fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <[email protected]>
  21  *  Edjard Mota <[email protected]>
  22  *  Ilias Biris <[email protected]>
  23  *  Mauricio Lin <[email protected]>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <[email protected]>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <[email protected]>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/generic-radix-tree.h>
  63 #include <linux/string.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/namei.h>
  66 #include <linux/mnt_namespace.h>
  67 #include <linux/mm.h>
  68 #include <linux/swap.h>
  69 #include <linux/rcupdate.h>
  70 #include <linux/kallsyms.h>
  71 #include <linux/stacktrace.h>
  72 #include <linux/resource.h>
  73 #include <linux/module.h>
  74 #include <linux/mount.h>
  75 #include <linux/security.h>
  76 #include <linux/ptrace.h>
  77 #include <linux/printk.h>
  78 #include <linux/cache.h>
  79 #include <linux/cgroup.h>
  80 #include <linux/cpuset.h>
  81 #include <linux/audit.h>
  82 #include <linux/poll.h>
  83 #include <linux/nsproxy.h>
  84 #include <linux/oom.h>
  85 #include <linux/elf.h>
  86 #include <linux/pid_namespace.h>
  87 #include <linux/user_namespace.h>
  88 #include <linux/fs_struct.h>
  89 #include <linux/slab.h>
  90 #include <linux/sched/autogroup.h>
  91 #include <linux/sched/mm.h>
  92 #include <linux/sched/coredump.h>
  93 #include <linux/sched/debug.h>
  94 #include <linux/sched/stat.h>
  95 #include <linux/posix-timers.h>
  96 #include <linux/time_namespace.h>
  97 #include <linux/resctrl.h>
  98 #include <linux/cn_proc.h>
  99 #include <linux/ksm.h>
 100 #include <trace/events/oom.h>
 101 #include "internal.h"
 102 #include "fd.h"
 103
 104 #include "../../lib/kstrtox.h"
 105
 106 /* NOTE:
 107  *      Implementing inode permission operations in /proc is almost
 108  *      certainly an error.  Permission checks need to happen during
 109  *      each system call not at open time.  The reason is that most of
 110  *      what we wish to check for permissions in /proc varies at runtime.
 111  *
 112  *      The classic example of a problem is opening file descriptors
 113  *      in /proc for a task before it execs a suid executable.
 114  */
 115
 116 static u8 nlink_tid __ro_after_init;
 117 static u8 nlink_tgid __ro_after_init;
 118
 119 struct pid_entry {
 120         const char *name;
 121         unsigned int len;
 122         umode_t mode;
 123         const struct inode_operations *iop;
 124         const struct file_operations *fop;
 125         union proc_op op;
 126 };
 127
 128 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 129         .name = (NAME),                                 \
 130         .len  = sizeof(NAME) - 1,                       \
 131         .mode = MODE,                                   \
 132         .iop  = IOP,                                    \
 133         .fop  = FOP,                                    \
 134         .op   = OP,                                     \
 135 }
 136
 137 #define DIR(NAME, MODE, iops, fops)     \
 138         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 139 #define LNK(NAME, get_link)                                     \
 140         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 141                 &proc_pid_link_inode_operations, NULL,          \
 142                 { .proc_get_link = get_link } )
 143 #define REG(NAME, MODE, fops)                           \
 144         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 145 #define ONE(NAME, MODE, show)                           \
 146         NOD(NAME, (S_IFREG|(MODE)),                     \
 147                 NULL, &proc_single_file_operations,     \
 148                 { .proc_show = show } )
 149 #define ATTR(LSM, NAME, MODE)                           \
 150         NOD(NAME, (S_IFREG|(MODE)),                     \
 151                 NULL, &proc_pid_attr_operations,        \
 152                 { .lsm = LSM })
 153
 154 /*
 155  * Count the number of hardlinks for the pid_entry table, excluding the .
 156  * and .. links.
 157  */
 158 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 159         unsigned int n)
 160 {
 161         unsigned int i;
 162         unsigned int count;
 163
 164         count = 2;
 165         for (i = 0; i < n; ++i) {
 166                 if (S_ISDIR(entries[i].mode))
 167                         ++count;
 168         }
 169
 170         return count;
 171 }
 172
 173 static int get_task_root(struct task_struct *task, struct path *root)
 174 {
 175         int result = -ENOENT;
 176
 177         task_lock(task);
 178         if (task->fs) {
 179                 get_fs_root(task->fs, root);
 180                 result = 0;
 181         }
 182         task_unlock(task);
 183         return result;
 184 }
 185
 186 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 187 {
 188         struct task_struct *task = get_proc_task(d_inode(dentry));
 189         int result = -ENOENT;
 190
 191         if (task) {
 192                 task_lock(task);
 193                 if (task->fs) {
 194                         get_fs_pwd(task->fs, path);
 195                         result = 0;
 196                 }
 197                 task_unlock(task);
 198                 put_task_struct(task);
 199         }
 200         return result;
 201 }
 202
 203 static int proc_root_link(struct dentry *dentry, struct path *path)
 204 {
 205         struct task_struct *task = get_proc_task(d_inode(dentry));
 206         int result = -ENOENT;
 207
 208         if (task) {
 209                 result = get_task_root(task, path);
 210                 put_task_struct(task);
 211         }
 212         return result;
 213 }
 214
 215 /*
 216  * If the user used setproctitle(), we just get the string from
 217  * user space at arg_start, and limit it to a maximum of one page.
 218  */
 219 static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
 220                                 size_t count, unsigned long pos,
 221                                 unsigned long arg_start)
 222 {
 223         char *page;
 224         int ret, got;
 225
 226         if (pos >= PAGE_SIZE)
 227                 return 0;
 228
 229         page = (char *)__get_free_page(GFP_KERNEL);
 230         if (!page)
 231                 return -ENOMEM;
 232
 233         ret = 0;
 234         got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
 235         if (got > 0) {
 236                 int len = strnlen(page, got);
 237
 238                 /* Include the NUL character if it was found */
 239                 if (len < got)
 240                         len++;
 241
 242                 if (len > pos) {
 243                         len -= pos;
 244                         if (len > count)
 245                                 len = count;
 246                         len -= copy_to_user(buf, page+pos, len);
 247                         if (!len)
 248                                 len = -EFAULT;
 249                         ret = len;
 250                 }
 251         }
 252         free_page((unsigned long)page);
 253         return ret;
 254 }
 255
 256 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 257                               size_t count, loff_t *ppos)
 258 {
 259         unsigned long arg_start, arg_end, env_start, env_end;
 260         unsigned long pos, len;
 261         char *page, c;
 262
 263         /* Check if process spawned far enough to have cmdline. */
 264         if (!mm->env_end)
 265                 return 0;
 266
 267         spin_lock(&mm->arg_lock);
 268         arg_start = mm->arg_start;
 269         arg_end = mm->arg_end;
 270         env_start = mm->env_start;
 271         env_end = mm->env_end;
 272         spin_unlock(&mm->arg_lock);
 273
 274         if (arg_start >= arg_end)
 275                 return 0;
 276
 277         /*
 278          * We allow setproctitle() to overwrite the argument
 279          * strings, and overflow past the original end. But
 280          * only when it overflows into the environment area.
 281          */
 282         if (env_start != arg_end || env_end < env_start)
 283                 env_start = env_end = arg_end;
 284         len = env_end - arg_start;
 285
 286         /* We're not going to care if "*ppos" has high bits set */
 287         pos = *ppos;
 288         if (pos >= len)
 289                 return 0;
 290         if (count > len - pos)
 291                 count = len - pos;
 292         if (!count)
 293                 return 0;
 294
 295         /*
 296          * Magical special case: if the argv[] end byte is not
 297          * zero, the user has overwritten it with setproctitle(3).
 298          *
 299          * Possible future enhancement: do this only once when
 300          * pos is 0, and set a flag in the 'struct file'.
 301          */
 302         if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
 303                 return get_mm_proctitle(mm, buf, count, pos, arg_start);
 304
 305         /*
 306          * For the non-setproctitle() case we limit things strictly
 307          * to the [arg_start, arg_end[ range.
 308          */
 309         pos += arg_start;
 310         if (pos < arg_start || pos >= arg_end)
 311                 return 0;
 312         if (count > arg_end - pos)
 313                 count = arg_end - pos;
 314
 315         page = (char *)__get_free_page(GFP_KERNEL);
 316         if (!page)
 317                 return -ENOMEM;
 318
 319         len = 0;
 320         while (count) {
 321                 int got;
 322                 size_t size = min_t(size_t, PAGE_SIZE, count);
 323
 324                 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
 325                 if (got <= 0)
 326                         break;
 327                 got -= copy_to_user(buf, page, got);
 328                 if (unlikely(!got)) {
 329                         if (!len)
 330                                 len = -EFAULT;
 331                         break;
 332                 }
 333                 pos += got;
 334                 buf += got;
 335                 len += got;
 336                 count -= got;
 337         }
 338
 339         free_page((unsigned long)page);
 340         return len;
 341 }
 342
 343 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 344                                 size_t count, loff_t *pos)
 345 {
 346         struct mm_struct *mm;
 347         ssize_t ret;
 348
 349         mm = get_task_mm(tsk);
 350         if (!mm)
 351                 return 0;
 352
 353         ret = get_mm_cmdline(mm, buf, count, pos);
 354         mmput(mm);
 355         return ret;
 356 }
 357
 358 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 359                                      size_t count, loff_t *pos)
 360 {
 361         struct task_struct *tsk;
 362         ssize_t ret;
 363
 364         BUG_ON(*pos < 0);
 365
 366         tsk = get_proc_task(file_inode(file));
 367         if (!tsk)
 368                 return -ESRCH;
 369         ret = get_task_cmdline(tsk, buf, count, pos);
 370         put_task_struct(tsk);
 371         if (ret > 0)
 372                 *pos += ret;
 373         return ret;
 374 }
 375
 376 static const struct file_operations proc_pid_cmdline_ops = {
 377         .read   = proc_pid_cmdline_read,
 378         .llseek = generic_file_llseek,
 379 };
 380
 381 #ifdef CONFIG_KALLSYMS
 382 /*
 383  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 384  * Returns the resolved symbol.  If that fails, simply return the address.
 385  */
 386 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 387                           struct pid *pid, struct task_struct *task)
 388 {
 389         unsigned long wchan;
 390         char symname[KSYM_NAME_LEN];
 391
 392         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 393                 goto print0;
 394
 395         wchan = get_wchan(task);
 396         if (wchan && !lookup_symbol_name(wchan, symname)) {
 397                 seq_puts(m, symname);
 398                 return 0;
 399         }
 400
 401 print0:
 402         seq_putc(m, '0');
 403         return 0;
 404 }
 405 #endif /* CONFIG_KALLSYMS */
 406
 407 static int lock_trace(struct task_struct *task)
 408 {
 409         int err = down_read_killable(&task->signal->exec_update_lock);
 410         if (err)
 411                 return err;
 412         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 413                 up_read(&task->signal->exec_update_lock);
 414                 return -EPERM;
 415         }
 416         return 0;
 417 }
 418
 419 static void unlock_trace(struct task_struct *task)
 420 {
 421         up_read(&task->signal->exec_update_lock);
 422 }
 423
 424 #ifdef CONFIG_STACKTRACE
 425
 426 #define MAX_STACK_TRACE_DEPTH   64
 427
 428 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 429                           struct pid *pid, struct task_struct *task)
 430 {
 431         unsigned long *entries;
 432         int err;
 433
 434         /*
 435          * The ability to racily run the kernel stack unwinder on a running task
 436          * and then observe the unwinder output is scary; while it is useful for
 437          * debugging kernel issues, it can also allow an attacker to leak kernel
 438          * stack contents.
 439          * Doing this in a manner that is at least safe from races would require
 440          * some work to ensure that the remote task can not be scheduled; and
 441          * even then, this would still expose the unwinder as local attack
 442          * surface.
 443          * Therefore, this interface is restricted to root.
 444          */
 445         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 446                 return -EACCES;
 447
 448         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 449                                 GFP_KERNEL);
 450         if (!entries)
 451                 return -ENOMEM;
 452
 453         err = lock_trace(task);
 454         if (!err) {
 455                 unsigned int i, nr_entries;
 456
 457                 nr_entries = stack_trace_save_tsk(task, entries,
 458                                                   MAX_STACK_TRACE_DEPTH, 0);
 459
 460                 for (i = 0; i < nr_entries; i++) {
 461                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 462                 }
 463
 464                 unlock_trace(task);
 465         }
 466         kfree(entries);
 467
 468         return err;
 469 }
 470 #endif
 471
 472 #ifdef CONFIG_SCHED_INFO
 473 /*
 474  * Provides /proc/PID/schedstat
 475  */
 476 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 477                               struct pid *pid, struct task_struct *task)
 478 {
 479         if (unlikely(!sched_info_on()))
 480                 seq_puts(m, "0 0 0\n");
 481         else
 482                 seq_printf(m, "%llu %llu %lu\n",
 483                    (unsigned long long)task->se.sum_exec_runtime,
 484                    (unsigned long long)task->sched_info.run_delay,
 485                    task->sched_info.pcount);
 486
 487         return 0;
 488 }
 489 #endif
 490
 491 #ifdef CONFIG_LATENCYTOP
 492 static int lstats_show_proc(struct seq_file *m, void *v)
 493 {
 494         int i;
 495         struct inode *inode = m->private;
 496         struct task_struct *task = get_proc_task(inode);
 497
 498         if (!task)
 499                 return -ESRCH;
 500         seq_puts(m, "Latency Top version : v0.1\n");
 501         for (i = 0; i < LT_SAVECOUNT; i++) {
 502                 struct latency_record *lr = &task->latency_record[i];
 503                 if (lr->backtrace[0]) {
 504                         int q;
 505                         seq_printf(m, "%i %li %li",
 506                                    lr->count, lr->time, lr->max);
 507                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 508                                 unsigned long bt = lr->backtrace[q];
 509
 510                                 if (!bt)
 511                                         break;
 512                                 seq_printf(m, " %ps", (void *)bt);
 513                         }
 514                         seq_putc(m, '\n');
 515                 }
 516
 517         }
 518         put_task_struct(task);
 519         return 0;
 520 }
 521
 522 static int lstats_open(struct inode *inode, struct file *file)
 523 {
 524         return single_open(file, lstats_show_proc, inode);
 525 }
 526
 527 static ssize_t lstats_write(struct file *file, const char __user *buf,
 528                             size_t count, loff_t *offs)
 529 {
 530         struct task_struct *task = get_proc_task(file_inode(file));
 531
 532         if (!task)
 533                 return -ESRCH;
 534         clear_tsk_latency_tracing(task);
 535         put_task_struct(task);
 536
 537         return count;
 538 }
 539
 540 static const struct file_operations proc_lstats_operations = {
 541         .open           = lstats_open,
 542         .read           = seq_read,
 543         .write          = lstats_write,
 544         .llseek         = seq_lseek,
 545         .release        = single_release,
 546 };
 547
 548 #endif
 549
 550 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 551                           struct pid *pid, struct task_struct *task)
 552 {
 553         unsigned long totalpages = totalram_pages() + total_swap_pages;
 554         unsigned long points = 0;
 555         long badness;
 556
 557         badness = oom_badness(task, totalpages);
 558         /*
 559          * Special case OOM_SCORE_ADJ_MIN for all others scale the
 560          * badness value into [0, 2000] range which we have been
 561          * exporting for a long time so userspace might depend on it.
