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 <linux/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/sched/autogroup.h>
  89 #include <linux/sched/mm.h>
  90 #include <linux/sched/coredump.h>
  91 #include <linux/sched/debug.h>
  92 #include <linux/sched/stat.h>
  93 #include <linux/flex_array.h>
  94 #include <linux/posix-timers.h>
  95 #ifdef CONFIG_HARDWALL
  96 #include <asm/hardwall.h>
  97 #endif
  98 #include <trace/events/oom.h>
  99 #include "internal.h"
 100 #include "fd.h"
 101
 102 /* NOTE:
 103  *      Implementing inode permission operations in /proc is almost
 104  *      certainly an error.  Permission checks need to happen during
 105  *      each system call not at open time.  The reason is that most of
 106  *      what we wish to check for permissions in /proc varies at runtime.
 107  *
 108  *      The classic example of a problem is opening file descriptors
 109  *      in /proc for a task before it execs a suid executable.
 110  */
 111
 112 static u8 nlink_tid;
 113 static u8 nlink_tgid;
 114
 115 struct pid_entry {
 116         const char *name;
 117         unsigned int len;
 118         umode_t mode;
 119         const struct inode_operations *iop;
 120         const struct file_operations *fop;
 121         union proc_op op;
 122 };
 123
 124 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 125         .name = (NAME),                                 \
 126         .len  = sizeof(NAME) - 1,                       \
 127         .mode = MODE,                                   \
 128         .iop  = IOP,                                    \
 129         .fop  = FOP,                                    \
 130         .op   = OP,                                     \
 131 }
 132
 133 #define DIR(NAME, MODE, iops, fops)     \
 134         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 135 #define LNK(NAME, get_link)                                     \
 136         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 137                 &proc_pid_link_inode_operations, NULL,          \
 138                 { .proc_get_link = get_link } )
 139 #define REG(NAME, MODE, fops)                           \
 140         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 141 #define ONE(NAME, MODE, show)                           \
 142         NOD(NAME, (S_IFREG|(MODE)),                     \
 143                 NULL, &proc_single_file_operations,     \
 144                 { .proc_show = show } )
 145
 146 /*
 147  * Count the number of hardlinks for the pid_entry table, excluding the .
 148  * and .. links.
 149  */
 150 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 151         unsigned int n)
 152 {
 153         unsigned int i;
 154         unsigned int count;
 155
 156         count = 2;
 157         for (i = 0; i < n; ++i) {
 158                 if (S_ISDIR(entries[i].mode))
 159                         ++count;
 160         }
 161
 162         return count;
 163 }
 164
 165 static int get_task_root(struct task_struct *task, struct path *root)
 166 {
 167         int result = -ENOENT;
 168
 169         task_lock(task);
 170         if (task->fs) {
 171                 get_fs_root(task->fs, root);
 172                 result = 0;
 173         }
 174         task_unlock(task);
 175         return result;
 176 }
 177
 178 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 179 {
 180         struct task_struct *task = get_proc_task(d_inode(dentry));
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 task_lock(task);
 185                 if (task->fs) {
 186                         get_fs_pwd(task->fs, path);
 187                         result = 0;
 188                 }
 189                 task_unlock(task);
 190                 put_task_struct(task);
 191         }
 192         return result;
 193 }
 194
 195 static int proc_root_link(struct dentry *dentry, struct path *path)
 196 {
 197         struct task_struct *task = get_proc_task(d_inode(dentry));
 198         int result = -ENOENT;
 199
 200         if (task) {
 201                 result = get_task_root(task, path);
 202                 put_task_struct(task);
 203         }
 204         return result;
 205 }
 206
 207 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 208                                      size_t _count, loff_t *pos)
 209 {
 210         struct task_struct *tsk;
 211         struct mm_struct *mm;
 212         char *page;
 213         unsigned long count = _count;
 214         unsigned long arg_start, arg_end, env_start, env_end;
 215         unsigned long len1, len2, len;
 216         unsigned long p;
 217         char c;
 218         ssize_t rv;
 219
 220         BUG_ON(*pos < 0);
 221
 222         tsk = get_proc_task(file_inode(file));
 223         if (!tsk)
 224                 return -ESRCH;
 225         mm = get_task_mm(tsk);
 226         put_task_struct(tsk);
 227         if (!mm)
 228                 return 0;
 229         /* Check if process spawned far enough to have cmdline. */
 230         if (!mm->env_end) {
 231                 rv = 0;
 232                 goto out_mmput;
 233         }
 234
 235         page = (char *)__get_free_page(GFP_TEMPORARY);
 236         if (!page) {
 237                 rv = -ENOMEM;
 238                 goto out_mmput;
 239         }
 240
 241         down_read(&mm->mmap_sem);
 242         arg_start = mm->arg_start;
 243         arg_end = mm->arg_end;
 244         env_start = mm->env_start;
 245         env_end = mm->env_end;
 246         up_read(&mm->mmap_sem);
 247
 248         BUG_ON(arg_start > arg_end);
 249         BUG_ON(env_start > env_end);
 250
 251         len1 = arg_end - arg_start;
 252         len2 = env_end - env_start;
 253
 254         /* Empty ARGV. */
 255         if (len1 == 0) {
 256                 rv = 0;
 257                 goto out_free_page;
 258         }
 259         /*
 260          * Inherently racy -- command line shares address space
 261          * with code and data.
 262          */
 263         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 264         if (rv <= 0)
 265                 goto out_free_page;
 266
 267         rv = 0;
 268
 269         if (c == '\0') {
 270                 /* Command line (set of strings) occupies whole ARGV. */
 271                 if (len1 <= *pos)
 272                         goto out_free_page;
 273
 274                 p = arg_start + *pos;
 275                 len = len1 - *pos;
 276                 while (count > 0 && len > 0) {
 277                         unsigned int _count;
 278                         int nr_read;
 279
 280                         _count = min3(count, len, PAGE_SIZE);
 281                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 282                         if (nr_read < 0)
 283                                 rv = nr_read;
 284                         if (nr_read <= 0)
 285                                 goto out_free_page;
 286
 287                         if (copy_to_user(buf, page, nr_read)) {
 288                                 rv = -EFAULT;
 289                                 goto out_free_page;
 290                         }
 291
 292                         p       += nr_read;
 293                         len     -= nr_read;
 294                         buf     += nr_read;
 295                         count   -= nr_read;
 296                         rv      += nr_read;
 297                 }
 298         } else {
 299                 /*
 300                  * Command line (1 string) occupies ARGV and
 301                  * extends into ENVP.
 302                  */
 303                 struct {
 304                         unsigned long p;
 305                         unsigned long len;
 306                 } cmdline[2] = {
 307                         { .p = arg_start, .len = len1 },
 308                         { .p = env_start, .len = len2 },
 309                 };
 310                 loff_t pos1 = *pos;
 311                 unsigned int i;
 312
 313                 i = 0;
 314                 while (i < 2 && pos1 >= cmdline[i].len) {
 315                         pos1 -= cmdline[i].len;
 316                         i++;
 317                 }
 318                 while (i < 2) {
 319                         p = cmdline[i].p + pos1;
 320                         len = cmdline[i].len - pos1;
 321                         while (count > 0 && len > 0) {
 322                                 unsigned int _count, l;
 323                                 int nr_read;
 324                                 bool final;
 325
 326                                 _count = min3(count, len, PAGE_SIZE);
 327                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
 328                                 if (nr_read < 0)
 329                                         rv = nr_read;
 330                                 if (nr_read <= 0)
 331                                         goto out_free_page;
 332
 333                                 /*
 334                                  * Command line can be shorter than whole ARGV
 335                                  * even if last "marker" byte says it is not.
 336                                  */
 337                                 final = false;
 338                                 l = strnlen(page, nr_read);
 339                                 if (l < nr_read) {
 340                                         nr_read = l;
 341                                         final = true;
 342                                 }
 343
 344                                 if (copy_to_user(buf, page, nr_read)) {
 345                                         rv = -EFAULT;
 346                                         goto out_free_page;
 347                                 }
 348
 349                                 p       += nr_read;
 350                                 len     -= nr_read;
 351                                 buf     += nr_read;
 352                                 count   -= nr_read;
 353                                 rv      += nr_read;
 354
 355                                 if (final)
 356                                         goto out_free_page;
 357                         }
 358
 359                         /* Only first chunk can be read partially. */
 360                         pos1 = 0;
 361                         i++;
 362                 }
 363         }
 364
 365 out_free_page:
 366         free_page((unsigned long)page);
 367 out_mmput:
 368         mmput(mm);
 369         if (rv > 0)
 370                 *pos += rv;
 371         return rv;
 372 }
 373
 374 static const struct file_operations proc_pid_cmdline_ops = {
 375         .read   = proc_pid_cmdline_read,
 376         .llseek = generic_file_llseek,
 377 };
 378
 379 #ifdef CONFIG_KALLSYMS
 380 /*
 381  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 382  * Returns the resolved symbol.  If that fails, simply return the address.
