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 <[email protected]>
  20  *  Edjard Mota <[email protected]>
  21  *  Ilias Biris <[email protected]>
  22  *  Mauricio Lin <[email protected]>
  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 <[email protected]>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <[email protected]>:
  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 /*
1036  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1037  * kernels.  The effective policy is defined by oom_score_adj, which has a
1038  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1039  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1040  * Processes that become oom disabled via oom_adj will still be oom disabled
1041  * with this implementation.
1042  *
1043  * oom_adj cannot be removed since existing userspace binaries use it.
1044  */
1045 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1046                              size_t count, loff_t *ppos)
1047 {
1048         struct task_struct *task;
1049         char buffer[PROC_NUMBUF];
1050         int oom_adj;
1051         unsigned long flags;
1052         int err;
1053
1054         memset(buffer, 0, sizeof(buffer));
1055         if (count > sizeof(buffer) - 1)
1056                 count = sizeof(buffer) - 1;
1057         if (copy_from_user(buffer, buf, count)) {
1058                 err = -EFAULT;
1059                 goto out;
1060         }
1061
1062         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1063         if (err)
1064                 goto out;
1065         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1066              oom_adj != OOM_DISABLE) {
1067                 err = -EINVAL;
1068                 goto out;
1069         }
1070
1071         task = get_proc_task(file_inode(file));
1072         if (!task) {
1073                 err = -ESRCH;
1074                 goto out;
1075         }
1076
1077         task_lock(task);
1078         if (!task->mm) {
1079                 err = -EINVAL;
1080                 goto err_task_lock;
1081         }
1082
1083         if (!lock_task_sighand(task, &flags)) {
1084                 err = -ESRCH;
1085                 goto err_task_lock;
1086         }
1087
1088         /*
1089          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1090          * value is always attainable.
1091          */
1092         if (oom_adj == OOM_ADJUST_MAX)
1093                 oom_adj = OOM_SCORE_ADJ_MAX;
1094         else
1095                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1096
1097         if (oom_adj < task->signal->oom_score_adj &&
1098             !capable(CAP_SYS_RESOURCE)) {
1099                 err = -EACCES;
1100                 goto err_sighand;
1101         }
1102
1103         /*
1104          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1105          * /proc/pid/oom_score_adj instead.
1106          */
1107         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1108                   current->comm, task_pid_nr(current), task_pid_nr(task),
1109                   task_pid_nr(task));
1110
1111         task->signal->oom_score_adj = oom_adj;
1112         trace_oom_score_adj_update(task);
1113 err_sighand:
1114         unlock_task_sighand(task, &flags);
1115 err_task_lock:
1116         task_unlock(task);
1117         put_task_struct(task);
1118 out:
1119         return err < 0 ? err : count;
1120 }
1121
1122 static const struct file_operations proc_oom_adj_operations = {
1123         .read           = oom_adj_read,
1124         .write          = oom_adj_write,
1125         .llseek         = generic_file_llseek,
1126 };
1127
1128 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1129                                         size_t count, loff_t *ppos)
1130 {
1131         struct task_struct *task = get_proc_task(file_inode(file));
1132         char buffer[PROC_NUMBUF];
1133         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1134         unsigned long flags;
1135         size_t len;
1136
1137         if (!task)
1138                 return -ESRCH;
1139         if (lock_task_sighand(task, &flags)) {
1140                 oom_score_adj = task->signal->oom_score_adj;
1141                 unlock_task_sighand(task, &flags);
1142         }
1143         put_task_struct(task);
1144         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1145         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1146 }
1147
1148 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1149                                         size_t count, loff_t *ppos)
1150 {
1151         struct task_struct *task;
1152         char buffer[PROC_NUMBUF];
1153         unsigned long flags;
1154         int oom_score_adj;
1155         int err;
1156
1157         memset(buffer, 0, sizeof(buffer));
1158         if (count > sizeof(buffer) - 1)
1159                 count = sizeof(buffer) - 1;
1160         if (copy_from_user(buffer, buf, count)) {
1161                 err = -EFAULT;
1162                 goto out;
1163         }
1164
1165         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1166         if (err)
1167                 goto out;
1168         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1169                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1170                 err = -EINVAL;
1171                 goto out;
1172         }
1173
1174         task = get_proc_task(file_inode(file));
1175         if (!task) {
1176                 err = -ESRCH;
1177                 goto out;
1178         }
1179
1180         task_lock(task);
1181         if (!task->mm) {
1182                 err = -EINVAL;
1183                 goto err_task_lock;
1184         }
1185
1186         if (!lock_task_sighand(task, &flags)) {
1187                 err = -ESRCH;
1188                 goto err_task_lock;
1189         }
1190
1191         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1192                         !capable(CAP_SYS_RESOURCE)) {
1193                 err = -EACCES;
1194                 goto err_sighand;
1195         }
1196
1197         task->signal->oom_score_adj = (short)oom_score_adj;
1198         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1199                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1200         trace_oom_score_adj_update(task);
1201
1202 err_sighand:
1203         unlock_task_sighand(task, &flags);
1204 err_task_lock:
1205         task_unlock(task);
1206         put_task_struct(task);
1207 out:
1208         return err < 0 ? err : count;
1209 }
1210
1211 static const struct file_operations proc_oom_score_adj_operations = {
1212         .read           = oom_score_adj_read,
1213         .write          = oom_score_adj_write,
1214         .llseek         = default_llseek,
1215 };
1216
1217 #ifdef CONFIG_AUDITSYSCALL
1218 #define TMPBUFLEN 21
1219 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1220                                   size_t count, loff_t *ppos)
1221 {
1222         struct inode * inode = file_inode(file);
1223         struct task_struct *task = get_proc_task(inode);
1224         ssize_t length;
1225         char tmpbuf[TMPBUFLEN];
1226
1227         if (!task)
1228                 return -ESRCH;
1229         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1230                            from_kuid(file->f_cred->user_ns,
1231                                      audit_get_loginuid(task)));
1232         put_task_struct(task);
1233         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1234 }
1235
1236 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1237                                    size_t count, loff_t *ppos)
1238 {
1239         struct inode * inode = file_inode(file);
1240         uid_t loginuid;
1241         kuid_t kloginuid;
1242         int rv;
1243
1244         rcu_read_lock();
1245         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1246                 rcu_read_unlock();
1247                 return -EPERM;
1248         }
1249         rcu_read_unlock();
1250
1251         if (*ppos != 0) {
1252                 /* No partial writes. */
1253                 return -EINVAL;
1254         }
1255
1256         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1257         if (rv < 0)
1258                 return rv;
1259
1260         /* is userspace tring to explicitly UNSET the loginuid? */
1261         if (loginuid == AUDIT_UID_UNSET) {
1262                 kloginuid = INVALID_UID;
1263         } else {
1264                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1265                 if (!uid_valid(kloginuid))
1266                         return -EINVAL;
1267         }
1268
1269         rv = audit_set_loginuid(kloginuid);
1270         if (rv < 0)
1271                 return rv;
1272         return count;
1273 }
1274
1275 static const struct file_operations proc_loginuid_operations = {
1276         .read           = proc_loginuid_read,
1277         .write          = proc_loginuid_write,
1278         .llseek         = generic_file_llseek,
1279 };
1280
1281 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1282                                   size_t count, loff_t *ppos)
1283 {
1284         struct inode * inode = file_inode(file);
1285         struct task_struct *task = get_proc_task(inode);
1286         ssize_t length;
1287         char tmpbuf[TMPBUFLEN];
1288
1289         if (!task)
1290                 return -ESRCH;
1291         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1292                                 audit_get_sessionid(task));
1293         put_task_struct(task);
1294         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1295 }
1296
1297 static const struct file_operations proc_sessionid_operations = {
1298         .read           = proc_sessionid_read,
1299         .llseek         = generic_file_llseek,
1300 };
1301 #endif
1302
1303 #ifdef CONFIG_FAULT_INJECTION
1304 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1305                                       size_t count, loff_t *ppos)
1306 {
1307         struct task_struct *task = get_proc_task(file_inode(file));
1308         char buffer[PROC_NUMBUF];
1309         size_t len;
1310         int make_it_fail;
1311
1312         if (!task)
1313                 return -ESRCH;
1314         make_it_fail = task->make_it_fail;