 562          */
 563         if (badness != LONG_MIN)
 564                 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
 565
 566         seq_printf(m, "%lu\n", points);
 567
 568         return 0;
 569 }
 570
 571 struct limit_names {
 572         const char *name;
 573         const char *unit;
 574 };
 575
 576 static const struct limit_names lnames[RLIM_NLIMITS] = {
 577         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 578         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 579         [RLIMIT_DATA] = {"Max data size", "bytes"},
 580         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 581         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 582         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 583         [RLIMIT_NPROC] = {"Max processes", "processes"},
 584         [RLIMIT_NOFILE] = {"Max open files", "files"},
 585         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 586         [RLIMIT_AS] = {"Max address space", "bytes"},
 587         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 588         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 589         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 590         [RLIMIT_NICE] = {"Max nice priority", NULL},
 591         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 592         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 593 };
 594
 595 /* Display limits for a process */
 596 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 597                            struct pid *pid, struct task_struct *task)
 598 {
 599         unsigned int i;
 600         unsigned long flags;
 601
 602         struct rlimit rlim[RLIM_NLIMITS];
 603
 604         if (!lock_task_sighand(task, &flags))
 605                 return 0;
 606         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 607         unlock_task_sighand(task, &flags);
 608
 609         /*
 610          * print the file header
 611          */
 612         seq_puts(m, "Limit                     "
 613                 "Soft Limit           "
 614                 "Hard Limit           "
 615                 "Units     \n");
 616
 617         for (i = 0; i < RLIM_NLIMITS; i++) {
 618                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 619                         seq_printf(m, "%-25s %-20s ",
 620                                    lnames[i].name, "unlimited");
 621                 else
 622                         seq_printf(m, "%-25s %-20lu ",
 623                                    lnames[i].name, rlim[i].rlim_cur);
 624
 625                 if (rlim[i].rlim_max == RLIM_INFINITY)
 626                         seq_printf(m, "%-20s ", "unlimited");
 627                 else
 628                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 629
 630                 if (lnames[i].unit)
 631                         seq_printf(m, "%-10s\n", lnames[i].unit);
 632                 else
 633                         seq_putc(m, '\n');
 634         }
 635
 636         return 0;
 637 }
 638
 639 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 640 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 641                             struct pid *pid, struct task_struct *task)
 642 {
 643         struct syscall_info info;
 644         u64 *args = &info.data.args[0];
 645         int res;
 646
 647         res = lock_trace(task);
 648         if (res)
 649                 return res;
 650
 651         if (task_current_syscall(task, &info))
 652                 seq_puts(m, "running\n");
 653         else if (info.data.nr < 0)
 654                 seq_printf(m, "%d 0x%llx 0x%llx\n",
 655                            info.data.nr, info.sp, info.data.instruction_pointer);
 656         else
 657                 seq_printf(m,
 658                        "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
 659                        info.data.nr,
 660                        args[0], args[1], args[2], args[3], args[4], args[5],
 661                        info.sp, info.data.instruction_pointer);
 662         unlock_trace(task);
 663
 664         return 0;
 665 }
 666 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 667
 668 /************************************************************************/
 669 /*                       Here the fs part begins                        */
 670 /************************************************************************/
 671
 672 /* permission checks */
 673 static bool proc_fd_access_allowed(struct inode *inode)
 674 {
 675         struct task_struct *task;
 676         bool allowed = false;
 677         /* Allow access to a task's file descriptors if it is us or we
 678          * may use ptrace attach to the process and find out that
 679          * information.
 680          */
 681         task = get_proc_task(inode);
 682         if (task) {
 683                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 684                 put_task_struct(task);
 685         }
 686         return allowed;
 687 }
 688
 689 int proc_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 690                  struct iattr *attr)
 691 {
 692         int error;
 693         struct inode *inode = d_inode(dentry);
 694
 695         if (attr->ia_valid & ATTR_MODE)
 696                 return -EPERM;
 697
 698         error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 699         if (error)
 700                 return error;
 701
 702         setattr_copy(&nop_mnt_idmap, inode, attr);
 703         return 0;
 704 }
 705
 706 /*
 707  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 708  * or euid/egid (for hide_pid_min=2)?
 709  */
 710 static bool has_pid_permissions(struct proc_fs_info *fs_info,
 711                                  struct task_struct *task,
 712                                  enum proc_hidepid hide_pid_min)
 713 {
 714         /*
 715          * If 'hidpid' mount option is set force a ptrace check,
 716          * we indicate that we are using a filesystem syscall
 717          * by passing PTRACE_MODE_READ_FSCREDS
 718          */
 719         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
 720                 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 721
 722         if (fs_info->hide_pid < hide_pid_min)
 723                 return true;
 724         if (in_group_p(fs_info->pid_gid))
 725                 return true;
 726         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 727 }
 728
 729
 730 static int proc_pid_permission(struct mnt_idmap *idmap,
 731                                struct inode *inode, int mask)
 732 {
 733         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
 734         struct task_struct *task;
 735         bool has_perms;
 736
 737         task = get_proc_task(inode);
 738         if (!task)
 739                 return -ESRCH;
 740         has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
 741         put_task_struct(task);
 742
 743         if (!has_perms) {
 744                 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
 745                         /*
 746                          * Let's make getdents(), stat(), and open()
 747                          * consistent with each other.  If a process
 748                          * may not stat() a file, it shouldn't be seen
 749                          * in procfs at all.
 750                          */
 751                         return -ENOENT;
 752                 }
 753
 754                 return -EPERM;
 755         }
 756         return generic_permission(&nop_mnt_idmap, inode, mask);
 757 }
 758
 759
 760
 761 static const struct inode_operations proc_def_inode_operations = {
 762         .setattr        = proc_setattr,
 763 };
 764
 765 static int proc_single_show(struct seq_file *m, void *v)
 766 {
 767         struct inode *inode = m->private;
 768         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
 769         struct pid *pid = proc_pid(inode);
 770         struct task_struct *task;
 771         int ret;
 772
 773         task = get_pid_task(pid, PIDTYPE_PID);
 774         if (!task)
 775                 return -ESRCH;
 776
 777         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 778
 779         put_task_struct(task);
 780         return ret;
 781 }
 782
 783 static int proc_single_open(struct inode *inode, struct file *filp)
 784 {
 785         return single_open(filp, proc_single_show, inode);
 786 }
 787
 788 static const struct file_operations proc_single_file_operations = {
 789         .open           = proc_single_open,
 790         .read           = seq_read,
 791         .llseek         = seq_lseek,
 792         .release        = single_release,
 793 };
 794
 795
 796 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 797 {
 798         struct task_struct *task = get_proc_task(inode);
 799         struct mm_struct *mm = ERR_PTR(-ESRCH);
 800
 801         if (task) {
 802                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 803                 put_task_struct(task);
 804
 805                 if (!IS_ERR_OR_NULL(mm)) {
 806                         /* ensure this mm_struct can't be freed */
 807                         mmgrab(mm);
 808                         /* but do not pin its memory */
 809                         mmput(mm);
 810                 }
 811         }
 812
 813         return mm;
 814 }
 815
 816 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 817 {
 818         struct mm_struct *mm = proc_mem_open(inode, mode);
 819
 820         if (IS_ERR(mm))
 821                 return PTR_ERR(mm);
 822
 823         file->private_data = mm;
 824         return 0;
 825 }
 826
 827 static int mem_open(struct inode *inode, struct file *file)
 828 {
 829         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 830
 831         /* OK to pass negative loff_t, we can catch out-of-range */
 832         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 833
 834         return ret;
 835 }
 836
 837 static ssize_t mem_rw(struct file *file, char __user *buf,
 838                         size_t count, loff_t *ppos, int write)
 839 {
 840         struct mm_struct *mm = file->private_data;
 841         unsigned long addr = *ppos;
 842         ssize_t copied;
 843         char *page;
 844         unsigned int flags;
 845
 846         if (!mm)
 847                 return 0;
 848
 849         page = (char *)__get_free_page(GFP_KERNEL);
 850         if (!page)
 851                 return -ENOMEM;
 852
 853         copied = 0;
 854         if (!mmget_not_zero(mm))
 855                 goto free;
 856
 857         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 858
 859         while (count > 0) {
 860                 size_t this_len = min_t(size_t, count, PAGE_SIZE);
 861
 862                 if (write && copy_from_user(page, buf, this_len)) {
 863                         copied = -EFAULT;
 864                         break;
 865                 }
 866
 867                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 868                 if (!this_len) {
 869                         if (!copied)
 870                                 copied = -EIO;
 871                         break;
 872                 }
 873
 874                 if (!write && copy_to_user(buf, page, this_len)) {
 875                         copied = -EFAULT;
 876                         break;
 877                 }
 878
 879                 buf += this_len;
 880                 addr += this_len;
 881                 copied += this_len;
 882                 count -= this_len;
 883         }
 884         *ppos = addr;
 885
 886         mmput(mm);
 887 free:
 888         free_page((unsigned long) page);
 889         return copied;
 890 }
 891
 892 static ssize_t mem_read(struct file *file, char __user *buf,
 893                         size_t count, loff_t *ppos)
 894 {
 895         return mem_rw(file, buf, count, ppos, 0);
 896 }
 897
 898 static ssize_t mem_write(struct file *file, const char __user *buf,
 899                          size_t count, loff_t *ppos)
 900 {
 901         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 902 }
 903
 904 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 905 {
 906         switch (orig) {
 907         case 0:
 908                 file->f_pos = offset;
 909                 break;
 910         case 1:
 911                 file->f_pos += offset;
 912                 break;
 913         default:
 914                 return -EINVAL;
 915         }
 916         force_successful_syscall_return();
 917         return file->f_pos;
 918 }
 919
 920 static int mem_release(struct inode *inode, struct file *file)
 921 {
 922         struct mm_struct *mm = file->private_data;
 923         if (mm)
 924                 mmdrop(mm);
 925         return 0;
 926 }
 927
 928 static const struct file_operations proc_mem_operations = {
 929         .llseek         = mem_lseek,
 930         .read           = mem_read,
 931         .write          = mem_write,
 932         .open           = mem_open,
 933         .release        = mem_release,
 934 };
 935
 936 static int environ_open(struct inode *inode, struct file *file)
 937 {
 938         return __mem_open(inode, file, PTRACE_MODE_READ);
 939 }
 940
 941 static ssize_t environ_read(struct file *file, char __user *buf,
 942                         size_t count, loff_t *ppos)
 943 {
 944         char *page;
 945         unsigned long src = *ppos;
 946         int ret = 0;
 947         struct mm_struct *mm = file->private_data;
 948         unsigned long env_start, env_end;
 949
 950         /* Ensure the process spawned far enough to have an environment. */
 951         if (!mm || !mm->env_end)
 952                 return 0;
 953
 954         page = (char *)__get_free_page(GFP_KERNEL);
 955         if (!page)
 956                 return -ENOMEM;
 957
 958         ret = 0;
 959         if (!mmget_not_zero(mm))
 960                 goto free;
 961
 962         spin_lock(&mm->arg_lock);
 963         env_start = mm->env_start;
 964         env_end = mm->env_end;
 965         spin_unlock(&mm->arg_lock);
 966
 967         while (count > 0) {
 968                 size_t this_len, max_len;
 969                 int retval;
 970
 971                 if (src >= (env_end - env_start))
 972                         break;
 973
 974                 this_len = env_end - (env_start + src);
 975
 976                 max_len = min_t(size_t, PAGE_SIZE, count);
 977                 this_len = min(max_len, this_len);
 978
 979                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 980
 981                 if (retval <= 0) {
 982                         ret = retval;
 983                         break;
 984                 }
 985
 986                 if (copy_to_user(buf, page, retval)) {
 987                         ret = -EFAULT;
 988                         break;
 989                 }
 990
 991                 ret += retval;
 992                 src += retval;
 993                 buf += retval;
 994                 count -= retval;
 995         }
 996         *ppos = src;
 997         mmput(mm);
 998
 999 free:
1000         free_page((unsigned long) page);
1001         return ret;
1002 }
1003
1004 static const struct file_operations proc_environ_operations = {
1005         .open           = environ_open,
1006         .read           = environ_read,
1007         .llseek         = generic_file_llseek,
1008         .release        = mem_release,
1009 };
1010
1011 static int auxv_open(struct inode *inode, struct file *file)
1012 {
1013         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1014 }
1015
1016 static ssize_t auxv_read(struct file *file, char __user *buf,
1017                         size_t count, loff_t *ppos)
1018 {
1019         struct mm_struct *mm = file->private_data;
1020         unsigned int nwords = 0;
1021
1022         if (!mm)
1023                 return 0;
1024         do {
1025                 nwords += 2;
1026         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1027         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1028                                        nwords * sizeof(mm->saved_auxv[0]));
1029 }
1030
1031 static const struct file_operations proc_auxv_operations = {
1032         .open           = auxv_open,
1033         .read           = auxv_read,
1034         .llseek         = generic_file_llseek,
1035         .release        = mem_release,
1036 };
1037
1038 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1039                             loff_t *ppos)
1040 {
1041         struct task_struct *task = get_proc_task(file_inode(file));
1042         char buffer[PROC_NUMBUF];
1043         int oom_adj = OOM_ADJUST_MIN;
1044         size_t len;
1045
1046         if (!task)
1047                 return -ESRCH;
1048         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1049                 oom_adj = OOM_ADJUST_MAX;
1050         else
1051                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1052                           OOM_SCORE_ADJ_MAX;
1053         put_task_struct(task);
1054         if (oom_adj > OOM_ADJUST_MAX)
1055                 oom_adj = OOM_ADJUST_MAX;
1056         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1057         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1058 }
1059
1060 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1061 {
1062         struct mm_struct *mm = NULL;
1063         struct task_struct *task;
1064         int err = 0;
1065
1066         task = get_proc_task(file_inode(file));
1067         if (!task)
1068                 return -ESRCH;
1069
1070         mutex_lock(&oom_adj_mutex);
1071         if (legacy) {
1072                 if (oom_adj < task->signal->oom_score_adj &&
1073                                 !capable(CAP_SYS_RESOURCE)) {
1074                         err = -EACCES;
1075                         goto err_unlock;
1076                 }
1077                 /*
1078                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1079                  * /proc/pid/oom_score_adj instead.
