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/generic-radix-tree.h>
  62 #include <linux/string.h>
  63 #include <linux/seq_file.h>
  64 #include <linux/namei.h>
  65 #include <linux/mnt_namespace.h>
  66 #include <linux/mm.h>
  67 #include <linux/swap.h>
  68 #include <linux/rcupdate.h>
  69 #include <linux/kallsyms.h>
  70 #include <linux/stacktrace.h>
  71 #include <linux/resource.h>
  72 #include <linux/module.h>
  73 #include <linux/mount.h>
  74 #include <linux/security.h>
  75 #include <linux/ptrace.h>
  76 #include <linux/printk.h>
  77 #include <linux/cache.h>
  78 #include <linux/cgroup.h>
  79 #include <linux/cpuset.h>
  80 #include <linux/audit.h>
  81 #include <linux/poll.h>
  82 #include <linux/nsproxy.h>
  83 #include <linux/oom.h>
  84 #include <linux/elf.h>
  85 #include <linux/pid_namespace.h>
  86 #include <linux/user_namespace.h>
  87 #include <linux/fs_parser.h>
  88 #include <linux/fs_struct.h>
  89 #include <linux/slab.h>
  90 #include <linux/sched/autogroup.h>
  91 #include <linux/sched/mm.h>
  92 #include <linux/sched/coredump.h>
  93 #include <linux/sched/debug.h>
  94 #include <linux/sched/stat.h>
  95 #include <linux/posix-timers.h>
  96 #include <linux/time_namespace.h>
  97 #include <linux/resctrl.h>
  98 #include <linux/cn_proc.h>
  99 #include <linux/ksm.h>
 100 #include <uapi/linux/lsm.h>
 101 #include <trace/events/oom.h>
 102 #include "internal.h"
 103 #include "fd.h"
 104
 105 #include "../../lib/kstrtox.h"
 106
 107 /* NOTE:
 108  *      Implementing inode permission operations in /proc is almost
 109  *      certainly an error.  Permission checks need to happen during
 110  *      each system call not at open time.  The reason is that most of
 111  *      what we wish to check for permissions in /proc varies at runtime.
 112  *
 113  *      The classic example of a problem is opening file descriptors
 114  *      in /proc for a task before it execs a suid executable.
 115  */
 116
 117 static u8 nlink_tid __ro_after_init;
 118 static u8 nlink_tgid __ro_after_init;
 119
 120 enum proc_mem_force {
 121         PROC_MEM_FORCE_ALWAYS,
 122         PROC_MEM_FORCE_PTRACE,
 123         PROC_MEM_FORCE_NEVER
 124 };
 125
 126 static enum proc_mem_force proc_mem_force_override __ro_after_init =
 127         IS_ENABLED(CONFIG_PROC_MEM_NO_FORCE) ? PROC_MEM_FORCE_NEVER :
 128         IS_ENABLED(CONFIG_PROC_MEM_FORCE_PTRACE) ? PROC_MEM_FORCE_PTRACE :
 129         PROC_MEM_FORCE_ALWAYS;
 130
 131 static const struct constant_table proc_mem_force_table[] __initconst = {
 132         { "always", PROC_MEM_FORCE_ALWAYS },
 133         { "ptrace", PROC_MEM_FORCE_PTRACE },
 134         { "never", PROC_MEM_FORCE_NEVER },
 135         { }
 136 };
 137
 138 static int __init early_proc_mem_force_override(char *buf)
 139 {
 140         if (!buf)
 141                 return -EINVAL;
 142
 143         /*
 144          * lookup_constant() defaults to proc_mem_force_override to preseve
 145          * the initial Kconfig choice in case an invalid param gets passed.
 146          */
 147         proc_mem_force_override = lookup_constant(proc_mem_force_table,
 148                                                   buf, proc_mem_force_override);
 149
 150         return 0;
 151 }
 152 early_param("proc_mem.force_override", early_proc_mem_force_override);
 153
 154 struct pid_entry {
 155         const char *name;
 156         unsigned int len;
 157         umode_t mode;
 158         const struct inode_operations *iop;
 159         const struct file_operations *fop;
 160         union proc_op op;
 161 };
 162
 163 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 164         .name = (NAME),                                 \
 165         .len  = sizeof(NAME) - 1,                       \
 166         .mode = MODE,                                   \
 167         .iop  = IOP,                                    \
 168         .fop  = FOP,                                    \
 169         .op   = OP,                                     \
 170 }
 171
 172 #define DIR(NAME, MODE, iops, fops)     \
 173         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 174 #define LNK(NAME, get_link)                                     \
 175         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 176                 &proc_pid_link_inode_operations, NULL,          \
 177                 { .proc_get_link = get_link } )
 178 #define REG(NAME, MODE, fops)                           \
 179         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 180 #define ONE(NAME, MODE, show)                           \
 181         NOD(NAME, (S_IFREG|(MODE)),                     \
 182                 NULL, &proc_single_file_operations,     \
 183                 { .proc_show = show } )
 184 #define ATTR(LSMID, NAME, MODE)                         \
 185         NOD(NAME, (S_IFREG|(MODE)),                     \
 186                 NULL, &proc_pid_attr_operations,        \
 187                 { .lsmid = LSMID })
 188
 189 /*
 190  * Count the number of hardlinks for the pid_entry table, excluding the .
