fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/flex_array.h>
  89 #include <linux/posix-timers.h>
  90 #ifdef CONFIG_HARDWALL
  91 #include <asm/hardwall.h>
  92 #endif
  93 #include <trace/events/oom.h>
  94 #include "internal.h"
  95 #include "fd.h"
  96
  97 /* NOTE:
  98  *      Implementing inode permission operations in /proc is almost
  99  *      certainly an error.  Permission checks need to happen during
 100  *      each system call not at open time.  The reason is that most of
 101  *      what we wish to check for permissions in /proc varies at runtime.
 102  *
 103  *      The classic example of a problem is opening file descriptors
 104  *      in /proc for a task before it execs a suid executable.
 105  */
 106
 107 struct pid_entry {
 108         const char *name;
 109         int len;
 110         umode_t mode;
 111         const struct inode_operations *iop;
 112         const struct file_operations *fop;
 113         union proc_op op;
 114 };
 115
 116 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 117         .name = (NAME),                                 \
 118         .len  = sizeof(NAME) - 1,                       \
 119         .mode = MODE,                                   \
 120         .iop  = IOP,                                    \
 121         .fop  = FOP,                                    \
 122         .op   = OP,                                     \
 123 }
 124
 125 #define DIR(NAME, MODE, iops, fops)     \
 126         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 127 #define LNK(NAME, get_link)                                     \
 128         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 129                 &proc_pid_link_inode_operations, NULL,          \
 130                 { .proc_get_link = get_link } )
 131 #define REG(NAME, MODE, fops)                           \
 132         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 133 #define ONE(NAME, MODE, show)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_single_file_operations,     \
 136                 { .proc_show = show } )
 137
 138 /*
 139  * Count the number of hardlinks for the pid_entry table, excluding the .
 140  * and .. links.
 141  */
 142 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 143         unsigned int n)
 144 {
 145         unsigned int i;
 146         unsigned int count;
 147
 148         count = 0;
 149         for (i = 0; i < n; ++i) {
 150                 if (S_ISDIR(entries[i].mode))
 151                         ++count;
 152         }
 153
 154         return count;
 155 }
 156
 157 static int get_task_root(struct task_struct *task, struct path *root)
 158 {
 159         int result = -ENOENT;
 160
 161         task_lock(task);
 162         if (task->fs) {
 163                 get_fs_root(task->fs, root);
 164                 result = 0;
 165         }
 166         task_unlock(task);
 167         return result;
 168 }
 169
 170 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 171 {
 172         struct task_struct *task = get_proc_task(d_inode(dentry));
 173         int result = -ENOENT;
 174
 175         if (task) {
 176                 task_lock(task);
 177                 if (task->fs) {
 178                         get_fs_pwd(task->fs, path);
 179                         result = 0;
 180                 }
 181                 task_unlock(task);
 182                 put_task_struct(task);
 183         }
 184         return result;
 185 }
 186
 187 static int proc_root_link(struct dentry *dentry, struct path *path)
 188 {
 189         struct task_struct *task = get_proc_task(d_inode(dentry));
 190         int result = -ENOENT;
 191
 192         if (task) {
 193                 result = get_task_root(task, path);
 194                 put_task_struct(task);
 195         }
 196         return result;
 197 }
 198
 199 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 200                                      size_t _count, loff_t *pos)
 201 {
 202         struct task_struct *tsk;
 203         struct mm_struct *mm;
 204         char *page;
 205         unsigned long count = _count;
 206         unsigned long arg_start, arg_end, env_start, env_end;
 207         unsigned long len1, len2, len;
 208         unsigned long p;
 209         char c;
 210         ssize_t rv;
 211
 212         BUG_ON(*pos < 0);
 213
 214         tsk = get_proc_task(file_inode(file));
 215         if (!tsk)
 216                 return -ESRCH;
 217         mm = get_task_mm(tsk);
 218         put_task_struct(tsk);
 219         if (!mm)
 220                 return 0;
 221         /* Check if process spawned far enough to have cmdline. */
 222         if (!mm->env_end) {
 223                 rv = 0;
 224                 goto out_mmput;
 225         }
 226
 227         page = (char *)__get_free_page(GFP_TEMPORARY);
 228         if (!page) {
 229                 rv = -ENOMEM;
 230                 goto out_mmput;
 231         }
 232
 233         down_read(&mm->mmap_sem);
 234         arg_start = mm->arg_start;
 235         arg_end = mm->arg_end;
 236         env_start = mm->env_start;
 237         env_end = mm->env_end;
 238         up_read(&mm->mmap_sem);
 239
 240         BUG_ON(arg_start > arg_end);
 241         BUG_ON(env_start > env_end);
 242
 243         len1 = arg_end - arg_start;
 244         len2 = env_end - env_start;
 245
 246         /* Empty ARGV. */
 247         if (len1 == 0) {
 248                 rv = 0;
 249                 goto out_free_page;
 250         }
 251         /*
 252          * Inherently racy -- command line shares address space
 253          * with code and data.
 254          */
 255         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 256         if (rv <= 0)
 257                 goto out_free_page;
 258
 259         rv = 0;
 260
 261         if (c == '\0') {
 262                 /* Command line (set of strings) occupies whole ARGV. */
 263                 if (len1 <= *pos)
 264                         goto out_free_page;
 265
 266                 p = arg_start + *pos;
 267                 len = len1 - *pos;
 268                 while (count > 0 && len > 0) {
 269                         unsigned int _count;
 270                         int nr_read;
 271
 272                         _count = min3(count, len, PAGE_SIZE);
 273                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 274                         if (nr_read < 0)
 275                                 rv = nr_read;
 276                         if (nr_read <= 0)
 277                                 goto out_free_page;
 278
 279                         if (copy_to_user(buf, page, nr_read)) {
 280                                 rv = -EFAULT;
 281                                 goto out_free_page;
 282                         }
 283
 284                         p       += nr_read;
 285                         len     -= nr_read;
 286                         buf     += nr_read;
 287                         count   -= nr_read;
 288                         rv      += nr_read;
 289                 }
 290         } else {
 291                 /*
 292                  * Command line (1 string) occupies ARGV and maybe
 293                  * extends into ENVP.
 294                  */
 295                 if (len1 + len2 <= *pos)
 296                         goto skip_argv_envp;
 297                 if (len1 <= *pos)
 298                         goto skip_argv;
 299
 300                 p = arg_start + *pos;
 301                 len = len1 - *pos;
 302                 while (count > 0 && len > 0) {
 303                         unsigned int _count, l;
 304                         int nr_read;
 305                         bool final;
 306
 307                         _count = min3(count, len, PAGE_SIZE);
 308                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 309                         if (nr_read < 0)
 310                                 rv = nr_read;
 311                         if (nr_read <= 0)
 312                                 goto out_free_page;
 313
 314                         /*
 315                          * Command line can be shorter than whole ARGV
 316                          * even if last "marker" byte says it is not.
 317                          */
 318                         final = false;
 319                         l = strnlen(page, nr_read);
 320                         if (l < nr_read) {
 321                                 nr_read = l;
 322                                 final = true;
 323                         }
 324
 325                         if (copy_to_user(buf, page, nr_read)) {
 326                                 rv = -EFAULT;
 327                                 goto out_free_page;
 328                         }
 329
 330                         p       += nr_read;
 331                         len     -= nr_read;
 332                         buf     += nr_read;
 333                         count   -= nr_read;
 334                         rv      += nr_read;
 335
 336                         if (final)
 337                                 goto out_free_page;
 338                 }
 339 skip_argv:
 340                 /*
 341                  * Command line (1 string) occupies ARGV and
 342                  * extends into ENVP.
 343                  */
 344                 if (len1 <= *pos) {
 345                         p = env_start + *pos - len1;
 346                         len = len1 + len2 - *pos;
 347                 } else {
 348                         p = env_start;
 349                         len = len2;
 350                 }
 351                 while (count > 0 && len > 0) {
 352                         unsigned int _count, l;
 353                         int nr_read;
 354                         bool final;
 355
 356                         _count = min3(count, len, PAGE_SIZE);
 357                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 358                         if (nr_read < 0)
 359                                 rv = nr_read;
 360                         if (nr_read <= 0)
 361                                 goto out_free_page;
 362
 363                         /* Find EOS. */
 364                         final = false;
 365                         l = strnlen(page, nr_read);
 366                         if (l < nr_read) {
 367                                 nr_read = l;
 368                                 final = true;
 369                         }
 370
 371                         if (copy_to_user(buf, page, nr_read)) {
 372                                 rv = -EFAULT;
 373                                 goto out_free_page;
 374                         }
 375
 376                         p       += nr_read;
 377                         len     -= nr_read;
 378                         buf     += nr_read;
 379                         count   -= nr_read;
 380                         rv      += nr_read;
 381
 382                         if (final)
 383                                 goto out_free_page;
 384                 }
 385 skip_argv_envp:
 386                 ;
 387         }
 388
 389 out_free_page:
 390         free_page((unsigned long)page);
 391 out_mmput:
 392         mmput(mm);
 393         if (rv > 0)
 394                 *pos += rv;
 395         return rv;
 396 }
 397
 398 static const struct file_operations proc_pid_cmdline_ops = {
 399         .read   = proc_pid_cmdline_read,
 400         .llseek = generic_file_llseek,
 401 };
 402
 403 static int proc_pid_auxv(struct seq_file *m, struct pid_namespace *ns,
 404                          struct pid *pid, struct task_struct *task)
 405 {
 406         struct mm_struct *mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
 407         if (mm && !IS_ERR(mm)) {
 408                 unsigned int nwords = 0;
 409                 do {
 410                         nwords += 2;
 411                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 412                 seq_write(m, mm->saved_auxv, nwords * sizeof(mm->saved_auxv[0]));
 413                 mmput(mm);
 414                 return 0;
 415         } else
 416                 return PTR_ERR(mm);
 417 }
 418
 419
 420 #ifdef CONFIG_KALLSYMS
 421 /*
 422  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 423  * Returns the resolved symbol.  If that fails, simply return the address.