1080                  */
1081                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1082                           current->comm, task_pid_nr(current), task_pid_nr(task),
1083                           task_pid_nr(task));
1084         } else {
1085                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1086                                 !capable(CAP_SYS_RESOURCE)) {
1087                         err = -EACCES;
1088                         goto err_unlock;
1089                 }
1090         }
1091
1092         /*
1093          * Make sure we will check other processes sharing the mm if this is
1094          * not vfrok which wants its own oom_score_adj.
1095          * pin the mm so it doesn't go away and get reused after task_unlock
1096          */
1097         if (!task->vfork_done) {
1098                 struct task_struct *p = find_lock_task_mm(task);
1099
1100                 if (p) {
1101                         if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1102                                 mm = p->mm;
1103                                 mmgrab(mm);
1104                         }
1105                         task_unlock(p);
1106                 }
1107         }
1108
1109         task->signal->oom_score_adj = oom_adj;
1110         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1111                 task->signal->oom_score_adj_min = (short)oom_adj;
1112         trace_oom_score_adj_update(task);
1113
1114         if (mm) {
1115                 struct task_struct *p;
1116
1117                 rcu_read_lock();
1118                 for_each_process(p) {
1119                         if (same_thread_group(task, p))
1120                                 continue;
1121
1122                         /* do not touch kernel threads or the global init */
1123                         if (p->flags & PF_KTHREAD || is_global_init(p))
1124                                 continue;
1125
1126                         task_lock(p);
1127                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1128                                 p->signal->oom_score_adj = oom_adj;
1129                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1130                                         p->signal->oom_score_adj_min = (short)oom_adj;
1131                         }
1132                         task_unlock(p);
1133                 }
1134                 rcu_read_unlock();
1135                 mmdrop(mm);
1136         }
1137 err_unlock:
1138         mutex_unlock(&oom_adj_mutex);
1139         put_task_struct(task);
1140         return err;
1141 }
1142
1143 /*
1144  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1145  * kernels.  The effective policy is defined by oom_score_adj, which has a
1146  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1147  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1148  * Processes that become oom disabled via oom_adj will still be oom disabled
1149  * with this implementation.
1150  *
1151  * oom_adj cannot be removed since existing userspace binaries use it.
1152  */
1153 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1154                              size_t count, loff_t *ppos)
1155 {
1156         char buffer[PROC_NUMBUF] = {};
1157         int oom_adj;
1158         int err;
1159
1160         if (count > sizeof(buffer) - 1)
1161                 count = sizeof(buffer) - 1;
1162         if (copy_from_user(buffer, buf, count)) {
1163                 err = -EFAULT;
1164                 goto out;
1165         }
1166
1167         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1168         if (err)
1169                 goto out;
1170         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1171              oom_adj != OOM_DISABLE) {
1172                 err = -EINVAL;
1173                 goto out;
1174         }
1175
1176         /*
1177          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1178          * value is always attainable.
1179          */
1180         if (oom_adj == OOM_ADJUST_MAX)
1181                 oom_adj = OOM_SCORE_ADJ_MAX;
1182         else
1183                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1184
1185         err = __set_oom_adj(file, oom_adj, true);
1186 out:
1187         return err < 0 ? err : count;
1188 }
1189
1190 static const struct file_operations proc_oom_adj_operations = {
1191         .read           = oom_adj_read,
1192         .write          = oom_adj_write,
1193         .llseek         = generic_file_llseek,
1194 };
1195
1196 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1197                                         size_t count, loff_t *ppos)
1198 {
1199         struct task_struct *task = get_proc_task(file_inode(file));
1200         char buffer[PROC_NUMBUF];
1201         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1202         size_t len;
1203
1204         if (!task)
1205                 return -ESRCH;
1206         oom_score_adj = task->signal->oom_score_adj;
1207         put_task_struct(task);
1208         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1209         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1210 }
1211
1212 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1213                                         size_t count, loff_t *ppos)
1214 {
1215         char buffer[PROC_NUMBUF] = {};
1216         int oom_score_adj;
1217         int err;
1218
1219         if (count > sizeof(buffer) - 1)
1220                 count = sizeof(buffer) - 1;
1221         if (copy_from_user(buffer, buf, count)) {
1222                 err = -EFAULT;
1223                 goto out;
1224         }
1225
1226         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1227         if (err)
1228                 goto out;
1229         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1230                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1231                 err = -EINVAL;
1232                 goto out;
1233         }
1234
1235         err = __set_oom_adj(file, oom_score_adj, false);
1236 out:
1237         return err < 0 ? err : count;
1238 }
1239
1240 static const struct file_operations proc_oom_score_adj_operations = {
1241         .read           = oom_score_adj_read,
1242         .write          = oom_score_adj_write,
1243         .llseek         = default_llseek,
1244 };
1245
1246 #ifdef CONFIG_AUDIT
1247 #define TMPBUFLEN 11
1248 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1249                                   size_t count, loff_t *ppos)
1250 {
1251         struct inode * inode = file_inode(file);
1252         struct task_struct *task = get_proc_task(inode);
1253         ssize_t length;
1254         char tmpbuf[TMPBUFLEN];
1255
1256         if (!task)
1257                 return -ESRCH;
1258         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1259                            from_kuid(file->f_cred->user_ns,
1260                                      audit_get_loginuid(task)));
1261         put_task_struct(task);
1262         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1263 }
1264
1265 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1266                                    size_t count, loff_t *ppos)
1267 {
1268         struct inode * inode = file_inode(file);
1269         uid_t loginuid;
1270         kuid_t kloginuid;
1271         int rv;
1272
1273         /* Don't let kthreads write their own loginuid */
1274         if (current->flags & PF_KTHREAD)
1275                 return -EPERM;
1276
1277         rcu_read_lock();
1278         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1279                 rcu_read_unlock();
1280                 return -EPERM;
1281         }
1282         rcu_read_unlock();
1283
1284         if (*ppos != 0) {
1285                 /* No partial writes. */
1286                 return -EINVAL;
1287         }
1288
1289         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1290         if (rv < 0)
1291                 return rv;
1292
1293         /* is userspace tring to explicitly UNSET the loginuid? */
1294         if (loginuid == AUDIT_UID_UNSET) {
1295                 kloginuid = INVALID_UID;
1296         } else {
1297                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1298                 if (!uid_valid(kloginuid))
1299                         return -EINVAL;
1300         }
1301
1302         rv = audit_set_loginuid(kloginuid);
1303         if (rv < 0)
1304                 return rv;
1305         return count;
1306 }
1307
1308 static const struct file_operations proc_loginuid_operations = {
1309         .read           = proc_loginuid_read,
1310         .write          = proc_loginuid_write,
1311         .llseek         = generic_file_llseek,
1312 };
1313
1314 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1315                                   size_t count, loff_t *ppos)
1316 {
1317         struct inode * inode = file_inode(file);
1318         struct task_struct *task = get_proc_task(inode);
1319         ssize_t length;
1320         char tmpbuf[TMPBUFLEN];
1321
1322         if (!task)
1323                 return -ESRCH;
1324         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1325                                 audit_get_sessionid(task));
1326         put_task_struct(task);
1327         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1328 }
1329
1330 static const struct file_operations proc_sessionid_operations = {
1331         .read           = proc_sessionid_read,
1332         .llseek         = generic_file_llseek,
1333 };
1334 #endif
1335
1336 #ifdef CONFIG_FAULT_INJECTION
1337 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1338                                       size_t count, loff_t *ppos)
1339 {
1340         struct task_struct *task = get_proc_task(file_inode(file));
1341         char buffer[PROC_NUMBUF];
1342         size_t len;
1343         int make_it_fail;
1344
1345         if (!task)
1346                 return -ESRCH;
1347         make_it_fail = task->make_it_fail;
1348         put_task_struct(task);
1349
1350         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1351
1352         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1353 }
1354
1355 static ssize_t proc_fault_inject_write(struct file * file,
1356                         const char __user * buf, size_t count, loff_t *ppos)
1357 {
1358         struct task_struct *task;
1359         char buffer[PROC_NUMBUF] = {};
1360         int make_it_fail;
1361         int rv;
1362
1363         if (!capable(CAP_SYS_RESOURCE))
1364                 return -EPERM;
1365
1366         if (count > sizeof(buffer) - 1)
1367                 count = sizeof(buffer) - 1;
1368         if (copy_from_user(buffer, buf, count))
1369                 return -EFAULT;
1370         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1371         if (rv < 0)
1372                 return rv;
1373         if (make_it_fail < 0 || make_it_fail > 1)
1374                 return -EINVAL;
1375
1376         task = get_proc_task(file_inode(file));
1377         if (!task)
1378                 return -ESRCH;
1379         task->make_it_fail = make_it_fail;
1380         put_task_struct(task);
1381
1382         return count;
1383 }
1384
1385 static const struct file_operations proc_fault_inject_operations = {
1386         .read           = proc_fault_inject_read,
1387         .write          = proc_fault_inject_write,
1388         .llseek         = generic_file_llseek,
1389 };
1390
1391 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1392                                    size_t count, loff_t *ppos)
1393 {
1394         struct task_struct *task;
1395         int err;
1396         unsigned int n;
1397
1398         err = kstrtouint_from_user(buf, count, 0, &n);
1399         if (err)
1400                 return err;
1401
1402         task = get_proc_task(file_inode(file));
1403         if (!task)
1404                 return -ESRCH;
1405         task->fail_nth = n;
1406         put_task_struct(task);
1407
1408         return count;
1409 }
1410
1411 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1412                                   size_t count, loff_t *ppos)
1413 {
1414         struct task_struct *task;
1415         char numbuf[PROC_NUMBUF];
1416         ssize_t len;
1417
1418         task = get_proc_task(file_inode(file));
1419         if (!task)
1420                 return -ESRCH;
1421         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1422         put_task_struct(task);
1423         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1424 }
1425
1426 static const struct file_operations proc_fail_nth_operations = {
1427         .read           = proc_fail_nth_read,
1428         .write          = proc_fail_nth_write,
1429 };
1430 #endif
1431
1432
1433 #ifdef CONFIG_SCHED_DEBUG
1434 /*
1435  * Print out various scheduling related per-task fields:
1436  */
1437 static int sched_show(struct seq_file *m, void *v)
1438 {
1439         struct inode *inode = m->private;
1440         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1441         struct task_struct *p;
1442
1443         p = get_proc_task(inode);
1444         if (!p)
1445                 return -ESRCH;
1446         proc_sched_show_task(p, ns, m);
1447
1448         put_task_struct(p);
1449
1450         return 0;
1451 }
1452
1453 static ssize_t
1454 sched_write(struct file *file, const char __user *buf,
1455             size_t count, loff_t *offset)
1456 {
1457         struct inode *inode = file_inode(file);
1458         struct task_struct *p;
1459
1460         p = get_proc_task(inode);
1461         if (!p)
1462                 return -ESRCH;
1463         proc_sched_set_task(p);
1464
1465         put_task_struct(p);
1466
1467         return count;
1468 }
1469
1470 static int sched_open(struct inode *inode, struct file *filp)
1471 {
1472         return single_open(filp, sched_show, inode);
1473 }
1474
1475 static const struct file_operations proc_pid_sched_operations = {
1476         .open           = sched_open,
1477         .read           = seq_read,
1478         .write          = sched_write,
1479         .llseek         = seq_lseek,
1480         .release        = single_release,
1481 };
1482
1483 #endif
1484
1485 #ifdef CONFIG_SCHED_AUTOGROUP
1486 /*
1487  * Print out autogroup related information:
1488  */
1489 static int sched_autogroup_show(struct seq_file *m, void *v)
1490 {
1491         struct inode *inode = m->private;
1492         struct task_struct *p;
1493
1494         p = get_proc_task(inode);
1495         if (!p)
1496                 return -ESRCH;
1497         proc_sched_autogroup_show_task(p, m);
1498
1499         put_task_struct(p);
1500
1501         return 0;
1502 }
1503
1504 static ssize_t
1505 sched_autogroup_write(struct file *file, const char __user *buf,
1506             size_t count, loff_t *offset)
1507 {
1508         struct inode *inode = file_inode(file);
1509         struct task_struct *p;
1510         char buffer[PROC_NUMBUF] = {};
1511         int nice;
1512         int err;
1513
1514         if (count > sizeof(buffer) - 1)
1515                 count = sizeof(buffer) - 1;
1516         if (copy_from_user(buffer, buf, count))
1517                 return -EFAULT;
1518
1519         err = kstrtoint(strstrip(buffer), 0, &nice);
1520         if (err < 0)
1521                 return err;
1522
1523         p = get_proc_task(inode);
1524         if (!p)
1525                 return -ESRCH;
1526
1527         err = proc_sched_autogroup_set_nice(p, nice);
1528         if (err)
1529                 count = err;
1530
1531         put_task_struct(p);
1532
1533         return count;
1534 }
1535
1536 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1537 {
1538         int ret;
1539
1540         ret = single_open(filp, sched_autogroup_show, NULL);
1541         if (!ret) {
1542                 struct seq_file *m = filp->private_data;
1543
1544                 m->private = inode;
1545         }
1546         return ret;
1547 }
1548
1549 static const struct file_operations proc_pid_sched_autogroup_operations = {
1550         .open           = sched_autogroup_open,
1551         .read           = seq_read,
1552         .write          = sched_autogroup_write,
