fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         const char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define ONE(NAME, MODE, show)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_single_file_operations,     \
 136                 { .proc_show = show } )
 137
 138 /*
 139  * Count the number of hardlinks for the pid_entry table, excluding the .
 140  * and .. links.
 141  */
 142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 143         unsigned int n)
 144 {
 145         unsigned int i;
 146         unsigned int count;
 147
 148         count = 0;
 149         for (i = 0; i < n; ++i) {
 150                 if (S_ISDIR(entries[i].mode))
 151                         ++count;
 152         }
 153
 154         return count;
 155 }
 156
 157 static int get_task_root(struct task_struct *task, struct path *root)
 158 {
 159         int result = -ENOENT;
 160
 161         task_lock(task);
 162         if (task->fs) {
 163                 get_fs_root(task->fs, root);
 164                 result = 0;
 165         }
 166         task_unlock(task);
 167         return result;
 168 }
 169
 170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 171 {
 172         struct task_struct *task = get_proc_task(d_inode(dentry));
 173         int result = -ENOENT;
 174
 175         if (task) {
 176                 task_lock(task);
 177                 if (task->fs) {
 178                         get_fs_pwd(task->fs, path);
 179                         result = 0;
 180                 }
 181                 task_unlock(task);
 182                 put_task_struct(task);
 183         }
 184         return result;
 185 }
 186
 187 static int proc_root_link(struct dentry *dentry, struct path *path)
 188 {
 189         struct task_struct *task = get_proc_task(d_inode(dentry));
 190         int result = -ENOENT;
 191
 192         if (task) {
 193                 result = get_task_root(task, path);
 194                 put_task_struct(task);
 195         }
 196         return result;
 197 }
 198
 199 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 200                                      size_t _count, loff_t *pos)
 201 {
 202         struct task_struct *tsk;
 203         struct mm_struct *mm;
 204         char *page;
 205         unsigned long count = _count;
 206         unsigned long arg_start, arg_end, env_start, env_end;
 207         unsigned long len1, len2, len;
 208         unsigned long p;
 209         char c;
 210         ssize_t rv;
 211
 212         BUG_ON(*pos < 0);
 213
 214         tsk = get_proc_task(file_inode(file));
 215         if (!tsk)
 216                 return -ESRCH;
 217         mm = get_task_mm(tsk);
 218         put_task_struct(tsk);
 219         if (!mm)
 220                 return 0;
 221         /* Check if process spawned far enough to have cmdline. */
 222         if (!mm->env_end) {
 223                 rv = 0;
 224                 goto out_mmput;
 225         }
 226
 227         page = (char *)__get_free_page(GFP_TEMPORARY);
 228         if (!page) {
 229                 rv = -ENOMEM;
 230                 goto out_mmput;
 231         }
 232
 233         down_read(&mm->mmap_sem);
 234         arg_start = mm->arg_start;
 235         arg_end = mm->arg_end;
 236         env_start = mm->env_start;
 237         env_end = mm->env_end;
 238         up_read(&mm->mmap_sem);
 239
 240         BUG_ON(arg_start > arg_end);
 241         BUG_ON(env_start > env_end);
 242
 243         len1 = arg_end - arg_start;
 244         len2 = env_end - env_start;
 245
 246         /*
 247          * Inherently racy -- command line shares address space
 248          * with code and data.
 249          */
 250         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 251         if (rv <= 0)
 252                 goto out_free_page;
 253
 254         rv = 0;
 255
 256         if (c == '\0') {
 257                 /* Command line (set of strings) occupies whole ARGV. */
 258                 if (len1 <= *pos)
 259                         goto out_free_page;
 260
 261                 p = arg_start + *pos;
 262                 len = len1 - *pos;
 263                 while (count > 0 && len > 0) {
 264                         unsigned int _count;
 265                         int nr_read;
 266
 267                         _count = min3(count, len, PAGE_SIZE);
 268                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 269                         if (nr_read < 0)
 270                                 rv = nr_read;
 271                         if (nr_read <= 0)
 272                                 goto out_free_page;
 273
 274                         if (copy_to_user(buf, page, nr_read)) {
 275                                 rv = -EFAULT;
 276                                 goto out_free_page;
 277                         }
 278
 279                         p       += nr_read;
 280                         len     -= nr_read;
 281                         buf     += nr_read;
 282                         count   -= nr_read;
 283                         rv      += nr_read;
 284                 }
 285         } else {
 286                 /*
 287                  * Command line (1 string) occupies ARGV and maybe
 288                  * extends into ENVP.
 289                  */
 290                 if (len1 + len2 <= *pos)
 291                         goto skip_argv_envp;
 292                 if (len1 <= *pos)
 293                         goto skip_argv;
 294
 295                 p = arg_start + *pos;
 296                 len = len1 - *pos;
 297                 while (count > 0 && len > 0) {
 298                         unsigned int _count, l;
 299                         int nr_read;
 300                         bool final;
 301
 302                         _count = min3(count, len, PAGE_SIZE);
 303                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 304                         if (nr_read < 0)
 305                                 rv = nr_read;
 306                         if (nr_read <= 0)
 307                                 goto out_free_page;
 308
 309                         /*
 310                          * Command line can be shorter than whole ARGV
 311                          * even if last "marker" byte says it is not.
 312                          */
 313                         final = false;
 314                         l = strnlen(page, nr_read);
 315                         if (l < nr_read) {
 316                                 nr_read = l;
 317                                 final = true;
 318                         }
 319
 320                         if (copy_to_user(buf, page, nr_read)) {
 321                                 rv = -EFAULT;
 322                                 goto out_free_page;
 323                         }
 324
 325                         p       += nr_read;
 326                         len     -= nr_read;
 327                         buf     += nr_read;
 328                         count   -= nr_read;
 329                         rv      += nr_read;
 330
 331                         if (final)
 332                                 goto out_free_page;
 333                 }
 334 skip_argv:
 335                 /*
 336                  * Command line (1 string) occupies ARGV and
 337                  * extends into ENVP.
 338                  */
 339                 if (len1 <= *pos) {
 340                         p = env_start + *pos - len1;
 341                         len = len1 + len2 - *pos;
 342                 } else {
 343                         p = env_start;
 344                         len = len2;
 345                 }
 346                 while (count > 0 && len > 0) {
 347                         unsigned int _count, l;
 348                         int nr_read;
 349                         bool final;
 350
 351                         _count = min3(count, len, PAGE_SIZE);
 352                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 353                         if (nr_read < 0)
 354                                 rv = nr_read;
 355                         if (nr_read <= 0)
 356                                 goto out_free_page;
 357
 358                         /* Find EOS. */
 359                         final = false;
 360                         l = strnlen(page, nr_read);
 361                         if (l < nr_read) {
 362                                 nr_read = l;
 363                                 final = true;
 364                         }
 365
 366                         if (copy_to_user(buf, page, nr_read)) {
 367                                 rv = -EFAULT;
 368                                 goto out_free_page;
 369                         }
 370
 371                         p       += nr_read;
 372                         len     -= nr_read;
 373                         buf     += nr_read;
 374                         count   -= nr_read;
 375                         rv      += nr_read;
 376
 377                         if (final)
 378                                 goto out_free_page;
 379                 }
 380 skip_argv_envp:
 381                 ;
 382         }
 383
 384 out_free_page:
 385         free_page((unsigned long)page);
 386 out_mmput:
 387         mmput(mm);
 388         if (rv > 0)
 389                 *pos += rv;
 390         return rv;
 391 }
 392
 393 static const struct file_operations proc_pid_cmdline_ops = {
 394         .read   = proc_pid_cmdline_read,
 395         .llseek = generic_file_llseek,
 396 };
 397
 398 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 399                          struct pid *pid, struct task_struct *task)
 400 {
 401         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ);
 402         if (mm && !IS_ERR(mm)) {
 403                 unsigned int nwords = 0;
 404                 do {
 405                         nwords += 2;
 406                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 407                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 408                 mmput(mm);
 409                 return 0;
 410         } else
 411                 return PTR_ERR(mm);
 412 }
 413
 414
 415 #ifdef CONFIG_KALLSYMS
 416 /*
 417  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 418  * Returns the resolved symbol.  If that fails, simply return the address.
 419  */
 420 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 421                           struct pid *pid, struct task_struct *task)
 422 {
 423         unsigned long wchan;
 424         char symname[KSYM_NAME_LEN];
 425
 426         wchan = get_wchan(task);
 427
 428         if (lookup_symbol_name(wchan, symname) < 0) {
 429                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 430                         return 0;
 431                 seq_printf(m, "%lu", wchan);
 432         } else {
 433                 seq_printf(m, "%s", symname);
 434         }
 435
 436         return 0;
 437 }
 438 #endif /* CONFIG_KALLSYMS */
 439
 440 static int lock_trace(struct task_struct *task)
 441 {
 442         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 443         if (err)
 444                 return err;
 445         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 446                 mutex_unlock(&task->signal->cred_guard_mutex);
 447                 return -EPERM;
 448         }
 449         return 0;
 450 }
 451
 452 static void unlock_trace(struct task_struct *task)
 453 {
 454         mutex_unlock(&task->signal->cred_guard_mutex);
 455 }
 456
 457 #ifdef CONFIG_STACKTRACE
 458
 459 #define MAX_STACK_TRACE_DEPTH   64
 460
 461 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 462                           struct pid *pid, struct task_struct *task)
 463 {
 464         struct stack_trace trace;
 465         unsigned long *entries;
 466         int err;
 467         int i;
 468
 469         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 470         if (!entries)
 471                 return -ENOMEM;
 472
 473         trace.nr_entries        = 0;
 474         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 475         trace.entries           = entries;
 476         trace.skip              = 0;
 477
 478         err = lock_trace(task);
 479         if (!err) {
 480                 save_stack_trace_tsk(task, &trace);
 481
 482                 for (i = 0; i < trace.nr_entries; i++) {
 483                         seq_printf(m, "[<%pK>] %pS\n",
 484                                    (void *)entries[i], (void *)entries[i]);
 485                 }
 486                 unlock_trace(task);
 487         }
 488         kfree(entries);
 489
 490         return err;
 491 }
 492 #endif
 493
 494 #ifdef CONFIG_SCHED_INFO
 495 /*
 496  * Provides /proc/PID/schedstat
 497  */
 498 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 499                               struct pid *pid, struct task_struct *task)
 500 {
 501         if (unlikely(!sched_info_on()))
 502                 seq_printf(m, "0 0 0\n");
 503         else
 504                 seq_printf(m, "%llu %llu %lu\n",
 505                    (unsigned long long)task->se.sum_exec_runtime,
 506                    (unsigned long long)task->sched_info.run_delay,
 507                    task->sched_info.pcount);
 508
 509         return 0;
 510 }
 511 #endif
 512
 513 #ifdef CONFIG_LATENCYTOP
 514 static int lstats_show_proc(struct seq_file *m, void *v)
 515 {
 516         int i;
 517         struct inode *inode = m->private;
 518         struct task_struct *task = get_proc_task(inode);
 519
 520         if (!task)
 521                 return -ESRCH;
 522         seq_puts(m, "Latency Top version : v0.1\n");
 523         for (i = 0; i < 32; i++) {
 524                 struct latency_record *lr = &task->latency_record[i];
 525                 if (lr->backtrace[0]) {
 526                         int q;
 527                         seq_printf(m, "%i %li %li",
 528                                    lr->count, lr->time, lr->max);
 529                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 530                                 unsigned long bt = lr->backtrace[q];
 531                                 if (!bt)
 532                                         break;
 533                                 if (bt == ULONG_MAX)
 534                                         break;
 535                                 seq_printf(m, " %ps", (void *)bt);
 536                         }
 537                         seq_putc(m, '\n');
 538                 }
 539
 540         }
 541         put_task_struct(task);
 542         return 0;
 543 }
 544
 545 static int lstats_open(struct inode *inode, struct file *file)
 546 {
 547         return single_open(file, lstats_show_proc, inode);
 548 }
 549
 550 static ssize_t lstats_write(struct file *file, const char __user *buf,
 551                             size_t count, loff_t *offs)
 552 {
 553         struct task_struct *task = get_proc_task(file_inode(file));
 554
 555         if (!task)
 556                 return -ESRCH;
 557         clear_all_latency_tracing(task);
 558         put_task_struct(task);
 559
 560         return count;
 561 }
 562
 563 static const struct file_operations proc_lstats_operations = {
 564         .open           = lstats_open,
 565         .read           = seq_read,
 566         .write          = lstats_write,
 567         .llseek         = seq_lseek,
 568         .release        = single_release,
 569 };
 570
 571 #endif
 572
 573 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 574                           struct pid *pid, struct task_struct *task)
 575 {
 576         unsigned long totalpages = totalram_pages + total_swap_pages;
 577         unsigned long points = 0;
 578
 579         read_lock(&tasklist_lock);
 580         if (pid_alive(task))
 581                 points = oom_badness(task, NULL, NULL, totalpages) *
 582                                                 1000 / totalpages;
 583         read_unlock(&tasklist_lock);
 584         seq_printf(m, "%lu\n", points);
 585
 586         return 0;
 587 }
 588
 589 struct limit_names {
 590         const char *name;
 591         const char *unit;
 592 };
 593
 594 static const struct limit_names lnames[RLIM_NLIMITS] = {
 595         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 596         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 597         [RLIMIT_DATA] = {"Max data size", "bytes"},
