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