 383  */
 384 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 385                           struct pid *pid, struct task_struct *task)
 386 {
 387         unsigned long wchan;
 388         char symname[KSYM_NAME_LEN];
 389
 390         wchan = get_wchan(task);
 391
 392         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 393                         && !lookup_symbol_name(wchan, symname))
 394                 seq_printf(m, "%s", symname);
 395         else
 396                 seq_putc(m, '0');
 397
 398         return 0;
 399 }
 400 #endif /* CONFIG_KALLSYMS */
 401
 402 static int lock_trace(struct task_struct *task)
 403 {
 404         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 405         if (err)
 406                 return err;
 407         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 408                 mutex_unlock(&task->signal->cred_guard_mutex);
 409                 return -EPERM;
 410         }
 411         return 0;
 412 }
 413
 414 static void unlock_trace(struct task_struct *task)
 415 {
 416         mutex_unlock(&task->signal->cred_guard_mutex);
 417 }
 418
 419 #ifdef CONFIG_STACKTRACE
 420
 421 #define MAX_STACK_TRACE_DEPTH   64
 422
 423 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 424                           struct pid *pid, struct task_struct *task)
 425 {
 426         struct stack_trace trace;
 427         unsigned long *entries;
 428         int err;
 429         int i;
 430
 431         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 432         if (!entries)
 433                 return -ENOMEM;
 434
 435         trace.nr_entries        = 0;
 436         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 437         trace.entries           = entries;
 438         trace.skip              = 0;
 439
 440         err = lock_trace(task);
 441         if (!err) {
 442                 save_stack_trace_tsk(task, &trace);
 443
 444                 for (i = 0; i < trace.nr_entries; i++) {
 445                         seq_printf(m, "[<%pK>] %pB\n",
 446                                    (void *)entries[i], (void *)entries[i]);
 447                 }
 448                 unlock_trace(task);
 449         }
 450         kfree(entries);
 451
 452         return err;
 453 }
 454 #endif
 455
 456 #ifdef CONFIG_SCHED_INFO
 457 /*
 458  * Provides /proc/PID/schedstat
 459  */
 460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 461                               struct pid *pid, struct task_struct *task)
 462 {
 463         if (unlikely(!sched_info_on()))
 464                 seq_printf(m, "0 0 0\n");
 465         else
 466                 seq_printf(m, "%llu %llu %lu\n",
 467                    (unsigned long long)task->se.sum_exec_runtime,
 468                    (unsigned long long)task->sched_info.run_delay,
 469                    task->sched_info.pcount);
 470
 471         return 0;
 472 }
 473 #endif
 474
 475 #ifdef CONFIG_LATENCYTOP
 476 static int lstats_show_proc(struct seq_file *m, void *v)
 477 {
 478         int i;
 479         struct inode *inode = m->private;
 480         struct task_struct *task = get_proc_task(inode);
 481
 482         if (!task)
 483                 return -ESRCH;
 484         seq_puts(m, "Latency Top version : v0.1\n");
 485         for (i = 0; i < 32; i++) {
 486                 struct latency_record *lr = &task->latency_record[i];
 487                 if (lr->backtrace[0]) {
 488                         int q;
 489                         seq_printf(m, "%i %li %li",
 490                                    lr->count, lr->time, lr->max);
 491                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 492                                 unsigned long bt = lr->backtrace[q];
 493                                 if (!bt)
 494                                         break;
 495                                 if (bt == ULONG_MAX)
 496                                         break;
 497                                 seq_printf(m, " %ps", (void *)bt);
 498                         }
 499                         seq_putc(m, '\n');
 500                 }
 501
 502         }
 503         put_task_struct(task);
 504         return 0;
 505 }
 506
 507 static int lstats_open(struct inode *inode, struct file *file)
 508 {
 509         return single_open(file, lstats_show_proc, inode);
 510 }
 511
 512 static ssize_t lstats_write(struct file *file, const char __user *buf,
 513                             size_t count, loff_t *offs)
 514 {
 515         struct task_struct *task = get_proc_task(file_inode(file));
 516
 517         if (!task)
 518                 return -ESRCH;
 519         clear_all_latency_tracing(task);
 520         put_task_struct(task);
 521
 522         return count;
 523 }
 524
 525 static const struct file_operations proc_lstats_operations = {
 526         .open           = lstats_open,
 527         .read           = seq_read,
 528         .write          = lstats_write,
 529         .llseek         = seq_lseek,
 530         .release        = single_release,
 531 };
 532
 533 #endif
 534
 535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 536                           struct pid *pid, struct task_struct *task)
 537 {
 538         unsigned long totalpages = totalram_pages + total_swap_pages;
 539         unsigned long points = 0;
 540
 541         points = oom_badness(task, NULL, NULL, totalpages) *
 542                                         1000 / totalpages;
 543         seq_printf(m, "%lu\n", points);
 544
 545         return 0;
 546 }
 547
 548 struct limit_names {
 549         const char *name;
 550         const char *unit;
 551 };
 552
 553 static const struct limit_names lnames[RLIM_NLIMITS] = {
 554         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 555         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 556         [RLIMIT_DATA] = {"Max data size", "bytes"},
 557         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 558         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 559         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 560         [RLIMIT_NPROC] = {"Max processes", "processes"},
 561         [RLIMIT_NOFILE] = {"Max open files", "files"},
 562         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 563         [RLIMIT_AS] = {"Max address space", "bytes"},
 564         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 565         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 566         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 567         [RLIMIT_NICE] = {"Max nice priority", NULL},
 568         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 569         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 570 };
 571
 572 /* Display limits for a process */
 573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 574                            struct pid *pid, struct task_struct *task)
 575 {
 576         unsigned int i;
 577         unsigned long flags;
 578
 579         struct rlimit rlim[RLIM_NLIMITS];
 580
 581         if (!lock_task_sighand(task, &flags))
 582                 return 0;
 583         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 584         unlock_task_sighand(task, &flags);
 585
 586         /*
 587          * print the file header
 588          */
 589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 590                   "Limit", "Soft Limit", "Hard Limit", "Units");
 591
 592         for (i = 0; i < RLIM_NLIMITS; i++) {
 593                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 594                         seq_printf(m, "%-25s %-20s ",
 595                                    lnames[i].name, "unlimited");
 596                 else
 597                         seq_printf(m, "%-25s %-20lu ",
 598                                    lnames[i].name, rlim[i].rlim_cur);
 599
 600                 if (rlim[i].rlim_max == RLIM_INFINITY)
 601                         seq_printf(m, "%-20s ", "unlimited");
 602                 else
 603                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 604
 605                 if (lnames[i].unit)
 606                         seq_printf(m, "%-10s\n", lnames[i].unit);
 607                 else
 608                         seq_putc(m, '\n');
 609         }
 610
 611         return 0;
 612 }
 613
 614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 616                             struct pid *pid, struct task_struct *task)
 617 {
 618         long nr;
 619         unsigned long args[6], sp, pc;
 620         int res;
 621
 622         res = lock_trace(task);
 623         if (res)
 624                 return res;
 625
 626         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 627                 seq_puts(m, "running\n");
 628         else if (nr < 0)
 629                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 630         else
 631                 seq_printf(m,
 632                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 633                        nr,
 634                        args[0], args[1], args[2], args[3], args[4], args[5],
 635                        sp, pc);
 636         unlock_trace(task);
 637
 638         return 0;
 639 }
 640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 641
 642 /************************************************************************/
 643 /*                       Here the fs part begins                        */
 644 /************************************************************************/
 645
 646 /* permission checks */
 647 static int proc_fd_access_allowed(struct inode *inode)
 648 {
 649         struct task_struct *task;
 650         int allowed = 0;
 651         /* Allow access to a task's file descriptors if it is us or we
 652          * may use ptrace attach to the process and find out that
 653          * information.
 654          */
 655         task = get_proc_task(inode);
 656         if (task) {
 657                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 658                 put_task_struct(task);
 659         }
 660         return allowed;
 661 }
 662
 663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 664 {
 665         int error;
 666         struct inode *inode = d_inode(dentry);
 667
 668         if (attr->ia_valid & ATTR_MODE)
 669                 return -EPERM;
 670
 671         error = setattr_prepare(dentry, attr);
 672         if (error)
 673                 return error;
 674
 675         setattr_copy(inode, attr);
 676         mark_inode_dirty(inode);
 677         return 0;
 678 }
 679
 680 /*
 681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 682  * or euid/egid (for hide_pid_min=2)?
 683  */
 684 static bool has_pid_permissions(struct pid_namespace *pid,
 685                                  struct task_struct *task,
 686                                  int hide_pid_min)
 687 {
 688         if (pid->hide_pid < hide_pid_min)
 689                 return true;
 690         if (in_group_p(pid->pid_gid))
 691                 return true;
 692         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 693 }
 694
 695
 696 static int proc_pid_permission(struct inode *inode, int mask)
 697 {
 698         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 699         struct task_struct *task;
 700         bool has_perms;
 701
 702         task = get_proc_task(inode);
 703         if (!task)
 704                 return -ESRCH;
 705         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 706         put_task_struct(task);
 707
 708         if (!has_perms) {
 709                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 710                         /*
 711                          * Let's make getdents(), stat(), and open()
 712                          * consistent with each other.  If a process
 713                          * may not stat() a file, it shouldn't be seen
 714                          * in procfs at all.
 715                          */
 716                         return -ENOENT;
 717                 }
 718
 719                 return -EPERM;
 720         }
 721         return generic_permission(inode, mask);
 722 }
 723
 724
 725
 726 static const struct inode_operations proc_def_inode_operations = {
 727         .setattr        = proc_setattr,
 728 };
 729
 730 static int proc_single_show(struct seq_file *m, void *v)
 731 {
 732         struct inode *inode = m->private;
 733         struct pid_namespace *ns;
 734         struct pid *pid;
 735         struct task_struct *task;
 736         int ret;
 737
 738         ns = inode->i_sb->s_fs_info;
 739         pid = proc_pid(inode);
 740         task = get_pid_task(pid, PIDTYPE_PID);
 741         if (!task)
 742                 return -ESRCH;
 743
 744         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 745
 746         put_task_struct(task);
 747         return ret;
 748 }
 749
 750 static int proc_single_open(struct inode *inode, struct file *filp)
 751 {
 752         return single_open(filp, proc_single_show, inode);
 753 }
 754
 755 static const struct file_operations proc_single_file_operations = {
 756         .open           = proc_single_open,
 757         .read           = seq_read,
 758         .llseek         = seq_lseek,
 759         .release        = single_release,
 760 };
 761
 762
 763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 764 {
 765         struct task_struct *task = get_proc_task(inode);
 766         struct mm_struct *mm = ERR_PTR(-ESRCH);
 767
 768         if (task) {
 769                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 770                 put_task_struct(task);
 771
 772                 if (!IS_ERR_OR_NULL(mm)) {
 773                         /* ensure this mm_struct can't be freed */
 774                         mmgrab(mm);
 775                         /* but do not pin its memory */
 776                         mmput(mm);
 777                 }
 778         }
 779
 780         return mm;
 781 }
 782
 783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 784 {
 785         struct mm_struct *mm = proc_mem_open(inode, mode);
 786
 787         if (IS_ERR(mm))
 788                 return PTR_ERR(mm);
 789
 790         file->private_data = mm;
 791         return 0;
 792 }
 793
 794 static int mem_open(struct inode *inode, struct file *file)
 795 {
 796         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 797
 798         /* OK to pass negative loff_t, we can catch out-of-range */
 799         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 800
 801         return ret;
 802 }
 803
 804 static ssize_t mem_rw(struct file *file, char __user *buf,
 805                         size_t count, loff_t *ppos, int write)
 806 {
 807         struct mm_struct *mm = file->private_data;
 808         unsigned long addr = *ppos;
 809         ssize_t copied;
 810         char *page;
 811         unsigned int flags;
 812
 813         if (!mm)
 814                 return 0;
 815
 816         page = (char *)__get_free_page(GFP_TEMPORARY);
 817         if (!page)
 818                 return -ENOMEM;
 819
 820         copied = 0;
 821         if (!mmget_not_zero(mm))
 822                 goto free;
 823
 824         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 825
 826         while (count > 0) {
 827                 int this_len = min_t(int, count, PAGE_SIZE);
 828
 829                 if (write && copy_from_user(page, buf, this_len)) {
 830                         copied = -EFAULT;
 831                         break;
 832                 }
 833
 834                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 835                 if (!this_len) {
 836                         if (!copied)
 837                                 copied = -EIO;
 838                         break;
 839                 }
 840
 841                 if (!write && copy_to_user(buf, page, this_len)) {
 842                         copied = -EFAULT;
 843                         break;
 844                 }
 845
 846                 buf += this_len;
 847                 addr += this_len;
 848                 copied += this_len;
 849                 count -= this_len;
 850         }
 851         *ppos = addr;
 852
 853         mmput(mm);
 854 free:
 855         free_page((unsigned long) page);
 856         return copied;
 857 }
 858
 859 static ssize_t mem_read(struct file *file, char __user *buf,
 860                         size_t count, loff_t *ppos)
 861 {
 862         return mem_rw(file, buf, count, ppos, 0);
 863 }
 864
 865 static ssize_t mem_write(struct file *file, const char __user *buf,
 866                          size_t count, loff_t *ppos)
 867 {
 868         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 869 }
 870
 871 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 872 {
 873         switch (orig) {
 874         case 0:
 875                 file->f_pos = offset;
 876                 break;
 877         case 1:
 878                 file->f_pos += offset;
 879                 break;
 880         default:
 881                 return -EINVAL;
 882         }
 883         force_successful_syscall_return();
 884         return file->f_pos;
 885 }
 886
 887 static int mem_release(struct inode *inode, struct file *file)
 888 {
 889         struct mm_struct *mm = file->private_data;
 890         if (mm)
 891                 mmdrop(mm);
 892         return 0;
 893 }
 894
 895 static const struct file_operations proc_mem_operations = {
 896         .llseek         = mem_lseek,
 897         .read           = mem_read,
 898         .write          = mem_write,
 899         .open           = mem_open,
 900         .release        = mem_release,
 901 };
 902
 903 static int environ_open(struct inode *inode, struct file *file)
 904 {
 905         return __mem_open(inode, file, PTRACE_MODE_READ);
 906 }
 907
 908 static ssize_t environ_read(struct file *file, char __user *buf,
 909                         size_t count, loff_t *ppos)
 910 {
 911         char *page;
 912         unsigned long src = *ppos;
 913         int ret = 0;
 914         struct mm_struct *mm = file->private_data;
 915         unsigned long env_start, env_end;
 916
 917         /* Ensure the process spawned far enough to have an environment. */
 918         if (!mm || !mm->env_end)
 919                 return 0;
 920
 921         page = (char *)__get_free_page(GFP_TEMPORARY);
 922         if (!page)
 923                 return -ENOMEM;
 924
 925         ret = 0;
 926         if (!mmget_not_zero(mm))
 927                 goto free;
 928
 929         down_read(&mm->mmap_sem);
 930         env_start = mm->env_start;
 931         env_end = mm->env_end;
 932         up_read(&mm->mmap_sem);
 933
 934         while (count > 0) {
 935                 size_t this_len, max_len;
 936                 int retval;
 937
 938                 if (src >= (env_end - env_start))
 939                         break;
 940
 941                 this_len = env_end - (env_start + src);
 942
 943                 max_len = min_t(size_t, PAGE_SIZE, count);
 944                 this_len = min(max_len, this_len);
 945
 946                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 947
 948                 if (retval <= 0) {
 949                         ret = retval;
 950                         break;
 951                 }
 952
 953                 if (copy_to_user(buf, page, retval)) {
 954                         ret = -EFAULT;
 955                         break;
 956                 }
 957
 958                 ret += retval;
 959                 src += retval;
 960                 buf += retval;
 961                 count -= retval;
 962         }
 963         *ppos = src;
 964         mmput(mm);
 965
 966 free:
 967         free_page((unsigned long) page);
 968         return ret;
 969 }
 970
 971 static const struct file_operations proc_environ_operations = {
 972         .open           = environ_open,
 973         .read           = environ_read,
 974         .llseek         = generic_file_llseek,
 975         .release        = mem_release,
 976 };
 977
 978 static int auxv_open(struct inode *inode, struct file *file)
 979 {
 980         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 981 }
 982
 983 static ssize_t auxv_read(struct file *file, char __user *buf,
 984                         size_t count, loff_t *ppos)
 985 {
 986         struct mm_struct *mm = file->private_data;
 987         unsigned int nwords = 0;
 988
 989         if (!mm)
 990                 return 0;
 991         do {
 992                 nwords += 2;
 993         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 994         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 995                                        nwords * sizeof(mm->saved_auxv[0]));
 996 }
 997
 998 static const struct file_operations proc_auxv_operations = {
 999         .open           = auxv_open,
1000         .read           = auxv_read,
1001         .llseek         = generic_file_llseek,
1002         .release        = mem_release,
1003 };
1004
1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006                             loff_t *ppos)
1007 {
1008         struct task_struct *task = get_proc_task(file_inode(file));
1009         char buffer[PROC_NUMBUF];
1010         int oom_adj = OOM_ADJUST_MIN;
1011         size_t len;
1012
1013         if (!task)
1014                 return -ESRCH;
1015         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1016                 oom_adj = OOM_ADJUST_MAX;
1017         else
1018                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1019                           OOM_SCORE_ADJ_MAX;
1020         put_task_struct(task);
1021         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1022         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 }
1024
1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1026 {
1027         static DEFINE_MUTEX(oom_adj_mutex);
1028         struct mm_struct *mm = NULL;
1029         struct task_struct *task;
1030         int err = 0;
1031
1032         task = get_proc_task(file_inode(file));
1033         if (!task)
1034                 return -ESRCH;
1035
1036         mutex_lock(&oom_adj_mutex);
1037         if (legacy) {
1038                 if (oom_adj < task->signal->oom_score_adj &&
1039                                 !capable(CAP_SYS_RESOURCE)) {
1040                         err = -EACCES;
1041                         goto err_unlock;
1042                 }
1043                 /*
1044                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1045                  * /proc/pid/oom_score_adj instead.