 598         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 599         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 600         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 601         [RLIMIT_NPROC] = {"Max processes", "processes"},
 602         [RLIMIT_NOFILE] = {"Max open files", "files"},
 603         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 604         [RLIMIT_AS] = {"Max address space", "bytes"},
 605         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 606         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 607         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 608         [RLIMIT_NICE] = {"Max nice priority", NULL},
 609         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 610         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 611 };
 612
 613 /* Display limits for a process */
 614 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 615                            struct pid *pid, struct task_struct *task)
 616 {
 617         unsigned int i;
 618         unsigned long flags;
 619
 620         struct rlimit rlim[RLIM_NLIMITS];
 621
 622         if (!lock_task_sighand(task, &flags))
 623                 return 0;
 624         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 625         unlock_task_sighand(task, &flags);
 626
 627         /*
 628          * print the file header
 629          */
 630        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 631                   "Limit", "Soft Limit", "Hard Limit", "Units");
 632
 633         for (i = 0; i < RLIM_NLIMITS; i++) {
 634                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 635                         seq_printf(m, "%-25s %-20s ",
 636                                    lnames[i].name, "unlimited");
 637                 else
 638                         seq_printf(m, "%-25s %-20lu ",
 639                                    lnames[i].name, rlim[i].rlim_cur);
 640
 641                 if (rlim[i].rlim_max == RLIM_INFINITY)
 642                         seq_printf(m, "%-20s ", "unlimited");
 643                 else
 644                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 645
 646                 if (lnames[i].unit)
 647                         seq_printf(m, "%-10s\n", lnames[i].unit);
 648                 else
 649                         seq_putc(m, '\n');
 650         }
 651
 652         return 0;
 653 }
 654
 655 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 656 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 657                             struct pid *pid, struct task_struct *task)
 658 {
 659         long nr;
 660         unsigned long args[6], sp, pc;
 661         int res;
 662
 663         res = lock_trace(task);
 664         if (res)
 665                 return res;
 666
 667         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 668                 seq_puts(m, "running\n");
 669         else if (nr < 0)
 670                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 671         else
 672                 seq_printf(m,
 673                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 674                        nr,
 675                        args[0], args[1], args[2], args[3], args[4], args[5],
 676                        sp, pc);
 677         unlock_trace(task);
 678
 679         return 0;
 680 }
 681 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 682
 683 /************************************************************************/
 684 /*                       Here the fs part begins                        */
 685 /************************************************************************/
 686
 687 /* permission checks */
 688 static int proc_fd_access_allowed(struct inode *inode)
 689 {
 690         struct task_struct *task;
 691         int allowed = 0;
 692         /* Allow access to a task's file descriptors if it is us or we
 693          * may use ptrace attach to the process and find out that
 694          * information.
 695          */
 696         task = get_proc_task(inode);
 697         if (task) {
 698                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 699                 put_task_struct(task);
 700         }
 701         return allowed;
 702 }
 703
 704 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 705 {
 706         int error;
 707         struct inode *inode = d_inode(dentry);
 708
 709         if (attr->ia_valid & ATTR_MODE)
 710                 return -EPERM;
 711
 712         error = inode_change_ok(inode, attr);
 713         if (error)
 714                 return error;
 715
 716         setattr_copy(inode, attr);
 717         mark_inode_dirty(inode);
 718         return 0;
 719 }
 720
 721 /*
 722  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 723  * or euid/egid (for hide_pid_min=2)?
 724  */
 725 static bool has_pid_permissions(struct pid_namespace *pid,
 726                                  struct task_struct *task,
 727                                  int hide_pid_min)
 728 {
 729         if (pid->hide_pid < hide_pid_min)
 730                 return true;
 731         if (in_group_p(pid->pid_gid))
 732                 return true;
 733         return ptrace_may_access(task, PTRACE_MODE_READ);
 734 }
 735
 736
 737 static int proc_pid_permission(struct inode *inode, int mask)
 738 {
 739         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 740         struct task_struct *task;
 741         bool has_perms;
 742
 743         task = get_proc_task(inode);
 744         if (!task)
 745                 return -ESRCH;
 746         has_perms = has_pid_permissions(pid, task, 1);
 747         put_task_struct(task);
 748
 749         if (!has_perms) {
 750                 if (pid->hide_pid == 2) {
 751                         /*
 752                          * Let's make getdents(), stat(), and open()
 753                          * consistent with each other.  If a process
 754                          * may not stat() a file, it shouldn't be seen
 755                          * in procfs at all.
 756                          */
 757                         return -ENOENT;
 758                 }
 759
 760                 return -EPERM;
 761         }
 762         return generic_permission(inode, mask);
 763 }
 764
 765
 766
 767 static const struct inode_operations proc_def_inode_operations = {
 768         .setattr        = proc_setattr,
 769 };
 770
 771 static int proc_single_show(struct seq_file *m, void *v)
 772 {
 773         struct inode *inode = m->private;
 774         struct pid_namespace *ns;
 775         struct pid *pid;
 776         struct task_struct *task;
 777         int ret;
 778
 779         ns = inode->i_sb->s_fs_info;
 780         pid = proc_pid(inode);
 781         task = get_pid_task(pid, PIDTYPE_PID);
 782         if (!task)
 783                 return -ESRCH;
 784
 785         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 786
 787         put_task_struct(task);
 788         return ret;
 789 }
 790
 791 static int proc_single_open(struct inode *inode, struct file *filp)
 792 {
 793         return single_open(filp, proc_single_show, inode);
 794 }
 795
 796 static const struct file_operations proc_single_file_operations = {
 797         .open           = proc_single_open,
 798         .read           = seq_read,
 799         .llseek         = seq_lseek,
 800         .release        = single_release,
 801 };
 802
 803
 804 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 805 {
 806         struct task_struct *task = get_proc_task(inode);
 807         struct mm_struct *mm = ERR_PTR(-ESRCH);
 808
 809         if (task) {
 810                 mm = mm_access(task, mode);
 811                 put_task_struct(task);
 812
 813                 if (!IS_ERR_OR_NULL(mm)) {
 814                         /* ensure this mm_struct can't be freed */
 815                         atomic_inc(&mm->mm_count);
 816                         /* but do not pin its memory */
 817                         mmput(mm);
 818                 }
 819         }
 820
 821         return mm;
 822 }
 823
 824 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 825 {
 826         struct mm_struct *mm = proc_mem_open(inode, mode);
 827
 828         if (IS_ERR(mm))
 829                 return PTR_ERR(mm);
 830
 831         file->private_data = mm;
 832         return 0;
 833 }
 834
 835 static int mem_open(struct inode *inode, struct file *file)
 836 {
 837         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 838
 839         /* OK to pass negative loff_t, we can catch out-of-range */
 840         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 841
 842         return ret;
 843 }
 844
 845 static ssize_t mem_rw(struct file *file, char __user *buf,
 846                         size_t count, loff_t *ppos, int write)
 847 {
 848         struct mm_struct *mm = file->private_data;
 849         unsigned long addr = *ppos;
 850         ssize_t copied;
 851         char *page;
 852
 853         if (!mm)
 854                 return 0;
 855
 856         page = (char *)__get_free_page(GFP_TEMPORARY);
 857         if (!page)
 858                 return -ENOMEM;
 859
 860         copied = 0;
 861         if (!atomic_inc_not_zero(&mm->mm_users))
 862                 goto free;
 863
 864         while (count > 0) {
 865                 int this_len = min_t(int, count, PAGE_SIZE);
 866
 867                 if (write && copy_from_user(page, buf, this_len)) {
 868                         copied = -EFAULT;
 869                         break;
 870                 }
 871
 872                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 873                 if (!this_len) {
 874                         if (!copied)
 875                                 copied = -EIO;
 876                         break;
 877                 }
 878
 879                 if (!write && copy_to_user(buf, page, this_len)) {
 880                         copied = -EFAULT;
 881                         break;
 882                 }
 883
 884                 buf += this_len;
 885                 addr += this_len;
 886                 copied += this_len;
 887                 count -= this_len;
 888         }
 889         *ppos = addr;
 890
 891         mmput(mm);
 892 free:
 893         free_page((unsigned long) page);
 894         return copied;
 895 }
 896
 897 static ssize_t mem_read(struct file *file, char __user *buf,
 898                         size_t count, loff_t *ppos)
 899 {
 900         return mem_rw(file, buf, count, ppos, 0);
 901 }
 902
 903 static ssize_t mem_write(struct file *file, const char __user *buf,
 904                          size_t count, loff_t *ppos)
 905 {
 906         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 907 }
 908
 909 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 910 {
 911         switch (orig) {
 912         case 0:
 913                 file->f_pos = offset;
 914                 break;
 915         case 1:
 916                 file->f_pos += offset;
 917                 break;
 918         default:
 919                 return -EINVAL;
 920         }
 921         force_successful_syscall_return();
 922         return file->f_pos;
 923 }
 924
 925 static int mem_release(struct inode *inode, struct file *file)
 926 {
 927         struct mm_struct *mm = file->private_data;
 928         if (mm)
 929                 mmdrop(mm);
 930         return 0;
 931 }
 932
 933 static const struct file_operations proc_mem_operations = {
 934         .llseek         = mem_lseek,
 935         .read           = mem_read,
 936         .write          = mem_write,
 937         .open           = mem_open,
 938         .release        = mem_release,
 939 };
 940
 941 static int environ_open(struct inode *inode, struct file *file)
 942 {
 943         return __mem_open(inode, file, PTRACE_MODE_READ);
 944 }
 945
 946 static ssize_t environ_read(struct file *file, char __user *buf,
 947                         size_t count, loff_t *ppos)
 948 {
 949         char *page;
 950         unsigned long src = *ppos;
 951         int ret = 0;
 952         struct mm_struct *mm = file->private_data;
 953
 954         if (!mm)
 955                 return 0;
 956
 957         page = (char *)__get_free_page(GFP_TEMPORARY);
 958         if (!page)
 959                 return -ENOMEM;
 960
 961         ret = 0;
 962         if (!atomic_inc_not_zero(&mm->mm_users))
 963                 goto free;
 964         while (count > 0) {
 965                 size_t this_len, max_len;
 966                 int retval;
 967
 968                 if (src >= (mm->env_end - mm->env_start))
 969                         break;
 970
 971                 this_len = mm->env_end - (mm->env_start + src);
 972
 973                 max_len = min_t(size_t, PAGE_SIZE, count);
 974                 this_len = min(max_len, this_len);
 975
 976                 retval = access_remote_vm(mm, (mm->env_start + src),
 977                         page, this_len, 0);
 978
 979                 if (retval <= 0) {
 980                         ret = retval;
 981                         break;
 982                 }
 983
 984                 if (copy_to_user(buf, page, retval)) {
 985                         ret = -EFAULT;
 986                         break;
 987                 }
 988
 989                 ret += retval;
 990                 src += retval;
 991                 buf += retval;
 992                 count -= retval;
 993         }
 994         *ppos = src;
 995         mmput(mm);
 996
 997 free:
 998         free_page((unsigned long) page);
 999         return ret;
1000 }
1001
1002 static const struct file_operations proc_environ_operations = {
1003         .open           = environ_open,
1004         .read           = environ_read,
1005         .llseek         = generic_file_llseek,
1006         .release        = mem_release,
1007 };
1008
1009 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1010                             loff_t *ppos)
1011 {
1012         struct task_struct *task = get_proc_task(file_inode(file));
1013         char buffer[PROC_NUMBUF];
1014         int oom_adj = OOM_ADJUST_MIN;
1015         size_t len;
1016         unsigned long flags;
1017
1018         if (!task)
1019                 return -ESRCH;
1020         if (lock_task_sighand(task, &flags)) {
1021                 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1022                         oom_adj = OOM_ADJUST_MAX;
1023                 else
1024                         oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1025                                   OOM_SCORE_ADJ_MAX;
1026                 unlock_task_sighand(task, &flags);
1027         }
1028         put_task_struct(task);
1029         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1030         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1031 }
1032
1033 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1034                              size_t count, loff_t *ppos)
1035 {
1036         struct task_struct *task;
1037         char buffer[PROC_NUMBUF];
1038         int oom_adj;
1039         unsigned long flags;
1040         int err;
1041
1042         memset(buffer, 0, sizeof(buffer));
1043         if (count > sizeof(buffer) - 1)
1044                 count = sizeof(buffer) - 1;
1045         if (copy_from_user(buffer, buf, count)) {
1046                 err = -EFAULT;
1047                 goto out;
1048         }
1049
1050         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1051         if (err)
1052                 goto out;
1053         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1054              oom_adj != OOM_DISABLE) {
1055                 err = -EINVAL;
1056                 goto out;
1057         }
1058
1059         task = get_proc_task(file_inode(file));
1060         if (!task) {
1061                 err = -ESRCH;
1062                 goto out;
1063         }
1064
1065         task_lock(task);
1066         if (!task->mm) {
1067                 err = -EINVAL;
1068                 goto err_task_lock;
1069         }
1070
1071         if (!lock_task_sighand(task, &flags)) {
1072                 err = -ESRCH;
1073                 goto err_task_lock;
1074         }
1075
1076         /*
1077          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1078          * value is always attainable.
