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         /* Empty ARGV. */
 247         if (len1 == 0) {
 248                 rv = 0;
 249                 goto out_free_page;
 250         }
 251         /*
 252          * Inherently racy -- command line shares address space
 253          * with code and data.
 254          */
 255         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 256         if (rv <= 0)
 257                 goto out_free_page;
 258
 259         rv = 0;
 260
 261         if (c == '\0') {
 262                 /* Command line (set of strings) occupies whole ARGV. */
 263                 if (len1 <= *pos)
 264                         goto out_free_page;
 265
 266                 p = arg_start + *pos;
 267                 len = len1 - *pos;
 268                 while (count > 0 && len > 0) {
 269                         unsigned int _count;
 270                         int nr_read;
 271
 272                         _count = min3(count, len, PAGE_SIZE);
 273                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 274                         if (nr_read < 0)
 275                                 rv = nr_read;
 276                         if (nr_read <= 0)
 277                                 goto out_free_page;
 278
 279                         if (copy_to_user(buf, page, nr_read)) {
 280                                 rv = -EFAULT;
 281                                 goto out_free_page;
 282                         }
 283
 284                         p       += nr_read;
 285                         len     -= nr_read;
 286                         buf     += nr_read;
 287                         count   -= nr_read;
 288                         rv      += nr_read;
 289                 }
 290         } else {
 291                 /*
 292                  * Command line (1 string) occupies ARGV and maybe
 293                  * extends into ENVP.
 294                  */
 295                 if (len1 + len2 <= *pos)
 296                         goto skip_argv_envp;
 297                 if (len1 <= *pos)
 298                         goto skip_argv;
 299
 300                 p = arg_start + *pos;
 301                 len = len1 - *pos;
 302                 while (count > 0 && len > 0) {
 303                         unsigned int _count, l;
 304                         int nr_read;
 305                         bool final;
 306
 307                         _count = min3(count, len, PAGE_SIZE);
 308                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 309                         if (nr_read < 0)
 310                                 rv = nr_read;
 311                         if (nr_read <= 0)
 312                                 goto out_free_page;
 313
 314                         /*
 315                          * Command line can be shorter than whole ARGV
 316                          * even if last "marker" byte says it is not.
 317                          */
 318                         final = false;
 319                         l = strnlen(page, nr_read);
 320                         if (l < nr_read) {
 321                                 nr_read = l;
 322                                 final = true;
 323                         }
 324
 325                         if (copy_to_user(buf, page, nr_read)) {
 326                                 rv = -EFAULT;
 327                                 goto out_free_page;
 328                         }
 329
 330                         p       += nr_read;
 331                         len     -= nr_read;
 332                         buf     += nr_read;
 333                         count   -= nr_read;
 334                         rv      += nr_read;
 335
 336                         if (final)
 337                                 goto out_free_page;
 338                 }
 339 skip_argv:
 340                 /*
 341                  * Command line (1 string) occupies ARGV and
 342                  * extends into ENVP.
 343                  */
 344                 if (len1 <= *pos) {
 345                         p = env_start + *pos - len1;
 346                         len = len1 + len2 - *pos;
 347                 } else {
 348                         p = env_start;
 349                         len = len2;
 350                 }
 351                 while (count > 0 && len > 0) {
 352                         unsigned int _count, l;
 353                         int nr_read;
 354                         bool final;
 355
 356                         _count = min3(count, len, PAGE_SIZE);
 357                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 358                         if (nr_read < 0)
 359                                 rv = nr_read;
 360                         if (nr_read <= 0)
 361                                 goto out_free_page;
 362
 363                         /* Find EOS. */
 364                         final = false;
 365                         l = strnlen(page, nr_read);
 366                         if (l < nr_read) {
 367                                 nr_read = l;
 368                                 final = true;
 369                         }
 370
 371                         if (copy_to_user(buf, page, nr_read)) {
 372                                 rv = -EFAULT;
 373                                 goto out_free_page;
 374                         }
 375
 376                         p       += nr_read;
 377                         len     -= nr_read;
 378                         buf     += nr_read;
 379                         count   -= nr_read;
 380                         rv      += nr_read;
 381
 382                         if (final)
 383                                 goto out_free_page;
 384                 }
 385 skip_argv_envp:
 386                 ;
 387         }
 388
 389 out_free_page:
 390         free_page((unsigned long)page);
 391 out_mmput:
 392         mmput(mm);
 393         if (rv > 0)
 394                 *pos += rv;
 395         return rv;
 396 }
 397
 398 static const struct file_operations proc_pid_cmdline_ops = {
 399         .read   = proc_pid_cmdline_read,
 400         .llseek = generic_file_llseek,
 401 };
 402
 403 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 404                          struct pid *pid, struct task_struct *task)
 405 {
 406         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 407         if (mm && !IS_ERR(mm)) {
 408                 unsigned int nwords = 0;
 409                 do {
 410                         nwords += 2;
 411                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 412                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 413                 mmput(mm);
 414                 return 0;
 415         } else
 416                 return PTR_ERR(mm);
 417 }
 418
 419
 420 #ifdef CONFIG_KALLSYMS
 421 /*
 422  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 423  * Returns the resolved symbol.  If that fails, simply return the address.
 424  */
 425 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 426                           struct pid *pid, struct task_struct *task)
 427 {
 428         unsigned long wchan;
 429         char symname[KSYM_NAME_LEN];
 430
 431         wchan = get_wchan(task);
 432
 433         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ) && !lookup_symbol_name(wchan, symname))
 434                 seq_printf(m, "%s", symname);
 435         else
 436                 seq_putc(m, '0');
 437
 438         return 0;
 439 }
 440 #endif /* CONFIG_KALLSYMS */
 441
 442 static int lock_trace(struct task_struct *task)
 443 {
 444         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 445         if (err)
 446                 return err;
 447         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 448                 mutex_unlock(&task->signal->cred_guard_mutex);
 449                 return -EPERM;
 450         }
 451         return 0;
 452 }
 453
 454 static void unlock_trace(struct task_struct *task)
 455 {
 456         mutex_unlock(&task->signal->cred_guard_mutex);
 457 }
 458
 459 #ifdef CONFIG_STACKTRACE
 460
 461 #define MAX_STACK_TRACE_DEPTH   64
 462
 463 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 464                           struct pid *pid, struct task_struct *task)
 465 {
 466         struct stack_trace trace;
 467         unsigned long *entries;
 468         int err;
 469         int i;
 470
 471         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 472         if (!entries)
 473                 return -ENOMEM;
 474
 475         trace.nr_entries        = 0;
 476         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 477         trace.entries           = entries;
 478         trace.skip              = 0;
 479
 480         err = lock_trace(task);
 481         if (!err) {
 482                 save_stack_trace_tsk(task, &trace);
 483
 484                 for (i = 0; i < trace.nr_entries; i++) {
 485                         seq_printf(m, "[<%pK>] %pS\n",
 486                                    (void *)entries[i], (void *)entries[i]);
 487                 }
 488                 unlock_trace(task);
 489         }
 490         kfree(entries);
 491
 492         return err;
 493 }
 494 #endif
 495
 496 #ifdef CONFIG_SCHED_INFO
 497 /*
 498  * Provides /proc/PID/schedstat
 499  */
 500 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 501                               struct pid *pid, struct task_struct *task)
 502 {
 503         if (unlikely(!sched_info_on()))
 504                 seq_printf(m, "0 0 0\n");
 505         else
 506                 seq_printf(m, "%llu %llu %lu\n",
 507                    (unsigned long long)task->se.sum_exec_runtime,
 508                    (unsigned long long)task->sched_info.run_delay,
 509                    task->sched_info.pcount);
 510
 511         return 0;
 512 }
 513 #endif
 514
 515 #ifdef CONFIG_LATENCYTOP
 516 static int lstats_show_proc(struct seq_file *m, void *v)
 517 {
 518         int i;
 519         struct inode *inode = m->private;
 520         struct task_struct *task = get_proc_task(inode);
 521
 522         if (!task)
 523                 return -ESRCH;
 524         seq_puts(m, "Latency Top version : v0.1\n");
 525         for (i = 0; i < 32; i++) {
 526                 struct latency_record *lr = &task->latency_record[i];
 527                 if (lr->backtrace[0]) {
 528                         int q;
 529                         seq_printf(m, "%i %li %li",
 530                                    lr->count, lr->time, lr->max);
 531                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 532                                 unsigned long bt = lr->backtrace[q];
 533                                 if (!bt)
 534                                         break;
 535                                 if (bt == ULONG_MAX)
 536                                         break;
 537                                 seq_printf(m, " %ps", (void *)bt);
 538                         }
 539                         seq_putc(m, '\n');
 540                 }
 541
 542         }
 543         put_task_struct(task);
 544         return 0;
 545 }
 546
 547 static int lstats_open(struct inode *inode, struct file *file)
 548 {
 549         return single_open(file, lstats_show_proc, inode);
 550 }
 551
 552 static ssize_t lstats_write(struct file *file, const char __user *buf,
 553                             size_t count, loff_t *offs)
 554 {
 555         struct task_struct *task = get_proc_task(file_inode(file));
 556
 557         if (!task)
 558                 return -ESRCH;
 559         clear_all_latency_tracing(task);
 560         put_task_struct(task);
 561
 562         return count;
 563 }
 564
 565 static const struct file_operations proc_lstats_operations = {
 566         .open           = lstats_open,
 567         .read           = seq_read,
 568         .write          = lstats_write,
 569         .llseek         = seq_lseek,
 570         .release        = single_release,
 571 };
 572
 573 #endif
 574
 575 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 576                           struct pid *pid, struct task_struct *task)
 577 {
 578         unsigned long totalpages = totalram_pages + total_swap_pages;
 579         unsigned long points = 0;
 580
 581         read_lock(&tasklist_lock);
 582         if (pid_alive(task))
 583                 points = oom_badness(task, NULL, NULL, totalpages) *
 584                                                 1000 / totalpages;
 585         read_unlock(&tasklist_lock);
 586         seq_printf(m, "%lu\n", points);
 587
 588         return 0;
 589 }
 590
 591 struct limit_names {
 592         const char *name;
 593         const char *unit;
 594 };
 595
 596 static const struct limit_names lnames[RLIM_NLIMITS] = {
 597         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 598         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 599         [RLIMIT_DATA] = {"Max data size", "bytes"},
 600         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 601         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 602         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 603         [RLIMIT_NPROC] = {"Max processes", "processes"},
 604         [RLIMIT_NOFILE] = {"Max open files", "files"},
 605         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 606         [RLIMIT_AS] = {"Max address space", "bytes"},
 607         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 608         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 609         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 610         [RLIMIT_NICE] = {"Max nice priority", NULL},
 611         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 612         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 613 };
 614
 615 /* Display limits for a process */
 616 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 617                            struct pid *pid, struct task_struct *task)
 618 {
 619         unsigned int i;
 620         unsigned long flags;
 621
 622         struct rlimit rlim[RLIM_NLIMITS];
 623
 624         if (!lock_task_sighand(task, &flags))
 625                 return 0;
 626         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 627         unlock_task_sighand(task, &flags);
 628
 629         /*
 630          * print the file header
 631          */
 632        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 633                   "Limit", "Soft Limit", "Hard Limit", "Units");
 634
 635         for (i = 0; i < RLIM_NLIMITS; i++) {
 636                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 637                         seq_printf(m, "%-25s %-20s ",
 638                                    lnames[i].name, "unlimited");
 639                 else
 640                         seq_printf(m, "%-25s %-20lu ",
 641                                    lnames[i].name, rlim[i].rlim_cur);
 642
 643                 if (rlim[i].rlim_max == RLIM_INFINITY)
 644                         seq_printf(m, "%-20s ", "unlimited");
 645                 else
 646                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 647
 648                 if (lnames[i].unit)
 649                         seq_printf(m, "%-10s\n", lnames[i].unit);
 650                 else
 651                         seq_putc(m, '\n');
 652         }
 653
 654         return 0;
 655 }
 656
 657 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 658 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 659                             struct pid *pid, struct task_struct *task)
 660 {
 661         long nr;
 662         unsigned long args[6], sp, pc;
 663         int res;
 664
 665         res = lock_trace(task);
 666         if (res)
 667                 return res;
 668
 669         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 670                 seq_puts(m, "running\n");
 671         else if (nr < 0)
 672                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 673         else
 674                 seq_printf(m,
 675                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 676                        nr,
 677                        args[0], args[1], args[2], args[3], args[4], args[5],
 678                        sp, pc);
 679         unlock_trace(task);
 680
 681         return 0;
 682 }
 683 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 684
 685 /************************************************************************/
 686 /*                       Here the fs part begins                        */
 687 /************************************************************************/
 688
 689 /* permission checks */
 690 static int proc_fd_access_allowed(struct inode *inode)
 691 {
 692         struct task_struct *task;
 693         int allowed = 0;
 694         /* Allow access to a task's file descriptors if it is us or we
 695          * may use ptrace attach to the process and find out that
 696          * information.