1553         .llseek         = seq_lseek,
1554         .release        = single_release,
1555 };
1556
1557 #endif /* CONFIG_SCHED_AUTOGROUP */
1558
1559 #ifdef CONFIG_TIME_NS
1560 static int timens_offsets_show(struct seq_file *m, void *v)
1561 {
1562         struct task_struct *p;
1563
1564         p = get_proc_task(file_inode(m->file));
1565         if (!p)
1566                 return -ESRCH;
1567         proc_timens_show_offsets(p, m);
1568
1569         put_task_struct(p);
1570
1571         return 0;
1572 }
1573
1574 static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1575                                     size_t count, loff_t *ppos)
1576 {
1577         struct inode *inode = file_inode(file);
1578         struct proc_timens_offset offsets[2];
1579         char *kbuf = NULL, *pos, *next_line;
1580         struct task_struct *p;
1581         int ret, noffsets;
1582
1583         /* Only allow < page size writes at the beginning of the file */
1584         if ((*ppos != 0) || (count >= PAGE_SIZE))
1585                 return -EINVAL;
1586
1587         /* Slurp in the user data */
1588         kbuf = memdup_user_nul(buf, count);
1589         if (IS_ERR(kbuf))
1590                 return PTR_ERR(kbuf);
1591
1592         /* Parse the user data */
1593         ret = -EINVAL;
1594         noffsets = 0;
1595         for (pos = kbuf; pos; pos = next_line) {
1596                 struct proc_timens_offset *off = &offsets[noffsets];
1597                 char clock[10];
1598                 int err;
1599
1600                 /* Find the end of line and ensure we don't look past it */
1601                 next_line = strchr(pos, '\n');
1602                 if (next_line) {
1603                         *next_line = '\0';
1604                         next_line++;
1605                         if (*next_line == '\0')
1606                                 next_line = NULL;
1607                 }
1608
1609                 err = sscanf(pos, "%9s %lld %lu", clock,
1610                                 &off->val.tv_sec, &off->val.tv_nsec);
1611                 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1612                         goto out;
1613
1614                 clock[sizeof(clock) - 1] = 0;
1615                 if (strcmp(clock, "monotonic") == 0 ||
1616                     strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1617                         off->clockid = CLOCK_MONOTONIC;
1618                 else if (strcmp(clock, "boottime") == 0 ||
1619                          strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1620                         off->clockid = CLOCK_BOOTTIME;
1621                 else
1622                         goto out;
1623
1624                 noffsets++;
1625                 if (noffsets == ARRAY_SIZE(offsets)) {
1626                         if (next_line)
1627                                 count = next_line - kbuf;
1628                         break;
1629                 }
1630         }
1631
1632         ret = -ESRCH;
1633         p = get_proc_task(inode);
1634         if (!p)
1635                 goto out;
1636         ret = proc_timens_set_offset(file, p, offsets, noffsets);
1637         put_task_struct(p);
1638         if (ret)
1639                 goto out;
1640
1641         ret = count;
1642 out:
1643         kfree(kbuf);
1644         return ret;
1645 }
1646
1647 static int timens_offsets_open(struct inode *inode, struct file *filp)
1648 {
1649         return single_open(filp, timens_offsets_show, inode);
1650 }
1651
1652 static const struct file_operations proc_timens_offsets_operations = {
1653         .open           = timens_offsets_open,
1654         .read           = seq_read,
1655         .write          = timens_offsets_write,
1656         .llseek         = seq_lseek,
1657         .release        = single_release,
1658 };
1659 #endif /* CONFIG_TIME_NS */
1660
1661 static ssize_t comm_write(struct file *file, const char __user *buf,
1662                                 size_t count, loff_t *offset)
1663 {
1664         struct inode *inode = file_inode(file);
1665         struct task_struct *p;
1666         char buffer[TASK_COMM_LEN] = {};
1667         const size_t maxlen = sizeof(buffer) - 1;
1668
1669         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1670                 return -EFAULT;
1671
1672         p = get_proc_task(inode);
1673         if (!p)
1674                 return -ESRCH;
1675
1676         if (same_thread_group(current, p)) {
1677                 set_task_comm(p, buffer);
1678                 proc_comm_connector(p);
1679         }
1680         else
1681                 count = -EINVAL;
1682
1683         put_task_struct(p);
1684
1685         return count;
1686 }
1687
1688 static int comm_show(struct seq_file *m, void *v)
1689 {
1690         struct inode *inode = m->private;
1691         struct task_struct *p;
1692
1693         p = get_proc_task(inode);
1694         if (!p)
1695                 return -ESRCH;
1696
1697         proc_task_name(m, p, false);
1698         seq_putc(m, '\n');
1699
1700         put_task_struct(p);
1701
1702         return 0;
1703 }
1704
1705 static int comm_open(struct inode *inode, struct file *filp)
1706 {
1707         return single_open(filp, comm_show, inode);
1708 }
1709
1710 static const struct file_operations proc_pid_set_comm_operations = {
1711         .open           = comm_open,
1712         .read           = seq_read,
1713         .write          = comm_write,
1714         .llseek         = seq_lseek,
1715         .release        = single_release,
1716 };
1717
1718 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1719 {
1720         struct task_struct *task;
1721         struct file *exe_file;
1722
1723         task = get_proc_task(d_inode(dentry));
1724         if (!task)
1725                 return -ENOENT;
1726         exe_file = get_task_exe_file(task);
1727         put_task_struct(task);
1728         if (exe_file) {
1729                 *exe_path = exe_file->f_path;
1730                 path_get(&exe_file->f_path);
1731                 fput(exe_file);
1732                 return 0;
1733         } else
1734                 return -ENOENT;
1735 }
1736
1737 static const char *proc_pid_get_link(struct dentry *dentry,
1738                                      struct inode *inode,
1739                                      struct delayed_call *done)
1740 {
1741         struct path path;
1742         int error = -EACCES;
1743
1744         if (!dentry)
1745                 return ERR_PTR(-ECHILD);
1746
1747         /* Are we allowed to snoop on the tasks file descriptors? */
1748         if (!proc_fd_access_allowed(inode))
1749                 goto out;
1750
1751         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1752         if (error)
1753                 goto out;
1754
1755         error = nd_jump_link(&path);
1756 out:
1757         return ERR_PTR(error);
1758 }
1759
1760 static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1761 {
1762         char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1763         char *pathname;
1764         int len;
1765
1766         if (!tmp)
1767                 return -ENOMEM;
1768
1769         pathname = d_path(path, tmp, PATH_MAX);
1770         len = PTR_ERR(pathname);
1771         if (IS_ERR(pathname))
1772                 goto out;
1773         len = tmp + PATH_MAX - 1 - pathname;
1774
1775         if (len > buflen)
1776                 len = buflen;
1777         if (copy_to_user(buffer, pathname, len))
1778                 len = -EFAULT;
1779  out:
1780         kfree(tmp);
1781         return len;
1782 }
1783
1784 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1785 {
1786         int error = -EACCES;
1787         struct inode *inode = d_inode(dentry);
1788         struct path path;
1789
1790         /* Are we allowed to snoop on the tasks file descriptors? */
1791         if (!proc_fd_access_allowed(inode))
1792                 goto out;
1793
1794         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1795         if (error)
1796                 goto out;
1797
1798         error = do_proc_readlink(&path, buffer, buflen);
1799         path_put(&path);
1800 out:
1801         return error;
1802 }
1803
1804 const struct inode_operations proc_pid_link_inode_operations = {
1805         .readlink       = proc_pid_readlink,
1806         .get_link       = proc_pid_get_link,
1807         .setattr        = proc_setattr,
1808 };
1809
1810
1811 /* building an inode */
1812
1813 void task_dump_owner(struct task_struct *task, umode_t mode,
1814                      kuid_t *ruid, kgid_t *rgid)
1815 {
1816         /* Depending on the state of dumpable compute who should own a
1817          * proc file for a task.
1818          */
1819         const struct cred *cred;
1820         kuid_t uid;
1821         kgid_t gid;
1822
1823         if (unlikely(task->flags & PF_KTHREAD)) {
1824                 *ruid = GLOBAL_ROOT_UID;
1825                 *rgid = GLOBAL_ROOT_GID;
1826                 return;
1827         }
1828
1829         /* Default to the tasks effective ownership */
1830         rcu_read_lock();
1831         cred = __task_cred(task);
1832         uid = cred->euid;
1833         gid = cred->egid;
1834         rcu_read_unlock();
1835
1836         /*
1837          * Before the /proc/pid/status file was created the only way to read
1838          * the effective uid of a /process was to stat /proc/pid.  Reading
1839          * /proc/pid/status is slow enough that procps and other packages
1840          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1841          * made this apply to all per process world readable and executable
1842          * directories.
1843          */
1844         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1845                 struct mm_struct *mm;
1846                 task_lock(task);
1847                 mm = task->mm;
1848                 /* Make non-dumpable tasks owned by some root */
1849                 if (mm) {
1850                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1851                                 struct user_namespace *user_ns = mm->user_ns;
1852
1853                                 uid = make_kuid(user_ns, 0);
1854                                 if (!uid_valid(uid))
1855                                         uid = GLOBAL_ROOT_UID;
1856
1857                                 gid = make_kgid(user_ns, 0);
1858                                 if (!gid_valid(gid))
1859                                         gid = GLOBAL_ROOT_GID;
1860                         }
1861                 } else {
1862                         uid = GLOBAL_ROOT_UID;
1863                         gid = GLOBAL_ROOT_GID;
1864                 }
1865                 task_unlock(task);
1866         }
1867         *ruid = uid;
1868         *rgid = gid;
1869 }
1870
1871 void proc_pid_evict_inode(struct proc_inode *ei)
1872 {
1873         struct pid *pid = ei->pid;
1874
1875         if (S_ISDIR(ei->vfs_inode.i_mode)) {
1876                 spin_lock(&pid->lock);
1877                 hlist_del_init_rcu(&ei->sibling_inodes);
1878                 spin_unlock(&pid->lock);
1879         }
1880
1881         put_pid(pid);
1882 }
1883
1884 struct inode *proc_pid_make_inode(struct super_block *sb,
1885                                   struct task_struct *task, umode_t mode)
1886 {
1887         struct inode * inode;
1888         struct proc_inode *ei;
1889         struct pid *pid;
1890
1891         /* We need a new inode */
1892
1893         inode = new_inode(sb);
1894         if (!inode)
1895                 goto out;
1896
1897         /* Common stuff */
1898         ei = PROC_I(inode);
1899         inode->i_mode = mode;
1900         inode->i_ino = get_next_ino();
1901         simple_inode_init_ts(inode);
1902         inode->i_op = &proc_def_inode_operations;
1903
1904         /*
1905          * grab the reference to task.
1906          */
1907         pid = get_task_pid(task, PIDTYPE_PID);
1908         if (!pid)
1909                 goto out_unlock;
1910
1911         /* Let the pid remember us for quick removal */
1912         ei->pid = pid;
1913
1914         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1915         security_task_to_inode(task, inode);
1916
1917 out:
1918         return inode;
1919
1920 out_unlock:
1921         iput(inode);
1922         return NULL;
1923 }
1924
1925 /*
1926  * Generating an inode and adding it into @pid->inodes, so that task will
1927  * invalidate inode's dentry before being released.
1928  *
1929  * This helper is used for creating dir-type entries under '/proc' and
1930  * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1931  * can be released by invalidating '/proc/<tgid>' dentry.
1932  * In theory, dentries under '/proc/<tgid>/task' can also be released by
1933  * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1934  * thread exiting situation: Any one of threads should invalidate its
1935  * '/proc/<tgid>/task/<pid>' dentry before released.
1936  */
1937 static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1938                                 struct task_struct *task, umode_t mode)
1939 {
1940         struct inode *inode;
1941         struct proc_inode *ei;
1942         struct pid *pid;
1943
1944         inode = proc_pid_make_inode(sb, task, mode);
1945         if (!inode)
1946                 return NULL;
1947
1948         /* Let proc_flush_pid find this directory inode */
1949         ei = PROC_I(inode);
1950         pid = ei->pid;
1951         spin_lock(&pid->lock);
1952         hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1953         spin_unlock(&pid->lock);
1954
1955         return inode;
1956 }
1957
1958 int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
1959                 struct kstat *stat, u32 request_mask, unsigned int query_flags)
1960 {
1961         struct inode *inode = d_inode(path->dentry);
1962         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
1963         struct task_struct *task;
1964
1965         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
1966
1967         stat->uid = GLOBAL_ROOT_UID;
1968         stat->gid = GLOBAL_ROOT_GID;
1969         rcu_read_lock();
1970         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1971         if (task) {
1972                 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
1973                         rcu_read_unlock();
1974                         /*
1975                          * This doesn't prevent learning whether PID exists,
1976                          * it only makes getattr() consistent with readdir().
1977                          */
1978                         return -ENOENT;
1979                 }
1980                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1981         }
1982         rcu_read_unlock();
1983         return 0;
1984 }
1985
1986 /* dentry stuff */
1987
1988 /*
1989  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1990  */
1991 void pid_update_inode(struct task_struct *task, struct inode *inode)
1992 {
1993         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1994
1995         inode->i_mode &= ~(S_ISUID | S_ISGID);
1996         security_task_to_inode(task, inode);
1997 }
1998
1999 /*
2000  * Rewrite the inode's ownerships here because the owning task may have
2001  * performed a setuid(), etc.
2002  *
2003  */
2004 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2005 {
2006         struct inode *inode;
2007         struct task_struct *task;
2008         int ret = 0;
2009
2010         rcu_read_lock();
2011         inode = d_inode_rcu(dentry);
2012         if (!inode)
2013                 goto out;
2014         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2015
2016         if (task) {
2017                 pid_update_inode(task, inode);
2018                 ret = 1;
2019         }
2020 out:
2021         rcu_read_unlock();
2022         return ret;
2023 }
2024
2025 static inline bool proc_inode_is_dead(struct inode *inode)
2026 {
2027         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2028 }
2029
2030 int pid_delete_dentry(const struct dentry *dentry)
2031 {
2032         /* Is the task we represent dead?
