fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <[email protected]>
  21  *  Edjard Mota <[email protected]>
  22  *  Ilias Biris <[email protected]>
  23  *  Mauricio Lin <[email protected]>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <[email protected]>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <[email protected]>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/generic-radix-tree.h>
  63 #include <linux/string.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/namei.h>
  66 #include <linux/mnt_namespace.h>
  67 #include <linux/mm.h>
  68 #include <linux/swap.h>
  69 #include <linux/rcupdate.h>
  70 #include <linux/kallsyms.h>
  71 #include <linux/stacktrace.h>
  72 #include <linux/resource.h>
  73 #include <linux/module.h>
  74 #include <linux/mount.h>
  75 #include <linux/security.h>
  76 #include <linux/ptrace.h>
  77 #include <linux/tracehook.h>
  78 #include <linux/printk.h>
  79 #include <linux/cache.h>
  80 #include <linux/cgroup.h>
  81 #include <linux/cpuset.h>
  82 #include <linux/audit.h>
  83 #include <linux/poll.h>
  84 #include <linux/nsproxy.h>
  85 #include <linux/oom.h>
  86 #include <linux/elf.h>
  87 #include <linux/pid_namespace.h>
  88 #include <linux/user_namespace.h>
  89 #include <linux/fs_struct.h>
  90 #include <linux/slab.h>
  91 #include <linux/sched/autogroup.h>
  92 #include <linux/sched/mm.h>
  93 #include <linux/sched/coredump.h>
  94 #include <linux/sched/debug.h>
  95 #include <linux/sched/stat.h>
  96 #include <linux/posix-timers.h>
  97 #include <linux/time_namespace.h>
  98 #include <linux/resctrl.h>
  99 #include <trace/events/oom.h>
 100 #include "internal.h"
 101 #include "fd.h"
 102
 103 #include "../../lib/kstrtox.h"
 104
 105 /* NOTE:
 106  *      Implementing inode permission operations in /proc is almost
 107  *      certainly an error.  Permission checks need to happen during
 108  *      each system call not at open time.  The reason is that most of
 109  *      what we wish to check for permissions in /proc varies at runtime.
 110  *
 111  *      The classic example of a problem is opening file descriptors
 112  *      in /proc for a task before it execs a suid executable.
 113  */
 114
 115 static u8 nlink_tid __ro_after_init;
 116 static u8 nlink_tgid __ro_after_init;
 117
 118 struct pid_entry {
 119         const char *name;
 120         unsigned int len;
 121         umode_t mode;
 122         const struct inode_operations *iop;
 123         const struct file_operations *fop;
 124         union proc_op op;
 125 };
 126
 127 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 128         .name = (NAME),                                 \
 129         .len  = sizeof(NAME) - 1,                       \
 130         .mode = MODE,                                   \
 131         .iop  = IOP,                                    \
 132         .fop  = FOP,                                    \
 133         .op   = OP,                                     \
 134 }
 135
 136 #define DIR(NAME, MODE, iops, fops)     \
 137         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 138 #define LNK(NAME, get_link)                                     \
 139         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 140                 &proc_pid_link_inode_operations, NULL,          \
 141                 { .proc_get_link = get_link } )
 142 #define REG(NAME, MODE, fops)                           \
 143         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 144 #define ONE(NAME, MODE, show)                           \
 145         NOD(NAME, (S_IFREG|(MODE)),                     \
 146                 NULL, &proc_single_file_operations,     \
 147                 { .proc_show = show } )
 148 #define ATTR(LSM, NAME, MODE)                           \
 149         NOD(NAME, (S_IFREG|(MODE)),                     \
 150                 NULL, &proc_pid_attr_operations,        \
 151                 { .lsm = LSM })
 152
 153 /*
 154  * Count the number of hardlinks for the pid_entry table, excluding the .
 155  * and .. links.
 156  */
 157 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 158         unsigned int n)
 159 {
 160         unsigned int i;
 161         unsigned int count;
 162
 163         count = 2;
 164         for (i = 0; i < n; ++i) {
 165                 if (S_ISDIR(entries[i].mode))
 166                         ++count;
 167         }
 168
 169         return count;
 170 }
 171
 172 static int get_task_root(struct task_struct *task, struct path *root)
 173 {
 174         int result = -ENOENT;
 175
 176         task_lock(task);
 177         if (task->fs) {
 178                 get_fs_root(task->fs, root);
 179                 result = 0;
 180         }
 181         task_unlock(task);
 182         return result;
 183 }
 184
 185 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 186 {
 187         struct task_struct *task = get_proc_task(d_inode(dentry));
 188         int result = -ENOENT;
 189
 190         if (task) {
 191                 task_lock(task);
 192                 if (task->fs) {
 193                         get_fs_pwd(task->fs, path);
 194                         result = 0;
 195                 }
 196                 task_unlock(task);
 197                 put_task_struct(task);
 198         }
 199         return result;
 200 }
 201
 202 static int proc_root_link(struct dentry *dentry, struct path *path)
 203 {
 204         struct task_struct *task = get_proc_task(d_inode(dentry));
 205         int result = -ENOENT;
 206
 207         if (task) {
 208                 result = get_task_root(task, path);
 209                 put_task_struct(task);
 210         }
 211         return result;
 212 }
 213
 214 /*
 215  * If the user used setproctitle(), we just get the string from
 216  * user space at arg_start, and limit it to a maximum of one page.
 217  */
 218 static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
 219                                 size_t count, unsigned long pos,
 220                                 unsigned long arg_start)
 221 {
 222         char *page;
 223         int ret, got;
 224
 225         if (pos >= PAGE_SIZE)
 226                 return 0;
 227
 228         page = (char *)__get_free_page(GFP_KERNEL);
 229         if (!page)
 230                 return -ENOMEM;
 231
 232         ret = 0;
 233         got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
 234         if (got > 0) {
 235                 int len = strnlen(page, got);
 236
 237                 /* Include the NUL character if it was found */
 238                 if (len < got)
 239                         len++;
 240
 241                 if (len > pos) {
 242                         len -= pos;
 243                         if (len > count)
 244                                 len = count;
 245                         len -= copy_to_user(buf, page+pos, len);
 246                         if (!len)
 247                                 len = -EFAULT;
 248                         ret = len;
 249                 }
 250         }
 251         free_page((unsigned long)page);
 252         return ret;
 253 }
 254
 255 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 256                               size_t count, loff_t *ppos)
 257 {
 258         unsigned long arg_start, arg_end, env_start, env_end;
 259         unsigned long pos, len;
 260         char *page, c;
 261
 262         /* Check if process spawned far enough to have cmdline. */
 263         if (!mm->env_end)
 264                 return 0;
 265
 266         spin_lock(&mm->arg_lock);
 267         arg_start = mm->arg_start;
 268         arg_end = mm->arg_end;
 269         env_start = mm->env_start;
 270         env_end = mm->env_end;
 271         spin_unlock(&mm->arg_lock);
 272
 273         if (arg_start >= arg_end)
 274                 return 0;
 275
 276         /*
 277          * We allow setproctitle() to overwrite the argument
 278          * strings, and overflow past the original end. But
 279          * only when it overflows into the environment area.
 280          */
 281         if (env_start != arg_end || env_end < env_start)
 282                 env_start = env_end = arg_end;
 283         len = env_end - arg_start;
 284
 285         /* We're not going to care if "*ppos" has high bits set */
 286         pos = *ppos;
 287         if (pos >= len)
 288                 return 0;
 289         if (count > len - pos)
 290                 count = len - pos;
 291         if (!count)
 292                 return 0;
 293
 294         /*
 295          * Magical special case: if the argv[] end byte is not
 296          * zero, the user has overwritten it with setproctitle(3).
 297          *
 298          * Possible future enhancement: do this only once when
 299          * pos is 0, and set a flag in the 'struct file'.
 300          */
 301         if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
 302                 return get_mm_proctitle(mm, buf, count, pos, arg_start);
 303
 304         /*
 305          * For the non-setproctitle() case we limit things strictly
 306          * to the [arg_start, arg_end[ range.
 307          */
 308         pos += arg_start;
 309         if (pos < arg_start || pos >= arg_end)
 310                 return 0;
 311         if (count > arg_end - pos)
 312                 count = arg_end - pos;
 313
 314         page = (char *)__get_free_page(GFP_KERNEL);
 315         if (!page)
 316                 return -ENOMEM;
 317
 318         len = 0;
 319         while (count) {
 320                 int got;
 321                 size_t size = min_t(size_t, PAGE_SIZE, count);
 322
 323                 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
 324                 if (got <= 0)
 325                         break;
 326                 got -= copy_to_user(buf, page, got);
 327                 if (unlikely(!got)) {
 328                         if (!len)
 329                                 len = -EFAULT;
 330                         break;
 331                 }
 332                 pos += got;
 333                 buf += got;
 334                 len += got;
 335                 count -= got;
 336         }
 337
 338         free_page((unsigned long)page);
 339         return len;
 340 }
 341
 342 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 343                                 size_t count, loff_t *pos)
 344 {
 345         struct mm_struct *mm;
 346         ssize_t ret;
 347
 348         mm = get_task_mm(tsk);
 349         if (!mm)
 350                 return 0;
 351
 352         ret = get_mm_cmdline(mm, buf, count, pos);
 353         mmput(mm);
 354         return ret;
 355 }
 356
 357 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 358                                      size_t count, loff_t *pos)
 359 {
 360         struct task_struct *tsk;
 361         ssize_t ret;
 362
 363         BUG_ON(*pos < 0);
 364
 365         tsk = get_proc_task(file_inode(file));
 366         if (!tsk)
 367                 return -ESRCH;
 368         ret = get_task_cmdline(tsk, buf, count, pos);
 369         put_task_struct(tsk);
 370         if (ret > 0)
 371                 *pos += ret;
 372         return ret;
 373 }
 374
 375 static const struct file_operations proc_pid_cmdline_ops = {
 376         .read   = proc_pid_cmdline_read,
 377         .llseek = generic_file_llseek,
 378 };
 379
 380 #ifdef CONFIG_KALLSYMS
 381 /*
 382  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 383  * Returns the resolved symbol.  If that fails, simply return the address.
 384  */
 385 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 386                           struct pid *pid, struct task_struct *task)
 387 {
 388         unsigned long wchan;
 389         char symname[KSYM_NAME_LEN];
 390
 391         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 392                 goto print0;
 393
 394         wchan = get_wchan(task);
 395         if (wchan && !lookup_symbol_name(wchan, symname)) {
 396                 seq_puts(m, symname);
 397                 return 0;
 398         }
 399
 400 print0:
 401         seq_putc(m, '0');
 402         return 0;
 403 }
 404 #endif /* CONFIG_KALLSYMS */
 405
 406 static int lock_trace(struct task_struct *task)
 407 {
 408         int err = down_read_killable(&task->signal->exec_update_lock);
 409         if (err)
 410                 return err;
 411         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 412                 up_read(&task->signal->exec_update_lock);
 413                 return -EPERM;
 414         }
 415         return 0;
 416 }
 417
 418 static void unlock_trace(struct task_struct *task)
 419 {
 420         up_read(&task->signal->exec_update_lock);
 421 }
 422
 423 #ifdef CONFIG_STACKTRACE
 424
 425 #define MAX_STACK_TRACE_DEPTH   64
 426
 427 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 428                           struct pid *pid, struct task_struct *task)
 429 {
 430         unsigned long *entries;
 431         int err;
 432
 433         /*
 434          * The ability to racily run the kernel stack unwinder on a running task
 435          * and then observe the unwinder output is scary; while it is useful for
 436          * debugging kernel issues, it can also allow an attacker to leak kernel
 437          * stack contents.
 438          * Doing this in a manner that is at least safe from races would require
 439          * some work to ensure that the remote task can not be scheduled; and
 440          * even then, this would still expose the unwinder as local attack
 441          * surface.
 442          * Therefore, this interface is restricted to root.
 443          */
 444         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 445                 return -EACCES;
 446
 447         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 448                                 GFP_KERNEL);
 449         if (!entries)
 450                 return -ENOMEM;
 451
 452         err = lock_trace(task);
 453         if (!err) {
 454                 unsigned int i, nr_entries;
 455
 456                 nr_entries = stack_trace_save_tsk(task, entries,
 457                                                   MAX_STACK_TRACE_DEPTH, 0);
 458
 459                 for (i = 0; i < nr_entries; i++) {
 460                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 461                 }
 462
 463                 unlock_trace(task);
 464         }
 465         kfree(entries);
 466
 467         return err;
 468 }
 469 #endif
 470
 471 #ifdef CONFIG_SCHED_INFO
 472 /*
 473  * Provides /proc/PID/schedstat
 474  */
 475 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 476                               struct pid *pid, struct task_struct *task)
 477 {
 478         if (unlikely(!sched_info_on()))
 479                 seq_puts(m, "0 0 0\n");
 480         else
 481                 seq_printf(m, "%llu %llu %lu\n",
 482                    (unsigned long long)task->se.sum_exec_runtime,
 483                    (unsigned long long)task->sched_info.run_delay,
 484                    task->sched_info.pcount);
 485
 486         return 0;
 487 }
 488 #endif
 489
 490 #ifdef CONFIG_LATENCYTOP
 491 static int lstats_show_proc(struct seq_file *m, void *v)
 492 {
 493         int i;
 494         struct inode *inode = m->private;
 495         struct task_struct *task = get_proc_task(inode);
 496
 497         if (!task)
 498                 return -ESRCH;
 499         seq_puts(m, "Latency Top version : v0.1\n");
 500         for (i = 0; i < LT_SAVECOUNT; i++) {
 501                 struct latency_record *lr = &task->latency_record[i];
 502                 if (lr->backtrace[0]) {
 503                         int q;
 504                         seq_printf(m, "%i %li %li",
 505                                    lr->count, lr->time, lr->max);
 506                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 507                                 unsigned long bt = lr->backtrace[q];
 508
 509                                 if (!bt)
 510                                         break;
 511                                 seq_printf(m, " %ps", (void *)bt);
 512                         }
 513                         seq_putc(m, '\n');
 514                 }
 515
 516         }
 517         put_task_struct(task);
 518         return 0;
 519 }
 520
 521 static int lstats_open(struct inode *inode, struct file *file)
 522 {
 523         return single_open(file, lstats_show_proc, inode);
 524 }
 525
 526 static ssize_t lstats_write(struct file *file, const char __user *buf,
 527                             size_t count, loff_t *offs)
 528 {
 529         struct task_struct *task = get_proc_task(file_inode(file));
 530
 531         if (!task)
 532                 return -ESRCH;
 533         clear_tsk_latency_tracing(task);
 534         put_task_struct(task);
 535
 536         return count;
 537 }
 538
 539 static const struct file_operations proc_lstats_operations = {
 540         .open           = lstats_open,
 541         .read           = seq_read,
 542         .write          = lstats_write,
 543         .llseek         = seq_lseek,
 544         .release        = single_release,
 545 };
 546
 547 #endif
 548
 549 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 550                           struct pid *pid, struct task_struct *task)
 551 {
 552         unsigned long totalpages = totalram_pages() + total_swap_pages;
 553         unsigned long points = 0;
 554         long badness;
 555
 556         badness = oom_badness(task, totalpages);
 557         /*
 558          * Special case OOM_SCORE_ADJ_MIN for all others scale the
 559          * badness value into [0, 2000] range which we have been
 560          * exporting for a long time so userspace might depend on it.