 697          */
 698         task = get_proc_task(inode);
 699         if (task) {
 700                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 701                 put_task_struct(task);
 702         }
 703         return allowed;
 704 }
 705
 706 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 707 {
 708         int error;
 709         struct inode *inode = d_inode(dentry);
 710
 711         if (attr->ia_valid & ATTR_MODE)
 712                 return -EPERM;
 713
 714         error = inode_change_ok(inode, attr);
 715         if (error)
 716                 return error;
 717
 718         setattr_copy(inode, attr);
 719         mark_inode_dirty(inode);
 720         return 0;
 721 }
 722
 723 /*
 724  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 725  * or euid/egid (for hide_pid_min=2)?
 726  */
 727 static bool has_pid_permissions(struct pid_namespace *pid,
 728                                  struct task_struct *task,
 729                                  int hide_pid_min)
 730 {
 731         if (pid->hide_pid < hide_pid_min)
 732                 return true;
 733         if (in_group_p(pid->pid_gid))
 734                 return true;
 735         return ptrace_may_access(task, PTRACE_MODE_READ);
 736 }
 737
 738
 739 static int proc_pid_permission(struct inode *inode, int mask)
 740 {
 741         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 742         struct task_struct *task;
 743         bool has_perms;
 744
 745         task = get_proc_task(inode);
 746         if (!task)
 747                 return -ESRCH;
 748         has_perms = has_pid_permissions(pid, task, 1);
 749         put_task_struct(task);
 750
 751         if (!has_perms) {
 752                 if (pid->hide_pid == 2) {
 753                         /*
 754                          * Let's make getdents(), stat(), and open()
 755                          * consistent with each other.  If a process
 756                          * may not stat() a file, it shouldn't be seen
 757                          * in procfs at all.
 758                          */
 759                         return -ENOENT;
 760                 }
 761
 762                 return -EPERM;
 763         }
 764         return generic_permission(inode, mask);
 765 }
 766
 767
 768
 769 static const struct inode_operations proc_def_inode_operations = {
 770         .setattr        = proc_setattr,
 771 };
 772
 773 static int proc_single_show(struct seq_file *m, void *v)
 774 {
 775         struct inode *inode = m->private;
 776         struct pid_namespace *ns;
 777         struct pid *pid;
 778         struct task_struct *task;
 779         int ret;
 780
 781         ns = inode->i_sb->s_fs_info;
 782         pid = proc_pid(inode);
 783         task = get_pid_task(pid, PIDTYPE_PID);
 784         if (!task)
 785                 return -ESRCH;
 786
 787         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 788
 789         put_task_struct(task);
 790         return ret;
 791 }
 792
 793 static int proc_single_open(struct inode *inode, struct file *filp)
 794 {
 795         return single_open(filp, proc_single_show, inode);
 796 }
 797
 798 static const struct file_operations proc_single_file_operations = {
 799         .open           = proc_single_open,
 800         .read           = seq_read,
 801         .llseek         = seq_lseek,
 802         .release        = single_release,
 803 };
 804
 805
 806 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 807 {
 808         struct task_struct *task = get_proc_task(inode);
 809         struct mm_struct *mm = ERR_PTR(-ESRCH);
 810
 811         if (task) {
 812                 mm = mm_access(task, mode);
 813                 put_task_struct(task);
 814
 815                 if (!IS_ERR_OR_NULL(mm)) {
 816                         /* ensure this mm_struct can't be freed */
 817                         atomic_inc(&mm->mm_count);
 818                         /* but do not pin its memory */
 819                         mmput(mm);
 820                 }
 821         }
 822
 823         return mm;
 824 }
 825
 826 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 827 {
 828         struct mm_struct *mm = proc_mem_open(inode, mode);
 829
 830         if (IS_ERR(mm))
 831                 return PTR_ERR(mm);
 832
 833         file->private_data = mm;
 834         return 0;
 835 }
 836
 837 static int mem_open(struct inode *inode, struct file *file)
 838 {
 839         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 840
 841         /* OK to pass negative loff_t, we can catch out-of-range */
 842         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 843
 844         return ret;
 845 }
 846
 847 static ssize_t mem_rw(struct file *file, char __user *buf,
 848                         size_t count, loff_t *ppos, int write)
 849 {
 850         struct mm_struct *mm = file->private_data;
 851         unsigned long addr = *ppos;
 852         ssize_t copied;
 853         char *page;
 854
 855         if (!mm)
 856                 return 0;
 857
 858         page = (char *)__get_free_page(GFP_TEMPORARY);
 859         if (!page)
 860                 return -ENOMEM;
 861
 862         copied = 0;
 863         if (!atomic_inc_not_zero(&mm->mm_users))
 864                 goto free;
 865
 866         while (count > 0) {
 867                 int this_len = min_t(int, count, PAGE_SIZE);
 868
 869                 if (write && copy_from_user(page, buf, this_len)) {
 870                         copied = -EFAULT;
 871                         break;
 872                 }
 873
 874                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 875                 if (!this_len) {
 876                         if (!copied)
 877                                 copied = -EIO;
 878                         break;
 879                 }
 880
 881                 if (!write && copy_to_user(buf, page, this_len)) {
 882                         copied = -EFAULT;
 883                         break;
 884                 }
 885
 886                 buf += this_len;
 887                 addr += this_len;
 888                 copied += this_len;
 889                 count -= this_len;
 890         }
 891         *ppos = addr;
 892
 893         mmput(mm);
 894 free:
 895         free_page((unsigned long) page);
 896         return copied;
 897 }
 898
 899 static ssize_t mem_read(struct file *file, char __user *buf,
 900                         size_t count, loff_t *ppos)
 901 {
 902         return mem_rw(file, buf, count, ppos, 0);
 903 }
 904
 905 static ssize_t mem_write(struct file *file, const char __user *buf,
 906                          size_t count, loff_t *ppos)
 907 {
 908         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 909 }
 910
 911 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 912 {
 913         switch (orig) {
 914         case 0:
 915                 file->f_pos = offset;
 916                 break;
 917         case 1:
 918                 file->f_pos += offset;
 919                 break;
 920         default:
 921                 return -EINVAL;
 922         }
 923         force_successful_syscall_return();
 924         return file->f_pos;
 925 }
 926
 927 static int mem_release(struct inode *inode, struct file *file)
 928 {
 929         struct mm_struct *mm = file->private_data;
 930         if (mm)
 931                 mmdrop(mm);
 932         return 0;
 933 }
 934
 935 static const struct file_operations proc_mem_operations = {
 936         .llseek         = mem_lseek,
 937         .read           = mem_read,
 938         .write          = mem_write,
 939         .open           = mem_open,
 940         .release        = mem_release,
 941 };
 942
 943 static int environ_open(struct inode *inode, struct file *file)
 944 {
 945         return __mem_open(inode, file, PTRACE_MODE_READ);
 946 }
 947
 948 static ssize_t environ_read(struct file *file, char __user *buf,
 949                         size_t count, loff_t *ppos)
 950 {
 951         char *page;
 952         unsigned long src = *ppos;
 953         int ret = 0;
 954         struct mm_struct *mm = file->private_data;
 955
 956         if (!mm)
 957                 return 0;
 958
 959         page = (char *)__get_free_page(GFP_TEMPORARY);
 960         if (!page)
 961                 return -ENOMEM;
 962
 963         ret = 0;
 964         if (!atomic_inc_not_zero(&mm->mm_users))
 965                 goto free;
 966         while (count > 0) {
 967                 size_t this_len, max_len;
 968                 int retval;
 969
 970                 if (src >= (mm->env_end - mm->env_start))
 971                         break;
 972
 973                 this_len = mm->env_end - (mm->env_start + src);
 974
 975                 max_len = min_t(size_t, PAGE_SIZE, count);
 976                 this_len = min(max_len, this_len);
 977
 978                 retval = access_remote_vm(mm, (mm->env_start + src),
 979                         page, this_len, 0);
 980
 981                 if (retval <= 0) {
 982                         ret = retval;
 983                         break;
 984                 }
 985
 986                 if (copy_to_user(buf, page, retval)) {
 987                         ret = -EFAULT;
 988                         break;
 989                 }
 990
 991                 ret += retval;
 992                 src += retval;
 993                 buf += retval;
 994                 count -= retval;
 995         }
 996         *ppos = src;
 997         mmput(mm);
 998
 999 free:
1000         free_page((unsigned long) page);
1001         return ret;
1002 }
1003
1004 static const struct file_operations proc_environ_operations = {
1005         .open           = environ_open,
1006         .read           = environ_read,
1007         .llseek         = generic_file_llseek,
1008         .release        = mem_release,
1009 };
1010
1011 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1012                             loff_t *ppos)
1013 {
1014         struct task_struct *task = get_proc_task(file_inode(file));
1015         char buffer[PROC_NUMBUF];
1016         int oom_adj = OOM_ADJUST_MIN;
1017         size_t len;
1018         unsigned long flags;
1019
1020         if (!task)
1021                 return -ESRCH;
1022         if (lock_task_sighand(task, &flags)) {
1023                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1024                         oom_adj = OOM_ADJUST_MAX;
1025                 else
1026                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1027                                   OOM_SCORE_ADJ_MAX;
1028                 unlock_task_sighand(task, &flags);
1029         }
1030         put_task_struct(task);
1031         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1032         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1033 }
1034
1035 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1036                              size_t count, loff_t *ppos)
1037 {
1038         struct task_struct *task;
1039         char buffer[PROC_NUMBUF];
1040         int oom_adj;
1041         unsigned long flags;
1042         int err;
1043
1044         memset(buffer, 0, sizeof(buffer));
1045         if (count > sizeof(buffer) - 1)
1046                 count = sizeof(buffer) - 1;
1047         if (copy_from_user(buffer, buf, count)) {
1048                 err = -EFAULT;
1049                 goto out;
1050         }
1051
1052         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1053         if (err)
1054                 goto out;
1055         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1056              oom_adj != OOM_DISABLE) {
1057                 err = -EINVAL;
1058                 goto out;
1059         }
1060
1061         task = get_proc_task(file_inode(file));
1062         if (!task) {
1063                 err = -ESRCH;
1064                 goto out;
1065         }
1066
1067         task_lock(task);
1068         if (!task->mm) {
1069                 err = -EINVAL;
1070                 goto err_task_lock;
1071         }
1072
1073         if (!lock_task_sighand(task, &flags)) {
1074                 err = -ESRCH;
1075                 goto err_task_lock;
1076         }
1077
1078         /*
1079          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1080          * value is always attainable.