 191  * and .. links.
 192  */
 193 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 194         unsigned int n)
 195 {
 196         unsigned int i;
 197         unsigned int count;
 198
 199         count = 2;
 200         for (i = 0; i < n; ++i) {
 201                 if (S_ISDIR(entries[i].mode))
 202                         ++count;
 203         }
 204
 205         return count;
 206 }
 207
 208 static int get_task_root(struct task_struct *task, struct path *root)
 209 {
 210         int result = -ENOENT;
 211
 212         task_lock(task);
 213         if (task->fs) {
 214                 get_fs_root(task->fs, root);
 215                 result = 0;
 216         }
 217         task_unlock(task);
 218         return result;
 219 }
 220
 221 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 222 {
 223         struct task_struct *task = get_proc_task(d_inode(dentry));
 224         int result = -ENOENT;
 225
 226         if (task) {
 227                 task_lock(task);
 228                 if (task->fs) {
 229                         get_fs_pwd(task->fs, path);
 230                         result = 0;
 231                 }
 232                 task_unlock(task);
 233                 put_task_struct(task);
 234         }
 235         return result;
 236 }
 237
 238 static int proc_root_link(struct dentry *dentry, struct path *path)
 239 {
 240         struct task_struct *task = get_proc_task(d_inode(dentry));
 241         int result = -ENOENT;
 242
 243         if (task) {
 244                 result = get_task_root(task, path);
 245                 put_task_struct(task);
 246         }
 247         return result;
 248 }
 249
 250 /*
 251  * If the user used setproctitle(), we just get the string from
 252  * user space at arg_start, and limit it to a maximum of one page.
 253  */
 254 static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
 255                                 size_t count, unsigned long pos,
 256                                 unsigned long arg_start)
 257 {
 258         char *page;
 259         int ret, got;
 260
 261         if (pos >= PAGE_SIZE)
 262                 return 0;
 263
 264         page = (char *)__get_free_page(GFP_KERNEL);
 265         if (!page)
 266                 return -ENOMEM;
 267
 268         ret = 0;
 269         got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
 270         if (got > 0) {
 271                 int len = strnlen(page, got);
 272
 273                 /* Include the NUL character if it was found */
 274                 if (len < got)
 275                         len++;
 276
 277                 if (len > pos) {
 278                         len -= pos;
 279                         if (len > count)
 280                                 len = count;
 281                         len -= copy_to_user(buf, page+pos, len);
 282                         if (!len)
 283                                 len = -EFAULT;
 284                         ret = len;
 285                 }
 286         }
 287         free_page((unsigned long)page);
 288         return ret;
 289 }
 290
 291 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 292                               size_t count, loff_t *ppos)
 293 {
 294         unsigned long arg_start, arg_end, env_start, env_end;
 295         unsigned long pos, len;
 296         char *page, c;
 297
 298         /* Check if process spawned far enough to have cmdline. */
 299         if (!mm->env_end)
 300                 return 0;
 301
 302         spin_lock(&mm->arg_lock);
 303         arg_start = mm->arg_start;
 304         arg_end = mm->arg_end;
 305         env_start = mm->env_start;
 306         env_end = mm->env_end;
 307         spin_unlock(&mm->arg_lock);
 308
 309         if (arg_start >= arg_end)
 310                 return 0;
 311
 312         /*
 313          * We allow setproctitle() to overwrite the argument
 314          * strings, and overflow past the original end. But
 315          * only when it overflows into the environment area.
 316          */
 317         if (env_start != arg_end || env_end < env_start)
 318                 env_start = env_end = arg_end;
 319         len = env_end - arg_start;
 320
 321         /* We're not going to care if "*ppos" has high bits set */
 322         pos = *ppos;
 323         if (pos >= len)
 324                 return 0;
 325         if (count > len - pos)
 326                 count = len - pos;
 327         if (!count)
 328                 return 0;
 329
 330         /*
 331          * Magical special case: if the argv[] end byte is not
 332          * zero, the user has overwritten it with setproctitle(3).
 333          *
 334          * Possible future enhancement: do this only once when
 335          * pos is 0, and set a flag in the 'struct file'.
 336          */
 337         if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
 338                 return get_mm_proctitle(mm, buf, count, pos, arg_start);
 339
 340         /*
 341          * For the non-setproctitle() case we limit things strictly
 342          * to the [arg_start, arg_end[ range.