 561          */
 562         if (badness != LONG_MIN)
 563                 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
 564
 565         seq_printf(m, "%lu\n", points);
 566
 567         return 0;
 568 }
 569
 570 struct limit_names {
 571         const char *name;
 572         const char *unit;
 573 };
 574
 575 static const struct limit_names lnames[RLIM_NLIMITS] = {
 576         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 577         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 578         [RLIMIT_DATA] = {"Max data size", "bytes"},
 579         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 580         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 581         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 582         [RLIMIT_NPROC] = {"Max processes", "processes"},
 583         [RLIMIT_NOFILE] = {"Max open files", "files"},
 584         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 585         [RLIMIT_AS] = {"Max address space", "bytes"},
 586         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 587         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 588         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 589         [RLIMIT_NICE] = {"Max nice priority", NULL},
 590         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 591         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 592 };
 593
 594 /* Display limits for a process */
 595 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 596                            struct pid *pid, struct task_struct *task)
 597 {
 598         unsigned int i;
 599         unsigned long flags;
 600
 601         struct rlimit rlim[RLIM_NLIMITS];
 602
 603         if (!lock_task_sighand(task, &flags))
 604                 return 0;
 605         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 606         unlock_task_sighand(task, &flags);
 607
 608         /*
 609          * print the file header
 610          */
 611         seq_puts(m, "Limit                     "
 612                 "Soft Limit           "
 613                 "Hard Limit           "
 614                 "Units     \n");
 615
 616         for (i = 0; i < RLIM_NLIMITS; i++) {
 617                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 618                         seq_printf(m, "%-25s %-20s ",
 619                                    lnames[i].name, "unlimited");
 620                 else
 621                         seq_printf(m, "%-25s %-20lu ",
 622                                    lnames[i].name, rlim[i].rlim_cur);
 623
 624                 if (rlim[i].rlim_max == RLIM_INFINITY)
 625                         seq_printf(m, "%-20s ", "unlimited");
 626                 else
 627                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 628
 629                 if (lnames[i].unit)
 630                         seq_printf(m, "%-10s\n", lnames[i].unit);
 631                 else
 632                         seq_putc(m, '\n');
 633         }
 634
 635         return 0;
 636 }
 637
 638 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 639 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 640                             struct pid *pid, struct task_struct *task)
 641 {
 642         struct syscall_info info;
 643         u64 *args = &info.data.args[0];
 644         int res;
 645
 646         res = lock_trace(task);
 647         if (res)
 648                 return res;
 649
 650         if (task_current_syscall(task, &info))
 651                 seq_puts(m, "running\n");
 652         else if (info.data.nr < 0)
 653                 seq_printf(m, "%d 0x%llx 0x%llx\n",
 654                            info.data.nr, info.sp, info.data.instruction_pointer);
 655         else
 656                 seq_printf(m,
 657                        "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
 658                        info.data.nr,
 659                        args[0], args[1], args[2], args[3], args[4], args[5],
 660                        info.sp, info.data.instruction_pointer);
 661         unlock_trace(task);
 662
 663         return 0;
 664 }
 665 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 666
 667 /************************************************************************/
 668 /*                       Here the fs part begins                        */
 669 /************************************************************************/
 670
 671 /* permission checks */
 672 static int proc_fd_access_allowed(struct inode *inode)
 673 {
 674         struct task_struct *task;
 675         int allowed = 0;
 676         /* Allow access to a task's file descriptors if it is us or we
 677          * may use ptrace attach to the process and find out that
 678          * information.
 679          */
 680         task = get_proc_task(inode);
 681         if (task) {
 682                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 683                 put_task_struct(task);
 684         }
 685         return allowed;
 686 }
 687
 688 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 689 {
 690         int error;
 691         struct inode *inode = d_inode(dentry);
 692
 693         if (attr->ia_valid & ATTR_MODE)
 694                 return -EPERM;
 695
 696         error = setattr_prepare(dentry, attr);
 697         if (error)
 698                 return error;
 699
 700         setattr_copy(inode, attr);
 701         mark_inode_dirty(inode);
 702         return 0;
 703 }
 704
 705 /*
 706  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 707  * or euid/egid (for hide_pid_min=2)?
 708  */
 709 static bool has_pid_permissions(struct proc_fs_info *fs_info,
 710                                  struct task_struct *task,
 711                                  enum proc_hidepid hide_pid_min)
 712 {
 713         /*
 714          * If 'hidpid' mount option is set force a ptrace check,
 715          * we indicate that we are using a filesystem syscall
 716          * by passing PTRACE_MODE_READ_FSCREDS
 717          */
 718         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
 719                 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 720
 721         if (fs_info->hide_pid < hide_pid_min)
 722                 return true;
 723         if (in_group_p(fs_info->pid_gid))
 724                 return true;
 725         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 726 }
 727
 728
 729 static int proc_pid_permission(struct inode *inode, int mask)
 730 {
 731         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
 732         struct task_struct *task;
 733         bool has_perms;
 734
 735         task = get_proc_task(inode);
 736         if (!task)
 737                 return -ESRCH;
 738         has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
 739         put_task_struct(task);
 740
 741         if (!has_perms) {
 742                 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
 743                         /*
 744                          * Let's make getdents(), stat(), and open()
 745                          * consistent with each other.  If a process
 746                          * may not stat() a file, it shouldn't be seen
 747                          * in procfs at all.
 748                          */
 749                         return -ENOENT;
 750                 }
 751
 752                 return -EPERM;
 753         }
 754         return generic_permission(inode, mask);
 755 }
 756
 757
 758
 759 static const struct inode_operations proc_def_inode_operations = {
 760         .setattr        = proc_setattr,
 761 };
 762
 763 static int proc_single_show(struct seq_file *m, void *v)
 764 {
 765         struct inode *inode = m->private;
 766         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
 767         struct pid *pid = proc_pid(inode);
 768         struct task_struct *task;
 769         int ret;
 770
 771         task = get_pid_task(pid, PIDTYPE_PID);
 772         if (!task)
 773                 return -ESRCH;
 774
 775         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 776
 777         put_task_struct(task);
 778         return ret;
 779 }
 780
 781 static int proc_single_open(struct inode *inode, struct file *filp)
 782 {
 783         return single_open(filp, proc_single_show, inode);
 784 }
 785
 786 static const struct file_operations proc_single_file_operations = {
 787         .open           = proc_single_open,
 788         .read           = seq_read,
 789         .llseek         = seq_lseek,
 790         .release        = single_release,
 791 };
 792
 793
 794 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 795 {
 796         struct task_struct *task = get_proc_task(inode);
 797         struct mm_struct *mm = ERR_PTR(-ESRCH);
 798
 799         if (task) {
 800                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 801                 put_task_struct(task);
 802
 803                 if (!IS_ERR_OR_NULL(mm)) {
 804                         /* ensure this mm_struct can't be freed */
 805                         mmgrab(mm);
 806                         /* but do not pin its memory */
 807                         mmput(mm);
 808                 }
 809         }
 810
 811         return mm;
 812 }
 813
 814 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 815 {
 816         struct mm_struct *mm = proc_mem_open(inode, mode);
 817
 818         if (IS_ERR(mm))
 819                 return PTR_ERR(mm);
 820
 821         file->private_data = mm;
 822         return 0;
 823 }
 824
 825 static int mem_open(struct inode *inode, struct file *file)
 826 {
 827         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 828
 829         /* OK to pass negative loff_t, we can catch out-of-range */
 830         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 831
 832         return ret;
 833 }
 834
 835 static ssize_t mem_rw(struct file *file, char __user *buf,
 836                         size_t count, loff_t *ppos, int write)
 837 {
 838         struct mm_struct *mm = file->private_data;
 839         unsigned long addr = *ppos;
 840         ssize_t copied;
 841         char *page;
 842         unsigned int flags;
 843
 844         if (!mm)
 845                 return 0;
 846
 847         page = (char *)__get_free_page(GFP_KERNEL);
 848         if (!page)
 849                 return -ENOMEM;
 850
 851         copied = 0;
 852         if (!mmget_not_zero(mm))
 853                 goto free;
 854
 855         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 856
 857         while (count > 0) {
 858                 int this_len = min_t(int, count, PAGE_SIZE);
 859
 860                 if (write && copy_from_user(page, buf, this_len)) {
 861                         copied = -EFAULT;
 862                         break;
 863                 }
 864
 865                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 866                 if (!this_len) {
 867                         if (!copied)
 868                                 copied = -EIO;
 869                         break;
 870                 }
 871
 872                 if (!write && copy_to_user(buf, page, this_len)) {
 873                         copied = -EFAULT;
 874                         break;
 875                 }
 876
 877                 buf += this_len;
 878                 addr += this_len;
 879                 copied += this_len;
 880                 count -= this_len;
 881         }
 882         *ppos = addr;
 883
 884         mmput(mm);
 885 free:
 886         free_page((unsigned long) page);
 887         return copied;
 888 }
 889
 890 static ssize_t mem_read(struct file *file, char __user *buf,
 891                         size_t count, loff_t *ppos)
 892 {
 893         return mem_rw(file, buf, count, ppos, 0);
 894 }
 895
 896 static ssize_t mem_write(struct file *file, const char __user *buf,
 897                          size_t count, loff_t *ppos)
 898 {
 899         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 900 }
 901
 902 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 903 {
 904         switch (orig) {
 905         case 0:
 906                 file->f_pos = offset;
 907                 break;
 908         case 1:
 909                 file->f_pos += offset;
 910                 break;
 911         default:
 912                 return -EINVAL;
 913         }
 914         force_successful_syscall_return();
 915         return file->f_pos;
 916 }
 917
 918 static int mem_release(struct inode *inode, struct file *file)
 919 {
 920         struct mm_struct *mm = file->private_data;
 921         if (mm)
 922                 mmdrop(mm);
 923         return 0;
 924 }
 925
 926 static const struct file_operations proc_mem_operations = {
 927         .llseek         = mem_lseek,
 928         .read           = mem_read,
 929         .write          = mem_write,
 930         .open           = mem_open,
 931         .release        = mem_release,
 932 };
 933
 934 static int environ_open(struct inode *inode, struct file *file)
 935 {
 936         return __mem_open(inode, file, PTRACE_MODE_READ);
 937 }
 938
 939 static ssize_t environ_read(struct file *file, char __user *buf,
 940                         size_t count, loff_t *ppos)
 941 {
 942         char *page;
 943         unsigned long src = *ppos;
 944         int ret = 0;
 945         struct mm_struct *mm = file->private_data;
 946         unsigned long env_start, env_end;
 947
 948         /* Ensure the process spawned far enough to have an environment. */
 949         if (!mm || !mm->env_end)
 950                 return 0;
 951
 952         page = (char *)__get_free_page(GFP_KERNEL);
 953         if (!page)
 954                 return -ENOMEM;
 955
 956         ret = 0;
 957         if (!mmget_not_zero(mm))
 958                 goto free;
 959
 960         spin_lock(&mm->arg_lock);
 961         env_start = mm->env_start;
 962         env_end = mm->env_end;
 963         spin_unlock(&mm->arg_lock);
 964
 965         while (count > 0) {
 966                 size_t this_len, max_len;
 967                 int retval;
 968
 969                 if (src >= (env_end - env_start))
 970                         break;
 971
 972                 this_len = env_end - (env_start + src);
 973
 974                 max_len = min_t(size_t, PAGE_SIZE, count);
 975                 this_len = min(max_len, this_len);
 976
 977                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 978
 979                 if (retval <= 0) {
 980                         ret = retval;
 981                         break;
 982                 }
 983
 984                 if (copy_to_user(buf, page, retval)) {
 985                         ret = -EFAULT;
 986                         break;
 987                 }
 988
 989                 ret += retval;
 990                 src += retval;
 991                 buf += retval;
 992                 count -= retval;
 993         }
 994         *ppos = src;
 995         mmput(mm);
 996
 997 free:
 998         free_page((unsigned long) page);
 999         return ret;
1000 }
1001
1002 static const struct file_operations proc_environ_operations = {
1003         .open           = environ_open,
1004         .read           = environ_read,
1005         .llseek         = generic_file_llseek,
1006         .release        = mem_release,
1007 };
1008
1009 static int auxv_open(struct inode *inode, struct file *file)
1010 {
1011         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1012 }
1013
1014 static ssize_t auxv_read(struct file *file, char __user *buf,
1015                         size_t count, loff_t *ppos)
1016 {
1017         struct mm_struct *mm = file->private_data;
1018         unsigned int nwords = 0;
1019
1020         if (!mm)
1021                 return 0;
1022         do {
1023                 nwords += 2;
1024         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1025         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1026                                        nwords * sizeof(mm->saved_auxv[0]));
1027 }
1028
1029 static const struct file_operations proc_auxv_operations = {
1030         .open           = auxv_open,
1031         .read           = auxv_read,
1032         .llseek         = generic_file_llseek,
1033         .release        = mem_release,
1034 };
1035
1036 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1037                             loff_t *ppos)
1038 {
1039         struct task_struct *task = get_proc_task(file_inode(file));
1040         char buffer[PROC_NUMBUF];
1041         int oom_adj = OOM_ADJUST_MIN;
1042         size_t len;
1043
1044         if (!task)
1045                 return -ESRCH;
1046         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1047                 oom_adj = OOM_ADJUST_MAX;
1048         else
1049                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1050                           OOM_SCORE_ADJ_MAX;
1051         put_task_struct(task);
1052         if (oom_adj > OOM_ADJUST_MAX)
1053                 oom_adj = OOM_ADJUST_MAX;
1054         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1055         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1056 }
1057
1058 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1059 {
1060         struct mm_struct *mm = NULL;
1061         struct task_struct *task;
1062         int err = 0;
1063
1064         task = get_proc_task(file_inode(file));
1065         if (!task)
1066                 return -ESRCH;
1067
1068         mutex_lock(&oom_adj_mutex);
1069         if (legacy) {
1070                 if (oom_adj < task->signal->oom_score_adj &&
1071                                 !capable(CAP_SYS_RESOURCE)) {
1072                         err = -EACCES;
1073                         goto err_unlock;
1074                 }
1075                 /*
1076                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1077                  * /proc/pid/oom_score_adj instead.