1046                  */
1047                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048                           current->comm, task_pid_nr(current), task_pid_nr(task),
1049                           task_pid_nr(task));
1050         } else {
1051                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1052                                 !capable(CAP_SYS_RESOURCE)) {
1053                         err = -EACCES;
1054                         goto err_unlock;
1055                 }
1056         }
1057
1058         /*
1059          * Make sure we will check other processes sharing the mm if this is
1060          * not vfrok which wants its own oom_score_adj.
1061          * pin the mm so it doesn't go away and get reused after task_unlock
1062          */
1063         if (!task->vfork_done) {
1064                 struct task_struct *p = find_lock_task_mm(task);
1065
1066                 if (p) {
1067                         if (atomic_read(&p->mm->mm_users) > 1) {
1068                                 mm = p->mm;
1069                                 mmgrab(mm);
1070                         }
1071                         task_unlock(p);
1072                 }
1073         }
1074
1075         task->signal->oom_score_adj = oom_adj;
1076         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1077                 task->signal->oom_score_adj_min = (short)oom_adj;
1078         trace_oom_score_adj_update(task);
1079
1080         if (mm) {
1081                 struct task_struct *p;
1082
1083                 rcu_read_lock();
1084                 for_each_process(p) {
1085                         if (same_thread_group(task, p))
1086                                 continue;
1087
1088                         /* do not touch kernel threads or the global init */
1089                         if (p->flags & PF_KTHREAD || is_global_init(p))
1090                                 continue;
1091
1092                         task_lock(p);
1093                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1094                                 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",
1095                                                 task_pid_nr(p), p->comm,
1096                                                 p->signal->oom_score_adj, oom_adj,
1097                                                 task_pid_nr(task), task->comm);
1098                                 p->signal->oom_score_adj = oom_adj;
1099                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1100                                         p->signal->oom_score_adj_min = (short)oom_adj;
1101                         }
1102                         task_unlock(p);
1103                 }
1104                 rcu_read_unlock();
1105                 mmdrop(mm);
1106         }
1107 err_unlock:
1108         mutex_unlock(&oom_adj_mutex);
1109         put_task_struct(task);
1110         return err;
1111 }
1112
1113 /*
1114  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1115  * kernels.  The effective policy is defined by oom_score_adj, which has a
1116  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1117  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1118  * Processes that become oom disabled via oom_adj will still be oom disabled
1119  * with this implementation.
1120  *
1121  * oom_adj cannot be removed since existing userspace binaries use it.
1122  */
1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1124                              size_t count, loff_t *ppos)
1125 {
1126         char buffer[PROC_NUMBUF];
1127         int oom_adj;
1128         int err;
1129
1130         memset(buffer, 0, sizeof(buffer));
1131         if (count > sizeof(buffer) - 1)
1132                 count = sizeof(buffer) - 1;
1133         if (copy_from_user(buffer, buf, count)) {
1134                 err = -EFAULT;
1135                 goto out;
1136         }
1137
1138         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139         if (err)
1140                 goto out;
1141         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1142              oom_adj != OOM_DISABLE) {
1143                 err = -EINVAL;
1144                 goto out;
1145         }
1146
1147         /*
1148          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1149          * value is always attainable.
1150          */
1151         if (oom_adj == OOM_ADJUST_MAX)
1152                 oom_adj = OOM_SCORE_ADJ_MAX;
1153         else
1154                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1155
1156         err = __set_oom_adj(file, oom_adj, true);
1157 out:
1158         return err < 0 ? err : count;
1159 }
1160
1161 static const struct file_operations proc_oom_adj_operations = {
1162         .read           = oom_adj_read,
1163         .write          = oom_adj_write,
1164         .llseek         = generic_file_llseek,
1165 };
1166
1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1168                                         size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file_inode(file));
1171         char buffer[PROC_NUMBUF];
1172         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1173         size_t len;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         oom_score_adj = task->signal->oom_score_adj;
1178         put_task_struct(task);
1179         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1180         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182
1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1184                                         size_t count, loff_t *ppos)
1185 {
1186         char buffer[PROC_NUMBUF];
1187         int oom_score_adj;
1188         int err;
1189
1190         memset(buffer, 0, sizeof(buffer));
1191         if (count > sizeof(buffer) - 1)
1192                 count = sizeof(buffer) - 1;
1193         if (copy_from_user(buffer, buf, count)) {
1194                 err = -EFAULT;
1195                 goto out;
1196         }
1197
1198         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199         if (err)
1200                 goto out;
1201         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1202                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1203                 err = -EINVAL;
1204                 goto out;
1205         }
1206
1207         err = __set_oom_adj(file, oom_score_adj, false);
1208 out:
1209         return err < 0 ? err : count;
1210 }
1211
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213         .read           = oom_score_adj_read,
1214         .write          = oom_score_adj_write,
1215         .llseek         = default_llseek,
1216 };
1217
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 11
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221                                   size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file_inode(file);
1224         struct task_struct *task = get_proc_task(inode);
1225         ssize_t length;
1226         char tmpbuf[TMPBUFLEN];
1227
1228         if (!task)
1229                 return -ESRCH;
1230         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231                            from_kuid(file->f_cred->user_ns,
1232                                      audit_get_loginuid(task)));
1233         put_task_struct(task);
1234         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238                                    size_t count, loff_t *ppos)
1239 {
1240         struct inode * inode = file_inode(file);
1241         uid_t loginuid;
1242         kuid_t kloginuid;
1243         int rv;
1244
1245         rcu_read_lock();
1246         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247                 rcu_read_unlock();
1248                 return -EPERM;
1249         }
1250         rcu_read_unlock();
1251
1252         if (*ppos != 0) {
1253                 /* No partial writes. */
1254                 return -EINVAL;
1255         }
1256
1257         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258         if (rv < 0)
1259                 return rv;
1260
1261         /* is userspace tring to explicitly UNSET the loginuid? */
1262         if (loginuid == AUDIT_UID_UNSET) {
1263                 kloginuid = INVALID_UID;
1264         } else {
1265                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266                 if (!uid_valid(kloginuid))
1267                         return -EINVAL;
1268         }
1269
1270         rv = audit_set_loginuid(kloginuid);
1271         if (rv < 0)
1272                 return rv;
1273         return count;
1274 }
1275
1276 static const struct file_operations proc_loginuid_operations = {
1277         .read           = proc_loginuid_read,
1278         .write          = proc_loginuid_write,
1279         .llseek         = generic_file_llseek,
1280 };
1281
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283                                   size_t count, loff_t *ppos)
1284 {
1285         struct inode * inode = file_inode(file);
1286         struct task_struct *task = get_proc_task(inode);
1287         ssize_t length;
1288         char tmpbuf[TMPBUFLEN];
1289
1290         if (!task)
1291                 return -ESRCH;
1292         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293                                 audit_get_sessionid(task));
1294         put_task_struct(task);
1295         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297
1298 static const struct file_operations proc_sessionid_operations = {
1299         .read           = proc_sessionid_read,
1300         .llseek         = generic_file_llseek,
1301 };
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306                                       size_t count, loff_t *ppos)
1307 {
1308         struct task_struct *task = get_proc_task(file_inode(file));
1309         char buffer[PROC_NUMBUF];
1310         size_t len;
1311         int make_it_fail;
1312
1313         if (!task)
1314                 return -ESRCH;
1315         make_it_fail = task->make_it_fail;
1316         put_task_struct(task);
1317
1318         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319
1320         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324                         const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326         struct task_struct *task;
1327         char buffer[PROC_NUMBUF];
1328         int make_it_fail;
1329         int rv;
1330
1331         if (!capable(CAP_SYS_RESOURCE))
1332                 return -EPERM;
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339         if (rv < 0)
1340                 return rv;
1341         if (make_it_fail < 0 || make_it_fail > 1)
1342                 return -EINVAL;
1343
1344         task = get_proc_task(file_inode(file));
1345         if (!task)
1346                 return -ESRCH;
1347         task->make_it_fail = make_it_fail;
1348         put_task_struct(task);
1349
1350         return count;
1351 }
1352
1353 static const struct file_operations proc_fault_inject_operations = {
1354         .read           = proc_fault_inject_read,
1355         .write          = proc_fault_inject_write,
1356         .llseek         = generic_file_llseek,
1357 };
1358
1359 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1360                                    size_t count, loff_t *ppos)
1361 {
1362         struct task_struct *task;
1363         int err;
1364         unsigned int n;
1365
1366         task = get_proc_task(file_inode(file));
1367         if (!task)
1368                 return -ESRCH;
1369         put_task_struct(task);
1370         if (task != current)
1371                 return -EPERM;
1372         err = kstrtouint_from_user(buf, count, 0, &n);
1373         if (err)
1374                 return err;
1375         current->fail_nth = n;
1376         return count;
1377 }
1378
1379 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1380                                   size_t count, loff_t *ppos)
1381 {
1382         struct task_struct *task;
1383         int err;
1384
1385         task = get_proc_task(file_inode(file));
1386         if (!task)
1387                 return -ESRCH;
1388         put_task_struct(task);
1389         if (task != current)
1390                 return -EPERM;
1391         if (count < 1)
1392                 return -EINVAL;
1393         err = put_user((char)(current->fail_nth ? 'N' : 'Y'), buf);
1394         if (err)
1395                 return err;
1396         current->fail_nth = 0;
1397         return 1;
1398 }
1399
1400 static const struct file_operations proc_fail_nth_operations = {
1401         .read           = proc_fail_nth_read,
1402         .write          = proc_fail_nth_write,
1403 };
1404 #endif
1405
1406
1407 #ifdef CONFIG_SCHED_DEBUG
1408 /*
1409  * Print out various scheduling related per-task fields:
1410  */
1411 static int sched_show(struct seq_file *m, void *v)
1412 {
1413         struct inode *inode = m->private;
1414         struct task_struct *p;
1415
1416         p = get_proc_task(inode);
1417         if (!p)
1418                 return -ESRCH;
1419         proc_sched_show_task(p, m);
1420
1421         put_task_struct(p);
1422
1423         return 0;
1424 }
1425
1426 static ssize_t
1427 sched_write(struct file *file, const char __user *buf,
1428             size_t count, loff_t *offset)
1429 {
1430         struct inode *inode = file_inode(file);
1431         struct task_struct *p;
1432
1433         p = get_proc_task(inode);
1434         if (!p)
1435                 return -ESRCH;
1436         proc_sched_set_task(p);
1437
1438         put_task_struct(p);
1439
1440         return count;
1441 }
1442
1443 static int sched_open(struct inode *inode, struct file *filp)
1444 {