1315         put_task_struct(task);
1316
1317         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1318
1319         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1320 }
1321
1322 static ssize_t proc_fault_inject_write(struct file * file,
1323                         const char __user * buf, size_t count, loff_t *ppos)
1324 {
1325         struct task_struct *task;
1326         char buffer[PROC_NUMBUF];
1327         int make_it_fail;
1328         int rv;
1329
1330         if (!capable(CAP_SYS_RESOURCE))
1331                 return -EPERM;
1332         memset(buffer, 0, sizeof(buffer));
1333         if (count > sizeof(buffer) - 1)
1334                 count = sizeof(buffer) - 1;
1335         if (copy_from_user(buffer, buf, count))
1336                 return -EFAULT;
1337         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1338         if (rv < 0)
1339                 return rv;
1340         if (make_it_fail < 0 || make_it_fail > 1)
1341                 return -EINVAL;
1342
1343         task = get_proc_task(file_inode(file));
1344         if (!task)
1345                 return -ESRCH;
1346         task->make_it_fail = make_it_fail;
1347         put_task_struct(task);
1348
1349         return count;
1350 }
1351
1352 static const struct file_operations proc_fault_inject_operations = {
1353         .read           = proc_fault_inject_read,
1354         .write          = proc_fault_inject_write,
1355         .llseek         = generic_file_llseek,
1356 };
1357 #endif
1358
1359
1360 #ifdef CONFIG_SCHED_DEBUG
1361 /*
1362  * Print out various scheduling related per-task fields:
1363  */
1364 static int sched_show(struct seq_file *m, void *v)
1365 {
1366         struct inode *inode = m->private;
1367         struct task_struct *p;
1368
1369         p = get_proc_task(inode);
1370         if (!p)
1371                 return -ESRCH;
1372         proc_sched_show_task(p, m);
1373
1374         put_task_struct(p);
1375
1376         return 0;
1377 }
1378
1379 static ssize_t
1380 sched_write(struct file *file, const char __user *buf,
1381             size_t count, loff_t *offset)
1382 {
1383         struct inode *inode = file_inode(file);
1384         struct task_struct *p;
1385
1386         p = get_proc_task(inode);
1387         if (!p)
1388                 return -ESRCH;
1389         proc_sched_set_task(p);
1390
1391         put_task_struct(p);
1392
1393         return count;
1394 }
1395
1396 static int sched_open(struct inode *inode, struct file *filp)
1397 {
1398         return single_open(filp, sched_show, inode);
1399 }
1400
1401 static const struct file_operations proc_pid_sched_operations = {
1402         .open           = sched_open,
1403         .read           = seq_read,
1404         .write          = sched_write,
1405         .llseek         = seq_lseek,
1406         .release        = single_release,
1407 };
1408
1409 #endif
1410
1411 #ifdef CONFIG_SCHED_AUTOGROUP
1412 /*
1413  * Print out autogroup related information:
1414  */
1415 static int sched_autogroup_show(struct seq_file *m, void *v)
1416 {
1417         struct inode *inode = m->private;
1418         struct task_struct *p;
1419
1420         p = get_proc_task(inode);
1421         if (!p)
1422                 return -ESRCH;
1423         proc_sched_autogroup_show_task(p, m);
1424
1425         put_task_struct(p);
1426
1427         return 0;
1428 }
1429
1430 static ssize_t
1431 sched_autogroup_write(struct file *file, const char __user *buf,
1432             size_t count, loff_t *offset)
1433 {
1434         struct inode *inode = file_inode(file);
1435         struct task_struct *p;
1436         char buffer[PROC_NUMBUF];
1437         int nice;
1438         int err;
1439
1440         memset(buffer, 0, sizeof(buffer));
1441         if (count > sizeof(buffer) - 1)
1442                 count = sizeof(buffer) - 1;
1443         if (copy_from_user(buffer, buf, count))
1444                 return -EFAULT;
1445
1446         err = kstrtoint(strstrip(buffer), 0, &nice);
1447         if (err < 0)
1448                 return err;
1449
1450         p = get_proc_task(inode);
1451         if (!p)
1452                 return -ESRCH;
1453
1454         err = proc_sched_autogroup_set_nice(p, nice);
1455         if (err)
1456                 count = err;
1457
1458         put_task_struct(p);
1459
1460         return count;
1461 }
1462
1463 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1464 {
1465         int ret;
1466
1467         ret = single_open(filp, sched_autogroup_show, NULL);
1468         if (!ret) {
1469                 struct seq_file *m = filp->private_data;
1470
1471                 m->private = inode;
1472         }
1473         return ret;
1474 }
1475
1476 static const struct file_operations proc_pid_sched_autogroup_operations = {
1477         .open           = sched_autogroup_open,
1478         .read           = seq_read,
1479         .write          = sched_autogroup_write,
1480         .llseek         = seq_lseek,
1481         .release        = single_release,
1482 };
1483
1484 #endif /* CONFIG_SCHED_AUTOGROUP */
1485
1486 static ssize_t comm_write(struct file *file, const char __user *buf,
1487                                 size_t count, loff_t *offset)
1488 {
1489         struct inode *inode = file_inode(file);
1490         struct task_struct *p;
1491         char buffer[TASK_COMM_LEN];
1492         const size_t maxlen = sizeof(buffer) - 1;
1493
1494         memset(buffer, 0, sizeof(buffer));
1495         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1496                 return -EFAULT;
1497
1498         p = get_proc_task(inode);
1499         if (!p)
1500                 return -ESRCH;
1501
1502         if (same_thread_group(current, p))
1503                 set_task_comm(p, buffer);
1504         else
1505                 count = -EINVAL;
1506
1507         put_task_struct(p);
1508
1509         return count;
1510 }
1511
1512 static int comm_show(struct seq_file *m, void *v)
1513 {
1514         struct inode *inode = m->private;
1515         struct task_struct *p;
1516
1517         p = get_proc_task(inode);
1518         if (!p)
1519                 return -ESRCH;
1520
1521         task_lock(p);
1522         seq_printf(m, "%s\n", p->comm);
1523         task_unlock(p);
1524
1525         put_task_struct(p);
1526
1527         return 0;
1528 }
1529
1530 static int comm_open(struct inode *inode, struct file *filp)
1531 {
1532         return single_open(filp, comm_show, inode);
1533 }
1534
1535 static const struct file_operations proc_pid_set_comm_operations = {
1536         .open           = comm_open,
1537         .read           = seq_read,
1538         .write          = comm_write,
1539         .llseek         = seq_lseek,
1540         .release        = single_release,
1541 };
1542
1543 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1544 {
1545         struct task_struct *task;
1546         struct mm_struct *mm;
1547         struct file *exe_file;
1548
1549         task = get_proc_task(d_inode(dentry));
1550         if (!task)
1551                 return -ENOENT;
1552         mm = get_task_mm(task);
1553         put_task_struct(task);
1554         if (!mm)
1555                 return -ENOENT;
1556         exe_file = get_mm_exe_file(mm);
1557         mmput(mm);
1558         if (exe_file) {
1559                 *exe_path = exe_file->f_path;
1560                 path_get(&exe_file->f_path);
1561                 fput(exe_file);
1562                 return 0;
1563         } else
1564                 return -ENOENT;
1565 }
1566
1567 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1568 {
1569         struct inode *inode = d_inode(dentry);
1570         struct path path;
1571         int error = -EACCES;
1572
1573         /* Are we allowed to snoop on the tasks file descriptors? */
1574         if (!proc_fd_access_allowed(inode))
1575                 goto out;
1576
1577         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1578         if (error)
1579                 goto out;
1580
1581         nd_jump_link(&path);
1582         return NULL;
1583 out:
1584         return ERR_PTR(error);
1585 }
1586
1587 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1588 {
1589         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1590         char *pathname;
1591         int len;
1592
1593         if (!tmp)
1594                 return -ENOMEM;
1595
1596         pathname = d_path(path, tmp, PAGE_SIZE);
1597         len = PTR_ERR(pathname);
1598         if (IS_ERR(pathname))
1599                 goto out;
1600         len = tmp + PAGE_SIZE - 1 - pathname;
1601
1602         if (len > buflen)
1603                 len = buflen;
1604         if (copy_to_user(buffer, pathname, len))
1605                 len = -EFAULT;
1606  out:
1607         free_page((unsigned long)tmp);
1608         return len;
1609 }
1610
1611 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1612 {
1613         int error = -EACCES;
1614         struct inode *inode = d_inode(dentry);
1615         struct path path;
1616
1617         /* Are we allowed to snoop on the tasks file descriptors? */
1618         if (!proc_fd_access_allowed(inode))
1619                 goto out;
1620
1621         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1622         if (error)
1623                 goto out;
1624
1625         error = do_proc_readlink(&path, buffer, buflen);
1626         path_put(&path);
1627 out:
1628         return error;
1629 }
1630
1631 const struct inode_operations proc_pid_link_inode_operations = {
1632         .readlink       = proc_pid_readlink,
1633         .follow_link    = proc_pid_follow_link,
1634         .setattr        = proc_setattr,
1635 };
1636
1637
1638 /* building an inode */
1639
1640 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1641 {
1642         struct inode * inode;
1643         struct proc_inode *ei;
1644         const struct cred *cred;
1645
1646         /* We need a new inode */
1647
1648         inode = new_inode(sb);
1649         if (!inode)
1650                 goto out;
1651
1652         /* Common stuff */
1653         ei = PROC_I(inode);
1654         inode->i_ino = get_next_ino();
1655         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1656         inode->i_op = &proc_def_inode_operations;
1657
1658         /*
1659          * grab the reference to task.