1079          */
1080         if (oom_adj == OOM_ADJUST_MAX)
1081                 oom_adj = OOM_SCORE_ADJ_MAX;
1082         else
1083                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1084
1085         if (oom_adj < task->signal->oom_score_adj &&
1086             !capable(CAP_SYS_RESOURCE)) {
1087                 err = -EACCES;
1088                 goto err_sighand;
1089         }
1090
1091         /*
1092          * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1093          * /proc/pid/oom_score_adj instead.
1094          */
1095         pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1096                   current->comm, task_pid_nr(current), task_pid_nr(task),
1097                   task_pid_nr(task));
1098
1099         task->signal->oom_score_adj = oom_adj;
1100         trace_oom_score_adj_update(task);
1101 err_sighand:
1102         unlock_task_sighand(task, &flags);
1103 err_task_lock:
1104         task_unlock(task);
1105         put_task_struct(task);
1106 out:
1107         return err < 0 ? err : count;
1108 }
1109
1110 static const struct file_operations proc_oom_adj_operations = {
1111         .read           = oom_adj_read,
1112         .write          = oom_adj_write,
1113         .llseek         = generic_file_llseek,
1114 };
1115
1116 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1117                                         size_t count, loff_t *ppos)
1118 {
1119         struct task_struct *task = get_proc_task(file_inode(file));
1120         char buffer[PROC_NUMBUF];
1121         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1122         unsigned long flags;
1123         size_t len;
1124
1125         if (!task)
1126                 return -ESRCH;
1127         if (lock_task_sighand(task, &flags)) {
1128                 oom_score_adj = task->signal->oom_score_adj;
1129                 unlock_task_sighand(task, &flags);
1130         }
1131         put_task_struct(task);
1132         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1133         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1134 }
1135
1136 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1137                                         size_t count, loff_t *ppos)
1138 {
1139         struct task_struct *task;
1140         char buffer[PROC_NUMBUF];
1141         unsigned long flags;
1142         int oom_score_adj;
1143         int err;
1144
1145         memset(buffer, 0, sizeof(buffer));
1146         if (count > sizeof(buffer) - 1)
1147                 count = sizeof(buffer) - 1;
1148         if (copy_from_user(buffer, buf, count)) {
1149                 err = -EFAULT;
1150                 goto out;
1151         }
1152
1153         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1154         if (err)
1155                 goto out;
1156         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1157                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1158                 err = -EINVAL;
1159                 goto out;
1160         }
1161
1162         task = get_proc_task(file_inode(file));
1163         if (!task) {
1164                 err = -ESRCH;
1165                 goto out;
1166         }
1167
1168         task_lock(task);
1169         if (!task->mm) {
1170                 err = -EINVAL;
1171                 goto err_task_lock;
1172         }
1173
1174         if (!lock_task_sighand(task, &flags)) {
1175                 err = -ESRCH;
1176                 goto err_task_lock;
1177         }
1178
1179         if ((short)oom_score_adj < task->signal->oom_score_adj_min &&
1180                         !capable(CAP_SYS_RESOURCE)) {
1181                 err = -EACCES;
1182                 goto err_sighand;
1183         }
1184
1185         task->signal->oom_score_adj = (short)oom_score_adj;
1186         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1187                 task->signal->oom_score_adj_min = (short)oom_score_adj;
1188         trace_oom_score_adj_update(task);
1189
1190 err_sighand:
1191         unlock_task_sighand(task, &flags);
1192 err_task_lock:
1193         task_unlock(task);
1194         put_task_struct(task);
1195 out:
1196         return err < 0 ? err : count;
1197 }
1198
1199 static const struct file_operations proc_oom_score_adj_operations = {
1200         .read           = oom_score_adj_read,
1201         .write          = oom_score_adj_write,
1202         .llseek         = default_llseek,
1203 };
1204
1205 #ifdef CONFIG_AUDITSYSCALL
1206 #define TMPBUFLEN 21
1207 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1208                                   size_t count, loff_t *ppos)
1209 {
1210         struct inode * inode = file_inode(file);
1211         struct task_struct *task = get_proc_task(inode);
1212         ssize_t length;
1213         char tmpbuf[TMPBUFLEN];
1214
1215         if (!task)
1216                 return -ESRCH;
1217         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1218                            from_kuid(file->f_cred->user_ns,
1219                                      audit_get_loginuid(task)));
1220         put_task_struct(task);
1221         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1222 }
1223
1224 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1225                                    size_t count, loff_t *ppos)
1226 {
1227         struct inode * inode = file_inode(file);
1228         char *page, *tmp;
1229         ssize_t length;
1230         uid_t loginuid;
1231         kuid_t kloginuid;
1232
1233         rcu_read_lock();
1234         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1235                 rcu_read_unlock();
1236                 return -EPERM;
1237         }
1238         rcu_read_unlock();
1239
1240         if (count >= PAGE_SIZE)
1241                 count = PAGE_SIZE - 1;
1242
1243         if (*ppos != 0) {
1244                 /* No partial writes. */
1245                 return -EINVAL;
1246         }
1247         page = (char*)__get_free_page(GFP_TEMPORARY);
1248         if (!page)
1249                 return -ENOMEM;
1250         length = -EFAULT;
1251         if (copy_from_user(page, buf, count))
1252                 goto out_free_page;
1253
1254         page[count] = '\0';
1255         loginuid = simple_strtoul(page, &tmp, 10);
1256         if (tmp == page) {
1257                 length = -EINVAL;
1258                 goto out_free_page;
1259
1260         }
1261
1262         /* is userspace tring to explicitly UNSET the loginuid? */
1263         if (loginuid == AUDIT_UID_UNSET) {
1264                 kloginuid = INVALID_UID;
1265         } else {
1266                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1267                 if (!uid_valid(kloginuid)) {
1268                         length = -EINVAL;
1269                         goto out_free_page;
1270                 }
1271         }
1272
1273         length = audit_set_loginuid(kloginuid);
1274         if (likely(length == 0))
1275                 length = count;
1276
1277 out_free_page:
1278         free_page((unsigned long) page);
1279         return length;
1280 }
1281
1282 static const struct file_operations proc_loginuid_operations = {
1283         .read           = proc_loginuid_read,
1284         .write          = proc_loginuid_write,
1285         .llseek         = generic_file_llseek,
1286 };
1287
1288 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1289                                   size_t count, loff_t *ppos)
1290 {
1291         struct inode * inode = file_inode(file);
1292         struct task_struct *task = get_proc_task(inode);
1293         ssize_t length;
1294         char tmpbuf[TMPBUFLEN];
1295
1296         if (!task)
1297                 return -ESRCH;
1298         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1299                                 audit_get_sessionid(task));
1300         put_task_struct(task);
1301         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1302 }
1303
1304 static const struct file_operations proc_sessionid_operations = {
1305         .read           = proc_sessionid_read,
1306         .llseek         = generic_file_llseek,
1307 };
1308 #endif
1309
1310 #ifdef CONFIG_FAULT_INJECTION
1311 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1312                                       size_t count, loff_t *ppos)
1313 {
1314         struct task_struct *task = get_proc_task(file_inode(file));
1315         char buffer[PROC_NUMBUF];
1316         size_t len;
1317         int make_it_fail;
1318
1319         if (!task)
1320                 return -ESRCH;
1321         make_it_fail = task->make_it_fail;
1322         put_task_struct(task);
1323
1324         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1325
1326         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1327 }
1328
1329 static ssize_t proc_fault_inject_write(struct file * file,
1330                         const char __user * buf, size_t count, loff_t *ppos)
1331 {
1332         struct task_struct *task;
1333         char buffer[PROC_NUMBUF], *end;
1334         int make_it_fail;
1335
1336         if (!capable(CAP_SYS_RESOURCE))
1337                 return -EPERM;
1338         memset(buffer, 0, sizeof(buffer));
1339         if (count > sizeof(buffer) - 1)
1340                 count = sizeof(buffer) - 1;
1341         if (copy_from_user(buffer, buf, count))
1342                 return -EFAULT;
1343         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1344         if (*end)
1345                 return -EINVAL;
1346         if (make_it_fail < 0 || make_it_fail > 1)
1347                 return -EINVAL;
1348
1349         task = get_proc_task(file_inode(file));
1350         if (!task)
1351                 return -ESRCH;
1352         task->make_it_fail = make_it_fail;
1353         put_task_struct(task);
1354
1355         return count;
1356 }
1357
1358 static const struct file_operations proc_fault_inject_operations = {
1359         .read           = proc_fault_inject_read,
1360         .write          = proc_fault_inject_write,
1361         .llseek         = generic_file_llseek,
1362 };
1363 #endif
1364
1365
1366 #ifdef CONFIG_SCHED_DEBUG
1367 /*
1368  * Print out various scheduling related per-task fields:
1369  */
1370 static int sched_show(struct seq_file *m, void *v)
1371 {
1372         struct inode *inode = m->private;
1373         struct task_struct *p;
1374
1375         p = get_proc_task(inode);
1376         if (!p)
1377                 return -ESRCH;
1378         proc_sched_show_task(p, m);
1379
1380         put_task_struct(p);
1381
1382         return 0;
1383 }
1384
1385 static ssize_t
1386 sched_write(struct file *file, const char __user *buf,
1387             size_t count, loff_t *offset)
1388 {
1389         struct inode *inode = file_inode(file);
1390         struct task_struct *p;
1391
1392         p = get_proc_task(inode);
1393         if (!p)
1394                 return -ESRCH;
1395         proc_sched_set_task(p);
1396
1397         put_task_struct(p);
1398
1399         return count;
1400 }
1401
1402 static int sched_open(struct inode *inode, struct file *filp)
1403 {
1404         return single_open(filp, sched_show, inode);
1405 }
1406
1407 static const struct file_operations proc_pid_sched_operations = {
1408         .open           = sched_open,
1409         .read           = seq_read,
1410         .write          = sched_write,
1411         .llseek         = seq_lseek,
1412         .release        = single_release,
1413 };
1414
1415 #endif
1416
1417 #ifdef CONFIG_SCHED_AUTOGROUP
1418 /*
1419  * Print out autogroup related information:
1420  */
1421 static int sched_autogroup_show(struct seq_file *m, void *v)
1422 {
1423         struct inode *inode = m->private;
1424         struct task_struct *p;
1425
1426         p = get_proc_task(inode);
1427         if (!p)
1428                 return -ESRCH;
1429         proc_sched_autogroup_show_task(p, m);
1430
1431         put_task_struct(p);
1432
