fs/proc/base.c

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