1660          */
1661         ei->pid = get_task_pid(task, PIDTYPE_PID);
1662         if (!ei->pid)
1663                 goto out_unlock;
1664
1665         if (task_dumpable(task)) {
1666                 rcu_read_lock();
1667                 cred = __task_cred(task);
1668                 inode->i_uid = cred->euid;
1669                 inode->i_gid = cred->egid;
1670                 rcu_read_unlock();
1671         }
1672         security_task_to_inode(task, inode);
1673
1674 out:
1675         return inode;
1676
1677 out_unlock:
1678         iput(inode);
1679         return NULL;
1680 }
1681
1682 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1683 {
1684         struct inode *inode = d_inode(dentry);
1685         struct task_struct *task;
1686         const struct cred *cred;
1687         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1688
1689         generic_fillattr(inode, stat);
1690
1691         rcu_read_lock();
1692         stat->uid = GLOBAL_ROOT_UID;
1693         stat->gid = GLOBAL_ROOT_GID;
1694         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1695         if (task) {
1696                 if (!has_pid_permissions(pid, task, 2)) {
1697                         rcu_read_unlock();
1698                         /*
1699                          * This doesn't prevent learning whether PID exists,
1700                          * it only makes getattr() consistent with readdir().
1701                          */
1702                         return -ENOENT;
1703                 }
1704                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1705                     task_dumpable(task)) {
1706                         cred = __task_cred(task);
1707                         stat->uid = cred->euid;
1708                         stat->gid = cred->egid;
1709                 }
1710         }
1711         rcu_read_unlock();
1712         return 0;
1713 }
1714
1715 /* dentry stuff */
1716
1717 /*
1718  *      Exceptional case: normally we are not allowed to unhash a busy
1719  * directory. In this case, however, we can do it - no aliasing problems
1720  * due to the way we treat inodes.
1721  *
1722  * Rewrite the inode's ownerships here because the owning task may have
1723  * performed a setuid(), etc.
1724  *
1725  * Before the /proc/pid/status file was created the only way to read
1726  * the effective uid of a /process was to stat /proc/pid.  Reading
1727  * /proc/pid/status is slow enough that procps and other packages
1728  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1729  * made this apply to all per process world readable and executable
1730  * directories.
1731  */
1732 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1733 {
1734         struct inode *inode;
1735         struct task_struct *task;
1736         const struct cred *cred;
1737
1738         if (flags & LOOKUP_RCU)
1739                 return -ECHILD;
1740
1741         inode = d_inode(dentry);
1742         task = get_proc_task(inode);
1743
1744         if (task) {
1745                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1746                     task_dumpable(task)) {
1747                         rcu_read_lock();
1748                         cred = __task_cred(task);
1749                         inode->i_uid = cred->euid;
1750                         inode->i_gid = cred->egid;
1751                         rcu_read_unlock();
1752                 } else {
1753                         inode->i_uid = GLOBAL_ROOT_UID;
1754                         inode->i_gid = GLOBAL_ROOT_GID;
1755                 }
1756                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1757                 security_task_to_inode(task, inode);
1758                 put_task_struct(task);
1759                 return 1;
1760         }
1761         return 0;
1762 }
1763
1764 static inline bool proc_inode_is_dead(struct inode *inode)
1765 {
1766         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1767 }
1768
1769 int pid_delete_dentry(const struct dentry *dentry)
1770 {
1771         /* Is the task we represent dead?
1772          * If so, then don't put the dentry on the lru list,
1773          * kill it immediately.
1774          */
1775         return proc_inode_is_dead(d_inode(dentry));
1776 }
1777
1778 const struct dentry_operations pid_dentry_operations =
1779 {
1780         .d_revalidate   = pid_revalidate,
1781         .d_delete       = pid_delete_dentry,
1782 };
1783
1784 /* Lookups */
1785
1786 /*
1787  * Fill a directory entry.
1788  *
1789  * If possible create the dcache entry and derive our inode number and
1790  * file type from dcache entry.
1791  *
1792  * Since all of the proc inode numbers are dynamically generated, the inode
1793  * numbers do not exist until the inode is cache.  This means creating the
1794  * the dcache entry in readdir is necessary to keep the inode numbers
1795  * reported by readdir in sync with the inode numbers reported
1796  * by stat.
1797  */
1798 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1799         const char *name, int len,
1800         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1801 {
1802         struct dentry *child, *dir = file->f_path.dentry;
1803         struct qstr qname = QSTR_INIT(name, len);
1804         struct inode *inode;
1805         unsigned type;
1806         ino_t ino;
1807
1808         child = d_hash_and_lookup(dir, &qname);
1809         if (!child) {
1810                 child = d_alloc(dir, &qname);
1811                 if (!child)
1812                         goto end_instantiate;
1813                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1814                         dput(child);
1815                         goto end_instantiate;
1816                 }
1817         }
1818         inode = d_inode(child);
1819         ino = inode->i_ino;
1820         type = inode->i_mode >> 12;
1821         dput(child);
1822         return dir_emit(ctx, name, len, ino, type);
1823
1824 end_instantiate:
1825         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1826 }
1827
1828 /*
1829  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1830  * which represent vma start and end addresses.
1831  */
1832 static int dname_to_vma_addr(struct dentry *dentry,
1833                              unsigned long *start, unsigned long *end)
1834 {
1835         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1836                 return -EINVAL;
1837
1838         return 0;
1839 }
1840
1841 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1842 {
1843         unsigned long vm_start, vm_end;
1844         bool exact_vma_exists = false;
1845         struct mm_struct *mm = NULL;
1846         struct task_struct *task;
1847         const struct cred *cred;
1848         struct inode *inode;
1849         int status = 0;
1850
1851         if (flags & LOOKUP_RCU)
1852                 return -ECHILD;
1853
1854         inode = d_inode(dentry);
1855         task = get_proc_task(inode);
1856         if (!task)
1857                 goto out_notask;
1858
1859         mm = mm_access(task, PTRACE_MODE_READ);
1860         if (IS_ERR_OR_NULL(mm))
1861                 goto out;
1862
1863         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1864                 down_read(&mm->mmap_sem);
1865                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1866                 up_read(&mm->mmap_sem);
1867         }
1868
1869         mmput(mm);
1870
1871         if (exact_vma_exists) {
1872                 if (task_dumpable(task)) {
1873                         rcu_read_lock();
1874                         cred = __task_cred(task);
1875                         inode->i_uid = cred->euid;
1876                         inode->i_gid = cred->egid;
1877                         rcu_read_unlock();
1878                 } else {
1879                         inode->i_uid = GLOBAL_ROOT_UID;
1880                         inode->i_gid = GLOBAL_ROOT_GID;
1881                 }
1882                 security_task_to_inode(task, inode);
1883                 status = 1;
1884         }
1885
1886 out:
1887         put_task_struct(task);
1888
1889 out_notask:
1890         return status;
1891 }
1892
1893 static const struct dentry_operations tid_map_files_dentry_operations = {
1894         .d_revalidate   = map_files_d_revalidate,
1895         .d_delete       = pid_delete_dentry,
1896 };
1897
1898 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1899 {
1900         unsigned long vm_start, vm_end;
1901         struct vm_area_struct *vma;
1902         struct task_struct *task;
1903         struct mm_struct *mm;
1904         int rc;
1905
1906         rc = -ENOENT;
1907         task = get_proc_task(d_inode(dentry));
1908         if (!task)
1909                 goto out;
1910
1911         mm = get_task_mm(task);
1912         put_task_struct(task);
1913         if (!mm)
1914                 goto out;
1915
1916         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1917         if (rc)
1918                 goto out_mmput;
1919
1920         rc = -ENOENT;
1921         down_read(&mm->mmap_sem);
1922         vma = find_exact_vma(mm, vm_start, vm_end);
1923         if (vma && vma->vm_file) {
1924                 *path = vma->vm_file->f_path;
1925                 path_get(path);
1926                 rc = 0;
1927         }
1928         up_read(&mm->mmap_sem);
1929
1930 out_mmput:
1931         mmput(mm);
1932 out:
1933         return rc;
1934 }
1935
1936 struct map_files_info {
1937         fmode_t         mode;
1938         unsigned long   len;
1939         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1940 };
1941
1942 /*
1943  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1944  * symlinks may be used to bypass permissions on ancestor directories in the
1945  * path to the file in question.