1081          */
1082         if (oom_adj == OOM_ADJUST_MAX)
1083                 oom_adj = OOM_SCORE_ADJ_MAX;
1084         else
1085                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1086
1087         if (oom_adj < task->signal->oom_score_adj &&
1088             !capable(CAP_SYS_RESOURCE)) {
1089                 err = -EACCES;
1090                 goto err_sighand;
1091         }
1092
1093         /*
1094          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1095          * /proc/pid/oom_score_adj instead.
1096          */
1097         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1098                   current->comm, task_pid_nr(current), task_pid_nr(task),
1099                   task_pid_nr(task));
1100
1101         task->signal->oom_score_adj = oom_adj;
1102         trace_oom_score_adj_update(task);
1103 err_sighand:
1104         unlock_task_sighand(task, &flags);
1105 err_task_lock:
1106         task_unlock(task);
1107         put_task_struct(task);
1108 out:
1109         return err < 0 ? err : count;
1110 }
1111
1112 static const struct file_operations proc_oom_adj_operations = {
1113         .read           = oom_adj_read,
1114         .write          = oom_adj_write,
1115         .llseek         = generic_file_llseek,
1116 };
1117
1118 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1119                                         size_t count, loff_t *ppos)
1120 {
1121         struct task_struct *task = get_proc_task(file_inode(file));
1122         char buffer[PROC_NUMBUF];
1123         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1124         unsigned long flags;
1125         size_t len;
1126
1127         if (!task)
1128                 return -ESRCH;
1129         if (lock_task_sighand(task, &flags)) {
1130                 oom_score_adj = task->signal->oom_score_adj;
1131                 unlock_task_sighand(task, &flags);
1132         }
1133         put_task_struct(task);
1134         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1135         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1136 }
1137
1138 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1139                                         size_t count, loff_t *ppos)
1140 {
1141         struct task_struct *task;
1142         char buffer[PROC_NUMBUF];
1143         unsigned long flags;
1144         int oom_score_adj;
1145         int err;
1146
1147         memset(buffer, 0, sizeof(buffer));
1148         if (count > sizeof(buffer) - 1)
1149                 count = sizeof(buffer) - 1;
1150         if (copy_from_user(buffer, buf, count)) {
1151                 err = -EFAULT;
1152                 goto out;
1153         }
1154
1155         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1156         if (err)
1157                 goto out;
1158         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1159                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1160                 err = -EINVAL;
1161                 goto out;
1162         }
1163
1164         task = get_proc_task(file_inode(file));
1165         if (!task) {
1166                 err = -ESRCH;
1167                 goto out;
1168         }
1169
1170         task_lock(task);
1171         if (!task->mm) {
1172                 err = -EINVAL;
1173                 goto err_task_lock;
1174         }
1175
1176         if (!lock_task_sighand(task, &flags)) {
1177                 err = -ESRCH;
1178                 goto err_task_lock;
1179         }
1180
1181         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1182                         !capable(CAP_SYS_RESOURCE)) {
1183                 err = -EACCES;
1184                 goto err_sighand;
1185         }
1186
1187         task->signal->oom_score_adj = (short)oom_score_adj;
1188         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1189                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1190         trace_oom_score_adj_update(task);
1191
1192 err_sighand:
1193         unlock_task_sighand(task, &flags);
1194 err_task_lock:
1195         task_unlock(task);
1196         put_task_struct(task);
1197 out:
1198         return err < 0 ? err : count;
1199 }
1200
1201 static const struct file_operations proc_oom_score_adj_operations = {
1202         .read           = oom_score_adj_read,
1203         .write          = oom_score_adj_write,
1204         .llseek         = default_llseek,
1205 };
1206
1207 #ifdef CONFIG_AUDITSYSCALL
1208 #define TMPBUFLEN 21
1209 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1210                                   size_t count, loff_t *ppos)
1211 {
1212         struct inode * inode = file_inode(file);
1213         struct task_struct *task = get_proc_task(inode);
1214         ssize_t length;
1215         char tmpbuf[TMPBUFLEN];
1216
1217         if (!task)
1218                 return -ESRCH;
1219         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1220                            from_kuid(file->f_cred->user_ns,
1221                                      audit_get_loginuid(task)));
1222         put_task_struct(task);
1223         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1224 }
1225
1226 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1227                                    size_t count, loff_t *ppos)
1228 {
1229         struct inode * inode = file_inode(file);
1230         uid_t loginuid;
1231         kuid_t kloginuid;
1232         int rv;
1233
1234         rcu_read_lock();
1235         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1236                 rcu_read_unlock();
1237                 return -EPERM;
1238         }
1239         rcu_read_unlock();
1240
1241         if (*ppos != 0) {
1242                 /* No partial writes. */
1243                 return -EINVAL;
1244         }
1245
1246         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1247         if (rv < 0)
1248                 return rv;
1249
1250         /* is userspace tring to explicitly UNSET the loginuid? */
1251         if (loginuid == AUDIT_UID_UNSET) {
1252                 kloginuid = INVALID_UID;
1253         } else {
1254                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1255                 if (!uid_valid(kloginuid))
1256                         return -EINVAL;
1257         }
1258
1259         rv = audit_set_loginuid(kloginuid);
1260         if (rv < 0)
1261                 return rv;
1262         return count;
1263 }
1264
1265 static const struct file_operations proc_loginuid_operations = {
1266         .read           = proc_loginuid_read,
1267         .write          = proc_loginuid_write,
1268         .llseek         = generic_file_llseek,
1269 };
1270
1271 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1272                                   size_t count, loff_t *ppos)
1273 {
1274         struct inode * inode = file_inode(file);
1275         struct task_struct *task = get_proc_task(inode);
1276         ssize_t length;
1277         char tmpbuf[TMPBUFLEN];
1278
1279         if (!task)
1280                 return -ESRCH;
1281         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1282                                 audit_get_sessionid(task));
1283         put_task_struct(task);
1284         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1285 }
1286
1287 static const struct file_operations proc_sessionid_operations = {
1288         .read           = proc_sessionid_read,
1289         .llseek         = generic_file_llseek,
1290 };
1291 #endif
1292
1293 #ifdef CONFIG_FAULT_INJECTION
1294 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1295                                       size_t count, loff_t *ppos)
1296 {
1297         struct task_struct *task = get_proc_task(file_inode(file));
1298         char buffer[PROC_NUMBUF];
1299         size_t len;
1300         int make_it_fail;
1301
1302         if (!task)
1303                 return -ESRCH;
1304         make_it_fail = task->make_it_fail;
1305         put_task_struct(task);
1306
1307         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1308
1309         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1310 }
1311
1312 static ssize_t proc_fault_inject_write(struct file * file,
1313                         const char __user * buf, size_t count, loff_t *ppos)
1314 {
1315         struct task_struct *task;
1316         char buffer[PROC_NUMBUF];
1317         int make_it_fail;
1318         int rv;
1319
1320         if (!capable(CAP_SYS_RESOURCE))
1321                 return -EPERM;
1322         memset(buffer, 0, sizeof(buffer));
1323         if (count > sizeof(buffer) - 1)
1324                 count = sizeof(buffer) - 1;
1325         if (copy_from_user(buffer, buf, count))
1326                 return -EFAULT;
1327         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1328         if (rv < 0)
1329                 return rv;
1330         if (make_it_fail < 0 || make_it_fail > 1)
1331                 return -EINVAL;
1332
1333         task = get_proc_task(file_inode(file));
1334         if (!task)
1335                 return -ESRCH;
1336         task->make_it_fail = make_it_fail;
1337         put_task_struct(task);
1338
1339         return count;
1340 }
1341
1342 static const struct file_operations proc_fault_inject_operations = {
1343         .read           = proc_fault_inject_read,
1344         .write          = proc_fault_inject_write,
1345         .llseek         = generic_file_llseek,
1346 };
1347 #endif
1348
1349
1350 #ifdef CONFIG_SCHED_DEBUG
1351 /*
1352  * Print out various scheduling related per-task fields:
1353  */
1354 static int sched_show(struct seq_file *m, void *v)
1355 {
1356         struct inode *inode = m->private;
1357         struct task_struct *p;
1358
1359         p = get_proc_task(inode);
1360         if (!p)
1361                 return -ESRCH;
1362         proc_sched_show_task(p, m);
1363
1364         put_task_struct(p);
1365
1366         return 0;
1367 }
1368
1369 static ssize_t
1370 sched_write(struct file *file, const char __user *buf,
1371             size_t count, loff_t *offset)
1372 {
1373         struct inode *inode = file_inode(file);
1374         struct task_struct *p;
1375
1376         p = get_proc_task(inode);
1377         if (!p)
1378                 return -ESRCH;
1379         proc_sched_set_task(p);
1380
1381         put_task_struct(p);
1382
1383         return count;
1384 }
1385
1386 static int sched_open(struct inode *inode, struct file *filp)
1387 {
1388         return single_open(filp, sched_show, inode);
1389 }
1390
1391 static const struct file_operations proc_pid_sched_operations = {
1392         .open           = sched_open,
1393         .read           = seq_read,
1394         .write          = sched_write,
1395         .llseek         = seq_lseek,
1396         .release        = single_release,
1397 };
1398
1399 #endif
1400
1401 #ifdef CONFIG_SCHED_AUTOGROUP
1402 /*
1403  * Print out autogroup related information:
1404  */
1405 static int sched_autogroup_show(struct seq_file *m, void *v)
1406 {
1407         struct inode *inode = m->private;
1408         struct task_struct *p;
1409
1410         p = get_proc_task(inode);
1411         if (!p)
1412                 return -ESRCH;
1413         proc_sched_autogroup_show_task(p, m);
1414
1415         put_task_struct(p);
1416
1417         return 0;
1418 }
1419
1420 static ssize_t
1421 sched_autogroup_write(struct file *file, const char __user *buf,
1422             size_t count, loff_t *offset)
1423 {
1424         struct inode *inode = file_inode(file);
1425         struct task_struct *p;
1426         char buffer[PROC_NUMBUF];
1427         int nice;
1428         int err;
1429
1430         memset(buffer, 0, sizeof(buffer));
1431         if (count > sizeof(buffer) - 1)
1432                 count = sizeof(buffer) - 1;
1433         if (copy_from_user(buffer, buf, count))
1434                 return -EFAULT;
1435
1436         err = kstrtoint(strstrip(buffer), 0, &nice);
1437         if (err < 0)
1438                 return err;
1439
1440         p = get_proc_task(inode);
1441         if (!p)
1442                 return -ESRCH;
1443
1444         err = proc_sched_autogroup_set_nice(p, nice);
1445         if (err)
1446                 count = err;