 424  */
 425 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 426                           struct pid *pid, struct task_struct *task)
 427 {
 428         unsigned long wchan;
 429         char symname[KSYM_NAME_LEN];
 430
 431         wchan = get_wchan(task);
 432
 433         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 434                         && !lookup_symbol_name(wchan, symname))
 435                 seq_printf(m, "%s", symname);
 436         else
 437                 seq_putc(m, '0');
 438
 439         return 0;
 440 }
 441 #endif /* CONFIG_KALLSYMS */
 442
 443 static int lock_trace(struct task_struct *task)
 444 {
 445         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 446         if (err)
 447                 return err;
 448         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 449                 mutex_unlock(&task->signal->cred_guard_mutex);
 450                 return -EPERM;
 451         }
 452         return 0;
 453 }
 454
 455 static void unlock_trace(struct task_struct *task)
 456 {
 457         mutex_unlock(&task->signal->cred_guard_mutex);
 458 }
 459
 460 #ifdef CONFIG_STACKTRACE
 461
 462 #define MAX_STACK_TRACE_DEPTH   64
 463
 464 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 465                           struct pid *pid, struct task_struct *task)
 466 {
 467         struct stack_trace trace;
 468         unsigned long *entries;
 469         int err;
 470         int i;
 471
 472         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 473         if (!entries)
 474                 return -ENOMEM;
 475
 476         trace.nr_entries        = 0;
 477         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 478         trace.entries           = entries;
 479         trace.skip              = 0;
 480
 481         err = lock_trace(task);
 482         if (!err) {
 483                 save_stack_trace_tsk(task, &trace);
 484
 485                 for (i = 0; i < trace.nr_entries; i++) {
 486                         seq_printf(m, "[<%pK>] %pS\n",
 487                                    (void *)entries[i], (void *)entries[i]);
 488                 }
 489                 unlock_trace(task);
 490         }
 491         kfree(entries);
 492
 493         return err;
 494 }
 495 #endif
 496
 497 #ifdef CONFIG_SCHED_INFO
 498 /*
 499  * Provides /proc/PID/schedstat
 500  */
 501 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 502                               struct pid *pid, struct task_struct *task)
 503 {
 504         if (unlikely(!sched_info_on()))
 505                 seq_printf(m, "0 0 0\n");
 506         else
 507                 seq_printf(m, "%llu %llu %lu\n",
 508                    (unsigned long long)task->se.sum_exec_runtime,
 509                    (unsigned long long)task->sched_info.run_delay,
 510                    task->sched_info.pcount);
 511
 512         return 0;
 513 }
 514 #endif
 515
 516 #ifdef CONFIG_LATENCYTOP
 517 static int lstats_show_proc(struct seq_file *m, void *v)
 518 {
 519         int i;
 520         struct inode *inode = m->private;
 521         struct task_struct *task = get_proc_task(inode);
 522
 523         if (!task)
 524                 return -ESRCH;
 525         seq_puts(m, "Latency Top version : v0.1\n");
 526         for (i = 0; i < 32; i++) {
 527                 struct latency_record *lr = &task->latency_record[i];
 528                 if (lr->backtrace[0]) {
 529                         int q;
 530                         seq_printf(m, "%i %li %li",
 531                                    lr->count, lr->time, lr->max);
 532                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 533                                 unsigned long bt = lr->backtrace[q];
 534                                 if (!bt)
 535                                         break;
 536                                 if (bt == ULONG_MAX)
 537                                         break;
 538                                 seq_printf(m, " %ps", (void *)bt);
 539                         }
 540                         seq_putc(m, '\n');
 541                 }
 542
 543         }
 544         put_task_struct(task);
 545         return 0;
 546 }
 547
 548 static int lstats_open(struct inode *inode, struct file *file)
 549 {
 550         return single_open(file, lstats_show_proc, inode);
 551 }
 552
 553 static ssize_t lstats_write(struct file *file, const char __user *buf,
 554                             size_t count, loff_t *offs)
 555 {
 556         struct task_struct *task = get_proc_task(file_inode(file));
 557
 558         if (!task)
 559                 return -ESRCH;
 560         clear_all_latency_tracing(task);
 561         put_task_struct(task);
 562
 563         return count;
 564 }
 565
 566 static const struct file_operations proc_lstats_operations = {
 567         .open           = lstats_open,
 568         .read           = seq_read,
 569         .write          = lstats_write,
 570         .llseek         = seq_lseek,
 571         .release        = single_release,
 572 };
 573
 574 #endif
 575
 576 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 577                           struct pid *pid, struct task_struct *task)
 578 {
 579         unsigned long totalpages = totalram_pages + total_swap_pages;
 580         unsigned long points = 0;
 581
 582         points = oom_badness(task, NULL, NULL, totalpages) *
 583                                         1000 / totalpages;
 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_FSCREDS);
 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_FSCREDS);
 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 | PTRACE_MODE_FSCREDS);
 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         unsigned long env_start, env_end;
 954
 955         /* Ensure the process spawned far enough to have an environment. */
 956         if (!mm || !mm->env_end)
 957                 return 0;
 958
 959         page = (char *)__get_free_page(GFP_TEMPORARY);
 960         if (!page)
 961                 return -ENOMEM;
 962
 963         ret = 0;
 964         if (!atomic_inc_not_zero(&mm->mm_users))
 965                 goto free;
 966
 967         down_read(&mm->mmap_sem);
 968         env_start = mm->env_start;
 969         env_end = mm->env_end;
 970         up_read(&mm->mmap_sem);
 971
 972         while (count > 0) {
 973                 size_t this_len, max_len;
 974                 int retval;
 975
 976                 if (src >= (env_end - env_start))
 977                         break;
 978
 979                 this_len = env_end - (env_start + src);
 980
 981                 max_len = min_t(size_t, PAGE_SIZE, count);
 982                 this_len = min(max_len, this_len);
 983
 984                 retval = access_remote_vm(mm, (env_start + src),
 985                         page, this_len, 0);
 986
 987                 if (retval <= 0) {
 988                         ret = retval;
 989                         break;
 990                 }
 991
 992                 if (copy_to_user(buf, page, retval)) {
 993                         ret = -EFAULT;
 994                         break;
 995                 }
 996
 997                 ret += retval;
 998                 src += retval;
 999                 buf += retval;
1000                 count -= retval;
1001         }
1002         *ppos = src;
1003         mmput(mm);
1004
1005 free:
1006         free_page((unsigned long) page);
1007         return ret;
1008 }
1009
1010 static const struct file_operations proc_environ_operations = {
1011         .open           = environ_open,
1012         .read           = environ_read,
1013         .llseek         = generic_file_llseek,
1014         .release        = mem_release,
1015 };
1016
1017 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1018                             loff_t *ppos)
1019 {
1020         struct task_struct *task = get_proc_task(file_inode(file));
1021         char buffer[PROC_NUMBUF];
1022         int oom_adj = OOM_ADJUST_MIN;
1023         size_t len;
1024
1025         if (!task)
1026                 return -ESRCH;
1027         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1028                 oom_adj = OOM_ADJUST_MAX;
1029         else
1030                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1031                           OOM_SCORE_ADJ_MAX;
1032         put_task_struct(task);
1033         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1034         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1035 }
1036
1037 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1038 {
1039         static DEFINE_MUTEX(oom_adj_mutex);
1040         struct mm_struct *mm = NULL;
1041         struct task_struct *task;
1042         int err = 0;
1043
1044         task = get_proc_task(file_inode(file));
1045         if (!task)
1046                 return -ESRCH;
1047
1048         mutex_lock(&oom_adj_mutex);
1049         if (legacy) {
1050                 if (oom_adj < task->signal->oom_score_adj &&
1051                                 !capable(CAP_SYS_RESOURCE)) {
1052                         err = -EACCES;
1053                         goto err_unlock;
1054                 }
1055                 /*
1056                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1057                  * /proc/pid/oom_score_adj instead.
1058                  */
1059                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1060                           current->comm, task_pid_nr(current), task_pid_nr(task),
1061                           task_pid_nr(task));
1062         } else {
1063                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1064                                 !capable(CAP_SYS_RESOURCE)) {
1065                         err = -EACCES;
1066                         goto err_unlock;
1067                 }
1068         }
1069
1070         /*
1071          * Make sure we will check other processes sharing the mm if this is
1072          * not vfrok which wants its own oom_score_adj.
1073          * pin the mm so it doesn't go away and get reused after task_unlock
1074          */
1075         if (!task->vfork_done) {
1076                 struct task_struct *p = find_lock_task_mm(task);
1077
1078                 if (p) {
1079                         if (atomic_read(&p->mm->mm_users) > 1) {
1080                                 mm = p->mm;
1081                                 atomic_inc(&mm->mm_count);
1082                         }
1083                         task_unlock(p);
1084                 }
1085         }
1086
1087         task->signal->oom_score_adj = oom_adj;
1088         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1089                 task->signal->oom_score_adj_min = (short)oom_adj;
1090         trace_oom_score_adj_update(task);
1091
1092         if (mm) {
1093                 struct task_struct *p;
1094
1095                 rcu_read_lock();
1096                 for_each_process(p) {
1097                         if (same_thread_group(task, p))
1098                                 continue;
1099
1100                         /* do not touch kernel threads or the global init */
1101                         if (p->flags & PF_KTHREAD || is_global_init(p))
1102                                 continue;
1103
1104                         task_lock(p);
1105                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1106                                 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1107                                                 task_pid_nr(p), p->comm,
1108                                                 p->signal->oom_score_adj, oom_adj,
1109                                                 task_pid_nr(task), task->comm);
1110                                 p->signal->oom_score_adj = oom_adj;
1111                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1112                                         p->signal->oom_score_adj_min = (short)oom_adj;
1113                         }
1114                         task_unlock(p);
1115                 }
1116                 rcu_read_unlock();
1117                 mmdrop(mm);
1118         }
1119 err_unlock:
1120         mutex_unlock(&oom_adj_mutex);
1121         put_task_struct(task);
1122         return err;
1123 }
1124
1125 /*
1126  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1127  * kernels.  The effective policy is defined by oom_score_adj, which has a
1128  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1129  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1130  * Processes that become oom disabled via oom_adj will still be oom disabled
1131  * with this implementation.
1132  *
1133  * oom_adj cannot be removed since existing userspace binaries use it.
1134  */
1135 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1136                              size_t count, loff_t *ppos)
1137 {
1138         char buffer[PROC_NUMBUF];
1139         int oom_adj;
1140         int err;
1141
1142         memset(buffer, 0, sizeof(buffer));
1143         if (count > sizeof(buffer) - 1)
1144                 count = sizeof(buffer) - 1;
1145         if (copy_from_user(buffer, buf, count)) {
1146                 err = -EFAULT;
1147                 goto out;
1148         }
1149
1150         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1151         if (err)
1152                 goto out;
1153         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1154              oom_adj != OOM_DISABLE) {
1155                 err = -EINVAL;
1156                 goto out;
1157         }
1158
1159         /*
1160          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1161          * value is always attainable.