1078                  */
1079                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1080                           current->comm, task_pid_nr(current), task_pid_nr(task),
1081                           task_pid_nr(task));
1082         } else {
1083                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1084                                 !capable(CAP_SYS_RESOURCE)) {
1085                         err = -EACCES;
1086                         goto err_unlock;
1087                 }
1088         }
1089
1090         /*
1091          * Make sure we will check other processes sharing the mm if this is
1092          * not vfrok which wants its own oom_score_adj.
1093          * pin the mm so it doesn't go away and get reused after task_unlock
1094          */
1095         if (!task->vfork_done) {
1096                 struct task_struct *p = find_lock_task_mm(task);
1097
1098                 if (p) {
1099                         if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1100                                 mm = p->mm;
1101                                 mmgrab(mm);
1102                         }
1103                         task_unlock(p);
1104                 }
1105         }
1106
1107         task->signal->oom_score_adj = oom_adj;
1108         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1109                 task->signal->oom_score_adj_min = (short)oom_adj;
1110         trace_oom_score_adj_update(task);
1111
1112         if (mm) {
1113                 struct task_struct *p;
1114
1115                 rcu_read_lock();
1116                 for_each_process(p) {
1117                         if (same_thread_group(task, p))
1118                                 continue;
1119
1120                         /* do not touch kernel threads or the global init */
1121                         if (p->flags & PF_KTHREAD || is_global_init(p))
1122                                 continue;
1123
1124                         task_lock(p);
1125                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1126                                 p->signal->oom_score_adj = oom_adj;
1127                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1128                                         p->signal->oom_score_adj_min = (short)oom_adj;
1129                         }
1130                         task_unlock(p);
1131                 }
1132                 rcu_read_unlock();
1133                 mmdrop(mm);
1134         }
1135 err_unlock:
1136         mutex_unlock(&oom_adj_mutex);
1137         put_task_struct(task);
1138         return err;
1139 }
1140
1141 /*
1142  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1143  * kernels.  The effective policy is defined by oom_score_adj, which has a
1144  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1145  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1146  * Processes that become oom disabled via oom_adj will still be oom disabled
1147  * with this implementation.
1148  *
1149  * oom_adj cannot be removed since existing userspace binaries use it.
1150  */
1151 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1152                              size_t count, loff_t *ppos)
1153 {
1154         char buffer[PROC_NUMBUF];
1155         int oom_adj;
1156         int err;
1157
1158         memset(buffer, 0, sizeof(buffer));
1159         if (count > sizeof(buffer) - 1)
1160                 count = sizeof(buffer) - 1;
1161         if (copy_from_user(buffer, buf, count)) {
1162                 err = -EFAULT;
1163                 goto out;
1164         }
1165
1166         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1167         if (err)
1168                 goto out;
1169         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1170              oom_adj != OOM_DISABLE) {
1171                 err = -EINVAL;
1172                 goto out;
1173         }
1174
1175         /*
1176          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1177          * value is always attainable.
1178          */
1179         if (oom_adj == OOM_ADJUST_MAX)
1180                 oom_adj = OOM_SCORE_ADJ_MAX;
1181         else
1182                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1183
1184         err = __set_oom_adj(file, oom_adj, true);
1185 out:
1186         return err < 0 ? err : count;
1187 }
1188
1189 static const struct file_operations proc_oom_adj_operations = {
1190         .read           = oom_adj_read,
1191         .write          = oom_adj_write,
1192         .llseek         = generic_file_llseek,
1193 };
1194
1195 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1196                                         size_t count, loff_t *ppos)
1197 {
1198         struct task_struct *task = get_proc_task(file_inode(file));
1199         char buffer[PROC_NUMBUF];
1200         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1201         size_t len;
1202
1203         if (!task)
1204                 return -ESRCH;
1205         oom_score_adj = task->signal->oom_score_adj;
1206         put_task_struct(task);
1207         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1208         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1209 }
1210
1211 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1212                                         size_t count, loff_t *ppos)
1213 {
1214         char buffer[PROC_NUMBUF];
1215         int oom_score_adj;
1216         int err;
1217
1218         memset(buffer, 0, sizeof(buffer));
1219         if (count > sizeof(buffer) - 1)
1220                 count = sizeof(buffer) - 1;
1221         if (copy_from_user(buffer, buf, count)) {
1222                 err = -EFAULT;
1223                 goto out;
1224         }
1225
1226         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1227         if (err)
1228                 goto out;
1229         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1230                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1231                 err = -EINVAL;
1232                 goto out;
1233         }
1234
1235         err = __set_oom_adj(file, oom_score_adj, false);
1236 out:
1237         return err < 0 ? err : count;
1238 }
1239
1240 static const struct file_operations proc_oom_score_adj_operations = {
1241         .read           = oom_score_adj_read,
1242         .write          = oom_score_adj_write,
1243         .llseek         = default_llseek,
1244 };
1245
1246 #ifdef CONFIG_AUDIT
1247 #define TMPBUFLEN 11
1248 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1249                                   size_t count, loff_t *ppos)
1250 {
1251         struct inode * inode = file_inode(file);
1252         struct task_struct *task = get_proc_task(inode);
1253         ssize_t length;
1254         char tmpbuf[TMPBUFLEN];
1255
1256         if (!task)
1257                 return -ESRCH;
1258         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1259                            from_kuid(file->f_cred->user_ns,
1260                                      audit_get_loginuid(task)));
1261         put_task_struct(task);
1262         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1263 }
1264
1265 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1266                                    size_t count, loff_t *ppos)
1267 {
1268         struct inode * inode = file_inode(file);
1269         uid_t loginuid;
1270         kuid_t kloginuid;
1271         int rv;
1272
1273         /* Don't let kthreads write their own loginuid */
1274         if (current->flags & PF_KTHREAD)
1275                 return -EPERM;
1276
1277         rcu_read_lock();
1278         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1279                 rcu_read_unlock();
1280                 return -EPERM;
1281         }
1282         rcu_read_unlock();
1283
1284         if (*ppos != 0) {
1285                 /* No partial writes. */
1286                 return -EINVAL;
1287         }
1288
1289         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1290         if (rv < 0)
1291                 return rv;
1292
1293         /* is userspace tring to explicitly UNSET the loginuid? */
1294         if (loginuid == AUDIT_UID_UNSET) {
1295                 kloginuid = INVALID_UID;
1296         } else {
1297                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1298                 if (!uid_valid(kloginuid))
1299                         return -EINVAL;
1300         }
1301
1302         rv = audit_set_loginuid(kloginuid);
1303         if (rv < 0)
1304                 return rv;
1305         return count;
1306 }
1307
1308 static const struct file_operations proc_loginuid_operations = {
1309         .read           = proc_loginuid_read,
1310         .write          = proc_loginuid_write,
1311         .llseek         = generic_file_llseek,
1312 };
1313
1314 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1315                                   size_t count, loff_t *ppos)
1316 {
1317         struct inode * inode = file_inode(file);
1318         struct task_struct *task = get_proc_task(inode);
1319         ssize_t length;
1320         char tmpbuf[TMPBUFLEN];
1321
1322         if (!task)
1323                 return -ESRCH;
1324         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1325                                 audit_get_sessionid(task));
1326         put_task_struct(task);
1327         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1328 }
1329
1330 static const struct file_operations proc_sessionid_operations = {
1331         .read           = proc_sessionid_read,
1332         .llseek         = generic_file_llseek,
1333 };
1334 #endif
1335
1336 #ifdef CONFIG_FAULT_INJECTION
1337 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1338                                       size_t count, loff_t *ppos)
1339 {
1340         struct task_struct *task = get_proc_task(file_inode(file));
1341         char buffer[PROC_NUMBUF];
1342         size_t len;
1343         int make_it_fail;
1344
1345         if (!task)
1346                 return -ESRCH;
1347         make_it_fail = task->make_it_fail;
1348         put_task_struct(task);
1349
1350         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1351
1352         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1353 }
1354
1355 static ssize_t proc_fault_inject_write(struct file * file,
1356                         const char __user * buf, size_t count, loff_t *ppos)
1357 {
1358         struct task_struct *task;
1359         char buffer[PROC_NUMBUF];
1360         int make_it_fail;
1361         int rv;
1362
1363         if (!capable(CAP_SYS_RESOURCE))
1364                 return -EPERM;
1365         memset(buffer, 0, sizeof(buffer));
1366         if (count > sizeof(buffer) - 1)
1367                 count = sizeof(buffer) - 1;
1368         if (copy_from_user(buffer, buf, count))
1369                 return -EFAULT;
1370         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1371         if (rv < 0)
1372                 return rv;
1373         if (make_it_fail < 0 || make_it_fail > 1)
1374                 return -EINVAL;
1375
1376         task = get_proc_task(file_inode(file));
1377         if (!task)
1378                 return -ESRCH;
1379         task->make_it_fail = make_it_fail;
1380         put_task_struct(task);
1381
1382         return count;
1383 }
1384
1385 static const struct file_operations proc_fault_inject_operations = {
1386         .read           = proc_fault_inject_read,
1387         .write          = proc_fault_inject_write,
1388         .llseek         = generic_file_llseek,
1389 };
1390
1391 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1392                                    size_t count, loff_t *ppos)
1393 {
1394         struct task_struct *task;
1395         int err;
1396         unsigned int n;
1397
1398         err = kstrtouint_from_user(buf, count, 0, &n);
1399         if (err)
1400                 return err;
1401
1402         task = get_proc_task(file_inode(file));
1403         if (!task)
1404                 return -ESRCH;
1405         task->fail_nth = n;
1406         put_task_struct(task);
1407
1408         return count;
1409 }
1410
1411 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1412                                   size_t count, loff_t *ppos)
1413 {
1414         struct task_struct *task;
1415         char numbuf[PROC_NUMBUF];
1416         ssize_t len;
1417
1418         task = get_proc_task(file_inode(file));
1419         if (!task)
1420                 return -ESRCH;
1421         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1422         put_task_struct(task);
1423         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1424 }
1425
1426 static const struct file_operations proc_fail_nth_operations = {
1427         .read           = proc_fail_nth_read,
1428         .write          = proc_fail_nth_write,
1429 };
1430 #endif
1431
1432
1433 #ifdef CONFIG_SCHED_DEBUG
1434 /*
1435  * Print out various scheduling related per-task fields:
1436  */
1437 static int sched_show(struct seq_file *m, void *v)
1438 {
1439         struct inode *inode = m->private;
1440         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1441         struct task_struct *p;
1442
1443         p = get_proc_task(inode);
1444         if (!p)
1445                 return -ESRCH;
1446         proc_sched_show_task(p, ns, m);
1447
1448         put_task_struct(p);
1449
1450         return 0;
1451 }
1452
1453 static ssize_t
1454 sched_write(struct file *file, const char __user *buf,
1455             size_t count, loff_t *offset)
1456 {
1457         struct inode *inode = file_inode(file);
1458         struct task_struct *p;
1459
1460         p = get_proc_task(inode);
1461         if (!p)
1462                 return -ESRCH;
1463         proc_sched_set_task(p);
1464
1465         put_task_struct(p);
1466
1467         return count;
1468 }
1469
1470 static int sched_open(struct inode *inode, struct file *filp)
1471 {
1472         return single_open(filp, sched_show, inode);
1473 }
1474
1475 static const struct file_operations proc_pid_sched_operations = {
1476         .open           = sched_open,
1477         .read           = seq_read,
1478         .write          = sched_write,
1479         .llseek         = seq_lseek,
1480         .release        = single_release,
1481 };
1482
1483 #endif
1484
1485 #ifdef CONFIG_SCHED_AUTOGROUP
1486 /*
1487  * Print out autogroup related information:
1488  */
1489 static int sched_autogroup_show(struct seq_file *m, void *v)
1490 {
1491         struct inode *inode = m->private;
1492         struct task_struct *p;
1493
1494         p = get_proc_task(inode);
1495         if (!p)
1496                 return -ESRCH;
1497         proc_sched_autogroup_show_task(p, m);
1498
1499         put_task_struct(p);
1500
1501         return 0;
1502 }
1503
1504 static ssize_t
1505 sched_autogroup_write(struct file *file, const char __user *buf,
1506             size_t count, loff_t *offset)
1507 {
1508         struct inode *inode = file_inode(file);
1509         struct task_struct *p;
1510         char buffer[PROC_NUMBUF];
1511         int nice;
1512         int err;
1513
1514         memset(buffer, 0, sizeof(buffer));
1515         if (count > sizeof(buffer) - 1)
1516                 count = sizeof(buffer) - 1;
