Latency Top version : v0.1\n");
 375         for (i = 0; i < 32; i++) {
 376                 if (task->latency_record[i].backtrace[0]) {
 377                         int q;
 378                         seq_printf(m, "%i %li %li ",
 379                                 task->latency_record[i].count,
 380                                 task->latency_record[i].time,
 381                                 task->latency_record[i].max);
 382                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 383                                 char sym[KSYM_SYMBOL_LEN];
 384                                 char *c;
 385                                 if (!task->latency_record[i].backtrace[q])
 386                                         break;
 387                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 388                                         break;
 389                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 390                                 c = strchr(sym, '+');
 391                                 if (c)
 392                                         *c = 0;
 393                                 seq_printf(m, "%s ", sym);
 394                         }
 395                         seq_printf(m, "\n");
 396                 }
 397
 398         }
 399         put_task_struct(task);
 400         return 0;
 401 }
 402
 403 static int lstats_open(struct inode *inode, struct file *file)
 404 {
 405         return single_open(file, lstats_show_proc, inode);
 406 }
 407
 408 static ssize_t lstats_write(struct file *file, const char __user *buf,
 409                             size_t count, loff_t *offs)
 410 {
 411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 412
 413         if (!task)
 414                 return -ESRCH;
 415         clear_all_latency_tracing(task);
 416         put_task_struct(task);
 417
 418         return count;
 419 }
 420
 421 static const struct file_operations proc_lstats_operations = {
 422         .open           = lstats_open,
 423         .read           = seq_read,
 424         .write          = lstats_write,
 425         .llseek         = seq_lseek,
 426         .release        = single_release,
 427 };
 428
 429 #endif
 430
 431 static int proc_oom_score(struct task_struct *task, char *buffer)
 432 {
 433         unsigned long points = 0;
 434
 435         read_lock(&tasklist_lock);
 436         if (pid_alive(task))
 437                 points = oom_badness(task, NULL, NULL,
 438                                         totalram_pages + total_swap_pages);
 439         read_unlock(&tasklist_lock);
 440         return sprintf(buffer, "%lu\n", points);
 441 }
 442
 443 struct limit_names {
 444         char *name;
 445         char *unit;
 446 };
 447
 448 static const struct limit_names lnames[RLIM_NLIMITS] = {
 449         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 450         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 451         [RLIMIT_DATA] = {"Max data size", "bytes"},
 452         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 453         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 454         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 455         [RLIMIT_NPROC] = {"Max processes", "processes"},
 456         [RLIMIT_NOFILE] = {"Max open files", "files"},
 457         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 458         [RLIMIT_AS] = {"Max address space", "bytes"},
 459         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 460         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 461         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 462         [RLIMIT_NICE] = {"Max nice priority", NULL},
 463         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 464         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 465 };
 466
 467 /* Display limits for a process */
 468 static int proc_pid_limits(struct task_struct *task, char *buffer)
 469 {
 470         unsigned int i;
 471         int count = 0;
 472         unsigned long flags;
 473         char *bufptr = buffer;
 474
 475         struct rlimit rlim[RLIM_NLIMITS];
 476
 477         if (!lock_task_sighand(task, &flags))
 478                 return 0;
 479         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 480         unlock_task_sighand(task, &flags);
 481
 482         /*
 483          * print the file header
 484          */
 485         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 486                         "Limit", "Soft Limit", "Hard Limit", "Units");
 487
 488         for (i = 0; i < RLIM_NLIMITS; i++) {
 489                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 490                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 491                                          lnames[i].name, "unlimited");
 492                 else
 493                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 494                                          lnames[i].name, rlim[i].rlim_cur);
 495
 496                 if (rlim[i].rlim_max == RLIM_INFINITY)
 497                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 498                 else
 499                         count += sprintf(&bufptr[count], "%-20lu ",
 500                                          rlim[i].rlim_max);
 501
 502                 if (lnames[i].unit)
 503                         count += sprintf(&bufptr[count], "%-10s\n",
 504                                          lnames[i].unit);
 505                 else
 506                         count += sprintf(&bufptr[count], "\n");
 507         }
 508
 509         return count;
 510 }
 511
 512 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 513 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 514 {
 515         long nr;
 516         unsigned long args[6], sp, pc;
 517
 518         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 519                 return sprintf(buffer, "running\n");
 520
 521         if (nr < 0)
 522                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 523
 524         return sprintf(buffer,
 525                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 526                        nr,
 527                        args[0], args[1], args[2], args[3], args[4], args[5],
 528                        sp, pc);
 529 }
 530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 531
 532 /************************************************************************/
 533 /*                       Here the fs part begins                        */
 534 /************************************************************************/
 535
 536 /* permission checks */
 537 static int proc_fd_access_allowed(struct inode *inode)
 538 {
 539         struct task_struct *task;
 540         int allowed = 0;
 541         /* Allow access to a task's file descriptors if it is us or we
 542          * may use ptrace attach to the process and find out that
 543          * information.
 544          */
 545         task = get_proc_task(inode);
 546         if (task) {
 547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 548                 put_task_struct(task);
 549         }
 550         return allowed;
 551 }
 552
 553 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 554 {
 555         int error;
 556         struct inode *inode = dentry->d_inode;
 557
 558         if (attr->ia_valid & ATTR_MODE)
 559                 return -EPERM;
 560
 561         error = inode_change_ok(inode, attr);
 562         if (error)
 563                 return error;
 564
 565         if ((attr->ia_valid & ATTR_SIZE) &&
 566             attr->ia_size != i_size_read(inode)) {
 567                 error = vmtruncate(inode, attr->ia_size);
 568                 if (error)
 569                         return error;
 570         }
 571
 572         setattr_copy(inode, attr);
 573         mark_inode_dirty(inode);
 574         return 0;
 575 }
 576
 577 static const struct inode_operations proc_def_inode_operations = {
 578         .setattr        = proc_setattr,
 579 };
 580
 581 static int mounts_open_common(struct inode *inode, struct file *file,
 582                               const struct seq_operations *op)
 583 {
 584         struct task_struct *task = get_proc_task(inode);
 585         struct nsproxy *nsp;
 586         struct mnt_namespace *ns = NULL;
 587         struct path root;
 588         struct proc_mounts *p;
 589         int ret = -EINVAL;
 590
 591         if (task) {
 592                 rcu_read_lock();
 593                 nsp = task_nsproxy(task);
 594                 if (nsp) {
 595                         ns = nsp->mnt_ns;
 596                         if (ns)
 597                                 get_mnt_ns(ns);
 598                 }
 599                 rcu_read_unlock();
 600                 if (ns && get_task_root(task, &root) == 0)
 601                         ret = 0;
 602                 put_task_struct(task);
 603         }
 604
 605         if (!ns)
 606                 goto err;
 607         if (ret)
 608                 goto err_put_ns;
 609
 610         ret = -ENOMEM;
 611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 612         if (!p)
 613                 goto err_put_path;
 614
 615         file->private_data = &p->m;
 616         ret = seq_open(file, op);
 617         if (ret)
 618                 goto err_free;
 619
 620         p->m.private = p;
 621         p->ns = ns;
 622         p->root = root;
 623         p->event = ns->event;
 624
 625         return 0;
 626
 627  err_free:
 628         kfree(p);
 629  err_put_path:
 630         path_put(&root);
 631  err_put_ns:
 632         put_mnt_ns(ns);
 633  err:
 634         return ret;
 635 }
 636
 637 static int mounts_release(struct inode *inode, struct file *file)
 638 {
 639         struct proc_mounts *p = file->private_data;
 640         path_put(&p->root);
 641         put_mnt_ns(p->ns);
 642         return seq_release(inode, file);
 643 }
 644
 645 static unsigned mounts_poll(struct file *file, poll_table *wait)
 646 {
 647         struct proc_mounts *p = file->private_data;
 648         unsigned res = POLLIN | POLLRDNORM;
 649
 650         poll_wait(file, &p->ns->poll, wait);
 651         if (mnt_had_events(p))
 652                 res |= POLLERR | POLLPRI;
 653
 654         return res;
 655 }
 656
 657 static int mounts_open(struct inode *inode, struct file *file)
 658 {
 659         return mounts_open_common(inode, file, &mounts_op);
 660 }
 661
 662 static const struct file_operations proc_mounts_operations = {
 663         .open           = mounts_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = mounts_release,
 667         .poll           = mounts_poll,
 668 };
 669
 670 static int mountinfo_open(struct inode *inode, struct file *file)
 671 {
 672         return mounts_open_common(inode, file, &mountinfo_op);
 673 }
 674
 675 static const struct file_operations proc_mountinfo_operations = {
 676         .open           = mountinfo_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = mounts_release,
 680         .poll           = mounts_poll,
 681 };
 682
 683 static int mountstats_open(struct inode *inode, struct file *file)
 684 {
 685         return mounts_open_common(inode, file, &mountstats_op);
 686 }
 687
 688 static const struct file_operations proc_mountstats_operations = {
 689         .open           = mountstats_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = mounts_release,
 693 };
 694
 695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 696
 697 static ssize_t proc_info_read(struct file * file, char __user * buf,
 698                           size_t count, loff_t *ppos)
 699 {
 700         struct inode * inode = file->f_path.dentry->d_inode;
 701         unsigned long page;
 702         ssize_t length;
 703         struct task_struct *task = get_proc_task(inode);
 704
 705         length = -ESRCH;
 706         if (!task)
 707                 goto out_no_task;
 708
 709         if (count > PROC_BLOCK_SIZE)
 710                 count = PROC_BLOCK_SIZE;
 711
 712         length = -ENOMEM;
 713         if (!(page = __get_free_page(GFP_TEMPORARY)))
 714                 goto out;
 715
 716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 717
 718         if (length >= 0)
 719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 720         free_page(page);
 721 out:
 722         put_task_struct(task);
 723 out_no_task:
 724         return length;
 725 }
 726
 727 static const struct file_operations proc_info_file_operations = {
 728         .read           = proc_info_read,
 729         .llseek         = generic_file_llseek,
 730 };
 731
 732 static int proc_single_show(struct seq_file *m, void *v)
 733 {
 734         struct inode *inode = m->private;
 735         struct pid_namespace *ns;
 736         struct pid *pid;
 737         struct task_struct *task;
 738         int ret;
 739
 740         ns = inode->i_sb->s_fs_info;
 741         pid = proc_pid(inode);
 742         task = get_pid_task(pid, PIDTYPE_PID);
 743         if (!task)
 744                 return -ESRCH;
 745
 746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 747
 748         put_task_struct(task);
 749         return ret;
 750 }
 751
 752 static int proc_single_open(struct inode *inode, struct file *filp)
 753 {
 754         int ret;
 755         ret = single_open(filp, proc_single_show, NULL);
 756         if (!ret) {
 757                 struct seq_file *m = filp->private_data;
 758
 759                 m->private = inode;
 760         }
 761         return ret;
 762 }
 763
 764 static const struct file_operations proc_single_file_operations = {
