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(d_inode(dentry));
 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(d_inode(dentry));
 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 ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 200                                      size_t _count, loff_t *pos)
 201 {
 202         struct task_struct *tsk;
 203         struct mm_struct *mm;
 204         char *page;
 205         unsigned long count = _count;
 206         unsigned long arg_start, arg_end, env_start, env_end;
 207         unsigned long len1, len2, len;
 208         unsigned long p;
 209         char c;
 210         ssize_t rv;
 211
 212         BUG_ON(*pos < 0);
 213
 214         tsk = get_proc_task(file_inode(file));
 215         if (!tsk)
 216                 return -ESRCH;
 217         mm = get_task_mm(tsk);
 218         put_task_struct(tsk);
 219         if (!mm)
 220                 return 0;
 221         /* Check if process spawned far enough to have cmdline. */
 222         if (!mm->env_end) {
 223                 rv = 0;
 224                 goto out_mmput;
 225         }
 226
 227         page = (char *)__get_free_page(GFP_TEMPORARY);
 228         if (!page) {
 229                 rv = -ENOMEM;
 230                 goto out_mmput;
 231         }
 232
 233         down_read(&mm->mmap_sem);
 234         arg_start = mm->arg_start;
 235         arg_end = mm->arg_end;
 236         env_start = mm->env_start;
 237         env_end = mm->env_end;
 238         up_read(&mm->mmap_sem);
 239
 240         BUG_ON(arg_start > arg_end);
 241         BUG_ON(env_start > env_end);
 242
 243         len1 = arg_end - arg_start;
 244         len2 = env_end - env_start;
 245
 246         /*
 247          * Inherently racy -- command line shares address space
 248          * with code and data.
 249          */
 250         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 251         if (rv <= 0)
 252                 goto out_free_page;
 253
 254         rv = 0;
 255
 256         if (c == '\0') {
 257                 /* Command line (set of strings) occupies whole ARGV. */
 258                 if (len1 <= *pos)
 259                         goto out_free_page;
 260
 261                 p = arg_start + *pos;
 262                 len = len1 - *pos;
 263                 while (count > 0 && len > 0) {
 264                         unsigned int _count;
 265                         int nr_read;
 266
 267                         _count = min3(count, len, PAGE_SIZE);
 268                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 269                         if (nr_read < 0)
 270                                 rv = nr_read;
 271                         if (nr_read <= 0)
 272                                 goto out_free_page;
 273
 274                         if (copy_to_user(buf, page, nr_read)) {
 275                                 rv = -EFAULT;
 276                                 goto out_free_page;
 277                         }
 278
 279                         p       += nr_read;
 280                         len     -= nr_read;
 281                         buf     += nr_read;
 282                         count   -= nr_read;
 283                         rv      += nr_read;
 284                 }
 285         } else {
 286                 /*
 287                  * Command line (1 string) occupies ARGV and maybe
 288                  * extends into ENVP.
 289                  */
 290                 if (len1 + len2 <= *pos)
 291                         goto skip_argv_envp;
 292                 if (len1 <= *pos)
 293                         goto skip_argv;
 294
 295                 p = arg_start + *pos;
 296                 len = len1 - *pos;
 297                 while (count > 0 && len > 0) {
 298                         unsigned int _count, l;
 299                         int nr_read;
 300                         bool final;
 301
 302                         _count = min3(count, len, PAGE_SIZE);
 303                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 304                         if (nr_read < 0)
 305                                 rv = nr_read;
 306                         if (nr_read <= 0)
 307                                 goto out_free_page;
 308
 309                         /*
 310                          * Command line can be shorter than whole ARGV
 311                          * even if last "marker" byte says it is not.
 312                          */
 313                         final = false;
 314                         l = strnlen(page, nr_read);
 315                         if (l < nr_read) {
 316                                 nr_read = l;
 317                                 final = true;
 318                         }
 319
 320                         if (copy_to_user(buf, page, nr_read)) {
 321                                 rv = -EFAULT;
 322                                 goto out_free_page;
 323                         }
 324
 325                         p       += nr_read;
 326                         len     -= nr_read;
 327                         buf     += nr_read;
 328                         count   -= nr_read;
 329                         rv      += nr_read;
 330
 331                         if (final)
 332                                 goto out_free_page;
 333                 }
 334 skip_argv:
 335                 /*
 336                  * Command line (1 string) occupies ARGV and
 337                  * extends into ENVP.
 338                  */
 339                 if (len1 <= *pos) {
 340                         p = env_start + *pos - len1;
 341                         len = len1 + len2 - *pos;
 342                 } else {
 343                         p = env_start;
 344                         len = len2;
 345                 }
 346                 while (count > 0 && len > 0) {
 347                         unsigned int _count, l;
 348                         int nr_read;
 349                         bool final;
 350
 351                         _count = min3(count, len, PAGE_SIZE);
 352                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 353                         if (nr_read < 0)
 354                                 rv = nr_read;
 355                         if (nr_read <= 0)
 356                                 goto out_free_page;
 357
 358                         /* Find EOS. */
 359                         final = false;
 360                         l = strnlen(page, nr_read);
 361                         if (l < nr_read) {
 362                                 nr_read = l;
 363                                 final = true;
 364                         }
 365
 366                         if (copy_to_user(buf, page, nr_read)) {
 367                                 rv = -EFAULT;
 368                                 goto out_free_page;
 369                         }
 370
 371                         p       += nr_read;
 372                         len     -= nr_read;
 373                         buf     += nr_read;
 374                         count   -= nr_read;
 375                         rv      += nr_read;
 376
 377                         if (final)
 378                                 goto out_free_page;
 379                 }
 380 skip_argv_envp:
 381                 ;
 382         }
 383
 384 out_free_page:
 385         free_page((unsigned long)page);
 386 out_mmput:
 387         mmput(mm);
 388         if (rv > 0)
 389                 *pos += rv;
 390         return rv;
 391 }
 392
 393 static const struct file_operations proc_pid_cmdline_ops = {
 394         .read   = proc_pid_cmdline_read,
 395         .llseek = generic_file_llseek,
 396 };
 397
 398 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 399                          struct pid *pid, struct task_struct *task)
 400 {
 401         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 402         if (mm && !IS_ERR(mm)) {
 403                 unsigned int nwords = 0;
 404                 do {
 405                         nwords += 2;
 406                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 407                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 408                 mmput(mm);
 409                 return 0;
 410         } else
 411                 return PTR_ERR(mm);
 412 }
 413
 414
 415 #ifdef CONFIG_KALLSYMS
 416 /*
 417  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 418  * Returns the resolved symbol.  If that fails, simply return the address.
 419  */
 420 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 421                           struct pid *pid, struct task_struct *task)
 422 {
 423         unsigned long wchan;
 424         char symname[KSYM_NAME_LEN];
 425
 426         wchan = get_wchan(task);
 427
 428         if (lookup_symbol_name(wchan, symname) < 0) {
 429                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 430                         return 0;
 431                 seq_printf(m, "%lu", wchan);
 432         } else {
 433                 seq_printf(m, "%s", symname);
 434         }
 435
 436         return 0;
 437 }
 438 #endif /* CONFIG_KALLSYMS */
 439
 440 static int lock_trace(struct task_struct *task)
 441 {
 442         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 443         if (err)
 444                 return err;
 445         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 446                 mutex_unlock(&task->signal->cred_guard_mutex);
 447                 return -EPERM;
 448         }
 449         return 0;
 450 }
 451
 452 static void unlock_trace(struct task_struct *task)
 453 {
 454         mutex_unlock(&task->signal->cred_guard_mutex);
 455 }
 456
 457 #ifdef CONFIG_STACKTRACE
 458
 459 #define MAX_STACK_TRACE_DEPTH   64
 460
 461 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 462                           struct pid *pid, struct task_struct *task)
 463 {
 464         struct stack_trace trace;
 465         unsigned long *entries;
 466         int err;
 467         int i;
 468
 469         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 470         if (!entries)
 471                 return -ENOMEM;
 472
 473         trace.nr_entries        = 0;
 474         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 475         trace.entries           = entries;
 476         trace.skip              = 0;
 477
 478         err = lock_trace(task);
 479         if (!err) {
 480                 save_stack_trace_tsk(task, &trace);
 481
 482                 for (i = 0; i < trace.nr_entries; i++) {
 483                         seq_printf(m, "[<%pK>] %pS\n",
 484                                    (void *)entries[i], (void *)entries[i]);
 485                 }
 486                 unlock_trace(task);
 487         }
 488         kfree(entries);
 489
 490         return err;
 491 }
 492 #endif
 493
 494 #ifdef CONFIG_SCHEDSTATS
 495 /*
 496  * Provides /proc/PID/schedstat
 497  */
 498 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 499                               struct pid *pid, struct task_struct *task)
 500 {
 501         seq_printf(m, "%llu %llu %lu\n",
 502                    (unsigned long long)task->se.sum_exec_runtime,
 503                    (unsigned long long)task->sched_info.run_delay,
 504                    task->sched_info.pcount);
 505
 506         return 0;
 507 }
 508 #endif
 509
 510 #ifdef CONFIG_LATENCYTOP
 511 static int lstats_show_proc(struct seq_file *m, void *v)
 512 {
 513         int i;
 514         struct inode *inode = m->private;
 515         struct task_struct *task = get_proc_task(inode);
 516
 517         if (!task)
 518                 return -ESRCH;
 519         seq_puts(m, "Latency Top version : v0.1\n");
 520         for (i = 0; i < 32; i++) {
 521                 struct latency_record *lr = &task->latency_record[i];
 522                 if (lr->backtrace[0]) {
 523                         int q;
 524                         seq_printf(m, "%i %li %li",
 525                                    lr->count, lr->time, lr->max);
 526                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 527                                 unsigned long bt = lr->backtrace[q];
 528                                 if (!bt)
 529                                         break;
 530                                 if (bt == ULONG_MAX)
 531                                         break;
 532                                 seq_printf(m, " %ps", (void *)bt);
 533                         }
 534                         seq_putc(m, '\n');
 535                 }
 536
 537         }
 538         put_task_struct(task);
 539         return 0;
 540 }
 541
 542 static int lstats_open(struct inode *inode, struct file *file)
 543 {
 544         return single_open(file, lstats_show_proc, inode);
 545 }
 546
 547 static ssize_t lstats_write(struct file *file, const char __user *buf,
 548                             size_t count, loff_t *offs)
 549 {
 550         struct task_struct *task = get_proc_task(file_inode(file));
 551
 552         if (!task)
 553                 return -ESRCH;
 554         clear_all_latency_tracing(task);
 555         put_task_struct(task);
 556
 557         return count;
 558 }
 559
 560 static const struct file_operations proc_lstats_operations = {
 561         .open           = lstats_open,
 562         .read           = seq_read,
 563         .write          = lstats_write,
 564         .llseek         = seq_lseek,
 565         .release        = single_release,
 566 };
 567
 568 #endif
 569
 570 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 571                           struct pid *pid, struct task_struct *task)
 572 {
 573         unsigned long totalpages = totalram_pages + total_swap_pages;
 574         unsigned long points = 0;
 575
 576         read_lock(&tasklist_lock);
 577         if (pid_alive(task))
 578                 points = oom_badness(task, NULL, NULL, totalpages) *
 579                                                 1000 / totalpages;
 580         read_unlock(&tasklist_lock);
 581         seq_printf(m, "%lu\n", points);
 582
 583         return 0;
 584 }
 585
 586 struct limit_names {
 587         const char *name;
 588         const char *unit;
 589 };
 590
 591 static const struct limit_names lnames[RLIM_NLIMITS] = {
 592         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 593         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 594         [RLIMIT_DATA] = {"Max data size", "bytes"},
 595         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 596         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 597         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 598         [RLIMIT_NPROC] = {"Max processes", "processes"},
 599         [RLIMIT_NOFILE] = {"Max open files", "files"},
 600         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 601         [RLIMIT_AS] = {"Max address space", "bytes"},
 602         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 603         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 604         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 605         [RLIMIT_NICE] = {"Max nice priority", NULL},
 606         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 607         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 608 };
 609
 610 /* Display limits for a process */
 611 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 612                            struct pid *pid, struct task_struct *task)
 613 {
 614         unsigned int i;
 615         unsigned long flags;
 616
 617         struct rlimit rlim[RLIM_NLIMITS];
 618
 619         if (!lock_task_sighand(task, &flags))
 620                 return 0;
 621         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 622         unlock_task_sighand(task, &flags);
 623
 624         /*
 625          * print the file header
 626          */
 627        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 628                   "Limit", "Soft Limit", "Hard Limit", "Units");
 629
 630         for (i = 0; i < RLIM_NLIMITS; i++) {
 631                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 632                         seq_printf(m, "%-25s %-20s ",
 633                                    lnames[i].name, "unlimited");
 634                 else
 635                         seq_printf(m, "%-25s %-20lu ",
 636                                    lnames[i].name, rlim[i].rlim_cur);
 637
 638                 if (rlim[i].rlim_max == RLIM_INFINITY)
 639                         seq_printf(m, "%-20s ", "unlimited");
 640                 else
 641                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 642
 643                 if (lnames[i].unit)
 644                         seq_printf(m, "%-10s\n", lnames[i].unit);
 645                 else
 646                         seq_putc(m, '\n');
 647         }
 648
 649         return 0;
 650 }
 651
 652 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 653 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 654                             struct pid *pid, struct task_struct *task)
 655 {
 656         long nr;
 657         unsigned long args[6], sp, pc;
 658         int res;
 659
 660         res = lock_trace(task);
 661         if (res)
 662                 return res;
 663
 664         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 665                 seq_puts(m, "running\n");
 666         else if (nr < 0)
 667                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 668         else
 669                 seq_printf(m,
 670                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 671                        nr,
 672                        args[0], args[1], args[2], args[3], args[4], args[5],
 673                        sp, pc);
 674         unlock_trace(task);
 675
 676         return 0;
 677 }
 678 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 679
 680 /************************************************************************/
 681 /*                       Here the fs part begins                        */
 682 /************************************************************************/
 683
 684 /* permission checks */
 685 static int proc_fd_access_allowed(struct inode *inode)
 686 {
 687         struct task_struct *task;
 688         int allowed = 0;
 689         /* Allow access to a task's file descriptors if it is us or we
 690          * may use ptrace attach to the process and find out that
 691          * information.