1517         if (copy_from_user(buffer, buf, count))
1518                 return -EFAULT;
1519
1520         err = kstrtoint(strstrip(buffer), 0, &nice);
1521         if (err < 0)
1522                 return err;
1523
1524         p = get_proc_task(inode);
1525         if (!p)
1526                 return -ESRCH;
1527
1528         err = proc_sched_autogroup_set_nice(p, nice);
1529         if (err)
1530                 count = err;
1531
1532         put_task_struct(p);
1533
1534         return count;
1535 }
1536
1537 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1538 {
1539         int ret;
1540
1541         ret = single_open(filp, sched_autogroup_show, NULL);
1542         if (!ret) {
1543                 struct seq_file *m = filp->private_data;
1544
1545                 m->private = inode;
1546         }
1547         return ret;
1548 }
1549
1550 static const struct file_operations proc_pid_sched_autogroup_operations = {
1551         .open           = sched_autogroup_open,
1552         .read           = seq_read,
1553         .write          = sched_autogroup_write,
1554         .llseek         = seq_lseek,
1555         .release        = single_release,
1556 };
1557
1558 #endif /* CONFIG_SCHED_AUTOGROUP */
1559
1560 #ifdef CONFIG_TIME_NS
1561 static int timens_offsets_show(struct seq_file *m, void *v)
1562 {
1563         struct task_struct *p;
1564
1565         p = get_proc_task(file_inode(m->file));
1566         if (!p)
1567                 return -ESRCH;
1568         proc_timens_show_offsets(p, m);
1569
1570         put_task_struct(p);
1571
1572         return 0;
1573 }
1574
1575 static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1576                                     size_t count, loff_t *ppos)
1577 {
1578         struct inode *inode = file_inode(file);
1579         struct proc_timens_offset offsets[2];
1580         char *kbuf = NULL, *pos, *next_line;
1581         struct task_struct *p;
1582         int ret, noffsets;
1583
1584         /* Only allow < page size writes at the beginning of the file */
1585         if ((*ppos != 0) || (count >= PAGE_SIZE))
1586                 return -EINVAL;
1587
1588         /* Slurp in the user data */
1589         kbuf = memdup_user_nul(buf, count);
1590         if (IS_ERR(kbuf))
1591                 return PTR_ERR(kbuf);
1592
1593         /* Parse the user data */
1594         ret = -EINVAL;
1595         noffsets = 0;
1596         for (pos = kbuf; pos; pos = next_line) {
1597                 struct proc_timens_offset *off = &offsets[noffsets];
1598                 char clock[10];
1599                 int err;
1600
1601                 /* Find the end of line and ensure we don't look past it */
1602                 next_line = strchr(pos, '\n');
1603                 if (next_line) {
1604                         *next_line = '\0';
1605                         next_line++;
1606                         if (*next_line == '\0')
1607                                 next_line = NULL;
1608                 }
1609
1610                 err = sscanf(pos, "%9s %lld %lu", clock,
1611                                 &off->val.tv_sec, &off->val.tv_nsec);
1612                 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1613                         goto out;
1614
1615                 clock[sizeof(clock) - 1] = 0;
1616                 if (strcmp(clock, "monotonic") == 0 ||
1617                     strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1618                         off->clockid = CLOCK_MONOTONIC;
1619                 else if (strcmp(clock, "boottime") == 0 ||
1620                          strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1621                         off->clockid = CLOCK_BOOTTIME;
1622                 else
1623                         goto out;
1624
1625                 noffsets++;
1626                 if (noffsets == ARRAY_SIZE(offsets)) {
1627                         if (next_line)
1628                                 count = next_line - kbuf;
1629                         break;
1630                 }
1631         }
1632
1633         ret = -ESRCH;
1634         p = get_proc_task(inode);
1635         if (!p)
1636                 goto out;
1637         ret = proc_timens_set_offset(file, p, offsets, noffsets);
1638         put_task_struct(p);
1639         if (ret)
1640                 goto out;
1641
1642         ret = count;
1643 out:
1644         kfree(kbuf);
1645         return ret;
1646 }
1647
1648 static int timens_offsets_open(struct inode *inode, struct file *filp)
1649 {
1650         return single_open(filp, timens_offsets_show, inode);
1651 }
1652
1653 static const struct file_operations proc_timens_offsets_operations = {
1654         .open           = timens_offsets_open,
1655         .read           = seq_read,
1656         .write          = timens_offsets_write,
1657         .llseek         = seq_lseek,
1658         .release        = single_release,
1659 };
1660 #endif /* CONFIG_TIME_NS */
1661
1662 static ssize_t comm_write(struct file *file, const char __user *buf,
1663                                 size_t count, loff_t *offset)
1664 {
1665         struct inode *inode = file_inode(file);
1666         struct task_struct *p;
1667         char buffer[TASK_COMM_LEN];
1668         const size_t maxlen = sizeof(buffer) - 1;
1669
1670         memset(buffer, 0, sizeof(buffer));
1671         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1672                 return -EFAULT;
1673
1674         p = get_proc_task(inode);
1675         if (!p)
1676                 return -ESRCH;
1677
1678         if (same_thread_group(current, p))
1679                 set_task_comm(p, buffer);
1680         else
1681                 count = -EINVAL;
1682
1683         put_task_struct(p);
1684
1685         return count;
1686 }
1687
1688 static int comm_show(struct seq_file *m, void *v)
1689 {
1690         struct inode *inode = m->private;
1691         struct task_struct *p;
1692
1693         p = get_proc_task(inode);
1694         if (!p)
1695                 return -ESRCH;
1696
1697         proc_task_name(m, p, false);
1698         seq_putc(m, '\n');
1699
1700         put_task_struct(p);
1701
1702         return 0;
1703 }
1704
1705 static int comm_open(struct inode *inode, struct file *filp)
1706 {
1707         return single_open(filp, comm_show, inode);
1708 }
1709
1710 static const struct file_operations proc_pid_set_comm_operations = {
1711         .open           = comm_open,
1712         .read           = seq_read,
1713         .write          = comm_write,
1714         .llseek         = seq_lseek,
1715         .release        = single_release,
1716 };
1717
1718 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1719 {
1720         struct task_struct *task;
1721         struct file *exe_file;
1722
1723         task = get_proc_task(d_inode(dentry));
1724         if (!task)
1725                 return -ENOENT;
1726         exe_file = get_task_exe_file(task);
1727         put_task_struct(task);
1728         if (exe_file) {
1729                 *exe_path = exe_file->f_path;
1730                 path_get(&exe_file->f_path);
1731                 fput(exe_file);
1732                 return 0;
1733         } else
1734                 return -ENOENT;
1735 }
1736
1737 static const char *proc_pid_get_link(struct dentry *dentry,
1738                                      struct inode *inode,
1739                                      struct delayed_call *done)
1740 {
1741         struct path path;
1742         int error = -EACCES;
1743
1744         if (!dentry)
1745                 return ERR_PTR(-ECHILD);
1746
1747         /* Are we allowed to snoop on the tasks file descriptors? */
1748         if (!proc_fd_access_allowed(inode))
1749                 goto out;
1750
1751         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1752         if (error)
1753                 goto out;
1754
1755         error = nd_jump_link(&path);
1756 out:
1757         return ERR_PTR(error);
1758 }
1759
1760 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1761 {
1762         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1763         char *pathname;
1764         int len;
1765
1766         if (!tmp)
1767                 return -ENOMEM;
1768
1769         pathname = d_path(path, tmp, PAGE_SIZE);
1770         len = PTR_ERR(pathname);
1771         if (IS_ERR(pathname))
1772                 goto out;
1773         len = tmp + PAGE_SIZE - 1 - pathname;
1774
1775         if (len > buflen)
1776                 len = buflen;
1777         if (copy_to_user(buffer, pathname, len))
1778                 len = -EFAULT;
1779  out:
1780         free_page((unsigned long)tmp);
1781         return len;
1782 }
1783
1784 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1785 {
1786         int error = -EACCES;
1787         struct inode *inode = d_inode(dentry);
1788         struct path path;
1789
1790         /* Are we allowed to snoop on the tasks file descriptors? */
1791         if (!proc_fd_access_allowed(inode))
1792                 goto out;
1793
1794         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1795         if (error)
1796                 goto out;
1797
1798         error = do_proc_readlink(&path, buffer, buflen);
1799         path_put(&path);
1800 out:
1801         return error;
1802 }
1803
1804 const struct inode_operations proc_pid_link_inode_operations = {
1805         .readlink       = proc_pid_readlink,
1806         .get_link       = proc_pid_get_link,
1807         .setattr        = proc_setattr,
1808 };
1809
1810
1811 /* building an inode */
1812
1813 void task_dump_owner(struct task_struct *task, umode_t mode,
1814                      kuid_t *ruid, kgid_t *rgid)
1815 {
1816         /* Depending on the state of dumpable compute who should own a
1817          * proc file for a task.
1818          */
1819         const struct cred *cred;
1820         kuid_t uid;
1821         kgid_t gid;
1822
1823         if (unlikely(task->flags & PF_KTHREAD)) {
1824                 *ruid = GLOBAL_ROOT_UID;
1825                 *rgid = GLOBAL_ROOT_GID;
1826                 return;
1827         }
1828
1829         /* Default to the tasks effective ownership */
1830         rcu_read_lock();
1831         cred = __task_cred(task);
1832         uid = cred->euid;
1833         gid = cred->egid;
1834         rcu_read_unlock();
1835
1836         /*
1837          * Before the /proc/pid/status file was created the only way to read
1838          * the effective uid of a /process was to stat /proc/pid.  Reading
1839          * /proc/pid/status is slow enough that procps and other packages
1840          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1841          * made this apply to all per process world readable and executable
1842          * directories.
1843          */
1844         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1845                 struct mm_struct *mm;
1846                 task_lock(task);
1847                 mm = task->mm;
1848                 /* Make non-dumpable tasks owned by some root */
1849                 if (mm) {
1850                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1851                                 struct user_namespace *user_ns = mm->user_ns;
1852
1853                                 uid = make_kuid(user_ns, 0);
1854                                 if (!uid_valid(uid))
1855                                         uid = GLOBAL_ROOT_UID;
1856
1857                                 gid = make_kgid(user_ns, 0);
1858                                 if (!gid_valid(gid))
1859                                         gid = GLOBAL_ROOT_GID;
1860                         }
1861                 } else {
1862                         uid = GLOBAL_ROOT_UID;
1863                         gid = GLOBAL_ROOT_GID;
1864                 }
1865                 task_unlock(task);
1866         }
1867         *ruid = uid;
1868         *rgid = gid;
1869 }
1870
1871 void proc_pid_evict_inode(struct proc_inode *ei)
1872 {
1873         struct pid *pid = ei->pid;
1874
1875         if (S_ISDIR(ei->vfs_inode.i_mode)) {
1876                 spin_lock(&pid->lock);
1877                 hlist_del_init_rcu(&ei->sibling_inodes);
1878                 spin_unlock(&pid->lock);
1879         }
1880
1881         put_pid(pid);
1882 }
1883
1884 struct inode *proc_pid_make_inode(struct super_block * sb,
1885                                   struct task_struct *task, umode_t mode)
1886 {
1887         struct inode * inode;
1888         struct proc_inode *ei;
1889         struct pid *pid;
1890
1891         /* We need a new inode */
1892
1893         inode = new_inode(sb);
1894         if (!inode)
1895                 goto out;
1896
1897         /* Common stuff */
1898         ei = PROC_I(inode);
1899         inode->i_mode = mode;
1900         inode->i_ino = get_next_ino();
1901         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1902         inode->i_op = &proc_def_inode_operations;
1903
1904         /*
1905          * grab the reference to task.
1906          */
1907         pid = get_task_pid(task, PIDTYPE_PID);
1908         if (!pid)
1909                 goto out_unlock;
1910
1911         /* Let the pid remember us for quick removal */
1912         ei->pid = pid;
1913         if (S_ISDIR(mode)) {
1914                 spin_lock(&pid->lock);
1915                 hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
1916                 spin_unlock(&pid->lock);
1917         }
1918
1919         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1920         security_task_to_inode(task, inode);
1921
1922 out:
1923         return inode;
1924
1925 out_unlock:
1926         iput(inode);
1927         return NULL;
1928 }
1929
1930 int pid_getattr(const struct path *path, struct kstat *stat,
1931                 u32 request_mask, unsigned int query_flags)
1932 {
1933         struct inode *inode = d_inode(path->dentry);
1934         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
1935         struct task_struct *task;
1936
1937         generic_fillattr(inode, stat);
1938
1939         stat->uid = GLOBAL_ROOT_UID;
1940         stat->gid = GLOBAL_ROOT_GID;
1941         rcu_read_lock();
1942         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1943         if (task) {
1944                 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
1945                         rcu_read_unlock();
1946                         /*
1947                          * This doesn't prevent learning whether PID exists,
1948                          * it only makes getattr() consistent with readdir().
1949                          */
1950                         return -ENOENT;
1951                 }
1952                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1953         }
1954         rcu_read_unlock();
1955         return 0;
1956 }
1957
1958 /* dentry stuff */
1959
1960 /*
1961  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1962  */
1963 void pid_update_inode(struct task_struct *task, struct inode *inode)
1964 {
1965         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1966
1967         inode->i_mode &= ~(S_ISUID | S_ISGID);
1968         security_task_to_inode(task, inode);
1969 }
1970
1971 /*
1972  * Rewrite the inode's ownerships here because the owning task may have
1973  * performed a setuid(), etc.