 343          */
 344         pos += arg_start;
 345         if (pos < arg_start || pos >= arg_end)
 346                 return 0;
 347         if (count > arg_end - pos)
 348                 count = arg_end - pos;
 349
 350         page = (char *)__get_free_page(GFP_KERNEL);
 351         if (!page)
 352                 return -ENOMEM;
 353
 354         len = 0;
 355         while (count) {
 356                 int got;
 357                 size_t size = min_t(size_t, PAGE_SIZE, count);
 358
 359                 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
 360                 if (got <= 0)
 361                         break;
 362                 got -= copy_to_user(buf, page, got);
 363                 if (unlikely(!got)) {
 364                         if (!len)
 365                                 len = -EFAULT;
 366                         break;
 367                 }
 368                 pos += got;
 369                 buf += got;
 370                 len += got;
 371                 count -= got;
 372         }
 373
 374         free_page((unsigned long)page);
 375         return len;
 376 }
 377
 378 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 379                                 size_t count, loff_t *pos)
 380 {
 381         struct mm_struct *mm;
 382         ssize_t ret;
 383
 384         mm = get_task_mm(tsk);
 385         if (!mm)
 386                 return 0;
 387
 388         ret = get_mm_cmdline(mm, buf, count, pos);
 389         mmput(mm);
 390         return ret;
 391 }
 392
 393 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 394                                      size_t count, loff_t *pos)
 395 {
 396         struct task_struct *tsk;
 397         ssize_t ret;
 398
 399         BUG_ON(*pos < 0);
 400
 401         tsk = get_proc_task(file_inode(file));
 402         if (!tsk)
 403                 return -ESRCH;
 404         ret = get_task_cmdline(tsk, buf, count, pos);
 405         put_task_struct(tsk);
 406         if (ret > 0)
 407                 *pos += ret;
 408         return ret;
 409 }
 410
 411 static const struct file_operations proc_pid_cmdline_ops = {
 412         .read   = proc_pid_cmdline_read,
 413         .llseek = generic_file_llseek,
 414 };
 415
 416 #ifdef CONFIG_KALLSYMS
 417 /*
 418  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 419  * Returns the resolved symbol.  If that fails, simply return the address.
 420  */
 421 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 422                           struct pid *pid, struct task_struct *task)
 423 {
 424         unsigned long wchan;
 425         char symname[KSYM_NAME_LEN];
 426
 427         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 428                 goto print0;
 429
 430         wchan = get_wchan(task);
 431         if (wchan && !lookup_symbol_name(wchan, symname)) {
 432                 seq_puts(m, symname);
 433                 return 0;
 434         }
 435
 436 print0:
 437         seq_putc(m, '0');
 438         return 0;
 439 }
 440 #endif /* CONFIG_KALLSYMS */
 441
 442 static int lock_trace(struct task_struct *task)
 443 {
 444         int err = down_read_killable(&task->signal->exec_update_lock);
 445         if (err)
 446                 return err;
 447         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 448                 up_read(&task->signal->exec_update_lock);
 449                 return -EPERM;
 450         }
 451         return 0;
 452 }
 453
 454 static void unlock_trace(struct task_struct *task)
 455 {
 456         up_read(&task->signal->exec_update_lock);
 457 }
 458
 459 #ifdef CONFIG_STACKTRACE
 460
 461 #define MAX_STACK_TRACE_DEPTH   64
 462
 463 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 464                           struct pid *pid, struct task_struct *task)
 465 {
 466         unsigned long *entries;
 467         int err;
 468
 469         /*
 470          * The ability to racily run the kernel stack unwinder on a running task
 471          * and then observe the unwinder output is scary; while it is useful for
 472          * debugging kernel issues, it can also allow an attacker to leak kernel
 473          * stack contents.
 474          * Doing this in a manner that is at least safe from races would require
 475          * some work to ensure that the remote task can not be scheduled; and
 476          * even then, this would still expose the unwinder as local attack
 477          * surface.
 478          * Therefore, this interface is restricted to root.
 479          */
 480         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 481                 return -EACCES;
 482
 483         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 484                                 GFP_KERNEL);
 485         if (!entries)
 486                 return -ENOMEM;
 487
 488         err = lock_trace(task);
 489         if (!err) {
 490                 unsigned int i, nr_entries;
 491
 492                 nr_entries = stack_trace_save_tsk(task, entries,
 493                                                   MAX_STACK_TRACE_DEPTH, 0);
 494
 495                 for (i = 0; i < nr_entries; i++) {
 496                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 497                 }
 498
 499                 unlock_trace(task);
 500         }
 501         kfree(entries);
 502
 503         return err;
 504 }
 505 #endif
 506
 507 #ifdef CONFIG_SCHED_INFO
 508 /*
 509  * Provides /proc/PID/schedstat
 510  */
 511 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 512                               struct pid *pid, struct task_struct *task)
 513 {
 514         if (unlikely(!sched_info_on()))
 515                 seq_puts(m, "0 0 0\n");
 516         else
 517                 seq_printf(m, "%llu %llu %lu\n",
 518                    (unsigned long long)task->se.sum_exec_runtime,
 519                    (unsigned long long)task->sched_info.run_delay,
 520                    task->sched_info.pcount);
 521
 522         return 0;
 523 }
 524 #endif
 525
 526 #ifdef CONFIG_LATENCYTOP
 527 static int lstats_show_proc(struct seq_file *m, void *v)
 528 {
 529         int i;
 530         struct inode *inode = m->private;
 531         struct task_struct *task = get_proc_task(inode);
 532
 533         if (!task)
 534                 return -ESRCH;
 535         seq_puts(m, "Latency Top version : v0.1\n");
 536         for (i = 0; i < LT_SAVECOUNT; i++) {
 537                 struct latency_record *lr = &task->latency_record[i];
 538                 if (lr->backtrace[0]) {
 539                         int q;
 540                         seq_printf(m, "%i %li %li",
 541                                    lr->count, lr->time, lr->max);
 542                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 543                                 unsigned long bt = lr->backtrace[q];
 544
 545                                 if (!bt)
 546                                         break;
 547                                 seq_printf(m, " %ps", (void *)bt);
 548                         }
 549                         seq_putc(m, '\n');
 550                 }
 551
 552         }
 553         put_task_struct(task);
 554         return 0;
 555 }
 556
 557 static int lstats_open(struct inode *inode, struct file *file)
 558 {
 559         return single_open(file, lstats_show_proc, inode);
 560 }
 561
 562 static ssize_t lstats_write(struct file *file, const char __user *buf,
 563                             size_t count, loff_t *offs)
 564 {
 565         struct task_struct *task = get_proc_task(file_inode(file));
 566
 567         if (!task)
 568                 return -ESRCH;
 569         clear_tsk_latency_tracing(task);
 570         put_task_struct(task);
 571
 572         return count;
 573 }
 574
 575 static const struct file_operations proc_lstats_operations = {
 576         .open           = lstats_open,
 577         .read           = seq_read,
 578         .write          = lstats_write,
 579         .llseek         = seq_lseek,
 580         .release        = single_release,
 581 };
 582
 583 #endif
 584
 585 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 586                           struct pid *pid, struct task_struct *task)
 587 {
 588         unsigned long totalpages = totalram_pages() + total_swap_pages;
 589         unsigned long points = 0;
 590         long badness;
 591
 592         badness = oom_badness(task, totalpages);
 593         /*
 594          * Special case OOM_SCORE_ADJ_MIN for all others scale the
 595          * badness value into [0, 2000] range which we have been
 596          * exporting for a long time so userspace might depend on it.