 692          */
 693         task = get_proc_task(inode);
 694         if (task) {
 695                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 696                 put_task_struct(task);
 697         }
 698         return allowed;
 699 }
 700
 701 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 702 {
 703         int error;
 704         struct inode *inode = d_inode(dentry);
 705
 706         if (attr->ia_valid & ATTR_MODE)
 707                 return -EPERM;
 708
 709         error = inode_change_ok(inode, attr);
 710         if (error)
 711                 return error;
 712
 713         setattr_copy(inode, attr);
 714         mark_inode_dirty(inode);
 715         return 0;
 716 }
 717
 718 /*
 719  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 720  * or euid/egid (for hide_pid_min=2)?
 721  */
 722 static bool has_pid_permissions(struct pid_namespace *pid,
 723                                  struct task_struct *task,
 724                                  int hide_pid_min)
 725 {
 726         if (pid->hide_pid < hide_pid_min)
 727                 return true;
 728         if (in_group_p(pid->pid_gid))
 729                 return true;
 730         return ptrace_may_access(task, PTRACE_MODE_READ);
 731 }
 732
 733
 734 static int proc_pid_permission(struct inode *inode, int mask)
 735 {
 736         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 737         struct task_struct *task;
 738         bool has_perms;
 739
 740         task = get_proc_task(inode);
 741         if (!task)
 742                 return -ESRCH;
 743         has_perms = has_pid_permissions(pid, task, 1);
 744         put_task_struct(task);
 745
 746         if (!has_perms) {
 747                 if (pid->hide_pid == 2) {
 748                         /*
 749                          * Let's make getdents(), stat(), and open()
 750                          * consistent with each other.  If a process
 751                          * may not stat() a file, it shouldn't be seen
 752                          * in procfs at all.
 753                          */
 754                         return -ENOENT;
 755                 }
 756
 757                 return -EPERM;
 758         }
 759         return generic_permission(inode, mask);
 760 }
 761
 762
 763
 764 static const struct inode_operations proc_def_inode_operations = {
 765         .setattr        = proc_setattr,
 766 };
 767
 768 static int proc_single_show(struct seq_file *m, void *v)
 769 {
 770         struct inode *inode = m->private;
 771         struct pid_namespace *ns;
 772         struct pid *pid;
 773         struct task_struct *task;
 774         int ret;
 775
 776         ns = inode->i_sb->s_fs_info;
 777         pid = proc_pid(inode);
 778         task = get_pid_task(pid, PIDTYPE_PID);
 779         if (!task)
 780                 return -ESRCH;
 781
 782         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 783
 784         put_task_struct(task);
 785         return ret;
 786 }
 787
 788 static int proc_single_open(struct inode *inode, struct file *filp)
 789 {
 790         return single_open(filp, proc_single_show, inode);
 791 }
 792
 793 static const struct file_operations proc_single_file_operations = {
 794         .open           = proc_single_open,
 795         .read           = seq_read,
 796         .llseek         = seq_lseek,
 797         .release        = single_release,
 798 };
 799
 800
 801 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 802 {
 803         struct task_struct *task = get_proc_task(inode);
 804         struct mm_struct *mm = ERR_PTR(-ESRCH);
 805
 806         if (task) {
 807                 mm = mm_access(task, mode);
 808                 put_task_struct(task);
 809
 810                 if (!IS_ERR_OR_NULL(mm)) {
 811                         /* ensure this mm_struct can't be freed */
 812                         atomic_inc(&mm->mm_count);
 813                         /* but do not pin its memory */
 814                         mmput(mm);
 815                 }
 816         }
 817
 818         return mm;
 819 }
 820
 821 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 822 {
 823         struct mm_struct *mm = proc_mem_open(inode, mode);
 824
 825         if (IS_ERR(mm))
 826                 return PTR_ERR(mm);
 827
 828         file->private_data = mm;
 829         return 0;
 830 }
 831
 832 static int mem_open(struct inode *inode, struct file *file)
 833 {
 834         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 835
 836         /* OK to pass negative loff_t, we can catch out-of-range */
 837         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 838
 839         return ret;
 840 }
 841
 842 static ssize_t mem_rw(struct file *file, char __user *buf,
 843                         size_t count, loff_t *ppos, int write)
 844 {
 845         struct mm_struct *mm = file->private_data;
 846         unsigned long addr = *ppos;
 847         ssize_t copied;
 848         char *page;
 849
 850         if (!mm)
 851                 return 0;
 852
 853         page = (char *)__get_free_page(GFP_TEMPORARY);
 854         if (!page)
 855                 return -ENOMEM;
 856
 857         copied = 0;
 858         if (!atomic_inc_not_zero(&mm->mm_users))
 859                 goto free;
 860
 861         while (count > 0) {
 862                 int this_len = min_t(int, count, PAGE_SIZE);
 863
 864                 if (write && copy_from_user(page, buf, this_len)) {
 865                         copied = -EFAULT;
 866                         break;
 867                 }
 868
 869                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 870                 if (!this_len) {
 871                         if (!copied)
 872                                 copied = -EIO;
 873                         break;
 874                 }
 875
 876                 if (!write && copy_to_user(buf, page, this_len)) {
 877                         copied = -EFAULT;
 878                         break;
 879                 }
 880
 881                 buf += this_len;
 882                 addr += this_len;
 883                 copied += this_len;
 884                 count -= this_len;
 885         }
 886         *ppos = addr;
 887
 888         mmput(mm);
 889 free:
 890         free_page((unsigned long) page);
 891         return copied;
 892 }
 893
 894 static ssize_t mem_read(struct file *file, char __user *buf,
 895                         size_t count, loff_t *ppos)
 896 {
 897         return mem_rw(file, buf, count, ppos, 0);
 898 }
 899
 900 static ssize_t mem_write(struct file *file, const char __user *buf,
 901                          size_t count, loff_t *ppos)
 902 {
 903         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 904 }
 905
 906 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 907 {
 908         switch (orig) {
 909         case 0:
 910                 file->f_pos = offset;
 911                 break;
 912         case 1:
 913                 file->f_pos += offset;
 914                 break;
 915         default:
 916                 return -EINVAL;
 917         }
 918         force_successful_syscall_return();
 919         return file->f_pos;
 920 }
 921
 922 static int mem_release(struct inode *inode, struct file *file)
 923 {
 924         struct mm_struct *mm = file->private_data;
 925         if (mm)
 926                 mmdrop(mm);
 927         return 0;
 928 }
 929
 930 static const struct file_operations proc_mem_operations = {
 931         .llseek         = mem_lseek,
 932         .read           = mem_read,
 933         .write          = mem_write,
 934         .open           = mem_open,
 935         .release        = mem_release,
 936 };
 937
 938 static int environ_open(struct inode *inode, struct file *file)
 939 {
 940         return __mem_open(inode, file, PTRACE_MODE_READ);
 941 }
 942
 943 static ssize_t environ_read(struct file *file, char __user *buf,
 944                         size_t count, loff_t *ppos)
 945 {
 946         char *page;
 947         unsigned long src = *ppos;
 948         int ret = 0;
 949         struct mm_struct *mm = file->private_data;
 950
 951         if (!mm)
 952                 return 0;
 953
 954         page = (char *)__get_free_page(GFP_TEMPORARY);
 955         if (!page)
 956                 return -ENOMEM;
 957
 958         ret = 0;
 959         if (!atomic_inc_not_zero(&mm->mm_users))
 960                 goto free;
 961         while (count > 0) {
 962                 size_t this_len, max_len;
 963                 int retval;
 964
 965                 if (src >= (mm->env_end - mm->env_start))
 966                         break;
 967
 968                 this_len = mm->env_end - (mm->env_start + src);
 969
 970                 max_len = min_t(size_t, PAGE_SIZE, count);
 971                 this_len = min(max_len, this_len);
 972
 973                 retval = access_remote_vm(mm, (mm->env_start + src),
 974                         page, this_len, 0);
 975
 976                 if (retval <= 0) {
 977                         ret = retval;
 978                         break;
 979                 }
 980
 981                 if (copy_to_user(buf, page, retval)) {
 982                         ret = -EFAULT;
 983                         break;
 984                 }
 985
 986                 ret += retval;
 987                 src += retval;
 988                 buf += retval;
 989                 count -= retval;
 990         }
 991         *ppos = src;
 992         mmput(mm);
 993
 994 free:
 995         free_page((unsigned long) page);
 996         return ret;
 997 }
 998
 999 static const struct file_operations proc_environ_operations = {
1000         .open           = environ_open,
1001         .read           = environ_read,
1002         .llseek         = generic_file_llseek,
1003         .release        = mem_release,
1004 };
1005
1006 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1007                             loff_t *ppos)
1008 {
1009         struct task_struct *task = get_proc_task(file_inode(file));
1010         char buffer[PROC_NUMBUF];
1011         int oom_adj = OOM_ADJUST_MIN;
1012         size_t len;
1013         unsigned long flags;
1014
1015         if (!task)
1016                 return -ESRCH;
1017         if (lock_task_sighand(task, &flags)) {
1018                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1019                         oom_adj = OOM_ADJUST_MAX;
1020                 else
1021                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1022                                   OOM_SCORE_ADJ_MAX;
1023                 unlock_task_sighand(task, &flags);
1024         }
1025         put_task_struct(task);
1026         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1027         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1028 }
1029
1030 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1031                              size_t count, loff_t *ppos)
1032 {
1033         struct task_struct *task;
1034         char buffer[PROC_NUMBUF];
1035         int oom_adj;
1036         unsigned long flags;
1037         int err;
1038
1039         memset(buffer, 0, sizeof(buffer));
1040         if (count > sizeof(buffer) - 1)
1041                 count = sizeof(buffer) - 1;
1042         if (copy_from_user(buffer, buf, count)) {
1043                 err = -EFAULT;
1044                 goto out;
1045         }
1046
1047         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1048         if (err)
1049                 goto out;
1050         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1051              oom_adj != OOM_DISABLE) {
1052                 err = -EINVAL;
1053                 goto out;
1054         }
1055
1056         task = get_proc_task(file_inode(file));
1057         if (!task) {
1058                 err = -ESRCH;
1059                 goto out;
1060         }
1061
1062         task_lock(task);
1063         if (!task->mm) {
1064                 err = -EINVAL;
1065                 goto err_task_lock;
1066         }
1067
1068         if (!lock_task_sighand(task, &flags)) {
1069                 err = -ESRCH;
1070                 goto err_task_lock;
1071         }
1072
1073         /*
1074          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1075          * value is always attainable.