2033          * If so, then don't put the dentry on the lru list,
2034          * kill it immediately.
2035          */
2036         return proc_inode_is_dead(d_inode(dentry));
2037 }
2038
2039 const struct dentry_operations pid_dentry_operations =
2040 {
2041         .d_revalidate   = pid_revalidate,
2042         .d_delete       = pid_delete_dentry,
2043 };
2044
2045 /* Lookups */
2046
2047 /*
2048  * Fill a directory entry.
2049  *
2050  * If possible create the dcache entry and derive our inode number and
2051  * file type from dcache entry.
2052  *
2053  * Since all of the proc inode numbers are dynamically generated, the inode
2054  * numbers do not exist until the inode is cache.  This means creating
2055  * the dcache entry in readdir is necessary to keep the inode numbers
2056  * reported by readdir in sync with the inode numbers reported
2057  * by stat.
2058  */
2059 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2060         const char *name, unsigned int len,
2061         instantiate_t instantiate, struct task_struct *task, const void *ptr)
2062 {
2063         struct dentry *child, *dir = file->f_path.dentry;
2064         struct qstr qname = QSTR_INIT(name, len);
2065         struct inode *inode;
2066         unsigned type = DT_UNKNOWN;
2067         ino_t ino = 1;
2068
2069         child = d_hash_and_lookup(dir, &qname);
2070         if (!child) {
2071                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2072                 child = d_alloc_parallel(dir, &qname, &wq);
2073                 if (IS_ERR(child))
2074                         goto end_instantiate;
2075                 if (d_in_lookup(child)) {
2076                         struct dentry *res;
2077                         res = instantiate(child, task, ptr);
2078                         d_lookup_done(child);
2079                         if (unlikely(res)) {
2080                                 dput(child);
2081                                 child = res;
2082                                 if (IS_ERR(child))
2083                                         goto end_instantiate;
2084                         }
2085                 }
2086         }
2087         inode = d_inode(child);
2088         ino = inode->i_ino;
2089         type = inode->i_mode >> 12;
2090         dput(child);
2091 end_instantiate:
2092         return dir_emit(ctx, name, len, ino, type);
2093 }
2094
2095 /*
2096  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2097  * which represent vma start and end addresses.
2098  */
2099 static int dname_to_vma_addr(struct dentry *dentry,
2100                              unsigned long *start, unsigned long *end)
2101 {
2102         const char *str = dentry->d_name.name;
2103         unsigned long long sval, eval;
2104         unsigned int len;
2105
2106         if (str[0] == '0' && str[1] != '-')
2107                 return -EINVAL;
2108         len = _parse_integer(str, 16, &sval);
2109         if (len & KSTRTOX_OVERFLOW)
2110                 return -EINVAL;
2111         if (sval != (unsigned long)sval)
2112                 return -EINVAL;
2113         str += len;
2114
2115         if (*str != '-')
2116                 return -EINVAL;
2117         str++;
2118
2119         if (str[0] == '0' && str[1])
2120                 return -EINVAL;
2121         len = _parse_integer(str, 16, &eval);
2122         if (len & KSTRTOX_OVERFLOW)
2123                 return -EINVAL;
2124         if (eval != (unsigned long)eval)
2125                 return -EINVAL;
2126         str += len;
2127
2128         if (*str != '\0')
2129                 return -EINVAL;
2130
2131         *start = sval;
2132         *end = eval;
2133
2134         return 0;
2135 }
2136
2137 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2138 {
2139         unsigned long vm_start, vm_end;
2140         bool exact_vma_exists = false;
2141         struct mm_struct *mm = NULL;
2142         struct task_struct *task;
2143         struct inode *inode;
2144         int status = 0;
2145
2146         if (flags & LOOKUP_RCU)
2147                 return -ECHILD;
2148
2149         inode = d_inode(dentry);
2150         task = get_proc_task(inode);
2151         if (!task)
2152                 goto out_notask;
2153
2154         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2155         if (IS_ERR_OR_NULL(mm))
2156                 goto out;
2157
2158         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2159                 status = mmap_read_lock_killable(mm);
2160                 if (!status) {
2161                         exact_vma_exists = !!find_exact_vma(mm, vm_start,
2162                                                             vm_end);
2163                         mmap_read_unlock(mm);
2164                 }
2165         }
2166
2167         mmput(mm);
2168
2169         if (exact_vma_exists) {
2170                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2171
2172                 security_task_to_inode(task, inode);
2173                 status = 1;
2174         }
2175
2176 out:
2177         put_task_struct(task);
2178
2179 out_notask:
2180         return status;
2181 }
2182
2183 static const struct dentry_operations tid_map_files_dentry_operations = {
2184         .d_revalidate   = map_files_d_revalidate,
2185         .d_delete       = pid_delete_dentry,
2186 };
2187
2188 static int map_files_get_link(struct dentry *dentry, struct path *path)
2189 {
2190         unsigned long vm_start, vm_end;
2191         struct vm_area_struct *vma;
2192         struct task_struct *task;
2193         struct mm_struct *mm;
2194         int rc;
2195
2196         rc = -ENOENT;
2197         task = get_proc_task(d_inode(dentry));
2198         if (!task)
2199                 goto out;
2200
2201         mm = get_task_mm(task);
2202         put_task_struct(task);
2203         if (!mm)
2204                 goto out;
2205
2206         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2207         if (rc)
2208                 goto out_mmput;
2209
2210         rc = mmap_read_lock_killable(mm);
2211         if (rc)
2212                 goto out_mmput;
2213
2214         rc = -ENOENT;
2215         vma = find_exact_vma(mm, vm_start, vm_end);
2216         if (vma && vma->vm_file) {
2217                 *path = *file_user_path(vma->vm_file);
2218                 path_get(path);
2219                 rc = 0;
2220         }
2221         mmap_read_unlock(mm);
2222
2223 out_mmput:
2224         mmput(mm);
2225 out:
2226         return rc;
2227 }
2228
2229 struct map_files_info {
2230         unsigned long   start;
2231         unsigned long   end;
2232         fmode_t         mode;
2233 };
2234
2235 /*
2236  * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2237  * to concerns about how the symlinks may be used to bypass permissions on
2238  * ancestor directories in the path to the file in question.
2239  */
2240 static const char *
2241 proc_map_files_get_link(struct dentry *dentry,
2242                         struct inode *inode,
2243                         struct delayed_call *done)
2244 {
2245         if (!checkpoint_restore_ns_capable(&init_user_ns))
2246                 return ERR_PTR(-EPERM);
2247
2248         return proc_pid_get_link(dentry, inode, done);
2249 }
2250
2251 /*
2252  * Identical to proc_pid_link_inode_operations except for get_link()
2253  */
2254 static const struct inode_operations proc_map_files_link_inode_operations = {
2255         .readlink       = proc_pid_readlink,
2256         .get_link       = proc_map_files_get_link,
2257         .setattr        = proc_setattr,
2258 };
2259
2260 static struct dentry *
2261 proc_map_files_instantiate(struct dentry *dentry,
2262                            struct task_struct *task, const void *ptr)
2263 {
2264         fmode_t mode = (fmode_t)(unsigned long)ptr;
2265         struct proc_inode *ei;
2266         struct inode *inode;
2267
2268         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2269                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2270                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2271         if (!inode)
2272                 return ERR_PTR(-ENOENT);
2273
2274         ei = PROC_I(inode);
2275         ei->op.proc_get_link = map_files_get_link;
2276
2277         inode->i_op = &proc_map_files_link_inode_operations;
2278         inode->i_size = 64;
2279
2280         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2281         return d_splice_alias(inode, dentry);
2282 }
2283
2284 static struct dentry *proc_map_files_lookup(struct inode *dir,
2285                 struct dentry *dentry, unsigned int flags)
2286 {
2287         unsigned long vm_start, vm_end;
2288         struct vm_area_struct *vma;
2289         struct task_struct *task;
2290         struct dentry *result;
2291         struct mm_struct *mm;
2292
2293         result = ERR_PTR(-ENOENT);
2294         task = get_proc_task(dir);
2295         if (!task)
2296                 goto out;
2297
2298         result = ERR_PTR(-EACCES);
2299         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2300                 goto out_put_task;
2301
2302         result = ERR_PTR(-ENOENT);
2303         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2304                 goto out_put_task;
2305
2306         mm = get_task_mm(task);
2307         if (!mm)
2308                 goto out_put_task;
2309
2310         result = ERR_PTR(-EINTR);
2311         if (mmap_read_lock_killable(mm))
2312                 goto out_put_mm;
2313
2314         result = ERR_PTR(-ENOENT);
2315         vma = find_exact_vma(mm, vm_start, vm_end);
2316         if (!vma)
2317                 goto out_no_vma;
2318
2319         if (vma->vm_file)
2320                 result = proc_map_files_instantiate(dentry, task,
2321                                 (void *)(unsigned long)vma->vm_file->f_mode);
2322
2323 out_no_vma:
2324         mmap_read_unlock(mm);
2325 out_put_mm:
2326         mmput(mm);
2327 out_put_task:
2328         put_task_struct(task);
2329 out:
2330         return result;
2331 }
2332
2333 static const struct inode_operations proc_map_files_inode_operations = {
2334         .lookup         = proc_map_files_lookup,
2335         .permission     = proc_fd_permission,
2336         .setattr        = proc_setattr,
2337 };
2338
2339 static int
2340 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2341 {
2342         struct vm_area_struct *vma;
2343         struct task_struct *task;
2344         struct mm_struct *mm;
2345         unsigned long nr_files, pos, i;
2346         GENRADIX(struct map_files_info) fa;
2347         struct map_files_info *p;
2348         int ret;
2349         struct vma_iterator vmi;
2350
2351         genradix_init(&fa);
2352
2353         ret = -ENOENT;
2354         task = get_proc_task(file_inode(file));
2355         if (!task)
2356                 goto out;
2357
2358         ret = -EACCES;
2359         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2360                 goto out_put_task;
2361
2362         ret = 0;
2363         if (!dir_emit_dots(file, ctx))
2364                 goto out_put_task;
2365
2366         mm = get_task_mm(task);
2367         if (!mm)
2368                 goto out_put_task;
2369
2370         ret = mmap_read_lock_killable(mm);
2371         if (ret) {
2372                 mmput(mm);
2373                 goto out_put_task;
2374         }
2375
2376         nr_files = 0;
2377
2378         /*
2379          * We need two passes here:
2380          *
2381          *  1) Collect vmas of mapped files with mmap_lock taken
2382          *  2) Release mmap_lock and instantiate entries
2383          *
2384          * otherwise we get lockdep complained, since filldir()
2385          * routine might require mmap_lock taken in might_fault().