1447
1448         put_task_struct(p);
1449
1450         return count;
1451 }
1452
1453 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1454 {
1455         int ret;
1456
1457         ret = single_open(filp, sched_autogroup_show, NULL);
1458         if (!ret) {
1459                 struct seq_file *m = filp->private_data;
1460
1461                 m->private = inode;
1462         }
1463         return ret;
1464 }
1465
1466 static const struct file_operations proc_pid_sched_autogroup_operations = {
1467         .open           = sched_autogroup_open,
1468         .read           = seq_read,
1469         .write          = sched_autogroup_write,
1470         .llseek         = seq_lseek,
1471         .release        = single_release,
1472 };
1473
1474 #endif /* CONFIG_SCHED_AUTOGROUP */
1475
1476 static ssize_t comm_write(struct file *file, const char __user *buf,
1477                                 size_t count, loff_t *offset)
1478 {
1479         struct inode *inode = file_inode(file);
1480         struct task_struct *p;
1481         char buffer[TASK_COMM_LEN];
1482         const size_t maxlen = sizeof(buffer) - 1;
1483
1484         memset(buffer, 0, sizeof(buffer));
1485         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1486                 return -EFAULT;
1487
1488         p = get_proc_task(inode);
1489         if (!p)
1490                 return -ESRCH;
1491
1492         if (same_thread_group(current, p))
1493                 set_task_comm(p, buffer);
1494         else
1495                 count = -EINVAL;
1496
1497         put_task_struct(p);
1498
1499         return count;
1500 }
1501
1502 static int comm_show(struct seq_file *m, void *v)
1503 {
1504         struct inode *inode = m->private;
1505         struct task_struct *p;
1506
1507         p = get_proc_task(inode);
1508         if (!p)
1509                 return -ESRCH;
1510
1511         task_lock(p);
1512         seq_printf(m, "%s\n", p->comm);
1513         task_unlock(p);
1514
1515         put_task_struct(p);
1516
1517         return 0;
1518 }
1519
1520 static int comm_open(struct inode *inode, struct file *filp)
1521 {
1522         return single_open(filp, comm_show, inode);
1523 }
1524
1525 static const struct file_operations proc_pid_set_comm_operations = {
1526         .open           = comm_open,
1527         .read           = seq_read,
1528         .write          = comm_write,
1529         .llseek         = seq_lseek,
1530         .release        = single_release,
1531 };
1532
1533 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1534 {
1535         struct task_struct *task;
1536         struct mm_struct *mm;
1537         struct file *exe_file;
1538
1539         task = get_proc_task(d_inode(dentry));
1540         if (!task)
1541                 return -ENOENT;
1542         mm = get_task_mm(task);
1543         put_task_struct(task);
1544         if (!mm)
1545                 return -ENOENT;
1546         exe_file = get_mm_exe_file(mm);
1547         mmput(mm);
1548         if (exe_file) {
1549                 *exe_path = exe_file->f_path;
1550                 path_get(&exe_file->f_path);
1551                 fput(exe_file);
1552                 return 0;
1553         } else
1554                 return -ENOENT;
1555 }
1556
1557 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1558 {
1559         struct inode *inode = d_inode(dentry);
1560         struct path path;
1561         int error = -EACCES;
1562
1563         /* Are we allowed to snoop on the tasks file descriptors? */
1564         if (!proc_fd_access_allowed(inode))
1565                 goto out;
1566
1567         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1568         if (error)
1569                 goto out;
1570
1571         nd_jump_link(&path);
1572         return NULL;
1573 out:
1574         return ERR_PTR(error);
1575 }
1576
1577 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1578 {
1579         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1580         char *pathname;
1581         int len;
1582
1583         if (!tmp)
1584                 return -ENOMEM;
1585
1586         pathname = d_path(path, tmp, PAGE_SIZE);
1587         len = PTR_ERR(pathname);
1588         if (IS_ERR(pathname))
1589                 goto out;
1590         len = tmp + PAGE_SIZE - 1 - pathname;
1591
1592         if (len > buflen)
1593                 len = buflen;
1594         if (copy_to_user(buffer, pathname, len))
1595                 len = -EFAULT;
1596  out:
1597         free_page((unsigned long)tmp);
1598         return len;
1599 }
1600
1601 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1602 {
1603         int error = -EACCES;
1604         struct inode *inode = d_inode(dentry);
1605         struct path path;
1606
1607         /* Are we allowed to snoop on the tasks file descriptors? */
1608         if (!proc_fd_access_allowed(inode))
1609                 goto out;
1610
1611         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1612         if (error)
1613                 goto out;
1614
1615         error = do_proc_readlink(&path, buffer, buflen);
1616         path_put(&path);
1617 out:
1618         return error;
1619 }
1620
1621 const struct inode_operations proc_pid_link_inode_operations = {
1622         .readlink       = proc_pid_readlink,
1623         .follow_link    = proc_pid_follow_link,
1624         .setattr        = proc_setattr,
1625 };
1626
1627
1628 /* building an inode */
1629
1630 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1631 {
1632         struct inode * inode;
1633         struct proc_inode *ei;
1634         const struct cred *cred;
1635
1636         /* We need a new inode */
1637
1638         inode = new_inode(sb);
1639         if (!inode)
1640                 goto out;
1641
1642         /* Common stuff */
1643         ei = PROC_I(inode);
1644         inode->i_ino = get_next_ino();
1645         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1646         inode->i_op = &proc_def_inode_operations;
1647
1648         /*
1649          * grab the reference to task.
1650          */
1651         ei->pid = get_task_pid(task, PIDTYPE_PID);
1652         if (!ei->pid)
1653                 goto out_unlock;
1654
1655         if (task_dumpable(task)) {
1656                 rcu_read_lock();
1657                 cred = __task_cred(task);
1658                 inode->i_uid = cred->euid;
1659                 inode->i_gid = cred->egid;
1660                 rcu_read_unlock();
1661         }
1662         security_task_to_inode(task, inode);
1663
1664 out:
1665         return inode;
1666
1667 out_unlock:
1668         iput(inode);
1669         return NULL;
1670 }
1671
1672 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1673 {
1674         struct inode *inode = d_inode(dentry);
1675         struct task_struct *task;
1676         const struct cred *cred;
1677         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1678
1679         generic_fillattr(inode, stat);
1680
1681         rcu_read_lock();
1682         stat->uid = GLOBAL_ROOT_UID;
1683         stat->gid = GLOBAL_ROOT_GID;
1684         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1685         if (task) {
1686                 if (!has_pid_permissions(pid, task, 2)) {
1687                         rcu_read_unlock();
1688                         /*
1689                          * This doesn't prevent learning whether PID exists,
1690                          * it only makes getattr() consistent with readdir().
1691                          */
1692                         return -ENOENT;
1693                 }
1694                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1695                     task_dumpable(task)) {
1696                         cred = __task_cred(task);
1697                         stat->uid = cred->euid;
1698                         stat->gid = cred->egid;
1699                 }
1700         }
1701         rcu_read_unlock();
1702         return 0;
1703 }
1704
1705 /* dentry stuff */
1706
1707 /*
1708  *      Exceptional case: normally we are not allowed to unhash a busy
1709  * directory. In this case, however, we can do it - no aliasing problems
1710  * due to the way we treat inodes.
1711  *
1712  * Rewrite the inode's ownerships here because the owning task may have
1713  * performed a setuid(), etc.
1714  *
1715  * Before the /proc/pid/status file was created the only way to read
1716  * the effective uid of a /process was to stat /proc/pid.  Reading
1717  * /proc/pid/status is slow enough that procps and other packages
1718  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1719  * made this apply to all per process world readable and executable
1720  * directories.
1721  */
1722 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1723 {
1724         struct inode *inode;
1725         struct task_struct *task;
1726         const struct cred *cred;
1727
1728         if (flags & LOOKUP_RCU)
1729                 return -ECHILD;
1730
1731         inode = d_inode(dentry);
1732         task = get_proc_task(inode);
1733
1734         if (task) {
1735                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1736                     task_dumpable(task)) {
1737                         rcu_read_lock();
1738                         cred = __task_cred(task);
1739                         inode->i_uid = cred->euid;
1740                         inode->i_gid = cred->egid;
1741                         rcu_read_unlock();
1742                 } else {
1743                         inode->i_uid = GLOBAL_ROOT_UID;
1744                         inode->i_gid = GLOBAL_ROOT_GID;
1745                 }
1746                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1747                 security_task_to_inode(task, inode);
1748                 put_task_struct(task);
1749                 return 1;
1750         }
1751         return 0;
1752 }
1753
1754 static inline bool proc_inode_is_dead(struct inode *inode)
1755 {
1756         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1757 }
1758
1759 int pid_delete_dentry(const struct dentry *dentry)
1760 {
1761         /* Is the task we represent dead?
1762          * If so, then don't put the dentry on the lru list,
1763          * kill it immediately.
1764          */
1765         return proc_inode_is_dead(d_inode(dentry));
1766 }
1767
1768 const struct dentry_operations pid_dentry_operations =
1769 {
1770         .d_revalidate   = pid_revalidate,
1771         .d_delete       = pid_delete_dentry,
1772 };
1773
1774 /* Lookups */
1775
1776 /*
1777  * Fill a directory entry.
1778  *
1779  * If possible create the dcache entry and derive our inode number and
1780  * file type from dcache entry.
1781  *
1782  * Since all of the proc inode numbers are dynamically generated, the inode
1783  * numbers do not exist until the inode is cache.  This means creating the
1784  * the dcache entry in readdir is necessary to keep the inode numbers
1785  * reported by readdir in sync with the inode numbers reported
1786  * by stat.
1787  */
1788 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1789         const char *name, int len,
1790         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1791 {
1792         struct dentry *child, *dir = file->f_path.dentry;
1793         struct qstr qname = QSTR_INIT(name, len);
1794         struct inode *inode;
1795         unsigned type;
1796         ino_t ino;
1797
1798         child = d_hash_and_lookup(dir, &qname);
1799         if (!child) {
1800                 child = d_alloc(dir, &qname);
1801                 if (!child)
1802                         goto end_instantiate;
1803                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1804                         dput(child);
1805                         goto end_instantiate;
1806                 }
1807         }
1808         inode = d_inode(child);
1809         ino = inode->i_ino;
1810         type = inode->i_mode >> 12;
1811         dput(child);
1812         return dir_emit(ctx, name, len, ino, type);
1813
1814 end_instantiate:
1815         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1816 }
1817
1818 /*
1819  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1820  * which represent vma start and end addresses.
