fs/proc/base.c

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