1445         return single_open(filp, sched_show, inode);
1446 }
1447
1448 static const struct file_operations proc_pid_sched_operations = {
1449         .open           = sched_open,
1450         .read           = seq_read,
1451         .write          = sched_write,
1452         .llseek         = seq_lseek,
1453         .release        = single_release,
1454 };
1455
1456 #endif
1457
1458 #ifdef CONFIG_SCHED_AUTOGROUP
1459 /*
1460  * Print out autogroup related information:
1461  */
1462 static int sched_autogroup_show(struct seq_file *m, void *v)
1463 {
1464         struct inode *inode = m->private;
1465         struct task_struct *p;
1466
1467         p = get_proc_task(inode);
1468         if (!p)
1469                 return -ESRCH;
1470         proc_sched_autogroup_show_task(p, m);
1471
1472         put_task_struct(p);
1473
1474         return 0;
1475 }
1476
1477 static ssize_t
1478 sched_autogroup_write(struct file *file, const char __user *buf,
1479             size_t count, loff_t *offset)
1480 {
1481         struct inode *inode = file_inode(file);
1482         struct task_struct *p;
1483         char buffer[PROC_NUMBUF];
1484         int nice;
1485         int err;
1486
1487         memset(buffer, 0, sizeof(buffer));
1488         if (count > sizeof(buffer) - 1)
1489                 count = sizeof(buffer) - 1;
1490         if (copy_from_user(buffer, buf, count))
1491                 return -EFAULT;
1492
1493         err = kstrtoint(strstrip(buffer), 0, &nice);
1494         if (err < 0)
1495                 return err;
1496
1497         p = get_proc_task(inode);
1498         if (!p)
1499                 return -ESRCH;
1500
1501         err = proc_sched_autogroup_set_nice(p, nice);
1502         if (err)
1503                 count = err;
1504
1505         put_task_struct(p);
1506
1507         return count;
1508 }
1509
1510 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1511 {
1512         int ret;
1513
1514         ret = single_open(filp, sched_autogroup_show, NULL);
1515         if (!ret) {
1516                 struct seq_file *m = filp->private_data;
1517
1518                 m->private = inode;
1519         }
1520         return ret;
1521 }
1522
1523 static const struct file_operations proc_pid_sched_autogroup_operations = {
1524         .open           = sched_autogroup_open,
1525         .read           = seq_read,
1526         .write          = sched_autogroup_write,
1527         .llseek         = seq_lseek,
1528         .release        = single_release,
1529 };
1530
1531 #endif /* CONFIG_SCHED_AUTOGROUP */
1532
1533 static ssize_t comm_write(struct file *file, const char __user *buf,
1534                                 size_t count, loff_t *offset)
1535 {
1536         struct inode *inode = file_inode(file);
1537         struct task_struct *p;
1538         char buffer[TASK_COMM_LEN];
1539         const size_t maxlen = sizeof(buffer) - 1;
1540
1541         memset(buffer, 0, sizeof(buffer));
1542         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1543                 return -EFAULT;
1544
1545         p = get_proc_task(inode);
1546         if (!p)
1547                 return -ESRCH;
1548
1549         if (same_thread_group(current, p))
1550                 set_task_comm(p, buffer);
1551         else
1552                 count = -EINVAL;
1553
1554         put_task_struct(p);
1555
1556         return count;
1557 }
1558
1559 static int comm_show(struct seq_file *m, void *v)
1560 {
1561         struct inode *inode = m->private;
1562         struct task_struct *p;
1563
1564         p = get_proc_task(inode);
1565         if (!p)
1566                 return -ESRCH;
1567
1568         task_lock(p);
1569         seq_printf(m, "%s\n", p->comm);
1570         task_unlock(p);
1571
1572         put_task_struct(p);
1573
1574         return 0;
1575 }
1576
1577 static int comm_open(struct inode *inode, struct file *filp)
1578 {
1579         return single_open(filp, comm_show, inode);
1580 }
1581
1582 static const struct file_operations proc_pid_set_comm_operations = {
1583         .open           = comm_open,
1584         .read           = seq_read,
1585         .write          = comm_write,
1586         .llseek         = seq_lseek,
1587         .release        = single_release,
1588 };
1589
1590 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1591 {
1592         struct task_struct *task;
1593         struct file *exe_file;
1594
1595         task = get_proc_task(d_inode(dentry));
1596         if (!task)
1597                 return -ENOENT;
1598         exe_file = get_task_exe_file(task);
1599         put_task_struct(task);
1600         if (exe_file) {
1601                 *exe_path = exe_file->f_path;
1602                 path_get(&exe_file->f_path);
1603                 fput(exe_file);
1604                 return 0;
1605         } else
1606                 return -ENOENT;
1607 }
1608
1609 static const char *proc_pid_get_link(struct dentry *dentry,
1610                                      struct inode *inode,
1611                                      struct delayed_call *done)
1612 {
1613         struct path path;
1614         int error = -EACCES;
1615
1616         if (!dentry)
1617                 return ERR_PTR(-ECHILD);
1618
1619         /* Are we allowed to snoop on the tasks file descriptors? */
1620         if (!proc_fd_access_allowed(inode))
1621                 goto out;
1622
1623         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1624         if (error)
1625                 goto out;
1626
1627         nd_jump_link(&path);
1628         return NULL;
1629 out:
1630         return ERR_PTR(error);
1631 }
1632
1633 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1634 {
1635         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1636         char *pathname;
1637         int len;
1638
1639         if (!tmp)
1640                 return -ENOMEM;
1641
1642         pathname = d_path(path, tmp, PAGE_SIZE);
1643         len = PTR_ERR(pathname);
1644         if (IS_ERR(pathname))
1645                 goto out;
1646         len = tmp + PAGE_SIZE - 1 - pathname;
1647
1648         if (len > buflen)
1649                 len = buflen;
1650         if (copy_to_user(buffer, pathname, len))
1651                 len = -EFAULT;
1652  out:
1653         free_page((unsigned long)tmp);
1654         return len;
1655 }
1656
1657 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1658 {
1659         int error = -EACCES;
1660         struct inode *inode = d_inode(dentry);
1661         struct path path;
1662
1663         /* Are we allowed to snoop on the tasks file descriptors? */
1664         if (!proc_fd_access_allowed(inode))
1665                 goto out;
1666
1667         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1668         if (error)
1669                 goto out;
1670
1671         error = do_proc_readlink(&path, buffer, buflen);
1672         path_put(&path);
1673 out:
1674         return error;
1675 }
1676
1677 const struct inode_operations proc_pid_link_inode_operations = {
1678         .readlink       = proc_pid_readlink,
1679         .get_link       = proc_pid_get_link,
1680         .setattr        = proc_setattr,
1681 };
1682
1683
1684 /* building an inode */
1685
1686 void task_dump_owner(struct task_struct *task, mode_t mode,
1687                      kuid_t *ruid, kgid_t *rgid)
1688 {
1689         /* Depending on the state of dumpable compute who should own a
1690          * proc file for a task.
1691          */
1692         const struct cred *cred;
1693         kuid_t uid;
1694         kgid_t gid;
1695
1696         /* Default to the tasks effective ownership */
1697         rcu_read_lock();
1698         cred = __task_cred(task);
1699         uid = cred->euid;
1700         gid = cred->egid;
1701         rcu_read_unlock();
1702
1703         /*
1704          * Before the /proc/pid/status file was created the only way to read
1705          * the effective uid of a /process was to stat /proc/pid.  Reading
1706          * /proc/pid/status is slow enough that procps and other packages
1707          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1708          * made this apply to all per process world readable and executable
1709          * directories.
1710          */
1711         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1712                 struct mm_struct *mm;
1713                 task_lock(task);
1714                 mm = task->mm;
1715                 /* Make non-dumpable tasks owned by some root */
1716                 if (mm) {
1717                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1718                                 struct user_namespace *user_ns = mm->user_ns;
1719
1720                                 uid = make_kuid(user_ns, 0);
1721                                 if (!uid_valid(uid))
1722                                         uid = GLOBAL_ROOT_UID;
1723
1724                                 gid = make_kgid(user_ns, 0);
1725                                 if (!gid_valid(gid))
1726                                         gid = GLOBAL_ROOT_GID;
1727                         }
1728                 } else {
1729                         uid = GLOBAL_ROOT_UID;
1730                         gid = GLOBAL_ROOT_GID;
1731                 }
1732                 task_unlock(task);
1733         }
1734         *ruid = uid;
1735         *rgid = gid;
1736 }
1737
1738 struct inode *proc_pid_make_inode(struct super_block * sb,
1739                                   struct task_struct *task, umode_t mode)
1740 {
1741         struct inode * inode;
1742         struct proc_inode *ei;
1743
1744         /* We need a new inode */
1745
1746         inode = new_inode(sb);
1747         if (!inode)
1748                 goto out;
1749
1750         /* Common stuff */
1751         ei = PROC_I(inode);
1752         inode->i_mode = mode;
1753         inode->i_ino = get_next_ino();
1754         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1755         inode->i_op = &proc_def_inode_operations;
1756
1757         /*
1758          * grab the reference to task.
1759          */
1760         ei->pid = get_task_pid(task, PIDTYPE_PID);
1761         if (!ei->pid)
1762                 goto out_unlock;
1763
1764         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1765         security_task_to_inode(task, inode);
1766
1767 out:
1768         return inode;
1769
1770 out_unlock:
1771         iput(inode);
1772         return NULL;
1773 }
1774
1775 int pid_getattr(const struct path *path, struct kstat *stat,
1776                 u32 request_mask, unsigned int query_flags)
1777 {
1778         struct inode *inode = d_inode(path->dentry);
1779         struct task_struct *task;
1780         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1781
1782         generic_fillattr(inode, stat);
1783
1784         rcu_read_lock();
1785         stat->uid = GLOBAL_ROOT_UID;
1786         stat->gid = GLOBAL_ROOT_GID;
1787         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1788         if (task) {
1789                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1790                         rcu_read_unlock();
1791                         /*
1792                          * This doesn't prevent learning whether PID exists,
1793                          * it only makes getattr() consistent with readdir().
1794                          */
1795                         return -ENOENT;
1796                 }
1797                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1798         }
1799         rcu_read_unlock();
1800         return 0;
1801 }
1802
1803 /* dentry stuff */
1804
1805 /*
1806  *      Exceptional case: normally we are not allowed to unhash a busy
1807  * directory. In this case, however, we can do it - no aliasing problems
1808  * due to the way we treat inodes.
1809  *
1810  * Rewrite the inode's ownerships here because the owning task may have
1811  * performed a setuid(), etc.
1812  *
1813  */
1814 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1815 {
1816         struct inode *inode;
1817         struct task_struct *task;
1818
1819         if (flags & LOOKUP_RCU)
1820                 return -ECHILD;
1821
1822         inode = d_inode(dentry);
1823         task = get_proc_task(inode);
1824
1825         if (task) {
1826                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1827
1828                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1829                 security_task_to_inode(task, inode);
1830                 put_task_struct(task);
1831                 return 1;
1832         }
1833         return 0;
1834 }
1835
1836 static inline bool proc_inode_is_dead(struct inode *inode)
1837 {
1838         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1839 }
1840
1841 int pid_delete_dentry(const struct dentry *dentry)
1842 {
1843         /* Is the task we represent dead?
1844          * If so, then don't put the dentry on the lru list,
1845          * kill it immediately.
1846          */
1847         return proc_inode_is_dead(d_inode(dentry));
1848 }
1849
1850 const struct dentry_operations pid_dentry_operations =
1851 {
1852         .d_revalidate   = pid_revalidate,
1853         .d_delete       = pid_delete_dentry,
1854 };
1855
1856 /* Lookups */
1857
1858 /*
1859  * Fill a directory entry.
1860  *
1861  * If possible create the dcache entry and derive our inode number and
1862  * file type from dcache entry.
1863  *
1864  * Since all of the proc inode numbers are dynamically generated, the inode
1865  * numbers do not exist until the inode is cache.  This means creating the
1866  * the dcache entry in readdir is necessary to keep the inode numbers
1867  * reported by readdir in sync with the inode numbers reported
1868  * by stat.