1821  */
1822 static int dname_to_vma_addr(struct dentry *dentry,
1823                              unsigned long *start, unsigned long *end)
1824 {
1825         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1826                 return -EINVAL;
1827
1828         return 0;
1829 }
1830
1831 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1832 {
1833         unsigned long vm_start, vm_end;
1834         bool exact_vma_exists = false;
1835         struct mm_struct *mm = NULL;
1836         struct task_struct *task;
1837         const struct cred *cred;
1838         struct inode *inode;
1839         int status = 0;
1840
1841         if (flags & LOOKUP_RCU)
1842                 return -ECHILD;
1843
1844         inode = d_inode(dentry);
1845         task = get_proc_task(inode);
1846         if (!task)
1847                 goto out_notask;
1848
1849         mm = mm_access(task, PTRACE_MODE_READ);
1850         if (IS_ERR_OR_NULL(mm))
1851                 goto out;
1852
1853         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1854                 down_read(&mm->mmap_sem);
1855                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1856                 up_read(&mm->mmap_sem);
1857         }
1858
1859         mmput(mm);
1860
1861         if (exact_vma_exists) {
1862                 if (task_dumpable(task)) {
1863                         rcu_read_lock();
1864                         cred = __task_cred(task);
1865                         inode->i_uid = cred->euid;
1866                         inode->i_gid = cred->egid;
1867                         rcu_read_unlock();
1868                 } else {
1869                         inode->i_uid = GLOBAL_ROOT_UID;
1870                         inode->i_gid = GLOBAL_ROOT_GID;
1871                 }
1872                 security_task_to_inode(task, inode);
1873                 status = 1;
1874         }
1875
1876 out:
1877         put_task_struct(task);
1878
1879 out_notask:
1880         return status;
1881 }
1882
1883 static const struct dentry_operations tid_map_files_dentry_operations = {
1884         .d_revalidate   = map_files_d_revalidate,
1885         .d_delete       = pid_delete_dentry,
1886 };
1887
1888 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1889 {
1890         unsigned long vm_start, vm_end;
1891         struct vm_area_struct *vma;
1892         struct task_struct *task;
1893         struct mm_struct *mm;
1894         int rc;
1895
1896         rc = -ENOENT;
1897         task = get_proc_task(d_inode(dentry));
1898         if (!task)
1899                 goto out;
1900
1901         mm = get_task_mm(task);
1902         put_task_struct(task);
1903         if (!mm)
1904                 goto out;
1905
1906         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1907         if (rc)
1908                 goto out_mmput;
1909
1910         rc = -ENOENT;
1911         down_read(&mm->mmap_sem);
1912         vma = find_exact_vma(mm, vm_start, vm_end);
1913         if (vma && vma->vm_file) {
1914                 *path = vma->vm_file->f_path;
1915                 path_get(path);
1916                 rc = 0;
1917         }
1918         up_read(&mm->mmap_sem);
1919
1920 out_mmput:
1921         mmput(mm);
1922 out:
1923         return rc;
1924 }
1925
1926 struct map_files_info {
1927         fmode_t         mode;
1928         unsigned long   len;
1929         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1930 };
1931
1932 /*
1933  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1934  * symlinks may be used to bypass permissions on ancestor directories in the
1935  * path to the file in question.
1936  */
1937 static const char *
1938 proc_map_files_follow_link(struct dentry *dentry, void **cookie)
1939 {
1940         if (!capable(CAP_SYS_ADMIN))
1941                 return ERR_PTR(-EPERM);
1942
1943         return proc_pid_follow_link(dentry, NULL);
1944 }
1945
1946 /*
1947  * Identical to proc_pid_link_inode_operations except for follow_link()
1948  */
1949 static const struct inode_operations proc_map_files_link_inode_operations = {
1950         .readlink       = proc_pid_readlink,
1951         .follow_link    = proc_map_files_follow_link,
1952         .setattr        = proc_setattr,
1953 };
1954
1955 static int
1956 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1957                            struct task_struct *task, const void *ptr)
1958 {
1959         fmode_t mode = (fmode_t)(unsigned long)ptr;
1960         struct proc_inode *ei;
1961         struct inode *inode;
1962
1963         inode = proc_pid_make_inode(dir->i_sb, task);
1964         if (!inode)
1965                 return -ENOENT;
1966
1967         ei = PROC_I(inode);
1968         ei->op.proc_get_link = proc_map_files_get_link;
1969
1970         inode->i_op = &proc_map_files_link_inode_operations;
1971         inode->i_size = 64;
1972         inode->i_mode = S_IFLNK;
1973
1974         if (mode & FMODE_READ)
1975                 inode->i_mode |= S_IRUSR;
1976         if (mode & FMODE_WRITE)
1977                 inode->i_mode |= S_IWUSR;
1978
1979         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1980         d_add(dentry, inode);
1981
1982         return 0;
1983 }
1984
1985 static struct dentry *proc_map_files_lookup(struct inode *dir,
1986                 struct dentry *dentry, unsigned int flags)
1987 {
1988         unsigned long vm_start, vm_end;
1989         struct vm_area_struct *vma;
1990         struct task_struct *task;
1991         int result;
1992         struct mm_struct *mm;
1993
1994         result = -ENOENT;
1995         task = get_proc_task(dir);
1996         if (!task)
1997                 goto out;
1998
1999         result = -EACCES;
2000         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2001                 goto out_put_task;
2002
2003         result = -ENOENT;
2004         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2005                 goto out_put_task;
2006
2007         mm = get_task_mm(task);
2008         if (!mm)
2009                 goto out_put_task;
2010
2011         down_read(&mm->mmap_sem);
2012         vma = find_exact_vma(mm, vm_start, vm_end);
2013         if (!vma)
2014                 goto out_no_vma;
2015
2016         if (vma->vm_file)
2017                 result = proc_map_files_instantiate(dir, dentry, task,
2018                                 (void *)(unsigned long)vma->vm_file->f_mode);
2019
2020 out_no_vma:
2021         up_read(&mm->mmap_sem);
2022         mmput(mm);
2023 out_put_task:
2024         put_task_struct(task);
2025 out:
2026         return ERR_PTR(result);
2027 }
2028
2029 static const struct inode_operations proc_map_files_inode_operations = {
2030         .lookup         = proc_map_files_lookup,
2031         .permission     = proc_fd_permission,
2032         .setattr        = proc_setattr,
2033 };
2034
2035 static int
2036 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2037 {
2038         struct vm_area_struct *vma;
2039         struct task_struct *task;
2040         struct mm_struct *mm;
2041         unsigned long nr_files, pos, i;
2042         struct flex_array *fa = NULL;
2043         struct map_files_info info;
2044         struct map_files_info *p;
2045         int ret;
2046
2047         ret = -ENOENT;
2048         task = get_proc_task(file_inode(file));
2049         if (!task)
2050                 goto out;
2051
2052         ret = -EACCES;
2053         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2054                 goto out_put_task;
2055
2056         ret = 0;
2057         if (!dir_emit_dots(file, ctx))
2058                 goto out_put_task;
2059
2060         mm = get_task_mm(task);
2061         if (!mm)
2062                 goto out_put_task;
2063         down_read(&mm->mmap_sem);
2064
2065         nr_files = 0;
2066
2067         /*
2068          * We need two passes here:
2069          *
2070          *  1) Collect vmas of mapped files with mmap_sem taken
2071          *  2) Release mmap_sem and instantiate entries
2072          *
2073          * otherwise we get lockdep complained, since filldir()
2074          * routine might require mmap_sem taken in might_fault().
2075          */
2076
2077         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2078                 if (vma->vm_file && ++pos > ctx->pos)
2079                         nr_files++;
2080         }
2081
2082         if (nr_files) {
2083                 fa = flex_array_alloc(sizeof(info), nr_files,
2084                                         GFP_KERNEL);
2085                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2086                                                 GFP_KERNEL)) {
2087                         ret = -ENOMEM;
2088                         if (fa)
2089                                 flex_array_free(fa);
2090                         up_read(&mm->mmap_sem);
2091                         mmput(mm);
2092                         goto out_put_task;
2093                 }
2094                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2095                                 vma = vma->vm_next) {
2096                         if (!vma->vm_file)
2097                                 continue;
2098                         if (++pos <= ctx->pos)
2099                                 continue;
2100
2101                         info.mode = vma->vm_file->f_mode;
2102                         info.len = snprintf(info.name,
2103                                         sizeof(info.name), "%lx-%lx",
2104                                         vma->vm_start, vma->vm_end);
2105                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2106                                 BUG();
2107                 }
2108         }
2109         up_read(&mm->mmap_sem);
2110
2111         for (i = 0; i < nr_files; i++) {
2112                 p = flex_array_get(fa, i);
2113                 if (!proc_fill_cache(file, ctx,
2114                                       p->name, p->len,
2115                                       proc_map_files_instantiate,
2116                                       task,
2117                                       (void *)(unsigned long)p->mode))
2118                         break;
2119                 ctx->pos++;
2120         }
2121         if (fa)
2122                 flex_array_free(fa);
2123         mmput(mm);
2124
2125 out_put_task:
2126         put_task_struct(task);
2127 out:
2128         return ret;
2129 }
2130
2131 static const struct file_operations proc_map_files_operations = {
2132         .read           = generic_read_dir,
2133         .iterate        = proc_map_files_readdir,
2134         .llseek         = default_llseek,
2135 };
2136
2137 struct timers_private {
2138         struct pid *pid;
2139         struct task_struct *task;
2140         struct sighand_struct *sighand;
2141         struct pid_namespace *ns;
2142         unsigned long flags;
2143 };
2144
2145 static void *timers_start(struct seq_file *m, loff_t *pos)
2146 {
2147         struct timers_private *tp = m->private;
2148
2149         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2150         if (!tp->task)
2151                 return ERR_PTR(-ESRCH);
2152
2153         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2154         if (!tp->sighand)
2155                 return ERR_PTR(-ESRCH);
2156
2157         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2158 }
2159
2160 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2161 {
2162         struct timers_private *tp = m->private;
2163         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2164 }
2165
2166 static void timers_stop(struct seq_file *m, void *v)
2167 {
2168         struct timers_private *tp = m->private;
2169
2170         if (tp->sighand) {
2171                 unlock_task_sighand(tp->task, &tp->flags);
2172                 tp->sighand = NULL;
2173         }
2174
2175         if (tp->task) {
2176                 put_task_struct(tp->task);
2177                 tp->task = NULL;
2178         }
2179 }
2180
2181 static int show_timer(struct seq_file *m, void *v)
2182 {
2183         struct k_itimer *timer;
2184         struct timers_private *tp = m->private;
2185         int notify;
2186         static const char * const nstr[] = {
2187                 [SIGEV_SIGNAL] = "signal",
2188                 [SIGEV_NONE] = "none",
2189                 [SIGEV_THREAD] = "thread",
2190         };
2191
2192         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2193         notify = timer->it_sigev_notify;
2194
2195         seq_printf(m, "ID: %d\n", timer->it_id);
2196         seq_printf(m, "signal: %d/%p\n",
2197                    timer->sigq->info.si_signo,
2198                    timer->sigq->info.si_value.sival_ptr);
2199         seq_printf(m, "notify: %s/%s.%d\n",
2200                    nstr[notify & ~SIGEV_THREAD_ID],
2201                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2202                    pid_nr_ns(timer->it_pid, tp->ns));
2203         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2204
2205         return 0;
2206 }
2207
2208 static const struct seq_operations proc_timers_seq_ops = {
2209         .start  = timers_start,
2210         .next   = timers_next,
2211         .stop   = timers_stop,
2212         .show   = show_timer,
2213 };
2214
2215 static int proc_timers_open(struct inode *inode, struct file *file)
2216 {
2217         struct timers_private *tp;
2218
2219         tp = __seq_open_private(file, &proc_timers_seq_ops,
2220                         sizeof(struct timers_private));
2221         if (!tp)
2222                 return -ENOMEM;
2223
2224         tp->pid = proc_pid(inode);
2225         tp->ns = inode->i_sb->s_fs_info;
2226         return 0;
2227 }
2228
2229 static const struct file_operations proc_timers_operations = {
2230         .open           = proc_timers_open,
2231         .read           = seq_read,
2232         .llseek         = seq_lseek,
2233         .release        = seq_release_private,
2234 };
2235
2236 static int proc_pident_instantiate(struct inode *dir,
2237         struct dentry *dentry, struct task_struct *task, const void *ptr)
2238 {
2239         const struct pid_entry *p = ptr;
2240         struct inode *inode;
2241         struct proc_inode *ei;
2242
2243         inode = proc_pid_make_inode(dir->i_sb, task);
2244         if (!inode)
2245                 goto out;
2246
2247         ei = PROC_I(inode);
2248         inode->i_mode = p->mode;
2249         if (S_ISDIR(inode->i_mode))
2250                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2251         if (p->iop)
2252                 inode->i_op = p->iop;
2253         if (p->fop)
2254                 inode->i_fop = p->fop;
2255         ei->op = p->op;
2256         d_set_d_op(dentry, &pid_dentry_operations);
2257         d_add(dentry, inode);
2258         /* Close the race of the process dying before we return the dentry */
2259         if (pid_revalidate(dentry, 0))
2260                 return 0;
2261 out:
2262         return -ENOENT;
2263 }
2264
2265 static struct dentry *proc_pident_lookup(struct inode *dir,
2266                                          struct dentry *dentry,
2267                                          const struct pid_entry *ents,
2268                                          unsigned int nents)
2269 {
2270         int error;
2271         struct task_struct *task = get_proc_task(dir);
2272         const struct pid_entry *p, *last;
2273
2274         error = -ENOENT;
2275
2276         if (!task)
2277                 goto out_no_task;
2278
2279         /*
2280          * Yes, it does not scale. And it should not. Don't add
2281          * new entries into /proc/<tgid>/ without very good reasons.
