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