 597          */
 598         if (badness != LONG_MIN)
 599                 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
 600
 601         seq_printf(m, "%lu\n", points);
 602
 603         return 0;
 604 }
 605
 606 struct limit_names {
 607         const char *name;
 608         const char *unit;
 609 };
 610
 611 static const struct limit_names lnames[RLIM_NLIMITS] = {
 612         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 613         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 614         [RLIMIT_DATA] = {"Max data size", "bytes"},
 615         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 616         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 617         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 618         [RLIMIT_NPROC] = {"Max processes", "processes"},
 619         [RLIMIT_NOFILE] = {"Max open files", "files"},
 620         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 621         [RLIMIT_AS] = {"Max address space", "bytes"},
 622         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 623         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 624         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 625         [RLIMIT_NICE] = {"Max nice priority", NULL},
 626         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 627         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 628 };
 629
 630 /* Display limits for a process */
 631 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 632                            struct pid *pid, struct task_struct *task)
 633 {
 634         unsigned int i;
 635         unsigned long flags;
 636
 637         struct rlimit rlim[RLIM_NLIMITS];
 638
 639         if (!lock_task_sighand(task, &flags))
 640                 return 0;
 641         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 642         unlock_task_sighand(task, &flags);
 643
 644         /*
 645          * print the file header
 646          */
 647         seq_puts(m, "Limit                     "
 648                 "Soft Limit           "
 649                 "Hard Limit           "
 650                 "Units     \n");
 651
 652         for (i = 0; i < RLIM_NLIMITS; i++) {
 653                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 654                         seq_printf(m, "%-25s %-20s ",
 655                                    lnames[i].name, "unlimited");
 656                 else
 657                         seq_printf(m, "%-25s %-20lu ",
 658                                    lnames[i].name, rlim[i].rlim_cur);
 659
 660                 if (rlim[i].rlim_max == RLIM_INFINITY)
 661                         seq_printf(m, "%-20s ", "unlimited");
 662                 else
 663                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 664
 665                 if (lnames[i].unit)
 666                         seq_printf(m, "%-10s\n", lnames[i].unit);
 667                 else
 668                         seq_putc(m, '\n');
 669         }
 670
 671         return 0;
 672 }
 673
 674 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 675 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 676                             struct pid *pid, struct task_struct *task)
 677 {
 678         struct syscall_info info;
 679         u64 *args = &info.data.args[0];
 680         int res;
 681
 682         res = lock_trace(task);
 683         if (res)
 684                 return res;
 685
 686         if (task_current_syscall(task, &info))
 687                 seq_puts(m, "running\n");
 688         else if (info.data.nr < 0)
 689                 seq_printf(m, "%d 0x%llx 0x%llx\n",
 690                            info.data.nr, info.sp, info.data.instruction_pointer);
 691         else
 692                 seq_printf(m,
 693                        "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
 694                        info.data.nr,
 695                        args[0], args[1], args[2], args[3], args[4], args[5],
 696                        info.sp, info.data.instruction_pointer);
 697         unlock_trace(task);
 698
 699         return 0;
 700 }
 701 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 702
 703 /************************************************************************/
 704 /*                       Here the fs part begins                        */
 705 /************************************************************************/
 706
 707 /* permission checks */
 708 static bool proc_fd_access_allowed(struct inode *inode)
 709 {
 710         struct task_struct *task;
 711         bool allowed = false;
 712         /* Allow access to a task's file descriptors if it is us or we
 713          * may use ptrace attach to the process and find out that
 714          * information.
 715          */
 716         task = get_proc_task(inode);
 717         if (task) {
 718                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 719                 put_task_struct(task);
 720         }
 721         return allowed;
 722 }
 723
 724 int proc_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 725                  struct iattr *attr)
 726 {
 727         int error;
 728         struct inode *inode = d_inode(dentry);
 729
 730         if (attr->ia_valid & ATTR_MODE)
 731                 return -EPERM;
 732
 733         error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 734         if (error)
 735                 return error;
 736
 737         setattr_copy(&nop_mnt_idmap, inode, attr);
 738         return 0;
 739 }
 740
 741 /*
 742  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 743  * or euid/egid (for hide_pid_min=2)?