1974  *
1975  */
1976 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1977 {
1978         struct inode *inode;
1979         struct task_struct *task;
1980
1981         if (flags & LOOKUP_RCU)
1982                 return -ECHILD;
1983
1984         inode = d_inode(dentry);
1985         task = get_proc_task(inode);
1986
1987         if (task) {
1988                 pid_update_inode(task, inode);
1989                 put_task_struct(task);
1990                 return 1;
1991         }
1992         return 0;
1993 }
1994
1995 static inline bool proc_inode_is_dead(struct inode *inode)
1996 {
1997         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1998 }
1999
2000 int pid_delete_dentry(const struct dentry *dentry)
2001 {
2002         /* Is the task we represent dead?
2003          * If so, then don't put the dentry on the lru list,
2004          * kill it immediately.
2005          */
2006         return proc_inode_is_dead(d_inode(dentry));
2007 }
2008
2009 const struct dentry_operations pid_dentry_operations =
2010 {
2011         .d_revalidate   = pid_revalidate,
2012         .d_delete       = pid_delete_dentry,
2013 };
2014
2015 /* Lookups */
2016
2017 /*
2018  * Fill a directory entry.
2019  *
2020  * If possible create the dcache entry and derive our inode number and
2021  * file type from dcache entry.
2022  *
2023  * Since all of the proc inode numbers are dynamically generated, the inode
2024  * numbers do not exist until the inode is cache.  This means creating
2025  * the dcache entry in readdir is necessary to keep the inode numbers
2026  * reported by readdir in sync with the inode numbers reported
2027  * by stat.
2028  */
2029 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2030         const char *name, unsigned int len,
2031         instantiate_t instantiate, struct task_struct *task, const void *ptr)
2032 {
2033         struct dentry *child, *dir = file->f_path.dentry;
2034         struct qstr qname = QSTR_INIT(name, len);
2035         struct inode *inode;
2036         unsigned type = DT_UNKNOWN;
2037         ino_t ino = 1;
2038
2039         child = d_hash_and_lookup(dir, &qname);
2040         if (!child) {
2041                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2042                 child = d_alloc_parallel(dir, &qname, &wq);
2043                 if (IS_ERR(child))
2044                         goto end_instantiate;
2045                 if (d_in_lookup(child)) {
2046                         struct dentry *res;
2047                         res = instantiate(child, task, ptr);
2048                         d_lookup_done(child);
2049                         if (unlikely(res)) {
2050                                 dput(child);
2051                                 child = res;
2052                                 if (IS_ERR(child))
2053                                         goto end_instantiate;
2054                         }
2055                 }
2056         }
2057         inode = d_inode(child);
2058         ino = inode->i_ino;
2059         type = inode->i_mode >> 12;
2060         dput(child);
2061 end_instantiate:
2062         return dir_emit(ctx, name, len, ino, type);
2063 }
2064
2065 /*
2066  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2067  * which represent vma start and end addresses.
2068  */
2069 static int dname_to_vma_addr(struct dentry *dentry,
2070                              unsigned long *start, unsigned long *end)
2071 {
2072         const char *str = dentry->d_name.name;
2073         unsigned long long sval, eval;
2074         unsigned int len;
2075
2076         if (str[0] == '0' && str[1] != '-')
2077                 return -EINVAL;
2078         len = _parse_integer(str, 16, &sval);
2079         if (len & KSTRTOX_OVERFLOW)
2080                 return -EINVAL;
2081         if (sval != (unsigned long)sval)
2082                 return -EINVAL;
2083         str += len;
2084
2085         if (*str != '-')
2086                 return -EINVAL;
2087         str++;
2088
2089         if (str[0] == '0' && str[1])
2090                 return -EINVAL;
2091         len = _parse_integer(str, 16, &eval);
2092         if (len & KSTRTOX_OVERFLOW)
2093                 return -EINVAL;
2094         if (eval != (unsigned long)eval)
2095                 return -EINVAL;
2096         str += len;
2097
2098         if (*str != '\0')
2099                 return -EINVAL;
2100
2101         *start = sval;
2102         *end = eval;
2103
2104         return 0;
2105 }
2106
2107 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2108 {
2109         unsigned long vm_start, vm_end;
2110         bool exact_vma_exists = false;
2111         struct mm_struct *mm = NULL;
2112         struct task_struct *task;
2113         struct inode *inode;
2114         int status = 0;
2115
2116         if (flags & LOOKUP_RCU)
2117                 return -ECHILD;
2118
2119         inode = d_inode(dentry);
2120         task = get_proc_task(inode);
2121         if (!task)
2122                 goto out_notask;
2123
2124         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2125         if (IS_ERR_OR_NULL(mm))
2126                 goto out;
2127
2128         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2129                 status = mmap_read_lock_killable(mm);
2130                 if (!status) {
2131                         exact_vma_exists = !!find_exact_vma(mm, vm_start,
2132                                                             vm_end);
2133                         mmap_read_unlock(mm);
2134                 }
2135         }
2136
2137         mmput(mm);
2138
2139         if (exact_vma_exists) {
2140                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2141
2142                 security_task_to_inode(task, inode);
2143                 status = 1;
2144         }
2145
2146 out:
2147         put_task_struct(task);
2148
2149 out_notask:
2150         return status;
2151 }
2152
2153 static const struct dentry_operations tid_map_files_dentry_operations = {
2154         .d_revalidate   = map_files_d_revalidate,
2155         .d_delete       = pid_delete_dentry,
2156 };
2157
2158 static int map_files_get_link(struct dentry *dentry, struct path *path)
2159 {
2160         unsigned long vm_start, vm_end;
2161         struct vm_area_struct *vma;
2162         struct task_struct *task;
2163         struct mm_struct *mm;
2164         int rc;
2165
2166         rc = -ENOENT;
2167         task = get_proc_task(d_inode(dentry));
2168         if (!task)
2169                 goto out;
2170
2171         mm = get_task_mm(task);
2172         put_task_struct(task);
2173         if (!mm)
2174                 goto out;
2175
2176         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2177         if (rc)
2178                 goto out_mmput;
2179
2180         rc = mmap_read_lock_killable(mm);
2181         if (rc)
2182                 goto out_mmput;
2183
2184         rc = -ENOENT;
2185         vma = find_exact_vma(mm, vm_start, vm_end);
2186         if (vma && vma->vm_file) {
2187                 *path = vma->vm_file->f_path;
2188                 path_get(path);
2189                 rc = 0;
2190         }
2191         mmap_read_unlock(mm);
2192
2193 out_mmput:
2194         mmput(mm);
2195 out:
2196         return rc;
2197 }
2198
2199 struct map_files_info {
2200         unsigned long   start;
2201         unsigned long   end;
2202         fmode_t         mode;
2203 };
2204
2205 /*
2206  * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2207  * to concerns about how the symlinks may be used to bypass permissions on
2208  * ancestor directories in the path to the file in question.
2209  */
2210 static const char *
2211 proc_map_files_get_link(struct dentry *dentry,
2212                         struct inode *inode,
2213                         struct delayed_call *done)
2214 {
2215         if (!checkpoint_restore_ns_capable(&init_user_ns))
2216                 return ERR_PTR(-EPERM);
2217
2218         return proc_pid_get_link(dentry, inode, done);
2219 }
2220
2221 /*
2222  * Identical to proc_pid_link_inode_operations except for get_link()
2223  */
2224 static const struct inode_operations proc_map_files_link_inode_operations = {
2225         .readlink       = proc_pid_readlink,
2226         .get_link       = proc_map_files_get_link,
2227         .setattr        = proc_setattr,
2228 };
2229
2230 static struct dentry *
2231 proc_map_files_instantiate(struct dentry *dentry,
2232                            struct task_struct *task, const void *ptr)
2233 {
2234         fmode_t mode = (fmode_t)(unsigned long)ptr;
2235         struct proc_inode *ei;
2236         struct inode *inode;
2237
2238         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2239                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2240                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2241         if (!inode)
2242                 return ERR_PTR(-ENOENT);
2243
2244         ei = PROC_I(inode);
2245         ei->op.proc_get_link = map_files_get_link;
2246
2247         inode->i_op = &proc_map_files_link_inode_operations;
2248         inode->i_size = 64;
2249
2250         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2251         return d_splice_alias(inode, dentry);
2252 }
2253
2254 static struct dentry *proc_map_files_lookup(struct inode *dir,
2255                 struct dentry *dentry, unsigned int flags)
2256 {
2257         unsigned long vm_start, vm_end;
2258         struct vm_area_struct *vma;
2259         struct task_struct *task;
2260         struct dentry *result;
2261         struct mm_struct *mm;
2262
2263         result = ERR_PTR(-ENOENT);
2264         task = get_proc_task(dir);
2265         if (!task)
2266                 goto out;
2267
2268         result = ERR_PTR(-EACCES);
2269         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2270                 goto out_put_task;
2271
2272         result = ERR_PTR(-ENOENT);
2273         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2274                 goto out_put_task;
2275
2276         mm = get_task_mm(task);
2277         if (!mm)
2278                 goto out_put_task;
2279
2280         result = ERR_PTR(-EINTR);
2281         if (mmap_read_lock_killable(mm))
2282                 goto out_put_mm;
2283
2284         result = ERR_PTR(-ENOENT);
2285         vma = find_exact_vma(mm, vm_start, vm_end);
2286         if (!vma)
2287                 goto out_no_vma;
2288
2289         if (vma->vm_file)
2290                 result = proc_map_files_instantiate(dentry, task,
2291                                 (void *)(unsigned long)vma->vm_file->f_mode);
2292
2293 out_no_vma:
2294         mmap_read_unlock(mm);
2295 out_put_mm:
2296         mmput(mm);
2297 out_put_task:
2298         put_task_struct(task);
2299 out:
2300         return result;
2301 }
2302
2303 static const struct inode_operations proc_map_files_inode_operations = {
2304         .lookup         = proc_map_files_lookup,
2305         .permission     = proc_fd_permission,
2306         .setattr        = proc_setattr,
2307 };
2308
2309 static int
2310 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2311 {
2312         struct vm_area_struct *vma;
2313         struct task_struct *task;
2314         struct mm_struct *mm;
2315         unsigned long nr_files, pos, i;
2316         GENRADIX(struct map_files_info) fa;
2317         struct map_files_info *p;
2318         int ret;
2319
2320         genradix_init(&fa);
2321
2322         ret = -ENOENT;
2323         task = get_proc_task(file_inode(file));
2324         if (!task)
2325                 goto out;
2326
2327         ret = -EACCES;
2328         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2329                 goto out_put_task;
2330
2331         ret = 0;
2332         if (!dir_emit_dots(file, ctx))
2333                 goto out_put_task;
2334
2335         mm = get_task_mm(task);
2336         if (!mm)
2337                 goto out_put_task;
2338
2339         ret = mmap_read_lock_killable(mm);
2340         if (ret) {
2341                 mmput(mm);
2342                 goto out_put_task;
2343         }
2344
2345         nr_files = 0;
2346
2347         /*
2348          * We need two passes here:
2349          *
2350          *  1) Collect vmas of mapped files with mmap_lock taken
2351          *  2) Release mmap_lock and instantiate entries
2352          *
2353          * otherwise we get lockdep complained, since filldir()
2354          * routine might require mmap_lock taken in might_fault().