1869  */
1870 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1871         const char *name, int len,
1872         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1873 {
1874         struct dentry *child, *dir = file->f_path.dentry;
1875         struct qstr qname = QSTR_INIT(name, len);
1876         struct inode *inode;
1877         unsigned type;
1878         ino_t ino;
1879
1880         child = d_hash_and_lookup(dir, &qname);
1881         if (!child) {
1882                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1883                 child = d_alloc_parallel(dir, &qname, &wq);
1884                 if (IS_ERR(child))
1885                         goto end_instantiate;
1886                 if (d_in_lookup(child)) {
1887                         int err = instantiate(d_inode(dir), child, task, ptr);
1888                         d_lookup_done(child);
1889                         if (err < 0) {
1890                                 dput(child);
1891                                 goto end_instantiate;
1892                         }
1893                 }
1894         }
1895         inode = d_inode(child);
1896         ino = inode->i_ino;
1897         type = inode->i_mode >> 12;
1898         dput(child);
1899         return dir_emit(ctx, name, len, ino, type);
1900
1901 end_instantiate:
1902         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1903 }
1904
1905 /*
1906  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1907  * which represent vma start and end addresses.
1908  */
1909 static int dname_to_vma_addr(struct dentry *dentry,
1910                              unsigned long *start, unsigned long *end)
1911 {
1912         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1913                 return -EINVAL;
1914
1915         return 0;
1916 }
1917
1918 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1919 {
1920         unsigned long vm_start, vm_end;
1921         bool exact_vma_exists = false;
1922         struct mm_struct *mm = NULL;
1923         struct task_struct *task;
1924         struct inode *inode;
1925         int status = 0;
1926
1927         if (flags & LOOKUP_RCU)
1928                 return -ECHILD;
1929
1930         inode = d_inode(dentry);
1931         task = get_proc_task(inode);
1932         if (!task)
1933                 goto out_notask;
1934
1935         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1936         if (IS_ERR_OR_NULL(mm))
1937                 goto out;
1938
1939         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1940                 down_read(&mm->mmap_sem);
1941                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1942                 up_read(&mm->mmap_sem);
1943         }
1944
1945         mmput(mm);
1946
1947         if (exact_vma_exists) {
1948                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1949
1950                 security_task_to_inode(task, inode);
1951                 status = 1;
1952         }
1953
1954 out:
1955         put_task_struct(task);
1956
1957 out_notask:
1958         return status;
1959 }
1960
1961 static const struct dentry_operations tid_map_files_dentry_operations = {
1962         .d_revalidate   = map_files_d_revalidate,
1963         .d_delete       = pid_delete_dentry,
1964 };
1965
1966 static int map_files_get_link(struct dentry *dentry, struct path *path)
1967 {
1968         unsigned long vm_start, vm_end;
1969         struct vm_area_struct *vma;
1970         struct task_struct *task;
1971         struct mm_struct *mm;
1972         int rc;
1973
1974         rc = -ENOENT;
1975         task = get_proc_task(d_inode(dentry));
1976         if (!task)
1977                 goto out;
1978
1979         mm = get_task_mm(task);
1980         put_task_struct(task);
1981         if (!mm)
1982                 goto out;
1983
1984         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1985         if (rc)
1986                 goto out_mmput;
1987
1988         rc = -ENOENT;
1989         down_read(&mm->mmap_sem);
1990         vma = find_exact_vma(mm, vm_start, vm_end);
1991         if (vma && vma->vm_file) {
1992                 *path = vma->vm_file->f_path;
1993                 path_get(path);
1994                 rc = 0;
1995         }
1996         up_read(&mm->mmap_sem);
1997
1998 out_mmput:
1999         mmput(mm);
2000 out:
2001         return rc;
2002 }
2003
2004 struct map_files_info {
2005         fmode_t         mode;
2006         unsigned int    len;
2007         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2008 };
2009
2010 /*
2011  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2012  * symlinks may be used to bypass permissions on ancestor directories in the
2013  * path to the file in question.
2014  */
2015 static const char *
2016 proc_map_files_get_link(struct dentry *dentry,
2017                         struct inode *inode,
2018                         struct delayed_call *done)
2019 {
2020         if (!capable(CAP_SYS_ADMIN))
2021                 return ERR_PTR(-EPERM);
2022
2023         return proc_pid_get_link(dentry, inode, done);
2024 }
2025
2026 /*
2027  * Identical to proc_pid_link_inode_operations except for get_link()
2028  */
2029 static const struct inode_operations proc_map_files_link_inode_operations = {
2030         .readlink       = proc_pid_readlink,
2031         .get_link       = proc_map_files_get_link,
2032         .setattr        = proc_setattr,
2033 };
2034
2035 static int
2036 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2037                            struct task_struct *task, const void *ptr)
2038 {
2039         fmode_t mode = (fmode_t)(unsigned long)ptr;
2040         struct proc_inode *ei;
2041         struct inode *inode;
2042
2043         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2044                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2045                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2046         if (!inode)
2047                 return -ENOENT;
2048
2049         ei = PROC_I(inode);
2050         ei->op.proc_get_link = map_files_get_link;
2051
2052         inode->i_op = &proc_map_files_link_inode_operations;
2053         inode->i_size = 64;
2054
2055         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2056         d_add(dentry, inode);
2057
2058         return 0;
2059 }
2060
2061 static struct dentry *proc_map_files_lookup(struct inode *dir,
2062                 struct dentry *dentry, unsigned int flags)
2063 {
2064         unsigned long vm_start, vm_end;
2065         struct vm_area_struct *vma;
2066         struct task_struct *task;
2067         int result;
2068         struct mm_struct *mm;
2069
2070         result = -ENOENT;
2071         task = get_proc_task(dir);
2072         if (!task)
2073                 goto out;
2074
2075         result = -EACCES;
2076         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2077                 goto out_put_task;
2078
2079         result = -ENOENT;
2080         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2081                 goto out_put_task;
2082
2083         mm = get_task_mm(task);
2084         if (!mm)
2085                 goto out_put_task;
2086
2087         down_read(&mm->mmap_sem);
2088         vma = find_exact_vma(mm, vm_start, vm_end);
2089         if (!vma)
2090                 goto out_no_vma;
2091
2092         if (vma->vm_file)
2093                 result = proc_map_files_instantiate(dir, dentry, task,
2094                                 (void *)(unsigned long)vma->vm_file->f_mode);
2095
2096 out_no_vma:
2097         up_read(&mm->mmap_sem);
2098         mmput(mm);
2099 out_put_task:
2100         put_task_struct(task);
2101 out:
2102         return ERR_PTR(result);
2103 }
2104
2105 static const struct inode_operations proc_map_files_inode_operations = {
2106         .lookup         = proc_map_files_lookup,
2107         .permission     = proc_fd_permission,
2108         .setattr        = proc_setattr,
2109 };
2110
2111 static int
2112 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2113 {
2114         struct vm_area_struct *vma;
2115         struct task_struct *task;
2116         struct mm_struct *mm;
2117         unsigned long nr_files, pos, i;
2118         struct flex_array *fa = NULL;
2119         struct map_files_info info;
2120         struct map_files_info *p;
2121         int ret;
2122
2123         ret = -ENOENT;
2124         task = get_proc_task(file_inode(file));
2125         if (!task)
2126                 goto out;
2127
2128         ret = -EACCES;
2129         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2130                 goto out_put_task;
2131
2132         ret = 0;
2133         if (!dir_emit_dots(file, ctx))
2134                 goto out_put_task;
2135
2136         mm = get_task_mm(task);
2137         if (!mm)
2138                 goto out_put_task;
2139         down_read(&mm->mmap_sem);
2140
2141         nr_files = 0;
2142
2143         /*
2144          * We need two passes here:
2145          *
2146          *  1) Collect vmas of mapped files with mmap_sem taken
2147          *  2) Release mmap_sem and instantiate entries
2148          *
2149          * otherwise we get lockdep complained, since filldir()
2150          * routine might require mmap_sem taken in might_fault().
2151          */
2152
2153         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2154                 if (vma->vm_file && ++pos > ctx->pos)
2155                         nr_files++;
2156         }
2157
2158         if (nr_files) {
2159                 fa = flex_array_alloc(sizeof(info), nr_files,
2160                                         GFP_KERNEL);
2161                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2162                                                 GFP_KERNEL)) {
2163                         ret = -ENOMEM;
2164                         if (fa)
2165                                 flex_array_free(fa);
2166                         up_read(&mm->mmap_sem);
2167                         mmput(mm);
2168                         goto out_put_task;
2169                 }
2170                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2171                                 vma = vma->vm_next) {
2172                         if (!vma->vm_file)
2173                                 continue;
2174                         if (++pos <= ctx->pos)
2175                                 continue;
2176
2177                         info.mode = vma->vm_file->f_mode;
2178                         info.len = snprintf(info.name,
2179                                         sizeof(info.name), "%lx-%lx",
2180                                         vma->vm_start, vma->vm_end);
2181                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2182                                 BUG();
2183                 }
2184         }
2185         up_read(&mm->mmap_sem);
2186
2187         for (i = 0; i < nr_files; i++) {
2188                 p = flex_array_get(fa, i);
2189                 if (!proc_fill_cache(file, ctx,
2190                                       p->name, p->len,
2191                                       proc_map_files_instantiate,
2192                                       task,
2193                                       (void *)(unsigned long)p->mode))
2194                         break;
2195                 ctx->pos++;
2196         }
2197         if (fa)
2198                 flex_array_free(fa);
2199         mmput(mm);
2200
2201 out_put_task:
2202         put_task_struct(task);
2203 out:
2204         return ret;
2205 }
2206
2207 static const struct file_operations proc_map_files_operations = {
2208         .read           = generic_read_dir,
2209         .iterate_shared = proc_map_files_readdir,
2210         .llseek         = generic_file_llseek,
2211 };
2212
2213 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2214 struct timers_private {
2215         struct pid *pid;
2216         struct task_struct *task;
2217         struct sighand_struct *sighand;
2218         struct pid_namespace *ns;
2219         unsigned long flags;
2220 };
2221
2222 static void *timers_start(struct seq_file *m, loff_t *pos)
2223 {
2224         struct timers_private *tp = m->private;
2225
2226         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2227         if (!tp->task)
2228                 return ERR_PTR(-ESRCH);
2229
2230         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2231         if (!tp->sighand)
2232                 return ERR_PTR(-ESRCH);
2233
2234         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2235 }
2236
2237 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2238 {
2239         struct timers_private *tp = m->private;
2240         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2241 }
2242
2243 static void timers_stop(struct seq_file *m, void *v)
2244 {
2245         struct timers_private *tp = m->private;
2246
2247         if (tp->sighand) {
2248                 unlock_task_sighand(tp->task, &tp->flags);
2249                 tp->sighand = NULL;
2250         }
2251
2252         if (tp->task) {
2253                 put_task_struct(tp->task);
2254                 tp->task = NULL;
2255         }
2256 }
2257
2258 static int show_timer(struct seq_file *m, void *v)
2259 {
2260         struct k_itimer *timer;
2261         struct timers_private *tp = m->private;
2262         int notify;
2263         static const char * const nstr[] = {
2264                 [SIGEV_SIGNAL] = "signal",
2265                 [SIGEV_NONE] = "none",
2266                 [SIGEV_THREAD] = "thread",
2267         };
2268
2269         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2270         notify = timer->it_sigev_notify;
2271
2272         seq_printf(m, "ID: %d\n", timer->it_id);
2273         seq_printf(m, "signal: %d/%p\n",
2274                    timer->sigq->info.si_signo,
2275                    timer->sigq->info.si_value.sival_ptr);
2276         seq_printf(m, "notify: %s/%s.%d\n",
2277                    nstr[notify & ~SIGEV_THREAD_ID],
2278                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2279                    pid_nr_ns(timer->it_pid, tp->ns));
2280         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2281
2282         return 0;
2283 }
2284
2285 static const struct seq_operations proc_timers_seq_ops = {
2286         .start  = timers_start,
2287         .next   = timers_next,
2288         .stop   = timers_stop,
2289         .show   = show_timer,
2290 };
2291
2292 static int proc_timers_open(struct inode *inode, struct file *file)
2293 {
2294         struct timers_private *tp;
2295
2296         tp = __seq_open_private(file, &proc_timers_seq_ops,
2297                         sizeof(struct timers_private));
2298         if (!tp)
2299                 return -ENOMEM;
2300
2301         tp->pid = proc_pid(inode);
2302         tp->ns = inode->i_sb->s_fs_info;
2303         return 0;
2304 }
2305
2306 static const struct file_operations proc_timers_operations = {
2307         .open           = proc_timers_open,
2308         .read           = seq_read,
2309         .llseek         = seq_lseek,
2310         .release        = seq_release_private,
2311 };
2312 #endif
2313
2314 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2315                                         size_t count, loff_t *offset)
2316 {
2317         struct inode *inode = file_inode(file);
2318         struct task_struct *p;
2319         u64 slack_ns;
2320         int err;
2321
2322         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2323         if (err < 0)
2324                 return err;
2325
2326         p = get_proc_task(inode);
2327         if (!p)
2328                 return -ESRCH;
2329
2330         if (p != current) {
2331                 if (!capable(CAP_SYS_NICE)) {
2332                         count = -EPERM;
2333                         goto out;
2334                 }
2335
2336                 err = security_task_setscheduler(p);
2337                 if (err) {
2338                         count = err;
2339                         goto out;
2340                 }
2341         }
2342
2343         task_lock(p);
2344         if (slack_ns == 0)
2345                 p->timer_slack_ns = p->default_timer_slack_ns;
2346         else
2347                 p->timer_slack_ns = slack_ns;
2348         task_unlock(p);
2349
2350 out:
2351         put_task_struct(p);
2352
2353         return count;
2354 }
2355
2356 static int timerslack_ns_show(struct seq_file *m, void *v)
2357 {
2358         struct inode *inode = m->private;
2359         struct task_struct *p;
2360         int err = 0;
2361
2362         p = get_proc_task(inode);
2363         if (!p)
2364                 return -ESRCH;
2365
2366         if (p != current) {
2367
2368                 if (!capable(CAP_SYS_NICE)) {
2369                         err = -EPERM;
2370                         goto out;
2371                 }
2372                 err = security_task_getscheduler(p);
2373                 if (err)
2374                         goto out;
2375         }
2376
2377         task_lock(p);
2378         seq_printf(m, "%llu\n", p->timer_slack_ns);
2379         task_unlock(p);
2380
2381 out:
2382         put_task_struct(p);
2383
2384         return err;
2385 }
2386
2387 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2388 {
2389         return single_open(filp, timerslack_ns_show, inode);
2390 }
2391
2392 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2393         .open           = timerslack_ns_open,
2394         .read           = seq_read,
2395         .write          = timerslack_ns_write,
2396         .llseek         = seq_lseek,
2397         .release        = single_release,
2398 };
2399
2400 static int proc_pident_instantiate(struct inode *dir,
2401         struct dentry *dentry, struct task_struct *task, const void *ptr)
2402 {
2403         const struct pid_entry *p = ptr;
2404         struct inode *inode;
2405         struct proc_inode *ei;
2406
2407         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2408         if (!inode)
2409                 goto out;
2410
2411         ei = PROC_I(inode);
2412         if (S_ISDIR(inode->i_mode))
2413                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2414         if (p->iop)
2415                 inode->i_op = p->iop;
2416         if (p->fop)
2417                 inode->i_fop = p->fop;
2418         ei->op = p->op;
2419         d_set_d_op(dentry, &pid_dentry_operations);
2420         d_add(dentry, inode);
2421         /* Close the race of the process dying before we return the dentry */
2422         if (pid_revalidate(dentry, 0))
2423                 return 0;
2424 out:
2425         return -ENOENT;
2426 }
2427
2428 static struct dentry *proc_pident_lookup(struct inode *dir,
2429                                          struct dentry *dentry,
2430                                          const struct pid_entry *ents,
2431                                          unsigned int nents)
2432 {
2433         int error;
2434         struct task_struct *task = get_proc_task(dir);
2435         const struct pid_entry *p, *last;
2436
2437         error = -ENOENT;
2438
2439         if (!task)
2440                 goto out_no_task;
2441
2442         /*
2443          * Yes, it does not scale. And it should not. Don't add
2444          * new entries into /proc/<tgid>/ without very good reasons.