2386          */
2387
2388         pos = 2;
2389         vma_iter_init(&vmi, mm, 0);
2390         for_each_vma(vmi, vma) {
2391                 if (!vma->vm_file)
2392                         continue;
2393                 if (++pos <= ctx->pos)
2394                         continue;
2395
2396                 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2397                 if (!p) {
2398                         ret = -ENOMEM;
2399                         mmap_read_unlock(mm);
2400                         mmput(mm);
2401                         goto out_put_task;
2402                 }
2403
2404                 p->start = vma->vm_start;
2405                 p->end = vma->vm_end;
2406                 p->mode = vma->vm_file->f_mode;
2407         }
2408         mmap_read_unlock(mm);
2409         mmput(mm);
2410
2411         for (i = 0; i < nr_files; i++) {
2412                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2413                 unsigned int len;
2414
2415                 p = genradix_ptr(&fa, i);
2416                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2417                 if (!proc_fill_cache(file, ctx,
2418                                       buf, len,
2419                                       proc_map_files_instantiate,
2420                                       task,
2421                                       (void *)(unsigned long)p->mode))
2422                         break;
2423                 ctx->pos++;
2424         }
2425
2426 out_put_task:
2427         put_task_struct(task);
2428 out:
2429         genradix_free(&fa);
2430         return ret;
2431 }
2432
2433 static const struct file_operations proc_map_files_operations = {
2434         .read           = generic_read_dir,
2435         .iterate_shared = proc_map_files_readdir,
2436         .llseek         = generic_file_llseek,
2437 };
2438
2439 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2440 struct timers_private {
2441         struct pid *pid;
2442         struct task_struct *task;
2443         struct sighand_struct *sighand;
2444         struct pid_namespace *ns;
2445         unsigned long flags;
2446 };
2447
2448 static void *timers_start(struct seq_file *m, loff_t *pos)
2449 {
2450         struct timers_private *tp = m->private;
2451
2452         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2453         if (!tp->task)
2454                 return ERR_PTR(-ESRCH);
2455
2456         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2457         if (!tp->sighand)
2458                 return ERR_PTR(-ESRCH);
2459
2460         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2461 }
2462
2463 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2464 {
2465         struct timers_private *tp = m->private;
2466         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2467 }
2468
2469 static void timers_stop(struct seq_file *m, void *v)
2470 {
2471         struct timers_private *tp = m->private;
2472
2473         if (tp->sighand) {
2474                 unlock_task_sighand(tp->task, &tp->flags);
2475                 tp->sighand = NULL;
2476         }
2477
2478         if (tp->task) {
2479                 put_task_struct(tp->task);
2480                 tp->task = NULL;
2481         }
2482 }
2483
2484 static int show_timer(struct seq_file *m, void *v)
2485 {
2486         struct k_itimer *timer;
2487         struct timers_private *tp = m->private;
2488         int notify;
2489         static const char * const nstr[] = {
2490                 [SIGEV_SIGNAL] = "signal",
2491                 [SIGEV_NONE] = "none",
2492                 [SIGEV_THREAD] = "thread",
2493         };
2494
2495         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2496         notify = timer->it_sigev_notify;
2497
2498         seq_printf(m, "ID: %d\n", timer->it_id);
2499         seq_printf(m, "signal: %d/%px\n",
2500                    timer->sigq->info.si_signo,
2501                    timer->sigq->info.si_value.sival_ptr);
2502         seq_printf(m, "notify: %s/%s.%d\n",
2503                    nstr[notify & ~SIGEV_THREAD_ID],
2504                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2505                    pid_nr_ns(timer->it_pid, tp->ns));
2506         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2507
2508         return 0;
2509 }
2510
2511 static const struct seq_operations proc_timers_seq_ops = {
2512         .start  = timers_start,
2513         .next   = timers_next,
2514         .stop   = timers_stop,
2515         .show   = show_timer,
2516 };
2517
2518 static int proc_timers_open(struct inode *inode, struct file *file)
2519 {
2520         struct timers_private *tp;
2521
2522         tp = __seq_open_private(file, &proc_timers_seq_ops,
2523                         sizeof(struct timers_private));
2524         if (!tp)
2525                 return -ENOMEM;
2526
2527         tp->pid = proc_pid(inode);
2528         tp->ns = proc_pid_ns(inode->i_sb);
2529         return 0;
2530 }
2531
2532 static const struct file_operations proc_timers_operations = {
2533         .open           = proc_timers_open,
2534         .read           = seq_read,
2535         .llseek         = seq_lseek,
2536         .release        = seq_release_private,
2537 };
2538 #endif
2539
2540 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2541                                         size_t count, loff_t *offset)
2542 {
2543         struct inode *inode = file_inode(file);
2544         struct task_struct *p;
2545         u64 slack_ns;
2546         int err;
2547
2548         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2549         if (err < 0)
2550                 return err;
2551
2552         p = get_proc_task(inode);
2553         if (!p)
2554                 return -ESRCH;
2555
2556         if (p != current) {
2557                 rcu_read_lock();
2558                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2559                         rcu_read_unlock();
2560                         count = -EPERM;
2561                         goto out;
2562                 }
2563                 rcu_read_unlock();
2564
2565                 err = security_task_setscheduler(p);
2566                 if (err) {
2567                         count = err;
2568                         goto out;
2569                 }
2570         }
2571
2572         task_lock(p);
2573         if (slack_ns == 0)
2574                 p->timer_slack_ns = p->default_timer_slack_ns;
2575         else
2576                 p->timer_slack_ns = slack_ns;
2577         task_unlock(p);
2578
2579 out:
2580         put_task_struct(p);
2581
2582         return count;
2583 }
2584
2585 static int timerslack_ns_show(struct seq_file *m, void *v)
2586 {
2587         struct inode *inode = m->private;
2588         struct task_struct *p;
2589         int err = 0;
2590
2591         p = get_proc_task(inode);
2592         if (!p)
2593                 return -ESRCH;
2594
2595         if (p != current) {
2596                 rcu_read_lock();
2597                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2598                         rcu_read_unlock();
2599                         err = -EPERM;
2600                         goto out;
2601                 }
2602                 rcu_read_unlock();
2603
2604                 err = security_task_getscheduler(p);
2605                 if (err)
2606                         goto out;
2607         }
2608
2609         task_lock(p);
2610         seq_printf(m, "%llu\n", p->timer_slack_ns);
2611         task_unlock(p);
2612
2613 out:
2614         put_task_struct(p);
2615
2616         return err;
2617 }
2618
2619 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2620 {
2621         return single_open(filp, timerslack_ns_show, inode);
2622 }
2623
2624 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2625         .open           = timerslack_ns_open,
2626         .read           = seq_read,
2627         .write          = timerslack_ns_write,
2628         .llseek         = seq_lseek,
2629         .release        = single_release,
2630 };
2631
2632 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2633         struct task_struct *task, const void *ptr)
2634 {
2635         const struct pid_entry *p = ptr;
2636         struct inode *inode;
2637         struct proc_inode *ei;
2638
2639         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2640         if (!inode)
2641                 return ERR_PTR(-ENOENT);
2642
2643         ei = PROC_I(inode);
2644         if (S_ISDIR(inode->i_mode))
2645                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2646         if (p->iop)
2647                 inode->i_op = p->iop;
2648         if (p->fop)
2649                 inode->i_fop = p->fop;
2650         ei->op = p->op;
2651         pid_update_inode(task, inode);
2652         d_set_d_op(dentry, &pid_dentry_operations);
2653         return d_splice_alias(inode, dentry);
2654 }
2655
2656 static struct dentry *proc_pident_lookup(struct inode *dir,
2657                                          struct dentry *dentry,
2658                                          const struct pid_entry *p,
2659                                          const struct pid_entry *end)
2660 {
2661         struct task_struct *task = get_proc_task(dir);
2662         struct dentry *res = ERR_PTR(-ENOENT);
2663
2664         if (!task)
2665                 goto out_no_task;
2666
2667         /*
2668          * Yes, it does not scale. And it should not. Don't add
2669          * new entries into /proc/<tgid>/ without very good reasons.
2670          */
2671         for (; p < end; p++) {
2672                 if (p->len != dentry->d_name.len)
2673                         continue;
2674                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2675                         res = proc_pident_instantiate(dentry, task, p);
2676                         break;
2677                 }
2678         }
2679         put_task_struct(task);
2680 out_no_task:
2681         return res;
2682 }
2683
2684 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2685                 const struct pid_entry *ents, unsigned int nents)
2686 {
2687         struct task_struct *task = get_proc_task(file_inode(file));
2688         const struct pid_entry *p;
2689
2690         if (!task)
2691                 return -ENOENT;
2692
2693         if (!dir_emit_dots(file, ctx))
2694                 goto out;
2695
2696         if (ctx->pos >= nents + 2)
2697                 goto out;
2698
2699         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2700                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2701                                 proc_pident_instantiate, task, p))
2702                         break;
2703                 ctx->pos++;
2704         }
2705 out:
2706         put_task_struct(task);
2707         return 0;
2708 }
2709
2710 #ifdef CONFIG_SECURITY
2711 static int proc_pid_attr_open(struct inode *inode, struct file *file)
2712 {
2713         file->private_data = NULL;
2714         __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2715         return 0;
2716 }
2717
2718 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2719                                   size_t count, loff_t *ppos)
2720 {
2721         struct inode * inode = file_inode(file);
2722         char *p = NULL;
2723         ssize_t length;
2724         struct task_struct *task = get_proc_task(inode);
2725
2726         if (!task)
2727                 return -ESRCH;
2728
2729         length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2730                                       file->f_path.dentry->d_name.name,
2731                                       &p);
2732         put_task_struct(task);
2733         if (length > 0)
2734                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2735         kfree(p);
2736         return length;
2737 }
2738
2739 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2740                                    size_t count, loff_t *ppos)
2741 {
2742         struct inode * inode = file_inode(file);
2743         struct task_struct *task;
2744         void *page;
2745         int rv;
2746
2747         /* A task may only write when it was the opener. */
2748         if (file->private_data != current->mm)
2749                 return -EPERM;
2750
2751         rcu_read_lock();
2752         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2753         if (!task) {
2754                 rcu_read_unlock();
2755                 return -ESRCH;
2756         }
2757         /* A task may only write its own attributes. */
2758         if (current != task) {
2759                 rcu_read_unlock();
2760                 return -EACCES;
2761         }
2762         /* Prevent changes to overridden credentials. */
2763         if (current_cred() != current_real_cred()) {
2764                 rcu_read_unlock();
2765                 return -EBUSY;
2766         }
2767         rcu_read_unlock();
2768
2769         if (count > PAGE_SIZE)
2770                 count = PAGE_SIZE;
2771
2772         /* No partial writes. */
2773         if (*ppos != 0)
2774                 return -EINVAL;
2775
2776         page = memdup_user(buf, count);
2777         if (IS_ERR(page)) {
2778                 rv = PTR_ERR(page);
2779                 goto out;
2780         }
2781
2782         /* Guard against adverse ptrace interaction */
2783         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2784         if (rv < 0)
2785                 goto out_free;
2786
2787         rv = security_setprocattr(PROC_I(inode)->op.lsm,
2788                                   file->f_path.dentry->d_name.name, page,
2789                                   count);
2790         mutex_unlock(&current->signal->cred_guard_mutex);
2791 out_free:
2792         kfree(page);
2793 out:
2794         return rv;
2795 }
2796
2797 static const struct file_operations proc_pid_attr_operations = {
2798         .open           = proc_pid_attr_open,
2799         .read           = proc_pid_attr_read,
2800         .write          = proc_pid_attr_write,
2801         .llseek         = generic_file_llseek,
2802         .release        = mem_release,
2803 };
2804
2805 #define LSM_DIR_OPS(LSM) \
2806 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2807                              struct dir_context *ctx) \
2808 { \
2809         return proc_pident_readdir(filp, ctx, \
2810                                    LSM##_attr_dir_stuff, \
2811                                    ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2812 } \
2813 \
2814 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2815         .read           = generic_read_dir, \
2816         .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2817         .llseek         = default_llseek, \
2818 }; \
2819 \
2820 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2821                                 struct dentry *dentry, unsigned int flags) \
2822 { \
2823         return proc_pident_lookup(dir, dentry, \
2824                                   LSM##_attr_dir_stuff, \
2825                                   LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2826 } \
2827 \
2828 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2829         .lookup         = proc_##LSM##_attr_dir_lookup, \
2830         .getattr        = pid_getattr, \
2831         .setattr        = proc_setattr, \
2832 }
2833
2834 #ifdef CONFIG_SECURITY_SMACK
2835 static const struct pid_entry smack_attr_dir_stuff[] = {
2836         ATTR("smack", "current",        0666),
2837 };
2838 LSM_DIR_OPS(smack);
2839 #endif
2840
2841 #ifdef CONFIG_SECURITY_APPARMOR
2842 static const struct pid_entry apparmor_attr_dir_stuff[] = {
2843         ATTR("apparmor", "current",     0666),
2844         ATTR("apparmor", "prev",        0444),
2845         ATTR("apparmor", "exec",        0666),
2846 };
2847 LSM_DIR_OPS(apparmor);
2848 #endif
2849
2850 static const struct pid_entry attr_dir_stuff[] = {
2851         ATTR(NULL, "current",           0666),
2852         ATTR(NULL, "prev",              0444),
2853         ATTR(NULL, "exec",              0666),
2854         ATTR(NULL, "fscreate",          0666),
2855         ATTR(NULL, "keycreate",         0666),
2856         ATTR(NULL, "sockcreate",        0666),
2857 #ifdef CONFIG_SECURITY_SMACK
2858         DIR("smack",                    0555,
2859             proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2860 #endif
2861 #ifdef CONFIG_SECURITY_APPARMOR
2862         DIR("apparmor",                 0555,
2863             proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2864 #endif
2865 };
2866
2867 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2868 {
2869         return proc_pident_readdir(file, ctx,
2870                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2871 }
2872
2873 static const struct file_operations proc_attr_dir_operations = {
2874         .read           = generic_read_dir,
2875         .iterate_shared = proc_attr_dir_readdir,
2876         .llseek         = generic_file_llseek,
2877 };
2878
2879 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2880                                 struct dentry *dentry, unsigned int flags)
2881 {
2882         return proc_pident_lookup(dir, dentry,
2883                                   attr_dir_stuff,
2884                                   attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2885 }
2886
2887 static const struct inode_operations proc_attr_dir_inode_operations = {
2888         .lookup         = proc_attr_dir_lookup,
2889         .getattr        = pid_getattr,
2890         .setattr        = proc_setattr,
2891 };
2892
2893 #endif
2894
2895 #ifdef CONFIG_ELF_CORE
2896 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2897                                          size_t count, loff_t *ppos)
2898 {
2899         struct task_struct *task = get_proc_task(file_inode(file));
2900         struct mm_struct *mm;
2901         char buffer[PROC_NUMBUF];
2902         size_t len;
2903         int ret;
2904
2905         if (!task)
2906                 return -ESRCH;
2907
2908         ret = 0;
2909         mm = get_task_mm(task);
2910         if (mm) {
2911                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2912                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2913                                 MMF_DUMP_FILTER_SHIFT));
2914                 mmput(mm);
2915                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2916         }
2917
2918         put_task_struct(task);
2919
2920         return ret;
2921 }