2282          */
2283         last = &ents[nents - 1];
2284         for (p = ents; p <= last; p++) {
2285                 if (p->len != dentry->d_name.len)
2286                         continue;
2287                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2288                         break;
2289         }
2290         if (p > last)
2291                 goto out;
2292
2293         error = proc_pident_instantiate(dir, dentry, task, p);
2294 out:
2295         put_task_struct(task);
2296 out_no_task:
2297         return ERR_PTR(error);
2298 }
2299
2300 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2301                 const struct pid_entry *ents, unsigned int nents)
2302 {
2303         struct task_struct *task = get_proc_task(file_inode(file));
2304         const struct pid_entry *p;
2305
2306         if (!task)
2307                 return -ENOENT;
2308
2309         if (!dir_emit_dots(file, ctx))
2310                 goto out;
2311
2312         if (ctx->pos >= nents + 2)
2313                 goto out;
2314
2315         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2316                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2317                                 proc_pident_instantiate, task, p))
2318                         break;
2319                 ctx->pos++;
2320         }
2321 out:
2322         put_task_struct(task);
2323         return 0;
2324 }
2325
2326 #ifdef CONFIG_SECURITY
2327 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2328                                   size_t count, loff_t *ppos)
2329 {
2330         struct inode * inode = file_inode(file);
2331         char *p = NULL;
2332         ssize_t length;
2333         struct task_struct *task = get_proc_task(inode);
2334
2335         if (!task)
2336                 return -ESRCH;
2337
2338         length = security_getprocattr(task,
2339                                       (char*)file->f_path.dentry->d_name.name,
2340                                       &p);
2341         put_task_struct(task);
2342         if (length > 0)
2343                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2344         kfree(p);
2345         return length;
2346 }
2347
2348 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2349                                    size_t count, loff_t *ppos)
2350 {
2351         struct inode * inode = file_inode(file);
2352         char *page;
2353         ssize_t length;
2354         struct task_struct *task = get_proc_task(inode);
2355
2356         length = -ESRCH;
2357         if (!task)
2358                 goto out_no_task;
2359         if (count > PAGE_SIZE)
2360                 count = PAGE_SIZE;
2361
2362         /* No partial writes. */
2363         length = -EINVAL;
2364         if (*ppos != 0)
2365                 goto out;
2366
2367         length = -ENOMEM;
2368         page = (char*)__get_free_page(GFP_TEMPORARY);
2369         if (!page)
2370                 goto out;
2371
2372         length = -EFAULT;
2373         if (copy_from_user(page, buf, count))
2374                 goto out_free;
2375
2376         /* Guard against adverse ptrace interaction */
2377         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2378         if (length < 0)
2379                 goto out_free;
2380
2381         length = security_setprocattr(task,
2382                                       (char*)file->f_path.dentry->d_name.name,
2383                                       (void*)page, count);
2384         mutex_unlock(&task->signal->cred_guard_mutex);
2385 out_free:
2386         free_page((unsigned long) page);
2387 out:
2388         put_task_struct(task);
2389 out_no_task:
2390         return length;
2391 }
2392
2393 static const struct file_operations proc_pid_attr_operations = {
2394         .read           = proc_pid_attr_read,
2395         .write          = proc_pid_attr_write,
2396         .llseek         = generic_file_llseek,
2397 };
2398
2399 static const struct pid_entry attr_dir_stuff[] = {
2400         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2401         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2402         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2403         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2404         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2405         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2406 };
2407
2408 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2409 {
2410         return proc_pident_readdir(file, ctx,
2411                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2412 }
2413
2414 static const struct file_operations proc_attr_dir_operations = {
2415         .read           = generic_read_dir,
2416         .iterate        = proc_attr_dir_readdir,
2417         .llseek         = default_llseek,
2418 };
2419
2420 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2421                                 struct dentry *dentry, unsigned int flags)
2422 {
2423         return proc_pident_lookup(dir, dentry,
2424                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2425 }
2426
2427 static const struct inode_operations proc_attr_dir_inode_operations = {
2428         .lookup         = proc_attr_dir_lookup,
2429         .getattr        = pid_getattr,
2430         .setattr        = proc_setattr,
2431 };
2432
2433 #endif
2434
2435 #ifdef CONFIG_ELF_CORE
2436 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2437                                          size_t count, loff_t *ppos)
2438 {
2439         struct task_struct *task = get_proc_task(file_inode(file));
2440         struct mm_struct *mm;
2441         char buffer[PROC_NUMBUF];
2442         size_t len;
2443         int ret;
2444
2445         if (!task)
2446                 return -ESRCH;
2447
2448         ret = 0;
2449         mm = get_task_mm(task);
2450         if (mm) {
2451                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2452                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2453                                 MMF_DUMP_FILTER_SHIFT));
2454                 mmput(mm);
2455                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2456         }
2457
2458         put_task_struct(task);
2459
2460         return ret;
2461 }
2462
2463 static ssize_t proc_coredump_filter_write(struct file *file,
2464                                           const char __user *buf,
2465                                           size_t count,
2466                                           loff_t *ppos)
2467 {
2468         struct task_struct *task;
2469         struct mm_struct *mm;
2470         unsigned int val;
2471         int ret;
2472         int i;
2473         unsigned long mask;
2474
2475         ret = kstrtouint_from_user(buf, count, 0, &val);
2476         if (ret < 0)
2477                 return ret;
2478
2479         ret = -ESRCH;
2480         task = get_proc_task(file_inode(file));
2481         if (!task)
2482                 goto out_no_task;
2483
2484         mm = get_task_mm(task);
2485         if (!mm)
2486                 goto out_no_mm;
2487
2488         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2489                 if (val & mask)
2490                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2491                 else
2492                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2493         }
2494
2495         mmput(mm);
2496  out_no_mm:
2497         put_task_struct(task);
2498  out_no_task:
2499         if (ret < 0)
2500                 return ret;
2501         return count;
2502 }
2503
2504 static const struct file_operations proc_coredump_filter_operations = {
2505         .read           = proc_coredump_filter_read,
2506         .write          = proc_coredump_filter_write,
2507         .llseek         = generic_file_llseek,
2508 };
2509 #endif
2510
2511 #ifdef CONFIG_TASK_IO_ACCOUNTING
2512 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2513 {
2514         struct task_io_accounting acct = task->ioac;
2515         unsigned long flags;
2516         int result;
2517
2518         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2519         if (result)
2520                 return result;
2521
2522         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2523                 result = -EACCES;
2524                 goto out_unlock;
2525         }
2526
2527         if (whole && lock_task_sighand(task, &flags)) {
2528                 struct task_struct *t = task;
2529
2530                 task_io_accounting_add(&acct, &task->signal->ioac);
2531                 while_each_thread(task, t)
2532                         task_io_accounting_add(&acct, &t->ioac);
2533
2534                 unlock_task_sighand(task, &flags);
2535         }
2536         seq_printf(m,
2537                    "rchar: %llu\n"
2538                    "wchar: %llu\n"
2539                    "syscr: %llu\n"
2540                    "syscw: %llu\n"
2541                    "read_bytes: %llu\n"
2542                    "write_bytes: %llu\n"
2543                    "cancelled_write_bytes: %llu\n",
2544                    (unsigned long long)acct.rchar,
2545                    (unsigned long long)acct.wchar,
2546                    (unsigned long long)acct.syscr,
2547                    (unsigned long long)acct.syscw,
2548                    (unsigned long long)acct.read_bytes,
2549                    (unsigned long long)acct.write_bytes,
2550                    (unsigned long long)acct.cancelled_write_bytes);
2551         result = 0;
2552
2553 out_unlock:
2554         mutex_unlock(&task->signal->cred_guard_mutex);
2555         return result;
2556 }
2557
2558 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2559                                   struct pid *pid, struct task_struct *task)
2560 {
2561         return do_io_accounting(task, m, 0);
2562 }
2563
2564 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2565                                    struct pid *pid, struct task_struct *task)
2566 {
2567         return do_io_accounting(task, m, 1);
2568 }
2569 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2570
2571 #ifdef CONFIG_USER_NS
2572 static int proc_id_map_open(struct inode *inode, struct file *file,
2573         const struct seq_operations *seq_ops)
2574 {
2575         struct user_namespace *ns = NULL;
2576         struct task_struct *task;
2577         struct seq_file *seq;
2578         int ret = -EINVAL;
2579
2580         task = get_proc_task(inode);
2581         if (task) {
2582                 rcu_read_lock();
2583                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2584                 rcu_read_unlock();
2585                 put_task_struct(task);
2586         }
2587         if (!ns)
2588                 goto err;
2589
2590         ret = seq_open(file, seq_ops);
2591         if (ret)
2592                 goto err_put_ns;
2593
2594         seq = file->private_data;
2595         seq->private = ns;
2596
2597         return 0;
2598 err_put_ns:
2599         put_user_ns(ns);
2600 err:
2601         return ret;
2602 }
2603
2604 static int proc_id_map_release(struct inode *inode, struct file *file)
2605 {
2606         struct seq_file *seq = file->private_data;
2607         struct user_namespace *ns = seq->private;
2608         put_user_ns(ns);
2609         return seq_release(inode, file);
2610 }
2611
2612 static int proc_uid_map_open(struct inode *inode, struct file *file)
2613 {
2614         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2615 }
2616
2617 static int proc_gid_map_open(struct inode *inode, struct file *file)
2618 {
2619         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2620 }
2621
2622 static int proc_projid_map_open(struct inode *inode, struct file *file)
2623 {
2624         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2625 }
2626
2627 static const struct file_operations proc_uid_map_operations = {
2628         .open           = proc_uid_map_open,
2629         .write          = proc_uid_map_write,
2630         .read           = seq_read,
2631         .llseek         = seq_lseek,
2632         .release        = proc_id_map_release,
2633 };
2634
2635 static const struct file_operations proc_gid_map_operations = {
2636         .open           = proc_gid_map_open,
2637         .write          = proc_gid_map_write,
2638         .read           = seq_read,
2639         .llseek         = seq_lseek,
2640         .release        = proc_id_map_release,
2641 };
2642
2643 static const struct file_operations proc_projid_map_operations = {
2644         .open           = proc_projid_map_open,
2645         .write          = proc_projid_map_write,
2646         .read           = seq_read,
2647         .llseek         = seq_lseek,
2648         .release        = proc_id_map_release,
2649 };
2650
2651 static int proc_setgroups_open(struct inode *inode, struct file *file)
2652 {
2653         struct user_namespace *ns = NULL;
2654         struct task_struct *task;
2655         int ret;
2656
2657         ret = -ESRCH;
2658         task = get_proc_task(inode);
2659         if (task) {