2445          */
2446         last = &ents[nents];
2447         for (p = ents; p < last; p++) {
2448                 if (p->len != dentry->d_name.len)
2449                         continue;
2450                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2451                         break;
2452         }
2453         if (p >= last)
2454                 goto out;
2455
2456         error = proc_pident_instantiate(dir, dentry, task, p);
2457 out:
2458         put_task_struct(task);
2459 out_no_task:
2460         return ERR_PTR(error);
2461 }
2462
2463 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2464                 const struct pid_entry *ents, unsigned int nents)
2465 {
2466         struct task_struct *task = get_proc_task(file_inode(file));
2467         const struct pid_entry *p;
2468
2469         if (!task)
2470                 return -ENOENT;
2471
2472         if (!dir_emit_dots(file, ctx))
2473                 goto out;
2474
2475         if (ctx->pos >= nents + 2)
2476                 goto out;
2477
2478         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2479                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2480                                 proc_pident_instantiate, task, p))
2481                         break;
2482                 ctx->pos++;
2483         }
2484 out:
2485         put_task_struct(task);
2486         return 0;
2487 }
2488
2489 #ifdef CONFIG_SECURITY
2490 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2491                                   size_t count, loff_t *ppos)
2492 {
2493         struct inode * inode = file_inode(file);
2494         char *p = NULL;
2495         ssize_t length;
2496         struct task_struct *task = get_proc_task(inode);
2497
2498         if (!task)
2499                 return -ESRCH;
2500
2501         length = security_getprocattr(task,
2502                                       (char*)file->f_path.dentry->d_name.name,
2503                                       &p);
2504         put_task_struct(task);
2505         if (length > 0)
2506                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2507         kfree(p);
2508         return length;
2509 }
2510
2511 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2512                                    size_t count, loff_t *ppos)
2513 {
2514         struct inode * inode = file_inode(file);
2515         void *page;
2516         ssize_t length;
2517         struct task_struct *task = get_proc_task(inode);
2518
2519         length = -ESRCH;
2520         if (!task)
2521                 goto out_no_task;
2522
2523         /* A task may only write its own attributes. */
2524         length = -EACCES;
2525         if (current != task)
2526                 goto out;
2527
2528         if (count > PAGE_SIZE)
2529                 count = PAGE_SIZE;
2530
2531         /* No partial writes. */
2532         length = -EINVAL;
2533         if (*ppos != 0)
2534                 goto out;
2535
2536         page = memdup_user(buf, count);
2537         if (IS_ERR(page)) {
2538                 length = PTR_ERR(page);
2539                 goto out;
2540         }
2541
2542         /* Guard against adverse ptrace interaction */
2543         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2544         if (length < 0)
2545                 goto out_free;
2546
2547         length = security_setprocattr(file->f_path.dentry->d_name.name,
2548                                       page, count);
2549         mutex_unlock(&current->signal->cred_guard_mutex);
2550 out_free:
2551         kfree(page);
2552 out:
2553         put_task_struct(task);
2554 out_no_task:
2555         return length;
2556 }
2557
2558 static const struct file_operations proc_pid_attr_operations = {
2559         .read           = proc_pid_attr_read,
2560         .write          = proc_pid_attr_write,
2561         .llseek         = generic_file_llseek,
2562 };
2563
2564 static const struct pid_entry attr_dir_stuff[] = {
2565         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2566         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2567         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2568         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2569         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2570         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2571 };
2572
2573 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2574 {
2575         return proc_pident_readdir(file, ctx,
2576                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2577 }
2578
2579 static const struct file_operations proc_attr_dir_operations = {
2580         .read           = generic_read_dir,
2581         .iterate_shared = proc_attr_dir_readdir,
2582         .llseek         = generic_file_llseek,
2583 };
2584
2585 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2586                                 struct dentry *dentry, unsigned int flags)
2587 {
2588         return proc_pident_lookup(dir, dentry,
2589                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2590 }
2591
2592 static const struct inode_operations proc_attr_dir_inode_operations = {
2593         .lookup         = proc_attr_dir_lookup,
2594         .getattr        = pid_getattr,
2595         .setattr        = proc_setattr,
2596 };
2597
2598 #endif
2599
2600 #ifdef CONFIG_ELF_CORE
2601 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2602                                          size_t count, loff_t *ppos)
2603 {
2604         struct task_struct *task = get_proc_task(file_inode(file));
2605         struct mm_struct *mm;
2606         char buffer[PROC_NUMBUF];
2607         size_t len;
2608         int ret;
2609
2610         if (!task)
2611                 return -ESRCH;
2612
2613         ret = 0;
2614         mm = get_task_mm(task);
2615         if (mm) {
2616                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2617                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2618                                 MMF_DUMP_FILTER_SHIFT));
2619                 mmput(mm);
2620                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2621         }
2622
2623         put_task_struct(task);
2624
2625         return ret;
2626 }
2627
2628 static ssize_t proc_coredump_filter_write(struct file *file,
2629                                           const char __user *buf,
2630                                           size_t count,
2631                                           loff_t *ppos)
2632 {
2633         struct task_struct *task;
2634         struct mm_struct *mm;
2635         unsigned int val;
2636         int ret;
2637         int i;
2638         unsigned long mask;
2639
2640         ret = kstrtouint_from_user(buf, count, 0, &val);
2641         if (ret < 0)
2642                 return ret;
2643
2644         ret = -ESRCH;
2645         task = get_proc_task(file_inode(file));
2646         if (!task)
2647                 goto out_no_task;
2648
2649         mm = get_task_mm(task);
2650         if (!mm)
2651                 goto out_no_mm;
2652         ret = 0;
2653
2654         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2655                 if (val & mask)
2656                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2657                 else
2658                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2659         }
2660
2661         mmput(mm);
2662  out_no_mm:
2663         put_task_struct(task);
2664  out_no_task:
2665         if (ret < 0)
2666                 return ret;
2667         return count;
2668 }
2669
2670 static const struct file_operations proc_coredump_filter_operations = {
2671         .read           = proc_coredump_filter_read,
2672         .write          = proc_coredump_filter_write,
2673         .llseek         = generic_file_llseek,
2674 };
2675 #endif
2676
2677 #ifdef CONFIG_TASK_IO_ACCOUNTING
2678 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2679 {
2680         struct task_io_accounting acct = task->ioac;
2681         unsigned long flags;
2682         int result;
2683
2684         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2685         if (result)
2686                 return result;
2687
2688         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2689                 result = -EACCES;
2690                 goto out_unlock;
2691         }
2692
2693         if (whole && lock_task_sighand(task, &flags)) {
2694                 struct task_struct *t = task;
2695
2696                 task_io_accounting_add(&acct, &task->signal->ioac);
2697                 while_each_thread(task, t)
2698                         task_io_accounting_add(&acct, &t->ioac);
2699
2700                 unlock_task_sighand(task, &flags);
2701         }
2702         seq_printf(m,
2703                    "rchar: %llu\n"
2704                    "wchar: %llu\n"
2705                    "syscr: %llu\n"
2706                    "syscw: %llu\n"
2707                    "read_bytes: %llu\n"
2708                    "write_bytes: %llu\n"
2709                    "cancelled_write_bytes: %llu\n",
2710                    (unsigned long long)acct.rchar,
2711                    (unsigned long long)acct.wchar,
2712                    (unsigned long long)acct.syscr,
2713                    (unsigned long long)acct.syscw,
2714                    (unsigned long long)acct.read_bytes,
2715                    (unsigned long long)acct.write_bytes,
2716                    (unsigned long long)acct.cancelled_write_bytes);
2717         result = 0;
2718
2719 out_unlock:
2720         mutex_unlock(&task->signal->cred_guard_mutex);
2721         return result;
2722 }
2723
2724 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2725                                   struct pid *pid, struct task_struct *task)
2726 {
2727         return do_io_accounting(task, m, 0);
2728 }
2729
2730 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2731                                    struct pid *pid, struct task_struct *task)
2732 {
2733         return do_io_accounting(task, m, 1);
2734 }
2735 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2736
2737 #ifdef CONFIG_USER_NS
2738 static int proc_id_map_open(struct inode *inode, struct file *file,
2739         const struct seq_operations *seq_ops)
2740 {
2741         struct user_namespace *ns = NULL;
2742         struct task_struct *task;
2743         struct seq_file *seq;
2744         int ret = -EINVAL;
2745
2746         task = get_proc_task(inode);
2747         if (task) {
2748                 rcu_read_lock();
2749                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2750                 rcu_read_unlock();
2751                 put_task_struct(task);
2752         }
2753         if (!ns)
2754                 goto err;
2755
2756         ret = seq_open(file, seq_ops);
2757         if (ret)
2758                 goto err_put_ns;
2759
2760         seq = file->private_data;
2761         seq->private = ns;
2762
2763         return 0;
2764 err_put_ns:
2765         put_user_ns(ns);
2766 err:
2767         return ret;
2768 }
2769
2770 static int proc_id_map_release(struct inode *inode, struct file *file)
2771 {
2772         struct seq_file *seq = file->private_data;
2773         struct user_namespace *ns = seq->private;
2774         put_user_ns(ns);
2775         return seq_release(inode, file);
2776 }
2777
2778 static int proc_uid_map_open(struct inode *inode, struct file *file)
2779 {
2780         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2781 }
2782
2783 static int proc_gid_map_open(struct inode *inode, struct file *file)
2784 {
2785         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2786 }
2787
2788 static int proc_projid_map_open(struct inode *inode, struct file *file)
2789 {
2790         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2791 }
2792
2793 static const struct file_operations proc_uid_map_operations = {
2794         .open           = proc_uid_map_open,
2795         .write          = proc_uid_map_write,
2796         .read           = seq_read,
2797         .llseek         = seq_lseek,
2798         .release        = proc_id_map_release,
2799 };
2800
2801 static const struct file_operations proc_gid_map_operations = {
2802         .open           = proc_gid_map_open,
2803         .write          = proc_gid_map_write,
2804         .read           = seq_read,
2805         .llseek         = seq_lseek,
2806         .release        = proc_id_map_release,
2807 };
2808
2809 static const struct file_operations proc_projid_map_operations = {
2810         .open           = proc_projid_map_open,
2811         .write          = proc_projid_map_write,
2812         .read           = seq_read,
2813         .llseek         = seq_lseek,
2814         .release        = proc_id_map_release,
2815 };
2816
2817 static int proc_setgroups_open(struct inode *inode, struct file *file)
2818 {
2819         struct user_namespace *ns = NULL;
2820         struct task_struct *task;
2821         int ret;
2822
2823         ret = -ESRCH;
2824         task = get_proc_task(inode);
2825         if (task) {
2826                 rcu_read_lock();
2827                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2828                 rcu_read_unlock();
2829                 put_task_struct(task);
2830         }