2355          */
2356
2357         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2358                 if (!vma->vm_file)
2359                         continue;
2360                 if (++pos <= ctx->pos)
2361                         continue;
2362
2363                 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2364                 if (!p) {
2365                         ret = -ENOMEM;
2366                         mmap_read_unlock(mm);
2367                         mmput(mm);
2368                         goto out_put_task;
2369                 }
2370
2371                 p->start = vma->vm_start;
2372                 p->end = vma->vm_end;
2373                 p->mode = vma->vm_file->f_mode;
2374         }
2375         mmap_read_unlock(mm);
2376         mmput(mm);
2377
2378         for (i = 0; i < nr_files; i++) {
2379                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2380                 unsigned int len;
2381
2382                 p = genradix_ptr(&fa, i);
2383                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2384                 if (!proc_fill_cache(file, ctx,
2385                                       buf, len,
2386                                       proc_map_files_instantiate,
2387                                       task,
2388                                       (void *)(unsigned long)p->mode))
2389                         break;
2390                 ctx->pos++;
2391         }
2392
2393 out_put_task:
2394         put_task_struct(task);
2395 out:
2396         genradix_free(&fa);
2397         return ret;
2398 }
2399
2400 static const struct file_operations proc_map_files_operations = {
2401         .read           = generic_read_dir,
2402         .iterate_shared = proc_map_files_readdir,
2403         .llseek         = generic_file_llseek,
2404 };
2405
2406 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2407 struct timers_private {
2408         struct pid *pid;
2409         struct task_struct *task;
2410         struct sighand_struct *sighand;
2411         struct pid_namespace *ns;
2412         unsigned long flags;
2413 };
2414
2415 static void *timers_start(struct seq_file *m, loff_t *pos)
2416 {
2417         struct timers_private *tp = m->private;
2418
2419         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2420         if (!tp->task)
2421                 return ERR_PTR(-ESRCH);
2422
2423         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2424         if (!tp->sighand)
2425                 return ERR_PTR(-ESRCH);
2426
2427         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2428 }
2429
2430 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2431 {
2432         struct timers_private *tp = m->private;
2433         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2434 }
2435
2436 static void timers_stop(struct seq_file *m, void *v)
2437 {
2438         struct timers_private *tp = m->private;
2439
2440         if (tp->sighand) {
2441                 unlock_task_sighand(tp->task, &tp->flags);
2442                 tp->sighand = NULL;
2443         }
2444
2445         if (tp->task) {
2446                 put_task_struct(tp->task);
2447                 tp->task = NULL;
2448         }
2449 }
2450
2451 static int show_timer(struct seq_file *m, void *v)
2452 {
2453         struct k_itimer *timer;
2454         struct timers_private *tp = m->private;
2455         int notify;
2456         static const char * const nstr[] = {
2457                 [SIGEV_SIGNAL] = "signal",
2458                 [SIGEV_NONE] = "none",
2459                 [SIGEV_THREAD] = "thread",
2460         };
2461
2462         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2463         notify = timer->it_sigev_notify;
2464
2465         seq_printf(m, "ID: %d\n", timer->it_id);
2466         seq_printf(m, "signal: %d/%px\n",
2467                    timer->sigq->info.si_signo,
2468                    timer->sigq->info.si_value.sival_ptr);
2469         seq_printf(m, "notify: %s/%s.%d\n",
2470                    nstr[notify & ~SIGEV_THREAD_ID],
2471                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2472                    pid_nr_ns(timer->it_pid, tp->ns));
2473         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2474
2475         return 0;
2476 }
2477
2478 static const struct seq_operations proc_timers_seq_ops = {
2479         .start  = timers_start,
2480         .next   = timers_next,
2481         .stop   = timers_stop,
2482         .show   = show_timer,
2483 };
2484
2485 static int proc_timers_open(struct inode *inode, struct file *file)
2486 {
2487         struct timers_private *tp;
2488
2489         tp = __seq_open_private(file, &proc_timers_seq_ops,
2490                         sizeof(struct timers_private));
2491         if (!tp)
2492                 return -ENOMEM;
2493
2494         tp->pid = proc_pid(inode);
2495         tp->ns = proc_pid_ns(inode->i_sb);
2496         return 0;
2497 }
2498
2499 static const struct file_operations proc_timers_operations = {
2500         .open           = proc_timers_open,
2501         .read           = seq_read,
2502         .llseek         = seq_lseek,
2503         .release        = seq_release_private,
2504 };
2505 #endif
2506
2507 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2508                                         size_t count, loff_t *offset)
2509 {
2510         struct inode *inode = file_inode(file);
2511         struct task_struct *p;
2512         u64 slack_ns;
2513         int err;
2514
2515         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2516         if (err < 0)
2517                 return err;
2518
2519         p = get_proc_task(inode);
2520         if (!p)
2521                 return -ESRCH;
2522
2523         if (p != current) {
2524                 rcu_read_lock();
2525                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2526                         rcu_read_unlock();
2527                         count = -EPERM;
2528                         goto out;
2529                 }
2530                 rcu_read_unlock();
2531
2532                 err = security_task_setscheduler(p);
2533                 if (err) {
2534                         count = err;
2535                         goto out;
2536                 }
2537         }
2538
2539         task_lock(p);
2540         if (slack_ns == 0)
2541                 p->timer_slack_ns = p->default_timer_slack_ns;
2542         else
2543                 p->timer_slack_ns = slack_ns;
2544         task_unlock(p);
2545
2546 out:
2547         put_task_struct(p);
2548
2549         return count;
2550 }
2551
2552 static int timerslack_ns_show(struct seq_file *m, void *v)
2553 {
2554         struct inode *inode = m->private;
2555         struct task_struct *p;
2556         int err = 0;
2557
2558         p = get_proc_task(inode);
2559         if (!p)
2560                 return -ESRCH;
2561
2562         if (p != current) {
2563                 rcu_read_lock();
2564                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2565                         rcu_read_unlock();
2566                         err = -EPERM;
2567                         goto out;
2568                 }
2569                 rcu_read_unlock();
2570
2571                 err = security_task_getscheduler(p);
2572                 if (err)
2573                         goto out;
2574         }
2575
2576         task_lock(p);
2577         seq_printf(m, "%llu\n", p->timer_slack_ns);
2578         task_unlock(p);
2579
2580 out:
2581         put_task_struct(p);
2582
2583         return err;
2584 }
2585
2586 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2587 {
2588         return single_open(filp, timerslack_ns_show, inode);
2589 }
2590
2591 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2592         .open           = timerslack_ns_open,
2593         .read           = seq_read,
2594         .write          = timerslack_ns_write,
2595         .llseek         = seq_lseek,
2596         .release        = single_release,
2597 };
2598
2599 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2600         struct task_struct *task, const void *ptr)
2601 {
2602         const struct pid_entry *p = ptr;
2603         struct inode *inode;
2604         struct proc_inode *ei;
2605
2606         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2607         if (!inode)
2608                 return ERR_PTR(-ENOENT);
2609
2610         ei = PROC_I(inode);
2611         if (S_ISDIR(inode->i_mode))
2612                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2613         if (p->iop)
2614                 inode->i_op = p->iop;
2615         if (p->fop)
2616                 inode->i_fop = p->fop;
2617         ei->op = p->op;
2618         pid_update_inode(task, inode);
2619         d_set_d_op(dentry, &pid_dentry_operations);
2620         return d_splice_alias(inode, dentry);
2621 }
2622
2623 static struct dentry *proc_pident_lookup(struct inode *dir,
2624                                          struct dentry *dentry,
2625                                          const struct pid_entry *p,
2626                                          const struct pid_entry *end)
2627 {
2628         struct task_struct *task = get_proc_task(dir);
2629         struct dentry *res = ERR_PTR(-ENOENT);
2630
2631         if (!task)
2632                 goto out_no_task;
2633
2634         /*
2635          * Yes, it does not scale. And it should not. Don't add
2636          * new entries into /proc/<tgid>/ without very good reasons.
2637          */
2638         for (; p < end; p++) {
2639                 if (p->len != dentry->d_name.len)
2640                         continue;
2641                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2642                         res = proc_pident_instantiate(dentry, task, p);
2643                         break;
2644                 }
2645         }
2646         put_task_struct(task);
2647 out_no_task:
2648         return res;
2649 }
2650
2651 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2652                 const struct pid_entry *ents, unsigned int nents)
2653 {
2654         struct task_struct *task = get_proc_task(file_inode(file));
2655         const struct pid_entry *p;
2656
2657         if (!task)
2658                 return -ENOENT;
2659
2660         if (!dir_emit_dots(file, ctx))
2661                 goto out;
2662
2663         if (ctx->pos >= nents + 2)
2664                 goto out;
2665
2666         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2667                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2668                                 proc_pident_instantiate, task, p))
2669                         break;
2670                 ctx->pos++;
2671         }
2672 out:
2673         put_task_struct(task);
2674         return 0;
2675 }
2676
2677 #ifdef CONFIG_SECURITY
2678 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2679                                   size_t count, loff_t *ppos)
2680 {
2681         struct inode * inode = file_inode(file);
2682         char *p = NULL;
2683         ssize_t length;
2684         struct task_struct *task = get_proc_task(inode);
2685
2686         if (!task)
2687                 return -ESRCH;
2688
2689         length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2690                                       (char*)file->f_path.dentry->d_name.name,
2691                                       &p);
2692         put_task_struct(task);
2693         if (length > 0)
2694                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2695         kfree(p);
2696         return length;
2697 }
2698
2699 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2700                                    size_t count, loff_t *ppos)
2701 {
2702         struct inode * inode = file_inode(file);
2703         struct task_struct *task;
2704         void *page;
2705         int rv;
2706
2707         rcu_read_lock();
2708         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2709         if (!task) {
2710                 rcu_read_unlock();
2711                 return -ESRCH;
2712         }
2713         /* A task may only write its own attributes. */
2714         if (current != task) {
2715                 rcu_read_unlock();
2716                 return -EACCES;
2717         }
2718         /* Prevent changes to overridden credentials. */
2719         if (current_cred() != current_real_cred()) {
2720                 rcu_read_unlock();
2721                 return -EBUSY;
2722         }
2723         rcu_read_unlock();
2724
2725         if (count > PAGE_SIZE)
2726                 count = PAGE_SIZE;
2727
2728         /* No partial writes. */
2729         if (*ppos != 0)
2730                 return -EINVAL;
2731
2732         page = memdup_user(buf, count);
2733         if (IS_ERR(page)) {
2734                 rv = PTR_ERR(page);
2735                 goto out;
2736         }
2737
2738         /* Guard against adverse ptrace interaction */
2739         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2740         if (rv < 0)
2741                 goto out_free;
2742
2743         rv = security_setprocattr(PROC_I(inode)->op.lsm,
2744                                   file->f_path.dentry->d_name.name, page,
2745                                   count);
2746         mutex_unlock(&current->signal->cred_guard_mutex);
2747 out_free:
2748         kfree(page);
2749 out:
2750         return rv;
2751 }
2752
2753 static const struct file_operations proc_pid_attr_operations = {
2754         .read           = proc_pid_attr_read,
2755         .write          = proc_pid_attr_write,
2756         .llseek         = generic_file_llseek,
2757 };
2758
2759 #define LSM_DIR_OPS(LSM) \
2760 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2761                              struct dir_context *ctx) \
2762 { \
2763         return proc_pident_readdir(filp, ctx, \
2764                                    LSM##_attr_dir_stuff, \
2765                                    ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2766 } \
2767 \
2768 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2769         .read           = generic_read_dir, \
2770         .iterate        = proc_##LSM##_attr_dir_iterate, \
2771         .llseek         = default_llseek, \
2772 }; \
2773 \
2774 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2775                                 struct dentry *dentry, unsigned int flags) \
2776 { \
2777         return proc_pident_lookup(dir, dentry, \
2778                                   LSM##_attr_dir_stuff, \
2779                                   LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2780 } \
2781 \
2782 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2783         .lookup         = proc_##LSM##_attr_dir_lookup, \
2784         .getattr        = pid_getattr, \
2785         .setattr        = proc_setattr, \
2786 }
2787
2788 #ifdef CONFIG_SECURITY_SMACK
2789 static const struct pid_entry smack_attr_dir_stuff[] = {
2790         ATTR("smack", "current",        0666),
2791 };
2792 LSM_DIR_OPS(smack);
2793 #endif
2794
2795 #ifdef CONFIG_SECURITY_APPARMOR
2796 static const struct pid_entry apparmor_attr_dir_stuff[] = {
2797         ATTR("apparmor", "current",     0666),
2798         ATTR("apparmor", "prev",        0444),
2799         ATTR("apparmor", "exec",        0666),
2800 };
2801 LSM_DIR_OPS(apparmor);
2802 #endif
2803
2804 static const struct pid_entry attr_dir_stuff[] = {
2805         ATTR(NULL, "current",           0666),
2806         ATTR(NULL, "prev",              0444),
2807         ATTR(NULL, "exec",              0666),
2808         ATTR(NULL, "fscreate",          0666),
2809         ATTR(NULL, "keycreate",         0666),
2810         ATTR(NULL, "sockcreate",        0666),
2811 #ifdef CONFIG_SECURITY_SMACK
2812         DIR("smack",                    0555,
2813             proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2814 #endif
2815 #ifdef CONFIG_SECURITY_APPARMOR
2816         DIR("apparmor",                 0555,
2817             proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2818 #endif
2819 };
2820
2821 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2822 {
2823         return proc_pident_readdir(file, ctx,
2824                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2825 }
2826
2827 static const struct file_operations proc_attr_dir_operations = {
2828         .read           = generic_read_dir,
2829         .iterate_shared = proc_attr_dir_readdir,
2830         .llseek         = generic_file_llseek,
2831 };
2832
2833 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2834                                 struct dentry *dentry, unsigned int flags)
2835 {
2836         return proc_pident_lookup(dir, dentry,
2837                                   attr_dir_stuff,
2838                                   attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2839 }
2840
2841 static const struct inode_operations proc_attr_dir_inode_operations = {
2842         .lookup         = proc_attr_dir_lookup,
2843         .getattr        = pid_getattr,
2844         .setattr        = proc_setattr,
2845 };
2846
2847 #endif
2848
2849 #ifdef CONFIG_ELF_CORE
2850 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2851                                          size_t count, loff_t *ppos)
2852 {
2853         struct task_struct *task = get_proc_task(file_inode(file));
2854         struct mm_struct *mm;
2855         char buffer[PROC_NUMBUF];