1076          */
1077         if (oom_adj == OOM_ADJUST_MAX)
1078                 oom_adj = OOM_SCORE_ADJ_MAX;
1079         else
1080                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1081
1082         if (oom_adj < task->signal->oom_score_adj &&
1083             !capable(CAP_SYS_RESOURCE)) {
1084                 err = -EACCES;
1085                 goto err_sighand;
1086         }
1087
1088         /*
1089          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1090          * /proc/pid/oom_score_adj instead.
1091          */
1092         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1093                   current->comm, task_pid_nr(current), task_pid_nr(task),
1094                   task_pid_nr(task));
1095
1096         task->signal->oom_score_adj = oom_adj;
1097         trace_oom_score_adj_update(task);
1098 err_sighand:
1099         unlock_task_sighand(task, &flags);
1100 err_task_lock:
1101         task_unlock(task);
1102         put_task_struct(task);
1103 out:
1104         return err < 0 ? err : count;
1105 }
1106
1107 static const struct file_operations proc_oom_adj_operations = {
1108         .read           = oom_adj_read,
1109         .write          = oom_adj_write,
1110         .llseek         = generic_file_llseek,
1111 };
1112
1113 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1114                                         size_t count, loff_t *ppos)
1115 {
1116         struct task_struct *task = get_proc_task(file_inode(file));
1117         char buffer[PROC_NUMBUF];
1118         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1119         unsigned long flags;
1120         size_t len;
1121
1122         if (!task)
1123                 return -ESRCH;
1124         if (lock_task_sighand(task, &flags)) {
1125                 oom_score_adj = task->signal->oom_score_adj;
1126                 unlock_task_sighand(task, &flags);
1127         }
1128         put_task_struct(task);
1129         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1130         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1131 }
1132
1133 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1134                                         size_t count, loff_t *ppos)
1135 {
1136         struct task_struct *task;
1137         char buffer[PROC_NUMBUF];
1138         unsigned long flags;
1139         int oom_score_adj;
1140         int err;
1141
1142         memset(buffer, 0, sizeof(buffer));
1143         if (count > sizeof(buffer) - 1)
1144                 count = sizeof(buffer) - 1;
1145         if (copy_from_user(buffer, buf, count)) {
1146                 err = -EFAULT;
1147                 goto out;
1148         }
1149
1150         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1151         if (err)
1152                 goto out;
1153         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1154                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1155                 err = -EINVAL;
1156                 goto out;
1157         }
1158
1159         task = get_proc_task(file_inode(file));
1160         if (!task) {
1161                 err = -ESRCH;
1162                 goto out;
1163         }
1164
1165         task_lock(task);
1166         if (!task->mm) {
1167                 err = -EINVAL;
1168                 goto err_task_lock;
1169         }
1170
1171         if (!lock_task_sighand(task, &flags)) {
1172                 err = -ESRCH;
1173                 goto err_task_lock;
1174         }
1175
1176         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1177                         !capable(CAP_SYS_RESOURCE)) {
1178                 err = -EACCES;
1179                 goto err_sighand;
1180         }
1181
1182         task->signal->oom_score_adj = (short)oom_score_adj;
1183         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1184                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1185         trace_oom_score_adj_update(task);
1186
1187 err_sighand:
1188         unlock_task_sighand(task, &flags);
1189 err_task_lock:
1190         task_unlock(task);
1191         put_task_struct(task);
1192 out:
1193         return err < 0 ? err : count;
1194 }
1195
1196 static const struct file_operations proc_oom_score_adj_operations = {
1197         .read           = oom_score_adj_read,
1198         .write          = oom_score_adj_write,
1199         .llseek         = default_llseek,
1200 };
1201
1202 #ifdef CONFIG_AUDITSYSCALL
1203 #define TMPBUFLEN 21
1204 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1205                                   size_t count, loff_t *ppos)
1206 {
1207         struct inode * inode = file_inode(file);
1208         struct task_struct *task = get_proc_task(inode);
1209         ssize_t length;
1210         char tmpbuf[TMPBUFLEN];
1211
1212         if (!task)
1213                 return -ESRCH;
1214         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1215                            from_kuid(file->f_cred->user_ns,
1216                                      audit_get_loginuid(task)));
1217         put_task_struct(task);
1218         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1219 }
1220
1221 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1222                                    size_t count, loff_t *ppos)
1223 {
1224         struct inode * inode = file_inode(file);
1225         char *page, *tmp;
1226         ssize_t length;
1227         uid_t loginuid;
1228         kuid_t kloginuid;
1229
1230         rcu_read_lock();
1231         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1232                 rcu_read_unlock();
1233                 return -EPERM;
1234         }
1235         rcu_read_unlock();
1236
1237         if (count >= PAGE_SIZE)
1238                 count = PAGE_SIZE - 1;
1239
1240         if (*ppos != 0) {
1241                 /* No partial writes. */
1242                 return -EINVAL;
1243         }
1244         page = (char*)__get_free_page(GFP_TEMPORARY);
1245         if (!page)
1246                 return -ENOMEM;
1247         length = -EFAULT;
1248         if (copy_from_user(page, buf, count))
1249                 goto out_free_page;
1250
1251         page[count] = '\0';
1252         loginuid = simple_strtoul(page, &tmp, 10);
1253         if (tmp == page) {
1254                 length = -EINVAL;
1255                 goto out_free_page;
1256
1257         }
1258
1259         /* is userspace tring to explicitly UNSET the loginuid? */
1260         if (loginuid == AUDIT_UID_UNSET) {
1261                 kloginuid = INVALID_UID;
1262         } else {
1263                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1264                 if (!uid_valid(kloginuid)) {
1265                         length = -EINVAL;
1266                         goto out_free_page;
1267                 }
1268         }
1269
1270         length = audit_set_loginuid(kloginuid);
1271         if (likely(length == 0))
1272                 length = count;
1273
1274 out_free_page:
1275         free_page((unsigned long) page);
1276         return length;
1277 }
1278
1279 static const struct file_operations proc_loginuid_operations = {
1280         .read           = proc_loginuid_read,
1281         .write          = proc_loginuid_write,
1282         .llseek         = generic_file_llseek,
1283 };
1284
1285 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1286                                   size_t count, loff_t *ppos)
1287 {
1288         struct inode * inode = file_inode(file);
1289         struct task_struct *task = get_proc_task(inode);
1290         ssize_t length;
1291         char tmpbuf[TMPBUFLEN];
1292
1293         if (!task)
1294                 return -ESRCH;
1295         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1296                                 audit_get_sessionid(task));
1297         put_task_struct(task);
1298         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1299 }
1300
1301 static const struct file_operations proc_sessionid_operations = {
1302         .read           = proc_sessionid_read,
1303         .llseek         = generic_file_llseek,
1304 };
1305 #endif
1306
1307 #ifdef CONFIG_FAULT_INJECTION
1308 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1309                                       size_t count, loff_t *ppos)
1310 {
1311         struct task_struct *task = get_proc_task(file_inode(file));
1312         char buffer[PROC_NUMBUF];
1313         size_t len;
1314         int make_it_fail;
1315
1316         if (!task)
1317                 return -ESRCH;
1318         make_it_fail = task->make_it_fail;
1319         put_task_struct(task);
1320
1321         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1322
1323         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1324 }
1325
1326 static ssize_t proc_fault_inject_write(struct file * file,
1327                         const char __user * buf, size_t count, loff_t *ppos)
1328 {
1329         struct task_struct *task;
1330         char buffer[PROC_NUMBUF], *end;
1331         int make_it_fail;
1332
1333         if (!capable(CAP_SYS_RESOURCE))
1334                 return -EPERM;
1335         memset(buffer, 0, sizeof(buffer));
1336         if (count > sizeof(buffer) - 1)
1337                 count = sizeof(buffer) - 1;
1338         if (copy_from_user(buffer, buf, count))
1339                 return -EFAULT;
1340         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1341         if (*end)
1342                 return -EINVAL;
1343         if (make_it_fail < 0 || make_it_fail > 1)
1344                 return -EINVAL;
1345
1346         task = get_proc_task(file_inode(file));
1347         if (!task)
1348                 return -ESRCH;
1349         task->make_it_fail = make_it_fail;
1350         put_task_struct(task);
1351
1352         return count;
1353 }
1354
1355 static const struct file_operations proc_fault_inject_operations = {
1356         .read           = proc_fault_inject_read,
1357         .write          = proc_fault_inject_write,
1358         .llseek         = generic_file_llseek,
1359 };
1360 #endif
1361
1362
1363 #ifdef CONFIG_SCHED_DEBUG
1364 /*
1365  * Print out various scheduling related per-task fields:
1366  */
1367 static int sched_show(struct seq_file *m, void *v)
1368 {
1369         struct inode *inode = m->private;
1370         struct task_struct *p;
1371
1372         p = get_proc_task(inode);
1373         if (!p)
1374                 return -ESRCH;
1375         proc_sched_show_task(p, m);
1376
1377         put_task_struct(p);
1378
1379         return 0;
1380 }
1381
1382 static ssize_t
1383 sched_write(struct file *file, const char __user *buf,
1384             size_t count, loff_t *offset)
1385 {
1386         struct inode *inode = file_inode(file);
1387         struct task_struct *p;
1388
1389         p = get_proc_task(inode);
1390         if (!p)
1391                 return -ESRCH;
1392         proc_sched_set_task(p);
1393
1394         put_task_struct(p);
1395
1396         return count;
1397 }
1398
1399 static int sched_open(struct inode *inode, struct file *filp)
1400 {
1401         return single_open(filp, sched_show, inode);
1402 }
1403
1404 static const struct file_operations proc_pid_sched_operations = {
1405         .open           = sched_open,
1406         .read           = seq_read,
1407         .write          = sched_write,
1408         .llseek         = seq_lseek,
1409         .release        = single_release,
1410 };
1411
1412 #endif
1413
1414 #ifdef CONFIG_SCHED_AUTOGROUP
1415 /*
1416  * Print out autogroup related information:
1417  */
1418 static int sched_autogroup_show(struct seq_file *m, void *v)
1419 {
1420         struct inode *inode = m->private;
1421         struct task_struct *p;
1422
1423         p = get_proc_task(inode);
1424         if (!p)
1425                 return -ESRCH;
1426         proc_sched_autogroup_show_task(p, m);
1427
1428         put_task_struct(p);
1429
1430         return 0;
1431 }
1432
1433 static ssize_t
1434 sched_autogroup_write(struct file *file, const char __user *buf,
1435             size_t count, loff_t *offset)
1436 {
1437         struct inode *inode = file_inode(file);
1438         struct task_struct *p;
1439         char buffer[PROC_NUMBUF];
1440         int nice;
1441         int err;
1442
1443         memset(buffer, 0, sizeof(buffer));
1444         if (count > sizeof(buffer) - 1)
1445                 count = sizeof(buffer) - 1;
1446         if (copy_from_user(buffer, buf, count))
1447                 return -EFAULT;
1448
1449         err = kstrtoint(strstrip(buffer), 0, &nice);