 744  */
 745 static bool has_pid_permissions(struct proc_fs_info *fs_info,
 746                                  struct task_struct *task,
 747                                  enum proc_hidepid hide_pid_min)
 748 {
 749         /*
 750          * If 'hidpid' mount option is set force a ptrace check,
 751          * we indicate that we are using a filesystem syscall
 752          * by passing PTRACE_MODE_READ_FSCREDS
 753          */
 754         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
 755                 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 756
 757         if (fs_info->hide_pid < hide_pid_min)
 758                 return true;
 759         if (in_group_p(fs_info->pid_gid))
 760                 return true;
 761         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 762 }
 763
 764
 765 static int proc_pid_permission(struct mnt_idmap *idmap,
 766                                struct inode *inode, int mask)
 767 {
 768         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
 769         struct task_struct *task;
 770         bool has_perms;
 771
 772         task = get_proc_task(inode);
 773         if (!task)
 774                 return -ESRCH;
 775         has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
 776         put_task_struct(task);
 777
 778         if (!has_perms) {
 779                 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
 780                         /*
 781                          * Let's make getdents(), stat(), and open()
 782                          * consistent with each other.  If a process
 783                          * may not stat() a file, it shouldn't be seen
 784                          * in procfs at all.
 785                          */
 786                         return -ENOENT;
 787                 }
 788
 789                 return -EPERM;
 790         }
 791         return generic_permission(&nop_mnt_idmap, inode, mask);
 792 }
 793
 794
 795
 796 static const struct inode_operations proc_def_inode_operations = {
 797         .setattr        = proc_setattr,
 798 };
 799
 800 static int proc_single_show(struct seq_file *m, void *v)
 801 {
 802         struct inode *inode = m->private;
 803         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
 804         struct pid *pid = proc_pid(inode);
 805         struct task_struct *task;
 806         int ret;
 807
 808         task = get_pid_task(pid, PIDTYPE_PID);
 809         if (!task)
 810                 return -ESRCH;
 811
 812         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 813
 814         put_task_struct(task);
 815         return ret;
 816 }
 817
 818 static int proc_single_open(struct inode *inode, struct file *filp)
 819 {
 820         return single_open(filp, proc_single_show, inode);
 821 }
 822
 823 static const struct file_operations proc_single_file_operations = {
 824         .open           = proc_single_open,
 825         .read           = seq_read,
 826         .llseek         = seq_lseek,
 827         .release        = single_release,
 828 };
 829
 830
 831 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 832 {
 833         struct task_struct *task = get_proc_task(inode);
 834         struct mm_struct *mm;
 835
 836         if (!task)
 837                 return ERR_PTR(-ESRCH);
 838
 839         mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 840         put_task_struct(task);
 841
 842         if (IS_ERR(mm))
 843                 return mm == ERR_PTR(-ESRCH) ? NULL : mm;
 844
 845         /* ensure this mm_struct can't be freed */
 846         mmgrab(mm);
 847         /* but do not pin its memory */
 848         mmput(mm);
 849
 850         return mm;
 851 }
 852
 853 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 854 {
 855         struct mm_struct *mm = proc_mem_open(inode, mode);
 856
 857         if (IS_ERR(mm))
 858                 return PTR_ERR(mm);
 859
 860         file->private_data = mm;
 861         return 0;
 862 }
 863
 864 static int mem_open(struct inode *inode, struct file *file)
 865 {
 866         if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
 867                 return -EINVAL;
 868         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
 869 }
 870
 871 static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
 872 {
 873         struct task_struct *task;
 874         bool ptrace_active = false;
 875
 876         switch (proc_mem_force_override) {
 877         case PROC_MEM_FORCE_NEVER:
 878                 return false;
 879         case PROC_MEM_FORCE_PTRACE:
 880                 task = get_proc_task(file_inode(file));
 881                 if (task) {
 882                         ptrace_active = READ_ONCE(task->ptrace) &&
 883                                         READ_ONCE(task->mm) == mm &&
 884                                         READ_ONCE(task->parent) == current;
 885                         put_task_struct(task);
 886                 }
 887                 return ptrace_active;
 888         default:
 889                 return true;
 890         }
 891 }
 892
 893 static ssize_t mem_rw(struct file *file, char __user *buf,
 894                         size_t count, loff_t *ppos, int write)
 895 {
 896         struct mm_struct *mm = file->private_data;
 897         unsigned long addr = *ppos;
 898         ssize_t copied;
 899         char *page;
 900         unsigned int flags;
 901
 902         if (!mm)
 903                 return 0;
 904
 905         page = (char *)__get_free_page(GFP_KERNEL);
 906         if (!page)
 907                 return -ENOMEM;
 908
 909         copied = 0;
 910         if (!mmget_not_zero(mm))
 911                 goto free;
 912
 913         flags = write ? FOLL_WRITE : 0;
 914         if (proc_mem_foll_force(file, mm))
 915                 flags |= FOLL_FORCE;
 916
 917         while (count > 0) {
 918                 size_t this_len = min_t(size_t, count, PAGE_SIZE);
 919
 920                 if (write && copy_from_user(page, buf, this_len)) {
 921                         copied = -EFAULT;
 922                         break;
 923                 }
 924
 925                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 926                 if (!this_len) {
 927                         if (!copied)
 928                                 copied = -EIO;
 929                         break;
 930                 }
 931
 932                 if (!write && copy_to_user(buf, page, this_len)) {
 933                         copied = -EFAULT;
 934                         break;
 935                 }
 936
 937                 buf += this_len;
 938                 addr += this_len;