2660                 rcu_read_lock();
2661                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2662                 rcu_read_unlock();
2663                 put_task_struct(task);
2664         }
2665         if (!ns)
2666                 goto err;
2667
2668         if (file->f_mode & FMODE_WRITE) {
2669                 ret = -EACCES;
2670                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2671                         goto err_put_ns;
2672         }
2673
2674         ret = single_open(file, &proc_setgroups_show, ns);
2675         if (ret)
2676                 goto err_put_ns;
2677
2678         return 0;
2679 err_put_ns:
2680         put_user_ns(ns);
2681 err:
2682         return ret;
2683 }
2684
2685 static int proc_setgroups_release(struct inode *inode, struct file *file)
2686 {
2687         struct seq_file *seq = file->private_data;
2688         struct user_namespace *ns = seq->private;
2689         int ret = single_release(inode, file);
2690         put_user_ns(ns);
2691         return ret;
2692 }
2693
2694 static const struct file_operations proc_setgroups_operations = {
2695         .open           = proc_setgroups_open,
2696         .write          = proc_setgroups_write,
2697         .read           = seq_read,
2698         .llseek         = seq_lseek,
2699         .release        = proc_setgroups_release,
2700 };
2701 #endif /* CONFIG_USER_NS */
2702
2703 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2704                                 struct pid *pid, struct task_struct *task)
2705 {
2706         int err = lock_trace(task);
2707         if (!err) {
2708                 seq_printf(m, "%08x\n", task->personality);
2709                 unlock_trace(task);
2710         }
2711         return err;
2712 }
2713
2714 /*
2715  * Thread groups
2716  */
2717 static const struct file_operations proc_task_operations;
2718 static const struct inode_operations proc_task_inode_operations;
2719
2720 static const struct pid_entry tgid_base_stuff[] = {
2721         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2722         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2723         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2724         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2725         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2726 #ifdef CONFIG_NET
2727         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2728 #endif
2729         REG("environ",    S_IRUSR, proc_environ_operations),
2730         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2731         ONE("status",     S_IRUGO, proc_pid_status),
2732         ONE("personality", S_IRUSR, proc_pid_personality),
2733         ONE("limits",     S_IRUGO, proc_pid_limits),
2734 #ifdef CONFIG_SCHED_DEBUG
2735         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2736 #endif
2737 #ifdef CONFIG_SCHED_AUTOGROUP
2738         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2739 #endif
2740         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2741 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2742         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2743 #endif
2744         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2745         ONE("stat",       S_IRUGO, proc_tgid_stat),
2746         ONE("statm",      S_IRUGO, proc_pid_statm),
2747         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2748 #ifdef CONFIG_NUMA
2749         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2750 #endif
2751         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2752         LNK("cwd",        proc_cwd_link),
2753         LNK("root",       proc_root_link),
2754         LNK("exe",        proc_exe_link),
2755         REG("mounts",     S_IRUGO, proc_mounts_operations),
2756         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2757         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2758 #ifdef CONFIG_PROC_PAGE_MONITOR
2759         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2760         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2761         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2762 #endif
2763 #ifdef CONFIG_SECURITY
2764         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2765 #endif
2766 #ifdef CONFIG_KALLSYMS
2767         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2768 #endif
2769 #ifdef CONFIG_STACKTRACE
2770         ONE("stack",      S_IRUSR, proc_pid_stack),
2771 #endif
2772 #ifdef CONFIG_SCHED_INFO
2773         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2774 #endif
2775 #ifdef CONFIG_LATENCYTOP
2776         REG("latency",  S_IRUGO, proc_lstats_operations),
2777 #endif
2778 #ifdef CONFIG_PROC_PID_CPUSET
2779         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2780 #endif
2781 #ifdef CONFIG_CGROUPS
2782         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2783 #endif
2784         ONE("oom_score",  S_IRUGO, proc_oom_score),
2785         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2786         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2787 #ifdef CONFIG_AUDITSYSCALL
2788         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2789         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2790 #endif
2791 #ifdef CONFIG_FAULT_INJECTION
2792         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2793 #endif
2794 #ifdef CONFIG_ELF_CORE
2795         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2796 #endif
2797 #ifdef CONFIG_TASK_IO_ACCOUNTING
2798         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2799 #endif
2800 #ifdef CONFIG_HARDWALL
2801         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2802 #endif
2803 #ifdef CONFIG_USER_NS
2804         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2805         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2806         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2807         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2808 #endif
2809 #ifdef CONFIG_CHECKPOINT_RESTORE
2810         REG("timers",     S_IRUGO, proc_timers_operations),
2811 #endif
2812 };
2813
2814 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2815 {
2816         return proc_pident_readdir(file, ctx,
2817                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2818 }
2819
2820 static const struct file_operations proc_tgid_base_operations = {
2821         .read           = generic_read_dir,
2822         .iterate        = proc_tgid_base_readdir,
2823         .llseek         = default_llseek,
2824 };
2825
2826 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2827 {
2828         return proc_pident_lookup(dir, dentry,
2829                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2830 }
2831
2832 static const struct inode_operations proc_tgid_base_inode_operations = {
2833         .lookup         = proc_tgid_base_lookup,
2834         .getattr        = pid_getattr,
2835         .setattr        = proc_setattr,
2836         .permission     = proc_pid_permission,
2837 };
2838
2839 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2840 {
2841         struct dentry *dentry, *leader, *dir;
2842         char buf[PROC_NUMBUF];
2843         struct qstr name;
2844
2845         name.name = buf;
2846         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2847         /* no ->d_hash() rejects on procfs */
2848         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2849         if (dentry) {
2850                 d_invalidate(dentry);
2851                 dput(dentry);
2852         }
2853
2854         if (pid == tgid)
2855                 return;
2856
2857         name.name = buf;
2858         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2859         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2860         if (!leader)
2861                 goto out;
2862
2863         name.name = "task";
2864         name.len = strlen(name.name);
2865         dir = d_hash_and_lookup(leader, &name);
2866         if (!dir)
2867                 goto out_put_leader;
2868
2869         name.name = buf;
2870         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2871         dentry = d_hash_and_lookup(dir, &name);
2872         if (dentry) {
2873                 d_invalidate(dentry);
2874                 dput(dentry);
2875         }
2876
2877         dput(dir);
2878 out_put_leader:
2879         dput(leader);
2880 out:
2881         return;
2882 }
2883
2884 /**
2885  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2886  * @task: task that should be flushed.
2887  *
2888  * When flushing dentries from proc, one needs to flush them from global
2889  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2890  * in. This call is supposed to do all of this job.
2891  *
2892  * Looks in the dcache for
2893  * /proc/@pid
2894  * /proc/@tgid/task/@pid
2895  * if either directory is present flushes it and all of it'ts children
2896  * from the dcache.
2897  *
2898  * It is safe and reasonable to cache /proc entries for a task until
2899  * that task exits.  After that they just clog up the dcache with
2900  * useless entries, possibly causing useful dcache entries to be
2901  * flushed instead.  This routine is proved to flush those useless
2902  * dcache entries at process exit time.
2903  *
2904  * NOTE: This routine is just an optimization so it does not guarantee
2905  *       that no dcache entries will exist at process exit time it
2906  *       just makes it very unlikely that any will persist.
2907  */
2908
2909 void proc_flush_task(struct task_struct *task)
2910 {
2911         int i;
2912         struct pid *pid, *tgid;
2913         struct upid *upid;
2914
2915         pid = task_pid(task);
2916         tgid = task_tgid(task);
2917
2918         for (i = 0; i <= pid->level; i++) {
2919                 upid = &pid->numbers[i];
2920                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2921                                         tgid->numbers[i].nr);
2922         }
2923 }
2924
2925 static int proc_pid_instantiate(struct inode *dir,
2926                                    struct dentry * dentry,
2927                                    struct task_struct *task, const void *ptr)
2928 {
2929         struct inode *inode;
2930
2931         inode = proc_pid_make_inode(dir->i_sb, task);
2932         if (!inode)
2933                 goto out;
2934
2935         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2936         inode->i_op = &proc_tgid_base_inode_operations;
2937         inode->i_fop = &proc_tgid_base_operations;
2938         inode->i_flags|=S_IMMUTABLE;
2939
2940         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2941                                                   ARRAY_SIZE(tgid_base_stuff)));
2942
2943         d_set_d_op(dentry, &pid_dentry_operations);
2944
2945         d_add(dentry, inode);
2946         /* Close the race of the process dying before we return the dentry */
2947         if (pid_revalidate(dentry, 0))
2948                 return 0;
2949 out:
2950         return -ENOENT;
2951 }
2952
2953 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2954 {
2955         int result = -ENOENT;
2956         struct task_struct *task;
2957         unsigned tgid;
2958         struct pid_namespace *ns;
2959
2960         tgid = name_to_int(&dentry->d_name);
2961         if (tgid == ~0U)
2962                 goto out;
2963
2964         ns = dentry->d_sb->s_fs_info;
2965         rcu_read_lock();
2966         task = find_task_by_pid_ns(tgid, ns);
2967         if (task)
2968                 get_task_struct(task);
2969         rcu_read_unlock();
2970         if (!task)
2971                 goto out;
2972
2973         result = proc_pid_instantiate(dir, dentry, task, NULL);
2974         put_task_struct(task);
2975 out:
2976         return ERR_PTR(result);
2977 }
2978
2979 /*
2980  * Find the first task with tgid >= tgid
2981  *
2982  */
2983 struct tgid_iter {
2984         unsigned int tgid;
2985         struct task_struct *task;
2986 };
2987 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2988 {
2989         struct pid *pid;
2990
2991         if (iter.task)
2992                 put_task_struct(iter.task);
2993         rcu_read_lock();
2994 retry:
2995         iter.task = NULL;
2996         pid = find_ge_pid(iter.tgid, ns);
2997         if (pid) {
2998                 iter.tgid = pid_nr_ns(pid, ns);
2999                 iter.task = pid_task(pid, PIDTYPE_PID);
3000                 /* What we to know is if the pid we have find is the
3001                  * pid of a thread_group_leader.  Testing for task
3002                  * being a thread_group_leader is the obvious thing
3003                  * todo but there is a window when it fails, due to
3004                  * the pid transfer logic in de_thread.
3005                  *
3006                  * So we perform the straight forward test of seeing
3007                  * if the pid we have found is the pid of a thread
3008                  * group leader, and don't worry if the task we have
3009                  * found doesn't happen to be a thread group leader.