 765         .open           = proc_single_open,
 766         .read           = seq_read,
 767         .llseek         = seq_lseek,
 768         .release        = single_release,
 769 };
 770
 771 static int mem_open(struct inode* inode, struct file* file)
 772 {
 773         file->private_data = (void*)((long)current->self_exec_id);
 774         /* OK to pass negative loff_t, we can catch out-of-range */
 775         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 776         return 0;
 777 }
 778
 779 static ssize_t mem_read(struct file * file, char __user * buf,
 780                         size_t count, loff_t *ppos)
 781 {
 782         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 783         char *page;
 784         unsigned long src = *ppos;
 785         int ret = -ESRCH;
 786         struct mm_struct *mm;
 787
 788         if (!task)
 789                 goto out_no_task;
 790
 791         if (check_mem_permission(task))
 792                 goto out;
 793
 794         ret = -ENOMEM;
 795         page = (char *)__get_free_page(GFP_TEMPORARY);
 796         if (!page)
 797                 goto out;
 798
 799         ret = 0;
 800
 801         mm = get_task_mm(task);
 802         if (!mm)
 803                 goto out_free;
 804
 805         ret = -EIO;
 806
 807         if (file->private_data != (void*)((long)current->self_exec_id))
 808                 goto out_put;
 809
 810         ret = 0;
 811
 812         while (count > 0) {
 813                 int this_len, retval;
 814
 815                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 816                 retval = access_process_vm(task, src, page, this_len, 0);
 817                 if (!retval || check_mem_permission(task)) {
 818                         if (!ret)
 819                                 ret = -EIO;
 820                         break;
 821                 }
 822
 823                 if (copy_to_user(buf, page, retval)) {
 824                         ret = -EFAULT;
 825                         break;
 826                 }
 827
 828                 ret += retval;
 829                 src += retval;
 830                 buf += retval;
 831                 count -= retval;
 832         }
 833         *ppos = src;
 834
 835 out_put:
 836         mmput(mm);
 837 out_free:
 838         free_page((unsigned long) page);
 839 out:
 840         put_task_struct(task);
 841 out_no_task:
 842         return ret;
 843 }
 844
 845 #define mem_write NULL
 846
 847 #ifndef mem_write
 848 /* This is a security hazard */
 849 static ssize_t mem_write(struct file * file, const char __user *buf,
 850                          size_t count, loff_t *ppos)
 851 {
 852         int copied;
 853         char *page;
 854         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 855         unsigned long dst = *ppos;
 856
 857         copied = -ESRCH;
 858         if (!task)
 859                 goto out_no_task;
 860
 861         if (check_mem_permission(task))
 862                 goto out;
 863
 864         copied = -ENOMEM;
 865         page = (char *)__get_free_page(GFP_TEMPORARY);
 866         if (!page)
 867                 goto out;
 868
 869         copied = 0;
 870         while (count > 0) {
 871                 int this_len, retval;
 872
 873                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 874                 if (copy_from_user(page, buf, this_len)) {
 875                         copied = -EFAULT;
 876                         break;
 877                 }
 878                 retval = access_process_vm(task, dst, page, this_len, 1);
 879                 if (!retval) {
 880                         if (!copied)
 881                                 copied = -EIO;
 882                         break;
 883                 }
 884                 copied += retval;
 885                 buf += retval;
 886                 dst += retval;
 887                 count -= retval;
 888         }
 889         *ppos = dst;
 890         free_page((unsigned long) page);
 891 out:
 892         put_task_struct(task);
 893 out_no_task:
 894         return copied;
 895 }
 896 #endif
 897
 898 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 899 {
 900         switch (orig) {
 901         case 0:
 902                 file->f_pos = offset;
 903                 break;
 904         case 1:
 905                 file->f_pos += offset;
 906                 break;
 907         default:
 908                 return -EINVAL;
 909         }
 910         force_successful_syscall_return();
 911         return file->f_pos;
 912 }
 913
 914 static const struct file_operations proc_mem_operations = {
 915         .llseek         = mem_lseek,
 916         .read           = mem_read,
 917         .write          = mem_write,
 918         .open           = mem_open,
 919 };
 920
 921 static ssize_t environ_read(struct file *file, char __user *buf,
 922                         size_t count, loff_t *ppos)
 923 {
 924         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 925         char *page;
 926         unsigned long src = *ppos;
 927         int ret = -ESRCH;
 928         struct mm_struct *mm;
 929
 930         if (!task)
 931                 goto out_no_task;
 932
 933         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 934                 goto out;
 935
 936         ret = -ENOMEM;
 937         page = (char *)__get_free_page(GFP_TEMPORARY);
 938         if (!page)
 939                 goto out;
 940
 941         ret = 0;
 942
 943         mm = get_task_mm(task);
 944         if (!mm)
 945                 goto out_free;
 946
 947         while (count > 0) {
 948                 int this_len, retval, max_len;
 949
 950                 this_len = mm->env_end - (mm->env_start + src);
 951
 952                 if (this_len <= 0)
 953                         break;
 954
 955                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 956                 this_len = (this_len > max_len) ? max_len : this_len;
 957
 958                 retval = access_process_vm(task, (mm->env_start + src),
 959                         page, this_len, 0);
 960
 961                 if (retval <= 0) {
 962                         ret = retval;
 963                         break;
 964                 }
 965
 966                 if (copy_to_user(buf, page, retval)) {
 967                         ret = -EFAULT;
 968                         break;
 969                 }
 970
 971                 ret += retval;
 972                 src += retval;
 973                 buf += retval;
 974                 count -= retval;
 975         }
 976         *ppos = src;
 977
 978         mmput(mm);
 979 out_free:
 980         free_page((unsigned long) page);
 981 out:
 982         put_task_struct(task);
 983 out_no_task:
 984         return ret;
 985 }
 986
 987 static const struct file_operations proc_environ_operations = {
 988         .read           = environ_read,
 989         .llseek         = generic_file_llseek,
 990 };
 991
 992 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 993                                 size_t count, loff_t *ppos)
 994 {
 995         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 996         char buffer[PROC_NUMBUF];
 997         size_t len;
 998         int oom_adjust = OOM_DISABLE;
 999         unsigned long flags;
1000
1001         if (!task)
1002                 return -ESRCH;
1003
1004         if (lock_task_sighand(task, &flags)) {
1005                 oom_adjust = task->signal->oom_adj;
1006                 unlock_task_sighand(task, &flags);
1007         }
1008
1009         put_task_struct(task);
1010
1011         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1012
1013         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1014 }
1015
1016 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1017                                 size_t count, loff_t *ppos)
1018 {
1019         struct task_struct *task;
1020         char buffer[PROC_NUMBUF];
1021         long oom_adjust;
1022         unsigned long flags;
1023         int err;
1024
1025         memset(buffer, 0, sizeof(buffer));
1026         if (count > sizeof(buffer) - 1)
1027                 count = sizeof(buffer) - 1;
1028         if (copy_from_user(buffer, buf, count)) {
1029                 err = -EFAULT;
1030                 goto out;
1031         }
1032
1033         err = strict_strtol(strstrip(buffer), 0, &oom_adjust);
1034         if (err)
1035                 goto out;
1036         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1037              oom_adjust != OOM_DISABLE) {
1038                 err = -EINVAL;
1039                 goto out;
1040         }
1041
1042         task = get_proc_task(file->f_path.dentry->d_inode);
1043         if (!task) {
1044                 err = -ESRCH;
1045                 goto out;
1046         }
1047
1048         task_lock(task);
1049         if (!task->mm) {
1050                 err = -EINVAL;
1051                 goto err_task_lock;
1052         }
1053
1054         if (!lock_task_sighand(task, &flags)) {
1055                 err = -ESRCH;
1056                 goto err_task_lock;
1057         }
1058
1059         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1060                 err = -EACCES;
1061                 goto err_sighand;
1062         }
1063
1064         if (oom_adjust != task->signal->oom_adj) {
1065                 if (oom_adjust == OOM_DISABLE)
1066                         atomic_inc(&task->mm->oom_disable_count);
1067                 if (task->signal->oom_adj == OOM_DISABLE)
1068                         atomic_dec(&task->mm->oom_disable_count);
1069         }
1070
1071         /*
1072          * Warn that /proc/pid/oom_adj is deprecated, see
1073          * Documentation/feature-removal-schedule.txt.
1074          */
1075         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, "
1076                         "please use /proc/%d/oom_score_adj instead.\n",
1077                         current->comm, task_pid_nr(current),
1078                         task_pid_nr(task), task_pid_nr(task));
1079         task->signal->oom_adj = oom_adjust;
1080         /*
1081          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1082          * value is always attainable.
1083          */
1084         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1085                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1086         else
1087                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1088                                                                 -OOM_DISABLE;
1089 err_sighand:
1090         unlock_task_sighand(task, &flags);
1091 err_task_lock:
1092         task_unlock(task);
1093         put_task_struct(task);
1094 out:
1095         return err < 0 ? err : count;
1096 }
1097
1098 static const struct file_operations proc_oom_adjust_operations = {
1099         .read           = oom_adjust_read,
1100         .write          = oom_adjust_write,
1101         .llseek         = generic_file_llseek,
1102 };
1103
1104 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1105                                         size_t count, loff_t *ppos)
1106 {
1107         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1108         char buffer[PROC_NUMBUF];
1109         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1110         unsigned long flags;
1111         size_t len;
1112
1113         if (!task)
1114                 return -ESRCH;
1115         if (lock_task_sighand(task, &flags)) {
1116                 oom_score_adj = task->signal->oom_score_adj;
1117                 unlock_task_sighand(task, &flags);
1118         }
1119         put_task_struct(task);
1120         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1121         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1122 }
1123
1124 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1125                                         size_t count, loff_t *ppos)
1126 {
1127         struct task_struct *task;
1128         char buffer[PROC_NUMBUF];
1129         unsigned long flags;
1130         long oom_score_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = strict_strtol(strstrip(buffer), 0, &oom_score_adj);
1142         if (err)
1143                 goto out;
1144         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1145                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         task = get_proc_task(file->f_path.dentry->d_inode);
1151         if (!task) {
1152                 err = -ESRCH;
1153                 goto out;
1154         }
1155
1156         task_lock(task);
1157         if (!task->mm) {
1158                 err = -EINVAL;
1159                 goto err_task_lock;
1160         }
1161
1162         if (!lock_task_sighand(task, &flags)) {
1163                 err = -ESRCH;
1164                 goto err_task_lock;
1165         }
1166
1167         if (oom_score_adj < task->signal->oom_score_adj &&
1168                         !capable(CAP_SYS_RESOURCE)) {
1169                 err = -EACCES;
1170                 goto err_sighand;
1171         }
1172
1173         if (oom_score_adj != task->signal->oom_score_adj) {
1174                 if (oom_score_adj == OOM_SCORE_ADJ_MIN)
1175                         atomic_inc(&task->mm->oom_disable_count);
1176                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1177                         atomic_dec(&task->mm->oom_disable_count);
1178         }
1179         task->signal->oom_score_adj = oom_score_adj;
1180         /*
1181          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1182          * always attainable.