1162          */
1163         if (oom_adj == OOM_ADJUST_MAX)
1164                 oom_adj = OOM_SCORE_ADJ_MAX;
1165         else
1166                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1167
1168         err = __set_oom_adj(file, oom_adj, true);
1169 out:
1170         return err < 0 ? err : count;
1171 }
1172
1173 static const struct file_operations proc_oom_adj_operations = {
1174         .read           = oom_adj_read,
1175         .write          = oom_adj_write,
1176         .llseek         = generic_file_llseek,
1177 };
1178
1179 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1180                                         size_t count, loff_t *ppos)
1181 {
1182         struct task_struct *task = get_proc_task(file_inode(file));
1183         char buffer[PROC_NUMBUF];
1184         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1185         size_t len;
1186
1187         if (!task)
1188                 return -ESRCH;
1189         oom_score_adj = task->signal->oom_score_adj;
1190         put_task_struct(task);
1191         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1192         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1193 }
1194
1195 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1196                                         size_t count, loff_t *ppos)
1197 {
1198         char buffer[PROC_NUMBUF];
1199         int oom_score_adj;
1200         int err;
1201
1202         memset(buffer, 0, sizeof(buffer));
1203         if (count > sizeof(buffer) - 1)
1204                 count = sizeof(buffer) - 1;
1205         if (copy_from_user(buffer, buf, count)) {
1206                 err = -EFAULT;
1207                 goto out;
1208         }
1209
1210         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1211         if (err)
1212                 goto out;
1213         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1214                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1215                 err = -EINVAL;
1216                 goto out;
1217         }
1218
1219         err = __set_oom_adj(file, oom_score_adj, false);
1220 out:
1221         return err < 0 ? err : count;
1222 }
1223
1224 static const struct file_operations proc_oom_score_adj_operations = {
1225         .read           = oom_score_adj_read,
1226         .write          = oom_score_adj_write,
1227         .llseek         = default_llseek,
1228 };
1229
1230 #ifdef CONFIG_AUDITSYSCALL
1231 #define TMPBUFLEN 21
1232 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1233                                   size_t count, loff_t *ppos)
1234 {
1235         struct inode * inode = file_inode(file);
1236         struct task_struct *task = get_proc_task(inode);
1237         ssize_t length;
1238         char tmpbuf[TMPBUFLEN];
1239
1240         if (!task)
1241                 return -ESRCH;
1242         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1243                            from_kuid(file->f_cred->user_ns,
1244                                      audit_get_loginuid(task)));
1245         put_task_struct(task);
1246         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1247 }
1248
1249 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1250                                    size_t count, loff_t *ppos)
1251 {
1252         struct inode * inode = file_inode(file);
1253         uid_t loginuid;
1254         kuid_t kloginuid;
1255         int rv;
1256
1257         rcu_read_lock();
1258         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1259                 rcu_read_unlock();
1260                 return -EPERM;
1261         }
1262         rcu_read_unlock();
1263
1264         if (*ppos != 0) {
1265                 /* No partial writes. */
1266                 return -EINVAL;
1267         }
1268
1269         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1270         if (rv < 0)
1271                 return rv;
1272
1273         /* is userspace tring to explicitly UNSET the loginuid? */
1274         if (loginuid == AUDIT_UID_UNSET) {
1275                 kloginuid = INVALID_UID;
1276         } else {
1277                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1278                 if (!uid_valid(kloginuid))
1279                         return -EINVAL;
1280         }
1281
1282         rv = audit_set_loginuid(kloginuid);
1283         if (rv < 0)
1284                 return rv;
1285         return count;
1286 }
1287
1288 static const struct file_operations proc_loginuid_operations = {
1289         .read           = proc_loginuid_read,
1290         .write          = proc_loginuid_write,
1291         .llseek         = generic_file_llseek,
1292 };
1293
1294 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1295                                   size_t count, loff_t *ppos)
1296 {
1297         struct inode * inode = file_inode(file);
1298         struct task_struct *task = get_proc_task(inode);
1299         ssize_t length;
1300         char tmpbuf[TMPBUFLEN];
1301
1302         if (!task)
1303                 return -ESRCH;
1304         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1305                                 audit_get_sessionid(task));
1306         put_task_struct(task);
1307         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1308 }
1309
1310 static const struct file_operations proc_sessionid_operations = {
1311         .read           = proc_sessionid_read,
1312         .llseek         = generic_file_llseek,
1313 };
1314 #endif
1315
1316 #ifdef CONFIG_FAULT_INJECTION
1317 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1318                                       size_t count, loff_t *ppos)
1319 {
1320         struct task_struct *task = get_proc_task(file_inode(file));
1321         char buffer[PROC_NUMBUF];
1322         size_t len;
1323         int make_it_fail;
1324
1325         if (!task)
1326                 return -ESRCH;
1327         make_it_fail = task->make_it_fail;
1328         put_task_struct(task);
1329
1330         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1331
1332         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1333 }
1334
1335 static ssize_t proc_fault_inject_write(struct file * file,
1336                         const char __user * buf, size_t count, loff_t *ppos)
1337 {
1338         struct task_struct *task;
1339         char buffer[PROC_NUMBUF];
1340         int make_it_fail;
1341         int rv;
1342
1343         if (!capable(CAP_SYS_RESOURCE))
1344                 return -EPERM;
1345         memset(buffer, 0, sizeof(buffer));
1346         if (count > sizeof(buffer) - 1)
1347                 count = sizeof(buffer) - 1;
1348         if (copy_from_user(buffer, buf, count))
1349                 return -EFAULT;
1350         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1351         if (rv < 0)
1352                 return rv;
1353         if (make_it_fail < 0 || make_it_fail > 1)
1354                 return -EINVAL;
1355
1356         task = get_proc_task(file_inode(file));
1357         if (!task)
1358                 return -ESRCH;
1359         task->make_it_fail = make_it_fail;
1360         put_task_struct(task);
1361
1362         return count;
1363 }
1364
1365 static const struct file_operations proc_fault_inject_operations = {
1366         .read           = proc_fault_inject_read,
1367         .write          = proc_fault_inject_write,
1368         .llseek         = generic_file_llseek,
1369 };
1370 #endif
1371
1372
1373 #ifdef CONFIG_SCHED_DEBUG
1374 /*
1375  * Print out various scheduling related per-task fields:
1376  */
1377 static int sched_show(struct seq_file *m, void *v)
1378 {
1379         struct inode *inode = m->private;
1380         struct task_struct *p;
1381
1382         p = get_proc_task(inode);
1383         if (!p)
1384                 return -ESRCH;
1385         proc_sched_show_task(p, m);
1386
1387         put_task_struct(p);
1388
1389         return 0;
1390 }
1391
1392 static ssize_t
1393 sched_write(struct file *file, const char __user *buf,
1394             size_t count, loff_t *offset)
1395 {
1396         struct inode *inode = file_inode(file);
1397         struct task_struct *p;
1398
1399         p = get_proc_task(inode);
1400         if (!p)
1401                 return -ESRCH;
1402         proc_sched_set_task(p);
1403
1404         put_task_struct(p);
1405
1406         return count;
1407 }
1408
1409 static int sched_open(struct inode *inode, struct file *filp)
1410 {
1411         return single_open(filp, sched_show, inode);
1412 }
1413
1414 static const struct file_operations proc_pid_sched_operations = {
1415         .open           = sched_open,
1416         .read           = seq_read,
1417         .write          = sched_write,
1418         .llseek         = seq_lseek,
1419         .release        = single_release,
1420 };
1421
1422 #endif
1423
1424 #ifdef CONFIG_SCHED_AUTOGROUP
1425 /*
1426  * Print out autogroup related information:
1427  */
1428 static int sched_autogroup_show(struct seq_file *m, void *v)
1429 {
1430         struct inode *inode = m->private;
1431         struct task_struct *p;
1432
1433         p = get_proc_task(inode);
1434         if (!p)
1435                 return -ESRCH;
1436         proc_sched_autogroup_show_task(p, m);
1437
1438         put_task_struct(p);
1439
1440         return 0;
1441 }
1442
1443 static ssize_t
1444 sched_autogroup_write(struct file *file, const char __user *buf,
1445             size_t count, loff_t *offset)
1446 {
1447         struct inode *inode = file_inode(file);
1448         struct task_struct *p;
1449         char buffer[PROC_NUMBUF];
1450         int nice;
1451         int err;
1452
1453         memset(buffer, 0, sizeof(buffer));
1454         if (count > sizeof(buffer) - 1)
1455                 count = sizeof(buffer) - 1;
1456         if (copy_from_user(buffer, buf, count))
1457                 return -EFAULT;
1458
1459         err = kstrtoint(strstrip(buffer), 0, &nice);
1460         if (err < 0)
1461                 return err;
1462
1463         p = get_proc_task(inode);
1464         if (!p)
1465                 return -ESRCH;
1466
1467         err = proc_sched_autogroup_set_nice(p, nice);
1468         if (err)
1469                 count = err;
1470
1471         put_task_struct(p);
1472
1473         return count;
1474 }
1475
1476 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1477 {
1478         int ret;
1479
1480         ret = single_open(filp, sched_autogroup_show, NULL);
1481         if (!ret) {
1482                 struct seq_file *m = filp->private_data;
1483
1484                 m->private = inode;
1485         }
1486         return ret;
1487 }
1488
1489 static const struct file_operations proc_pid_sched_autogroup_operations = {
1490         .open           = sched_autogroup_open,
1491         .read           = seq_read,
1492         .write          = sched_autogroup_write,
1493         .llseek         = seq_lseek,
1494         .release        = single_release,
1495 };
1496
1497 #endif /* CONFIG_SCHED_AUTOGROUP */
1498
1499 static ssize_t comm_write(struct file *file, const char __user *buf,
1500                                 size_t count, loff_t *offset)
1501 {
1502         struct inode *inode = file_inode(file);
1503         struct task_struct *p;
1504         char buffer[TASK_COMM_LEN];
1505         const size_t maxlen = sizeof(buffer) - 1;
1506
1507         memset(buffer, 0, sizeof(buffer));
1508         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1509                 return -EFAULT;
1510
1511         p = get_proc_task(inode);
1512         if (!p)
1513                 return -ESRCH;
1514
1515         if (same_thread_group(current, p))
1516                 set_task_comm(p, buffer);
1517         else
1518                 count = -EINVAL;
1519
1520         put_task_struct(p);
1521
1522         return count;
1523 }
1524
1525 static int comm_show(struct seq_file *m, void *v)
1526 {
1527         struct inode *inode = m->private;
1528         struct task_struct *p;
1529
1530         p = get_proc_task(inode);
1531         if (!p)
1532                 return -ESRCH;
1533
1534         task_lock(p);
1535         seq_printf(m, "%s\n", p->comm);
1536         task_unlock(p);
1537
1538         put_task_struct(p);
1539
1540         return 0;
1541 }
1542
1543 static int comm_open(struct inode *inode, struct file *filp)
1544 {
1545         return single_open(filp, comm_show, inode);
1546 }
1547
1548 static const struct file_operations proc_pid_set_comm_operations = {
1549         .open           = comm_open,
1550         .read           = seq_read,
1551         .write          = comm_write,
1552         .llseek         = seq_lseek,
1553         .release        = single_release,
1554 };
1555
1556 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1557 {
1558         struct task_struct *task;
1559         struct mm_struct *mm;
1560         struct file *exe_file;
1561
1562         task = get_proc_task(d_inode(dentry));
1563         if (!task)
1564                 return -ENOENT;
1565         mm = get_task_mm(task);
1566         put_task_struct(task);
1567         if (!mm)
1568                 return -ENOENT;
1569         exe_file = get_mm_exe_file(mm);
1570         mmput(mm);
1571         if (exe_file) {
1572                 *exe_path = exe_file->f_path;
1573                 path_get(&exe_file->f_path);
1574                 fput(exe_file);
1575                 return 0;
1576         } else
1577                 return -ENOENT;
1578 }
1579
1580 static const char *proc_pid_get_link(struct dentry *dentry,
1581                                      struct inode *inode,
1582                                      struct delayed_call *done)
1583 {
1584         struct path path;
1585         int error = -EACCES;
1586
1587         if (!dentry)
1588                 return ERR_PTR(-ECHILD);
1589
1590         /* Are we allowed to snoop on the tasks file descriptors? */
1591         if (!proc_fd_access_allowed(inode))
1592                 goto out;
1593
1594         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1595         if (error)
1596                 goto out;
1597
1598         nd_jump_link(&path);
1599         return NULL;
1600 out:
1601         return ERR_PTR(error);
1602 }
1603
1604 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1605 {
1606         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1607         char *pathname;
1608         int len;
1609
1610         if (!tmp)
1611                 return -ENOMEM;
1612
1613         pathname = d_path(path, tmp, PAGE_SIZE);
1614         len = PTR_ERR(pathname);
1615         if (IS_ERR(pathname))
1616                 goto out;
1617         len = tmp + PAGE_SIZE - 1 - pathname;
1618
1619         if (len > buflen)
1620                 len = buflen;
1621         if (copy_to_user(buffer, pathname, len))
1622                 len = -EFAULT;
1623  out:
1624         free_page((unsigned long)tmp);
1625         return len;
1626 }
1627
1628 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1629 {
1630         int error = -EACCES;
1631         struct inode *inode = d_inode(dentry);
1632         struct path path;
1633
1634         /* Are we allowed to snoop on the tasks file descriptors? */
1635         if (!proc_fd_access_allowed(inode))
1636                 goto out;
1637
1638         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1639         if (error)
1640                 goto out;
1641
1642         error = do_proc_readlink(&path, buffer, buflen);
1643         path_put(&path);
1644 out:
1645         return error;
1646 }
1647
1648 const struct inode_operations proc_pid_link_inode_operations = {
1649         .readlink       = proc_pid_readlink,
1650         .get_link       = proc_pid_get_link,
1651         .setattr        = proc_setattr,
1652 };
1653
1654
1655 /* building an inode */
1656
1657 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1658 {
1659         struct inode * inode;
1660         struct proc_inode *ei;
1661         const struct cred *cred;
1662
1663         /* We need a new inode */
1664
1665         inode = new_inode(sb);
1666         if (!inode)
1667                 goto out;
1668
1669         /* Common stuff */
1670         ei = PROC_I(inode);
1671         inode->i_ino = get_next_ino();
1672         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1673         inode->i_op = &proc_def_inode_operations;
1674
1675         /*
1676          * grab the reference to task.