2922
2923 static ssize_t proc_coredump_filter_write(struct file *file,
2924                                           const char __user *buf,
2925                                           size_t count,
2926                                           loff_t *ppos)
2927 {
2928         struct task_struct *task;
2929         struct mm_struct *mm;
2930         unsigned int val;
2931         int ret;
2932         int i;
2933         unsigned long mask;
2934
2935         ret = kstrtouint_from_user(buf, count, 0, &val);
2936         if (ret < 0)
2937                 return ret;
2938
2939         ret = -ESRCH;
2940         task = get_proc_task(file_inode(file));
2941         if (!task)
2942                 goto out_no_task;
2943
2944         mm = get_task_mm(task);
2945         if (!mm)
2946                 goto out_no_mm;
2947         ret = 0;
2948
2949         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2950                 if (val & mask)
2951                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2952                 else
2953                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2954         }
2955
2956         mmput(mm);
2957  out_no_mm:
2958         put_task_struct(task);
2959  out_no_task:
2960         if (ret < 0)
2961                 return ret;
2962         return count;
2963 }
2964
2965 static const struct file_operations proc_coredump_filter_operations = {
2966         .read           = proc_coredump_filter_read,
2967         .write          = proc_coredump_filter_write,
2968         .llseek         = generic_file_llseek,
2969 };
2970 #endif
2971
2972 #ifdef CONFIG_TASK_IO_ACCOUNTING
2973 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2974 {
2975         struct task_io_accounting acct;
2976         int result;
2977
2978         result = down_read_killable(&task->signal->exec_update_lock);
2979         if (result)
2980                 return result;
2981
2982         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2983                 result = -EACCES;
2984                 goto out_unlock;
2985         }
2986
2987         if (whole) {
2988                 struct signal_struct *sig = task->signal;
2989                 struct task_struct *t;
2990                 unsigned int seq = 1;
2991                 unsigned long flags;
2992
2993                 rcu_read_lock();
2994                 do {
2995                         seq++; /* 2 on the 1st/lockless path, otherwise odd */
2996                         flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
2997
2998                         acct = sig->ioac;
2999                         __for_each_thread(sig, t)
3000                                 task_io_accounting_add(&acct, &t->ioac);
3001
3002                 } while (need_seqretry(&sig->stats_lock, seq));
3003                 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
3004                 rcu_read_unlock();
3005         } else {
3006                 acct = task->ioac;
3007         }
3008
3009         seq_printf(m,
3010                    "rchar: %llu\n"
3011                    "wchar: %llu\n"
3012                    "syscr: %llu\n"
3013                    "syscw: %llu\n"
3014                    "read_bytes: %llu\n"
3015                    "write_bytes: %llu\n"
3016                    "cancelled_write_bytes: %llu\n",
3017                    (unsigned long long)acct.rchar,
3018                    (unsigned long long)acct.wchar,
3019                    (unsigned long long)acct.syscr,
3020                    (unsigned long long)acct.syscw,
3021                    (unsigned long long)acct.read_bytes,
3022                    (unsigned long long)acct.write_bytes,
3023                    (unsigned long long)acct.cancelled_write_bytes);
3024         result = 0;
3025
3026 out_unlock:
3027         up_read(&task->signal->exec_update_lock);
3028         return result;
3029 }
3030
3031 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3032                                   struct pid *pid, struct task_struct *task)
3033 {
3034         return do_io_accounting(task, m, 0);
3035 }
3036
3037 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3038                                    struct pid *pid, struct task_struct *task)
3039 {
3040         return do_io_accounting(task, m, 1);
3041 }
3042 #endif /* CONFIG_TASK_IO_ACCOUNTING */
3043
3044 #ifdef CONFIG_USER_NS
3045 static int proc_id_map_open(struct inode *inode, struct file *file,
3046         const struct seq_operations *seq_ops)
3047 {
3048         struct user_namespace *ns = NULL;
3049         struct task_struct *task;
3050         struct seq_file *seq;
3051         int ret = -EINVAL;
3052
3053         task = get_proc_task(inode);
3054         if (task) {
3055                 rcu_read_lock();
3056                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3057                 rcu_read_unlock();
3058                 put_task_struct(task);
3059         }
3060         if (!ns)
3061                 goto err;
3062
3063         ret = seq_open(file, seq_ops);
3064         if (ret)
3065                 goto err_put_ns;
3066
3067         seq = file->private_data;
3068         seq->private = ns;
3069
3070         return 0;
3071 err_put_ns:
3072         put_user_ns(ns);
3073 err:
3074         return ret;
3075 }
3076
3077 static int proc_id_map_release(struct inode *inode, struct file *file)
3078 {
3079         struct seq_file *seq = file->private_data;
3080         struct user_namespace *ns = seq->private;
3081         put_user_ns(ns);
3082         return seq_release(inode, file);
3083 }
3084
3085 static int proc_uid_map_open(struct inode *inode, struct file *file)
3086 {
3087         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3088 }
3089
3090 static int proc_gid_map_open(struct inode *inode, struct file *file)
3091 {
3092         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3093 }
3094
3095 static int proc_projid_map_open(struct inode *inode, struct file *file)
3096 {
3097         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3098 }
3099
3100 static const struct file_operations proc_uid_map_operations = {
3101         .open           = proc_uid_map_open,
3102         .write          = proc_uid_map_write,
3103         .read           = seq_read,
3104         .llseek         = seq_lseek,
3105         .release        = proc_id_map_release,
3106 };
3107
3108 static const struct file_operations proc_gid_map_operations = {
3109         .open           = proc_gid_map_open,
3110         .write          = proc_gid_map_write,
3111         .read           = seq_read,
3112         .llseek         = seq_lseek,
3113         .release        = proc_id_map_release,
3114 };
3115
3116 static const struct file_operations proc_projid_map_operations = {
3117         .open           = proc_projid_map_open,
3118         .write          = proc_projid_map_write,
3119         .read           = seq_read,
3120         .llseek         = seq_lseek,
3121         .release        = proc_id_map_release,
3122 };
3123
3124 static int proc_setgroups_open(struct inode *inode, struct file *file)
3125 {
3126         struct user_namespace *ns = NULL;
3127         struct task_struct *task;
3128         int ret;
3129
3130         ret = -ESRCH;
3131         task = get_proc_task(inode);
3132         if (task) {
3133                 rcu_read_lock();
3134                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3135                 rcu_read_unlock();
3136                 put_task_struct(task);
3137         }
3138         if (!ns)
3139                 goto err;
3140
3141         if (file->f_mode & FMODE_WRITE) {
3142                 ret = -EACCES;
3143                 if (!ns_capable(ns, CAP_SYS_ADMIN))
3144                         goto err_put_ns;
3145         }
3146
3147         ret = single_open(file, &proc_setgroups_show, ns);
3148         if (ret)
3149                 goto err_put_ns;
3150
3151         return 0;
3152 err_put_ns:
3153         put_user_ns(ns);
3154 err:
3155         return ret;
3156 }
3157
3158 static int proc_setgroups_release(struct inode *inode, struct file *file)
3159 {
3160         struct seq_file *seq = file->private_data;
3161         struct user_namespace *ns = seq->private;
3162         int ret = single_release(inode, file);
3163         put_user_ns(ns);
3164         return ret;
3165 }
3166
3167 static const struct file_operations proc_setgroups_operations = {
3168         .open           = proc_setgroups_open,
3169         .write          = proc_setgroups_write,
3170         .read           = seq_read,
3171         .llseek         = seq_lseek,
3172         .release        = proc_setgroups_release,
3173 };
3174 #endif /* CONFIG_USER_NS */
3175
3176 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3177                                 struct pid *pid, struct task_struct *task)
3178 {
3179         int err = lock_trace(task);
3180         if (!err) {
3181                 seq_printf(m, "%08x\n", task->personality);
3182                 unlock_trace(task);
3183         }
3184         return err;
3185 }
3186
3187 #ifdef CONFIG_LIVEPATCH
3188 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3189                                 struct pid *pid, struct task_struct *task)
3190 {
3191         seq_printf(m, "%d\n", task->patch_state);
3192         return 0;
3193 }
3194 #endif /* CONFIG_LIVEPATCH */
3195
3196 #ifdef CONFIG_KSM
3197 static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3198                                 struct pid *pid, struct task_struct *task)
3199 {
3200         struct mm_struct *mm;
3201
3202         mm = get_task_mm(task);
3203         if (mm) {
3204                 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3205                 mmput(mm);
3206         }
3207
3208         return 0;
3209 }
3210 static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3211                                 struct pid *pid, struct task_struct *task)
3212 {
3213         struct mm_struct *mm;
3214
3215         mm = get_task_mm(task);
3216         if (mm) {
3217                 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3218                 seq_printf(m, "ksm_zero_pages %lu\n", mm->ksm_zero_pages);
3219                 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3220                 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3221                 mmput(mm);
3222         }
3223
3224         return 0;
3225 }
3226 #endif /* CONFIG_KSM */
3227
3228 #ifdef CONFIG_STACKLEAK_METRICS
3229 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3230                                 struct pid *pid, struct task_struct *task)
3231 {
3232         unsigned long prev_depth = THREAD_SIZE -
3233                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3234         unsigned long depth = THREAD_SIZE -
3235                                 (task->lowest_stack & (THREAD_SIZE - 1));
3236
3237         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3238                                                         prev_depth, depth);
3239         return 0;
3240 }
3241 #endif /* CONFIG_STACKLEAK_METRICS */
3242
3243 /*
3244  * Thread groups
3245  */
3246 static const struct file_operations proc_task_operations;
3247 static const struct inode_operations proc_task_inode_operations;
3248
3249 static const struct pid_entry tgid_base_stuff[] = {
3250         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3251         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3252         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3253         DIR("fdinfo",     S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3254         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3255 #ifdef CONFIG_NET
3256         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3257 #endif
3258         REG("environ",    S_IRUSR, proc_environ_operations),
3259         REG("auxv",       S_IRUSR, proc_auxv_operations),
3260         ONE("status",     S_IRUGO, proc_pid_status),
3261         ONE("personality", S_IRUSR, proc_pid_personality),
3262         ONE("limits",     S_IRUGO, proc_pid_limits),
3263 #ifdef CONFIG_SCHED_DEBUG
3264         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3265 #endif
3266 #ifdef CONFIG_SCHED_AUTOGROUP
3267         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3268 #endif
3269 #ifdef CONFIG_TIME_NS
3270         REG("timens_offsets",  S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3271 #endif
3272         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3273 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3274         ONE("syscall",    S_IRUSR, proc_pid_syscall),
3275 #endif
3276         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3277         ONE("stat",       S_IRUGO, proc_tgid_stat),
3278         ONE("statm",      S_IRUGO, proc_pid_statm),
3279         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3280 #ifdef CONFIG_NUMA
3281         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3282 #endif
3283         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3284         LNK("cwd",        proc_cwd_link),
3285         LNK("root",       proc_root_link),
3286         LNK("exe",        proc_exe_link),
3287         REG("mounts",     S_IRUGO, proc_mounts_operations),
3288         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3289         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3290 #ifdef CONFIG_PROC_PAGE_MONITOR
3291         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3292         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3293         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3294         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3295 #endif
3296 #ifdef CONFIG_SECURITY
3297         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3298 #endif
3299 #ifdef CONFIG_KALLSYMS
3300         ONE("wchan",      S_IRUGO, proc_pid_wchan),
3301 #endif
3302 #ifdef CONFIG_STACKTRACE
3303         ONE("stack",      S_IRUSR, proc_pid_stack),
3304 #endif
3305 #ifdef CONFIG_SCHED_INFO
3306         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3307 #endif
3308 #ifdef CONFIG_LATENCYTOP
3309         REG("latency",  S_IRUGO, proc_lstats_operations),
3310 #endif
3311 #ifdef CONFIG_PROC_PID_CPUSET
3312         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3313 #endif
3314 #ifdef CONFIG_CGROUPS
3315         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3316 #endif
3317 #ifdef CONFIG_PROC_CPU_RESCTRL
3318         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3319 #endif
3320         ONE("oom_score",  S_IRUGO, proc_oom_score),
3321         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3322         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3323 #ifdef CONFIG_AUDIT
3324         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3325         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3326 #endif
3327 #ifdef CONFIG_FAULT_INJECTION
3328         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3329         REG("fail-nth", 0644, proc_fail_nth_operations),
3330 #endif
3331 #ifdef CONFIG_ELF_CORE
3332         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3333 #endif
3334 #ifdef CONFIG_TASK_IO_ACCOUNTING
3335         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3336 #endif
3337 #ifdef CONFIG_USER_NS
3338         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3339         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3340         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3341         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3342 #endif
3343 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3344         REG("timers",     S_IRUGO, proc_timers_operations),
3345 #endif
3346         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3347 #ifdef CONFIG_LIVEPATCH
3348         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3349 #endif
3350 #ifdef CONFIG_STACKLEAK_METRICS
3351         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3352 #endif
3353 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3354         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3355 #endif
3356 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3357         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3358 #endif
3359 #ifdef CONFIG_KSM
3360         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3361         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3362 #endif
3363 };
3364
3365 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3366 {
3367         return proc_pident_readdir(file, ctx,
3368                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3369 }
3370
3371 static const struct file_operations proc_tgid_base_operations = {
3372         .read           = generic_read_dir,
3373         .iterate_shared = proc_tgid_base_readdir,
3374         .llseek         = generic_file_llseek,
3375 };
3376
3377 struct pid *tgid_pidfd_to_pid(const struct file *file)
3378 {
3379         if (file->f_op != &proc_tgid_base_operations)
3380                 return ERR_PTR(-EBADF);
3381
3382         return proc_pid(file_inode(file));
3383 }
3384
3385 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3386 {
3387         return proc_pident_lookup(dir, dentry,
3388                                   tgid_base_stuff,
3389                                   tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3390 }
3391
3392 static const struct inode_operations proc_tgid_base_inode_operations = {
3393         .lookup         = proc_tgid_base_lookup,
3394         .getattr        = pid_getattr,
3395         .setattr        = proc_setattr,
3396         .permission     = proc_pid_permission,
3397 };
3398
3399 /**
3400  * proc_flush_pid -  Remove dcache entries for @pid from the /proc dcache.
3401  * @pid: pid that should be flushed.
3402  *
3403  * This function walks a list of inodes (that belong to any proc
3404  * filesystem) that are attached to the pid and flushes them from
3405  * the dentry cache.
3406  *
3407  * It is safe and reasonable to cache /proc entries for a task until
3408  * that task exits.  After that they just clog up the dcache with
3409  * useless entries, possibly causing useful dcache entries to be
3410  * flushed instead.  This routine is provided to flush those useless
3411  * dcache entries when a process is reaped.
3412  *
3413  * NOTE: This routine is just an optimization so it does not guarantee
3414  *       that no dcache entries will exist after a process is reaped
3415  *       it just makes it very unlikely that any will persist.