1450         if (err < 0)
1451                 return err;
1452
1453         p = get_proc_task(inode);
1454         if (!p)
1455                 return -ESRCH;
1456
1457         err = proc_sched_autogroup_set_nice(p, nice);
1458         if (err)
1459                 count = err;
1460
1461         put_task_struct(p);
1462
1463         return count;
1464 }
1465
1466 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1467 {
1468         int ret;
1469
1470         ret = single_open(filp, sched_autogroup_show, NULL);
1471         if (!ret) {
1472                 struct seq_file *m = filp->private_data;
1473
1474                 m->private = inode;
1475         }
1476         return ret;
1477 }
1478
1479 static const struct file_operations proc_pid_sched_autogroup_operations = {
1480         .open           = sched_autogroup_open,
1481         .read           = seq_read,
1482         .write          = sched_autogroup_write,
1483         .llseek         = seq_lseek,
1484         .release        = single_release,
1485 };
1486
1487 #endif /* CONFIG_SCHED_AUTOGROUP */
1488
1489 static ssize_t comm_write(struct file *file, const char __user *buf,
1490                                 size_t count, loff_t *offset)
1491 {
1492         struct inode *inode = file_inode(file);
1493         struct task_struct *p;
1494         char buffer[TASK_COMM_LEN];
1495         const size_t maxlen = sizeof(buffer) - 1;
1496
1497         memset(buffer, 0, sizeof(buffer));
1498         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1499                 return -EFAULT;
1500
1501         p = get_proc_task(inode);
1502         if (!p)
1503                 return -ESRCH;
1504
1505         if (same_thread_group(current, p))
1506                 set_task_comm(p, buffer);
1507         else
1508                 count = -EINVAL;
1509
1510         put_task_struct(p);
1511
1512         return count;
1513 }
1514
1515 static int comm_show(struct seq_file *m, void *v)
1516 {
1517         struct inode *inode = m->private;
1518         struct task_struct *p;
1519
1520         p = get_proc_task(inode);
1521         if (!p)
1522                 return -ESRCH;
1523
1524         task_lock(p);
1525         seq_printf(m, "%s\n", p->comm);
1526         task_unlock(p);
1527
1528         put_task_struct(p);
1529
1530         return 0;
1531 }
1532
1533 static int comm_open(struct inode *inode, struct file *filp)
1534 {
1535         return single_open(filp, comm_show, inode);
1536 }
1537
1538 static const struct file_operations proc_pid_set_comm_operations = {
1539         .open           = comm_open,
1540         .read           = seq_read,
1541         .write          = comm_write,
1542         .llseek         = seq_lseek,
1543         .release        = single_release,
1544 };
1545
1546 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1547 {
1548         struct task_struct *task;
1549         struct mm_struct *mm;
1550         struct file *exe_file;
1551
1552         task = get_proc_task(d_inode(dentry));
1553         if (!task)
1554                 return -ENOENT;
1555         mm = get_task_mm(task);
1556         put_task_struct(task);
1557         if (!mm)
1558                 return -ENOENT;
1559         exe_file = get_mm_exe_file(mm);
1560         mmput(mm);
1561         if (exe_file) {
1562                 *exe_path = exe_file->f_path;
1563                 path_get(&exe_file->f_path);
1564                 fput(exe_file);
1565                 return 0;
1566         } else
1567                 return -ENOENT;
1568 }
1569
1570 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1571 {
1572         struct inode *inode = d_inode(dentry);
1573         struct path path;
1574         int error = -EACCES;
1575
1576         /* Are we allowed to snoop on the tasks file descriptors? */
1577         if (!proc_fd_access_allowed(inode))
1578                 goto out;
1579
1580         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1581         if (error)
1582                 goto out;
1583
1584         nd_jump_link(&path);
1585         return NULL;
1586 out:
1587         return ERR_PTR(error);
1588 }
1589
1590 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1591 {
1592         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1593         char *pathname;
1594         int len;
1595
1596         if (!tmp)
1597                 return -ENOMEM;
1598
1599         pathname = d_path(path, tmp, PAGE_SIZE);
1600         len = PTR_ERR(pathname);
1601         if (IS_ERR(pathname))
1602                 goto out;
1603         len = tmp + PAGE_SIZE - 1 - pathname;
1604
1605         if (len > buflen)
1606                 len = buflen;
1607         if (copy_to_user(buffer, pathname, len))
1608                 len = -EFAULT;
1609  out:
1610         free_page((unsigned long)tmp);
1611         return len;
1612 }
1613
1614 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1615 {
1616         int error = -EACCES;
1617         struct inode *inode = d_inode(dentry);
1618         struct path path;
1619
1620         /* Are we allowed to snoop on the tasks file descriptors? */
1621         if (!proc_fd_access_allowed(inode))
1622                 goto out;
1623
1624         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1625         if (error)
1626                 goto out;
1627
1628         error = do_proc_readlink(&path, buffer, buflen);
1629         path_put(&path);
1630 out:
1631         return error;
1632 }
1633
1634 const struct inode_operations proc_pid_link_inode_operations = {
1635         .readlink       = proc_pid_readlink,
1636         .follow_link    = proc_pid_follow_link,
1637         .setattr        = proc_setattr,
1638 };
1639
1640
1641 /* building an inode */
1642
1643 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1644 {
1645         struct inode * inode;
1646         struct proc_inode *ei;
1647         const struct cred *cred;
1648
1649         /* We need a new inode */
1650
1651         inode = new_inode(sb);
1652         if (!inode)
1653                 goto out;
1654
1655         /* Common stuff */
1656         ei = PROC_I(inode);
1657         inode->i_ino = get_next_ino();
1658         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1659         inode->i_op = &proc_def_inode_operations;
1660
1661         /*
1662          * grab the reference to task.
1663          */
1664         ei->pid = get_task_pid(task, PIDTYPE_PID);
1665         if (!ei->pid)
1666                 goto out_unlock;
1667
1668         if (task_dumpable(task)) {
1669                 rcu_read_lock();
1670                 cred = __task_cred(task);
1671                 inode->i_uid = cred->euid;
1672                 inode->i_gid = cred->egid;
1673                 rcu_read_unlock();
1674         }
1675         security_task_to_inode(task, inode);
1676
1677 out:
1678         return inode;
1679
1680 out_unlock:
1681         iput(inode);
1682         return NULL;
1683 }
1684
1685 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1686 {
1687         struct inode *inode = d_inode(dentry);
1688         struct task_struct *task;
1689         const struct cred *cred;
1690         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1691
1692         generic_fillattr(inode, stat);
1693
1694         rcu_read_lock();
1695         stat->uid = GLOBAL_ROOT_UID;
1696         stat->gid = GLOBAL_ROOT_GID;
1697         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1698         if (task) {
1699                 if (!has_pid_permissions(pid, task, 2)) {
1700                         rcu_read_unlock();
1701                         /*
1702                          * This doesn't prevent learning whether PID exists,
1703                          * it only makes getattr() consistent with readdir().
1704                          */
1705                         return -ENOENT;
1706                 }
1707                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1708                     task_dumpable(task)) {
1709                         cred = __task_cred(task);
1710                         stat->uid = cred->euid;
1711                         stat->gid = cred->egid;
1712                 }
1713         }
1714         rcu_read_unlock();
1715         return 0;
1716 }
1717
1718 /* dentry stuff */
1719
1720 /*
1721  *      Exceptional case: normally we are not allowed to unhash a busy
1722  * directory. In this case, however, we can do it - no aliasing problems
1723  * due to the way we treat inodes.
1724  *
1725  * Rewrite the inode's ownerships here because the owning task may have
1726  * performed a setuid(), etc.
1727  *
1728  * Before the /proc/pid/status file was created the only way to read
1729  * the effective uid of a /process was to stat /proc/pid.  Reading
1730  * /proc/pid/status is slow enough that procps and other packages
1731  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1732  * made this apply to all per process world readable and executable
1733  * directories.
1734  */
1735 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1736 {
1737         struct inode *inode;
1738         struct task_struct *task;
1739         const struct cred *cred;
1740
1741         if (flags & LOOKUP_RCU)
1742                 return -ECHILD;
1743
1744         inode = d_inode(dentry);
1745         task = get_proc_task(inode);
1746
1747         if (task) {
1748                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1749                     task_dumpable(task)) {
1750                         rcu_read_lock();
1751                         cred = __task_cred(task);
1752                         inode->i_uid = cred->euid;
1753                         inode->i_gid = cred->egid;
1754                         rcu_read_unlock();
1755                 } else {
1756                         inode->i_uid = GLOBAL_ROOT_UID;
1757                         inode->i_gid = GLOBAL_ROOT_GID;
1758                 }
1759                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1760                 security_task_to_inode(task, inode);
1761                 put_task_struct(task);
1762                 return 1;
1763         }
1764         return 0;
1765 }
1766
1767 static inline bool proc_inode_is_dead(struct inode *inode)
1768 {
1769         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1770 }
1771
1772 int pid_delete_dentry(const struct dentry *dentry)
1773 {
1774         /* Is the task we represent dead?
1775          * If so, then don't put the dentry on the lru list,
1776          * kill it immediately.
1777          */
1778         return proc_inode_is_dead(d_inode(dentry));
1779 }
1780
1781 const struct dentry_operations pid_dentry_operations =
1782 {
1783         .d_revalidate   = pid_revalidate,
1784         .d_delete       = pid_delete_dentry,
1785 };
1786
1787 /* Lookups */
1788
1789 /*
1790  * Fill a directory entry.
1791  *
1792  * If possible create the dcache entry and derive our inode number and
1793  * file type from dcache entry.
1794  *
1795  * Since all of the proc inode numbers are dynamically generated, the inode
1796  * numbers do not exist until the inode is cache.  This means creating the
1797  * the dcache entry in readdir is necessary to keep the inode numbers
1798  * reported by readdir in sync with the inode numbers reported
1799  * by stat.
1800  */
1801 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1802         const char *name, int len,
1803         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1804 {
1805         struct dentry *child, *dir = file->f_path.dentry;
1806         struct qstr qname = QSTR_INIT(name, len);
1807         struct inode *inode;
1808         unsigned type;
1809         ino_t ino;
1810
1811         child = d_hash_and_lookup(dir, &qname);
1812         if (!child) {
1813                 child = d_alloc(dir, &qname);
1814                 if (!child)
1815                         goto end_instantiate;
1816                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1817                         dput(child);
1818                         goto end_instantiate;
1819                 }
1820         }
1821         inode = d_inode(child);
1822         ino = inode->i_ino;
1823         type = inode->i_mode >> 12;
1824         dput(child);
1825         return dir_emit(ctx, name, len, ino, type);
1826
1827 end_instantiate:
1828         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1829 }
1830
1831 #ifdef CONFIG_CHECKPOINT_RESTORE
1832
1833 /*
1834  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1835  * which represent vma start and end addresses.