2856         size_t len;
2857         int ret;
2858
2859         if (!task)
2860                 return -ESRCH;
2861
2862         ret = 0;
2863         mm = get_task_mm(task);
2864         if (mm) {
2865                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2866                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2867                                 MMF_DUMP_FILTER_SHIFT));
2868                 mmput(mm);
2869                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2870         }
2871
2872         put_task_struct(task);
2873
2874         return ret;
2875 }
2876
2877 static ssize_t proc_coredump_filter_write(struct file *file,
2878                                           const char __user *buf,
2879                                           size_t count,
2880                                           loff_t *ppos)
2881 {
2882         struct task_struct *task;
2883         struct mm_struct *mm;
2884         unsigned int val;
2885         int ret;
2886         int i;
2887         unsigned long mask;
2888
2889         ret = kstrtouint_from_user(buf, count, 0, &val);
2890         if (ret < 0)
2891                 return ret;
2892
2893         ret = -ESRCH;
2894         task = get_proc_task(file_inode(file));
2895         if (!task)
2896                 goto out_no_task;
2897
2898         mm = get_task_mm(task);
2899         if (!mm)
2900                 goto out_no_mm;
2901         ret = 0;
2902
2903         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2904                 if (val & mask)
2905                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2906                 else
2907                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2908         }
2909
2910         mmput(mm);
2911  out_no_mm:
2912         put_task_struct(task);
2913  out_no_task:
2914         if (ret < 0)
2915                 return ret;
2916         return count;
2917 }
2918
2919 static const struct file_operations proc_coredump_filter_operations = {
2920         .read           = proc_coredump_filter_read,
2921         .write          = proc_coredump_filter_write,
2922         .llseek         = generic_file_llseek,
2923 };
2924 #endif
2925
2926 #ifdef CONFIG_TASK_IO_ACCOUNTING
2927 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2928 {
2929         struct task_io_accounting acct = task->ioac;
2930         unsigned long flags;
2931         int result;
2932
2933         result = down_read_killable(&task->signal->exec_update_lock);
2934         if (result)
2935                 return result;
2936
2937         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2938                 result = -EACCES;
2939                 goto out_unlock;
2940         }
2941
2942         if (whole && lock_task_sighand(task, &flags)) {
2943                 struct task_struct *t = task;
2944
2945                 task_io_accounting_add(&acct, &task->signal->ioac);
2946                 while_each_thread(task, t)
2947                         task_io_accounting_add(&acct, &t->ioac);
2948
2949                 unlock_task_sighand(task, &flags);
2950         }
2951         seq_printf(m,
2952                    "rchar: %llu\n"
2953                    "wchar: %llu\n"
2954                    "syscr: %llu\n"
2955                    "syscw: %llu\n"
2956                    "read_bytes: %llu\n"
2957                    "write_bytes: %llu\n"
2958                    "cancelled_write_bytes: %llu\n",
2959                    (unsigned long long)acct.rchar,
2960                    (unsigned long long)acct.wchar,
2961                    (unsigned long long)acct.syscr,
2962                    (unsigned long long)acct.syscw,
2963                    (unsigned long long)acct.read_bytes,
2964                    (unsigned long long)acct.write_bytes,
2965                    (unsigned long long)acct.cancelled_write_bytes);
2966         result = 0;
2967
2968 out_unlock:
2969         up_read(&task->signal->exec_update_lock);
2970         return result;
2971 }
2972
2973 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2974                                   struct pid *pid, struct task_struct *task)
2975 {
2976         return do_io_accounting(task, m, 0);
2977 }
2978
2979 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2980                                    struct pid *pid, struct task_struct *task)
2981 {
2982         return do_io_accounting(task, m, 1);
2983 }
2984 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2985
2986 #ifdef CONFIG_USER_NS
2987 static int proc_id_map_open(struct inode *inode, struct file *file,
2988         const struct seq_operations *seq_ops)
2989 {
2990         struct user_namespace *ns = NULL;
2991         struct task_struct *task;
2992         struct seq_file *seq;
2993         int ret = -EINVAL;
2994
2995         task = get_proc_task(inode);
2996         if (task) {
2997                 rcu_read_lock();
2998                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2999                 rcu_read_unlock();
3000                 put_task_struct(task);
3001         }
3002         if (!ns)
3003                 goto err;
3004
3005         ret = seq_open(file, seq_ops);
3006         if (ret)
3007                 goto err_put_ns;
3008
3009         seq = file->private_data;
3010         seq->private = ns;
3011
3012         return 0;
3013 err_put_ns:
3014         put_user_ns(ns);
3015 err:
3016         return ret;
3017 }
3018
3019 static int proc_id_map_release(struct inode *inode, struct file *file)
3020 {
3021         struct seq_file *seq = file->private_data;
3022         struct user_namespace *ns = seq->private;
3023         put_user_ns(ns);
3024         return seq_release(inode, file);
3025 }
3026
3027 static int proc_uid_map_open(struct inode *inode, struct file *file)
3028 {
3029         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3030 }
3031
3032 static int proc_gid_map_open(struct inode *inode, struct file *file)
3033 {
3034         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3035 }
3036
3037 static int proc_projid_map_open(struct inode *inode, struct file *file)
3038 {
3039         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3040 }
3041
3042 static const struct file_operations proc_uid_map_operations = {
3043         .open           = proc_uid_map_open,
3044         .write          = proc_uid_map_write,
3045         .read           = seq_read,
3046         .llseek         = seq_lseek,
3047         .release        = proc_id_map_release,
3048 };
3049
3050 static const struct file_operations proc_gid_map_operations = {
3051         .open           = proc_gid_map_open,
3052         .write          = proc_gid_map_write,
3053         .read           = seq_read,
3054         .llseek         = seq_lseek,
3055         .release        = proc_id_map_release,
3056 };
3057
3058 static const struct file_operations proc_projid_map_operations = {
3059         .open           = proc_projid_map_open,
3060         .write          = proc_projid_map_write,
3061         .read           = seq_read,
3062         .llseek         = seq_lseek,
3063         .release        = proc_id_map_release,
3064 };
3065
3066 static int proc_setgroups_open(struct inode *inode, struct file *file)
3067 {
3068         struct user_namespace *ns = NULL;
3069         struct task_struct *task;
3070         int ret;
3071
3072         ret = -ESRCH;
3073         task = get_proc_task(inode);
3074         if (task) {
3075                 rcu_read_lock();
3076                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3077                 rcu_read_unlock();
3078                 put_task_struct(task);
3079         }
3080         if (!ns)
3081                 goto err;
3082
3083         if (file->f_mode & FMODE_WRITE) {
3084                 ret = -EACCES;
3085                 if (!ns_capable(ns, CAP_SYS_ADMIN))
3086                         goto err_put_ns;
3087         }
3088
3089         ret = single_open(file, &proc_setgroups_show, ns);
3090         if (ret)
3091                 goto err_put_ns;
3092
3093         return 0;
3094 err_put_ns:
3095         put_user_ns(ns);
3096 err:
3097         return ret;
3098 }
3099
3100 static int proc_setgroups_release(struct inode *inode, struct file *file)
3101 {
3102         struct seq_file *seq = file->private_data;
3103         struct user_namespace *ns = seq->private;
3104         int ret = single_release(inode, file);
3105         put_user_ns(ns);
3106         return ret;
3107 }
3108
3109 static const struct file_operations proc_setgroups_operations = {
3110         .open           = proc_setgroups_open,
3111         .write          = proc_setgroups_write,
3112         .read           = seq_read,
3113         .llseek         = seq_lseek,
3114         .release        = proc_setgroups_release,
3115 };
3116 #endif /* CONFIG_USER_NS */
3117
3118 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3119                                 struct pid *pid, struct task_struct *task)
3120 {
3121         int err = lock_trace(task);
3122         if (!err) {
3123                 seq_printf(m, "%08x\n", task->personality);
3124                 unlock_trace(task);
3125         }
3126         return err;
3127 }
3128
3129 #ifdef CONFIG_LIVEPATCH
3130 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3131                                 struct pid *pid, struct task_struct *task)
3132 {
3133         seq_printf(m, "%d\n", task->patch_state);
3134         return 0;
3135 }
3136 #endif /* CONFIG_LIVEPATCH */
3137
3138 #ifdef CONFIG_STACKLEAK_METRICS
3139 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3140                                 struct pid *pid, struct task_struct *task)
3141 {
3142         unsigned long prev_depth = THREAD_SIZE -
3143                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3144         unsigned long depth = THREAD_SIZE -
3145                                 (task->lowest_stack & (THREAD_SIZE - 1));
3146
3147         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3148                                                         prev_depth, depth);
3149         return 0;
3150 }
3151 #endif /* CONFIG_STACKLEAK_METRICS */
3152
3153 /*
3154  * Thread groups
3155  */
3156 static const struct file_operations proc_task_operations;
3157 static const struct inode_operations proc_task_inode_operations;
3158
3159 static const struct pid_entry tgid_base_stuff[] = {
3160         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3161         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3162         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3163         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3164         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3165 #ifdef CONFIG_NET
3166         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3167 #endif
3168         REG("environ",    S_IRUSR, proc_environ_operations),
3169         REG("auxv",       S_IRUSR, proc_auxv_operations),
3170         ONE("status",     S_IRUGO, proc_pid_status),
3171         ONE("personality", S_IRUSR, proc_pid_personality),
3172         ONE("limits",     S_IRUGO, proc_pid_limits),
3173 #ifdef CONFIG_SCHED_DEBUG
3174         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3175 #endif
3176 #ifdef CONFIG_SCHED_AUTOGROUP
3177         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3178 #endif
3179 #ifdef CONFIG_TIME_NS
3180         REG("timens_offsets",  S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3181 #endif
3182         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3183 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3184         ONE("syscall",    S_IRUSR, proc_pid_syscall),
3185 #endif
3186         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3187         ONE("stat",       S_IRUGO, proc_tgid_stat),
3188         ONE("statm",      S_IRUGO, proc_pid_statm),
3189         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3190 #ifdef CONFIG_NUMA
3191         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3192 #endif
3193         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3194         LNK("cwd",        proc_cwd_link),
3195         LNK("root",       proc_root_link),
3196         LNK("exe",        proc_exe_link),
3197         REG("mounts",     S_IRUGO, proc_mounts_operations),
3198         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3199         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3200 #ifdef CONFIG_PROC_PAGE_MONITOR
3201         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3202         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3203         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3204         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3205 #endif
3206 #ifdef CONFIG_SECURITY
3207         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3208 #endif
3209 #ifdef CONFIG_KALLSYMS
3210         ONE("wchan",      S_IRUGO, proc_pid_wchan),
3211 #endif
3212 #ifdef CONFIG_STACKTRACE
3213         ONE("stack",      S_IRUSR, proc_pid_stack),
3214 #endif
3215 #ifdef CONFIG_SCHED_INFO
3216         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3217 #endif
3218 #ifdef CONFIG_LATENCYTOP
3219         REG("latency",  S_IRUGO, proc_lstats_operations),
3220 #endif
3221 #ifdef CONFIG_PROC_PID_CPUSET
3222         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3223 #endif
3224 #ifdef CONFIG_CGROUPS
3225         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3226 #endif
3227 #ifdef CONFIG_PROC_CPU_RESCTRL
3228         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3229 #endif
3230         ONE("oom_score",  S_IRUGO, proc_oom_score),
3231         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3232         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3233 #ifdef CONFIG_AUDIT
3234         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3235         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3236 #endif
3237 #ifdef CONFIG_FAULT_INJECTION
3238         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3239         REG("fail-nth", 0644, proc_fail_nth_operations),
3240 #endif
3241 #ifdef CONFIG_ELF_CORE
3242         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3243 #endif
3244 #ifdef CONFIG_TASK_IO_ACCOUNTING
3245         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3246 #endif
3247 #ifdef CONFIG_USER_NS
3248         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3249         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3250         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3251         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3252 #endif
3253 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3254         REG("timers",     S_IRUGO, proc_timers_operations),
3255 #endif
3256         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3257 #ifdef CONFIG_LIVEPATCH
3258         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3259 #endif
3260 #ifdef CONFIG_STACKLEAK_METRICS
3261         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3262 #endif
3263 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3264         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3265 #endif
3266 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3267         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3268 #endif
3269 };
3270
3271 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3272 {
3273         return proc_pident_readdir(file, ctx,
3274                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3275 }
3276
3277 static const struct file_operations proc_tgid_base_operations = {
3278         .read           = generic_read_dir,
3279         .iterate_shared = proc_tgid_base_readdir,
3280         .llseek         = generic_file_llseek,
3281 };
3282
3283 struct pid *tgid_pidfd_to_pid(const struct file *file)
3284 {
3285         if (file->f_op != &proc_tgid_base_operations)
3286                 return ERR_PTR(-EBADF);
3287
3288         return proc_pid(file_inode(file));
3289 }
3290
3291 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3292 {
3293         return proc_pident_lookup(dir, dentry,
3294                                   tgid_base_stuff,
3295                                   tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3296 }
3297
3298 static const struct inode_operations proc_tgid_base_inode_operations = {
3299         .lookup         = proc_tgid_base_lookup,
3300         .getattr        = pid_getattr,
3301         .setattr        = proc_setattr,
3302         .permission     = proc_pid_permission,
3303 };
3304
3305 /**
3306  * proc_flush_pid -  Remove dcache entries for @pid from the /proc dcache.
3307  * @pid: pid that should be flushed.
3308  *
3309  * This function walks a list of inodes (that belong to any proc
3310  * filesystem) that are attached to the pid and flushes them from
3311  * the dentry cache.
3312  *
3313  * It is safe and reasonable to cache /proc entries for a task until
3314  * that task exits.  After that they just clog up the dcache with
3315  * useless entries, possibly causing useful dcache entries to be
3316  * flushed instead.  This routine is provided to flush those useless
3317  * dcache entries when a process is reaped.
3318  *
3319  * NOTE: This routine is just an optimization so it does not guarantee
3320  *       that no dcache entries will exist after a process is reaped
3321  *       it just makes it very unlikely that any will persist.