3416  */
3417
3418 void proc_flush_pid(struct pid *pid)
3419 {
3420         proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3421 }
3422
3423 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3424                                    struct task_struct *task, const void *ptr)
3425 {
3426         struct inode *inode;
3427
3428         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3429                                          S_IFDIR | S_IRUGO | S_IXUGO);
3430         if (!inode)
3431                 return ERR_PTR(-ENOENT);
3432
3433         inode->i_op = &proc_tgid_base_inode_operations;
3434         inode->i_fop = &proc_tgid_base_operations;
3435         inode->i_flags|=S_IMMUTABLE;
3436
3437         set_nlink(inode, nlink_tgid);
3438         pid_update_inode(task, inode);
3439
3440         d_set_d_op(dentry, &pid_dentry_operations);
3441         return d_splice_alias(inode, dentry);
3442 }
3443
3444 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3445 {
3446         struct task_struct *task;
3447         unsigned tgid;
3448         struct proc_fs_info *fs_info;
3449         struct pid_namespace *ns;
3450         struct dentry *result = ERR_PTR(-ENOENT);
3451
3452         tgid = name_to_int(&dentry->d_name);
3453         if (tgid == ~0U)
3454                 goto out;
3455
3456         fs_info = proc_sb_info(dentry->d_sb);
3457         ns = fs_info->pid_ns;
3458         rcu_read_lock();
3459         task = find_task_by_pid_ns(tgid, ns);
3460         if (task)
3461                 get_task_struct(task);
3462         rcu_read_unlock();
3463         if (!task)
3464                 goto out;
3465
3466         /* Limit procfs to only ptraceable tasks */
3467         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3468                 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3469                         goto out_put_task;
3470         }
3471
3472         result = proc_pid_instantiate(dentry, task, NULL);
3473 out_put_task:
3474         put_task_struct(task);
3475 out:
3476         return result;
3477 }
3478
3479 /*
3480  * Find the first task with tgid >= tgid
3481  *
3482  */
3483 struct tgid_iter {
3484         unsigned int tgid;
3485         struct task_struct *task;
3486 };
3487 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3488 {
3489         struct pid *pid;
3490
3491         if (iter.task)
3492                 put_task_struct(iter.task);
3493         rcu_read_lock();
3494 retry:
3495         iter.task = NULL;
3496         pid = find_ge_pid(iter.tgid, ns);
3497         if (pid) {
3498                 iter.tgid = pid_nr_ns(pid, ns);
3499                 iter.task = pid_task(pid, PIDTYPE_TGID);
3500                 if (!iter.task) {
3501                         iter.tgid += 1;
3502                         goto retry;
3503                 }
3504                 get_task_struct(iter.task);
3505         }
3506         rcu_read_unlock();
3507         return iter;
3508 }
3509
3510 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3511
3512 /* for the /proc/ directory itself, after non-process stuff has been done */
3513 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3514 {
3515         struct tgid_iter iter;
3516         struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3517         struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3518         loff_t pos = ctx->pos;
3519
3520         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3521                 return 0;
3522
3523         if (pos == TGID_OFFSET - 2) {
3524                 struct inode *inode = d_inode(fs_info->proc_self);
3525                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3526                         return 0;
3527                 ctx->pos = pos = pos + 1;
3528         }
3529         if (pos == TGID_OFFSET - 1) {
3530                 struct inode *inode = d_inode(fs_info->proc_thread_self);
3531                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3532                         return 0;
3533                 ctx->pos = pos = pos + 1;
3534         }
3535         iter.tgid = pos - TGID_OFFSET;
3536         iter.task = NULL;
3537         for (iter = next_tgid(ns, iter);
3538              iter.task;
3539              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3540                 char name[10 + 1];
3541                 unsigned int len;
3542
3543                 cond_resched();
3544                 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3545                         continue;
3546
3547                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3548                 ctx->pos = iter.tgid + TGID_OFFSET;
3549                 if (!proc_fill_cache(file, ctx, name, len,
3550                                      proc_pid_instantiate, iter.task, NULL)) {
3551                         put_task_struct(iter.task);
3552                         return 0;
3553                 }
3554         }
3555         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3556         return 0;
3557 }
3558
3559 /*
3560  * proc_tid_comm_permission is a special permission function exclusively
3561  * used for the node /proc/<pid>/task/<tid>/comm.
3562  * It bypasses generic permission checks in the case where a task of the same
3563  * task group attempts to access the node.
3564  * The rationale behind this is that glibc and bionic access this node for
3565  * cross thread naming (pthread_set/getname_np(!self)). However, if
3566  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3567  * which locks out the cross thread naming implementation.
3568  * This function makes sure that the node is always accessible for members of
3569  * same thread group.
3570  */
3571 static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3572                                     struct inode *inode, int mask)
3573 {
3574         bool is_same_tgroup;
3575         struct task_struct *task;
3576
3577         task = get_proc_task(inode);
3578         if (!task)
3579                 return -ESRCH;
3580         is_same_tgroup = same_thread_group(current, task);
3581         put_task_struct(task);
3582
3583         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3584                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3585                  * read or written by the members of the corresponding
3586                  * thread group.
3587                  */
3588                 return 0;
3589         }
3590
3591         return generic_permission(&nop_mnt_idmap, inode, mask);
3592 }
3593
3594 static const struct inode_operations proc_tid_comm_inode_operations = {
3595                 .setattr        = proc_setattr,
3596                 .permission     = proc_tid_comm_permission,
3597 };
3598
3599 /*
3600  * Tasks
3601  */
3602 static const struct pid_entry tid_base_stuff[] = {
3603         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3604         DIR("fdinfo",    S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3605         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3606 #ifdef CONFIG_NET
3607         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3608 #endif
3609         REG("environ",   S_IRUSR, proc_environ_operations),
3610         REG("auxv",      S_IRUSR, proc_auxv_operations),
3611         ONE("status",    S_IRUGO, proc_pid_status),
3612         ONE("personality", S_IRUSR, proc_pid_personality),
3613         ONE("limits",    S_IRUGO, proc_pid_limits),
3614 #ifdef CONFIG_SCHED_DEBUG
3615         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3616 #endif
3617         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3618                          &proc_tid_comm_inode_operations,
3619                          &proc_pid_set_comm_operations, {}),
3620 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3621         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3622 #endif
3623         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3624         ONE("stat",      S_IRUGO, proc_tid_stat),
3625         ONE("statm",     S_IRUGO, proc_pid_statm),
3626         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3627 #ifdef CONFIG_PROC_CHILDREN
3628         REG("children",  S_IRUGO, proc_tid_children_operations),
3629 #endif
3630 #ifdef CONFIG_NUMA
3631         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3632 #endif
3633         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3634         LNK("cwd",       proc_cwd_link),
3635         LNK("root",      proc_root_link),
3636         LNK("exe",       proc_exe_link),
3637         REG("mounts",    S_IRUGO, proc_mounts_operations),
3638         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3639 #ifdef CONFIG_PROC_PAGE_MONITOR
3640         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3641         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3642         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3643         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3644 #endif
3645 #ifdef CONFIG_SECURITY
3646         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3647 #endif
3648 #ifdef CONFIG_KALLSYMS
3649         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3650 #endif
3651 #ifdef CONFIG_STACKTRACE
3652         ONE("stack",      S_IRUSR, proc_pid_stack),
3653 #endif
3654 #ifdef CONFIG_SCHED_INFO
3655         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3656 #endif
3657 #ifdef CONFIG_LATENCYTOP
3658         REG("latency",  S_IRUGO, proc_lstats_operations),
3659 #endif
3660 #ifdef CONFIG_PROC_PID_CPUSET
3661         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3662 #endif
3663 #ifdef CONFIG_CGROUPS
3664         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3665 #endif
3666 #ifdef CONFIG_PROC_CPU_RESCTRL
3667         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3668 #endif
3669         ONE("oom_score", S_IRUGO, proc_oom_score),
3670         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3671         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3672 #ifdef CONFIG_AUDIT
3673         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3674         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3675 #endif
3676 #ifdef CONFIG_FAULT_INJECTION
3677         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3678         REG("fail-nth", 0644, proc_fail_nth_operations),
3679 #endif
3680 #ifdef CONFIG_TASK_IO_ACCOUNTING
3681         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3682 #endif
3683 #ifdef CONFIG_USER_NS
3684         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3685         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3686         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3687         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3688 #endif
3689 #ifdef CONFIG_LIVEPATCH
3690         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3691 #endif
3692 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3693         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3694 #endif
3695 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3696         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3697 #endif
3698 #ifdef CONFIG_KSM
3699         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3700         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3701 #endif
3702 };
3703
3704 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3705 {
3706         return proc_pident_readdir(file, ctx,
3707                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3708 }
3709
3710 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3711 {
3712         return proc_pident_lookup(dir, dentry,
3713                                   tid_base_stuff,
3714                                   tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3715 }
3716
3717 static const struct file_operations proc_tid_base_operations = {
3718         .read           = generic_read_dir,
3719         .iterate_shared = proc_tid_base_readdir,
3720         .llseek         = generic_file_llseek,
3721 };
3722
3723 static const struct inode_operations proc_tid_base_inode_operations = {
3724         .lookup         = proc_tid_base_lookup,
3725         .getattr        = pid_getattr,
3726         .setattr        = proc_setattr,
3727 };
3728
3729 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3730         struct task_struct *task, const void *ptr)
3731 {
3732         struct inode *inode;
3733         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3734                                          S_IFDIR | S_IRUGO | S_IXUGO);
3735         if (!inode)
3736                 return ERR_PTR(-ENOENT);
3737
3738         inode->i_op = &proc_tid_base_inode_operations;
3739         inode->i_fop = &proc_tid_base_operations;
3740         inode->i_flags |= S_IMMUTABLE;
3741
3742         set_nlink(inode, nlink_tid);
3743         pid_update_inode(task, inode);
3744
3745         d_set_d_op(dentry, &pid_dentry_operations);
3746         return d_splice_alias(inode, dentry);
3747 }
3748
3749 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3750 {
3751         struct task_struct *task;
3752         struct task_struct *leader = get_proc_task(dir);
3753         unsigned tid;
3754         struct proc_fs_info *fs_info;
3755         struct pid_namespace *ns;
3756         struct dentry *result = ERR_PTR(-ENOENT);
3757
3758         if (!leader)
3759                 goto out_no_task;
3760
3761         tid = name_to_int(&dentry->d_name);
3762         if (tid == ~0U)
3763                 goto out;
3764
3765         fs_info = proc_sb_info(dentry->d_sb);
3766         ns = fs_info->pid_ns;
3767         rcu_read_lock();
3768         task = find_task_by_pid_ns(tid, ns);
3769         if (task)
3770                 get_task_struct(task);
3771         rcu_read_unlock();
3772         if (!task)
3773                 goto out;
3774         if (!same_thread_group(leader, task))
3775                 goto out_drop_task;
3776
3777         result = proc_task_instantiate(dentry, task, NULL);
3778 out_drop_task:
3779         put_task_struct(task);
3780 out:
3781         put_task_struct(leader);
3782 out_no_task:
3783         return result;
3784 }
3785
3786 /*
3787  * Find the first tid of a thread group to return to user space.
3788  *
3789  * Usually this is just the thread group leader, but if the users
3790  * buffer was too small or there was a seek into the middle of the
3791  * directory we have more work todo.
3792  *
3793  * In the case of a short read we start with find_task_by_pid.
3794  *
3795  * In the case of a seek we start with the leader and walk nr
3796  * threads past it.
3797  */
3798 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3799                                         struct pid_namespace *ns)
3800 {
3801         struct task_struct *pos, *task;
3802         unsigned long nr = f_pos;
3803
3804         if (nr != f_pos)        /* 32bit overflow? */
3805                 return NULL;
3806
3807         rcu_read_lock();
3808         task = pid_task(pid, PIDTYPE_PID);
3809         if (!task)
3810                 goto fail;
3811
3812         /* Attempt to start with the tid of a thread */
3813         if (tid && nr) {
3814                 pos = find_task_by_pid_ns(tid, ns);
3815                 if (pos && same_thread_group(pos, task))
3816                         goto found;
3817         }
3818
3819         /* If nr exceeds the number of threads there is nothing todo */
3820         if (nr >= get_nr_threads(task))
3821                 goto fail;
3822
3823         /* If we haven't found our starting place yet start
3824          * with the leader and walk nr threads forward.
3825          */
3826         for_each_thread(task, pos) {
3827                 if (!nr--)
3828                         goto found;
3829         }
3830 fail:
3831         pos = NULL;
3832         goto out;
3833 found:
3834         get_task_struct(pos);
3835 out:
3836         rcu_read_unlock();
3837         return pos;
3838 }
3839
3840 /*
3841  * Find the next thread in the thread list.
3842  * Return NULL if there is an error or no next thread.
3843  *
3844  * The reference to the input task_struct is released.
3845  */
3846 static struct task_struct *next_tid(struct task_struct *start)
3847 {
3848         struct task_struct *pos = NULL;
3849         rcu_read_lock();
3850         if (pid_alive(start)) {
3851                 pos = __next_thread(start);
3852                 if (pos)
3853                         get_task_struct(pos);
3854         }
3855         rcu_read_unlock();
3856         put_task_struct(start);
3857         return pos;
3858 }
3859
3860 /* for the /proc/TGID/task/ directories */
3861 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3862 {
3863         struct inode *inode = file_inode(file);
3864         struct task_struct *task;
3865         struct pid_namespace *ns;
3866         int tid;
3867
3868         if (proc_inode_is_dead(inode))
3869                 return -ENOENT;
3870
3871         if (!dir_emit_dots(file, ctx))
3872                 return 0;
3873
3874         /* f_version caches the tgid value that the last readdir call couldn't
3875          * return. lseek aka telldir automagically resets f_version to 0.
3876          */
3877         ns = proc_pid_ns(inode->i_sb);
3878         tid = (int)file->f_version;
3879         file->f_version = 0;
3880         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3881              task;
3882              task = next_tid(task), ctx->pos++) {
3883                 char name[10 + 1];
3884                 unsigned int len;
3885
3886                 tid = task_pid_nr_ns(task, ns);
3887                 if (!tid)
3888                         continue;       /* The task has just exited. */
3889                 len = snprintf(name, sizeof(name), "%u", tid);
3890                 if (!proc_fill_cache(file, ctx, name, len,
3891                                 proc_task_instantiate, task, NULL)) {
3892                         /* returning this tgid failed, save it as the first
3893                          * pid for the next readir call */
3894                         file->f_version = (u64)tid;
3895                         put_task_struct(task);
3896                         break;
3897                 }
3898         }
3899
3900         return 0;
3901 }
3902
3903 static int proc_task_getattr(struct mnt_idmap *idmap,
3904                              const struct path *path, struct kstat *stat,
3905                              u32 request_mask, unsigned int query_flags)
3906 {
3907         struct inode *inode = d_inode(path->dentry);
3908         struct task_struct *p = get_proc_task(inode);
3909         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3910
3911         if (p) {
3912                 stat->nlink += get_nr_threads(p);
3913                 put_task_struct(p);
3914         }
3915
3916         return 0;
3917 }
3918
3919 static const struct inode_operations proc_task_inode_operations = {
3920         .lookup         = proc_task_lookup,
3921         .getattr        = proc_task_getattr,
3922         .setattr        = proc_setattr,
3923         .permission     = proc_pid_permission,
3924 };
3925
3926 static const struct file_operations proc_task_operations = {
3927         .read           = generic_read_dir,
3928         .iterate_shared = proc_task_readdir,
3929         .llseek         = generic_file_llseek,
3930 };
3931
3932 void __init set_proc_pid_nlink(void)
3933 {
3934         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3935         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3936 }