1836  */
1837 static int dname_to_vma_addr(struct dentry *dentry,
1838                              unsigned long *start, unsigned long *end)
1839 {
1840         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1841                 return -EINVAL;
1842
1843         return 0;
1844 }
1845
1846 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1847 {
1848         unsigned long vm_start, vm_end;
1849         bool exact_vma_exists = false;
1850         struct mm_struct *mm = NULL;
1851         struct task_struct *task;
1852         const struct cred *cred;
1853         struct inode *inode;
1854         int status = 0;
1855
1856         if (flags & LOOKUP_RCU)
1857                 return -ECHILD;
1858
1859         if (!capable(CAP_SYS_ADMIN)) {
1860                 status = -EPERM;
1861                 goto out_notask;
1862         }
1863
1864         inode = d_inode(dentry);
1865         task = get_proc_task(inode);
1866         if (!task)
1867                 goto out_notask;
1868
1869         mm = mm_access(task, PTRACE_MODE_READ);
1870         if (IS_ERR_OR_NULL(mm))
1871                 goto out;
1872
1873         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1874                 down_read(&mm->mmap_sem);
1875                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1876                 up_read(&mm->mmap_sem);
1877         }
1878
1879         mmput(mm);
1880
1881         if (exact_vma_exists) {
1882                 if (task_dumpable(task)) {
1883                         rcu_read_lock();
1884                         cred = __task_cred(task);
1885                         inode->i_uid = cred->euid;
1886                         inode->i_gid = cred->egid;
1887                         rcu_read_unlock();
1888                 } else {
1889                         inode->i_uid = GLOBAL_ROOT_UID;
1890                         inode->i_gid = GLOBAL_ROOT_GID;
1891                 }
1892                 security_task_to_inode(task, inode);
1893                 status = 1;
1894         }
1895
1896 out:
1897         put_task_struct(task);
1898
1899 out_notask:
1900         return status;
1901 }
1902
1903 static const struct dentry_operations tid_map_files_dentry_operations = {
1904         .d_revalidate   = map_files_d_revalidate,
1905         .d_delete       = pid_delete_dentry,
1906 };
1907
1908 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1909 {
1910         unsigned long vm_start, vm_end;
1911         struct vm_area_struct *vma;
1912         struct task_struct *task;
1913         struct mm_struct *mm;
1914         int rc;
1915
1916         rc = -ENOENT;
1917         task = get_proc_task(d_inode(dentry));
1918         if (!task)
1919                 goto out;
1920
1921         mm = get_task_mm(task);
1922         put_task_struct(task);
1923         if (!mm)
1924                 goto out;
1925
1926         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1927         if (rc)
1928                 goto out_mmput;
1929
1930         rc = -ENOENT;
1931         down_read(&mm->mmap_sem);
1932         vma = find_exact_vma(mm, vm_start, vm_end);
1933         if (vma && vma->vm_file) {
1934                 *path = vma->vm_file->f_path;
1935                 path_get(path);
1936                 rc = 0;
1937         }
1938         up_read(&mm->mmap_sem);
1939
1940 out_mmput:
1941         mmput(mm);
1942 out:
1943         return rc;
1944 }
1945
1946 struct map_files_info {
1947         fmode_t         mode;
1948         unsigned long   len;
1949         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1950 };
1951
1952 static int
1953 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1954                            struct task_struct *task, const void *ptr)
1955 {
1956         fmode_t mode = (fmode_t)(unsigned long)ptr;
1957         struct proc_inode *ei;
1958         struct inode *inode;
1959
1960         inode = proc_pid_make_inode(dir->i_sb, task);
1961         if (!inode)
1962                 return -ENOENT;
1963
1964         ei = PROC_I(inode);
1965         ei->op.proc_get_link = proc_map_files_get_link;
1966
1967         inode->i_op = &proc_pid_link_inode_operations;
1968         inode->i_size = 64;
1969         inode->i_mode = S_IFLNK;
1970
1971         if (mode & FMODE_READ)
1972                 inode->i_mode |= S_IRUSR;
1973         if (mode & FMODE_WRITE)
1974                 inode->i_mode |= S_IWUSR;
1975
1976         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1977         d_add(dentry, inode);
1978
1979         return 0;
1980 }
1981
1982 static struct dentry *proc_map_files_lookup(struct inode *dir,
1983                 struct dentry *dentry, unsigned int flags)
1984 {
1985         unsigned long vm_start, vm_end;
1986         struct vm_area_struct *vma;
1987         struct task_struct *task;
1988         int result;
1989         struct mm_struct *mm;
1990
1991         result = -EPERM;
1992         if (!capable(CAP_SYS_ADMIN))
1993                 goto out;
1994
1995         result = -ENOENT;
1996         task = get_proc_task(dir);
1997         if (!task)
1998                 goto out;
1999
2000         result = -EACCES;
2001         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2002                 goto out_put_task;
2003
2004         result = -ENOENT;
2005         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2006                 goto out_put_task;
2007
2008         mm = get_task_mm(task);
2009         if (!mm)
2010                 goto out_put_task;
2011
2012         down_read(&mm->mmap_sem);
2013         vma = find_exact_vma(mm, vm_start, vm_end);
2014         if (!vma)
2015                 goto out_no_vma;
2016
2017         if (vma->vm_file)
2018                 result = proc_map_files_instantiate(dir, dentry, task,
2019                                 (void *)(unsigned long)vma->vm_file->f_mode);
2020
2021 out_no_vma:
2022         up_read(&mm->mmap_sem);
2023         mmput(mm);
2024 out_put_task:
2025         put_task_struct(task);
2026 out:
2027         return ERR_PTR(result);
2028 }
2029
2030 static const struct inode_operations proc_map_files_inode_operations = {
2031         .lookup         = proc_map_files_lookup,
2032         .permission     = proc_fd_permission,
2033         .setattr        = proc_setattr,
2034 };
2035
2036 static int
2037 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2038 {
2039         struct vm_area_struct *vma;
2040         struct task_struct *task;
2041         struct mm_struct *mm;
2042         unsigned long nr_files, pos, i;
2043         struct flex_array *fa = NULL;
2044         struct map_files_info info;
2045         struct map_files_info *p;
2046         int ret;
2047
2048         ret = -EPERM;
2049         if (!capable(CAP_SYS_ADMIN))
2050                 goto out;
2051
2052         ret = -ENOENT;
2053         task = get_proc_task(file_inode(file));
2054         if (!task)
2055                 goto out;
2056
2057         ret = -EACCES;
2058         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2059                 goto out_put_task;
2060
2061         ret = 0;
2062         if (!dir_emit_dots(file, ctx))
2063                 goto out_put_task;
2064
2065         mm = get_task_mm(task);
2066         if (!mm)
2067                 goto out_put_task;
2068         down_read(&mm->mmap_sem);
2069
2070         nr_files = 0;
2071
2072         /*
2073          * We need two passes here:
2074          *
2075          *  1) Collect vmas of mapped files with mmap_sem taken
2076          *  2) Release mmap_sem and instantiate entries
2077          *
2078          * otherwise we get lockdep complained, since filldir()
2079          * routine might require mmap_sem taken in might_fault().
2080          */
2081
2082         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2083                 if (vma->vm_file && ++pos > ctx->pos)
2084                         nr_files++;
2085         }
2086
2087         if (nr_files) {
2088                 fa = flex_array_alloc(sizeof(info), nr_files,
2089                                         GFP_KERNEL);
2090                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2091                                                 GFP_KERNEL)) {
2092                         ret = -ENOMEM;
2093                         if (fa)
2094                                 flex_array_free(fa);
2095                         up_read(&mm->mmap_sem);
2096                         mmput(mm);
2097                         goto out_put_task;
2098                 }
2099                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2100                                 vma = vma->vm_next) {
2101                         if (!vma->vm_file)
2102                                 continue;
2103                         if (++pos <= ctx->pos)
2104                                 continue;
2105
2106                         info.mode = vma->vm_file->f_mode;
2107                         info.len = snprintf(info.name,
2108                                         sizeof(info.name), "%lx-%lx",
2109                                         vma->vm_start, vma->vm_end);
2110                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2111                                 BUG();
2112                 }
2113         }
2114         up_read(&mm->mmap_sem);
2115
2116         for (i = 0; i < nr_files; i++) {
2117                 p = flex_array_get(fa, i);
2118                 if (!proc_fill_cache(file, ctx,
2119                                       p->name, p->len,
2120                                       proc_map_files_instantiate,
2121                                       task,
2122                                       (void *)(unsigned long)p->mode))
2123                         break;
2124                 ctx->pos++;
2125         }
2126         if (fa)
2127                 flex_array_free(fa);
2128         mmput(mm);
2129
2130 out_put_task:
2131         put_task_struct(task);
2132 out:
2133         return ret;
2134 }
2135
2136 static const struct file_operations proc_map_files_operations = {
2137         .read           = generic_read_dir,
2138         .iterate        = proc_map_files_readdir,
2139         .llseek         = default_llseek,
2140 };
2141
2142 struct timers_private {
2143         struct pid *pid;
2144         struct task_struct *task;
2145         struct sighand_struct *sighand;
2146         struct pid_namespace *ns;
2147         unsigned long flags;
2148 };
2149
2150 static void *timers_start(struct seq_file *m, loff_t *pos)
2151 {
2152         struct timers_private *tp = m->private;
2153
2154         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2155         if (!tp->task)
2156                 return ERR_PTR(-ESRCH);
2157
2158         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2159         if (!tp->sighand)
2160                 return ERR_PTR(-ESRCH);
2161
2162         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2163 }
2164
2165 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2166 {
2167         struct timers_private *tp = m->private;
2168         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2169 }
2170
2171 static void timers_stop(struct seq_file *m, void *v)
2172 {
2173         struct timers_private *tp = m->private;
2174
2175         if (tp->sighand) {
2176                 unlock_task_sighand(tp->task, &tp->flags);
2177                 tp->sighand = NULL;
2178         }
2179
2180         if (tp->task) {
2181                 put_task_struct(tp->task);
2182                 tp->task = NULL;
2183         }
2184 }
2185
2186 static int show_timer(struct seq_file *m, void *v)
2187 {
2188         struct k_itimer *timer;
2189         struct timers_private *tp = m->private;
2190         int notify;
2191         static const char * const nstr[] = {
2192                 [SIGEV_SIGNAL] = "signal",
2193                 [SIGEV_NONE] = "none",
2194                 [SIGEV_THREAD] = "thread",
2195         };
2196
2197         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2198         notify = timer->it_sigev_notify;
2199
2200         seq_printf(m, "ID: %d\n", timer->it_id);
2201         seq_printf(m, "signal: %d/%p\n",
2202                    timer->sigq->info.si_signo,
2203                    timer->sigq->info.si_value.sival_ptr);
2204         seq_printf(m, "notify: %s/%s.%d\n",
2205                    nstr[notify & ~SIGEV_THREAD_ID],
2206                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2207                    pid_nr_ns(timer->it_pid, tp->ns));
2208         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2209
2210         return 0;
2211 }
2212
2213 static const struct seq_operations proc_timers_seq_ops = {
2214         .start  = timers_start,
2215         .next   = timers_next,
2216         .stop   = timers_stop,
2217         .show   = show_timer,
2218 };
2219
2220 static int proc_timers_open(struct inode *inode, struct file *file)
2221 {
2222         struct timers_private *tp;
2223
2224         tp = __seq_open_private(file, &proc_timers_seq_ops,
2225                         sizeof(struct timers_private));
2226         if (!tp)
2227                 return -ENOMEM;
2228
2229         tp->pid = proc_pid(inode);
2230         tp->ns = inode->i_sb->s_fs_info;
2231         return 0;
2232 }
2233
2234 static const struct file_operations proc_timers_operations = {
2235         .open           = proc_timers_open,
2236         .read           = seq_read,
2237         .llseek         = seq_lseek,
2238         .release        = seq_release_private,
2239 };
2240 #endif /* CONFIG_CHECKPOINT_RESTORE */
2241
2242 static int proc_pident_instantiate(struct inode *dir,
2243         struct dentry *dentry, struct task_struct *task, const void *ptr)
2244 {
2245         const struct pid_entry *p = ptr;
2246         struct inode *inode;
2247         struct proc_inode *ei;
2248
2249         inode = proc_pid_make_inode(dir->i_sb, task);
2250         if (!inode)
2251                 goto out;
2252
2253         ei = PROC_I(inode);
2254         inode->i_mode = p->mode;
2255         if (S_ISDIR(inode->i_mode))
2256                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2257         if (p->iop)
2258                 inode->i_op = p->iop;
2259         if (p->fop)
2260                 inode->i_fop = p->fop;
2261         ei->op = p->op;
2262         d_set_d_op(dentry, &pid_dentry_operations);
2263         d_add(dentry, inode);
2264         /* Close the race of the process dying before we return the dentry */
2265         if (pid_revalidate(dentry, 0))
2266                 return 0;
2267 out:
2268         return -ENOENT;
2269 }
2270
2271 static struct dentry *proc_pident_lookup(struct inode *dir,
2272                                          struct dentry *dentry,
2273                                          const struct pid_entry *ents,
2274                                          unsigned int nents)
2275 {
2276         int error;
2277         struct task_struct *task = get_proc_task(dir);
2278         const struct pid_entry *p, *last;
2279
2280         error = -ENOENT;
2281
2282         if (!task)
2283                 goto out_no_task;
2284
2285         /*
2286          * Yes, it does not scale. And it should not. Don't add
2287          * new entries into /proc/<tgid>/ without very good reasons.