1433         return 0;
1434 }
1435
1436 static ssize_t
1437 sched_autogroup_write(struct file *file, const char __user *buf,
1438             size_t count, loff_t *offset)
1439 {
1440         struct inode *inode = file_inode(file);
1441         struct task_struct *p;
1442         char buffer[PROC_NUMBUF];
1443         int nice;
1444         int err;
1445
1446         memset(buffer, 0, sizeof(buffer));
1447         if (count > sizeof(buffer) - 1)
1448                 count = sizeof(buffer) - 1;
1449         if (copy_from_user(buffer, buf, count))
1450                 return -EFAULT;
1451
1452         err = kstrtoint(strstrip(buffer), 0, &nice);
1453         if (err < 0)
1454                 return err;
1455
1456         p = get_proc_task(inode);
1457         if (!p)
1458                 return -ESRCH;
1459
1460         err = proc_sched_autogroup_set_nice(p, nice);
1461         if (err)
1462                 count = err;
1463
1464         put_task_struct(p);
1465
1466         return count;
1467 }
1468
1469 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1470 {
1471         int ret;
1472
1473         ret = single_open(filp, sched_autogroup_show, NULL);
1474         if (!ret) {
1475                 struct seq_file *m = filp->private_data;
1476
1477                 m->private = inode;
1478         }
1479         return ret;
1480 }
1481
1482 static const struct file_operations proc_pid_sched_autogroup_operations = {
1483         .open           = sched_autogroup_open,
1484         .read           = seq_read,
1485         .write          = sched_autogroup_write,
1486         .llseek         = seq_lseek,
1487         .release        = single_release,
1488 };
1489
1490 #endif /* CONFIG_SCHED_AUTOGROUP */
1491
1492 static ssize_t comm_write(struct file *file, const char __user *buf,
1493                                 size_t count, loff_t *offset)
1494 {
1495         struct inode *inode = file_inode(file);
1496         struct task_struct *p;
1497         char buffer[TASK_COMM_LEN];
1498         const size_t maxlen = sizeof(buffer) - 1;
1499
1500         memset(buffer, 0, sizeof(buffer));
1501         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1502                 return -EFAULT;
1503
1504         p = get_proc_task(inode);
1505         if (!p)
1506                 return -ESRCH;
1507
1508         if (same_thread_group(current, p))
1509                 set_task_comm(p, buffer);
1510         else
1511                 count = -EINVAL;
1512
1513         put_task_struct(p);
1514
1515         return count;
1516 }
1517
1518 static int comm_show(struct seq_file *m, void *v)
1519 {
1520         struct inode *inode = m->private;
1521         struct task_struct *p;
1522
1523         p = get_proc_task(inode);
1524         if (!p)
1525                 return -ESRCH;
1526
1527         task_lock(p);
1528         seq_printf(m, "%s\n", p->comm);
1529         task_unlock(p);
1530
1531         put_task_struct(p);
1532
1533         return 0;
1534 }
1535
1536 static int comm_open(struct inode *inode, struct file *filp)
1537 {
1538         return single_open(filp, comm_show, inode);
1539 }
1540
1541 static const struct file_operations proc_pid_set_comm_operations = {
1542         .open           = comm_open,
1543         .read           = seq_read,
1544         .write          = comm_write,
1545         .llseek         = seq_lseek,
1546         .release        = single_release,
1547 };
1548
1549 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1550 {
1551         struct task_struct *task;
1552         struct mm_struct *mm;
1553         struct file *exe_file;
1554
1555         task = get_proc_task(d_inode(dentry));
1556         if (!task)
1557                 return -ENOENT;
1558         mm = get_task_mm(task);
1559         put_task_struct(task);
1560         if (!mm)
1561                 return -ENOENT;
1562         exe_file = get_mm_exe_file(mm);
1563         mmput(mm);
1564         if (exe_file) {
1565                 *exe_path = exe_file->f_path;
1566                 path_get(&exe_file->f_path);
1567                 fput(exe_file);
1568                 return 0;
1569         } else
1570                 return -ENOENT;
1571 }
1572
1573 static const char *proc_pid_follow_link(struct dentry *dentry, void **cookie)
1574 {
1575         struct inode *inode = d_inode(dentry);
1576         struct path path;
1577         int error = -EACCES;
1578
1579         /* Are we allowed to snoop on the tasks file descriptors? */
1580         if (!proc_fd_access_allowed(inode))
1581                 goto out;
1582
1583         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1584         if (error)
1585                 goto out;
1586
1587         nd_jump_link(&path);
1588         return NULL;
1589 out:
1590         return ERR_PTR(error);
1591 }
1592
1593 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1594 {
1595         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1596         char *pathname;
1597         int len;
1598
1599         if (!tmp)
1600                 return -ENOMEM;
1601
1602         pathname = d_path(path, tmp, PAGE_SIZE);
1603         len = PTR_ERR(pathname);
1604         if (IS_ERR(pathname))
1605                 goto out;
1606         len = tmp + PAGE_SIZE - 1 - pathname;
1607
1608         if (len > buflen)
1609                 len = buflen;
1610         if (copy_to_user(buffer, pathname, len))
1611                 len = -EFAULT;
1612  out:
1613         free_page((unsigned long)tmp);
1614         return len;
1615 }
1616
1617 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1618 {
1619         int error = -EACCES;
1620         struct inode *inode = d_inode(dentry);
1621         struct path path;
1622
1623         /* Are we allowed to snoop on the tasks file descriptors? */
1624         if (!proc_fd_access_allowed(inode))
1625                 goto out;
1626
1627         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1628         if (error)
1629                 goto out;
1630
1631         error = do_proc_readlink(&path, buffer, buflen);
1632         path_put(&path);
1633 out:
1634         return error;
1635 }
1636
1637 const struct inode_operations proc_pid_link_inode_operations = {
1638         .readlink       = proc_pid_readlink,
1639         .follow_link    = proc_pid_follow_link,
1640         .setattr        = proc_setattr,
1641 };
1642
1643
1644 /* building an inode */
1645
1646 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1647 {
1648         struct inode * inode;
1649         struct proc_inode *ei;
1650         const struct cred *cred;
1651
1652         /* We need a new inode */
1653
1654         inode = new_inode(sb);
1655         if (!inode)
1656                 goto out;
1657
1658         /* Common stuff */
1659         ei = PROC_I(inode);
1660         inode->i_ino = get_next_ino();
1661         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1662         inode->i_op = &proc_def_inode_operations;
1663
1664         /*
1665          * grab the reference to task.
1666          */
1667         ei->pid = get_task_pid(task, PIDTYPE_PID);
1668         if (!ei->pid)
1669                 goto out_unlock;
1670
1671         if (task_dumpable(task)) {
1672                 rcu_read_lock();
1673                 cred = __task_cred(task);
1674                 inode->i_uid = cred->euid;
1675                 inode->i_gid = cred->egid;
1676                 rcu_read_unlock();
1677         }
1678         security_task_to_inode(task, inode);
1679
1680 out:
1681         return inode;
1682
1683 out_unlock:
1684         iput(inode);
1685         return NULL;
1686 }
1687
1688 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1689 {
1690         struct inode *inode = d_inode(dentry);
1691         struct task_struct *task;
1692         const struct cred *cred;
1693         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1694
1695         generic_fillattr(inode, stat);
1696
1697         rcu_read_lock();
1698         stat->uid = GLOBAL_ROOT_UID;
1699         stat->gid = GLOBAL_ROOT_GID;
1700         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1701         if (task) {
1702                 if (!has_pid_permissions(pid, task, 2)) {
1703                         rcu_read_unlock();
1704                         /*
1705                          * This doesn't prevent learning whether PID exists,
1706                          * it only makes getattr() consistent with readdir().
1707                          */
1708                         return -ENOENT;
1709                 }
1710                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1711                     task_dumpable(task)) {
1712                         cred = __task_cred(task);
1713                         stat->uid = cred->euid;
1714                         stat->gid = cred->egid;
1715                 }
1716         }
1717         rcu_read_unlock();
1718         return 0;
1719 }
1720
1721 /* dentry stuff */
1722
1723 /*
1724  *      Exceptional case: normally we are not allowed to unhash a busy
1725  * directory. In this case, however, we can do it - no aliasing problems
1726  * due to the way we treat inodes.
1727  *
1728  * Rewrite the inode's ownerships here because the owning task may have
1729  * performed a setuid(), etc.
1730  *
1731  * Before the /proc/pid/status file was created the only way to read
1732  * the effective uid of a /process was to stat /proc/pid.  Reading
1733  * /proc/pid/status is slow enough that procps and other packages
1734  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1735  * made this apply to all per process world readable and executable
1736  * directories.
1737  */
1738 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1739 {
1740         struct inode *inode;
1741         struct task_struct *task;
1742         const struct cred *cred;
1743
1744         if (flags & LOOKUP_RCU)
1745                 return -ECHILD;
1746
1747         inode = d_inode(dentry);
1748         task = get_proc_task(inode);
1749
1750         if (task) {
1751                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1752                     task_dumpable(task)) {
1753                         rcu_read_lock();
1754                         cred = __task_cred(task);
1755                         inode->i_uid = cred->euid;
1756                         inode->i_gid = cred->egid;
1757                         rcu_read_unlock();
1758                 } else {
1759                         inode->i_uid = GLOBAL_ROOT_UID;
1760                         inode->i_gid = GLOBAL_ROOT_GID;
1761                 }
1762                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1763                 security_task_to_inode(task, inode);
1764                 put_task_struct(task);
1765                 return 1;
1766         }
1767         return 0;
1768 }
1769
1770 static inline bool proc_inode_is_dead(struct inode *inode)
1771 {
1772         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1773 }
1774
1775 int pid_delete_dentry(const struct dentry *dentry)
1776 {
1777         /* Is the task we represent dead?
1778          * If so, then don't put the dentry on the lru list,
1779          * kill it immediately.
1780          */
1781         return proc_inode_is_dead(d_inode(dentry));
1782 }
1783
1784 const struct dentry_operations pid_dentry_operations =
1785 {
1786         .d_revalidate   = pid_revalidate,
1787         .d_delete       = pid_delete_dentry,
1788 };
1789
1790 /* Lookups */
1791
1792 /*
1793  * Fill a directory entry.
1794  *
1795  * If possible create the dcache entry and derive our inode number and
1796  * file type from dcache entry.
1797  *
1798  * Since all of the proc inode numbers are dynamically generated, the inode
1799  * numbers do not exist until the inode is cache.  This means creating the
1800  * the dcache entry in readdir is necessary to keep the inode numbers
1801  * reported by readdir in sync with the inode numbers reported
1802  * by stat.