1183          */
1184         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1185                 task->signal->oom_adj = OOM_DISABLE;
1186         else
1187                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1188                                                         OOM_SCORE_ADJ_MAX;
1189 err_sighand:
1190         unlock_task_sighand(task, &flags);
1191 err_task_lock:
1192         task_unlock(task);
1193         put_task_struct(task);
1194 out:
1195         return err < 0 ? err : count;
1196 }
1197
1198 static const struct file_operations proc_oom_score_adj_operations = {
1199         .read           = oom_score_adj_read,
1200         .write          = oom_score_adj_write,
1201         .llseek         = default_llseek,
1202 };
1203
1204 #ifdef CONFIG_AUDITSYSCALL
1205 #define TMPBUFLEN 21
1206 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1207                                   size_t count, loff_t *ppos)
1208 {
1209         struct inode * inode = file->f_path.dentry->d_inode;
1210         struct task_struct *task = get_proc_task(inode);
1211         ssize_t length;
1212         char tmpbuf[TMPBUFLEN];
1213
1214         if (!task)
1215                 return -ESRCH;
1216         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1217                                 audit_get_loginuid(task));
1218         put_task_struct(task);
1219         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1220 }
1221
1222 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1223                                    size_t count, loff_t *ppos)
1224 {
1225         struct inode * inode = file->f_path.dentry->d_inode;
1226         char *page, *tmp;
1227         ssize_t length;
1228         uid_t loginuid;
1229
1230         if (!capable(CAP_AUDIT_CONTROL))
1231                 return -EPERM;
1232
1233         rcu_read_lock();
1234         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1235                 rcu_read_unlock();
1236                 return -EPERM;
1237         }
1238         rcu_read_unlock();
1239
1240         if (count >= PAGE_SIZE)
1241                 count = PAGE_SIZE - 1;
1242
1243         if (*ppos != 0) {
1244                 /* No partial writes. */
1245                 return -EINVAL;
1246         }
1247         page = (char*)__get_free_page(GFP_TEMPORARY);
1248         if (!page)
1249                 return -ENOMEM;
1250         length = -EFAULT;
1251         if (copy_from_user(page, buf, count))
1252                 goto out_free_page;
1253
1254         page[count] = '\0';
1255         loginuid = simple_strtoul(page, &tmp, 10);
1256         if (tmp == page) {
1257                 length = -EINVAL;
1258                 goto out_free_page;
1259
1260         }
1261         length = audit_set_loginuid(current, loginuid);
1262         if (likely(length == 0))
1263                 length = count;
1264
1265 out_free_page:
1266         free_page((unsigned long) page);
1267         return length;
1268 }
1269
1270 static const struct file_operations proc_loginuid_operations = {
1271         .read           = proc_loginuid_read,
1272         .write          = proc_loginuid_write,
1273         .llseek         = generic_file_llseek,
1274 };
1275
1276 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1277                                   size_t count, loff_t *ppos)
1278 {
1279         struct inode * inode = file->f_path.dentry->d_inode;
1280         struct task_struct *task = get_proc_task(inode);
1281         ssize_t length;
1282         char tmpbuf[TMPBUFLEN];
1283
1284         if (!task)
1285                 return -ESRCH;
1286         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1287                                 audit_get_sessionid(task));
1288         put_task_struct(task);
1289         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1290 }
1291
1292 static const struct file_operations proc_sessionid_operations = {
1293         .read           = proc_sessionid_read,
1294         .llseek         = generic_file_llseek,
1295 };
1296 #endif
1297
1298 #ifdef CONFIG_FAULT_INJECTION
1299 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1300                                       size_t count, loff_t *ppos)
1301 {
1302         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1303         char buffer[PROC_NUMBUF];
1304         size_t len;
1305         int make_it_fail;
1306
1307         if (!task)
1308                 return -ESRCH;
1309         make_it_fail = task->make_it_fail;
1310         put_task_struct(task);
1311
1312         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1313
1314         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1315 }
1316
1317 static ssize_t proc_fault_inject_write(struct file * file,
1318                         const char __user * buf, size_t count, loff_t *ppos)
1319 {
1320         struct task_struct *task;
1321         char buffer[PROC_NUMBUF], *end;
1322         int make_it_fail;
1323
1324         if (!capable(CAP_SYS_RESOURCE))
1325                 return -EPERM;
1326         memset(buffer, 0, sizeof(buffer));
1327         if (count > sizeof(buffer) - 1)
1328                 count = sizeof(buffer) - 1;
1329         if (copy_from_user(buffer, buf, count))
1330                 return -EFAULT;
1331         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1332         if (*end)
1333                 return -EINVAL;
1334         task = get_proc_task(file->f_dentry->d_inode);
1335         if (!task)
1336                 return -ESRCH;
1337         task->make_it_fail = make_it_fail;
1338         put_task_struct(task);
1339
1340         return count;
1341 }
1342
1343 static const struct file_operations proc_fault_inject_operations = {
1344         .read           = proc_fault_inject_read,
1345         .write          = proc_fault_inject_write,
1346         .llseek         = generic_file_llseek,
1347 };
1348 #endif
1349
1350
1351 #ifdef CONFIG_SCHED_DEBUG
1352 /*
1353  * Print out various scheduling related per-task fields:
1354  */
1355 static int sched_show(struct seq_file *m, void *v)
1356 {
1357         struct inode *inode = m->private;
1358         struct task_struct *p;
1359
1360         p = get_proc_task(inode);
1361         if (!p)
1362                 return -ESRCH;
1363         proc_sched_show_task(p, m);
1364
1365         put_task_struct(p);
1366
1367         return 0;
1368 }
1369
1370 static ssize_t
1371 sched_write(struct file *file, const char __user *buf,
1372             size_t count, loff_t *offset)
1373 {
1374         struct inode *inode = file->f_path.dentry->d_inode;
1375         struct task_struct *p;
1376
1377         p = get_proc_task(inode);
1378         if (!p)
1379                 return -ESRCH;
1380         proc_sched_set_task(p);
1381
1382         put_task_struct(p);
1383
1384         return count;
1385 }
1386
1387 static int sched_open(struct inode *inode, struct file *filp)
1388 {
1389         int ret;
1390
1391         ret = single_open(filp, sched_show, NULL);
1392         if (!ret) {
1393                 struct seq_file *m = filp->private_data;
1394
1395                 m->private = inode;
1396         }
1397         return ret;
1398 }
1399
1400 static const struct file_operations proc_pid_sched_operations = {
1401         .open           = sched_open,
1402         .read           = seq_read,
1403         .write          = sched_write,
1404         .llseek         = seq_lseek,
1405         .release        = single_release,
1406 };
1407
1408 #endif
1409
1410 static ssize_t comm_write(struct file *file, const char __user *buf,
1411                                 size_t count, loff_t *offset)
1412 {
1413         struct inode *inode = file->f_path.dentry->d_inode;
1414         struct task_struct *p;
1415         char buffer[TASK_COMM_LEN];
1416
1417         memset(buffer, 0, sizeof(buffer));
1418         if (count > sizeof(buffer) - 1)
1419                 count = sizeof(buffer) - 1;
1420         if (copy_from_user(buffer, buf, count))
1421                 return -EFAULT;
1422
1423         p = get_proc_task(inode);
1424         if (!p)
1425                 return -ESRCH;
1426
1427         if (same_thread_group(current, p))
1428                 set_task_comm(p, buffer);
1429         else
1430                 count = -EINVAL;
1431
1432         put_task_struct(p);
1433
1434         return count;
1435 }
1436
1437 static int comm_show(struct seq_file *m, void *v)
1438 {
1439         struct inode *inode = m->private;
1440         struct task_struct *p;
1441
1442         p = get_proc_task(inode);
1443         if (!p)
1444                 return -ESRCH;
1445
1446         task_lock(p);
1447         seq_printf(m, "%s\n", p->comm);
1448         task_unlock(p);
1449
1450         put_task_struct(p);
1451
1452         return 0;
1453 }
1454
1455 static int comm_open(struct inode *inode, struct file *filp)
1456 {
1457         int ret;
1458
1459         ret = single_open(filp, comm_show, NULL);
1460         if (!ret) {
1461                 struct seq_file *m = filp->private_data;
1462
1463                 m->private = inode;
1464         }
1465         return ret;
1466 }
1467
1468 static const struct file_operations proc_pid_set_comm_operations = {
1469         .open           = comm_open,
1470         .read           = seq_read,
1471         .write          = comm_write,
1472         .llseek         = seq_lseek,
1473         .release        = single_release,
1474 };
1475
1476 /*
1477  * We added or removed a vma mapping the executable. The vmas are only mapped
1478  * during exec and are not mapped with the mmap system call.
1479  * Callers must hold down_write() on the mm's mmap_sem for these
1480  */
1481 void added_exe_file_vma(struct mm_struct *mm)
1482 {
1483         mm->num_exe_file_vmas++;
1484 }
1485
1486 void removed_exe_file_vma(struct mm_struct *mm)
1487 {
1488         mm->num_exe_file_vmas--;
1489         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1490                 fput(mm->exe_file);
1491                 mm->exe_file = NULL;
1492         }
1493
1494 }
1495
1496 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1497 {
1498         if (new_exe_file)
1499                 get_file(new_exe_file);
1500         if (mm->exe_file)
1501                 fput(mm->exe_file);
1502         mm->exe_file = new_exe_file;
1503         mm->num_exe_file_vmas = 0;
1504 }
1505
1506 struct file *get_mm_exe_file(struct mm_struct *mm)
1507 {
1508         struct file *exe_file;
1509
1510         /* We need mmap_sem to protect against races with removal of
1511          * VM_EXECUTABLE vmas */
1512         down_read(&mm->mmap_sem);
1513         exe_file = mm->exe_file;
1514         if (exe_file)
1515                 get_file(exe_file);
1516         up_read(&mm->mmap_sem);
1517         return exe_file;
1518 }
1519
1520 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1521 {
1522         /* It's safe to write the exe_file pointer without exe_file_lock because
1523          * this is called during fork when the task is not yet in /proc */
1524         newmm->exe_file = get_mm_exe_file(oldmm);
1525 }
1526
1527 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1528 {
1529         struct task_struct *task;
1530         struct mm_struct *mm;
1531         struct file *exe_file;
1532
1533         task = get_proc_task(inode);
1534         if (!task)
1535                 return -ENOENT;
1536         mm = get_task_mm(task);
1537         put_task_struct(task);
1538         if (!mm)
1539                 return -ENOENT;
1540         exe_file = get_mm_exe_file(mm);
1541         mmput(mm);
1542         if (exe_file) {
1543                 *exe_path = exe_file->f_path;
1544                 path_get(&exe_file->f_path);
1545                 fput(exe_file);
1546                 return 0;
1547         } else
1548                 return -ENOENT;
1549 }
1550
1551 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1552 {
1553         struct inode *inode = dentry->d_inode;
1554         int error = -EACCES;
1555
1556         /* We don't need a base pointer in the /proc filesystem */
1557         path_put(&nd->path);
1558
1559         /* Are we allowed to snoop on the tasks file descriptors? */
1560         if (!proc_fd_access_allowed(inode))
1561                 goto out;
1562
1563         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1564 out:
1565         return ERR_PTR(error);
1566 }
1567
1568 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1569 {
1570         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1571         char *pathname;
1572         int len;
1573
1574         if (!tmp)
1575                 return -ENOMEM;
1576
1577         pathname = d_path(path, tmp, PAGE_SIZE);
1578         len = PTR_ERR(pathname);
1579         if (IS_ERR(pathname))
1580                 goto out;
1581         len = tmp + PAGE_SIZE - 1 - pathname;
1582
1583         if (len > buflen)
1584                 len = buflen;
1585         if (copy_to_user(buffer, pathname, len))
1586                 len = -EFAULT;
1587  out:
1588         free_page((unsigned long)tmp);
1589         return len;
1590 }
1591
1592 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1593 {
1594         int error = -EACCES;
1595         struct inode *inode = dentry->d_inode;
1596         struct path path;
1597
1598         /* Are we allowed to snoop on the tasks file descriptors? */
1599         if (!proc_fd_access_allowed(inode))
1600                 goto out;
1601
1602         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1603         if (error)
1604                 goto out;
1605
1606         error = do_proc_readlink(&path, buffer, buflen);
1607         path_put(&path);
1608 out:
1609         return error;
1610 }
1611
1612 static const struct inode_operations proc_pid_link_inode_operations = {
1613         .readlink       = proc_pid_readlink,
1614         .follow_link    = proc_pid_follow_link,
1615         .setattr        = proc_setattr,
1616 };
1617
1618
1619 /* building an inode */
1620
1621 static int task_dumpable(struct task_struct *task)
1622 {
1623         int dumpable = 0;
1624         struct mm_struct *mm;
1625
1626         task_lock(task);
1627         mm = task->mm;
1628         if (mm)
1629                 dumpable = get_dumpable(mm);
1630         task_unlock(task);
1631         if(dumpable == 1)
1632                 return 1;
1633         return 0;
1634 }
1635
1636
1637 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1638 {
1639         struct inode * inode;
1640         struct proc_inode *ei;
1641         const struct cred *cred;
1642
1643         /* We need a new inode */
1644
1645         inode = new_inode(sb);
1646         if (!inode)
1647                 goto out;
1648
1649         /* Common stuff */
1650         ei = PROC_I(inode);
1651         inode->i_ino = get_next_ino();
1652         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1653         inode->i_op = &proc_def_inode_operations;
1654
1655         /*
1656          * grab the reference to task.