3322  */
3323
3324 void proc_flush_pid(struct pid *pid)
3325 {
3326         proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3327 }
3328
3329 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3330                                    struct task_struct *task, const void *ptr)
3331 {
3332         struct inode *inode;
3333
3334         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3335         if (!inode)
3336                 return ERR_PTR(-ENOENT);
3337
3338         inode->i_op = &proc_tgid_base_inode_operations;
3339         inode->i_fop = &proc_tgid_base_operations;
3340         inode->i_flags|=S_IMMUTABLE;
3341
3342         set_nlink(inode, nlink_tgid);
3343         pid_update_inode(task, inode);
3344
3345         d_set_d_op(dentry, &pid_dentry_operations);
3346         return d_splice_alias(inode, dentry);
3347 }
3348
3349 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3350 {
3351         struct task_struct *task;
3352         unsigned tgid;
3353         struct proc_fs_info *fs_info;
3354         struct pid_namespace *ns;
3355         struct dentry *result = ERR_PTR(-ENOENT);
3356
3357         tgid = name_to_int(&dentry->d_name);
3358         if (tgid == ~0U)
3359                 goto out;
3360
3361         fs_info = proc_sb_info(dentry->d_sb);
3362         ns = fs_info->pid_ns;
3363         rcu_read_lock();
3364         task = find_task_by_pid_ns(tgid, ns);
3365         if (task)
3366                 get_task_struct(task);
3367         rcu_read_unlock();
3368         if (!task)
3369                 goto out;
3370
3371         /* Limit procfs to only ptraceable tasks */
3372         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3373                 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3374                         goto out_put_task;
3375         }
3376
3377         result = proc_pid_instantiate(dentry, task, NULL);
3378 out_put_task:
3379         put_task_struct(task);
3380 out:
3381         return result;
3382 }
3383
3384 /*
3385  * Find the first task with tgid >= tgid
3386  *
3387  */
3388 struct tgid_iter {
3389         unsigned int tgid;
3390         struct task_struct *task;
3391 };
3392 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3393 {
3394         struct pid *pid;
3395
3396         if (iter.task)
3397                 put_task_struct(iter.task);
3398         rcu_read_lock();
3399 retry:
3400         iter.task = NULL;
3401         pid = find_ge_pid(iter.tgid, ns);
3402         if (pid) {
3403                 iter.tgid = pid_nr_ns(pid, ns);
3404                 iter.task = pid_task(pid, PIDTYPE_TGID);
3405                 if (!iter.task) {
3406                         iter.tgid += 1;
3407                         goto retry;
3408                 }
3409                 get_task_struct(iter.task);
3410         }
3411         rcu_read_unlock();
3412         return iter;
3413 }
3414
3415 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3416
3417 /* for the /proc/ directory itself, after non-process stuff has been done */
3418 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3419 {
3420         struct tgid_iter iter;
3421         struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3422         struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3423         loff_t pos = ctx->pos;
3424
3425         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3426                 return 0;
3427
3428         if (pos == TGID_OFFSET - 2) {
3429                 struct inode *inode = d_inode(fs_info->proc_self);
3430                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3431                         return 0;
3432                 ctx->pos = pos = pos + 1;
3433         }
3434         if (pos == TGID_OFFSET - 1) {
3435                 struct inode *inode = d_inode(fs_info->proc_thread_self);
3436                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3437                         return 0;
3438                 ctx->pos = pos = pos + 1;
3439         }
3440         iter.tgid = pos - TGID_OFFSET;
3441         iter.task = NULL;
3442         for (iter = next_tgid(ns, iter);
3443              iter.task;
3444              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3445                 char name[10 + 1];
3446                 unsigned int len;
3447
3448                 cond_resched();
3449                 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3450                         continue;
3451
3452                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3453                 ctx->pos = iter.tgid + TGID_OFFSET;
3454                 if (!proc_fill_cache(file, ctx, name, len,
3455                                      proc_pid_instantiate, iter.task, NULL)) {
3456                         put_task_struct(iter.task);
3457                         return 0;
3458                 }
3459         }
3460         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3461         return 0;
3462 }
3463
3464 /*
3465  * proc_tid_comm_permission is a special permission function exclusively
3466  * used for the node /proc/<pid>/task/<tid>/comm.
3467  * It bypasses generic permission checks in the case where a task of the same
3468  * task group attempts to access the node.
3469  * The rationale behind this is that glibc and bionic access this node for
3470  * cross thread naming (pthread_set/getname_np(!self)). However, if
3471  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3472  * which locks out the cross thread naming implementation.
3473  * This function makes sure that the node is always accessible for members of
3474  * same thread group.
3475  */
3476 static int proc_tid_comm_permission(struct inode *inode, int mask)
3477 {
3478         bool is_same_tgroup;
3479         struct task_struct *task;
3480
3481         task = get_proc_task(inode);
3482         if (!task)
3483                 return -ESRCH;
3484         is_same_tgroup = same_thread_group(current, task);
3485         put_task_struct(task);
3486
3487         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3488                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3489                  * read or written by the members of the corresponding
3490                  * thread group.
3491                  */
3492                 return 0;
3493         }
3494
3495         return generic_permission(inode, mask);
3496 }
3497
3498 static const struct inode_operations proc_tid_comm_inode_operations = {
3499                 .permission = proc_tid_comm_permission,
3500 };
3501
3502 /*
3503  * Tasks
3504  */
3505 static const struct pid_entry tid_base_stuff[] = {
3506         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3507         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3508         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3509 #ifdef CONFIG_NET
3510         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3511 #endif
3512         REG("environ",   S_IRUSR, proc_environ_operations),
3513         REG("auxv",      S_IRUSR, proc_auxv_operations),
3514         ONE("status",    S_IRUGO, proc_pid_status),
3515         ONE("personality", S_IRUSR, proc_pid_personality),
3516         ONE("limits",    S_IRUGO, proc_pid_limits),
3517 #ifdef CONFIG_SCHED_DEBUG
3518         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3519 #endif
3520         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3521                          &proc_tid_comm_inode_operations,
3522                          &proc_pid_set_comm_operations, {}),
3523 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3524         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3525 #endif
3526         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3527         ONE("stat",      S_IRUGO, proc_tid_stat),
3528         ONE("statm",     S_IRUGO, proc_pid_statm),
3529         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3530 #ifdef CONFIG_PROC_CHILDREN
3531         REG("children",  S_IRUGO, proc_tid_children_operations),
3532 #endif
3533 #ifdef CONFIG_NUMA
3534         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3535 #endif
3536         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3537         LNK("cwd",       proc_cwd_link),
3538         LNK("root",      proc_root_link),
3539         LNK("exe",       proc_exe_link),
3540         REG("mounts",    S_IRUGO, proc_mounts_operations),
3541         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3542 #ifdef CONFIG_PROC_PAGE_MONITOR
3543         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3544         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3545         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3546         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3547 #endif
3548 #ifdef CONFIG_SECURITY
3549         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3550 #endif
3551 #ifdef CONFIG_KALLSYMS
3552         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3553 #endif
3554 #ifdef CONFIG_STACKTRACE
3555         ONE("stack",      S_IRUSR, proc_pid_stack),
3556 #endif
3557 #ifdef CONFIG_SCHED_INFO
3558         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3559 #endif
3560 #ifdef CONFIG_LATENCYTOP
3561         REG("latency",  S_IRUGO, proc_lstats_operations),
3562 #endif
3563 #ifdef CONFIG_PROC_PID_CPUSET
3564         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3565 #endif
3566 #ifdef CONFIG_CGROUPS
3567         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3568 #endif
3569 #ifdef CONFIG_PROC_CPU_RESCTRL
3570         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3571 #endif
3572         ONE("oom_score", S_IRUGO, proc_oom_score),
3573         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3574         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3575 #ifdef CONFIG_AUDIT
3576         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3577         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3578 #endif
3579 #ifdef CONFIG_FAULT_INJECTION
3580         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3581         REG("fail-nth", 0644, proc_fail_nth_operations),
3582 #endif
3583 #ifdef CONFIG_TASK_IO_ACCOUNTING
3584         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3585 #endif
3586 #ifdef CONFIG_USER_NS
3587         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3588         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3589         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3590         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3591 #endif
3592 #ifdef CONFIG_LIVEPATCH
3593         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3594 #endif
3595 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3596         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3597 #endif
3598 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3599         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3600 #endif
3601 };
3602
3603 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3604 {
3605         return proc_pident_readdir(file, ctx,
3606                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3607 }
3608
3609 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3610 {
3611         return proc_pident_lookup(dir, dentry,
3612                                   tid_base_stuff,
3613                                   tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3614 }
3615
3616 static const struct file_operations proc_tid_base_operations = {
3617         .read           = generic_read_dir,
3618         .iterate_shared = proc_tid_base_readdir,
3619         .llseek         = generic_file_llseek,
3620 };
3621
3622 static const struct inode_operations proc_tid_base_inode_operations = {
3623         .lookup         = proc_tid_base_lookup,
3624         .getattr        = pid_getattr,
3625         .setattr        = proc_setattr,
3626 };
3627
3628 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3629         struct task_struct *task, const void *ptr)
3630 {
3631         struct inode *inode;
3632         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3633         if (!inode)
3634                 return ERR_PTR(-ENOENT);
3635
3636         inode->i_op = &proc_tid_base_inode_operations;
3637         inode->i_fop = &proc_tid_base_operations;
3638         inode->i_flags |= S_IMMUTABLE;
3639
3640         set_nlink(inode, nlink_tid);
3641         pid_update_inode(task, inode);
3642
3643         d_set_d_op(dentry, &pid_dentry_operations);
3644         return d_splice_alias(inode, dentry);
3645 }
3646
3647 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3648 {
3649         struct task_struct *task;
3650         struct task_struct *leader = get_proc_task(dir);
3651         unsigned tid;
3652         struct proc_fs_info *fs_info;
3653         struct pid_namespace *ns;
3654         struct dentry *result = ERR_PTR(-ENOENT);
3655
3656         if (!leader)
3657                 goto out_no_task;
3658
3659         tid = name_to_int(&dentry->d_name);
3660         if (tid == ~0U)
3661                 goto out;
3662
3663         fs_info = proc_sb_info(dentry->d_sb);
3664         ns = fs_info->pid_ns;
3665         rcu_read_lock();
3666         task = find_task_by_pid_ns(tid, ns);
3667         if (task)
3668                 get_task_struct(task);
3669         rcu_read_unlock();
3670         if (!task)
3671                 goto out;
3672         if (!same_thread_group(leader, task))
3673                 goto out_drop_task;
3674
3675         result = proc_task_instantiate(dentry, task, NULL);
3676 out_drop_task:
3677         put_task_struct(task);
3678 out:
3679         put_task_struct(leader);
3680 out_no_task:
3681         return result;
3682 }
3683
3684 /*
3685  * Find the first tid of a thread group to return to user space.
3686  *
3687  * Usually this is just the thread group leader, but if the users
3688  * buffer was too small or there was a seek into the middle of the
3689  * directory we have more work todo.
3690  *
3691  * In the case of a short read we start with find_task_by_pid.
3692  *
3693  * In the case of a seek we start with the leader and walk nr
3694  * threads past it.
3695  */
3696 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3697                                         struct pid_namespace *ns)
3698 {
3699         struct task_struct *pos, *task;
3700         unsigned long nr = f_pos;
3701
3702         if (nr != f_pos)        /* 32bit overflow? */
3703                 return NULL;
3704
3705         rcu_read_lock();
3706         task = pid_task(pid, PIDTYPE_PID);
3707         if (!task)
3708                 goto fail;
3709
3710         /* Attempt to start with the tid of a thread */
3711         if (tid && nr) {
3712                 pos = find_task_by_pid_ns(tid, ns);
3713                 if (pos && same_thread_group(pos, task))
3714                         goto found;
3715         }
3716
3717         /* If nr exceeds the number of threads there is nothing todo */
3718         if (nr >= get_nr_threads(task))
3719                 goto fail;
3720
3721         /* If we haven't found our starting place yet start
3722          * with the leader and walk nr threads forward.
3723          */
3724         pos = task = task->group_leader;
3725         do {
3726                 if (!nr--)
3727                         goto found;
3728         } while_each_thread(task, pos);
3729 fail:
3730         pos = NULL;
3731         goto out;
3732 found:
3733         get_task_struct(pos);
3734 out:
3735         rcu_read_unlock();
3736         return pos;
3737 }
3738
3739 /*
3740  * Find the next thread in the thread list.
3741  * Return NULL if there is an error or no next thread.
3742  *
3743  * The reference to the input task_struct is released.
3744  */
3745 static struct task_struct *next_tid(struct task_struct *start)
3746 {
3747         struct task_struct *pos = NULL;
3748         rcu_read_lock();
3749         if (pid_alive(start)) {
3750                 pos = next_thread(start);
3751                 if (thread_group_leader(pos))
3752                         pos = NULL;
3753                 else
3754                         get_task_struct(pos);
3755         }
3756         rcu_read_unlock();
3757         put_task_struct(start);
3758         return pos;
3759 }
3760
3761 /* for the /proc/TGID/task/ directories */
3762 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3763 {
3764         struct inode *inode = file_inode(file);
3765         struct task_struct *task;
3766         struct pid_namespace *ns;
3767         int tid;
3768
3769         if (proc_inode_is_dead(inode))
3770                 return -ENOENT;
3771
3772         if (!dir_emit_dots(file, ctx))
3773                 return 0;
3774
3775         /* f_version caches the tgid value that the last readdir call couldn't
3776          * return. lseek aka telldir automagically resets f_version to 0.
3777          */
3778         ns = proc_pid_ns(inode->i_sb);
3779         tid = (int)file->f_version;
3780         file->f_version = 0;
3781         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3782              task;
3783              task = next_tid(task), ctx->pos++) {
3784                 char name[10 + 1];
3785                 unsigned int len;
3786                 tid = task_pid_nr_ns(task, ns);
3787                 len = snprintf(name, sizeof(name), "%u", tid);
3788                 if (!proc_fill_cache(file, ctx, name, len,
3789                                 proc_task_instantiate, task, NULL)) {
3790                         /* returning this tgid failed, save it as the first
3791                          * pid for the next readir call */
3792                         file->f_version = (u64)tid;
3793                         put_task_struct(task);
3794                         break;
3795                 }
3796         }
3797
3798         return 0;
3799 }
3800
3801 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3802                              u32 request_mask, unsigned int query_flags)
3803 {
3804         struct inode *inode = d_inode(path->dentry);
3805         struct task_struct *p = get_proc_task(inode);
3806         generic_fillattr(inode, stat);
3807
3808         if (p) {
3809                 stat->nlink += get_nr_threads(p);
3810                 put_task_struct(p);
3811         }
3812
3813         return 0;
3814 }
3815
3816 static const struct inode_operations proc_task_inode_operations = {
3817         .lookup         = proc_task_lookup,
3818         .getattr        = proc_task_getattr,
3819         .setattr        = proc_setattr,
3820         .permission     = proc_pid_permission,
3821 };
3822
3823 static const struct file_operations proc_task_operations = {
3824         .read           = generic_read_dir,
3825         .iterate_shared = proc_task_readdir,
3826         .llseek         = generic_file_llseek,
3827 };
3828
3829 void __init set_proc_pid_nlink(void)
3830 {
3831         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3832         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3833 }