1803  */
1804 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1805         const char *name, int len,
1806         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1807 {
1808         struct dentry *child, *dir = file->f_path.dentry;
1809         struct qstr qname = QSTR_INIT(name, len);
1810         struct inode *inode;
1811         unsigned type;
1812         ino_t ino;
1813
1814         child = d_hash_and_lookup(dir, &qname);
1815         if (!child) {
1816                 child = d_alloc(dir, &qname);
1817                 if (!child)
1818                         goto end_instantiate;
1819                 if (instantiate(d_inode(dir), child, task, ptr) < 0) {
1820                         dput(child);
1821                         goto end_instantiate;
1822                 }
1823         }
1824         inode = d_inode(child);
1825         ino = inode->i_ino;
1826         type = inode->i_mode >> 12;
1827         dput(child);
1828         return dir_emit(ctx, name, len, ino, type);
1829
1830 end_instantiate:
1831         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1832 }
1833
1834 #ifdef CONFIG_CHECKPOINT_RESTORE
1835
1836 /*
1837  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1838  * which represent vma start and end addresses.
1839  */
1840 static int dname_to_vma_addr(struct dentry *dentry,
1841                              unsigned long *start, unsigned long *end)
1842 {
1843         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1844                 return -EINVAL;
1845
1846         return 0;
1847 }
1848
1849 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1850 {
1851         unsigned long vm_start, vm_end;
1852         bool exact_vma_exists = false;
1853         struct mm_struct *mm = NULL;
1854         struct task_struct *task;
1855         const struct cred *cred;
1856         struct inode *inode;
1857         int status = 0;
1858
1859         if (flags & LOOKUP_RCU)
1860                 return -ECHILD;
1861
1862         if (!capable(CAP_SYS_ADMIN)) {
1863                 status = -EPERM;
1864                 goto out_notask;
1865         }
1866
1867         inode = d_inode(dentry);
1868         task = get_proc_task(inode);
1869         if (!task)
1870                 goto out_notask;
1871
1872         mm = mm_access(task, PTRACE_MODE_READ);
1873         if (IS_ERR_OR_NULL(mm))
1874                 goto out;
1875
1876         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1877                 down_read(&mm->mmap_sem);
1878                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1879                 up_read(&mm->mmap_sem);
1880         }
1881
1882         mmput(mm);
1883
1884         if (exact_vma_exists) {
1885                 if (task_dumpable(task)) {
1886                         rcu_read_lock();
1887                         cred = __task_cred(task);
1888                         inode->i_uid = cred->euid;
1889                         inode->i_gid = cred->egid;
1890                         rcu_read_unlock();
1891                 } else {
1892                         inode->i_uid = GLOBAL_ROOT_UID;
1893                         inode->i_gid = GLOBAL_ROOT_GID;
1894                 }
1895                 security_task_to_inode(task, inode);
1896                 status = 1;
1897         }
1898
1899 out:
1900         put_task_struct(task);
1901
1902 out_notask:
1903         return status;
1904 }
1905
1906 static const struct dentry_operations tid_map_files_dentry_operations = {
1907         .d_revalidate   = map_files_d_revalidate,
1908         .d_delete       = pid_delete_dentry,
1909 };
1910
1911 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
1912 {
1913         unsigned long vm_start, vm_end;
1914         struct vm_area_struct *vma;
1915         struct task_struct *task;
1916         struct mm_struct *mm;
1917         int rc;
1918
1919         rc = -ENOENT;
1920         task = get_proc_task(d_inode(dentry));
1921         if (!task)
1922                 goto out;
1923
1924         mm = get_task_mm(task);
1925         put_task_struct(task);
1926         if (!mm)
1927                 goto out;
1928
1929         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1930         if (rc)
1931                 goto out_mmput;
1932
1933         rc = -ENOENT;
1934         down_read(&mm->mmap_sem);
1935         vma = find_exact_vma(mm, vm_start, vm_end);
1936         if (vma && vma->vm_file) {
1937                 *path = vma->vm_file->f_path;
1938                 path_get(path);
1939                 rc = 0;
1940         }
1941         up_read(&mm->mmap_sem);
1942
1943 out_mmput:
1944         mmput(mm);
1945 out:
1946         return rc;
1947 }
1948
1949 struct map_files_info {
1950         fmode_t         mode;
1951         unsigned long   len;
1952         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1953 };
1954
1955 static int
1956 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1957                            struct task_struct *task, const void *ptr)
1958 {
1959         fmode_t mode = (fmode_t)(unsigned long)ptr;
1960         struct proc_inode *ei;
1961         struct inode *inode;
1962
1963         inode = proc_pid_make_inode(dir->i_sb, task);
1964         if (!inode)
1965                 return -ENOENT;
1966
1967         ei = PROC_I(inode);
1968         ei->op.proc_get_link = proc_map_files_get_link;
1969
1970         inode->i_op = &proc_pid_link_inode_operations;
1971         inode->i_size = 64;
1972         inode->i_mode = S_IFLNK;
1973
1974         if (mode & FMODE_READ)
1975                 inode->i_mode |= S_IRUSR;
1976         if (mode & FMODE_WRITE)
1977                 inode->i_mode |= S_IWUSR;
1978
1979         d_set_d_op(dentry, &tid_map_files_dentry_operations);
1980         d_add(dentry, inode);
1981
1982         return 0;
1983 }
1984
1985 static struct dentry *proc_map_files_lookup(struct inode *dir,
1986                 struct dentry *dentry, unsigned int flags)
1987 {
1988         unsigned long vm_start, vm_end;
1989         struct vm_area_struct *vma;
1990         struct task_struct *task;
1991         int result;
1992         struct mm_struct *mm;
1993
1994         result = -EPERM;
1995         if (!capable(CAP_SYS_ADMIN))
1996                 goto out;
1997
1998         result = -ENOENT;
1999         task = get_proc_task(dir);
2000         if (!task)
2001                 goto out;
2002
2003         result = -EACCES;
2004         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2005                 goto out_put_task;
2006
2007         result = -ENOENT;
2008         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2009                 goto out_put_task;
2010
2011         mm = get_task_mm(task);
2012         if (!mm)
2013                 goto out_put_task;
2014
2015         down_read(&mm->mmap_sem);
2016         vma = find_exact_vma(mm, vm_start, vm_end);
2017         if (!vma)
2018                 goto out_no_vma;
2019
2020         if (vma->vm_file)
2021                 result = proc_map_files_instantiate(dir, dentry, task,
2022                                 (void *)(unsigned long)vma->vm_file->f_mode);
2023
2024 out_no_vma:
2025         up_read(&mm->mmap_sem);
2026         mmput(mm);
2027 out_put_task:
2028         put_task_struct(task);
2029 out:
2030         return ERR_PTR(result);
2031 }
2032
2033 static const struct inode_operations proc_map_files_inode_operations = {
2034         .lookup         = proc_map_files_lookup,
2035         .permission     = proc_fd_permission,
2036         .setattr        = proc_setattr,
2037 };
2038
2039 static int
2040 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2041 {
2042         struct vm_area_struct *vma;
2043         struct task_struct *task;
2044         struct mm_struct *mm;
2045         unsigned long nr_files, pos, i;
2046         struct flex_array *fa = NULL;
2047         struct map_files_info info;
2048         struct map_files_info *p;
2049         int ret;
2050
2051         ret = -EPERM;
2052         if (!capable(CAP_SYS_ADMIN))
2053                 goto out;
2054
2055         ret = -ENOENT;
2056         task = get_proc_task(file_inode(file));
2057         if (!task)
2058                 goto out;
2059
2060         ret = -EACCES;
2061         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2062                 goto out_put_task;
2063
2064         ret = 0;
2065         if (!dir_emit_dots(file, ctx))
2066                 goto out_put_task;
2067
2068         mm = get_task_mm(task);
2069         if (!mm)
2070                 goto out_put_task;
2071         down_read(&mm->mmap_sem);
2072
2073         nr_files = 0;
2074
2075         /*
2076          * We need two passes here:
2077          *
2078          *  1) Collect vmas of mapped files with mmap_sem taken
2079          *  2) Release mmap_sem and instantiate entries
2080          *
2081          * otherwise we get lockdep complained, since filldir()
2082          * routine might require mmap_sem taken in might_fault().
2083          */
2084
2085         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2086                 if (vma->vm_file && ++pos > ctx->pos)
2087                         nr_files++;
2088         }
2089
2090         if (nr_files) {
2091                 fa = flex_array_alloc(sizeof(info), nr_files,
2092                                         GFP_KERNEL);
2093                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2094                                                 GFP_KERNEL)) {
2095                         ret = -ENOMEM;
2096                         if (fa)
2097                                 flex_array_free(fa);
2098                         up_read(&mm->mmap_sem);
2099                         mmput(mm);
2100                         goto out_put_task;
2101                 }
2102                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2103                                 vma = vma->vm_next) {
2104                         if (!vma->vm_file)
2105                                 continue;
2106                         if (++pos <= ctx->pos)
2107                                 continue;
2108
2109                         info.mode = vma->vm_file->f_mode;
2110                         info.len = snprintf(info.name,
2111                                         sizeof(info.name), "%lx-%lx",
2112                                         vma->vm_start, vma->vm_end);
2113                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2114                                 BUG();
2115                 }
2116         }
2117         up_read(&mm->mmap_sem);
2118
2119         for (i = 0; i < nr_files; i++) {
2120                 p = flex_array_get(fa, i);
2121                 if (!proc_fill_cache(file, ctx,
2122                                       p->name, p->len,
2123                                       proc_map_files_instantiate,
2124                                       task,
2125                                       (void *)(unsigned long)p->mode))
2126                         break;
2127                 ctx->pos++;
2128         }
2129         if (fa)
2130                 flex_array_free(fa);
2131         mmput(mm);
2132
2133 out_put_task:
2134         put_task_struct(task);
2135 out:
2136         return ret;
2137 }
2138
2139 static const struct file_operations proc_map_files_operations = {
2140         .read           = generic_read_dir,
2141         .iterate        = proc_map_files_readdir,
2142         .llseek         = default_llseek,
2143 };
2144
2145 struct timers_private {
2146         struct pid *pid;
2147         struct task_struct *task;
2148         struct sighand_struct *sighand;
2149         struct pid_namespace *ns;
2150         unsigned long flags;
2151 };
2152
2153 static void *timers_start(struct seq_file *m, loff_t *pos)
2154 {
2155         struct timers_private *tp = m->private;
2156
2157         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2158         if (!tp->task)
2159                 return ERR_PTR(-ESRCH);
2160
2161         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2162         if (!tp->sighand)
2163                 return ERR_PTR(-ESRCH);
2164
2165         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2166 }
2167
2168 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2169 {
2170         struct timers_private *tp = m->private;
2171         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2172 }
2173
2174 static void timers_stop(struct seq_file *m, void *v)
2175 {
2176         struct timers_private *tp = m->private;
2177
2178         if (tp->sighand) {
2179                 unlock_task_sighand(tp->task, &tp->flags);
2180                 tp->sighand = NULL;
2181         }
2182
2183         if (tp->task) {
2184                 put_task_struct(tp->task);
2185                 tp->task = NULL;
2186         }
2187 }
2188
2189 static int show_timer(struct seq_file *m, void *v)
2190 {
2191         struct k_itimer *timer;
2192         struct timers_private *tp = m->private;
2193         int notify;
2194         static const char * const nstr[] = {
2195                 [SIGEV_SIGNAL] = "signal",
2196                 [SIGEV_NONE] = "none",
2197                 [SIGEV_THREAD] = "thread",
2198         };
2199
2200         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2201         notify = timer->it_sigev_notify;
2202
2203         seq_printf(m, "ID: %d\n", timer->it_id);
2204         seq_printf(m, "signal: %d/%p\n",
2205                    timer->sigq->info.si_signo,
2206                    timer->sigq->info.si_value.sival_ptr);
2207         seq_printf(m, "notify: %s/%s.%d\n",
2208                    nstr[notify & ~SIGEV_THREAD_ID],
2209                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2210                    pid_nr_ns(timer->it_pid, tp->ns));
2211         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2212
2213         return 0;
2214 }
2215
2216 static const struct seq_operations proc_timers_seq_ops = {
2217         .start  = timers_start,
2218         .next   = timers_next,
2219         .stop   = timers_stop,
2220         .show   = show_timer,
2221 };
2222
2223 static int proc_timers_open(struct inode *inode, struct file *file)
2224 {
2225         struct timers_private *tp;
2226
2227         tp = __seq_open_private(file, &proc_timers_seq_ops,
2228                         sizeof(struct timers_private));
2229         if (!tp)
2230                 return -ENOMEM;
2231
2232         tp->pid = proc_pid(inode);
2233         tp->ns = inode->i_sb->s_fs_info;
2234         return 0;
2235 }
2236
2237 static const struct file_operations proc_timers_operations = {
2238         .open           = proc_timers_open,
2239         .read           = seq_read,
2240         .llseek         = seq_lseek,
2241         .release        = seq_release_private,
2242 };
2243 #endif /* CONFIG_CHECKPOINT_RESTORE */
2244
2245 static int proc_pident_instantiate(struct inode *dir,
2246         struct dentry *dentry, struct task_struct *task, const void *ptr)
2247 {
2248         const struct pid_entry *p = ptr;
2249         struct inode *inode;
2250         struct proc_inode *ei;
2251
2252         inode = proc_pid_make_inode(dir->i_sb, task);
2253         if (!inode)
2254                 goto out;
2255
2256         ei = PROC_I(inode);
2257         inode->i_mode = p->mode;
2258         if (S_ISDIR(inode->i_mode))
2259                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2260         if (p->iop)
2261                 inode->i_op = p->iop;
2262         if (p->fop)
2263                 inode->i_fop = p->fop;
2264         ei->op = p->op;
2265         d_set_d_op(dentry, &pid_dentry_operations);
2266         d_add(dentry, inode);
2267         /* Close the race of the process dying before we return the dentry */
2268         if (pid_revalidate(dentry, 0))
2269                 return 0;
2270 out:
2271         return -ENOENT;
2272 }
2273
2274 static struct dentry *proc_pident_lookup(struct inode *dir,
2275                                          struct dentry *dentry,
2276                                          const struct pid_entry *ents,
2277                                          unsigned int nents)
2278 {
2279         int error;
2280         struct task_struct *task = get_proc_task(dir);
2281         const struct pid_entry *p, *last;
2282
2283         error = -ENOENT;
2284
2285         if (!task)
2286                 goto out_no_task;
2287
2288         /*
2289          * Yes, it does not scale. And it should not. Don't add
2290          * new entries into /proc/<tgid>/ without very good reasons.