1946  */
1947 static const char *
1948 proc_map_files_follow_link(struct dentry *dentry, void **cookie)
1949 {
1950         if (!capable(CAP_SYS_ADMIN))
1951                 return ERR_PTR(-EPERM);
1952
1953         return proc_pid_follow_link(dentry, NULL);
1954 }
1955
1956 /*
1957  * Identical to proc_pid_link_inode_operations except for follow_link()
1958  */
1959 static const struct inode_operations proc_map_files_link_inode_operations = {
1960         .readlink       = proc_pid_readlink,
1961         .follow_link    = proc_map_files_follow_link,
1962         .setattr        = proc_setattr,
1963 };
1964
1965 static int
1966 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1967                            struct task_struct *task, const void *ptr)
1968 {
1969         fmode_t mode = (fmode_t)(unsigned long)ptr;
1970         struct proc_inode *ei;
1971         struct inode *inode;
1972
1973         inode = proc_pid_make_inode(dir->i_sb, task);
1974         if (!inode)
1975                 return -ENOENT;
1976
1977         ei = PROC_I(inode);
1978         ei->op.proc_get_link = proc_map_files_get_link;
1979
1980         inode->i_op = &proc_map_files_link_inode_operations;
1981         inode->i_size = 64;
1982         inode->i_mode = S_IFLNK;
1983
1984         if (mode & FMODE_READ)
1985                 inode->i_mode |= S_IRUSR;
1986         if (mode & FMODE_WRITE)
1987                 inode->i_mode |= S_IWUSR;
1988
1989         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1990         d_add(dentry, inode);
1991
1992         return 0;
1993 }
1994
1995 static struct dentry *proc_map_files_lookup(struct inode *dir,
1996                 struct dentry *dentry, unsigned int flags)
1997 {
1998         unsigned long vm_start, vm_end;
1999         struct vm_area_struct *vma;
2000         struct task_struct *task;
2001         int result;
2002         struct mm_struct *mm;
2003
2004         result = -ENOENT;
2005         task = get_proc_task(dir);
2006         if (!task)
2007                 goto out;
2008
2009         result = -EACCES;
2010         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2011                 goto out_put_task;
2012
2013         result = -ENOENT;
2014         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2015                 goto out_put_task;
2016
2017         mm = get_task_mm(task);
2018         if (!mm)
2019                 goto out_put_task;
2020
2021         down_read(&mm->mmap_sem);
2022         vma = find_exact_vma(mm, vm_start, vm_end);
2023         if (!vma)
2024                 goto out_no_vma;
2025
2026         if (vma->vm_file)
2027                 result = proc_map_files_instantiate(dir, dentry, task,
2028                                 (void *)(unsigned long)vma->vm_file->f_mode);
2029
2030 out_no_vma:
2031         up_read(&mm->mmap_sem);
2032         mmput(mm);
2033 out_put_task:
2034         put_task_struct(task);
2035 out:
2036         return ERR_PTR(result);
2037 }
2038
2039 static const struct inode_operations proc_map_files_inode_operations = {
2040         .lookup         = proc_map_files_lookup,
2041         .permission     = proc_fd_permission,
2042         .setattr        = proc_setattr,
2043 };
2044
2045 static int
2046 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2047 {
2048         struct vm_area_struct *vma;
2049         struct task_struct *task;
2050         struct mm_struct *mm;
2051         unsigned long nr_files, pos, i;
2052         struct flex_array *fa = NULL;
2053         struct map_files_info info;
2054         struct map_files_info *p;
2055         int ret;
2056
2057         ret = -ENOENT;
2058         task = get_proc_task(file_inode(file));
2059         if (!task)
2060                 goto out;
2061
2062         ret = -EACCES;
2063         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2064                 goto out_put_task;
2065
2066         ret = 0;
2067         if (!dir_emit_dots(file, ctx))
2068                 goto out_put_task;
2069
2070         mm = get_task_mm(task);
2071         if (!mm)
2072                 goto out_put_task;
2073         down_read(&mm->mmap_sem);
2074
2075         nr_files = 0;
2076
2077         /*
2078          * We need two passes here:
2079          *
2080          *  1) Collect vmas of mapped files with mmap_sem taken
2081          *  2) Release mmap_sem and instantiate entries
2082          *
2083          * otherwise we get lockdep complained, since filldir()
2084          * routine might require mmap_sem taken in might_fault().
2085          */
2086
2087         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2088                 if (vma->vm_file && ++pos > ctx->pos)
2089                         nr_files++;
2090         }
2091
2092         if (nr_files) {
2093                 fa = flex_array_alloc(sizeof(info), nr_files,
2094                                         GFP_KERNEL);
2095                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2096                                                 GFP_KERNEL)) {
2097                         ret = -ENOMEM;
2098                         if (fa)
2099                                 flex_array_free(fa);
2100                         up_read(&mm->mmap_sem);
2101                         mmput(mm);
2102                         goto out_put_task;
2103                 }
2104                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2105                                 vma = vma->vm_next) {
2106                         if (!vma->vm_file)
2107                                 continue;
2108                         if (++pos <= ctx->pos)
2109                                 continue;
2110
2111                         info.mode = vma->vm_file->f_mode;
2112                         info.len = snprintf(info.name,
2113                                         sizeof(info.name), "%lx-%lx",
2114                                         vma->vm_start, vma->vm_end);
2115                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2116                                 BUG();
2117                 }
2118         }
2119         up_read(&mm->mmap_sem);
2120
2121         for (i = 0; i < nr_files; i++) {
2122                 p = flex_array_get(fa, i);
2123                 if (!proc_fill_cache(file, ctx,
2124                                       p->name, p->len,
2125                                       proc_map_files_instantiate,
2126                                       task,
2127                                       (void *)(unsigned long)p->mode))
2128                         break;
2129                 ctx->pos++;
2130         }
2131         if (fa)
2132                 flex_array_free(fa);
2133         mmput(mm);
2134
2135 out_put_task:
2136         put_task_struct(task);
2137 out:
2138         return ret;
2139 }
2140
2141 static const struct file_operations proc_map_files_operations = {
2142         .read           = generic_read_dir,
2143         .iterate        = proc_map_files_readdir,
2144         .llseek         = default_llseek,
2145 };
2146
2147 struct timers_private {
2148         struct pid *pid;
2149         struct task_struct *task;
2150         struct sighand_struct *sighand;
2151         struct pid_namespace *ns;
2152         unsigned long flags;
2153 };
2154
2155 static void *timers_start(struct seq_file *m, loff_t *pos)
2156 {
2157         struct timers_private *tp = m->private;
2158
2159         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2160         if (!tp->task)
2161                 return ERR_PTR(-ESRCH);
2162
2163         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2164         if (!tp->sighand)
2165                 return ERR_PTR(-ESRCH);
2166
2167         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2168 }
2169
2170 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2171 {
2172         struct timers_private *tp = m->private;
2173         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2174 }
2175
2176 static void timers_stop(struct seq_file *m, void *v)
2177 {
2178         struct timers_private *tp = m->private;
2179
2180         if (tp->sighand) {
2181                 unlock_task_sighand(tp->task, &tp->flags);
2182                 tp->sighand = NULL;
2183         }
2184
2185         if (tp->task) {
2186                 put_task_struct(tp->task);
2187                 tp->task = NULL;
2188         }
2189 }
2190
2191 static int show_timer(struct seq_file *m, void *v)
2192 {
2193         struct k_itimer *timer;
2194         struct timers_private *tp = m->private;
2195         int notify;
2196         static const char * const nstr[] = {
2197                 [SIGEV_SIGNAL] = "signal",
2198                 [SIGEV_NONE] = "none",
2199                 [SIGEV_THREAD] = "thread",
2200         };
2201
2202         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2203         notify = timer->it_sigev_notify;
2204
2205         seq_printf(m, "ID: %d\n", timer->it_id);
2206         seq_printf(m, "signal: %d/%p\n",
2207                    timer->sigq->info.si_signo,
2208                    timer->sigq->info.si_value.sival_ptr);
2209         seq_printf(m, "notify: %s/%s.%d\n",
2210                    nstr[notify & ~SIGEV_THREAD_ID],
2211                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2212                    pid_nr_ns(timer->it_pid, tp->ns));
2213         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2214
2215         return 0;
2216 }
2217
2218 static const struct seq_operations proc_timers_seq_ops = {
2219         .start  = timers_start,
2220         .next   = timers_next,
2221         .stop   = timers_stop,
2222         .show   = show_timer,
2223 };
2224
2225 static int proc_timers_open(struct inode *inode, struct file *file)
2226 {
2227         struct timers_private *tp;
2228
2229         tp = __seq_open_private(file, &proc_timers_seq_ops,
2230                         sizeof(struct timers_private));
2231         if (!tp)
2232                 return -ENOMEM;
2233
2234         tp->pid = proc_pid(inode);
2235         tp->ns = inode->i_sb->s_fs_info;
2236         return 0;
2237 }
2238
2239 static const struct file_operations proc_timers_operations = {
2240         .open           = proc_timers_open,
2241         .read           = seq_read,
2242         .llseek         = seq_lseek,
2243         .release        = seq_release_private,
2244 };
2245
2246 static int proc_pident_instantiate(struct inode *dir,
2247         struct dentry *dentry, struct task_struct *task, const void *ptr)
2248 {
2249         const struct pid_entry *p = ptr;
2250         struct inode *inode;
2251         struct proc_inode *ei;
2252
2253         inode = proc_pid_make_inode(dir->i_sb, task);
2254         if (!inode)
2255                 goto out;
2256
2257         ei = PROC_I(inode);
2258         inode->i_mode = p->mode;
2259         if (S_ISDIR(inode->i_mode))
2260                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2261         if (p->iop)
2262                 inode->i_op = p->iop;
2263         if (p->fop)
2264                 inode->i_fop = p->fop;
2265         ei->op = p->op;
2266         d_set_d_op(dentry, &pid_dentry_operations);
2267         d_add(dentry, inode);
2268         /* Close the race of the process dying before we return the dentry */
2269         if (pid_revalidate(dentry, 0))
2270                 return 0;
2271 out:
2272         return -ENOENT;
2273 }
2274
2275 static struct dentry *proc_pident_lookup(struct inode *dir,
2276                                          struct dentry *dentry,
2277                                          const struct pid_entry *ents,
2278                                          unsigned int nents)
2279 {
2280         int error;
2281         struct task_struct *task = get_proc_task(dir);
2282         const struct pid_entry *p, *last;
2283
2284         error = -ENOENT;
2285
2286         if (!task)
2287                 goto out_no_task;
2288
2289         /*
2290          * Yes, it does not scale. And it should not. Don't add
2291          * new entries into /proc/<tgid>/ without very good reasons.