1677          */
1678         ei->pid = get_task_pid(task, PIDTYPE_PID);
1679         if (!ei->pid)
1680                 goto out_unlock;
1681
1682         if (task_dumpable(task)) {
1683                 rcu_read_lock();
1684                 cred = __task_cred(task);
1685                 inode->i_uid = cred->euid;
1686                 inode->i_gid = cred->egid;
1687                 rcu_read_unlock();
1688         }
1689         security_task_to_inode(task, inode);
1690
1691 out:
1692         return inode;
1693
1694 out_unlock:
1695         iput(inode);
1696         return NULL;
1697 }
1698
1699 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1700 {
1701         struct inode *inode = d_inode(dentry);
1702         struct task_struct *task;
1703         const struct cred *cred;
1704         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1705
1706         generic_fillattr(inode, stat);
1707
1708         rcu_read_lock();
1709         stat->uid = GLOBAL_ROOT_UID;
1710         stat->gid = GLOBAL_ROOT_GID;
1711         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1712         if (task) {
1713                 if (!has_pid_permissions(pid, task, 2)) {
1714                         rcu_read_unlock();
1715                         /*
1716                          * This doesn't prevent learning whether PID exists,
1717                          * it only makes getattr() consistent with readdir().
1718                          */
1719                         return -ENOENT;
1720                 }
1721                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1722                     task_dumpable(task)) {
1723                         cred = __task_cred(task);
1724                         stat->uid = cred->euid;
1725                         stat->gid = cred->egid;
1726                 }
1727         }
1728         rcu_read_unlock();
1729         return 0;
1730 }
1731
1732 /* dentry stuff */
1733
1734 /*
1735  *      Exceptional case: normally we are not allowed to unhash a busy
1736  * directory. In this case, however, we can do it - no aliasing problems
1737  * due to the way we treat inodes.
1738  *
1739  * Rewrite the inode's ownerships here because the owning task may have
1740  * performed a setuid(), etc.
1741  *
1742  * Before the /proc/pid/status file was created the only way to read
1743  * the effective uid of a /process was to stat /proc/pid.  Reading
1744  * /proc/pid/status is slow enough that procps and other packages
1745  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1746  * made this apply to all per process world readable and executable
1747  * directories.
1748  */
1749 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1750 {
1751         struct inode *inode;
1752         struct task_struct *task;
1753         const struct cred *cred;
1754
1755         if (flags & LOOKUP_RCU)
1756                 return -ECHILD;
1757
1758         inode = d_inode(dentry);
1759         task = get_proc_task(inode);
1760
1761         if (task) {
1762                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1763                     task_dumpable(task)) {
1764                         rcu_read_lock();
1765                         cred = __task_cred(task);
1766                         inode->i_uid = cred->euid;
1767                         inode->i_gid = cred->egid;
1768                         rcu_read_unlock();
1769                 } else {
1770                         inode->i_uid = GLOBAL_ROOT_UID;
1771                         inode->i_gid = GLOBAL_ROOT_GID;
1772                 }
1773                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1774                 security_task_to_inode(task, inode);
1775                 put_task_struct(task);
1776                 return 1;
1777         }
1778         return 0;
1779 }
1780
1781 static inline bool proc_inode_is_dead(struct inode *inode)
1782 {
1783         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1784 }
1785
1786 int pid_delete_dentry(const struct dentry *dentry)
1787 {
1788         /* Is the task we represent dead?
1789          * If so, then don't put the dentry on the lru list,
1790          * kill it immediately.
1791          */
1792         return proc_inode_is_dead(d_inode(dentry));
1793 }
1794
1795 const struct dentry_operations pid_dentry_operations =
1796 {
1797         .d_revalidate   = pid_revalidate,
1798         .d_delete       = pid_delete_dentry,
1799 };
1800
1801 /* Lookups */
1802
1803 /*
1804  * Fill a directory entry.
1805  *
1806  * If possible create the dcache entry and derive our inode number and
1807  * file type from dcache entry.
1808  *
1809  * Since all of the proc inode numbers are dynamically generated, the inode
1810  * numbers do not exist until the inode is cache.  This means creating the
1811  * the dcache entry in readdir is necessary to keep the inode numbers
1812  * reported by readdir in sync with the inode numbers reported
1813  * by stat.
1814  */
1815 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1816         const char *name, int len,
1817         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1818 {
1819         struct dentry *child, *dir = file->f_path.dentry;
1820         struct qstr qname = QSTR_INIT(name, len);
1821         struct inode *inode;
1822         unsigned type;
1823         ino_t ino;
1824
1825         child = d_hash_and_lookup(dir, &qname);
1826         if (!child) {
1827                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1828                 child = d_alloc_parallel(dir, &qname, &wq);
1829                 if (IS_ERR(child))
1830                         goto end_instantiate;
1831                 if (d_in_lookup(child)) {
1832                         int err = instantiate(d_inode(dir), child, task, ptr);
1833                         d_lookup_done(child);
1834                         if (err < 0) {
1835                                 dput(child);
1836                                 goto end_instantiate;
1837                         }
1838                 }
1839         }
1840         inode = d_inode(child);
1841         ino = inode->i_ino;
1842         type = inode->i_mode >> 12;
1843         dput(child);
1844         return dir_emit(ctx, name, len, ino, type);
1845
1846 end_instantiate:
1847         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1848 }
1849
1850 /*
1851  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1852  * which represent vma start and end addresses.
1853  */
1854 static int dname_to_vma_addr(struct dentry *dentry,
1855                              unsigned long *start, unsigned long *end)
1856 {
1857         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1858                 return -EINVAL;
1859
1860         return 0;
1861 }
1862
1863 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1864 {
1865         unsigned long vm_start, vm_end;
1866         bool exact_vma_exists = false;
1867         struct mm_struct *mm = NULL;
1868         struct task_struct *task;
1869         const struct cred *cred;
1870         struct inode *inode;
1871         int status = 0;
1872
1873         if (flags & LOOKUP_RCU)
1874                 return -ECHILD;
1875
1876         inode = d_inode(dentry);
1877         task = get_proc_task(inode);
1878         if (!task)
1879                 goto out_notask;
1880
1881         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1882         if (IS_ERR_OR_NULL(mm))
1883                 goto out;
1884
1885         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1886                 down_read(&mm->mmap_sem);
1887                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1888                 up_read(&mm->mmap_sem);
1889         }
1890
1891         mmput(mm);
1892
1893         if (exact_vma_exists) {
1894                 if (task_dumpable(task)) {
1895                         rcu_read_lock();
1896                         cred = __task_cred(task);
1897                         inode->i_uid = cred->euid;
1898                         inode->i_gid = cred->egid;
1899                         rcu_read_unlock();
1900                 } else {
1901                         inode->i_uid = GLOBAL_ROOT_UID;
1902                         inode->i_gid = GLOBAL_ROOT_GID;
1903                 }
1904                 security_task_to_inode(task, inode);
1905                 status = 1;
1906         }
1907
1908 out:
1909         put_task_struct(task);
1910
1911 out_notask:
1912         return status;
1913 }
1914
1915 static const struct dentry_operations tid_map_files_dentry_operations = {
1916         .d_revalidate   = map_files_d_revalidate,
1917         .d_delete       = pid_delete_dentry,
1918 };
1919
1920 static int map_files_get_link(struct dentry *dentry, struct path *path)
1921 {
1922         unsigned long vm_start, vm_end;
1923         struct vm_area_struct *vma;
1924         struct task_struct *task;
1925         struct mm_struct *mm;
1926         int rc;
1927
1928         rc = -ENOENT;
1929         task = get_proc_task(d_inode(dentry));
1930         if (!task)
1931                 goto out;
1932
1933         mm = get_task_mm(task);
1934         put_task_struct(task);
1935         if (!mm)
1936                 goto out;
1937
1938         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1939         if (rc)
1940                 goto out_mmput;
1941
1942         rc = -ENOENT;
1943         down_read(&mm->mmap_sem);
1944         vma = find_exact_vma(mm, vm_start, vm_end);
1945         if (vma && vma->vm_file) {
1946                 *path = vma->vm_file->f_path;
1947                 path_get(path);
1948                 rc = 0;
1949         }
1950         up_read(&mm->mmap_sem);
1951
1952 out_mmput:
1953         mmput(mm);
1954 out:
1955         return rc;
1956 }
1957
1958 struct map_files_info {
1959         fmode_t         mode;
1960         unsigned long   len;
1961         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1962 };
1963
1964 /*
1965  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1966  * symlinks may be used to bypass permissions on ancestor directories in the
1967  * path to the file in question.