2291          */
2292         last = &ents[nents - 1];
2293         for (p = ents; p <= last; p++) {
2294                 if (p->len != dentry->d_name.len)
2295                         continue;
2296                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2297                         break;
2298         }
2299         if (p > last)
2300                 goto out;
2301
2302         error = proc_pident_instantiate(dir, dentry, task, p);
2303 out:
2304         put_task_struct(task);
2305 out_no_task:
2306         return ERR_PTR(error);
2307 }
2308
2309 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2310                 const struct pid_entry *ents, unsigned int nents)
2311 {
2312         struct task_struct *task = get_proc_task(file_inode(file));
2313         const struct pid_entry *p;
2314
2315         if (!task)
2316                 return -ENOENT;
2317
2318         if (!dir_emit_dots(file, ctx))
2319                 goto out;
2320
2321         if (ctx->pos >= nents + 2)
2322                 goto out;
2323
2324         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2325                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2326                                 proc_pident_instantiate, task, p))
2327                         break;
2328                 ctx->pos++;
2329         }
2330 out:
2331         put_task_struct(task);
2332         return 0;
2333 }
2334
2335 #ifdef CONFIG_SECURITY
2336 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2337                                   size_t count, loff_t *ppos)
2338 {
2339         struct inode * inode = file_inode(file);
2340         char *p = NULL;
2341         ssize_t length;
2342         struct task_struct *task = get_proc_task(inode);
2343
2344         if (!task)
2345                 return -ESRCH;
2346
2347         length = security_getprocattr(task,
2348                                       (char*)file->f_path.dentry->d_name.name,
2349                                       &p);
2350         put_task_struct(task);
2351         if (length > 0)
2352                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2353         kfree(p);
2354         return length;
2355 }
2356
2357 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2358                                    size_t count, loff_t *ppos)
2359 {
2360         struct inode * inode = file_inode(file);
2361         char *page;
2362         ssize_t length;
2363         struct task_struct *task = get_proc_task(inode);
2364
2365         length = -ESRCH;
2366         if (!task)
2367                 goto out_no_task;
2368         if (count > PAGE_SIZE)
2369                 count = PAGE_SIZE;
2370
2371         /* No partial writes. */
2372         length = -EINVAL;
2373         if (*ppos != 0)
2374                 goto out;
2375
2376         length = -ENOMEM;
2377         page = (char*)__get_free_page(GFP_TEMPORARY);
2378         if (!page)
2379                 goto out;
2380
2381         length = -EFAULT;
2382         if (copy_from_user(page, buf, count))
2383                 goto out_free;
2384
2385         /* Guard against adverse ptrace interaction */
2386         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2387         if (length < 0)
2388                 goto out_free;
2389
2390         length = security_setprocattr(task,
2391                                       (char*)file->f_path.dentry->d_name.name,
2392                                       (void*)page, count);
2393         mutex_unlock(&task->signal->cred_guard_mutex);
2394 out_free:
2395         free_page((unsigned long) page);
2396 out:
2397         put_task_struct(task);
2398 out_no_task:
2399         return length;
2400 }
2401
2402 static const struct file_operations proc_pid_attr_operations = {
2403         .read           = proc_pid_attr_read,
2404         .write          = proc_pid_attr_write,
2405         .llseek         = generic_file_llseek,
2406 };
2407
2408 static const struct pid_entry attr_dir_stuff[] = {
2409         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2410         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2411         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2412         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2413         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2414         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2415 };
2416
2417 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2418 {
2419         return proc_pident_readdir(file, ctx,
2420                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2421 }
2422
2423 static const struct file_operations proc_attr_dir_operations = {
2424         .read           = generic_read_dir,
2425         .iterate        = proc_attr_dir_readdir,
2426         .llseek         = default_llseek,
2427 };
2428
2429 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2430                                 struct dentry *dentry, unsigned int flags)
2431 {
2432         return proc_pident_lookup(dir, dentry,
2433                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2434 }
2435
2436 static const struct inode_operations proc_attr_dir_inode_operations = {
2437         .lookup         = proc_attr_dir_lookup,
2438         .getattr        = pid_getattr,
2439         .setattr        = proc_setattr,
2440 };
2441
2442 #endif
2443
2444 #ifdef CONFIG_ELF_CORE
2445 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2446                                          size_t count, loff_t *ppos)
2447 {
2448         struct task_struct *task = get_proc_task(file_inode(file));
2449         struct mm_struct *mm;
2450         char buffer[PROC_NUMBUF];
2451         size_t len;
2452         int ret;
2453
2454         if (!task)
2455                 return -ESRCH;
2456
2457         ret = 0;
2458         mm = get_task_mm(task);
2459         if (mm) {
2460                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2461                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2462                                 MMF_DUMP_FILTER_SHIFT));
2463                 mmput(mm);
2464                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2465         }
2466
2467         put_task_struct(task);
2468
2469         return ret;
2470 }
2471
2472 static ssize_t proc_coredump_filter_write(struct file *file,
2473                                           const char __user *buf,
2474                                           size_t count,
2475                                           loff_t *ppos)
2476 {
2477         struct task_struct *task;
2478         struct mm_struct *mm;
2479         char buffer[PROC_NUMBUF], *end;
2480         unsigned int val;
2481         int ret;
2482         int i;
2483         unsigned long mask;
2484
2485         ret = -EFAULT;
2486         memset(buffer, 0, sizeof(buffer));
2487         if (count > sizeof(buffer) - 1)
2488                 count = sizeof(buffer) - 1;
2489         if (copy_from_user(buffer, buf, count))
2490                 goto out_no_task;
2491
2492         ret = -EINVAL;
2493         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2494         if (*end == '\n')
2495                 end++;
2496         if (end - buffer == 0)
2497                 goto out_no_task;
2498
2499         ret = -ESRCH;
2500         task = get_proc_task(file_inode(file));
2501         if (!task)
2502                 goto out_no_task;
2503
2504         ret = end - buffer;
2505         mm = get_task_mm(task);
2506         if (!mm)
2507                 goto out_no_mm;
2508
2509         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2510                 if (val & mask)
2511                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2512                 else
2513                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2514         }
2515
2516         mmput(mm);
2517  out_no_mm:
2518         put_task_struct(task);
2519  out_no_task:
2520         return ret;
2521 }
2522
2523 static const struct file_operations proc_coredump_filter_operations = {
2524         .read           = proc_coredump_filter_read,
2525         .write          = proc_coredump_filter_write,
2526         .llseek         = generic_file_llseek,
2527 };
2528 #endif
2529
2530 #ifdef CONFIG_TASK_IO_ACCOUNTING
2531 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2532 {
2533         struct task_io_accounting acct = task->ioac;
2534         unsigned long flags;
2535         int result;
2536
2537         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2538         if (result)
2539                 return result;
2540
2541         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2542                 result = -EACCES;
2543                 goto out_unlock;
2544         }
2545
2546         if (whole && lock_task_sighand(task, &flags)) {
2547                 struct task_struct *t = task;
2548
2549                 task_io_accounting_add(&acct, &task->signal->ioac);
2550                 while_each_thread(task, t)
2551                         task_io_accounting_add(&acct, &t->ioac);
2552
2553                 unlock_task_sighand(task, &flags);
2554         }
2555         seq_printf(m,
2556                    "rchar: %llu\n"
2557                    "wchar: %llu\n"
2558                    "syscr: %llu\n"
2559                    "syscw: %llu\n"
2560                    "read_bytes: %llu\n"
2561                    "write_bytes: %llu\n"
2562                    "cancelled_write_bytes: %llu\n",
2563                    (unsigned long long)acct.rchar,
2564                    (unsigned long long)acct.wchar,
2565                    (unsigned long long)acct.syscr,
2566                    (unsigned long long)acct.syscw,
2567                    (unsigned long long)acct.read_bytes,
2568                    (unsigned long long)acct.write_bytes,
2569                    (unsigned long long)acct.cancelled_write_bytes);
2570         result = 0;
2571
2572 out_unlock:
2573         mutex_unlock(&task->signal->cred_guard_mutex);
2574         return result;
2575 }
2576
2577 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2578                                   struct pid *pid, struct task_struct *task)
2579 {
2580         return do_io_accounting(task, m, 0);
2581 }
2582
2583 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2584                                    struct pid *pid, struct task_struct *task)
2585 {
2586         return do_io_accounting(task, m, 1);
2587 }
2588 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2589
2590 #ifdef CONFIG_USER_NS
2591 static int proc_id_map_open(struct inode *inode, struct file *file,
2592         const struct seq_operations *seq_ops)
2593 {
2594         struct user_namespace *ns = NULL;
2595         struct task_struct *task;
2596         struct seq_file *seq;
2597         int ret = -EINVAL;
2598
2599         task = get_proc_task(inode);
2600         if (task) {
2601                 rcu_read_lock();
2602                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2603                 rcu_read_unlock();
2604                 put_task_struct(task);
2605         }
2606         if (!ns)
2607                 goto err;
2608
2609         ret = seq_open(file, seq_ops);
2610         if (ret)
2611                 goto err_put_ns;
2612
2613         seq = file->private_data;
2614         seq->private = ns;
2615
2616         return 0;
2617 err_put_ns:
2618         put_user_ns(ns);
2619 err:
2620         return ret;
2621 }
2622
2623 static int proc_id_map_release(struct inode *inode, struct file *file)
2624 {
2625         struct seq_file *seq = file->private_data;
2626         struct user_namespace *ns = seq->private;
2627         put_user_ns(ns);
2628         return seq_release(inode, file);
2629 }
2630
2631 static int proc_uid_map_open(struct inode *inode, struct file *file)
2632 {
2633         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2634 }
2635
2636 static int proc_gid_map_open(struct inode *inode, struct file *file)
2637 {
2638         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2639 }
2640
2641 static int proc_projid_map_open(struct inode *inode, struct file *file)
2642 {
2643         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2644 }
2645
2646 static const struct file_operations proc_uid_map_operations = {
2647         .open           = proc_uid_map_open,
2648         .write          = proc_uid_map_write,
2649         .read           = seq_read,
2650         .llseek         = seq_lseek,
2651         .release        = proc_id_map_release,
2652 };
2653
2654 static const struct file_operations proc_gid_map_operations = {
2655         .open           = proc_gid_map_open,
2656         .write          = proc_gid_map_write,
2657         .read           = seq_read,
2658         .llseek         = seq_lseek,
2659         .release        = proc_id_map_release,
2660 };
2661
2662 static const struct file_operations proc_projid_map_operations = {
2663         .open           = proc_projid_map_open,
2664         .write          = proc_projid_map_write,
2665         .read           = seq_read,
2666         .llseek         = seq_lseek,
2667         .release        = proc_id_map_release,