2292          */
2293         last = &ents[nents - 1];
2294         for (p = ents; p <= last; p++) {
2295                 if (p->len != dentry->d_name.len)
2296                         continue;
2297                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2298                         break;
2299         }
2300         if (p > last)
2301                 goto out;
2302
2303         error = proc_pident_instantiate(dir, dentry, task, p);
2304 out:
2305         put_task_struct(task);
2306 out_no_task:
2307         return ERR_PTR(error);
2308 }
2309
2310 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2311                 const struct pid_entry *ents, unsigned int nents)
2312 {
2313         struct task_struct *task = get_proc_task(file_inode(file));
2314         const struct pid_entry *p;
2315
2316         if (!task)
2317                 return -ENOENT;
2318
2319         if (!dir_emit_dots(file, ctx))
2320                 goto out;
2321
2322         if (ctx->pos >= nents + 2)
2323                 goto out;
2324
2325         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2326                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2327                                 proc_pident_instantiate, task, p))
2328                         break;
2329                 ctx->pos++;
2330         }
2331 out:
2332         put_task_struct(task);
2333         return 0;
2334 }
2335
2336 #ifdef CONFIG_SECURITY
2337 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2338                                   size_t count, loff_t *ppos)
2339 {
2340         struct inode * inode = file_inode(file);
2341         char *p = NULL;
2342         ssize_t length;
2343         struct task_struct *task = get_proc_task(inode);
2344
2345         if (!task)
2346                 return -ESRCH;
2347
2348         length = security_getprocattr(task,
2349                                       (char*)file->f_path.dentry->d_name.name,
2350                                       &p);
2351         put_task_struct(task);
2352         if (length > 0)
2353                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2354         kfree(p);
2355         return length;
2356 }
2357
2358 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2359                                    size_t count, loff_t *ppos)
2360 {
2361         struct inode * inode = file_inode(file);
2362         char *page;
2363         ssize_t length;
2364         struct task_struct *task = get_proc_task(inode);
2365
2366         length = -ESRCH;
2367         if (!task)
2368                 goto out_no_task;
2369         if (count > PAGE_SIZE)
2370                 count = PAGE_SIZE;
2371
2372         /* No partial writes. */
2373         length = -EINVAL;
2374         if (*ppos != 0)
2375                 goto out;
2376
2377         length = -ENOMEM;
2378         page = (char*)__get_free_page(GFP_TEMPORARY);
2379         if (!page)
2380                 goto out;
2381
2382         length = -EFAULT;
2383         if (copy_from_user(page, buf, count))
2384                 goto out_free;
2385
2386         /* Guard against adverse ptrace interaction */
2387         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2388         if (length < 0)
2389                 goto out_free;
2390
2391         length = security_setprocattr(task,
2392                                       (char*)file->f_path.dentry->d_name.name,
2393                                       (void*)page, count);
2394         mutex_unlock(&task->signal->cred_guard_mutex);
2395 out_free:
2396         free_page((unsigned long) page);
2397 out:
2398         put_task_struct(task);
2399 out_no_task:
2400         return length;
2401 }
2402
2403 static const struct file_operations proc_pid_attr_operations = {
2404         .read           = proc_pid_attr_read,
2405         .write          = proc_pid_attr_write,
2406         .llseek         = generic_file_llseek,
2407 };
2408
2409 static const struct pid_entry attr_dir_stuff[] = {
2410         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2411         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2412         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2413         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2414         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2415         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2416 };
2417
2418 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2419 {
2420         return proc_pident_readdir(file, ctx,
2421                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2422 }
2423
2424 static const struct file_operations proc_attr_dir_operations = {
2425         .read           = generic_read_dir,
2426         .iterate        = proc_attr_dir_readdir,
2427         .llseek         = default_llseek,
2428 };
2429
2430 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2431                                 struct dentry *dentry, unsigned int flags)
2432 {
2433         return proc_pident_lookup(dir, dentry,
2434                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2435 }
2436
2437 static const struct inode_operations proc_attr_dir_inode_operations = {
2438         .lookup         = proc_attr_dir_lookup,
2439         .getattr        = pid_getattr,
2440         .setattr        = proc_setattr,
2441 };
2442
2443 #endif
2444
2445 #ifdef CONFIG_ELF_CORE
2446 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2447                                          size_t count, loff_t *ppos)
2448 {
2449         struct task_struct *task = get_proc_task(file_inode(file));
2450         struct mm_struct *mm;
2451         char buffer[PROC_NUMBUF];
2452         size_t len;
2453         int ret;
2454
2455         if (!task)
2456                 return -ESRCH;
2457
2458         ret = 0;
2459         mm = get_task_mm(task);
2460         if (mm) {
2461                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2462                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2463                                 MMF_DUMP_FILTER_SHIFT));
2464                 mmput(mm);
2465                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2466         }
2467
2468         put_task_struct(task);
2469
2470         return ret;
2471 }
2472
2473 static ssize_t proc_coredump_filter_write(struct file *file,
2474                                           const char __user *buf,
2475                                           size_t count,
2476                                           loff_t *ppos)
2477 {
2478         struct task_struct *task;
2479         struct mm_struct *mm;
2480         unsigned int val;
2481         int ret;
2482         int i;
2483         unsigned long mask;
2484
2485         ret = kstrtouint_from_user(buf, count, 0, &val);
2486         if (ret < 0)
2487                 return ret;
2488
2489         ret = -ESRCH;
2490         task = get_proc_task(file_inode(file));
2491         if (!task)
2492                 goto out_no_task;
2493
2494         mm = get_task_mm(task);
2495         if (!mm)
2496                 goto out_no_mm;
2497
2498         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2499                 if (val & mask)
2500                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2501                 else
2502                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2503         }
2504
2505         mmput(mm);
2506  out_no_mm:
2507         put_task_struct(task);
2508  out_no_task:
2509         if (ret < 0)
2510                 return ret;
2511         return count;
2512 }
2513
2514 static const struct file_operations proc_coredump_filter_operations = {
2515         .read           = proc_coredump_filter_read,
2516         .write          = proc_coredump_filter_write,
2517         .llseek         = generic_file_llseek,
2518 };
2519 #endif
2520
2521 #ifdef CONFIG_TASK_IO_ACCOUNTING
2522 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2523 {
2524         struct task_io_accounting acct = task->ioac;
2525         unsigned long flags;
2526         int result;
2527
2528         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2529         if (result)
2530                 return result;
2531
2532         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2533                 result = -EACCES;
2534                 goto out_unlock;
2535         }
2536
2537         if (whole && lock_task_sighand(task, &flags)) {
2538                 struct task_struct *t = task;
2539
2540                 task_io_accounting_add(&acct, &task->signal->ioac);
2541                 while_each_thread(task, t)
2542                         task_io_accounting_add(&acct, &t->ioac);
2543
2544                 unlock_task_sighand(task, &flags);
2545         }
2546         seq_printf(m,
2547                    "rchar: %llu\n"
2548                    "wchar: %llu\n"
2549                    "syscr: %llu\n"
2550                    "syscw: %llu\n"
2551                    "read_bytes: %llu\n"
2552                    "write_bytes: %llu\n"
2553                    "cancelled_write_bytes: %llu\n",
2554                    (unsigned long long)acct.rchar,
2555                    (unsigned long long)acct.wchar,
2556                    (unsigned long long)acct.syscr,
2557                    (unsigned long long)acct.syscw,
2558                    (unsigned long long)acct.read_bytes,
2559                    (unsigned long long)acct.write_bytes,
2560                    (unsigned long long)acct.cancelled_write_bytes);
2561         result = 0;
2562
2563 out_unlock:
2564         mutex_unlock(&task->signal->cred_guard_mutex);
2565         return result;
2566 }
2567
2568 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2569                                   struct pid *pid, struct task_struct *task)
2570 {
2571         return do_io_accounting(task, m, 0);
2572 }
2573
2574 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2575                                    struct pid *pid, struct task_struct *task)
2576 {
2577         return do_io_accounting(task, m, 1);
2578 }
2579 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2580
2581 #ifdef CONFIG_USER_NS
2582 static int proc_id_map_open(struct inode *inode, struct file *file,
2583         const struct seq_operations *seq_ops)
2584 {
2585         struct user_namespace *ns = NULL;
2586         struct task_struct *task;
2587         struct seq_file *seq;
2588         int ret = -EINVAL;
2589
2590         task = get_proc_task(inode);
2591         if (task) {
2592                 rcu_read_lock();
2593                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2594                 rcu_read_unlock();
2595                 put_task_struct(task);
2596         }
2597         if (!ns)
2598                 goto err;
2599
2600         ret = seq_open(file, seq_ops);