1968  */
1969 static const char *
1970 proc_map_files_get_link(struct dentry *dentry,
1971                         struct inode *inode,
1972                         struct delayed_call *done)
1973 {
1974         if (!capable(CAP_SYS_ADMIN))
1975                 return ERR_PTR(-EPERM);
1976
1977         return proc_pid_get_link(dentry, inode, done);
1978 }
1979
1980 /*
1981  * Identical to proc_pid_link_inode_operations except for get_link()
1982  */
1983 static const struct inode_operations proc_map_files_link_inode_operations = {
1984         .readlink       = proc_pid_readlink,
1985         .get_link       = proc_map_files_get_link,
1986         .setattr        = proc_setattr,
1987 };
1988
1989 static int
1990 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1991                            struct task_struct *task, const void *ptr)
1992 {
1993         fmode_t mode = (fmode_t)(unsigned long)ptr;
1994         struct proc_inode *ei;
1995         struct inode *inode;
1996
1997         inode = proc_pid_make_inode(dir->i_sb, task);
1998         if (!inode)
1999                 return -ENOENT;
2000
2001         ei = PROC_I(inode);
2002         ei->op.proc_get_link = map_files_get_link;
2003
2004         inode->i_op = &proc_map_files_link_inode_operations;
2005         inode->i_size = 64;
2006         inode->i_mode = S_IFLNK;
2007
2008         if (mode & FMODE_READ)
2009                 inode->i_mode |= S_IRUSR;
2010         if (mode & FMODE_WRITE)
2011                 inode->i_mode |= S_IWUSR;
2012
2013         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2014         d_add(dentry, inode);
2015
2016         return 0;
2017 }
2018
2019 static struct dentry *proc_map_files_lookup(struct inode *dir,
2020                 struct dentry *dentry, unsigned int flags)
2021 {
2022         unsigned long vm_start, vm_end;
2023         struct vm_area_struct *vma;
2024         struct task_struct *task;
2025         int result;
2026         struct mm_struct *mm;
2027
2028         result = -ENOENT;
2029         task = get_proc_task(dir);
2030         if (!task)
2031                 goto out;
2032
2033         result = -EACCES;
2034         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2035                 goto out_put_task;
2036
2037         result = -ENOENT;
2038         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2039                 goto out_put_task;
2040
2041         mm = get_task_mm(task);
2042         if (!mm)
2043                 goto out_put_task;
2044
2045         down_read(&mm->mmap_sem);
2046         vma = find_exact_vma(mm, vm_start, vm_end);
2047         if (!vma)
2048                 goto out_no_vma;
2049
2050         if (vma->vm_file)
2051                 result = proc_map_files_instantiate(dir, dentry, task,
2052                                 (void *)(unsigned long)vma->vm_file->f_mode);
2053
2054 out_no_vma:
2055         up_read(&mm->mmap_sem);
2056         mmput(mm);
2057 out_put_task:
2058         put_task_struct(task);
2059 out:
2060         return ERR_PTR(result);
2061 }
2062
2063 static const struct inode_operations proc_map_files_inode_operations = {
2064         .lookup         = proc_map_files_lookup,
2065         .permission     = proc_fd_permission,
2066         .setattr        = proc_setattr,
2067 };
2068
2069 static int
2070 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2071 {
2072         struct vm_area_struct *vma;
2073         struct task_struct *task;
2074         struct mm_struct *mm;
2075         unsigned long nr_files, pos, i;
2076         struct flex_array *fa = NULL;
2077         struct map_files_info info;
2078         struct map_files_info *p;
2079         int ret;
2080
2081         ret = -ENOENT;
2082         task = get_proc_task(file_inode(file));
2083         if (!task)
2084                 goto out;
2085
2086         ret = -EACCES;
2087         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2088                 goto out_put_task;
2089
2090         ret = 0;
2091         if (!dir_emit_dots(file, ctx))
2092                 goto out_put_task;
2093
2094         mm = get_task_mm(task);
2095         if (!mm)
2096                 goto out_put_task;
2097         down_read(&mm->mmap_sem);
2098
2099         nr_files = 0;
2100
2101         /*
2102          * We need two passes here:
2103          *
2104          *  1) Collect vmas of mapped files with mmap_sem taken
2105          *  2) Release mmap_sem and instantiate entries
2106          *
2107          * otherwise we get lockdep complained, since filldir()
2108          * routine might require mmap_sem taken in might_fault().
2109          */
2110
2111         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2112                 if (vma->vm_file && ++pos > ctx->pos)
2113                         nr_files++;
2114         }
2115
2116         if (nr_files) {
2117                 fa = flex_array_alloc(sizeof(info), nr_files,
2118                                         GFP_KERNEL);
2119                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2120                                                 GFP_KERNEL)) {
2121                         ret = -ENOMEM;
2122                         if (fa)
2123                                 flex_array_free(fa);
2124                         up_read(&mm->mmap_sem);
2125                         mmput(mm);
2126                         goto out_put_task;
2127                 }
2128                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2129                                 vma = vma->vm_next) {
2130                         if (!vma->vm_file)
2131                                 continue;
2132                         if (++pos <= ctx->pos)
2133                                 continue;
2134
2135                         info.mode = vma->vm_file->f_mode;
2136                         info.len = snprintf(info.name,
2137                                         sizeof(info.name), "%lx-%lx",
2138                                         vma->vm_start, vma->vm_end);
2139                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2140                                 BUG();
2141                 }
2142         }
2143         up_read(&mm->mmap_sem);
2144
2145         for (i = 0; i < nr_files; i++) {
2146                 p = flex_array_get(fa, i);
2147                 if (!proc_fill_cache(file, ctx,
2148                                       p->name, p->len,
2149                                       proc_map_files_instantiate,
2150                                       task,
2151                                       (void *)(unsigned long)p->mode))
2152                         break;
2153                 ctx->pos++;
2154         }
2155         if (fa)
2156                 flex_array_free(fa);
2157         mmput(mm);
2158
2159 out_put_task:
2160         put_task_struct(task);
2161 out:
2162         return ret;
2163 }
2164
2165 static const struct file_operations proc_map_files_operations = {
2166         .read           = generic_read_dir,
2167         .iterate_shared = proc_map_files_readdir,
2168         .llseek         = generic_file_llseek,
2169 };
2170
2171 #ifdef CONFIG_CHECKPOINT_RESTORE
2172 struct timers_private {
2173         struct pid *pid;
2174         struct task_struct *task;
2175         struct sighand_struct *sighand;
2176         struct pid_namespace *ns;
2177         unsigned long flags;
2178 };
2179
2180 static void *timers_start(struct seq_file *m, loff_t *pos)
2181 {
2182         struct timers_private *tp = m->private;
2183
2184         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2185         if (!tp->task)
2186                 return ERR_PTR(-ESRCH);
2187
2188         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2189         if (!tp->sighand)
2190                 return ERR_PTR(-ESRCH);
2191
2192         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2193 }
2194
2195 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2196 {
2197         struct timers_private *tp = m->private;
2198         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2199 }
2200
2201 static void timers_stop(struct seq_file *m, void *v)
2202 {
2203         struct timers_private *tp = m->private;
2204
2205         if (tp->sighand) {
2206                 unlock_task_sighand(tp->task, &tp->flags);
2207                 tp->sighand = NULL;
2208         }
2209
2210         if (tp->task) {
2211                 put_task_struct(tp->task);
2212                 tp->task = NULL;
2213         }
2214 }
2215
2216 static int show_timer(struct seq_file *m, void *v)
2217 {
2218         struct k_itimer *timer;
2219         struct timers_private *tp = m->private;
2220         int notify;
2221         static const char * const nstr[] = {
2222                 [SIGEV_SIGNAL] = "signal",
2223                 [SIGEV_NONE] = "none",
2224                 [SIGEV_THREAD] = "thread",
2225         };
2226
2227         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2228         notify = timer->it_sigev_notify;
2229
2230         seq_printf(m, "ID: %d\n", timer->it_id);
2231         seq_printf(m, "signal: %d/%p\n",
2232                    timer->sigq->info.si_signo,
2233                    timer->sigq->info.si_value.sival_ptr);
2234         seq_printf(m, "notify: %s/%s.%d\n",
2235                    nstr[notify & ~SIGEV_THREAD_ID],
2236                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2237                    pid_nr_ns(timer->it_pid, tp->ns));
2238         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2239
2240         return 0;
2241 }
2242
2243 static const struct seq_operations proc_timers_seq_ops = {
2244         .start  = timers_start,
2245         .next   = timers_next,
2246         .stop   = timers_stop,
2247         .show   = show_timer,
2248 };
2249
2250 static int proc_timers_open(struct inode *inode, struct file *file)
2251 {
2252         struct timers_private *tp;
2253
2254         tp = __seq_open_private(file, &proc_timers_seq_ops,
2255                         sizeof(struct timers_private));
2256         if (!tp)
2257                 return -ENOMEM;
2258
2259         tp->pid = proc_pid(inode);
2260         tp->ns = inode->i_sb->s_fs_info;
2261         return 0;
2262 }
2263
2264 static const struct file_operations proc_timers_operations = {
2265         .open           = proc_timers_open,
2266         .read           = seq_read,
2267         .llseek         = seq_lseek,
2268         .release        = seq_release_private,
2269 };
2270 #endif
2271
2272 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2273                                         size_t count, loff_t *offset)
2274 {
2275         struct inode *inode = file_inode(file);
2276         struct task_struct *p;
2277         u64 slack_ns;
2278         int err;
2279
2280         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2281         if (err < 0)
2282                 return err;
2283
2284         p = get_proc_task(inode);
2285         if (!p)
2286                 return -ESRCH;
2287
2288         if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2289                 task_lock(p);
2290                 if (slack_ns == 0)
2291                         p->timer_slack_ns = p->default_timer_slack_ns;
2292                 else
2293                         p->timer_slack_ns = slack_ns;
2294                 task_unlock(p);
2295         } else
2296                 count = -EPERM;
2297
2298         put_task_struct(p);
2299
2300         return count;
2301 }
2302
2303 static int timerslack_ns_show(struct seq_file *m, void *v)
2304 {
2305         struct inode *inode = m->private;
2306         struct task_struct *p;
2307         int err =  0;
2308
2309         p = get_proc_task(inode);
2310         if (!p)
2311                 return -ESRCH;
2312
2313         if (ptrace_may_access(p, PTRACE_MODE_ATTACH_FSCREDS)) {
2314                 task_lock(p);
2315                 seq_printf(m, "%llu\n", p->timer_slack_ns);
2316                 task_unlock(p);
2317         } else
2318                 err = -EPERM;
2319
2320         put_task_struct(p);
2321
2322         return err;
2323 }
2324
2325 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2326 {
2327         return single_open(filp, timerslack_ns_show, inode);
2328 }
2329
2330 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2331         .open           = timerslack_ns_open,
2332         .read           = seq_read,
2333         .write          = timerslack_ns_write,
2334         .llseek         = seq_lseek,
2335         .release        = single_release,
2336 };
2337
2338 static int proc_pident_instantiate(struct inode *dir,
2339         struct dentry *dentry, struct task_struct *task, const void *ptr)
2340 {
2341         const struct pid_entry *p = ptr;
2342         struct inode *inode;
2343         struct proc_inode *ei;
2344
2345         inode = proc_pid_make_inode(dir->i_sb, task);
2346         if (!inode)
2347                 goto out;
2348
2349         ei = PROC_I(inode);
2350         inode->i_mode = p->mode;
2351         if (S_ISDIR(inode->i_mode))
2352                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2353         if (p->iop)
2354                 inode->i_op = p->iop;
2355         if (p->fop)
2356                 inode->i_fop = p->fop;
2357         ei->op = p->op;
2358         d_set_d_op(dentry, &pid_dentry_operations);
2359         d_add(dentry, inode);
2360         /* Close the race of the process dying before we return the dentry */
2361         if (pid_revalidate(dentry, 0))
2362                 return 0;
2363 out:
2364         return -ENOENT;
2365 }
2366
2367 static struct dentry *proc_pident_lookup(struct inode *dir,
2368                                          struct dentry *dentry,
2369                                          const struct pid_entry *ents,
2370                                          unsigned int nents)
2371 {
2372         int error;
2373         struct task_struct *task = get_proc_task(dir);
2374         const struct pid_entry *p, *last;
2375
2376         error = -ENOENT;
2377
2378         if (!task)
2379                 goto out_no_task;
2380
2381         /*
2382          * Yes, it does not scale. And it should not. Don't add
2383          * new entries into /proc/<tgid>/ without very good reasons.