 939                 copied += this_len;
 940                 count -= this_len;
 941         }
 942         *ppos = addr;
 943
 944         mmput(mm);
 945 free:
 946         free_page((unsigned long) page);
 947         return copied;
 948 }
 949
 950 static ssize_t mem_read(struct file *file, char __user *buf,
 951                         size_t count, loff_t *ppos)
 952 {
 953         return mem_rw(file, buf, count, ppos, 0);
 954 }
 955
 956 static ssize_t mem_write(struct file *file, const char __user *buf,
 957                          size_t count, loff_t *ppos)
 958 {
 959         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 960 }
 961
 962 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 963 {
 964         switch (orig) {
 965         case 0:
 966                 file->f_pos = offset;
 967                 break;
 968         case 1:
 969                 file->f_pos += offset;
 970                 break;
 971         default:
 972                 return -EINVAL;
 973         }
 974         force_successful_syscall_return();
 975         return file->f_pos;
 976 }
 977
 978 static int mem_release(struct inode *inode, struct file *file)
 979 {
 980         struct mm_struct *mm = file->private_data;
 981         if (mm)
 982                 mmdrop(mm);
 983         return 0;
 984 }
 985
 986 static const struct file_operations proc_mem_operations = {
 987         .llseek         = mem_lseek,
 988         .read           = mem_read,
 989         .write          = mem_write,
 990         .open           = mem_open,
 991         .release        = mem_release,
 992         .fop_flags      = FOP_UNSIGNED_OFFSET,
 993 };
 994
 995 static int environ_open(struct inode *inode, struct file *file)
 996 {
 997         return __mem_open(inode, file, PTRACE_MODE_READ);
 998 }
 999
1000 static ssize_t environ_read(struct file *file, char __user *buf,
1001                         size_t count, loff_t *ppos)
1002 {
1003         char *page;
1004         unsigned long src = *ppos;
1005         int ret = 0;
1006         struct mm_struct *mm = file->private_data;
1007         unsigned long env_start, env_end;
1008
1009         /* Ensure the process spawned far enough to have an environment. */
1010         if (!mm || !mm->env_end)
1011                 return 0;
1012
1013         page = (char *)__get_free_page(GFP_KERNEL);
1014         if (!page)
1015                 return -ENOMEM;
1016
1017         ret = 0;
1018         if (!mmget_not_zero(mm))
1019                 goto free;
1020
1021         spin_lock(&mm->arg_lock);
1022         env_start = mm->env_start;
1023         env_end = mm->env_end;
1024         spin_unlock(&mm->arg_lock);
1025
1026         while (count > 0) {
1027                 size_t this_len, max_len;
1028                 int retval;
1029
1030                 if (src >= (env_end - env_start))
1031                         break;
1032
1033                 this_len = env_end - (env_start + src);
1034
1035                 max_len = min_t(size_t, PAGE_SIZE, count);
1036                 this_len = min(max_len, this_len);
1037
1038                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
1039
1040                 if (retval <= 0) {
1041                         ret = retval;
1042                         break;
1043                 }
1044
1045                 if (copy_to_user(buf, page, retval)) {
1046                         ret = -EFAULT;
1047                         break;
1048                 }
1049
1050                 ret += retval;
1051                 src += retval;
1052                 buf += retval;
1053                 count -= retval;
1054         }
1055         *ppos = src;
1056         mmput(mm);
1057
1058 free:
1059         free_page((unsigned long) page);
1060         return ret;
1061 }
1062
1063 static const struct file_operations proc_environ_operations = {
1064         .open           = environ_open,
1065         .read           = environ_read,
1066         .llseek         = generic_file_llseek,
1067         .release        = mem_release,
1068 };
1069
1070 static int auxv_open(struct inode *inode, struct file *file)
1071 {
1072         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1073 }
1074
1075 static ssize_t auxv_read(struct file *file, char __user *buf,
1076                         size_t count, loff_t *ppos)
1077 {
1078         struct mm_struct *mm = file->private_data;
1079         unsigned int nwords = 0;
1080
1081         if (!mm)
1082                 return 0;
1083         do {
1084                 nwords += 2;
1085         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1086         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1087                                        nwords * sizeof(mm->saved_auxv[0]));
1088 }
1089
1090 static const struct file_operations proc_auxv_operations = {
1091         .open           = auxv_open,
1092         .read           = auxv_read,
1093         .llseek         = generic_file_llseek,
1094         .release        = mem_release,
1095 };
1096
1097 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1098                             loff_t *ppos)
1099 {
1100         struct task_struct *task = get_proc_task(file_inode(file));
1101         char buffer[PROC_NUMBUF];
1102         int oom_adj = OOM_ADJUST_MIN;
1103         size_t len;
1104
1105         if (!task)
1106                 return -ESRCH;
1107         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1108                 oom_adj = OOM_ADJUST_MAX;
1109         else
1110                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1111                           OOM_SCORE_ADJ_MAX;
1112         put_task_struct(task);
1113         if (oom_adj > OOM_ADJUST_MAX)
1114                 oom_adj = OOM_ADJUST_MAX;
1115         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1116         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1117 }
1118
1119 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1120 {
1121         struct mm_struct *mm = NULL;
1122         struct task_struct *task;
1123         int err = 0;
1124
1125         task = get_proc_task(file_inode(file));
1126         if (!task)
1127                 return -ESRCH;
1128
1129         mutex_lock(&oom_adj_mutex);
1130         if (legacy) {
1131                 if (oom_adj < task->signal->oom_score_adj &&
1132                                 !capable(CAP_SYS_RESOURCE)) {
1133                         err = -EACCES;
1134                         goto err_unlock;
1135                 }
1136                 /*
1137                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1138                  * /proc/pid/oom_score_adj instead.