2288          */
2289         last = &ents[nents - 1];
2290         for (p = ents; p <= last; p++) {
2291                 if (p->len != dentry->d_name.len)
2292                         continue;
2293                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2294                         break;
2295         }
2296         if (p > last)
2297                 goto out;
2298
2299         error = proc_pident_instantiate(dir, dentry, task, p);
2300 out:
2301         put_task_struct(task);
2302 out_no_task:
2303         return ERR_PTR(error);
2304 }
2305
2306 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2307                 const struct pid_entry *ents, unsigned int nents)
2308 {
2309         struct task_struct *task = get_proc_task(file_inode(file));
2310         const struct pid_entry *p;
2311
2312         if (!task)
2313                 return -ENOENT;
2314
2315         if (!dir_emit_dots(file, ctx))
2316                 goto out;
2317
2318         if (ctx->pos >= nents + 2)
2319                 goto out;
2320
2321         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2322                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2323                                 proc_pident_instantiate, task, p))
2324                         break;
2325                 ctx->pos++;
2326         }
2327 out:
2328         put_task_struct(task);
2329         return 0;
2330 }
2331
2332 #ifdef CONFIG_SECURITY
2333 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2334                                   size_t count, loff_t *ppos)
2335 {
2336         struct inode * inode = file_inode(file);
2337         char *p = NULL;
2338         ssize_t length;
2339         struct task_struct *task = get_proc_task(inode);
2340
2341         if (!task)
2342                 return -ESRCH;
2343
2344         length = security_getprocattr(task,
2345                                       (char*)file->f_path.dentry->d_name.name,
2346                                       &p);
2347         put_task_struct(task);
2348         if (length > 0)
2349                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2350         kfree(p);
2351         return length;
2352 }
2353
2354 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2355                                    size_t count, loff_t *ppos)
2356 {
2357         struct inode * inode = file_inode(file);
2358         char *page;
2359         ssize_t length;
2360         struct task_struct *task = get_proc_task(inode);
2361
2362         length = -ESRCH;
2363         if (!task)
2364                 goto out_no_task;
2365         if (count > PAGE_SIZE)
2366                 count = PAGE_SIZE;
2367
2368         /* No partial writes. */
2369         length = -EINVAL;
2370         if (*ppos != 0)
2371                 goto out;
2372
2373         length = -ENOMEM;
2374         page = (char*)__get_free_page(GFP_TEMPORARY);
2375         if (!page)
2376                 goto out;
2377
2378         length = -EFAULT;
2379         if (copy_from_user(page, buf, count))
2380                 goto out_free;
2381
2382         /* Guard against adverse ptrace interaction */
2383         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2384         if (length < 0)
2385                 goto out_free;
2386
2387         length = security_setprocattr(task,
2388                                       (char*)file->f_path.dentry->d_name.name,
2389                                       (void*)page, count);
2390         mutex_unlock(&task->signal->cred_guard_mutex);
2391 out_free:
2392         free_page((unsigned long) page);
2393 out:
2394         put_task_struct(task);
2395 out_no_task:
2396         return length;
2397 }
2398
2399 static const struct file_operations proc_pid_attr_operations = {
2400         .read           = proc_pid_attr_read,
2401         .write          = proc_pid_attr_write,
2402         .llseek         = generic_file_llseek,
2403 };
2404
2405 static const struct pid_entry attr_dir_stuff[] = {
2406         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2407         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2408         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2409         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2410         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2411         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2412 };
2413
2414 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2415 {
2416         return proc_pident_readdir(file, ctx,
2417                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2418 }
2419
2420 static const struct file_operations proc_attr_dir_operations = {
2421         .read           = generic_read_dir,
2422         .iterate        = proc_attr_dir_readdir,
2423         .llseek         = default_llseek,
2424 };
2425
2426 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2427                                 struct dentry *dentry, unsigned int flags)
2428 {
2429         return proc_pident_lookup(dir, dentry,
2430                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2431 }
2432
2433 static const struct inode_operations proc_attr_dir_inode_operations = {
2434         .lookup         = proc_attr_dir_lookup,
2435         .getattr        = pid_getattr,
2436         .setattr        = proc_setattr,
2437 };
2438
2439 #endif
2440
2441 #ifdef CONFIG_ELF_CORE
2442 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2443                                          size_t count, loff_t *ppos)
2444 {
2445         struct task_struct *task = get_proc_task(file_inode(file));
2446         struct mm_struct *mm;
2447         char buffer[PROC_NUMBUF];
2448         size_t len;
2449         int ret;
2450
2451         if (!task)
2452                 return -ESRCH;
2453
2454         ret = 0;
2455         mm = get_task_mm(task);
2456         if (mm) {
2457                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2458                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2459                                 MMF_DUMP_FILTER_SHIFT));
2460                 mmput(mm);
2461                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2462         }
2463
2464         put_task_struct(task);
2465
2466         return ret;
2467 }
2468
2469 static ssize_t proc_coredump_filter_write(struct file *file,
2470                                           const char __user *buf,
2471                                           size_t count,
2472                                           loff_t *ppos)
2473 {
2474         struct task_struct *task;
2475         struct mm_struct *mm;
2476         char buffer[PROC_NUMBUF], *end;
2477         unsigned int val;
2478         int ret;
2479         int i;
2480         unsigned long mask;
2481
2482         ret = -EFAULT;
2483         memset(buffer, 0, sizeof(buffer));
2484         if (count > sizeof(buffer) - 1)
2485                 count = sizeof(buffer) - 1;
2486         if (copy_from_user(buffer, buf, count))
2487                 goto out_no_task;
2488
2489         ret = -EINVAL;
2490         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2491         if (*end == '\n')
2492                 end++;
2493         if (end - buffer == 0)
2494                 goto out_no_task;
2495
2496         ret = -ESRCH;
2497         task = get_proc_task(file_inode(file));
2498         if (!task)
2499                 goto out_no_task;
2500
2501         ret = end - buffer;
2502         mm = get_task_mm(task);
2503         if (!mm)
2504                 goto out_no_mm;
2505
2506         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2507                 if (val & mask)
2508                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2509                 else
2510                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2511         }
2512
2513         mmput(mm);
2514  out_no_mm:
2515         put_task_struct(task);
2516  out_no_task:
2517         return ret;
2518 }
2519
2520 static const struct file_operations proc_coredump_filter_operations = {
2521         .read           = proc_coredump_filter_read,
2522         .write          = proc_coredump_filter_write,
2523         .llseek         = generic_file_llseek,
2524 };
2525 #endif
2526
2527 #ifdef CONFIG_TASK_IO_ACCOUNTING
2528 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2529 {
2530         struct task_io_accounting acct = task->ioac;
2531         unsigned long flags;
2532         int result;
2533
2534         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2535         if (result)
2536                 return result;
2537
2538         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2539                 result = -EACCES;
2540                 goto out_unlock;
2541         }
2542
2543         if (whole && lock_task_sighand(task, &flags)) {
2544                 struct task_struct *t = task;
2545
2546                 task_io_accounting_add(&acct, &task->signal->ioac);
2547                 while_each_thread(task, t)
2548                         task_io_accounting_add(&acct, &t->ioac);
2549
2550                 unlock_task_sighand(task, &flags);
2551         }
2552         seq_printf(m,
2553                    "rchar: %llu\n"
2554                    "wchar: %llu\n"
2555                    "syscr: %llu\n"
2556                    "syscw: %llu\n"
2557                    "read_bytes: %llu\n"
2558                    "write_bytes: %llu\n"
2559                    "cancelled_write_bytes: %llu\n",
2560                    (unsigned long long)acct.rchar,
2561                    (unsigned long long)acct.wchar,
2562                    (unsigned long long)acct.syscr,
2563                    (unsigned long long)acct.syscw,
2564                    (unsigned long long)acct.read_bytes,
2565                    (unsigned long long)acct.write_bytes,
2566                    (unsigned long long)acct.cancelled_write_bytes);
2567         result = 0;
2568
2569 out_unlock:
2570         mutex_unlock(&task->signal->cred_guard_mutex);
2571         return result;
2572 }
2573
2574 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2575                                   struct pid *pid, struct task_struct *task)
2576 {
2577         return do_io_accounting(task, m, 0);
2578 }
2579
2580 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2581                                    struct pid *pid, struct task_struct *task)
2582 {
2583         return do_io_accounting(task, m, 1);
2584 }
2585 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2586
2587 #ifdef CONFIG_USER_NS
2588 static int proc_id_map_open(struct inode *inode, struct file *file,
2589         const struct seq_operations *seq_ops)
2590 {
2591         struct user_namespace *ns = NULL;
2592         struct task_struct *task;
2593         struct seq_file *seq;
2594         int ret = -EINVAL;
2595
2596         task = get_proc_task(inode);
2597         if (task) {
2598                 rcu_read_lock();
2599                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2600                 rcu_read_unlock();
2601                 put_task_struct(task);
2602         }
2603         if (!ns)
2604                 goto err;
2605
2606         ret = seq_open(file, seq_ops);
2607         if (ret)
2608                 goto err_put_ns;
2609
2610         seq = file->private_data;
2611         seq->private = ns;
2612
2613         return 0;
2614 err_put_ns:
2615         put_user_ns(ns);
2616 err:
2617         return ret;
2618 }
2619
2620 static int proc_id_map_release(struct inode *inode, struct file *file)
2621 {
2622         struct seq_file *seq = file->private_data;
2623         struct user_namespace *ns = seq->private;
2624         put_user_ns(ns);
2625         return seq_release(inode, file);
2626 }
2627
2628 static int proc_uid_map_open(struct inode *inode, struct file *file)
2629 {
2630         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2631 }
2632
2633 static int proc_gid_map_open(struct inode *inode, struct file *file)
2634 {
2635         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2636 }
2637
2638 static int proc_projid_map_open(struct inode *inode, struct file *file)
2639 {
2640         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2641 }
2642
2643 static const struct file_operations proc_uid_map_operations = {
2644         .open           = proc_uid_map_open,
2645         .write          = proc_uid_map_write,
2646         .read           = seq_read,
2647         .llseek         = seq_lseek,
2648         .release        = proc_id_map_release,
2649 };
2650
2651 static const struct file_operations proc_gid_map_operations = {
2652         .open           = proc_gid_map_open,
2653         .write          = proc_gid_map_write,
2654         .read           = seq_read,
2655         .llseek         = seq_lseek,
2656         .release        = proc_id_map_release,
2657 };
2658
2659 static const struct file_operations proc_projid_map_operations = {
2660         .open           = proc_projid_map_open,
2661         .write          = proc_projid_map_write,
2662         .read           = seq_read,
2663         .llseek         = seq_lseek,
2664         .release        = proc_id_map_release,