2668 };
2669
2670 static int proc_setgroups_open(struct inode *inode, struct file *file)
2671 {
2672         struct user_namespace *ns = NULL;
2673         struct task_struct *task;
2674         int ret;
2675
2676         ret = -ESRCH;
2677         task = get_proc_task(inode);
2678         if (task) {
2679                 rcu_read_lock();
2680                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2681                 rcu_read_unlock();
2682                 put_task_struct(task);
2683         }
2684         if (!ns)
2685                 goto err;
2686
2687         if (file->f_mode & FMODE_WRITE) {
2688                 ret = -EACCES;
2689                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2690                         goto err_put_ns;
2691         }
2692
2693         ret = single_open(file, &proc_setgroups_show, ns);
2694         if (ret)
2695                 goto err_put_ns;
2696
2697         return 0;
2698 err_put_ns:
2699         put_user_ns(ns);
2700 err:
2701         return ret;
2702 }
2703
2704 static int proc_setgroups_release(struct inode *inode, struct file *file)
2705 {
2706         struct seq_file *seq = file->private_data;
2707         struct user_namespace *ns = seq->private;
2708         int ret = single_release(inode, file);
2709         put_user_ns(ns);
2710         return ret;
2711 }
2712
2713 static const struct file_operations proc_setgroups_operations = {
2714         .open           = proc_setgroups_open,
2715         .write          = proc_setgroups_write,
2716         .read           = seq_read,
2717         .llseek         = seq_lseek,
2718         .release        = proc_setgroups_release,
2719 };
2720 #endif /* CONFIG_USER_NS */
2721
2722 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2723                                 struct pid *pid, struct task_struct *task)
2724 {
2725         int err = lock_trace(task);
2726         if (!err) {
2727                 seq_printf(m, "%08x\n", task->personality);
2728                 unlock_trace(task);
2729         }
2730         return err;
2731 }
2732
2733 /*
2734  * Thread groups
2735  */
2736 static const struct file_operations proc_task_operations;
2737 static const struct inode_operations proc_task_inode_operations;
2738
2739 static const struct pid_entry tgid_base_stuff[] = {
2740         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2741         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2742 #ifdef CONFIG_CHECKPOINT_RESTORE
2743         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2744 #endif
2745         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2746         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2747 #ifdef CONFIG_NET
2748         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2749 #endif
2750         REG("environ",    S_IRUSR, proc_environ_operations),
2751         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2752         ONE("status",     S_IRUGO, proc_pid_status),
2753         ONE("personality", S_IRUSR, proc_pid_personality),
2754         ONE("limits",     S_IRUGO, proc_pid_limits),
2755 #ifdef CONFIG_SCHED_DEBUG
2756         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2757 #endif
2758 #ifdef CONFIG_SCHED_AUTOGROUP
2759         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2760 #endif
2761         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2762 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2763         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2764 #endif
2765         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2766         ONE("stat",       S_IRUGO, proc_tgid_stat),
2767         ONE("statm",      S_IRUGO, proc_pid_statm),
2768         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2769 #ifdef CONFIG_NUMA
2770         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2771 #endif
2772         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2773         LNK("cwd",        proc_cwd_link),
2774         LNK("root",       proc_root_link),
2775         LNK("exe",        proc_exe_link),
2776         REG("mounts",     S_IRUGO, proc_mounts_operations),
2777         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2778         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2779 #ifdef CONFIG_PROC_PAGE_MONITOR
2780         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2781         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2782         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2783 #endif
2784 #ifdef CONFIG_SECURITY
2785         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2786 #endif
2787 #ifdef CONFIG_KALLSYMS
2788         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2789 #endif
2790 #ifdef CONFIG_STACKTRACE
2791         ONE("stack",      S_IRUSR, proc_pid_stack),
2792 #endif
2793 #ifdef CONFIG_SCHED_INFO
2794         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2795 #endif
2796 #ifdef CONFIG_LATENCYTOP
2797         REG("latency",  S_IRUGO, proc_lstats_operations),
2798 #endif
2799 #ifdef CONFIG_PROC_PID_CPUSET
2800         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2801 #endif
2802 #ifdef CONFIG_CGROUPS
2803         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2804 #endif
2805         ONE("oom_score",  S_IRUGO, proc_oom_score),
2806         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2807         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2808 #ifdef CONFIG_AUDITSYSCALL
2809         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2810         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2811 #endif
2812 #ifdef CONFIG_FAULT_INJECTION
2813         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2814 #endif
2815 #ifdef CONFIG_ELF_CORE
2816         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2817 #endif
2818 #ifdef CONFIG_TASK_IO_ACCOUNTING
2819         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2820 #endif
2821 #ifdef CONFIG_HARDWALL
2822         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2823 #endif
2824 #ifdef CONFIG_USER_NS
2825         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2826         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2827         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2828         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2829 #endif
2830 #ifdef CONFIG_CHECKPOINT_RESTORE
2831         REG("timers",     S_IRUGO, proc_timers_operations),
2832 #endif
2833 };
2834
2835 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2836 {
2837         return proc_pident_readdir(file, ctx,
2838                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2839 }
2840
2841 static const struct file_operations proc_tgid_base_operations = {
2842         .read           = generic_read_dir,
2843         .iterate        = proc_tgid_base_readdir,
2844         .llseek         = default_llseek,
2845 };
2846
2847 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2848 {
2849         return proc_pident_lookup(dir, dentry,
2850                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2851 }
2852
2853 static const struct inode_operations proc_tgid_base_inode_operations = {
2854         .lookup         = proc_tgid_base_lookup,
2855         .getattr        = pid_getattr,
2856         .setattr        = proc_setattr,
2857         .permission     = proc_pid_permission,
2858 };
2859
2860 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2861 {
2862         struct dentry *dentry, *leader, *dir;
2863         char buf[PROC_NUMBUF];
2864         struct qstr name;
2865
2866         name.name = buf;
2867         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2868         /* no ->d_hash() rejects on procfs */
2869         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2870         if (dentry) {
2871                 d_invalidate(dentry);
2872                 dput(dentry);
2873         }
2874
2875         if (pid == tgid)
2876                 return;
2877
2878         name.name = buf;
2879         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2880         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2881         if (!leader)
2882                 goto out;
2883
2884         name.name = "task";
2885         name.len = strlen(name.name);
2886         dir = d_hash_and_lookup(leader, &name);
2887         if (!dir)
2888                 goto out_put_leader;
2889
2890         name.name = buf;
2891         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2892         dentry = d_hash_and_lookup(dir, &name);
2893         if (dentry) {
2894                 d_invalidate(dentry);
2895                 dput(dentry);
2896         }
2897
2898         dput(dir);
2899 out_put_leader:
2900         dput(leader);
2901 out:
2902         return;
2903 }
2904
2905 /**
2906  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2907  * @task: task that should be flushed.
2908  *
2909  * When flushing dentries from proc, one needs to flush them from global
2910  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2911  * in. This call is supposed to do all of this job.
2912  *
2913  * Looks in the dcache for
2914  * /proc/@pid
2915  * /proc/@tgid/task/@pid
2916  * if either directory is present flushes it and all of it'ts children
2917  * from the dcache.
2918  *
2919  * It is safe and reasonable to cache /proc entries for a task until
2920  * that task exits.  After that they just clog up the dcache with
2921  * useless entries, possibly causing useful dcache entries to be
2922  * flushed instead.  This routine is proved to flush those useless
2923  * dcache entries at process exit time.
2924  *
2925  * NOTE: This routine is just an optimization so it does not guarantee
2926  *       that no dcache entries will exist at process exit time it
2927  *       just makes it very unlikely that any will persist.
2928  */
2929
2930 void proc_flush_task(struct task_struct *task)
2931 {
2932         int i;
2933         struct pid *pid, *tgid;
2934         struct upid *upid;
2935
2936         pid = task_pid(task);
2937         tgid = task_tgid(task);
2938
2939         for (i = 0; i <= pid->level; i++) {
2940                 upid = &pid->numbers[i];
2941                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2942                                         tgid->numbers[i].nr);
2943         }
2944 }
2945
2946 static int proc_pid_instantiate(struct inode *dir,
2947                                    struct dentry * dentry,
2948                                    struct task_struct *task, const void *ptr)
2949 {
2950         struct inode *inode;
2951
2952         inode = proc_pid_make_inode(dir->i_sb, task);
2953         if (!inode)
2954                 goto out;
2955
2956         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2957         inode->i_op = &proc_tgid_base_inode_operations;
2958         inode->i_fop = &proc_tgid_base_operations;
2959         inode->i_flags|=S_IMMUTABLE;
2960
2961         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2962                                                   ARRAY_SIZE(tgid_base_stuff)));
2963
2964         d_set_d_op(dentry, &pid_dentry_operations);
2965
2966         d_add(dentry, inode);
2967         /* Close the race of the process dying before we return the dentry */
2968         if (pid_revalidate(dentry, 0))
2969                 return 0;
2970 out:
2971         return -ENOENT;
2972 }
2973
2974 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
2975 {
2976         int result = -ENOENT;
2977         struct task_struct *task;
2978         unsigned tgid;
2979         struct pid_namespace *ns;
2980
2981         tgid = name_to_int(&dentry->d_name);
2982         if (tgid == ~0U)
2983                 goto out;
2984
2985         ns = dentry->d_sb->s_fs_info;
2986         rcu_read_lock();
2987         task = find_task_by_pid_ns(tgid, ns);
2988         if (task)
2989                 get_task_struct(task);
2990         rcu_read_unlock();
2991         if (!task)
2992                 goto out;
2993
2994         result = proc_pid_instantiate(dir, dentry, task, NULL);
2995         put_task_struct(task);
2996 out:
2997         return ERR_PTR(result);
2998 }
2999
3000 /*
3001  * Find the first task with tgid >= tgid
3002  *
3003  */
3004 struct tgid_iter {
3005         unsigned int tgid;
3006         struct task_struct *task;
3007 };
3008 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3009 {
3010         struct pid *pid;
3011
3012         if (iter.task)
3013                 put_task_struct(iter.task);
3014         rcu_read_lock();
3015 retry:
3016         iter.task = NULL;
3017         pid = find_ge_pid(iter.tgid, ns);
3018         if (pid) {
3019                 iter.tgid = pid_nr_ns(pid, ns);
3020                 iter.task = pid_task(pid, PIDTYPE_PID);
3021                 /* What we to know is if the pid we have find is the
3022                  * pid of a thread_group_leader.  Testing for task
3023                  * being a thread_group_leader is the obvious thing
3024                  * todo but there is a window when it fails, due to
3025                  * the pid transfer logic in de_thread.