2384          */
2385         last = &ents[nents - 1];
2386         for (p = ents; p <= last; p++) {
2387                 if (p->len != dentry->d_name.len)
2388                         continue;
2389                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2390                         break;
2391         }
2392         if (p > last)
2393                 goto out;
2394
2395         error = proc_pident_instantiate(dir, dentry, task, p);
2396 out:
2397         put_task_struct(task);
2398 out_no_task:
2399         return ERR_PTR(error);
2400 }
2401
2402 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2403                 const struct pid_entry *ents, unsigned int nents)
2404 {
2405         struct task_struct *task = get_proc_task(file_inode(file));
2406         const struct pid_entry *p;
2407
2408         if (!task)
2409                 return -ENOENT;
2410
2411         if (!dir_emit_dots(file, ctx))
2412                 goto out;
2413
2414         if (ctx->pos >= nents + 2)
2415                 goto out;
2416
2417         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2418                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2419                                 proc_pident_instantiate, task, p))
2420                         break;
2421                 ctx->pos++;
2422         }
2423 out:
2424         put_task_struct(task);
2425         return 0;
2426 }
2427
2428 #ifdef CONFIG_SECURITY
2429 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2430                                   size_t count, loff_t *ppos)
2431 {
2432         struct inode * inode = file_inode(file);
2433         char *p = NULL;
2434         ssize_t length;
2435         struct task_struct *task = get_proc_task(inode);
2436
2437         if (!task)
2438                 return -ESRCH;
2439
2440         length = security_getprocattr(task,
2441                                       (char*)file->f_path.dentry->d_name.name,
2442                                       &p);
2443         put_task_struct(task);
2444         if (length > 0)
2445                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2446         kfree(p);
2447         return length;
2448 }
2449
2450 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2451                                    size_t count, loff_t *ppos)
2452 {
2453         struct inode * inode = file_inode(file);
2454         void *page;
2455         ssize_t length;
2456         struct task_struct *task = get_proc_task(inode);
2457
2458         length = -ESRCH;
2459         if (!task)
2460                 goto out_no_task;
2461         if (count > PAGE_SIZE)
2462                 count = PAGE_SIZE;
2463
2464         /* No partial writes. */
2465         length = -EINVAL;
2466         if (*ppos != 0)
2467                 goto out;
2468
2469         page = memdup_user(buf, count);
2470         if (IS_ERR(page)) {
2471                 length = PTR_ERR(page);
2472                 goto out;
2473         }
2474
2475         /* Guard against adverse ptrace interaction */
2476         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2477         if (length < 0)
2478                 goto out_free;
2479
2480         length = security_setprocattr(task,
2481                                       (char*)file->f_path.dentry->d_name.name,
2482                                       page, count);
2483         mutex_unlock(&task->signal->cred_guard_mutex);
2484 out_free:
2485         kfree(page);
2486 out:
2487         put_task_struct(task);
2488 out_no_task:
2489         return length;
2490 }
2491
2492 static const struct file_operations proc_pid_attr_operations = {
2493         .read           = proc_pid_attr_read,
2494         .write          = proc_pid_attr_write,
2495         .llseek         = generic_file_llseek,
2496 };
2497
2498 static const struct pid_entry attr_dir_stuff[] = {
2499         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2500         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2501         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2502         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2503         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2504         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2505 };
2506
2507 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2508 {
2509         return proc_pident_readdir(file, ctx,
2510                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2511 }
2512
2513 static const struct file_operations proc_attr_dir_operations = {
2514         .read           = generic_read_dir,
2515         .iterate_shared = proc_attr_dir_readdir,
2516         .llseek         = generic_file_llseek,
2517 };
2518
2519 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2520                                 struct dentry *dentry, unsigned int flags)
2521 {
2522         return proc_pident_lookup(dir, dentry,
2523                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2524 }
2525
2526 static const struct inode_operations proc_attr_dir_inode_operations = {
2527         .lookup         = proc_attr_dir_lookup,
2528         .getattr        = pid_getattr,
2529         .setattr        = proc_setattr,
2530 };
2531
2532 #endif
2533
2534 #ifdef CONFIG_ELF_CORE
2535 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2536                                          size_t count, loff_t *ppos)
2537 {
2538         struct task_struct *task = get_proc_task(file_inode(file));
2539         struct mm_struct *mm;
2540         char buffer[PROC_NUMBUF];
2541         size_t len;
2542         int ret;
2543
2544         if (!task)
2545                 return -ESRCH;
2546
2547         ret = 0;
2548         mm = get_task_mm(task);
2549         if (mm) {
2550                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2551                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2552                                 MMF_DUMP_FILTER_SHIFT));
2553                 mmput(mm);
2554                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2555         }
2556
2557         put_task_struct(task);
2558
2559         return ret;
2560 }
2561
2562 static ssize_t proc_coredump_filter_write(struct file *file,
2563                                           const char __user *buf,
2564                                           size_t count,
2565                                           loff_t *ppos)
2566 {
2567         struct task_struct *task;
2568         struct mm_struct *mm;
2569         unsigned int val;
2570         int ret;
2571         int i;
2572         unsigned long mask;
2573
2574         ret = kstrtouint_from_user(buf, count, 0, &val);
2575         if (ret < 0)
2576                 return ret;
2577
2578         ret = -ESRCH;
2579         task = get_proc_task(file_inode(file));
2580         if (!task)
2581                 goto out_no_task;
2582
2583         mm = get_task_mm(task);
2584         if (!mm)
2585                 goto out_no_mm;
2586         ret = 0;
2587
2588         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2589                 if (val & mask)
2590                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2591                 else
2592                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2593         }
2594
2595         mmput(mm);
2596  out_no_mm:
2597         put_task_struct(task);
2598  out_no_task:
2599         if (ret < 0)
2600                 return ret;
2601         return count;
2602 }
2603
2604 static const struct file_operations proc_coredump_filter_operations = {
2605         .read           = proc_coredump_filter_read,
2606         .write          = proc_coredump_filter_write,
2607         .llseek         = generic_file_llseek,
2608 };
2609 #endif
2610
2611 #ifdef CONFIG_TASK_IO_ACCOUNTING
2612 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2613 {
2614         struct task_io_accounting acct = task->ioac;
2615         unsigned long flags;
2616         int result;
2617
2618         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2619         if (result)
2620                 return result;
2621
2622         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2623                 result = -EACCES;
2624                 goto out_unlock;
2625         }
2626
2627         if (whole && lock_task_sighand(task, &flags)) {
2628                 struct task_struct *t = task;
2629
2630                 task_io_accounting_add(&acct, &task->signal->ioac);
2631                 while_each_thread(task, t)
2632                         task_io_accounting_add(&acct, &t->ioac);
2633
2634                 unlock_task_sighand(task, &flags);
2635         }
2636         seq_printf(m,
2637                    "rchar: %llu\n"
2638                    "wchar: %llu\n"
2639                    "syscr: %llu\n"
2640                    "syscw: %llu\n"
2641                    "read_bytes: %llu\n"
2642                    "write_bytes: %llu\n"
2643                    "cancelled_write_bytes: %llu\n",
2644                    (unsigned long long)acct.rchar,
2645                    (unsigned long long)acct.wchar,
2646                    (unsigned long long)acct.syscr,
2647                    (unsigned long long)acct.syscw,
2648                    (unsigned long long)acct.read_bytes,
2649                    (unsigned long long)acct.write_bytes,
2650                    (unsigned long long)acct.cancelled_write_bytes);
2651         result = 0;
2652
2653 out_unlock:
2654         mutex_unlock(&task->signal->cred_guard_mutex);
2655         return result;
2656 }
2657
2658 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2659                                   struct pid *pid, struct task_struct *task)
2660 {
2661         return do_io_accounting(task, m, 0);
2662 }
2663
2664 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2665                                    struct pid *pid, struct task_struct *task)
2666 {
2667         return do_io_accounting(task, m, 1);
2668 }
2669 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2670
2671 #ifdef CONFIG_USER_NS
2672 static int proc_id_map_open(struct inode *inode, struct file *file,
2673         const struct seq_operations *seq_ops)
2674 {
2675         struct user_namespace *ns = NULL;
2676         struct task_struct *task;
2677         struct seq_file *seq;
2678         int ret = -EINVAL;
2679
2680         task = get_proc_task(inode);
2681         if (task) {
2682                 rcu_read_lock();
2683                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2684                 rcu_read_unlock();
2685                 put_task_struct(task);
2686         }
2687         if (!ns)
2688                 goto err;
2689
2690         ret = seq_open(file, seq_ops);
2691         if (ret)
2692                 goto err_put_ns;
2693
2694         seq = file->private_data;
2695         seq->private = 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_id_map_release(struct inode *inode, struct file *file)
2705 {
2706         struct seq_file *seq = file->private_data;
2707         struct user_namespace *ns = seq->private;
2708         put_user_ns(ns);
2709         return seq_release(inode, file);
2710 }
2711
2712 static int proc_uid_map_open(struct inode *inode, struct file *file)
2713 {
2714         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2715 }
2716
2717 static int proc_gid_map_open(struct inode *inode, struct file *file)
2718 {
2719         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2720 }
2721
2722 static int proc_projid_map_open(struct inode *inode, struct file *file)
2723 {
2724         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2725 }
2726
2727 static const struct file_operations proc_uid_map_operations = {
2728         .open           = proc_uid_map_open,
2729         .write          = proc_uid_map_write,
2730         .read           = seq_read,
2731         .llseek         = seq_lseek,
2732         .release        = proc_id_map_release,
2733 };
2734
2735 static const struct file_operations proc_gid_map_operations = {
2736         .open           = proc_gid_map_open,
2737         .write          = proc_gid_map_write,
2738         .read           = seq_read,
2739         .llseek         = seq_lseek,
2740         .release        = proc_id_map_release,
2741 };
2742
2743 static const struct file_operations proc_projid_map_operations = {
2744         .open           = proc_projid_map_open,
2745         .write          = proc_projid_map_write,
2746         .read           = seq_read,
2747         .llseek         = seq_lseek,
2748         .release        = proc_id_map_release,
2749 };
2750
2751 static int proc_setgroups_open(struct inode *inode, struct file *file)
2752 {
2753         struct user_namespace *ns = NULL;
2754         struct task_struct *task;
2755         int ret;
2756
2757         ret = -ESRCH;
2758         task = get_proc_task(inode);
2759         if (task) {
2760                 rcu_read_lock();
2761                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2762                 rcu_read_unlock();
2763                 put_task_struct(task);
2764         }
2765         if (!ns)
2766                 goto err;
2767
2768         if (file->f_mode & FMODE_WRITE) {
2769                 ret = -EACCES;
2770                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2771                         goto err_put_ns;
2772         }
2773
2774         ret = single_open(file, &proc_setgroups_show, ns);
2775         if (ret)
2776                 goto err_put_ns;
2777
2778         return 0;
2779 err_put_ns:
2780         put_user_ns(ns);
2781 err:
2782         return ret;
2783 }
2784
2785 static int proc_setgroups_release(struct inode *inode, struct file *file)
2786 {
2787         struct seq_file *seq = file->private_data;
2788         struct user_namespace *ns = seq->private;
2789         int ret = single_release(inode, file);
2790         put_user_ns(ns);
2791         return ret;
2792 }
2793
2794 static const struct file_operations proc_setgroups_operations = {
2795         .open           = proc_setgroups_open,
2796         .write          = proc_setgroups_write,
2797         .read           = seq_read,
2798         .llseek         = seq_lseek,
2799         .release        = proc_setgroups_release,
2800 };
2801 #endif /* CONFIG_USER_NS */