2601         if (ret)
2602                 goto err_put_ns;
2603
2604         seq = file->private_data;
2605         seq->private = ns;
2606
2607         return 0;
2608 err_put_ns:
2609         put_user_ns(ns);
2610 err:
2611         return ret;
2612 }
2613
2614 static int proc_id_map_release(struct inode *inode, struct file *file)
2615 {
2616         struct seq_file *seq = file->private_data;
2617         struct user_namespace *ns = seq->private;
2618         put_user_ns(ns);
2619         return seq_release(inode, file);
2620 }
2621
2622 static int proc_uid_map_open(struct inode *inode, struct file *file)
2623 {
2624         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2625 }
2626
2627 static int proc_gid_map_open(struct inode *inode, struct file *file)
2628 {
2629         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2630 }
2631
2632 static int proc_projid_map_open(struct inode *inode, struct file *file)
2633 {
2634         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2635 }
2636
2637 static const struct file_operations proc_uid_map_operations = {
2638         .open           = proc_uid_map_open,
2639         .write          = proc_uid_map_write,
2640         .read           = seq_read,
2641         .llseek         = seq_lseek,
2642         .release        = proc_id_map_release,
2643 };
2644
2645 static const struct file_operations proc_gid_map_operations = {
2646         .open           = proc_gid_map_open,
2647         .write          = proc_gid_map_write,
2648         .read           = seq_read,
2649         .llseek         = seq_lseek,
2650         .release        = proc_id_map_release,
2651 };
2652
2653 static const struct file_operations proc_projid_map_operations = {
2654         .open           = proc_projid_map_open,
2655         .write          = proc_projid_map_write,
2656         .read           = seq_read,
2657         .llseek         = seq_lseek,
2658         .release        = proc_id_map_release,
2659 };
2660
2661 static int proc_setgroups_open(struct inode *inode, struct file *file)
2662 {
2663         struct user_namespace *ns = NULL;
2664         struct task_struct *task;
2665         int ret;
2666
2667         ret = -ESRCH;
2668         task = get_proc_task(inode);
2669         if (task) {
2670                 rcu_read_lock();
2671                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2672                 rcu_read_unlock();
2673                 put_task_struct(task);
2674         }
2675         if (!ns)
2676                 goto err;
2677
2678         if (file->f_mode & FMODE_WRITE) {
2679                 ret = -EACCES;
2680                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2681                         goto err_put_ns;
2682         }
2683
2684         ret = single_open(file, &proc_setgroups_show, ns);
2685         if (ret)
2686                 goto err_put_ns;
2687
2688         return 0;
2689 err_put_ns:
2690         put_user_ns(ns);
2691 err:
2692         return ret;
2693 }
2694
2695 static int proc_setgroups_release(struct inode *inode, struct file *file)
2696 {
2697         struct seq_file *seq = file->private_data;
2698         struct user_namespace *ns = seq->private;
2699         int ret = single_release(inode, file);
2700         put_user_ns(ns);
2701         return ret;
2702 }
2703
2704 static const struct file_operations proc_setgroups_operations = {
2705         .open           = proc_setgroups_open,
2706         .write          = proc_setgroups_write,
2707         .read           = seq_read,
2708         .llseek         = seq_lseek,
2709         .release        = proc_setgroups_release,
2710 };
2711 #endif /* CONFIG_USER_NS */
2712
2713 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2714                                 struct pid *pid, struct task_struct *task)
2715 {
2716         int err = lock_trace(task);
2717         if (!err) {
2718                 seq_printf(m, "%08x\n", task->personality);
2719                 unlock_trace(task);
2720         }
2721         return err;
2722 }
2723
2724 /*
2725  * Thread groups
2726  */
2727 static const struct file_operations proc_task_operations;
2728 static const struct inode_operations proc_task_inode_operations;
2729
2730 static const struct pid_entry tgid_base_stuff[] = {
2731         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2732         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2733         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2734         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2735         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2736 #ifdef CONFIG_NET
2737         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2738 #endif
2739         REG("environ",    S_IRUSR, proc_environ_operations),
2740         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2741         ONE("status",     S_IRUGO, proc_pid_status),
2742         ONE("personality", S_IRUSR, proc_pid_personality),
2743         ONE("limits",     S_IRUGO, proc_pid_limits),
2744 #ifdef CONFIG_SCHED_DEBUG
2745         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2746 #endif
2747 #ifdef CONFIG_SCHED_AUTOGROUP
2748         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2749 #endif
2750         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2751 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2752         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2753 #endif
2754         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2755         ONE("stat",       S_IRUGO, proc_tgid_stat),
2756         ONE("statm",      S_IRUGO, proc_pid_statm),
2757         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2758 #ifdef CONFIG_NUMA
2759         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2760 #endif
2761         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2762         LNK("cwd",        proc_cwd_link),
2763         LNK("root",       proc_root_link),
2764         LNK("exe",        proc_exe_link),
2765         REG("mounts",     S_IRUGO, proc_mounts_operations),
2766         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2767         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2768 #ifdef CONFIG_PROC_PAGE_MONITOR
2769         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2770         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2771         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2772 #endif
2773 #ifdef CONFIG_SECURITY
2774         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2775 #endif
2776 #ifdef CONFIG_KALLSYMS
2777         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2778 #endif
2779 #ifdef CONFIG_STACKTRACE
2780         ONE("stack",      S_IRUSR, proc_pid_stack),
2781 #endif
2782 #ifdef CONFIG_SCHED_INFO
2783         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2784 #endif
2785 #ifdef CONFIG_LATENCYTOP
2786         REG("latency",  S_IRUGO, proc_lstats_operations),
2787 #endif
2788 #ifdef CONFIG_PROC_PID_CPUSET
2789         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2790 #endif
2791 #ifdef CONFIG_CGROUPS
2792         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2793 #endif
2794         ONE("oom_score",  S_IRUGO, proc_oom_score),
2795         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2796         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2797 #ifdef CONFIG_AUDITSYSCALL
2798         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2799         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2800 #endif
2801 #ifdef CONFIG_FAULT_INJECTION
2802         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2803 #endif
2804 #ifdef CONFIG_ELF_CORE
2805         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2806 #endif
2807 #ifdef CONFIG_TASK_IO_ACCOUNTING
2808         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2809 #endif
2810 #ifdef CONFIG_HARDWALL
2811         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2812 #endif
2813 #ifdef CONFIG_USER_NS
2814         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2815         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2816         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2817         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2818 #endif
2819 #ifdef CONFIG_CHECKPOINT_RESTORE
2820         REG("timers",     S_IRUGO, proc_timers_operations),
2821 #endif
2822 };
2823
2824 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2825 {
2826         return proc_pident_readdir(file, ctx,
2827                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2828 }
2829
2830 static const struct file_operations proc_tgid_base_operations = {
2831         .read           = generic_read_dir,
2832         .iterate        = proc_tgid_base_readdir,
2833         .llseek         = default_llseek,
2834 };
2835
2836 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2837 {
2838         return proc_pident_lookup(dir, dentry,
2839                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2840 }
2841
2842 static const struct inode_operations proc_tgid_base_inode_operations = {
2843         .lookup         = proc_tgid_base_lookup,
2844         .getattr        = pid_getattr,
2845         .setattr        = proc_setattr,
2846         .permission     = proc_pid_permission,
2847 };
2848
2849 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2850 {
2851         struct dentry *dentry, *leader, *dir;
2852         char buf[PROC_NUMBUF];
2853         struct qstr name;
2854
2855         name.name = buf;
2856         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2857         /* no ->d_hash() rejects on procfs */
2858         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2859         if (dentry) {
2860                 d_invalidate(dentry);
2861                 dput(dentry);
2862         }
2863
2864         if (pid == tgid)
2865                 return;
2866
2867         name.name = buf;
2868         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2869         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2870         if (!leader)
2871                 goto out;
2872
2873         name.name = "task";
2874         name.len = strlen(name.name);
2875         dir = d_hash_and_lookup(leader, &name);
2876         if (!dir)
2877                 goto out_put_leader;
2878
2879         name.name = buf;
2880         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2881         dentry = d_hash_and_lookup(dir, &name);
2882         if (dentry) {
2883                 d_invalidate(dentry);
2884                 dput(dentry);
2885         }
2886
2887         dput(dir);
2888 out_put_leader:
2889         dput(leader);
2890 out:
2891         return;
2892 }
2893
2894 /**
2895  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2896  * @task: task that should be flushed.
