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