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