1657          */
1658         ei->pid = get_task_pid(task, PIDTYPE_PID);
1659         if (!ei->pid)
1660                 goto out_unlock;
1661
1662         if (task_dumpable(task)) {
1663                 rcu_read_lock();
1664                 cred = __task_cred(task);
1665                 inode->i_uid = cred->euid;
1666                 inode->i_gid = cred->egid;
1667                 rcu_read_unlock();
1668         }
1669         security_task_to_inode(task, inode);
1670
1671 out:
1672         return inode;
1673
1674 out_unlock:
1675         iput(inode);
1676         return NULL;
1677 }
1678
1679 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1680 {
1681         struct inode *inode = dentry->d_inode;
1682         struct task_struct *task;
1683         const struct cred *cred;
1684
1685         generic_fillattr(inode, stat);
1686
1687         rcu_read_lock();
1688         stat->uid = 0;
1689         stat->gid = 0;
1690         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1691         if (task) {
1692                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1693                     task_dumpable(task)) {
1694                         cred = __task_cred(task);
1695                         stat->uid = cred->euid;
1696                         stat->gid = cred->egid;
1697                 }
1698         }
1699         rcu_read_unlock();
1700         return 0;
1701 }
1702
1703 /* dentry stuff */
1704
1705 /*
1706  *      Exceptional case: normally we are not allowed to unhash a busy
1707  * directory. In this case, however, we can do it - no aliasing problems
1708  * due to the way we treat inodes.
1709  *
1710  * Rewrite the inode's ownerships here because the owning task may have
1711  * performed a setuid(), etc.
1712  *
1713  * Before the /proc/pid/status file was created the only way to read
1714  * the effective uid of a /process was to stat /proc/pid.  Reading
1715  * /proc/pid/status is slow enough that procps and other packages
1716  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1717  * made this apply to all per process world readable and executable
1718  * directories.
1719  */
1720 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1721 {
1722         struct inode *inode;
1723         struct task_struct *task;
1724         const struct cred *cred;
1725
1726         if (nd && nd->flags & LOOKUP_RCU)
1727                 return -ECHILD;
1728
1729         inode = dentry->d_inode;
1730         task = get_proc_task(inode);
1731
1732         if (task) {
1733                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1734                     task_dumpable(task)) {
1735                         rcu_read_lock();
1736                         cred = __task_cred(task);
1737                         inode->i_uid = cred->euid;
1738                         inode->i_gid = cred->egid;
1739                         rcu_read_unlock();
1740                 } else {
1741                         inode->i_uid = 0;
1742                         inode->i_gid = 0;
1743                 }
1744                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1745                 security_task_to_inode(task, inode);
1746                 put_task_struct(task);
1747                 return 1;
1748         }
1749         d_drop(dentry);
1750         return 0;
1751 }
1752
1753 static int pid_delete_dentry(const struct dentry * dentry)
1754 {
1755         /* Is the task we represent dead?
1756          * If so, then don't put the dentry on the lru list,
1757          * kill it immediately.
1758          */
1759         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1760 }
1761
1762 static const struct dentry_operations pid_dentry_operations =
1763 {
1764         .d_revalidate   = pid_revalidate,
1765         .d_delete       = pid_delete_dentry,
1766 };
1767
1768 /* Lookups */
1769
1770 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1771                                 struct task_struct *, const void *);
1772
1773 /*
1774  * Fill a directory entry.
1775  *
1776  * If possible create the dcache entry and derive our inode number and
1777  * file type from dcache entry.
1778  *
1779  * Since all of the proc inode numbers are dynamically generated, the inode
1780  * numbers do not exist until the inode is cache.  This means creating the
1781  * the dcache entry in readdir is necessary to keep the inode numbers
1782  * reported by readdir in sync with the inode numbers reported
1783  * by stat.
1784  */
1785 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1786         char *name, int len,
1787         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1788 {
1789         struct dentry *child, *dir = filp->f_path.dentry;
1790         struct inode *inode;
1791         struct qstr qname;
1792         ino_t ino = 0;
1793         unsigned type = DT_UNKNOWN;
1794
1795         qname.name = name;
1796         qname.len  = len;
1797         qname.hash = full_name_hash(name, len);
1798
1799         child = d_lookup(dir, &qname);
1800         if (!child) {
1801                 struct dentry *new;
1802                 new = d_alloc(dir, &qname);
1803                 if (new) {
1804                         child = instantiate(dir->d_inode, new, task, ptr);
1805                         if (child)
1806                                 dput(new);
1807                         else
1808                                 child = new;
1809                 }
1810         }
1811         if (!child || IS_ERR(child) || !child->d_inode)
1812                 goto end_instantiate;
1813         inode = child->d_inode;
1814         if (inode) {
1815                 ino = inode->i_ino;
1816                 type = inode->i_mode >> 12;
1817         }
1818         dput(child);
1819 end_instantiate:
1820         if (!ino)
1821                 ino = find_inode_number(dir, &qname);
1822         if (!ino)
1823                 ino = 1;
1824         return filldir(dirent, name, len, filp->f_pos, ino, type);
1825 }
1826
1827 static unsigned name_to_int(struct dentry *dentry)
1828 {
1829         const char *name = dentry->d_name.name;
1830         int len = dentry->d_name.len;
1831         unsigned n = 0;
1832
1833         if (len > 1 && *name == '0')
1834                 goto out;
1835         while (len-- > 0) {
1836                 unsigned c = *name++ - '0';
1837                 if (c > 9)
1838                         goto out;
1839                 if (n >= (~0U-9)/10)
1840                         goto out;
1841                 n *= 10;
1842                 n += c;
1843         }
1844         return n;
1845 out:
1846         return ~0U;
1847 }
1848
1849 #define PROC_FDINFO_MAX 64
1850
1851 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1852 {
1853         struct task_struct *task = get_proc_task(inode);
1854         struct files_struct *files = NULL;
1855         struct file *file;
1856         int fd = proc_fd(inode);
1857
1858         if (task) {
1859                 files = get_files_struct(task);
1860                 put_task_struct(task);
1861         }
1862         if (files) {
1863                 /*
1864                  * We are not taking a ref to the file structure, so we must
1865                  * hold ->file_lock.
1866                  */
1867                 spin_lock(&files->file_lock);
1868                 file = fcheck_files(files, fd);
1869                 if (file) {
1870                         if (path) {
1871                                 *path = file->f_path;
1872                                 path_get(&file->f_path);
1873                         }
1874                         if (info)
1875                                 snprintf(info, PROC_FDINFO_MAX,
1876                                          "pos:\t%lli\n"
1877                                          "flags:\t0%o\n",
1878                                          (long long) file->f_pos,
1879                                          file->f_flags);
1880                         spin_unlock(&files->file_lock);
1881                         put_files_struct(files);
1882                         return 0;
1883                 }
1884                 spin_unlock(&files->file_lock);
1885                 put_files_struct(files);
1886         }
1887         return -ENOENT;
1888 }
1889
1890 static int proc_fd_link(struct inode *inode, struct path *path)
1891 {
1892         return proc_fd_info(inode, path, NULL);
1893 }
1894
1895 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1896 {
1897         struct inode *inode;
1898         struct task_struct *task;
1899         int fd;
1900         struct files_struct *files;
1901         const struct cred *cred;
1902
1903         if (nd && nd->flags & LOOKUP_RCU)
1904                 return -ECHILD;
1905
1906         inode = dentry->d_inode;
1907         task = get_proc_task(inode);
1908         fd = proc_fd(inode);
1909
1910         if (task) {
1911                 files = get_files_struct(task);
1912                 if (files) {
1913                         rcu_read_lock();
1914                         if (fcheck_files(files, fd)) {
1915                                 rcu_read_unlock();
1916                                 put_files_struct(files);
1917                                 if (task_dumpable(task)) {
1918                                         rcu_read_lock();
1919                                         cred = __task_cred(task);
1920                                         inode->i_uid = cred->euid;
1921                                         inode->i_gid = cred->egid;
1922                                         rcu_read_unlock();
1923                                 } else {
1924                                         inode->i_uid = 0;
1925                                         inode->i_gid = 0;
1926                                 }
1927                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1928                                 security_task_to_inode(task, inode);
1929                                 put_task_struct(task);
1930                                 return 1;
1931                         }
1932                         rcu_read_unlock();
1933                         put_files_struct(files);
1934                 }
1935                 put_task_struct(task);
1936         }
1937         d_drop(dentry);
1938         return 0;
1939 }
1940
1941 static const struct dentry_operations tid_fd_dentry_operations =
1942 {
1943         .d_revalidate   = tid_fd_revalidate,
1944         .d_delete       = pid_delete_dentry,
1945 };
1946
1947 static struct dentry *proc_fd_instantiate(struct inode *dir,
1948         struct dentry *dentry, struct task_struct *task, const void *ptr)
1949 {
1950         unsigned fd = *(const unsigned *)ptr;
1951         struct file *file;
1952         struct files_struct *files;
1953         struct inode *inode;
1954         struct proc_inode *ei;
1955         struct dentry *error = ERR_PTR(-ENOENT);
1956
1957         inode = proc_pid_make_inode(dir->i_sb, task);
1958         if (!inode)
1959                 goto out;
1960         ei = PROC_I(inode);
1961         ei->fd = fd;
1962         files = get_files_struct(task);
1963         if (!files)
1964                 goto out_iput;
1965         inode->i_mode = S_IFLNK;
1966
1967         /*
1968          * We are not taking a ref to the file structure, so we must
1969          * hold ->file_lock.