2665 };
2666
2667 static int proc_setgroups_open(struct inode *inode, struct file *file)
2668 {
2669         struct user_namespace *ns = NULL;
2670         struct task_struct *task;
2671         int ret;
2672
2673         ret = -ESRCH;
2674         task = get_proc_task(inode);
2675         if (task) {
2676                 rcu_read_lock();
2677                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2678                 rcu_read_unlock();
2679                 put_task_struct(task);
2680         }
2681         if (!ns)
2682                 goto err;
2683
2684         if (file->f_mode & FMODE_WRITE) {
2685                 ret = -EACCES;
2686                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2687                         goto err_put_ns;
2688         }
2689
2690         ret = single_open(file, &proc_setgroups_show, ns);
2691         if (ret)
2692                 goto err_put_ns;
2693
2694         return 0;
2695 err_put_ns:
2696         put_user_ns(ns);
2697 err:
2698         return ret;
2699 }
2700
2701 static int proc_setgroups_release(struct inode *inode, struct file *file)
2702 {
2703         struct seq_file *seq = file->private_data;
2704         struct user_namespace *ns = seq->private;
2705         int ret = single_release(inode, file);
2706         put_user_ns(ns);
2707         return ret;
2708 }
2709
2710 static const struct file_operations proc_setgroups_operations = {
2711         .open           = proc_setgroups_open,
2712         .write          = proc_setgroups_write,
2713         .read           = seq_read,
2714         .llseek         = seq_lseek,
2715         .release        = proc_setgroups_release,
2716 };
2717 #endif /* CONFIG_USER_NS */
2718
2719 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2720                                 struct pid *pid, struct task_struct *task)
2721 {
2722         int err = lock_trace(task);
2723         if (!err) {
2724                 seq_printf(m, "%08x\n", task->personality);
2725                 unlock_trace(task);
2726         }
2727         return err;
2728 }
2729
2730 /*
2731  * Thread groups
2732  */
2733 static const struct file_operations proc_task_operations;
2734 static const struct inode_operations proc_task_inode_operations;
2735
2736 static const struct pid_entry tgid_base_stuff[] = {
2737         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2738         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2739 #ifdef CONFIG_CHECKPOINT_RESTORE
2740         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2741 #endif
2742         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2743         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2744 #ifdef CONFIG_NET
2745         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2746 #endif
2747         REG("environ",    S_IRUSR, proc_environ_operations),
2748         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2749         ONE("status",     S_IRUGO, proc_pid_status),
2750         ONE("personality", S_IRUSR, proc_pid_personality),
2751         ONE("limits",     S_IRUGO, proc_pid_limits),
2752 #ifdef CONFIG_SCHED_DEBUG
2753         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2754 #endif
2755 #ifdef CONFIG_SCHED_AUTOGROUP
2756         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2757 #endif
2758         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2759 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2760         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2761 #endif
2762         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2763         ONE("stat",       S_IRUGO, proc_tgid_stat),
2764         ONE("statm",      S_IRUGO, proc_pid_statm),
2765         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2766 #ifdef CONFIG_NUMA
2767         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2768 #endif
2769         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2770         LNK("cwd",        proc_cwd_link),
2771         LNK("root",       proc_root_link),
2772         LNK("exe",        proc_exe_link),
2773         REG("mounts",     S_IRUGO, proc_mounts_operations),
2774         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2775         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2776 #ifdef CONFIG_PROC_PAGE_MONITOR
2777         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2778         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2779         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2780 #endif
2781 #ifdef CONFIG_SECURITY
2782         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2783 #endif
2784 #ifdef CONFIG_KALLSYMS
2785         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2786 #endif
2787 #ifdef CONFIG_STACKTRACE
2788         ONE("stack",      S_IRUSR, proc_pid_stack),
2789 #endif
2790 #ifdef CONFIG_SCHEDSTATS
2791         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2792 #endif
2793 #ifdef CONFIG_LATENCYTOP
2794         REG("latency",  S_IRUGO, proc_lstats_operations),
2795 #endif
2796 #ifdef CONFIG_PROC_PID_CPUSET
2797         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2798 #endif
2799 #ifdef CONFIG_CGROUPS
2800         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2801 #endif
2802         ONE("oom_score",  S_IRUGO, proc_oom_score),
2803         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2804         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2805 #ifdef CONFIG_AUDITSYSCALL
2806         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2807         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2808 #endif
2809 #ifdef CONFIG_FAULT_INJECTION
2810         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2811 #endif
2812 #ifdef CONFIG_ELF_CORE
2813         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2814 #endif
2815 #ifdef CONFIG_TASK_IO_ACCOUNTING
2816         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2817 #endif
2818 #ifdef CONFIG_HARDWALL
2819         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2820 #endif
2821 #ifdef CONFIG_USER_NS
2822         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2823         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2824         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2825         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2826 #endif
2827 #ifdef CONFIG_CHECKPOINT_RESTORE
2828         REG("timers",     S_IRUGO, proc_timers_operations),
2829 #endif
2830 };
2831
2832 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2833 {
2834         return proc_pident_readdir(file, ctx,
2835                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2836 }
2837
2838 static const struct file_operations proc_tgid_base_operations = {
2839         .read           = generic_read_dir,
2840         .iterate        = proc_tgid_base_readdir,
2841         .llseek         = default_llseek,
2842 };
2843
2844 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2845 {
2846         return proc_pident_lookup(dir, dentry,
2847                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2848 }
2849
2850 static const struct inode_operations proc_tgid_base_inode_operations = {
2851         .lookup         = proc_tgid_base_lookup,
2852         .getattr        = pid_getattr,
2853         .setattr        = proc_setattr,
2854         .permission     = proc_pid_permission,
2855 };
2856
2857 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2858 {
2859         struct dentry *dentry, *leader, *dir;
2860         char buf[PROC_NUMBUF];
2861         struct qstr name;
2862
2863         name.name = buf;
2864         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2865         /* no ->d_hash() rejects on procfs */
2866         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2867         if (dentry) {
2868                 d_invalidate(dentry);
2869                 dput(dentry);
2870         }
2871
2872         if (pid == tgid)
2873                 return;
2874
2875         name.name = buf;
2876         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2877         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2878         if (!leader)
2879                 goto out;
2880
2881         name.name = "task";
2882         name.len = strlen(name.name);
2883         dir = d_hash_and_lookup(leader, &name);
2884         if (!dir)
2885                 goto out_put_leader;
2886
2887         name.name = buf;
2888         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2889         dentry = d_hash_and_lookup(dir, &name);
2890         if (dentry) {
2891                 d_invalidate(dentry);
2892                 dput(dentry);
2893         }
2894
2895         dput(dir);
2896 out_put_leader:
2897         dput(leader);
2898 out:
2899         return;
2900 }
2901
2902 /**
2903  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2904  * @task: task that should be flushed.
2905  *
2906  * When flushing dentries from proc, one needs to flush them from global
2907  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2908  * in. This call is supposed to do all of this job.
2909  *
2910  * Looks in the dcache for
2911  * /proc/@pid
2912  * /proc/@tgid/task/@pid
2913  * if either directory is present flushes it and all of it'ts children
2914  * from the dcache.
2915  *
2916  * It is safe and reasonable to cache /proc entries for a task until
2917  * that task exits.  After that they just clog up the dcache with
2918  * useless entries, possibly causing useful dcache entries to be
2919  * flushed instead.  This routine is proved to flush those useless
2920  * dcache entries at process exit time.
2921  *
2922  * NOTE: This routine is just an optimization so it does not guarantee
2923  *       that no dcache entries will exist at process exit time it
2924  *       just makes it very unlikely that any will persist.
2925  */
2926
2927 void proc_flush_task(struct task_struct *task)
2928 {
2929         int i;
2930         struct pid *pid, *tgid;
2931         struct upid *upid;
2932
2933         pid = task_pid(task);
2934         tgid = task_tgid(task);
2935
2936         for (i = 0; i <= pid->level; i++) {
2937                 upid = &pid->numbers[i];
2938                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2939                                         tgid->numbers[i].nr);
2940         }
2941 }
2942
2943 static int proc_pid_instantiate(struct inode *dir,
2944                                    struct dentry * dentry,
2945                                    struct task_struct *task, const void *ptr)
2946 {
2947         struct inode *inode;
2948
2949         inode = proc_pid_make_inode(dir->i_sb, task);
2950         if (!inode)
2951                 goto out;
2952
2953         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2954         inode->i_op = &proc_tgid_base_inode_operations;
2955         inode->i_fop = &proc_tgid_base_operations;
2956         inode->i_flags|=S_IMMUTABLE;
2957
2958         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2959                                                   ARRAY_SIZE(tgid_base_stuff)));
2960
2961         d_set_d_op(dentry, &pid_dentry_operations);
2962
2963         d_add(dentry, inode);
2964         /* Close the race of the process dying before we return the dentry */
2965         if (pid_revalidate(dentry, 0))
2966                 return 0;
2967 out:
2968         return -ENOENT;
2969 }
2970
2971 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2972 {
2973         int result = -ENOENT;
2974         struct task_struct *task;
2975         unsigned tgid;
2976         struct pid_namespace *ns;
2977
2978         tgid = name_to_int(&dentry->d_name);
2979         if (tgid == ~0U)
2980                 goto out;
2981
2982         ns = dentry->d_sb->s_fs_info;
2983         rcu_read_lock();
2984         task = find_task_by_pid_ns(tgid, ns);
2985         if (task)
2986                 get_task_struct(task);
2987         rcu_read_unlock();
2988         if (!task)
2989                 goto out;
2990
2991         result = proc_pid_instantiate(dir, dentry, task, NULL);
2992         put_task_struct(task);
2993 out:
2994         return ERR_PTR(result);
2995 }
2996
2997 /*
2998  * Find the first task with tgid >= tgid
2999  *
3000  */
3001 struct tgid_iter {
3002         unsigned int tgid;
3003         struct task_struct *task;
3004 };
3005 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3006 {
3007         struct pid *pid;
3008
3009         if (iter.task)
3010                 put_task_struct(iter.task);
3011         rcu_read_lock();
3012 retry:
3013         iter.task = NULL;
3014         pid = find_ge_pid(iter.tgid, ns);
3015         if (pid) {
3016                 iter.tgid = pid_nr_ns(pid, ns);
3017                 iter.task = pid_task(pid, PIDTYPE_PID);
3018                 /* What we to know is if the pid we have find is the
3019                  * pid of a thread_group_leader.  Testing for task
3020                  * being a thread_group_leader is the obvious thing
3021                  * todo but there is a window when it fails, due to
3022                  * the pid transfer logic in de_thread.
