fs/proc/base.c

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