1970          */
1971         spin_lock(&files->file_lock);
1972         file = fcheck_files(files, fd);
1973         if (!file)
1974                 goto out_unlock;
1975         if (file->f_mode & FMODE_READ)
1976                 inode->i_mode |= S_IRUSR | S_IXUSR;
1977         if (file->f_mode & FMODE_WRITE)
1978                 inode->i_mode |= S_IWUSR | S_IXUSR;
1979         spin_unlock(&files->file_lock);
1980         put_files_struct(files);
1981
1982         inode->i_op = &proc_pid_link_inode_operations;
1983         inode->i_size = 64;
1984         ei->op.proc_get_link = proc_fd_link;
1985         d_set_d_op(dentry, &tid_fd_dentry_operations);
1986         d_add(dentry, inode);
1987         /* Close the race of the process dying before we return the dentry */
1988         if (tid_fd_revalidate(dentry, NULL))
1989                 error = NULL;
1990
1991  out:
1992         return error;
1993 out_unlock:
1994         spin_unlock(&files->file_lock);
1995         put_files_struct(files);
1996 out_iput:
1997         iput(inode);
1998         goto out;
1999 }
2000
2001 static struct dentry *proc_lookupfd_common(struct inode *dir,
2002                                            struct dentry *dentry,
2003                                            instantiate_t instantiate)
2004 {
2005         struct task_struct *task = get_proc_task(dir);
2006         unsigned fd = name_to_int(dentry);
2007         struct dentry *result = ERR_PTR(-ENOENT);
2008
2009         if (!task)
2010                 goto out_no_task;
2011         if (fd == ~0U)
2012                 goto out;
2013
2014         result = instantiate(dir, dentry, task, &fd);
2015 out:
2016         put_task_struct(task);
2017 out_no_task:
2018         return result;
2019 }
2020
2021 static int proc_readfd_common(struct file * filp, void * dirent,
2022                               filldir_t filldir, instantiate_t instantiate)
2023 {
2024         struct dentry *dentry = filp->f_path.dentry;
2025         struct inode *inode = dentry->d_inode;
2026         struct task_struct *p = get_proc_task(inode);
2027         unsigned int fd, ino;
2028         int retval;
2029         struct files_struct * files;
2030
2031         retval = -ENOENT;
2032         if (!p)
2033                 goto out_no_task;
2034         retval = 0;
2035
2036         fd = filp->f_pos;
2037         switch (fd) {
2038                 case 0:
2039                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2040                                 goto out;
2041                         filp->f_pos++;
2042                 case 1:
2043                         ino = parent_ino(dentry);
2044                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2045                                 goto out;
2046                         filp->f_pos++;
2047                 default:
2048                         files = get_files_struct(p);
2049                         if (!files)
2050                                 goto out;
2051                         rcu_read_lock();
2052                         for (fd = filp->f_pos-2;
2053                              fd < files_fdtable(files)->max_fds;
2054                              fd++, filp->f_pos++) {
2055                                 char name[PROC_NUMBUF];
2056                                 int len;
2057
2058                                 if (!fcheck_files(files, fd))
2059                                         continue;
2060                                 rcu_read_unlock();
2061
2062                                 len = snprintf(name, sizeof(name), "%d", fd);
2063                                 if (proc_fill_cache(filp, dirent, filldir,
2064                                                     name, len, instantiate,
2065                                                     p, &fd) < 0) {
2066                                         rcu_read_lock();
2067                                         break;
2068                                 }
2069                                 rcu_read_lock();
2070                         }
2071                         rcu_read_unlock();
2072                         put_files_struct(files);
2073         }
2074 out:
2075         put_task_struct(p);
2076 out_no_task:
2077         return retval;
2078 }
2079
2080 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2081                                     struct nameidata *nd)
2082 {
2083         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2084 }
2085
2086 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2087 {
2088         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2089 }
2090
2091 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2092                                       size_t len, loff_t *ppos)
2093 {
2094         char tmp[PROC_FDINFO_MAX];
2095         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2096         if (!err)
2097                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2098         return err;
2099 }
2100
2101 static const struct file_operations proc_fdinfo_file_operations = {
2102         .open           = nonseekable_open,
2103         .read           = proc_fdinfo_read,
2104         .llseek         = no_llseek,
2105 };
2106
2107 static const struct file_operations proc_fd_operations = {
2108         .read           = generic_read_dir,
2109         .readdir        = proc_readfd,
2110         .llseek         = default_llseek,
2111 };
2112
2113 /*
2114  * /proc/pid/fd needs a special permission handler so that a process can still
2115  * access /proc/self/fd after it has executed a setuid().
2116  */
2117 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2118 {
2119         int rv;
2120
2121         if (flags & IPERM_FLAG_RCU)
2122                 return -ECHILD;
2123         rv = generic_permission(inode, mask, flags, NULL);
2124         if (rv == 0)
2125                 return 0;
2126         if (task_pid(current) == proc_pid(inode))
2127                 rv = 0;
2128         return rv;
2129 }
2130
2131 /*
2132  * proc directories can do almost nothing..
2133  */
2134 static const struct inode_operations proc_fd_inode_operations = {
2135         .lookup         = proc_lookupfd,
2136         .permission     = proc_fd_permission,
2137         .setattr        = proc_setattr,
2138 };
2139
2140 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2141         struct dentry *dentry, struct task_struct *task, const void *ptr)
2142 {
2143         unsigned fd = *(unsigned *)ptr;
2144         struct inode *inode;
2145         struct proc_inode *ei;
2146         struct dentry *error = ERR_PTR(-ENOENT);
2147
2148         inode = proc_pid_make_inode(dir->i_sb, task);
2149         if (!inode)
2150                 goto out;
2151         ei = PROC_I(inode);
2152         ei->fd = fd;
2153         inode->i_mode = S_IFREG | S_IRUSR;
2154         inode->i_fop = &proc_fdinfo_file_operations;
2155         d_set_d_op(dentry, &tid_fd_dentry_operations);
2156         d_add(dentry, inode);
2157         /* Close the race of the process dying before we return the dentry */
2158         if (tid_fd_revalidate(dentry, NULL))
2159                 error = NULL;
2160
2161  out:
2162         return error;
2163 }
2164
2165 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2166                                         struct dentry *dentry,
2167                                         struct nameidata *nd)
2168 {
2169         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2170 }
2171
2172 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2173 {
2174         return proc_readfd_common(filp, dirent, filldir,
2175                                   proc_fdinfo_instantiate);
2176 }
2177
2178 static const struct file_operations proc_fdinfo_operations = {
2179         .read           = generic_read_dir,
2180         .readdir        = proc_readfdinfo,
2181         .llseek         = default_llseek,
2182 };
2183
2184 /*
2185  * proc directories can do almost nothing..
2186  */
2187 static const struct inode_operations proc_fdinfo_inode_operations = {
2188         .lookup         = proc_lookupfdinfo,
2189         .setattr        = proc_setattr,
2190 };
2191
2192
2193 static struct dentry *proc_pident_instantiate(struct inode *dir,
2194         struct dentry *dentry, struct task_struct *task, const void *ptr)
2195 {
2196         const struct pid_entry *p = ptr;
2197         struct inode *inode;
2198         struct proc_inode *ei;
2199         struct dentry *error = ERR_PTR(-ENOENT);
2200
2201         inode = proc_pid_make_inode(dir->i_sb, task);
2202         if (!inode)
2203                 goto out;
2204
2205         ei = PROC_I(inode);
2206         inode->i_mode = p->mode;
2207         if (S_ISDIR(inode->i_mode))
2208                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2209         if (p->iop)
2210                 inode->i_op = p->iop;
2211         if (p->fop)
2212                 inode->i_fop = p->fop;
2213         ei->op = p->op;
2214         d_set_d_op(dentry, &pid_dentry_operations);
2215         d_add(dentry, inode);
2216         /* Close the race of the process dying before we return the dentry */
2217         if (pid_revalidate(dentry, NULL))
2218                 error = NULL;
2219 out:
2220         return error;
2221 }
2222
2223 static struct dentry *proc_pident_lookup(struct inode *dir,
2224                                          struct dentry *dentry,
2225                                          const struct pid_entry *ents,
2226                                          unsigned int nents)
2227 {
2228         struct dentry *error;
2229         struct task_struct *task = get_proc_task(dir);
2230         const struct pid_entry *p, *last;
2231
2232         error = ERR_PTR(-ENOENT);
2233
2234         if (!task)
2235                 goto out_no_task;
2236
2237         /*
2238          * Yes, it does not scale. And it should not. Don't add
2239          * new entries into /proc/<tgid>/ without very good reasons.