2831         if (!ns)
2832                 goto err;
2833
2834         if (file->f_mode & FMODE_WRITE) {
2835                 ret = -EACCES;
2836                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2837                         goto err_put_ns;
2838         }
2839
2840         ret = single_open(file, &proc_setgroups_show, ns);
2841         if (ret)
2842                 goto err_put_ns;
2843
2844         return 0;
2845 err_put_ns:
2846         put_user_ns(ns);
2847 err:
2848         return ret;
2849 }
2850
2851 static int proc_setgroups_release(struct inode *inode, struct file *file)
2852 {
2853         struct seq_file *seq = file->private_data;
2854         struct user_namespace *ns = seq->private;
2855         int ret = single_release(inode, file);
2856         put_user_ns(ns);
2857         return ret;
2858 }
2859
2860 static const struct file_operations proc_setgroups_operations = {
2861         .open           = proc_setgroups_open,
2862         .write          = proc_setgroups_write,
2863         .read           = seq_read,
2864         .llseek         = seq_lseek,
2865         .release        = proc_setgroups_release,
2866 };
2867 #endif /* CONFIG_USER_NS */
2868
2869 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2870                                 struct pid *pid, struct task_struct *task)
2871 {
2872         int err = lock_trace(task);
2873         if (!err) {
2874                 seq_printf(m, "%08x\n", task->personality);
2875                 unlock_trace(task);
2876         }
2877         return err;
2878 }
2879
2880 #ifdef CONFIG_LIVEPATCH
2881 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2882                                 struct pid *pid, struct task_struct *task)
2883 {
2884         seq_printf(m, "%d\n", task->patch_state);
2885         return 0;
2886 }
2887 #endif /* CONFIG_LIVEPATCH */
2888
2889 /*
2890  * Thread groups
2891  */
2892 static const struct file_operations proc_task_operations;
2893 static const struct inode_operations proc_task_inode_operations;
2894
2895 static const struct pid_entry tgid_base_stuff[] = {
2896         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2897         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2898         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2899         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2900         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2901 #ifdef CONFIG_NET
2902         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2903 #endif
2904         REG("environ",    S_IRUSR, proc_environ_operations),
2905         REG("auxv",       S_IRUSR, proc_auxv_operations),
2906         ONE("status",     S_IRUGO, proc_pid_status),
2907         ONE("personality", S_IRUSR, proc_pid_personality),
2908         ONE("limits",     S_IRUGO, proc_pid_limits),
2909 #ifdef CONFIG_SCHED_DEBUG
2910         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2911 #endif
2912 #ifdef CONFIG_SCHED_AUTOGROUP
2913         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2914 #endif
2915         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2916 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2917         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2918 #endif
2919         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2920         ONE("stat",       S_IRUGO, proc_tgid_stat),
2921         ONE("statm",      S_IRUGO, proc_pid_statm),
2922         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2923 #ifdef CONFIG_NUMA
2924         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2925 #endif
2926         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2927         LNK("cwd",        proc_cwd_link),
2928         LNK("root",       proc_root_link),
2929         LNK("exe",        proc_exe_link),
2930         REG("mounts",     S_IRUGO, proc_mounts_operations),
2931         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2932         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2933 #ifdef CONFIG_PROC_PAGE_MONITOR
2934         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2935         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2936         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2937 #endif
2938 #ifdef CONFIG_SECURITY
2939         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2940 #endif
2941 #ifdef CONFIG_KALLSYMS
2942         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2943 #endif
2944 #ifdef CONFIG_STACKTRACE
2945         ONE("stack",      S_IRUSR, proc_pid_stack),
2946 #endif
2947 #ifdef CONFIG_SCHED_INFO
2948         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2949 #endif
2950 #ifdef CONFIG_LATENCYTOP
2951         REG("latency",  S_IRUGO, proc_lstats_operations),
2952 #endif
2953 #ifdef CONFIG_PROC_PID_CPUSET
2954         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2955 #endif
2956 #ifdef CONFIG_CGROUPS
2957         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2958 #endif
2959         ONE("oom_score",  S_IRUGO, proc_oom_score),
2960         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2961         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2962 #ifdef CONFIG_AUDITSYSCALL
2963         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2964         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2965 #endif
2966 #ifdef CONFIG_FAULT_INJECTION
2967         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2968 #endif
2969 #ifdef CONFIG_ELF_CORE
2970         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2971 #endif
2972 #ifdef CONFIG_TASK_IO_ACCOUNTING
2973         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2974 #endif
2975 #ifdef CONFIG_HARDWALL
2976         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2977 #endif
2978 #ifdef CONFIG_USER_NS
2979         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2980         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2981         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2982         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2983 #endif
2984 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2985         REG("timers",     S_IRUGO, proc_timers_operations),
2986 #endif
2987         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2988 #ifdef CONFIG_LIVEPATCH
2989         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2990 #endif
2991 };
2992
2993 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2994 {
2995         return proc_pident_readdir(file, ctx,
2996                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2997 }
2998
2999 static const struct file_operations proc_tgid_base_operations = {
3000         .read           = generic_read_dir,
3001         .iterate_shared = proc_tgid_base_readdir,
3002         .llseek         = generic_file_llseek,
3003 };
3004
3005 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3006 {
3007         return proc_pident_lookup(dir, dentry,
3008                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3009 }
3010
3011 static const struct inode_operations proc_tgid_base_inode_operations = {
3012         .lookup         = proc_tgid_base_lookup,
3013         .getattr        = pid_getattr,
3014         .setattr        = proc_setattr,
3015         .permission     = proc_pid_permission,
3016 };
3017
3018 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3019 {
3020         struct dentry *dentry, *leader, *dir;
3021         char buf[PROC_NUMBUF];
3022         struct qstr name;
3023
3024         name.name = buf;
3025         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3026         /* no ->d_hash() rejects on procfs */
3027         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3028         if (dentry) {
3029                 d_invalidate(dentry);
3030                 dput(dentry);
3031         }
3032
3033         if (pid == tgid)
3034                 return;
3035
3036         name.name = buf;
3037         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3038         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3039         if (!leader)
3040                 goto out;
3041
3042         name.name = "task";
3043         name.len = strlen(name.name);
3044         dir = d_hash_and_lookup(leader, &name);
3045         if (!dir)
3046                 goto out_put_leader;
3047
3048         name.name = buf;
3049         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3050         dentry = d_hash_and_lookup(dir, &name);
3051         if (dentry) {
3052                 d_invalidate(dentry);
3053                 dput(dentry);
3054         }
3055
3056         dput(dir);
3057 out_put_leader:
3058         dput(leader);
3059 out:
3060         return;
3061 }
3062
3063 /**
3064  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3065  * @task: task that should be flushed.
3066  *
3067  * When flushing dentries from proc, one needs to flush them from global
3068  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3069  * in. This call is supposed to do all of this job.
3070  *
3071  * Looks in the dcache for
3072  * /proc/@pid
3073  * /proc/@tgid/task/@pid
3074  * if either directory is present flushes it and all of it'ts children
3075  * from the dcache.
3076  *
3077  * It is safe and reasonable to cache /proc entries for a task until
3078  * that task exits.  After that they just clog up the dcache with
3079  * useless entries, possibly causing useful dcache entries to be
3080  * flushed instead.  This routine is proved to flush those useless
3081  * dcache entries at process exit time.
3082  *
3083  * NOTE: This routine is just an optimization so it does not guarantee
3084  *       that no dcache entries will exist at process exit time it
3085  *       just makes it very unlikely that any will persist.
3086  */
3087
3088 void proc_flush_task(struct task_struct *task)
3089 {
3090         int i;
3091         struct pid *pid, *tgid;
3092         struct upid *upid;
3093
3094         pid = task_pid(task);
3095         tgid = task_tgid(task);
3096
3097         for (i = 0; i <= pid->level; i++) {
3098                 upid = &pid->numbers[i];
3099                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3100                                         tgid->numbers[i].nr);
3101         }
3102 }
3103
3104 static int proc_pid_instantiate(struct inode *dir,
3105                                    struct dentry * dentry,
3106                                    struct task_struct *task, const void *ptr)
3107 {
3108         struct inode *inode;
3109
3110         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3111         if (!inode)
3112                 goto out;
3113
3114         inode->i_op = &proc_tgid_base_inode_operations;
3115         inode->i_fop = &proc_tgid_base_operations;
3116         inode->i_flags|=S_IMMUTABLE;
3117
3118         set_nlink(inode, nlink_tgid);
3119
3120         d_set_d_op(dentry, &pid_dentry_operations);
3121
3122         d_add(dentry, inode);
3123         /* Close the race of the process dying before we return the dentry */
3124         if (pid_revalidate(dentry, 0))
3125                 return 0;
3126 out:
3127         return -ENOENT;
3128 }
3129
3130 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3131 {
3132         int result = -ENOENT;
3133         struct task_struct *task;
3134         unsigned tgid;
3135         struct pid_namespace *ns;
3136
3137         tgid = name_to_int(&dentry->d_name);
3138         if (tgid == ~0U)
3139                 goto out;
3140
3141         ns = dentry->d_sb->s_fs_info;
3142         rcu_read_lock();
3143         task = find_task_by_pid_ns(tgid, ns);
3144         if (task)
3145                 get_task_struct(task);
3146         rcu_read_unlock();
3147         if (!task)
3148                 goto out;
3149
3150         result = proc_pid_instantiate(dir, dentry, task, NULL);
3151         put_task_struct(task);
3152 out:
3153         return ERR_PTR(result);
3154 }
3155
3156 /*
3157  * Find the first task with tgid >= tgid
3158  *
3159  */
3160 struct tgid_iter {
3161         unsigned int tgid;
3162         struct task_struct *task;
3163 };
3164 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3165 {
3166         struct pid *pid;
3167
3168         if (iter.task)
3169                 put_task_struct(iter.task);
3170         rcu_read_lock();
3171 retry:
3172         iter.task = NULL;
3173         pid = find_ge_pid(iter.tgid, ns);
3174         if (pid) {
3175                 iter.tgid = pid_nr_ns(pid, ns);
3176                 iter.task = pid_task(pid, PIDTYPE_PID);
3177                 /* What we to know is if the pid we have find is the
3178                  * pid of a thread_group_leader.  Testing for task
3179                  * being a thread_group_leader is the obvious thing
3180                  * todo but there is a window when it fails, due to
3181                  * the pid transfer logic in de_thread.
3182                  *
3183                  * So we perform the straight forward test of seeing
3184                  * if the pid we have found is the pid of a thread
3185                  * group leader, and don't worry if the task we have
3186                  * found doesn't happen to be a thread group leader.
3187                  * As we don't care in the case of readdir.