2897  *
2898  * When flushing dentries from proc, one needs to flush them from global
2899  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2900  * in. This call is supposed to do all of this job.
2901  *
2902  * Looks in the dcache for
2903  * /proc/@pid
2904  * /proc/@tgid/task/@pid
2905  * if either directory is present flushes it and all of it'ts children
2906  * from the dcache.
2907  *
2908  * It is safe and reasonable to cache /proc entries for a task until
2909  * that task exits.  After that they just clog up the dcache with
2910  * useless entries, possibly causing useful dcache entries to be
2911  * flushed instead.  This routine is proved to flush those useless
2912  * dcache entries at process exit time.
2913  *
2914  * NOTE: This routine is just an optimization so it does not guarantee
2915  *       that no dcache entries will exist at process exit time it
2916  *       just makes it very unlikely that any will persist.
2917  */
2918
2919 void proc_flush_task(struct task_struct *task)
2920 {
2921         int i;
2922         struct pid *pid, *tgid;
2923         struct upid *upid;
2924
2925         pid = task_pid(task);
2926         tgid = task_tgid(task);
2927
2928         for (i = 0; i <= pid->level; i++) {
2929                 upid = &pid->numbers[i];
2930                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2931                                         tgid->numbers[i].nr);
2932         }
2933 }
2934
2935 static int proc_pid_instantiate(struct inode *dir,
2936                                    struct dentry * dentry,
2937                                    struct task_struct *task, const void *ptr)
2938 {
2939         struct inode *inode;
2940
2941         inode = proc_pid_make_inode(dir->i_sb, task);
2942         if (!inode)
2943                 goto out;
2944
2945         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2946         inode->i_op = &proc_tgid_base_inode_operations;
2947         inode->i_fop = &proc_tgid_base_operations;
2948         inode->i_flags|=S_IMMUTABLE;
2949
2950         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2951                                                   ARRAY_SIZE(tgid_base_stuff)));
2952
2953         d_set_d_op(dentry, &pid_dentry_operations);
2954
2955         d_add(dentry, inode);
2956         /* Close the race of the process dying before we return the dentry */
2957         if (pid_revalidate(dentry, 0))
2958                 return 0;
2959 out:
2960         return -ENOENT;
2961 }
2962
2963 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2964 {
2965         int result = -ENOENT;
2966         struct task_struct *task;
2967         unsigned tgid;
2968         struct pid_namespace *ns;
2969
2970         tgid = name_to_int(&dentry->d_name);
2971         if (tgid == ~0U)
2972                 goto out;
2973
2974         ns = dentry->d_sb->s_fs_info;
2975         rcu_read_lock();
2976         task = find_task_by_pid_ns(tgid, ns);
2977         if (task)
2978                 get_task_struct(task);
2979         rcu_read_unlock();
2980         if (!task)
2981                 goto out;
2982
2983         result = proc_pid_instantiate(dir, dentry, task, NULL);
2984         put_task_struct(task);
2985 out:
2986         return ERR_PTR(result);
2987 }
2988
2989 /*
2990  * Find the first task with tgid >= tgid
2991  *
2992  */
2993 struct tgid_iter {
2994         unsigned int tgid;
2995         struct task_struct *task;
2996 };
2997 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2998 {
2999         struct pid *pid;
3000
3001         if (iter.task)
3002                 put_task_struct(iter.task);
3003         rcu_read_lock();
3004 retry:
3005         iter.task = NULL;
3006         pid = find_ge_pid(iter.tgid, ns);
3007         if (pid) {
3008                 iter.tgid = pid_nr_ns(pid, ns);
3009                 iter.task = pid_task(pid, PIDTYPE_PID);
3010                 /* What we to know is if the pid we have find is the
3011                  * pid of a thread_group_leader.  Testing for task
3012                  * being a thread_group_leader is the obvious thing
3013                  * todo but there is a window when it fails, due to
3014                  * the pid transfer logic in de_thread.
3015                  *
3016                  * So we perform the straight forward test of seeing
3017                  * if the pid we have found is the pid of a thread
3018                  * group leader, and don't worry if the task we have
3019                  * found doesn't happen to be a thread group leader.
3020                  * As we don't care in the case of readdir.
3021                  */
3022                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3023                         iter.tgid += 1;
3024                         goto retry;
3025                 }
3026                 get_task_struct(iter.task);
3027         }
3028         rcu_read_unlock();
3029         return iter;
3030 }
3031
3032 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3033
3034 /* for the /proc/ directory itself, after non-process stuff has been done */
3035 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3036 {
3037         struct tgid_iter iter;
3038         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3039         loff_t pos = ctx->pos;
3040
3041         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3042                 return 0;
3043
3044         if (pos == TGID_OFFSET - 2) {
3045                 struct inode *inode = d_inode(ns->proc_self);
3046                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3047                         return 0;
3048                 ctx->pos = pos = pos + 1;
3049         }
3050         if (pos == TGID_OFFSET - 1) {
3051                 struct inode *inode = d_inode(ns->proc_thread_self);
3052                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3053                         return 0;
3054                 ctx->pos = pos = pos + 1;
3055         }
3056         iter.tgid = pos - TGID_OFFSET;
3057         iter.task = NULL;
3058         for (iter = next_tgid(ns, iter);
3059              iter.task;
3060              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3061                 char name[PROC_NUMBUF];
3062                 int len;
3063                 if (!has_pid_permissions(ns, iter.task, 2))
3064                         continue;
3065
3066                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3067                 ctx->pos = iter.tgid + TGID_OFFSET;
3068                 if (!proc_fill_cache(file, ctx, name, len,
3069                                      proc_pid_instantiate, iter.task, NULL)) {
3070                         put_task_struct(iter.task);
3071                         return 0;
3072                 }
3073         }
3074         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3075         return 0;
3076 }
3077
3078 /*
3079  * Tasks
3080  */
3081 static const struct pid_entry tid_base_stuff[] = {
3082         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3083         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3084         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3085 #ifdef CONFIG_NET
3086         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3087 #endif
3088         REG("environ",   S_IRUSR, proc_environ_operations),
3089         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3090         ONE("status",    S_IRUGO, proc_pid_status),
3091         ONE("personality", S_IRUSR, proc_pid_personality),
3092         ONE("limits",    S_IRUGO, proc_pid_limits),
3093 #ifdef CONFIG_SCHED_DEBUG
3094         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3095 #endif
3096         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3097 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3098         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3099 #endif
3100         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3101         ONE("stat",      S_IRUGO, proc_tid_stat),
3102         ONE("statm",     S_IRUGO, proc_pid_statm),
3103         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3104 #ifdef CONFIG_PROC_CHILDREN
3105         REG("children",  S_IRUGO, proc_tid_children_operations),
3106 #endif
3107 #ifdef CONFIG_NUMA
3108         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3109 #endif
3110         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3111         LNK("cwd",       proc_cwd_link),
3112         LNK("root",      proc_root_link),
3113         LNK("exe",       proc_exe_link),
3114         REG("mounts",    S_IRUGO, proc_mounts_operations),
3115         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3116 #ifdef CONFIG_PROC_PAGE_MONITOR
3117         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3118         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3119         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3120 #endif
3121 #ifdef CONFIG_SECURITY
3122         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3123 #endif
3124 #ifdef CONFIG_KALLSYMS
3125         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3126 #endif
3127 #ifdef CONFIG_STACKTRACE
3128         ONE("stack",      S_IRUSR, proc_pid_stack),
3129 #endif
3130 #ifdef CONFIG_SCHED_INFO
3131         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3132 #endif