1139                  */
1140                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1141                           current->comm, task_pid_nr(current), task_pid_nr(task),
1142                           task_pid_nr(task));
1143         } else {
1144                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1145                                 !capable(CAP_SYS_RESOURCE)) {
1146                         err = -EACCES;
1147                         goto err_unlock;
1148                 }
1149         }
1150
1151         /*
1152          * Make sure we will check other processes sharing the mm if this is
1153          * not vfrok which wants its own oom_score_adj.
1154          * pin the mm so it doesn't go away and get reused after task_unlock
1155          */
1156         if (!task->vfork_done) {
1157                 struct task_struct *p = find_lock_task_mm(task);
1158
1159                 if (p) {
1160                         if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1161                                 mm = p->mm;
1162                                 mmgrab(mm);
1163                         }
1164                         task_unlock(p);
1165                 }
1166         }
1167
1168         task->signal->oom_score_adj = oom_adj;
1169         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1170                 task->signal->oom_score_adj_min = (short)oom_adj;
1171         trace_oom_score_adj_update(task);
1172
1173         if (mm) {
1174                 struct task_struct *p;
1175
1176                 rcu_read_lock();
1177                 for_each_process(p) {
1178                         if (same_thread_group(task, p))
1179                                 continue;
1180
1181                         /* do not touch kernel threads or the global init */
1182                         if (p->flags & PF_KTHREAD || is_global_init(p))
1183                                 continue;
1184
1185                         task_lock(p);
1186                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1187                                 p->signal->oom_score_adj = oom_adj;
1188                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1189                                         p->signal->oom_score_adj_min = (short)oom_adj;
1190                         }
1191                         task_unlock(p);
1192                 }
1193                 rcu_read_unlock();
1194                 mmdrop(mm);
1195         }
1196 err_unlock:
1197         mutex_unlock(&oom_adj_mutex);
1198         put_task_struct(task);
1199         return err;
1200 }
1201
1202 /*
1203  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1204  * kernels.  The effective policy is defined by oom_score_adj, which has a
1205  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1206  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1207  * Processes that become oom disabled via oom_adj will still be oom disabled
1208  * with this implementation.
1209  *
1210  * oom_adj cannot be removed since existing userspace binaries use it.
1211  */
1212 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1213                              size_t count, loff_t *ppos)
1214 {
1215         char buffer[PROC_NUMBUF] = {};
1216         int oom_adj;
1217         int err;
1218
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_adj);
1227         if (err)
1228                 goto out;
1229         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1230              oom_adj != OOM_DISABLE) {
1231                 err = -EINVAL;
1232                 goto out;
1233         }
1234
1235         /*
1236          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1237          * value is always attainable.
1238          */
1239         if (oom_adj == OOM_ADJUST_MAX)
1240                 oom_adj = OOM_SCORE_ADJ_MAX;
1241         else
1242                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1243
1244         err = __set_oom_adj(file, oom_adj, true);
1245 out:
1246         return err < 0 ? err : count;
1247 }
1248
1249 static const struct file_operations proc_oom_adj_operations = {
1250         .read           = oom_adj_read,
1251         .write          = oom_adj_write,
1252         .llseek         = generic_file_llseek,
1253 };
1254
1255 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1256                                         size_t count, loff_t *ppos)
1257 {
1258         struct task_struct *task = get_proc_task(file_inode(file));
1259         char buffer[PROC_NUMBUF];
1260         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1261         size_t len;
1262
1263         if (!task)
1264                 return -ESRCH;
1265         oom_score_adj = task->signal->oom_score_adj;
1266         put_task_struct(task);
1267         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1268         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1269 }
1270
1271 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1272                                         size_t count, loff_t *ppos)
1273 {
1274         char buffer[PROC_NUMBUF] = {};
1275         int oom_score_adj;
1276         int err;
1277
1278         if (count > sizeof(buffer) - 1)
1279                 count = sizeof(buffer) - 1;
1280         if (copy_from_user(buffer, buf, count)) {
1281                 err = -EFAULT;
1282                 goto out;
1283         }
1284
1285         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1286         if (err)
1287                 goto out;
1288         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1289                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1290                 err = -EINVAL;
1291                 goto out;
1292         }
1293
1294         err = __set_oom_adj(file, oom_score_adj, false);
1295 out:
1296         return err < 0 ? err : count;
1297 }
1298
1299 static const struct file_operations proc_oom_score_adj_operations = {
1300         .read           = oom_score_adj_read,