2240          */
2241         last = &ents[nents - 1];
2242         for (p = ents; p <= last; p++) {
2243                 if (p->len != dentry->d_name.len)
2244                         continue;
2245                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2246                         break;
2247         }
2248         if (p > last)
2249                 goto out;
2250
2251         error = proc_pident_instantiate(dir, dentry, task, p);
2252 out:
2253         put_task_struct(task);
2254 out_no_task:
2255         return error;
2256 }
2257
2258 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2259         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2260 {
2261         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2262                                 proc_pident_instantiate, task, p);
2263 }
2264
2265 static int proc_pident_readdir(struct file *filp,
2266                 void *dirent, filldir_t filldir,
2267                 const struct pid_entry *ents, unsigned int nents)
2268 {
2269         int i;
2270         struct dentry *dentry = filp->f_path.dentry;
2271         struct inode *inode = dentry->d_inode;
2272         struct task_struct *task = get_proc_task(inode);
2273         const struct pid_entry *p, *last;
2274         ino_t ino;
2275         int ret;
2276
2277         ret = -ENOENT;
2278         if (!task)
2279                 goto out_no_task;
2280
2281         ret = 0;
2282         i = filp->f_pos;
2283         switch (i) {
2284         case 0:
2285                 ino = inode->i_ino;
2286                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2287                         goto out;
2288                 i++;
2289                 filp->f_pos++;
2290                 /* fall through */
2291         case 1:
2292                 ino = parent_ino(dentry);
2293                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2294                         goto out;
2295                 i++;
2296                 filp->f_pos++;
2297                 /* fall through */
2298         default:
2299                 i -= 2;
2300                 if (i >= nents) {
2301                         ret = 1;
2302                         goto out;
2303                 }
2304                 p = ents + i;
2305                 last = &ents[nents - 1];
2306                 while (p <= last) {
2307                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2308                                 goto out;
2309                         filp->f_pos++;
2310                         p++;
2311                 }
2312         }
2313
2314         ret = 1;
2315 out:
2316         put_task_struct(task);
2317 out_no_task:
2318         return ret;
2319 }
2320
2321 #ifdef CONFIG_SECURITY
2322 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2323                                   size_t count, loff_t *ppos)
2324 {
2325         struct inode * inode = file->f_path.dentry->d_inode;
2326         char *p = NULL;
2327         ssize_t length;
2328         struct task_struct *task = get_proc_task(inode);
2329
2330         if (!task)
2331                 return -ESRCH;
2332
2333         length = security_getprocattr(task,
2334                                       (char*)file->f_path.dentry->d_name.name,
2335                                       &p);
2336         put_task_struct(task);
2337         if (length > 0)
2338                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2339         kfree(p);
2340         return length;
2341 }
2342
2343 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2344                                    size_t count, loff_t *ppos)
2345 {
2346         struct inode * inode = file->f_path.dentry->d_inode;
2347         char *page;
2348         ssize_t length;
2349         struct task_struct *task = get_proc_task(inode);
2350
2351         length = -ESRCH;
2352         if (!task)
2353                 goto out_no_task;
2354         if (count > PAGE_SIZE)
2355                 count = PAGE_SIZE;
2356
2357         /* No partial writes. */
2358         length = -EINVAL;
2359         if (*ppos != 0)
2360                 goto out;
2361
2362         length = -ENOMEM;
2363         page = (char*)__get_free_page(GFP_TEMPORARY);
2364         if (!page)
2365                 goto out;
2366
2367         length = -EFAULT;
2368         if (copy_from_user(page, buf, count))
2369                 goto out_free;
2370
2371         /* Guard against adverse ptrace interaction */
2372         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2373         if (length < 0)
2374                 goto out_free;
2375
2376         length = security_setprocattr(task,
2377                                       (char*)file->f_path.dentry->d_name.name,
2378                                       (void*)page, count);
2379         mutex_unlock(&task->signal->cred_guard_mutex);
2380 out_free:
2381         free_page((unsigned long) page);
2382 out:
2383         put_task_struct(task);
2384 out_no_task:
2385         return length;
2386 }
2387
2388 static const struct file_operations proc_pid_attr_operations = {
2389         .read           = proc_pid_attr_read,
2390         .write          = proc_pid_attr_write,
2391         .llseek         = generic_file_llseek,
2392 };
2393
2394 static const struct pid_entry attr_dir_stuff[] = {
2395         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2396         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2397         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2398         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2399         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2400         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2401 };
2402
2403 static int proc_attr_dir_readdir(struct file * filp,
2404                              void * dirent, filldir_t filldir)
2405 {
2406         return proc_pident_readdir(filp,dirent,filldir,
2407                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2408 }
2409
2410 static const struct file_operations proc_attr_dir_operations = {
2411         .read           = generic_read_dir,
2412         .readdir        = proc_attr_dir_readdir,
2413         .llseek         = default_llseek,
2414 };
2415
2416 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2417                                 struct dentry *dentry, struct nameidata *nd)
2418 {
2419         return proc_pident_lookup(dir, dentry,
2420                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2421 }
2422
2423 static const struct inode_operations proc_attr_dir_inode_operations = {
2424         .lookup         = proc_attr_dir_lookup,
2425         .getattr        = pid_getattr,
2426         .setattr        = proc_setattr,
2427 };
2428
2429 #endif
2430
2431 #ifdef CONFIG_ELF_CORE
2432 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2433                                          size_t count, loff_t *ppos)
2434 {
2435         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2436         struct mm_struct *mm;
2437         char buffer[PROC_NUMBUF];
2438         size_t len;
2439         int ret;
2440
2441         if (!task)
2442                 return -ESRCH;
2443
2444         ret = 0;
2445         mm = get_task_mm(task);
2446         if (mm) {
2447                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2448                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2449                                 MMF_DUMP_FILTER_SHIFT));
2450                 mmput(mm);
2451                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2452         }
2453
2454         put_task_struct(task);
2455
2456         return ret;
2457 }
2458
2459 static ssize_t proc_coredump_filter_write(struct file *file,
2460                                           const char __user *buf,
2461                                           size_t count,
2462                                           loff_t *ppos)
2463 {
2464         struct task_struct *task;
2465         struct mm_struct *mm;
2466         char buffer[PROC_NUMBUF], *end;
2467         unsigned int val;
2468         int ret;
2469         int i;
2470         unsigned long mask;
2471
2472         ret = -EFAULT;
2473         memset(buffer, 0, sizeof(buffer));
2474         if (count > sizeof(buffer) - 1)
2475                 count = sizeof(buffer) - 1;
2476         if (copy_from_user(buffer, buf, count))
2477                 goto out_no_task;
2478
2479         ret = -EINVAL;
2480         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2481         if (*end == '\n')
2482                 end++;
2483         if (end - buffer == 0)
2484                 goto out_no_task;
2485
2486         ret = -ESRCH;
2487         task = get_proc_task(file->f_dentry->d_inode);
2488         if (!task)
2489                 goto out_no_task;
2490
2491         ret = end - buffer;
2492         mm = get_task_mm(task);
2493         if (!mm)
2494                 goto out_no_mm;
2495
2496         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2497                 if (val & mask)
2498                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2499                 else
2500                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2501         }
2502
2503         mmput(mm);
2504  out_no_mm:
2505         put_task_struct(task);
2506  out_no_task:
2507         return ret;
2508 }
2509
2510 static const struct file_operations proc_coredump_filter_operations = {
2511         .read           = proc_coredump_filter_read,
2512         .write          = proc_coredump_filter_write,
2513         .llseek         = generic_file_llseek,
2514 };
2515 #endif
2516
2517 /*
2518  * /proc/self:
2519  */
2520 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2521                               int buflen)
2522 {
2523         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2524         pid_t tgid = task_tgid_nr_ns(current, ns);
2525         char tmp[PROC_NUMBUF];
2526         if (!tgid)
2527                 return -ENOENT;
2528         sprintf(tmp, "%d", tgid);
2529         return vfs_readlink(dentry,buffer,buflen,tmp);
2530 }
2531
2532 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2533 {
2534         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2535         pid_t tgid = task_tgid_nr_ns(current, ns);
2536         char *name = ERR_PTR(-ENOENT);
2537         if (tgid) {
2538                 name = __getname();
2539                 if (!name)
2540                         name = ERR_PTR(-ENOMEM);
2541                 else
2542                         sprintf(name, "%d", tgid);
2543         }
2544         nd_set_link(nd, name);
2545         return NULL;
2546 }
2547
2548 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2549                                 void *cookie)
2550 {
2551         char *s = nd_get_link(nd);
2552         if (!IS_ERR(s))
2553                 __putname(s);
2554 }
2555
2556 static const struct inode_operations proc_self_inode_operations = {
2557         .readlink       = proc_self_readlink,
2558         .follow_link    = proc_self_follow_link,
2559         .put_link       = proc_self_put_link,
2560 };
2561
2562 /*
2563  * proc base
2564  *
2565  * These are the directory entries in the root directory of /proc
2566  * that properly belong to the /proc filesystem, as they describe
2567  * describe something that is process related.
2568  */
2569 static const struct pid_entry proc_base_stuff[] = {
2570         NOD("self", S_IFLNK|S_IRWXUGO,
2571                 &proc_self_inode_operations, NULL, {}),
2572 };
2573
2574 /*
2575  *      Exceptional case: normally we are not allowed to unhash a busy
2576  * directory. In this case, however, we can do it - no aliasing problems
2577  * due to the way we treat inodes.
2578  */
2579 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2580 {
2581         struct inode *inode;
2582         struct task_struct *task;
2583
2584         if (nd->flags & LOOKUP_RCU)
2585                 return -ECHILD;
2586
2587         inode = dentry->d_inode;
2588         task = get_proc_task(inode);
2589         if (task) {
2590                 put_task_struct(task);
2591                 return 1;
2592         }
2593         d_drop(dentry);
2594         return 0;
2595 }
2596
2597 static const struct dentry_operations proc_base_dentry_operations =
2598 {
2599         .d_revalidate   = proc_base_revalidate,
2600         .d_delete       = pid_delete_dentry,
2601 };
2602
2603 static struct dentry *proc_base_instantiate(struct inode *dir,
2604         struct dentry *dentry, struct task_struct *task, const void *ptr)
2605 {
2606         const struct pid_entry *p = ptr;
2607         struct inode *inode;
2608         struct proc_inode *ei;
2609         struct dentry *error;
2610
2611         /* Allocate the inode */
2612         error = ERR_PTR(-ENOMEM);
2613         inode = new_inode(dir->i_sb);
2614         if (!inode)
2615                 goto out;
2616
2617         /* Initialize the inode */
2618         ei = PROC_I(inode);
2619         inode->i_ino = get_next_ino();
2620         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2621
2622         /*
2623          * grab the reference to the task.