2802
2803 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2804                                 struct pid *pid, struct task_struct *task)
2805 {
2806         int err = lock_trace(task);
2807         if (!err) {
2808                 seq_printf(m, "%08x\n", task->personality);
2809                 unlock_trace(task);
2810         }
2811         return err;
2812 }
2813
2814 /*
2815  * Thread groups
2816  */
2817 static const struct file_operations proc_task_operations;
2818 static const struct inode_operations proc_task_inode_operations;
2819
2820 static const struct pid_entry tgid_base_stuff[] = {
2821         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2822         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2823         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2824         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2825         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2826 #ifdef CONFIG_NET
2827         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2828 #endif
2829         REG("environ",    S_IRUSR, proc_environ_operations),
2830         ONE("auxv",       S_IRUSR, proc_pid_auxv),
2831         ONE("status",     S_IRUGO, proc_pid_status),
2832         ONE("personality", S_IRUSR, proc_pid_personality),
2833         ONE("limits",     S_IRUGO, proc_pid_limits),
2834 #ifdef CONFIG_SCHED_DEBUG
2835         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2836 #endif
2837 #ifdef CONFIG_SCHED_AUTOGROUP
2838         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2839 #endif
2840         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2841 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2842         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2843 #endif
2844         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2845         ONE("stat",       S_IRUGO, proc_tgid_stat),
2846         ONE("statm",      S_IRUGO, proc_pid_statm),
2847         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2848 #ifdef CONFIG_NUMA
2849         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2850 #endif
2851         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2852         LNK("cwd",        proc_cwd_link),
2853         LNK("root",       proc_root_link),
2854         LNK("exe",        proc_exe_link),
2855         REG("mounts",     S_IRUGO, proc_mounts_operations),
2856         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2857         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2858 #ifdef CONFIG_PROC_PAGE_MONITOR
2859         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2860         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2861         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2862 #endif
2863 #ifdef CONFIG_SECURITY
2864         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2865 #endif
2866 #ifdef CONFIG_KALLSYMS
2867         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2868 #endif
2869 #ifdef CONFIG_STACKTRACE
2870         ONE("stack",      S_IRUSR, proc_pid_stack),
2871 #endif
2872 #ifdef CONFIG_SCHED_INFO
2873         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2874 #endif
2875 #ifdef CONFIG_LATENCYTOP
2876         REG("latency",  S_IRUGO, proc_lstats_operations),
2877 #endif
2878 #ifdef CONFIG_PROC_PID_CPUSET
2879         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2880 #endif
2881 #ifdef CONFIG_CGROUPS
2882         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2883 #endif
2884         ONE("oom_score",  S_IRUGO, proc_oom_score),
2885         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2886         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2887 #ifdef CONFIG_AUDITSYSCALL
2888         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2889         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2890 #endif
2891 #ifdef CONFIG_FAULT_INJECTION
2892         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2893 #endif
2894 #ifdef CONFIG_ELF_CORE
2895         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2896 #endif
2897 #ifdef CONFIG_TASK_IO_ACCOUNTING
2898         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2899 #endif
2900 #ifdef CONFIG_HARDWALL
2901         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2902 #endif
2903 #ifdef CONFIG_USER_NS
2904         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2905         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2906         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2907         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2908 #endif
2909 #ifdef CONFIG_CHECKPOINT_RESTORE
2910         REG("timers",     S_IRUGO, proc_timers_operations),
2911 #endif
2912         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2913 };
2914
2915 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2916 {
2917         return proc_pident_readdir(file, ctx,
2918                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2919 }
2920
2921 static const struct file_operations proc_tgid_base_operations = {
2922         .read           = generic_read_dir,
2923         .iterate_shared = proc_tgid_base_readdir,
2924         .llseek         = generic_file_llseek,
2925 };
2926
2927 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2928 {
2929         return proc_pident_lookup(dir, dentry,
2930                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2931 }
2932
2933 static const struct inode_operations proc_tgid_base_inode_operations = {
2934         .lookup         = proc_tgid_base_lookup,
2935         .getattr        = pid_getattr,
2936         .setattr        = proc_setattr,
2937         .permission     = proc_pid_permission,
2938 };
2939
2940 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2941 {
2942         struct dentry *dentry, *leader, *dir;
2943         char buf[PROC_NUMBUF];
2944         struct qstr name;
2945
2946         name.name = buf;
2947         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2948         /* no ->d_hash() rejects on procfs */
2949         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2950         if (dentry) {
2951                 d_invalidate(dentry);
2952                 dput(dentry);
2953         }
2954
2955         if (pid == tgid)
2956                 return;
2957
2958         name.name = buf;
2959         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2960         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2961         if (!leader)
2962                 goto out;
2963
2964         name.name = "task";
2965         name.len = strlen(name.name);
2966         dir = d_hash_and_lookup(leader, &name);
2967         if (!dir)
2968                 goto out_put_leader;
2969
2970         name.name = buf;
2971         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2972         dentry = d_hash_and_lookup(dir, &name);
2973         if (dentry) {
2974                 d_invalidate(dentry);
2975                 dput(dentry);
2976         }
2977
2978         dput(dir);
2979 out_put_leader:
2980         dput(leader);
2981 out:
2982         return;
2983 }
2984
2985 /**
2986  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2987  * @task: task that should be flushed.
2988  *
2989  * When flushing dentries from proc, one needs to flush them from global
2990  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2991  * in. This call is supposed to do all of this job.
2992  *
2993  * Looks in the dcache for
2994  * /proc/@pid
2995  * /proc/@tgid/task/@pid
2996  * if either directory is present flushes it and all of it'ts children
2997  * from the dcache.
2998  *
2999  * It is safe and reasonable to cache /proc entries for a task until
3000  * that task exits.  After that they just clog up the dcache with
3001  * useless entries, possibly causing useful dcache entries to be
3002  * flushed instead.  This routine is proved to flush those useless
3003  * dcache entries at process exit time.
3004  *
3005  * NOTE: This routine is just an optimization so it does not guarantee
3006  *       that no dcache entries will exist at process exit time it
3007  *       just makes it very unlikely that any will persist.
3008  */
3009
3010 void proc_flush_task(struct task_struct *task)
3011 {
3012         int i;
3013         struct pid *pid, *tgid;
3014         struct upid *upid;
3015
3016         pid = task_pid(task);
3017         tgid = task_tgid(task);
3018
3019         for (i = 0; i <= pid->level; i++) {
3020                 upid = &pid->numbers[i];
3021                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3022                                         tgid->numbers[i].nr);
3023         }
3024 }
3025
3026 static int proc_pid_instantiate(struct inode *dir,
3027                                    struct dentry * dentry,
3028                                    struct task_struct *task, const void *ptr)
3029 {
3030         struct inode *inode;
3031
3032         inode = proc_pid_make_inode(dir->i_sb, task);
3033         if (!inode)
3034                 goto out;
3035
3036         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3037         inode->i_op = &proc_tgid_base_inode_operations;
3038         inode->i_fop = &proc_tgid_base_operations;
3039         inode->i_flags|=S_IMMUTABLE;
3040
3041         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3042                                                   ARRAY_SIZE(tgid_base_stuff)));
3043
3044         d_set_d_op(dentry, &pid_dentry_operations);
3045
3046         d_add(dentry, inode);
3047         /* Close the race of the process dying before we return the dentry */
3048         if (pid_revalidate(dentry, 0))
3049                 return 0;
3050 out:
3051         return -ENOENT;
3052 }
3053
3054 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3055 {
3056         int result = -ENOENT;
3057         struct task_struct *task;
3058         unsigned tgid;
3059         struct pid_namespace *ns;
3060
3061         tgid = name_to_int(&dentry->d_name);
3062         if (tgid == ~0U)
3063                 goto out;
3064
3065         ns = dentry->d_sb->s_fs_info;
3066         rcu_read_lock();
3067         task = find_task_by_pid_ns(tgid, ns);
3068         if (task)
3069                 get_task_struct(task);
3070         rcu_read_unlock();
3071         if (!task)
3072                 goto out;
3073
3074         result = proc_pid_instantiate(dir, dentry, task, NULL);
3075         put_task_struct(task);
3076 out:
3077         return ERR_PTR(result);
3078 }
3079
3080 /*
3081  * Find the first task with tgid >= tgid
3082  *
3083  */
3084 struct tgid_iter {
3085         unsigned int tgid;
3086         struct task_struct *task;
3087 };
3088 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3089 {
3090         struct pid *pid;
3091
3092         if (iter.task)
3093                 put_task_struct(iter.task);
3094         rcu_read_lock();
3095 retry:
3096         iter.task = NULL;
3097         pid = find_ge_pid(iter.tgid, ns);
3098         if (pid) {
3099                 iter.tgid = pid_nr_ns(pid, ns);
3100                 iter.task = pid_task(pid, PIDTYPE_PID);
3101                 /* What we to know is if the pid we have find is the
3102                  * pid of a thread_group_leader.  Testing for task
3103                  * being a thread_group_leader is the obvious thing
3104                  * todo but there is a window when it fails, due to
3105                  * the pid transfer logic in de_thread.
3106                  *
3107                  * So we perform the straight forward test of seeing
3108                  * if the pid we have found is the pid of a thread
3109                  * group leader, and don't worry if the task we have
3110                  * found doesn't happen to be a thread group leader.
3111                  * As we don't care in the case of readdir.
3112                  */
3113                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3114                         iter.tgid += 1;
3115                         goto retry;
3116                 }
3117                 get_task_struct(iter.task);
3118         }
3119         rcu_read_unlock();
3120         return iter;
3121 }
3122
3123 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3124
3125 /* for the /proc/ directory itself, after non-process stuff has been done */
3126 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3127 {
3128         struct tgid_iter iter;
3129         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3130         loff_t pos = ctx->pos;
3131
3132         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3133                 return 0;
3134
3135         if (pos == TGID_OFFSET - 2) {
3136                 struct inode *inode = d_inode(ns->proc_self);
3137                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3138                         return 0;
3139                 ctx->pos = pos = pos + 1;
3140         }
3141         if (pos == TGID_OFFSET - 1) {
3142                 struct inode *inode = d_inode(ns->proc_thread_self);
3143                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3144                         return 0;
3145                 ctx->pos = pos = pos + 1;
3146         }
3147         iter.tgid = pos - TGID_OFFSET;
3148         iter.task = NULL;
3149         for (iter = next_tgid(ns, iter);
3150              iter.task;
3151              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3152                 char name[PROC_NUMBUF];
3153                 int len;
3154                 if (!has_pid_permissions(ns, iter.task, 2))
3155                         continue;
3156
3157                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3158                 ctx->pos = iter.tgid + TGID_OFFSET;
3159                 if (!proc_fill_cache(file, ctx, name, len,
3160                                      proc_pid_instantiate, iter.task, NULL)) {
3161                         put_task_struct(iter.task);
3162                         return 0;
3163                 }
3164         }
3165         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3166         return 0;
3167 }
3168
3169 /*
3170  * proc_tid_comm_permission is a special permission function exclusively
3171  * used for the node /proc/<pid>/task/<tid>/comm.
