fs/proc/base.c

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