3010                  * As we don't care in the case of readdir.
3011                  */
3012                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3013                         iter.tgid += 1;
3014                         goto retry;
3015                 }
3016                 get_task_struct(iter.task);
3017         }
3018         rcu_read_unlock();
3019         return iter;
3020 }
3021
3022 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3023
3024 /* for the /proc/ directory itself, after non-process stuff has been done */
3025 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3026 {
3027         struct tgid_iter iter;
3028         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3029         loff_t pos = ctx->pos;
3030
3031         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3032                 return 0;
3033
3034         if (pos == TGID_OFFSET - 2) {
3035                 struct inode *inode = d_inode(ns->proc_self);
3036                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3037                         return 0;
3038                 ctx->pos = pos = pos + 1;
3039         }
3040         if (pos == TGID_OFFSET - 1) {
3041                 struct inode *inode = d_inode(ns->proc_thread_self);
3042                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3043                         return 0;
3044                 ctx->pos = pos = pos + 1;
3045         }
3046         iter.tgid = pos - TGID_OFFSET;
3047         iter.task = NULL;
3048         for (iter = next_tgid(ns, iter);
3049              iter.task;
3050              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3051                 char name[PROC_NUMBUF];
3052                 int len;
3053                 if (!has_pid_permissions(ns, iter.task, 2))
3054                         continue;
3055
3056                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3057                 ctx->pos = iter.tgid + TGID_OFFSET;
3058                 if (!proc_fill_cache(file, ctx, name, len,
3059                                      proc_pid_instantiate, iter.task, NULL)) {
3060                         put_task_struct(iter.task);
3061                         return 0;
3062                 }
3063         }
3064         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3065         return 0;
3066 }
3067
3068 /*
3069  * Tasks
3070  */
3071 static const struct pid_entry tid_base_stuff[] = {
3072         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3073         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3074         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3075 #ifdef CONFIG_NET
3076         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3077 #endif
3078         REG("environ",   S_IRUSR, proc_environ_operations),
3079         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3080         ONE("status",    S_IRUGO, proc_pid_status),
3081         ONE("personality", S_IRUSR, proc_pid_personality),
3082         ONE("limits",    S_IRUGO, proc_pid_limits),
3083 #ifdef CONFIG_SCHED_DEBUG
3084         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3085 #endif
3086         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3087 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3088         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3089 #endif
3090         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3091         ONE("stat",      S_IRUGO, proc_tid_stat),
3092         ONE("statm",     S_IRUGO, proc_pid_statm),
3093         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3094 #ifdef CONFIG_PROC_CHILDREN
3095         REG("children",  S_IRUGO, proc_tid_children_operations),
3096 #endif
3097 #ifdef CONFIG_NUMA
3098         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3099 #endif
3100         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3101         LNK("cwd",       proc_cwd_link),
3102         LNK("root",      proc_root_link),
3103         LNK("exe",       proc_exe_link),
3104         REG("mounts",    S_IRUGO, proc_mounts_operations),
3105         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3106 #ifdef CONFIG_PROC_PAGE_MONITOR
3107         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3108         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3109         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3110 #endif
3111 #ifdef CONFIG_SECURITY
3112         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3113 #endif
3114 #ifdef CONFIG_KALLSYMS
3115         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3116 #endif
3117 #ifdef CONFIG_STACKTRACE
3118         ONE("stack",      S_IRUSR, proc_pid_stack),
3119 #endif
3120 #ifdef CONFIG_SCHED_INFO
3121         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3122 #endif
3123 #ifdef CONFIG_LATENCYTOP
3124         REG("latency",  S_IRUGO, proc_lstats_operations),
3125 #endif
3126 #ifdef CONFIG_PROC_PID_CPUSET
3127         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3128 #endif
3129 #ifdef CONFIG_CGROUPS
3130         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3131 #endif
3132         ONE("oom_score", S_IRUGO, proc_oom_score),
3133         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3134         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3135 #ifdef CONFIG_AUDITSYSCALL
3136         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3137         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3138 #endif
3139 #ifdef CONFIG_FAULT_INJECTION
3140         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3141 #endif
3142 #ifdef CONFIG_TASK_IO_ACCOUNTING
3143         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3144 #endif
3145 #ifdef CONFIG_HARDWALL
3146         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3147 #endif
3148 #ifdef CONFIG_USER_NS
3149         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3150         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3151         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3152         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3153 #endif
3154 };
3155
3156 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3157 {
3158         return proc_pident_readdir(file, ctx,
3159                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3160 }
3161
3162 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3163 {
3164         return proc_pident_lookup(dir, dentry,
3165                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3166 }
3167
3168 static const struct file_operations proc_tid_base_operations = {
3169         .read           = generic_read_dir,
3170         .iterate        = proc_tid_base_readdir,
3171         .llseek         = default_llseek,
3172 };
3173
3174 static const struct inode_operations proc_tid_base_inode_operations = {
3175         .lookup         = proc_tid_base_lookup,
3176         .getattr        = pid_getattr,
3177         .setattr        = proc_setattr,
3178 };
3179
3180 static int proc_task_instantiate(struct inode *dir,
3181         struct dentry *dentry, struct task_struct *task, const void *ptr)
3182 {
3183         struct inode *inode;
3184         inode = proc_pid_make_inode(dir->i_sb, task);
3185
3186         if (!inode)
3187                 goto out;
3188         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3189         inode->i_op = &proc_tid_base_inode_operations;
3190         inode->i_fop = &proc_tid_base_operations;
3191         inode->i_flags|=S_IMMUTABLE;
3192
3193         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3194                                                   ARRAY_SIZE(tid_base_stuff)));
3195
3196         d_set_d_op(dentry, &pid_dentry_operations);
3197
3198         d_add(dentry, inode);
3199         /* Close the race of the process dying before we return the dentry */
3200         if (pid_revalidate(dentry, 0))
3201                 return 0;
3202 out:
3203         return -ENOENT;
3204 }
3205
3206 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3207 {
3208         int result = -ENOENT;
3209         struct task_struct *task;
3210         struct task_struct *leader = get_proc_task(dir);
3211         unsigned tid;
3212         struct pid_namespace *ns;
3213
3214         if (!leader)
3215                 goto out_no_task;
3216
3217         tid = name_to_int(&dentry->d_name);
3218         if (tid == ~0U)
3219                 goto out;
3220
3221         ns = dentry->d_sb->s_fs_info;
3222         rcu_read_lock();
3223         task = find_task_by_pid_ns(tid, ns);
3224         if (task)
3225                 get_task_struct(task);
3226         rcu_read_unlock();
3227         if (!task)
3228                 goto out;
3229         if (!same_thread_group(leader, task))
3230                 goto out_drop_task;
3231
3232         result = proc_task_instantiate(dir, dentry, task, NULL);
3233 out_drop_task:
3234         put_task_struct(task);
3235 out:
3236         put_task_struct(leader);
3237 out_no_task:
3238         return ERR_PTR(result);
3239 }
3240
3241 /*
3242  * Find the first tid of a thread group to return to user space.
3243  *
3244  * Usually this is just the thread group leader, but if the users
3245  * buffer was too small or there was a seek into the middle of the
3246  * directory we have more work todo.
3247  *
3248  * In the case of a short read we start with find_task_by_pid.
3249  *
3250  * In the case of a seek we start with the leader and walk nr
3251  * threads past it.
3252  */
3253 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3254                                         struct pid_namespace *ns)
3255 {
3256         struct task_struct *pos, *task;
3257         unsigned long nr = f_pos;
3258
3259         if (nr != f_pos)        /* 32bit overflow? */
3260                 return NULL;
3261
3262         rcu_read_lock();
3263         task = pid_task(pid, PIDTYPE_PID);
3264         if (!task)
3265                 goto fail;
3266
3267         /* Attempt to start with the tid of a thread */
3268         if (tid && nr) {
3269                 pos = find_task_by_pid_ns(tid, ns);
3270                 if (pos && same_thread_group(pos, task))
3271                         goto found;
3272         }
3273
3274         /* If nr exceeds the number of threads there is nothing todo */
3275         if (nr >= get_nr_threads(task))
3276                 goto fail;
3277
3278         /* If we haven't found our starting place yet start
3279          * with the leader and walk nr threads forward.
3280          */
3281         pos = task = task->group_leader;
3282         do {
3283                 if (!nr--)
3284                         goto found;
3285         } while_each_thread(task, pos);
3286 fail:
3287         pos = NULL;
3288         goto out;
3289 found:
3290         get_task_struct(pos);
3291 out:
3292         rcu_read_unlock();
3293         return pos;
3294 }
3295
3296 /*
3297  * Find the next thread in the thread list.
3298  * Return NULL if there is an error or no next thread.
3299  *
3300  * The reference to the input task_struct is released.
3301  */
3302 static struct task_struct *next_tid(struct task_struct *start)
3303 {
3304         struct task_struct *pos = NULL;
3305         rcu_read_lock();
3306         if (pid_alive(start)) {
3307                 pos = next_thread(start);
3308                 if (thread_group_leader(pos))
3309                         pos = NULL;
3310                 else
3311                         get_task_struct(pos);
3312         }
3313         rcu_read_unlock();
3314         put_task_struct(start);
3315         return pos;
3316 }
3317
3318 /* for the /proc/TGID/task/ directories */
3319 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3320 {
3321         struct inode *inode = file_inode(file);
3322         struct task_struct *task;
3323         struct pid_namespace *ns;
3324         int tid;
3325
3326         if (proc_inode_is_dead(inode))
3327                 return -ENOENT;
3328
3329         if (!dir_emit_dots(file, ctx))
3330                 return 0;
3331
3332         /* f_version caches the tgid value that the last readdir call couldn't
3333          * return. lseek aka telldir automagically resets f_version to 0.
3334          */
3335         ns = inode->i_sb->s_fs_info;
3336         tid = (int)file->f_version;
3337         file->f_version = 0;
3338         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3339              task;
3340              task = next_tid(task), ctx->pos++) {
3341                 char name[PROC_NUMBUF];
3342                 int len;
3343                 tid = task_pid_nr_ns(task, ns);
3344                 len = snprintf(name, sizeof(name), "%d", tid);
3345                 if (!proc_fill_cache(file, ctx, name, len,
3346                                 proc_task_instantiate, task, NULL)) {
3347                         /* returning this tgid failed, save it as the first
3348                          * pid for the next readir call */
3349                         file->f_version = (u64)tid;
3350                         put_task_struct(task);
3351                         break;
3352                 }
3353         }
3354
3355         return 0;
3356 }
3357
3358 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3359 {
3360         struct inode *inode = d_inode(dentry);
3361         struct task_struct *p = get_proc_task(inode);
3362         generic_fillattr(inode, stat);
3363
3364         if (p) {
3365                 stat->nlink += get_nr_threads(p);
3366                 put_task_struct(p);
3367         }
3368
3369         return 0;
3370 }
3371
3372 static const struct inode_operations proc_task_inode_operations = {
3373         .lookup         = proc_task_lookup,
3374         .getattr        = proc_task_getattr,
3375         .setattr        = proc_setattr,
3376         .permission     = proc_pid_permission,
3377 };
3378
3379 static const struct file_operations proc_task_operations = {
3380         .read           = generic_read_dir,
3381         .iterate        = proc_task_readdir,
3382         .llseek         = default_llseek,
3383 };