2624          */
2625         ei->pid = get_task_pid(task, PIDTYPE_PID);
2626         if (!ei->pid)
2627                 goto out_iput;
2628
2629         inode->i_mode = p->mode;
2630         if (S_ISDIR(inode->i_mode))
2631                 inode->i_nlink = 2;
2632         if (S_ISLNK(inode->i_mode))
2633                 inode->i_size = 64;
2634         if (p->iop)
2635                 inode->i_op = p->iop;
2636         if (p->fop)
2637                 inode->i_fop = p->fop;
2638         ei->op = p->op;
2639         d_set_d_op(dentry, &proc_base_dentry_operations);
2640         d_add(dentry, inode);
2641         error = NULL;
2642 out:
2643         return error;
2644 out_iput:
2645         iput(inode);
2646         goto out;
2647 }
2648
2649 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2650 {
2651         struct dentry *error;
2652         struct task_struct *task = get_proc_task(dir);
2653         const struct pid_entry *p, *last;
2654
2655         error = ERR_PTR(-ENOENT);
2656
2657         if (!task)
2658                 goto out_no_task;
2659
2660         /* Lookup the directory entry */
2661         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2662         for (p = proc_base_stuff; p <= last; p++) {
2663                 if (p->len != dentry->d_name.len)
2664                         continue;
2665                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2666                         break;
2667         }
2668         if (p > last)
2669                 goto out;
2670
2671         error = proc_base_instantiate(dir, dentry, task, p);
2672
2673 out:
2674         put_task_struct(task);
2675 out_no_task:
2676         return error;
2677 }
2678
2679 static int proc_base_fill_cache(struct file *filp, void *dirent,
2680         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2681 {
2682         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2683                                 proc_base_instantiate, task, p);
2684 }
2685
2686 #ifdef CONFIG_TASK_IO_ACCOUNTING
2687 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2688 {
2689         struct task_io_accounting acct = task->ioac;
2690         unsigned long flags;
2691
2692         if (whole && lock_task_sighand(task, &flags)) {
2693                 struct task_struct *t = task;
2694
2695                 task_io_accounting_add(&acct, &task->signal->ioac);
2696                 while_each_thread(task, t)
2697                         task_io_accounting_add(&acct, &t->ioac);
2698
2699                 unlock_task_sighand(task, &flags);
2700         }
2701         return sprintf(buffer,
2702                         "rchar: %llu\n"
2703                         "wchar: %llu\n"
2704                         "syscr: %llu\n"
2705                         "syscw: %llu\n"
2706                         "read_bytes: %llu\n"
2707                         "write_bytes: %llu\n"
2708                         "cancelled_write_bytes: %llu\n",
2709                         (unsigned long long)acct.rchar,
2710                         (unsigned long long)acct.wchar,
2711                         (unsigned long long)acct.syscr,
2712                         (unsigned long long)acct.syscw,
2713                         (unsigned long long)acct.read_bytes,
2714                         (unsigned long long)acct.write_bytes,
2715                         (unsigned long long)acct.cancelled_write_bytes);
2716 }
2717
2718 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2719 {
2720         return do_io_accounting(task, buffer, 0);
2721 }
2722
2723 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2724 {
2725         return do_io_accounting(task, buffer, 1);
2726 }
2727 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2728
2729 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2730                                 struct pid *pid, struct task_struct *task)
2731 {
2732         seq_printf(m, "%08x\n", task->personality);
2733         return 0;
2734 }
2735
2736 /*
2737  * Thread groups
2738  */
2739 static const struct file_operations proc_task_operations;
2740 static const struct inode_operations proc_task_inode_operations;
2741
2742 static const struct pid_entry tgid_base_stuff[] = {
2743         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2744         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2745         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2746 #ifdef CONFIG_NET
2747         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2748 #endif
2749         REG("environ",    S_IRUSR, proc_environ_operations),
2750         INF("auxv",       S_IRUSR, proc_pid_auxv),
2751         ONE("status",     S_IRUGO, proc_pid_status),
2752         ONE("personality", S_IRUSR, proc_pid_personality),
2753         INF("limits",     S_IRUGO, proc_pid_limits),
2754 #ifdef CONFIG_SCHED_DEBUG
2755         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2756 #endif
2757         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2758 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2759         INF("syscall",    S_IRUSR, proc_pid_syscall),
2760 #endif
2761         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2762         ONE("stat",       S_IRUGO, proc_tgid_stat),
2763         ONE("statm",      S_IRUGO, proc_pid_statm),
2764         REG("maps",       S_IRUGO, proc_maps_operations),
2765 #ifdef CONFIG_NUMA
2766         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2767 #endif
2768         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2769         LNK("cwd",        proc_cwd_link),
2770         LNK("root",       proc_root_link),
2771         LNK("exe",        proc_exe_link),
2772         REG("mounts",     S_IRUGO, proc_mounts_operations),
2773         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2774         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2775 #ifdef CONFIG_PROC_PAGE_MONITOR
2776         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2777         REG("smaps",      S_IRUGO, proc_smaps_operations),
2778         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2779 #endif
2780 #ifdef CONFIG_SECURITY
2781         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2782 #endif
2783 #ifdef CONFIG_KALLSYMS
2784         INF("wchan",      S_IRUGO, proc_pid_wchan),
2785 #endif
2786 #ifdef CONFIG_STACKTRACE
2787         ONE("stack",      S_IRUSR, proc_pid_stack),
2788 #endif
2789 #ifdef CONFIG_SCHEDSTATS
2790         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2791 #endif
2792 #ifdef CONFIG_LATENCYTOP
2793         REG("latency",  S_IRUGO, proc_lstats_operations),
2794 #endif
2795 #ifdef CONFIG_PROC_PID_CPUSET
2796         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2797 #endif
2798 #ifdef CONFIG_CGROUPS
2799         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2800 #endif
2801         INF("oom_score",  S_IRUGO, proc_oom_score),
2802         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2803         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2804 #ifdef CONFIG_AUDITSYSCALL
2805         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2806         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2807 #endif
2808 #ifdef CONFIG_FAULT_INJECTION
2809         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2810 #endif
2811 #ifdef CONFIG_ELF_CORE
2812         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2813 #endif
2814 #ifdef CONFIG_TASK_IO_ACCOUNTING
2815         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2816 #endif
2817 };
2818
2819 static int proc_tgid_base_readdir(struct file * filp,
2820                              void * dirent, filldir_t filldir)
2821 {
2822         return proc_pident_readdir(filp,dirent,filldir,
2823                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2824 }
2825
2826 static const struct file_operations proc_tgid_base_operations = {
2827         .read           = generic_read_dir,
2828         .readdir        = proc_tgid_base_readdir,
2829         .llseek         = default_llseek,
2830 };
2831
2832 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2833         return proc_pident_lookup(dir, dentry,
2834                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2835 }
2836
2837 static const struct inode_operations proc_tgid_base_inode_operations = {
2838         .lookup         = proc_tgid_base_lookup,
2839         .getattr        = pid_getattr,
2840         .setattr        = proc_setattr,
2841 };
2842
2843 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2844 {
2845         struct dentry *dentry, *leader, *dir;
2846         char buf[PROC_NUMBUF];
2847         struct qstr name;
2848
2849         name.name = buf;
2850         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2851         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2852         if (dentry) {
2853                 shrink_dcache_parent(dentry);
2854                 d_drop(dentry);
2855                 dput(dentry);
2856         }
2857
2858         name.name = buf;
2859         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2860         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2861         if (!leader)
2862                 goto out;
2863
2864         name.name = "task";
2865         name.len = strlen(name.name);
2866         dir = d_hash_and_lookup(leader, &name);
2867         if (!dir)
2868                 goto out_put_leader;
2869
2870         name.name = buf;
2871         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2872         dentry = d_hash_and_lookup(dir, &name);
2873         if (dentry) {
2874                 shrink_dcache_parent(dentry);
2875                 d_drop(dentry);
2876                 dput(dentry);
2877         }
2878
2879         dput(dir);
2880 out_put_leader:
2881         dput(leader);
2882 out:
2883         return;
2884 }
2885
2886 /**
2887  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2888  * @task: task that should be flushed.
2889  *
2890  * When flushing dentries from proc, one needs to flush them from global
2891  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2892  * in. This call is supposed to do all of this job.
2893  *
2894  * Looks in the dcache for
2895  * /proc/@pid
2896  * /proc/@tgid/task/@pid
2897  * if either directory is present flushes it and all of it'ts children
2898  * from the dcache.
2899  *
2900  * It is safe and reasonable to cache /proc entries for a task until
2901  * that task exits.  After that they just clog up the dcache with
2902  * useless entries, possibly causing useful dcache entries to be
2903  * flushed instead.  This routine is proved to flush those useless
2904  * dcache entries at process exit time.
2905  *
2906  * NOTE: This routine is just an optimization so it does not guarantee
2907  *       that no dcache entries will exist at process exit time it
2908  *       just makes it very unlikely that any will persist.
2909  */
2910
2911 void proc_flush_task(struct task_struct *task)
2912 {
2913         int i;
2914         struct pid *pid, *tgid;
2915         struct upid *upid;
2916
2917         pid = task_pid(task);
2918         tgid = task_tgid(task);
2919
2920         for (i = 0; i <= pid->level; i++) {
2921                 upid = &pid->numbers[i];
2922                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2923                                         tgid->numbers[i].nr);
2924         }
2925
2926         upid = &pid->numbers[pid->level];
2927         if (upid->nr == 1)
2928                 pid_ns_release_proc(upid->ns);
2929 }
2930
2931 static struct dentry *proc_pid_instantiate(struct inode *dir,
2932                                            struct dentry * dentry,
2933                                            struct task_struct *task, const void *ptr)
2934 {
2935         struct dentry *error = ERR_PTR(-ENOENT);
2936         struct inode *inode;
2937
2938         inode = proc_pid_make_inode(dir->i_sb, task);
2939         if (!inode)
2940                 goto out;
2941
2942         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2943         inode->i_op = &proc_tgid_base_inode_operations;
2944         inode->i_fop = &proc_tgid_base_operations;
2945         inode->i_flags|=S_IMMUTABLE;
2946
2947         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2948                 ARRAY_SIZE(tgid_base_stuff));
2949
2950         d_set_d_op(dentry, &pid_dentry_operations);
2951
2952         d_add(dentry, inode);
2953         /* Close the race of the process dying before we return the dentry */
2954         if (pid_revalidate(dentry, NULL))
2955                 error = NULL;
2956 out:
2957         return error;
2958 }
2959
2960 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2961 {
2962         struct dentry *result;
2963         struct task_struct *task;
2964         unsigned tgid;
2965         struct pid_namespace *ns;
2966
2967         result = proc_base_lookup(dir, dentry);
2968         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2969                 goto out;
2970
2971         tgid = name_to_int(dentry);
2972         if (tgid == ~0U)
2973                 goto out;
2974
2975         ns = dentry->d_sb->s_fs_info;
2976         rcu_read_lock();
2977         task = find_task_by_pid_ns(tgid, ns);
2978         if (task)
2979                 get_task_struct(task);
2980         rcu_read_unlock();
2981         if (!task)
2982                 goto out;
2983
2984         result = proc_pid_instantiate(dir, dentry, task, NULL);
2985         put_task_struct(task);
2986 out:
2987         return result;
2988 }
2989
2990 /*
2991  * Find the first task with tgid >= tgid
2992  *
2993  */
2994 struct tgid_iter {
2995         unsigned int tgid;
2996         struct task_struct *task;
2997 };
2998 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2999 {
3000         struct pid *pid;
3001
3002         if (iter.task)
3003                 put_task_struct(iter.task);
3004         rcu_read_lock();
3005 retry:
3006         iter.task = NULL;
3007         pid = find_ge_pid(iter.tgid, ns);
3008         if (pid) {
3009                 iter.tgid = pid_nr_ns(pid, ns);
3010                 iter.task = pid_task(pid, PIDTYPE_PID);
3011                 /* What we to know is if the pid we have find is the
3012                  * pid of a thread_group_leader.  Testing for task
3013                  * being a thread_group_leader is the obvious thing
3014                  * todo but there is a window when it fails, due to
3015                  * the pid transfer logic in de_thread.
3016                  *
3017                  * So we perform the straight forward test of seeing
3018                  * if the pid we have found is the pid of a thread
3019                  * group leader, and don't worry if the task we have
3020                  * found doesn't happen to be a thread group leader.
3021                  * As we don't care in the case of readdir.