3026                  *
3027                  * So we perform the straight forward test of seeing
3028                  * if the pid we have found is the pid of a thread
3029                  * group leader, and don't worry if the task we have
3030                  * found doesn't happen to be a thread group leader.
3031                  * As we don't care in the case of readdir.
3032                  */
3033                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3034                         iter.tgid += 1;
3035                         goto retry;
3036                 }
3037                 get_task_struct(iter.task);
3038         }
3039         rcu_read_unlock();
3040         return iter;
3041 }
3042
3043 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3044
3045 /* for the /proc/ directory itself, after non-process stuff has been done */
3046 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3047 {
3048         struct tgid_iter iter;
3049         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3050         loff_t pos = ctx->pos;
3051
3052         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3053                 return 0;
3054
3055         if (pos == TGID_OFFSET - 2) {
3056                 struct inode *inode = d_inode(ns->proc_self);
3057                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3058                         return 0;
3059                 ctx->pos = pos = pos + 1;
3060         }
3061         if (pos == TGID_OFFSET - 1) {
3062                 struct inode *inode = d_inode(ns->proc_thread_self);
3063                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3064                         return 0;
3065                 ctx->pos = pos = pos + 1;
3066         }
3067         iter.tgid = pos - TGID_OFFSET;
3068         iter.task = NULL;
3069         for (iter = next_tgid(ns, iter);
3070              iter.task;
3071              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3072                 char name[PROC_NUMBUF];
3073                 int len;
3074                 if (!has_pid_permissions(ns, iter.task, 2))
3075                         continue;
3076
3077                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3078                 ctx->pos = iter.tgid + TGID_OFFSET;
3079                 if (!proc_fill_cache(file, ctx, name, len,
3080                                      proc_pid_instantiate, iter.task, NULL)) {
3081                         put_task_struct(iter.task);
3082                         return 0;
3083                 }
3084         }
3085         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3086         return 0;
3087 }
3088
3089 /*
3090  * Tasks
3091  */
3092 static const struct pid_entry tid_base_stuff[] = {
3093         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3094         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3095         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3096 #ifdef CONFIG_NET
3097         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3098 #endif
3099         REG("environ",   S_IRUSR, proc_environ_operations),
3100         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3101         ONE("status",    S_IRUGO, proc_pid_status),
3102         ONE("personality", S_IRUSR, proc_pid_personality),
3103         ONE("limits",    S_IRUGO, proc_pid_limits),
3104 #ifdef CONFIG_SCHED_DEBUG
3105         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3106 #endif
3107         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3108 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3109         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3110 #endif
3111         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3112         ONE("stat",      S_IRUGO, proc_tid_stat),
3113         ONE("statm",     S_IRUGO, proc_pid_statm),
3114         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3115 #ifdef CONFIG_PROC_CHILDREN
3116         REG("children",  S_IRUGO, proc_tid_children_operations),
3117 #endif
3118 #ifdef CONFIG_NUMA
3119         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3120 #endif
3121         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3122         LNK("cwd",       proc_cwd_link),
3123         LNK("root",      proc_root_link),
3124         LNK("exe",       proc_exe_link),
3125         REG("mounts",    S_IRUGO, proc_mounts_operations),
3126         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3127 #ifdef CONFIG_PROC_PAGE_MONITOR
3128         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3129         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3130         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3131 #endif
3132 #ifdef CONFIG_SECURITY
3133         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3134 #endif
3135 #ifdef CONFIG_KALLSYMS
3136         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3137 #endif
3138 #ifdef CONFIG_STACKTRACE
3139         ONE("stack",      S_IRUSR, proc_pid_stack),
3140 #endif
3141 #ifdef CONFIG_SCHED_INFO
3142         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3143 #endif
3144 #ifdef CONFIG_LATENCYTOP
3145         REG("latency",  S_IRUGO, proc_lstats_operations),
3146 #endif
3147 #ifdef CONFIG_PROC_PID_CPUSET
3148         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3149 #endif
3150 #ifdef CONFIG_CGROUPS
3151         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3152 #endif
3153         ONE("oom_score", S_IRUGO, proc_oom_score),
3154         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3155         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3156 #ifdef CONFIG_AUDITSYSCALL
3157         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3158         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3159 #endif
3160 #ifdef CONFIG_FAULT_INJECTION
3161         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3162 #endif
3163 #ifdef CONFIG_TASK_IO_ACCOUNTING
3164         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3165 #endif
3166 #ifdef CONFIG_HARDWALL
3167         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3168 #endif
3169 #ifdef CONFIG_USER_NS
3170         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3171         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3172         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3173         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3174 #endif
3175 };
3176
3177 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3178 {
3179         return proc_pident_readdir(file, ctx,
3180                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3181 }
3182
3183 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3184 {
3185         return proc_pident_lookup(dir, dentry,
3186                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3187 }
3188
3189 static const struct file_operations proc_tid_base_operations = {
3190         .read           = generic_read_dir,
3191         .iterate        = proc_tid_base_readdir,
3192         .llseek         = default_llseek,
3193 };
3194
3195 static const struct inode_operations proc_tid_base_inode_operations = {
3196         .lookup         = proc_tid_base_lookup,
3197         .getattr        = pid_getattr,
3198         .setattr        = proc_setattr,
3199 };
3200
3201 static int proc_task_instantiate(struct inode *dir,
3202         struct dentry *dentry, struct task_struct *task, const void *ptr)
3203 {
3204         struct inode *inode;
3205         inode = proc_pid_make_inode(dir->i_sb, task);
3206
3207         if (!inode)
3208                 goto out;
3209         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3210         inode->i_op = &proc_tid_base_inode_operations;
3211         inode->i_fop = &proc_tid_base_operations;
3212         inode->i_flags|=S_IMMUTABLE;
3213
3214         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3215                                                   ARRAY_SIZE(tid_base_stuff)));
3216
3217         d_set_d_op(dentry, &pid_dentry_operations);
3218
3219         d_add(dentry, inode);
3220         /* Close the race of the process dying before we return the dentry */
3221         if (pid_revalidate(dentry, 0))
3222                 return 0;
3223 out:
3224         return -ENOENT;
3225 }
3226
3227 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3228 {
3229         int result = -ENOENT;
3230         struct task_struct *task;
3231         struct task_struct *leader = get_proc_task(dir);
3232         unsigned tid;
3233         struct pid_namespace *ns;
3234
3235         if (!leader)
3236                 goto out_no_task;
3237
3238         tid = name_to_int(&dentry->d_name);
3239         if (tid == ~0U)
3240                 goto out;
3241
3242         ns = dentry->d_sb->s_fs_info;
3243         rcu_read_lock();
3244         task = find_task_by_pid_ns(tid, ns);
3245         if (task)
3246                 get_task_struct(task);
3247         rcu_read_unlock();
3248         if (!task)
3249                 goto out;
3250         if (!same_thread_group(leader, task))
3251                 goto out_drop_task;
3252
3253         result = proc_task_instantiate(dir, dentry, task, NULL);
3254 out_drop_task:
3255         put_task_struct(task);
3256 out:
3257         put_task_struct(leader);
3258 out_no_task:
3259         return ERR_PTR(result);
3260 }
3261
3262 /*
3263  * Find the first tid of a thread group to return to user space.
3264  *
3265  * Usually this is just the thread group leader, but if the users
3266  * buffer was too small or there was a seek into the middle of the
3267  * directory we have more work todo.
3268  *
3269  * In the case of a short read we start with find_task_by_pid.
3270  *
3271  * In the case of a seek we start with the leader and walk nr
3272  * threads past it.
3273  */
3274 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3275                                         struct pid_namespace *ns)
3276 {
3277         struct task_struct *pos, *task;
3278         unsigned long nr = f_pos;
3279
3280         if (nr != f_pos)        /* 32bit overflow? */
3281                 return NULL;
3282
3283         rcu_read_lock();
3284         task = pid_task(pid, PIDTYPE_PID);
3285         if (!task)
3286                 goto fail;
3287
3288         /* Attempt to start with the tid of a thread */
3289         if (tid && nr) {
3290                 pos = find_task_by_pid_ns(tid, ns);
3291                 if (pos && same_thread_group(pos, task))
3292                         goto found;
3293         }
3294
3295         /* If nr exceeds the number of threads there is nothing todo */
3296         if (nr >= get_nr_threads(task))
3297                 goto fail;
3298
3299         /* If we haven't found our starting place yet start
3300          * with the leader and walk nr threads forward.
3301          */
3302         pos = task = task->group_leader;
3303         do {
3304                 if (!nr--)
3305                         goto found;
3306         } while_each_thread(task, pos);
3307 fail:
3308         pos = NULL;
3309         goto out;
3310 found:
3311         get_task_struct(pos);
3312 out:
3313         rcu_read_unlock();
3314         return pos;
3315 }
3316
3317 /*
3318  * Find the next thread in the thread list.
3319  * Return NULL if there is an error or no next thread.
3320  *
3321  * The reference to the input task_struct is released.
3322  */
3323 static struct task_struct *next_tid(struct task_struct *start)
3324 {
3325         struct task_struct *pos = NULL;
3326         rcu_read_lock();
3327         if (pid_alive(start)) {
3328                 pos = next_thread(start);
3329                 if (thread_group_leader(pos))
3330                         pos = NULL;
3331                 else
3332                         get_task_struct(pos);
3333         }
3334         rcu_read_unlock();
3335         put_task_struct(start);
3336         return pos;
3337 }
3338
3339 /* for the /proc/TGID/task/ directories */
3340 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3341 {
3342         struct inode *inode = file_inode(file);
3343         struct task_struct *task;
3344         struct pid_namespace *ns;
3345         int tid;
3346
3347         if (proc_inode_is_dead(inode))
3348                 return -ENOENT;
3349
3350         if (!dir_emit_dots(file, ctx))
3351                 return 0;
3352
3353         /* f_version caches the tgid value that the last readdir call couldn't
3354          * return. lseek aka telldir automagically resets f_version to 0.
3355          */
3356         ns = inode->i_sb->s_fs_info;
3357         tid = (int)file->f_version;
3358         file->f_version = 0;
3359         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3360              task;
3361              task = next_tid(task), ctx->pos++) {
3362                 char name[PROC_NUMBUF];
3363                 int len;
3364                 tid = task_pid_nr_ns(task, ns);
3365                 len = snprintf(name, sizeof(name), "%d", tid);
3366                 if (!proc_fill_cache(file, ctx, name, len,
3367                                 proc_task_instantiate, task, NULL)) {
3368                         /* returning this tgid failed, save it as the first
3369                          * pid for the next readir call */
3370                         file->f_version = (u64)tid;
3371                         put_task_struct(task);
3372                         break;
3373                 }
3374         }
3375
3376         return 0;
3377 }
3378
3379 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3380 {
3381         struct inode *inode = d_inode(dentry);
3382         struct task_struct *p = get_proc_task(inode);
3383         generic_fillattr(inode, stat);
3384
3385         if (p) {
3386                 stat->nlink += get_nr_threads(p);
3387                 put_task_struct(p);
3388         }
3389
3390         return 0;
3391 }
3392
3393 static const struct inode_operations proc_task_inode_operations = {
3394         .lookup         = proc_task_lookup,
3395         .getattr        = proc_task_getattr,
3396         .setattr        = proc_setattr,
3397         .permission     = proc_pid_permission,
3398 };
3399
3400 static const struct file_operations proc_task_operations = {
3401         .read           = generic_read_dir,
3402         .iterate        = proc_task_readdir,
3403         .llseek         = default_llseek,
3404 };