3172  * It bypasses generic permission checks in the case where a task of the same
3173  * task group attempts to access the node.
3174  * The rationale behind this is that glibc and bionic access this node for
3175  * cross thread naming (pthread_set/getname_np(!self)). However, if
3176  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3177  * which locks out the cross thread naming implementation.
3178  * This function makes sure that the node is always accessible for members of
3179  * same thread group.
3180  */
3181 static int proc_tid_comm_permission(struct inode *inode, int mask)
3182 {
3183         bool is_same_tgroup;
3184         struct task_struct *task;
3185
3186         task = get_proc_task(inode);
3187         if (!task)
3188                 return -ESRCH;
3189         is_same_tgroup = same_thread_group(current, task);
3190         put_task_struct(task);
3191
3192         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3193                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3194                  * read or written by the members of the corresponding
3195                  * thread group.
3196                  */
3197                 return 0;
3198         }
3199
3200         return generic_permission(inode, mask);
3201 }
3202
3203 static const struct inode_operations proc_tid_comm_inode_operations = {
3204                 .permission = proc_tid_comm_permission,
3205 };
3206
3207 /*
3208  * Tasks
3209  */
3210 static const struct pid_entry tid_base_stuff[] = {
3211         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3212         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3213         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3214 #ifdef CONFIG_NET
3215         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3216 #endif
3217         REG("environ",   S_IRUSR, proc_environ_operations),
3218         ONE("auxv",      S_IRUSR, proc_pid_auxv),
3219         ONE("status",    S_IRUGO, proc_pid_status),
3220         ONE("personality", S_IRUSR, proc_pid_personality),
3221         ONE("limits",    S_IRUGO, proc_pid_limits),
3222 #ifdef CONFIG_SCHED_DEBUG
3223         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3224 #endif
3225         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3226                          &proc_tid_comm_inode_operations,
3227                          &proc_pid_set_comm_operations, {}),
3228 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3229         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3230 #endif
3231         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3232         ONE("stat",      S_IRUGO, proc_tid_stat),
3233         ONE("statm",     S_IRUGO, proc_pid_statm),
3234         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3235 #ifdef CONFIG_PROC_CHILDREN
3236         REG("children",  S_IRUGO, proc_tid_children_operations),
3237 #endif
3238 #ifdef CONFIG_NUMA
3239         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3240 #endif
3241         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3242         LNK("cwd",       proc_cwd_link),
3243         LNK("root",      proc_root_link),
3244         LNK("exe",       proc_exe_link),
3245         REG("mounts",    S_IRUGO, proc_mounts_operations),
3246         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3247 #ifdef CONFIG_PROC_PAGE_MONITOR
3248         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3249         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3250         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3251 #endif
3252 #ifdef CONFIG_SECURITY
3253         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3254 #endif
3255 #ifdef CONFIG_KALLSYMS
3256         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3257 #endif
3258 #ifdef CONFIG_STACKTRACE
3259         ONE("stack",      S_IRUSR, proc_pid_stack),
3260 #endif
3261 #ifdef CONFIG_SCHED_INFO
3262         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3263 #endif
3264 #ifdef CONFIG_LATENCYTOP
3265         REG("latency",  S_IRUGO, proc_lstats_operations),
3266 #endif
3267 #ifdef CONFIG_PROC_PID_CPUSET
3268         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3269 #endif
3270 #ifdef CONFIG_CGROUPS
3271         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3272 #endif
3273         ONE("oom_score", S_IRUGO, proc_oom_score),
3274         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3275         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3276 #ifdef CONFIG_AUDITSYSCALL
3277         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3278         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3279 #endif
3280 #ifdef CONFIG_FAULT_INJECTION
3281         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3282 #endif
3283 #ifdef CONFIG_TASK_IO_ACCOUNTING
3284         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3285 #endif
3286 #ifdef CONFIG_HARDWALL
3287         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3288 #endif
3289 #ifdef CONFIG_USER_NS
3290         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3291         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3292         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3293         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3294 #endif
3295 };
3296
3297 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3298 {
3299         return proc_pident_readdir(file, ctx,
3300                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3301 }
3302
3303 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3304 {
3305         return proc_pident_lookup(dir, dentry,
3306                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3307 }
3308
3309 static const struct file_operations proc_tid_base_operations = {
3310         .read           = generic_read_dir,
3311         .iterate_shared = proc_tid_base_readdir,
3312         .llseek         = generic_file_llseek,
3313 };
3314
3315 static const struct inode_operations proc_tid_base_inode_operations = {
3316         .lookup         = proc_tid_base_lookup,
3317         .getattr        = pid_getattr,
3318         .setattr        = proc_setattr,
3319 };
3320
3321 static int proc_task_instantiate(struct inode *dir,
3322         struct dentry *dentry, struct task_struct *task, const void *ptr)
3323 {
3324         struct inode *inode;
3325         inode = proc_pid_make_inode(dir->i_sb, task);
3326
3327         if (!inode)
3328                 goto out;
3329         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3330         inode->i_op = &proc_tid_base_inode_operations;
3331         inode->i_fop = &proc_tid_base_operations;
3332         inode->i_flags|=S_IMMUTABLE;
3333
3334         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3335                                                   ARRAY_SIZE(tid_base_stuff)));
3336
3337         d_set_d_op(dentry, &pid_dentry_operations);
3338
3339         d_add(dentry, inode);
3340         /* Close the race of the process dying before we return the dentry */
3341         if (pid_revalidate(dentry, 0))
3342                 return 0;
3343 out:
3344         return -ENOENT;
3345 }
3346
3347 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3348 {
3349         int result = -ENOENT;
3350         struct task_struct *task;
3351         struct task_struct *leader = get_proc_task(dir);
3352         unsigned tid;
3353         struct pid_namespace *ns;
3354
3355         if (!leader)
3356                 goto out_no_task;
3357
3358         tid = name_to_int(&dentry->d_name);
3359         if (tid == ~0U)
3360                 goto out;
3361
3362         ns = dentry->d_sb->s_fs_info;
3363         rcu_read_lock();
3364         task = find_task_by_pid_ns(tid, ns);
3365         if (task)
3366                 get_task_struct(task);
3367         rcu_read_unlock();
3368         if (!task)
3369                 goto out;
3370         if (!same_thread_group(leader, task))
3371                 goto out_drop_task;
3372
3373         result = proc_task_instantiate(dir, dentry, task, NULL);
3374 out_drop_task:
3375         put_task_struct(task);
3376 out:
3377         put_task_struct(leader);
3378 out_no_task:
3379         return ERR_PTR(result);
3380 }
3381
3382 /*
3383  * Find the first tid of a thread group to return to user space.
3384  *
3385  * Usually this is just the thread group leader, but if the users
3386  * buffer was too small or there was a seek into the middle of the
3387  * directory we have more work todo.
3388  *
3389  * In the case of a short read we start with find_task_by_pid.
3390  *
3391  * In the case of a seek we start with the leader and walk nr
3392  * threads past it.
3393  */
3394 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3395                                         struct pid_namespace *ns)
3396 {
3397         struct task_struct *pos, *task;
3398         unsigned long nr = f_pos;
3399
3400         if (nr != f_pos)        /* 32bit overflow? */
3401                 return NULL;
3402
3403         rcu_read_lock();
3404         task = pid_task(pid, PIDTYPE_PID);
3405         if (!task)
3406                 goto fail;
3407
3408         /* Attempt to start with the tid of a thread */
3409         if (tid && nr) {
3410                 pos = find_task_by_pid_ns(tid, ns);
3411                 if (pos && same_thread_group(pos, task))
3412                         goto found;
3413         }
3414
3415         /* If nr exceeds the number of threads there is nothing todo */
3416         if (nr >= get_nr_threads(task))
3417                 goto fail;
3418
3419         /* If we haven't found our starting place yet start
3420          * with the leader and walk nr threads forward.
3421          */
3422         pos = task = task->group_leader;
3423         do {
3424                 if (!nr--)
3425                         goto found;
3426         } while_each_thread(task, pos);
3427 fail:
3428         pos = NULL;
3429         goto out;
3430 found:
3431         get_task_struct(pos);
3432 out:
3433         rcu_read_unlock();
3434         return pos;
3435 }
3436
3437 /*
3438  * Find the next thread in the thread list.
3439  * Return NULL if there is an error or no next thread.
3440  *
3441  * The reference to the input task_struct is released.
3442  */
3443 static struct task_struct *next_tid(struct task_struct *start)
3444 {
3445         struct task_struct *pos = NULL;
3446         rcu_read_lock();
3447         if (pid_alive(start)) {
3448                 pos = next_thread(start);
3449                 if (thread_group_leader(pos))
3450                         pos = NULL;
3451                 else
3452                         get_task_struct(pos);
3453         }
3454         rcu_read_unlock();
3455         put_task_struct(start);
3456         return pos;
3457 }
3458
3459 /* for the /proc/TGID/task/ directories */
3460 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3461 {
3462         struct inode *inode = file_inode(file);
3463         struct task_struct *task;
3464         struct pid_namespace *ns;
3465         int tid;
3466
3467         if (proc_inode_is_dead(inode))
3468                 return -ENOENT;
3469
3470         if (!dir_emit_dots(file, ctx))
3471                 return 0;
3472
3473         /* f_version caches the tgid value that the last readdir call couldn't
3474          * return. lseek aka telldir automagically resets f_version to 0.
3475          */
3476         ns = inode->i_sb->s_fs_info;
3477         tid = (int)file->f_version;
3478         file->f_version = 0;
3479         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3480              task;
3481              task = next_tid(task), ctx->pos++) {
3482                 char name[PROC_NUMBUF];
3483                 int len;
3484                 tid = task_pid_nr_ns(task, ns);
3485                 len = snprintf(name, sizeof(name), "%d", tid);
3486                 if (!proc_fill_cache(file, ctx, name, len,
3487                                 proc_task_instantiate, task, NULL)) {
3488                         /* returning this tgid failed, save it as the first
3489                          * pid for the next readir call */
3490                         file->f_version = (u64)tid;
3491                         put_task_struct(task);
3492                         break;
3493                 }
3494         }
3495
3496         return 0;
3497 }
3498
3499 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3500 {
3501         struct inode *inode = d_inode(dentry);
3502         struct task_struct *p = get_proc_task(inode);
3503         generic_fillattr(inode, stat);
3504
3505         if (p) {
3506                 stat->nlink += get_nr_threads(p);
3507                 put_task_struct(p);
3508         }
3509
3510         return 0;
3511 }
3512
3513 static const struct inode_operations proc_task_inode_operations = {
3514         .lookup         = proc_task_lookup,
3515         .getattr        = proc_task_getattr,
3516         .setattr        = proc_setattr,
3517         .permission     = proc_pid_permission,
3518 };
3519
3520 static const struct file_operations proc_task_operations = {
3521         .read           = generic_read_dir,
3522         .iterate_shared = proc_task_readdir,
3523         .llseek         = generic_file_llseek,
3524 };