3023                  *
3024                  * So we perform the straight forward test of seeing
3025                  * if the pid we have found is the pid of a thread
3026                  * group leader, and don't worry if the task we have
3027                  * found doesn't happen to be a thread group leader.
3028                  * As we don't care in the case of readdir.
3029                  */
3030                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3031                         iter.tgid += 1;
3032                         goto retry;
3033                 }
3034                 get_task_struct(iter.task);
3035         }
3036         rcu_read_unlock();
3037         return iter;
3038 }
3039
3040 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3041
3042 /* for the /proc/ directory itself, after non-process stuff has been done */
3043 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3044 {
3045         struct tgid_iter iter;
3046         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3047         loff_t pos = ctx->pos;
3048
3049         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3050                 return 0;
3051
3052         if (pos == TGID_OFFSET - 2) {
3053                 struct inode *inode = d_inode(ns->proc_self);
3054                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3055                         return 0;
3056                 ctx->pos = pos = pos + 1;
3057         }
3058         if (pos == TGID_OFFSET - 1) {
3059                 struct inode *inode = d_inode(ns->proc_thread_self);
3060                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3061                         return 0;
3062                 ctx->pos = pos = pos + 1;
3063         }
3064         iter.tgid = pos - TGID_OFFSET;
3065         iter.task = NULL;
3066         for (iter = next_tgid(ns, iter);
3067              iter.task;
3068              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3069                 char name[PROC_NUMBUF];
3070                 int len;
3071                 if (!has_pid_permissions(ns, iter.task, 2))
3072                         continue;
3073
3074                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3075                 ctx->pos = iter.tgid + TGID_OFFSET;
3076                 if (!proc_fill_cache(file, ctx, name, len,
3077                                      proc_pid_instantiate, iter.task, NULL)) {
3078                         put_task_struct(iter.task);
3079                         return 0;
3080                 }
3081         }
3082         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3083         return 0;
3084 }
3085
3086 /*
3087  * Tasks
3088  */
3089 static const struct pid_entry tid_base_stuff[] = {
3090         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3091         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3092         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3093 #ifdef CONFIG_NET
3094         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3095 #endif
3096         REG("environ",   S_IRUSR, proc_environ_operations),
3097         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3098         ONE("status",    S_IRUGO, proc_pid_status),
3099         ONE("personality", S_IRUSR, proc_pid_personality),
3100         ONE("limits",    S_IRUGO, proc_pid_limits),
3101 #ifdef CONFIG_SCHED_DEBUG
3102         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3103 #endif
3104         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3105 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3106         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3107 #endif
3108         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3109         ONE("stat",      S_IRUGO, proc_tid_stat),
3110         ONE("statm",     S_IRUGO, proc_pid_statm),
3111         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3112 #ifdef CONFIG_PROC_CHILDREN
3113         REG("children",  S_IRUGO, proc_tid_children_operations),
3114 #endif
3115 #ifdef CONFIG_NUMA
3116         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3117 #endif
3118         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3119         LNK("cwd",       proc_cwd_link),
3120         LNK("root",      proc_root_link),
3121         LNK("exe",       proc_exe_link),
3122         REG("mounts",    S_IRUGO, proc_mounts_operations),
3123         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3124 #ifdef CONFIG_PROC_PAGE_MONITOR
3125         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3126         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3127         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3128 #endif
3129 #ifdef CONFIG_SECURITY
3130         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3131 #endif
3132 #ifdef CONFIG_KALLSYMS
3133         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3134 #endif
3135 #ifdef CONFIG_STACKTRACE
3136         ONE("stack",      S_IRUSR, proc_pid_stack),
3137 #endif
3138 #ifdef CONFIG_SCHEDSTATS
3139         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3140 #endif
3141 #ifdef CONFIG_LATENCYTOP
3142         REG("latency",  S_IRUGO, proc_lstats_operations),
3143 #endif
3144 #ifdef CONFIG_PROC_PID_CPUSET
3145         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3146 #endif
3147 #ifdef CONFIG_CGROUPS
3148         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3149 #endif
3150         ONE("oom_score", S_IRUGO, proc_oom_score),
3151         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3152         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3153 #ifdef CONFIG_AUDITSYSCALL
3154         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3155         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3156 #endif
3157 #ifdef CONFIG_FAULT_INJECTION
3158         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3159 #endif
3160 #ifdef CONFIG_TASK_IO_ACCOUNTING
3161         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3162 #endif
3163 #ifdef CONFIG_HARDWALL
3164         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3165 #endif
3166 #ifdef CONFIG_USER_NS
3167         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3168         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3169         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3170         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3171 #endif
3172 };
3173
3174 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3175 {
3176         return proc_pident_readdir(file, ctx,
3177                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3178 }
3179
3180 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3181 {
3182         return proc_pident_lookup(dir, dentry,
3183                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3184 }
3185
3186 static const struct file_operations proc_tid_base_operations = {
3187         .read           = generic_read_dir,
3188         .iterate        = proc_tid_base_readdir,
3189         .llseek         = default_llseek,
3190 };
3191
3192 static const struct inode_operations proc_tid_base_inode_operations = {
3193         .lookup         = proc_tid_base_lookup,
3194         .getattr        = pid_getattr,
3195         .setattr        = proc_setattr,
3196 };
3197
3198 static int proc_task_instantiate(struct inode *dir,
3199         struct dentry *dentry, struct task_struct *task, const void *ptr)
3200 {
3201         struct inode *inode;
3202         inode = proc_pid_make_inode(dir->i_sb, task);
3203
3204         if (!inode)
3205                 goto out;
3206         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3207         inode->i_op = &proc_tid_base_inode_operations;
3208         inode->i_fop = &proc_tid_base_operations;
3209         inode->i_flags|=S_IMMUTABLE;
3210
3211         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3212                                                   ARRAY_SIZE(tid_base_stuff)));
3213
3214         d_set_d_op(dentry, &pid_dentry_operations);
3215
3216         d_add(dentry, inode);
3217         /* Close the race of the process dying before we return the dentry */
3218         if (pid_revalidate(dentry, 0))
3219                 return 0;
3220 out:
3221         return -ENOENT;
3222 }
3223
3224 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3225 {
3226         int result = -ENOENT;
3227         struct task_struct *task;
3228         struct task_struct *leader = get_proc_task(dir);
3229         unsigned tid;
3230         struct pid_namespace *ns;
3231
3232         if (!leader)
3233                 goto out_no_task;
3234
3235         tid = name_to_int(&dentry->d_name);
3236         if (tid == ~0U)
3237                 goto out;
3238
3239         ns = dentry->d_sb->s_fs_info;
3240         rcu_read_lock();
3241         task = find_task_by_pid_ns(tid, ns);
3242         if (task)
3243                 get_task_struct(task);
3244         rcu_read_unlock();
3245         if (!task)
3246                 goto out;
3247         if (!same_thread_group(leader, task))
3248                 goto out_drop_task;
3249
3250         result = proc_task_instantiate(dir, dentry, task, NULL);
3251 out_drop_task:
3252         put_task_struct(task);
3253 out:
3254         put_task_struct(leader);
3255 out_no_task:
3256         return ERR_PTR(result);
3257 }
3258
3259 /*
3260  * Find the first tid of a thread group to return to user space.
3261  *
3262  * Usually this is just the thread group leader, but if the users
3263  * buffer was too small or there was a seek into the middle of the
3264  * directory we have more work todo.
3265  *
3266  * In the case of a short read we start with find_task_by_pid.
3267  *
3268  * In the case of a seek we start with the leader and walk nr
3269  * threads past it.
3270  */
3271 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3272                                         struct pid_namespace *ns)
3273 {
3274         struct task_struct *pos, *task;
3275         unsigned long nr = f_pos;
3276
3277         if (nr != f_pos)        /* 32bit overflow? */
3278                 return NULL;
3279
3280         rcu_read_lock();
3281         task = pid_task(pid, PIDTYPE_PID);
3282         if (!task)
3283                 goto fail;
3284
3285         /* Attempt to start with the tid of a thread */
3286         if (tid && nr) {
3287                 pos = find_task_by_pid_ns(tid, ns);
3288                 if (pos && same_thread_group(pos, task))
3289                         goto found;
3290         }
3291
3292         /* If nr exceeds the number of threads there is nothing todo */
3293         if (nr >= get_nr_threads(task))
3294                 goto fail;
3295
3296         /* If we haven't found our starting place yet start
3297          * with the leader and walk nr threads forward.
3298          */
3299         pos = task = task->group_leader;
3300         do {
3301                 if (!nr--)
3302                         goto found;
3303         } while_each_thread(task, pos);
3304 fail:
3305         pos = NULL;
3306         goto out;
3307 found:
3308         get_task_struct(pos);
3309 out:
3310         rcu_read_unlock();
3311         return pos;
3312 }
3313
3314 /*
3315  * Find the next thread in the thread list.
3316  * Return NULL if there is an error or no next thread.
3317  *
3318  * The reference to the input task_struct is released.
3319  */
3320 static struct task_struct *next_tid(struct task_struct *start)
3321 {
3322         struct task_struct *pos = NULL;
3323         rcu_read_lock();
3324         if (pid_alive(start)) {
3325                 pos = next_thread(start);
3326                 if (thread_group_leader(pos))
3327                         pos = NULL;
3328                 else
3329                         get_task_struct(pos);
3330         }
3331         rcu_read_unlock();
3332         put_task_struct(start);
3333         return pos;
3334 }
3335
3336 /* for the /proc/TGID/task/ directories */
3337 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3338 {
3339         struct inode *inode = file_inode(file);
3340         struct task_struct *task;
3341         struct pid_namespace *ns;
3342         int tid;
3343
3344         if (proc_inode_is_dead(inode))
3345                 return -ENOENT;
3346
3347         if (!dir_emit_dots(file, ctx))
3348                 return 0;
3349
3350         /* f_version caches the tgid value that the last readdir call couldn't
3351          * return. lseek aka telldir automagically resets f_version to 0.
3352          */
3353         ns = inode->i_sb->s_fs_info;
3354         tid = (int)file->f_version;
3355         file->f_version = 0;
3356         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3357              task;
3358              task = next_tid(task), ctx->pos++) {
3359                 char name[PROC_NUMBUF];
3360                 int len;
3361                 tid = task_pid_nr_ns(task, ns);
3362                 len = snprintf(name, sizeof(name), "%d", tid);
3363                 if (!proc_fill_cache(file, ctx, name, len,
3364                                 proc_task_instantiate, task, NULL)) {
3365                         /* returning this tgid failed, save it as the first
3366                          * pid for the next readir call */
3367                         file->f_version = (u64)tid;
3368                         put_task_struct(task);
3369                         break;
3370                 }
3371         }
3372
3373         return 0;
3374 }
3375
3376 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3377 {
3378         struct inode *inode = d_inode(dentry);
3379         struct task_struct *p = get_proc_task(inode);
3380         generic_fillattr(inode, stat);
3381
3382         if (p) {
3383                 stat->nlink += get_nr_threads(p);
3384                 put_task_struct(p);
3385         }
3386
3387         return 0;
3388 }
3389
3390 static const struct inode_operations proc_task_inode_operations = {
3391         .lookup         = proc_task_lookup,
3392         .getattr        = proc_task_getattr,
3393         .setattr        = proc_setattr,
3394         .permission     = proc_pid_permission,
3395 };
3396
3397 static const struct file_operations proc_task_operations = {
3398         .read           = generic_read_dir,
3399         .iterate        = proc_task_readdir,
3400         .llseek         = default_llseek,
3401 };