3022                  */
3023                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3024                         iter.tgid += 1;
3025                         goto retry;
3026                 }
3027                 get_task_struct(iter.task);
3028         }
3029         rcu_read_unlock();
3030         return iter;
3031 }
3032
3033 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3034
3035 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3036         struct tgid_iter iter)
3037 {
3038         char name[PROC_NUMBUF];
3039         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3040         return proc_fill_cache(filp, dirent, filldir, name, len,
3041                                 proc_pid_instantiate, iter.task, NULL);
3042 }
3043
3044 /* for the /proc/ directory itself, after non-process stuff has been done */
3045 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3046 {
3047         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3048         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
3049         struct tgid_iter iter;
3050         struct pid_namespace *ns;
3051
3052         if (!reaper)
3053                 goto out_no_task;
3054
3055         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3056                 const struct pid_entry *p = &proc_base_stuff[nr];
3057                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3058                         goto out;
3059         }
3060
3061         ns = filp->f_dentry->d_sb->s_fs_info;
3062         iter.task = NULL;
3063         iter.tgid = filp->f_pos - TGID_OFFSET;
3064         for (iter = next_tgid(ns, iter);
3065              iter.task;
3066              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3067                 filp->f_pos = iter.tgid + TGID_OFFSET;
3068                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3069                         put_task_struct(iter.task);
3070                         goto out;
3071                 }
3072         }
3073         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3074 out:
3075         put_task_struct(reaper);
3076 out_no_task:
3077         return 0;
3078 }
3079
3080 /*
3081  * Tasks
3082  */
3083 static const struct pid_entry tid_base_stuff[] = {
3084         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3085         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3086         REG("environ",   S_IRUSR, proc_environ_operations),
3087         INF("auxv",      S_IRUSR, proc_pid_auxv),
3088         ONE("status",    S_IRUGO, proc_pid_status),
3089         ONE("personality", S_IRUSR, proc_pid_personality),
3090         INF("limits",    S_IRUGO, proc_pid_limits),
3091 #ifdef CONFIG_SCHED_DEBUG
3092         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3093 #endif
3094         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3095 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3096         INF("syscall",   S_IRUSR, proc_pid_syscall),
3097 #endif
3098         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3099         ONE("stat",      S_IRUGO, proc_tid_stat),
3100         ONE("statm",     S_IRUGO, proc_pid_statm),
3101         REG("maps",      S_IRUGO, proc_maps_operations),
3102 #ifdef CONFIG_NUMA
3103         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3104 #endif
3105         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3106         LNK("cwd",       proc_cwd_link),
3107         LNK("root",      proc_root_link),
3108         LNK("exe",       proc_exe_link),
3109         REG("mounts",    S_IRUGO, proc_mounts_operations),
3110         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3111 #ifdef CONFIG_PROC_PAGE_MONITOR
3112         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3113         REG("smaps",     S_IRUGO, proc_smaps_operations),
3114         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3115 #endif
3116 #ifdef CONFIG_SECURITY
3117         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3118 #endif
3119 #ifdef CONFIG_KALLSYMS
3120         INF("wchan",     S_IRUGO, proc_pid_wchan),
3121 #endif
3122 #ifdef CONFIG_STACKTRACE
3123         ONE("stack",      S_IRUSR, proc_pid_stack),
3124 #endif
3125 #ifdef CONFIG_SCHEDSTATS
3126         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3127 #endif
3128 #ifdef CONFIG_LATENCYTOP
3129         REG("latency",  S_IRUGO, proc_lstats_operations),
3130 #endif
3131 #ifdef CONFIG_PROC_PID_CPUSET
3132         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3133 #endif
3134 #ifdef CONFIG_CGROUPS
3135         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3136 #endif
3137         INF("oom_score", S_IRUGO, proc_oom_score),
3138         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3139         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3140 #ifdef CONFIG_AUDITSYSCALL
3141         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3142         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
3143 #endif
3144 #ifdef CONFIG_FAULT_INJECTION
3145         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3146 #endif
3147 #ifdef CONFIG_TASK_IO_ACCOUNTING
3148         INF("io",       S_IRUGO, proc_tid_io_accounting),
3149 #endif
3150 };
3151
3152 static int proc_tid_base_readdir(struct file * filp,
3153                              void * dirent, filldir_t filldir)
3154 {
3155         return proc_pident_readdir(filp,dirent,filldir,
3156                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3157 }
3158
3159 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3160         return proc_pident_lookup(dir, dentry,
3161                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3162 }
3163
3164 static const struct file_operations proc_tid_base_operations = {
3165         .read           = generic_read_dir,
3166         .readdir        = proc_tid_base_readdir,
3167         .llseek         = default_llseek,
3168 };
3169
3170 static const struct inode_operations proc_tid_base_inode_operations = {
3171         .lookup         = proc_tid_base_lookup,
3172         .getattr        = pid_getattr,
3173         .setattr        = proc_setattr,
3174 };
3175
3176 static struct dentry *proc_task_instantiate(struct inode *dir,
3177         struct dentry *dentry, struct task_struct *task, const void *ptr)
3178 {
3179         struct dentry *error = ERR_PTR(-ENOENT);
3180         struct inode *inode;
3181         inode = proc_pid_make_inode(dir->i_sb, task);
3182
3183         if (!inode)
3184                 goto out;
3185         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3186         inode->i_op = &proc_tid_base_inode_operations;
3187         inode->i_fop = &proc_tid_base_operations;
3188         inode->i_flags|=S_IMMUTABLE;
3189
3190         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3191                 ARRAY_SIZE(tid_base_stuff));
3192
3193         d_set_d_op(dentry, &pid_dentry_operations);
3194
3195         d_add(dentry, inode);
3196         /* Close the race of the process dying before we return the dentry */
3197         if (pid_revalidate(dentry, NULL))
3198                 error = NULL;
3199 out:
3200         return error;
3201 }
3202
3203 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3204 {
3205         struct dentry *result = ERR_PTR(-ENOENT);
3206         struct task_struct *task;
3207         struct task_struct *leader = get_proc_task(dir);
3208         unsigned tid;
3209         struct pid_namespace *ns;
3210
3211         if (!leader)
3212                 goto out_no_task;
3213
3214         tid = name_to_int(dentry);
3215         if (tid == ~0U)
3216                 goto out;
3217
3218         ns = dentry->d_sb->s_fs_info;
3219         rcu_read_lock();
3220         task = find_task_by_pid_ns(tid, ns);
3221         if (task)
3222                 get_task_struct(task);
3223         rcu_read_unlock();
3224         if (!task)
3225                 goto out;
3226         if (!same_thread_group(leader, task))
3227                 goto out_drop_task;
3228
3229         result = proc_task_instantiate(dir, dentry, task, NULL);
3230 out_drop_task:
3231         put_task_struct(task);
3232 out:
3233         put_task_struct(leader);
3234 out_no_task:
3235         return result;
3236 }
3237
3238 /*
3239  * Find the first tid of a thread group to return to user space.
3240  *
3241  * Usually this is just the thread group leader, but if the users
3242  * buffer was too small or there was a seek into the middle of the
3243  * directory we have more work todo.
3244  *
3245  * In the case of a short read we start with find_task_by_pid.
3246  *
3247  * In the case of a seek we start with the leader and walk nr
3248  * threads past it.
3249  */
3250 static struct task_struct *first_tid(struct task_struct *leader,
3251                 int tid, int nr, struct pid_namespace *ns)
3252 {
3253         struct task_struct *pos;
3254
3255         rcu_read_lock();
3256         /* Attempt to start with the pid of a thread */
3257         if (tid && (nr > 0)) {
3258                 pos = find_task_by_pid_ns(tid, ns);
3259                 if (pos && (pos->group_leader == leader))
3260                         goto found;
3261         }
3262
3263         /* If nr exceeds the number of threads there is nothing todo */
3264         pos = NULL;
3265         if (nr && nr >= get_nr_threads(leader))
3266                 goto out;
3267
3268         /* If we haven't found our starting place yet start
3269          * with the leader and walk nr threads forward.
3270          */
3271         for (pos = leader; nr > 0; --nr) {
3272                 pos = next_thread(pos);
3273                 if (pos == leader) {
3274                         pos = NULL;
3275                         goto out;
3276                 }
3277         }
3278 found:
3279         get_task_struct(pos);
3280 out:
3281         rcu_read_unlock();
3282         return pos;
3283 }
3284
3285 /*
3286  * Find the next thread in the thread list.
3287  * Return NULL if there is an error or no next thread.
3288  *
3289  * The reference to the input task_struct is released.
3290  */
3291 static struct task_struct *next_tid(struct task_struct *start)
3292 {
3293         struct task_struct *pos = NULL;
3294         rcu_read_lock();
3295         if (pid_alive(start)) {
3296                 pos = next_thread(start);
3297                 if (thread_group_leader(pos))
3298                         pos = NULL;
3299                 else
3300                         get_task_struct(pos);
3301         }
3302         rcu_read_unlock();
3303         put_task_struct(start);
3304         return pos;
3305 }
3306
3307 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3308         struct task_struct *task, int tid)
3309 {
3310         char name[PROC_NUMBUF];
3311         int len = snprintf(name, sizeof(name), "%d", tid);
3312         return proc_fill_cache(filp, dirent, filldir, name, len,
3313                                 proc_task_instantiate, task, NULL);
3314 }
3315
3316 /* for the /proc/TGID/task/ directories */
3317 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3318 {
3319         struct dentry *dentry = filp->f_path.dentry;
3320         struct inode *inode = dentry->d_inode;
3321         struct task_struct *leader = NULL;
3322         struct task_struct *task;
3323         int retval = -ENOENT;
3324         ino_t ino;
3325         int tid;
3326         struct pid_namespace *ns;
3327
3328         task = get_proc_task(inode);
3329         if (!task)
3330                 goto out_no_task;
3331         rcu_read_lock();
3332         if (pid_alive(task)) {
3333                 leader = task->group_leader;
3334                 get_task_struct(leader);
3335         }
3336         rcu_read_unlock();
3337         put_task_struct(task);
3338         if (!leader)
3339                 goto out_no_task;
3340         retval = 0;
3341
3342         switch ((unsigned long)filp->f_pos) {
3343         case 0:
3344                 ino = inode->i_ino;
3345                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3346                         goto out;
3347                 filp->f_pos++;
3348                 /* fall through */
3349         case 1:
3350                 ino = parent_ino(dentry);
3351                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3352                         goto out;
3353                 filp->f_pos++;
3354                 /* fall through */
3355         }
3356
3357         /* f_version caches the tgid value that the last readdir call couldn't
3358          * return. lseek aka telldir automagically resets f_version to 0.
3359          */
3360         ns = filp->f_dentry->d_sb->s_fs_info;
3361         tid = (int)filp->f_version;
3362         filp->f_version = 0;
3363         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3364              task;
3365              task = next_tid(task), filp->f_pos++) {
3366                 tid = task_pid_nr_ns(task, ns);
3367                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3368                         /* returning this tgid failed, save it as the first
3369                          * pid for the next readir call */
3370                         filp->f_version = (u64)tid;
3371                         put_task_struct(task);
3372                         break;
3373                 }
3374         }
3375 out:
3376         put_task_struct(leader);
3377 out_no_task:
3378         return retval;
3379 }
3380
3381 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3382 {
3383         struct inode *inode = dentry->d_inode;
3384         struct task_struct *p = get_proc_task(inode);
3385         generic_fillattr(inode, stat);
3386
3387         if (p) {
3388                 stat->nlink += get_nr_threads(p);
3389                 put_task_struct(p);
3390         }
3391
3392         return 0;
3393 }
3394
3395 static const struct inode_operations proc_task_inode_operations = {
3396         .lookup         = proc_task_lookup,
3397         .getattr        = proc_task_getattr,
3398         .setattr        = proc_setattr,
3399 };
3400
3401 static const struct file_operations proc_task_operations = {
3402         .read           = generic_read_dir,
3403         .readdir        = proc_task_readdir,
3404         .llseek         = default_llseek,
3405 };