3133 #ifdef CONFIG_LATENCYTOP
3134         REG("latency",  S_IRUGO, proc_lstats_operations),
3135 #endif
3136 #ifdef CONFIG_PROC_PID_CPUSET
3137         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3138 #endif
3139 #ifdef CONFIG_CGROUPS
3140         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3141 #endif
3142         ONE("oom_score", S_IRUGO, proc_oom_score),
3143         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3144         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3145 #ifdef CONFIG_AUDITSYSCALL
3146         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3147         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3148 #endif
3149 #ifdef CONFIG_FAULT_INJECTION
3150         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3151 #endif
3152 #ifdef CONFIG_TASK_IO_ACCOUNTING
3153         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3154 #endif
3155 #ifdef CONFIG_HARDWALL
3156         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3157 #endif
3158 #ifdef CONFIG_USER_NS
3159         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3160         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3161         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3162         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3163 #endif
3164 };
3165
3166 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3167 {
3168         return proc_pident_readdir(file, ctx,
3169                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3170 }
3171
3172 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3173 {
3174         return proc_pident_lookup(dir, dentry,
3175                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3176 }
3177
3178 static const struct file_operations proc_tid_base_operations = {
3179         .read           = generic_read_dir,
3180         .iterate        = proc_tid_base_readdir,
3181         .llseek         = default_llseek,
3182 };
3183
3184 static const struct inode_operations proc_tid_base_inode_operations = {
3185         .lookup         = proc_tid_base_lookup,
3186         .getattr        = pid_getattr,
3187         .setattr        = proc_setattr,
3188 };
3189
3190 static int proc_task_instantiate(struct inode *dir,
3191         struct dentry *dentry, struct task_struct *task, const void *ptr)
3192 {
3193         struct inode *inode;
3194         inode = proc_pid_make_inode(dir->i_sb, task);
3195
3196         if (!inode)
3197                 goto out;
3198         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3199         inode->i_op = &proc_tid_base_inode_operations;
3200         inode->i_fop = &proc_tid_base_operations;
3201         inode->i_flags|=S_IMMUTABLE;
3202
3203         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3204                                                   ARRAY_SIZE(tid_base_stuff)));
3205
3206         d_set_d_op(dentry, &pid_dentry_operations);
3207
3208         d_add(dentry, inode);
3209         /* Close the race of the process dying before we return the dentry */
3210         if (pid_revalidate(dentry, 0))
3211                 return 0;
3212 out:
3213         return -ENOENT;
3214 }
3215
3216 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3217 {
3218         int result = -ENOENT;
3219         struct task_struct *task;
3220         struct task_struct *leader = get_proc_task(dir);
3221         unsigned tid;
3222         struct pid_namespace *ns;
3223
3224         if (!leader)
3225                 goto out_no_task;
3226
3227         tid = name_to_int(&dentry->d_name);
3228         if (tid == ~0U)
3229                 goto out;
3230
3231         ns = dentry->d_sb->s_fs_info;
3232         rcu_read_lock();
3233         task = find_task_by_pid_ns(tid, ns);
3234         if (task)
3235                 get_task_struct(task);
3236         rcu_read_unlock();
3237         if (!task)
3238                 goto out;
3239         if (!same_thread_group(leader, task))
3240                 goto out_drop_task;
3241
3242         result = proc_task_instantiate(dir, dentry, task, NULL);
3243 out_drop_task:
3244         put_task_struct(task);
3245 out:
3246         put_task_struct(leader);
3247 out_no_task:
3248         return ERR_PTR(result);
3249 }
3250
3251 /*
3252  * Find the first tid of a thread group to return to user space.
3253  *
3254  * Usually this is just the thread group leader, but if the users
3255  * buffer was too small or there was a seek into the middle of the
3256  * directory we have more work todo.
3257  *
3258  * In the case of a short read we start with find_task_by_pid.
3259  *
3260  * In the case of a seek we start with the leader and walk nr
3261  * threads past it.
3262  */
3263 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3264                                         struct pid_namespace *ns)
3265 {
3266         struct task_struct *pos, *task;
3267         unsigned long nr = f_pos;
3268
3269         if (nr != f_pos)        /* 32bit overflow? */
3270                 return NULL;
3271
3272         rcu_read_lock();
3273         task = pid_task(pid, PIDTYPE_PID);
3274         if (!task)
3275                 goto fail;
3276
3277         /* Attempt to start with the tid of a thread */
3278         if (tid && nr) {
3279                 pos = find_task_by_pid_ns(tid, ns);
3280                 if (pos && same_thread_group(pos, task))
3281                         goto found;
3282         }
3283
3284         /* If nr exceeds the number of threads there is nothing todo */
3285         if (nr >= get_nr_threads(task))
3286                 goto fail;
3287
3288         /* If we haven't found our starting place yet start
3289          * with the leader and walk nr threads forward.
3290          */
3291         pos = task = task->group_leader;
3292         do {
3293                 if (!nr--)
3294                         goto found;
3295         } while_each_thread(task, pos);
3296 fail:
3297         pos = NULL;
3298         goto out;
3299 found:
3300         get_task_struct(pos);
3301 out:
3302         rcu_read_unlock();
3303         return pos;
3304 }
3305
3306 /*
3307  * Find the next thread in the thread list.
3308  * Return NULL if there is an error or no next thread.
3309  *
3310  * The reference to the input task_struct is released.
3311  */
3312 static struct task_struct *next_tid(struct task_struct *start)
3313 {
3314         struct task_struct *pos = NULL;
3315         rcu_read_lock();
3316         if (pid_alive(start)) {
3317                 pos = next_thread(start);
3318                 if (thread_group_leader(pos))
3319                         pos = NULL;
3320                 else
3321                         get_task_struct(pos);
3322         }
3323         rcu_read_unlock();
3324         put_task_struct(start);
3325         return pos;
3326 }
3327
3328 /* for the /proc/TGID/task/ directories */
3329 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3330 {
3331         struct inode *inode = file_inode(file);
3332         struct task_struct *task;
3333         struct pid_namespace *ns;
3334         int tid;
3335
3336         if (proc_inode_is_dead(inode))
3337                 return -ENOENT;
3338
3339         if (!dir_emit_dots(file, ctx))
3340                 return 0;
3341
3342         /* f_version caches the tgid value that the last readdir call couldn't
3343          * return. lseek aka telldir automagically resets f_version to 0.
3344          */
3345         ns = inode->i_sb->s_fs_info;
3346         tid = (int)file->f_version;
3347         file->f_version = 0;
3348         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3349              task;
3350              task = next_tid(task), ctx->pos++) {
3351                 char name[PROC_NUMBUF];
3352                 int len;
3353                 tid = task_pid_nr_ns(task, ns);
3354                 len = snprintf(name, sizeof(name), "%d", tid);
3355                 if (!proc_fill_cache(file, ctx, name, len,
3356                                 proc_task_instantiate, task, NULL)) {
3357                         /* returning this tgid failed, save it as the first
3358                          * pid for the next readir call */
3359                         file->f_version = (u64)tid;
3360                         put_task_struct(task);
3361                         break;
3362                 }
3363         }
3364
3365         return 0;
3366 }
3367
3368 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3369 {
3370         struct inode *inode = d_inode(dentry);
3371         struct task_struct *p = get_proc_task(inode);
3372         generic_fillattr(inode, stat);
3373
3374         if (p) {
3375                 stat->nlink += get_nr_threads(p);
3376                 put_task_struct(p);
3377         }
3378
3379         return 0;
3380 }
3381
3382 static const struct inode_operations proc_task_inode_operations = {
3383         .lookup         = proc_task_lookup,
3384         .getattr        = proc_task_getattr,
3385         .setattr        = proc_setattr,
3386         .permission     = proc_pid_permission,
3387 };
3388
3389 static const struct file_operations proc_task_operations = {
3390         .read           = generic_read_dir,
3391         .iterate        = proc_task_readdir,
3392         .llseek         = default_llseek,
3393 };