3188                  */
3189                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3190                         iter.tgid += 1;
3191                         goto retry;
3192                 }
3193                 get_task_struct(iter.task);
3194         }
3195         rcu_read_unlock();
3196         return iter;
3197 }
3198
3199 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3200
3201 /* for the /proc/ directory itself, after non-process stuff has been done */
3202 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3203 {
3204         struct tgid_iter iter;
3205         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3206         loff_t pos = ctx->pos;
3207
3208         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3209                 return 0;
3210
3211         if (pos == TGID_OFFSET - 2) {
3212                 struct inode *inode = d_inode(ns->proc_self);
3213                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3214                         return 0;
3215                 ctx->pos = pos = pos + 1;
3216         }
3217         if (pos == TGID_OFFSET - 1) {
3218                 struct inode *inode = d_inode(ns->proc_thread_self);
3219                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3220                         return 0;
3221                 ctx->pos = pos = pos + 1;
3222         }
3223         iter.tgid = pos - TGID_OFFSET;
3224         iter.task = NULL;
3225         for (iter = next_tgid(ns, iter);
3226              iter.task;
3227              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3228                 char name[PROC_NUMBUF];
3229                 int len;
3230
3231                 cond_resched();
3232                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3233                         continue;
3234
3235                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3236                 ctx->pos = iter.tgid + TGID_OFFSET;
3237                 if (!proc_fill_cache(file, ctx, name, len,
3238                                      proc_pid_instantiate, iter.task, NULL)) {
3239                         put_task_struct(iter.task);
3240                         return 0;
3241                 }
3242         }
3243         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3244         return 0;
3245 }
3246
3247 /*
3248  * proc_tid_comm_permission is a special permission function exclusively
3249  * used for the node /proc/<pid>/task/<tid>/comm.
3250  * It bypasses generic permission checks in the case where a task of the same
3251  * task group attempts to access the node.
3252  * The rationale behind this is that glibc and bionic access this node for
3253  * cross thread naming (pthread_set/getname_np(!self)). However, if
3254  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3255  * which locks out the cross thread naming implementation.
3256  * This function makes sure that the node is always accessible for members of
3257  * same thread group.
3258  */
3259 static int proc_tid_comm_permission(struct inode *inode, int mask)
3260 {
3261         bool is_same_tgroup;
3262         struct task_struct *task;
3263
3264         task = get_proc_task(inode);
3265         if (!task)
3266                 return -ESRCH;
3267         is_same_tgroup = same_thread_group(current, task);
3268         put_task_struct(task);
3269
3270         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3271                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3272                  * read or written by the members of the corresponding
3273                  * thread group.
3274                  */
3275                 return 0;
3276         }
3277
3278         return generic_permission(inode, mask);
3279 }
3280
3281 static const struct inode_operations proc_tid_comm_inode_operations = {
3282                 .permission = proc_tid_comm_permission,
3283 };
3284
3285 /*
3286  * Tasks
3287  */
3288 static const struct pid_entry tid_base_stuff[] = {
3289         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3290         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3291         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3292 #ifdef CONFIG_NET
3293         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3294 #endif
3295         REG("environ",   S_IRUSR, proc_environ_operations),
3296         REG("auxv",      S_IRUSR, proc_auxv_operations),
3297         ONE("status",    S_IRUGO, proc_pid_status),
3298         ONE("personality", S_IRUSR, proc_pid_personality),
3299         ONE("limits",    S_IRUGO, proc_pid_limits),
3300 #ifdef CONFIG_SCHED_DEBUG
3301         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3302 #endif
3303         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3304                          &proc_tid_comm_inode_operations,
3305                          &proc_pid_set_comm_operations, {}),
3306 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3307         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3308 #endif
3309         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3310         ONE("stat",      S_IRUGO, proc_tid_stat),
3311         ONE("statm",     S_IRUGO, proc_pid_statm),
3312         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3313 #ifdef CONFIG_PROC_CHILDREN
3314         REG("children",  S_IRUGO, proc_tid_children_operations),
3315 #endif
3316 #ifdef CONFIG_NUMA
3317         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3318 #endif
3319         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3320         LNK("cwd",       proc_cwd_link),
3321         LNK("root",      proc_root_link),
3322         LNK("exe",       proc_exe_link),
3323         REG("mounts",    S_IRUGO, proc_mounts_operations),
3324         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3325 #ifdef CONFIG_PROC_PAGE_MONITOR
3326         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3327         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3328         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3329 #endif
3330 #ifdef CONFIG_SECURITY
3331         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3332 #endif
3333 #ifdef CONFIG_KALLSYMS
3334         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3335 #endif
3336 #ifdef CONFIG_STACKTRACE
3337         ONE("stack",      S_IRUSR, proc_pid_stack),
3338 #endif
3339 #ifdef CONFIG_SCHED_INFO
3340         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3341 #endif
3342 #ifdef CONFIG_LATENCYTOP
3343         REG("latency",  S_IRUGO, proc_lstats_operations),
3344 #endif
3345 #ifdef CONFIG_PROC_PID_CPUSET
3346         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3347 #endif
3348 #ifdef CONFIG_CGROUPS
3349         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3350 #endif
3351         ONE("oom_score", S_IRUGO, proc_oom_score),
3352         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3353         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3354 #ifdef CONFIG_AUDITSYSCALL
3355         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3356         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3357 #endif
3358 #ifdef CONFIG_FAULT_INJECTION
3359         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3360         /*
3361          * Operations on the file check that the task is current,
3362          * so we create it with 0666 to support testing under unprivileged user.
3363          */
3364         REG("fail-nth", 0666, proc_fail_nth_operations),
3365 #endif
3366 #ifdef CONFIG_TASK_IO_ACCOUNTING
3367         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3368 #endif
3369 #ifdef CONFIG_HARDWALL
3370         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3371 #endif
3372 #ifdef CONFIG_USER_NS
3373         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3374         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3375         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3376         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3377 #endif
3378 #ifdef CONFIG_LIVEPATCH
3379         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3380 #endif
3381 };
3382
3383 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3384 {
3385         return proc_pident_readdir(file, ctx,
3386                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3387 }
3388
3389 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3390 {
3391         return proc_pident_lookup(dir, dentry,
3392                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3393 }
3394
3395 static const struct file_operations proc_tid_base_operations = {
3396         .read           = generic_read_dir,
3397         .iterate_shared = proc_tid_base_readdir,
3398         .llseek         = generic_file_llseek,
3399 };
3400
3401 static const struct inode_operations proc_tid_base_inode_operations = {
3402         .lookup         = proc_tid_base_lookup,
3403         .getattr        = pid_getattr,
3404         .setattr        = proc_setattr,
3405 };
3406
3407 static int proc_task_instantiate(struct inode *dir,
3408         struct dentry *dentry, struct task_struct *task, const void *ptr)
3409 {
3410         struct inode *inode;
3411         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3412
3413         if (!inode)
3414                 goto out;
3415         inode->i_op = &proc_tid_base_inode_operations;
3416         inode->i_fop = &proc_tid_base_operations;
3417         inode->i_flags|=S_IMMUTABLE;
3418
3419         set_nlink(inode, nlink_tid);
3420
3421         d_set_d_op(dentry, &pid_dentry_operations);
3422
3423         d_add(dentry, inode);
3424         /* Close the race of the process dying before we return the dentry */
3425         if (pid_revalidate(dentry, 0))
3426                 return 0;
3427 out:
3428         return -ENOENT;
3429 }
3430
3431 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3432 {
3433         int result = -ENOENT;
3434         struct task_struct *task;
3435         struct task_struct *leader = get_proc_task(dir);
3436         unsigned tid;
3437         struct pid_namespace *ns;
3438
3439         if (!leader)
3440                 goto out_no_task;
3441
3442         tid = name_to_int(&dentry->d_name);
3443         if (tid == ~0U)
3444                 goto out;
3445
3446         ns = dentry->d_sb->s_fs_info;
3447         rcu_read_lock();
3448         task = find_task_by_pid_ns(tid, ns);
3449         if (task)
3450                 get_task_struct(task);
3451         rcu_read_unlock();
3452         if (!task)
3453                 goto out;
3454         if (!same_thread_group(leader, task))
3455                 goto out_drop_task;
3456
3457         result = proc_task_instantiate(dir, dentry, task, NULL);
3458 out_drop_task:
3459         put_task_struct(task);
3460 out:
3461         put_task_struct(leader);
3462 out_no_task:
3463         return ERR_PTR(result);
3464 }
3465
3466 /*
3467  * Find the first tid of a thread group to return to user space.
3468  *
3469  * Usually this is just the thread group leader, but if the users
3470  * buffer was too small or there was a seek into the middle of the
3471  * directory we have more work todo.
3472  *
3473  * In the case of a short read we start with find_task_by_pid.
3474  *
3475  * In the case of a seek we start with the leader and walk nr
3476  * threads past it.
3477  */
3478 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3479                                         struct pid_namespace *ns)
3480 {
3481         struct task_struct *pos, *task;
3482         unsigned long nr = f_pos;
3483
3484         if (nr != f_pos)        /* 32bit overflow? */
3485                 return NULL;
3486
3487         rcu_read_lock();
3488         task = pid_task(pid, PIDTYPE_PID);
3489         if (!task)
3490                 goto fail;
3491
3492         /* Attempt to start with the tid of a thread */
3493         if (tid && nr) {
3494                 pos = find_task_by_pid_ns(tid, ns);
3495                 if (pos && same_thread_group(pos, task))
3496                         goto found;
3497         }
3498
3499         /* If nr exceeds the number of threads there is nothing todo */
3500         if (nr >= get_nr_threads(task))
3501                 goto fail;
3502
3503         /* If we haven't found our starting place yet start
3504          * with the leader and walk nr threads forward.
3505          */
3506         pos = task = task->group_leader;
3507         do {
3508                 if (!nr--)
3509                         goto found;
3510         } while_each_thread(task, pos);
3511 fail:
3512         pos = NULL;
3513         goto out;
3514 found:
3515         get_task_struct(pos);
3516 out:
3517         rcu_read_unlock();
3518         return pos;
3519 }
3520
3521 /*
3522  * Find the next thread in the thread list.
3523  * Return NULL if there is an error or no next thread.
3524  *
3525  * The reference to the input task_struct is released.
3526  */
3527 static struct task_struct *next_tid(struct task_struct *start)
3528 {
3529         struct task_struct *pos = NULL;
3530         rcu_read_lock();
3531         if (pid_alive(start)) {
3532                 pos = next_thread(start);
3533                 if (thread_group_leader(pos))
3534                         pos = NULL;
3535                 else
3536                         get_task_struct(pos);
3537         }
3538         rcu_read_unlock();
3539         put_task_struct(start);
3540         return pos;
3541 }
3542
3543 /* for the /proc/TGID/task/ directories */
3544 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3545 {
3546         struct inode *inode = file_inode(file);
3547         struct task_struct *task;
3548         struct pid_namespace *ns;
3549         int tid;
3550
3551         if (proc_inode_is_dead(inode))
3552                 return -ENOENT;
3553
3554         if (!dir_emit_dots(file, ctx))
3555                 return 0;
3556
3557         /* f_version caches the tgid value that the last readdir call couldn't
3558          * return. lseek aka telldir automagically resets f_version to 0.
3559          */
3560         ns = inode->i_sb->s_fs_info;
3561         tid = (int)file->f_version;
3562         file->f_version = 0;
3563         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3564              task;
3565              task = next_tid(task), ctx->pos++) {
3566                 char name[PROC_NUMBUF];
3567                 int len;
3568                 tid = task_pid_nr_ns(task, ns);
3569                 len = snprintf(name, sizeof(name), "%d", tid);
3570                 if (!proc_fill_cache(file, ctx, name, len,
3571                                 proc_task_instantiate, task, NULL)) {
3572                         /* returning this tgid failed, save it as the first
3573                          * pid for the next readir call */
3574                         file->f_version = (u64)tid;
3575                         put_task_struct(task);
3576                         break;
3577                 }
3578         }
3579
3580         return 0;
3581 }
3582
3583 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3584                              u32 request_mask, unsigned int query_flags)
3585 {
3586         struct inode *inode = d_inode(path->dentry);
3587         struct task_struct *p = get_proc_task(inode);
3588         generic_fillattr(inode, stat);
3589
3590         if (p) {
3591                 stat->nlink += get_nr_threads(p);
3592                 put_task_struct(p);
3593         }
3594
3595         return 0;
3596 }
3597
3598 static const struct inode_operations proc_task_inode_operations = {
3599         .lookup         = proc_task_lookup,
3600         .getattr        = proc_task_getattr,
3601         .setattr        = proc_setattr,
3602         .permission     = proc_pid_permission,
3603 };
3604
3605 static const struct file_operations proc_task_operations = {
3606         .read           = generic_read_dir,
3607         .iterate_shared = proc_task_readdir,
3608         .llseek         = generic_file_llseek,
3609 };
3610
3611 void __init set_proc_pid_nlink(void)
3612 {
3613         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3614         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3615 }