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