1301         .write          = oom_score_adj_write,
1302         .llseek         = default_llseek,
1303 };
1304
1305 #ifdef CONFIG_AUDIT
1306 #define TMPBUFLEN 11
1307 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1308                                   size_t count, loff_t *ppos)
1309 {
1310         struct inode * inode = file_inode(file);
1311         struct task_struct *task = get_proc_task(inode);
1312         ssize_t length;
1313         char tmpbuf[TMPBUFLEN];
1314
1315         if (!task)
1316                 return -ESRCH;
1317         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1318                            from_kuid(file->f_cred->user_ns,
1319                                      audit_get_loginuid(task)));
1320         put_task_struct(task);
1321         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1322 }
1323
1324 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1325                                    size_t count, loff_t *ppos)
1326 {
1327         struct inode * inode = file_inode(file);
1328         uid_t loginuid;
1329         kuid_t kloginuid;
1330         int rv;
1331
1332         /* Don't let kthreads write their own loginuid */
1333         if (current->flags & PF_KTHREAD)
1334                 return -EPERM;
1335
1336         rcu_read_lock();
1337         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1338                 rcu_read_unlock();
1339                 return -EPERM;
1340         }
1341         rcu_read_unlock();
1342
1343         if (*ppos != 0) {
1344                 /* No partial writes. */
1345                 return -EINVAL;
1346         }
1347
1348         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1349         if (rv < 0)
1350                 return rv;
1351
1352         /* is userspace tring to explicitly UNSET the loginuid? */
1353         if (loginuid == AUDIT_UID_UNSET) {
1354                 kloginuid = INVALID_UID;
1355         } else {
1356                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1357                 if (!uid_valid(kloginuid))
1358                         return -EINVAL;
1359         }
1360
1361         rv = audit_set_loginuid(kloginuid);
1362         if (rv < 0)
1363                 return rv;
1364         return count;
1365 }
1366
1367 static const struct file_operations proc_loginuid_operations = {
1368         .read           = proc_loginuid_read,
1369         .write          = proc_loginuid_write,
1370         .llseek         = generic_file_llseek,
1371 };
1372
1373 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1374                                   size_t count, loff_t *ppos)
1375 {
1376         struct inode * inode = file_inode(file);
1377         struct task_struct *task = get_proc_task(inode);
1378         ssize_t length;
1379         char tmpbuf[TMPBUFLEN];
1380
1381         if (!task)
1382                 return -ESRCH;
1383         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1384                                 audit_get_sessionid(task));
1385         put_task_struct(task);
1386         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1387 }
1388
1389 static const struct file_operations proc_sessionid_operations = {
1390         .read           = proc_sessionid_read,
1391         .llseek         = generic_file_llseek,
1392 };
1393 #endif
1394
1395 #ifdef CONFIG_FAULT_INJECTION
1396 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1397                                       size_t count, loff_t *ppos)
1398 {
1399         struct task_struct *task = get_proc_task(file_inode(file));
1400         char buffer[PROC_NUMBUF];
1401         size_t len;
1402         int make_it_fail;
1403
1404         if (!task)
1405                 return -ESRCH;
1406         make_it_fail = task->make_it_fail;
1407         put_task_struct(task);
1408
1409         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1410
1411         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1412 }
1413
1414 static ssize_t proc_fault_inject_write(struct file * file,
1415                         const char __user * buf, size_t count, loff_t *ppos)
1416 {
1417         struct task_struct *task;
1418         char buffer[PROC_NUMBUF] = {};
1419         int make_it_fail;
1420         int rv;
1421
1422         if (!capable(CAP_SYS_RESOURCE))
1423                 return -EPERM;
1424
1425         if (count > sizeof(buffer) - 1)
1426                 count = sizeof(buffer) - 1;
1427         if (copy_from_user(buffer, buf, count))
1428                 return -EFAULT;
1429         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1430         if (rv < 0)
1431                 return rv;
1432         if (make_it_fail < 0 || make_it_fail > 1)
1433                 return -EINVAL;
1434
1435         task = get_proc_task(file_inode(file));
1436         if (!task)
1437                 return -ESRCH;
1438         task->make_it_fail = make_it_fail;
1439         put_task_struct(task);
1440
1441         return count;
1442 }
1443
1444 static const struct file_operations proc_fault_inject_operations = {
1445         .read           = proc_fault_inject_read,
1446         .write          = proc_fault_inject_write,
1447         .llseek         = generic_file_llseek,
1448 };
1449
1450 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1451                                    size_t count, loff_t *ppos)
1452 {
1453         struct task_struct *task;
1454         int err;
1455         unsigned int n;
1456
1457         err = kstrtouint_from_user(buf, count, 0, &n);
1458         if (err)
1459                 return err;
1460
1461         task = get_proc_task(file_inode(file));
1462         if (!task)
1463                 return -ESRCH;
1464         task->fail_nth = n;
1465         put_task_struct(task);
1466
1467         return count;
1468 }
1469
1470 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1471                                   size_t count, loff_t *ppos)
1472 {
1473         struct task_struct *task;
1474         char numbuf[PROC_NUMBUF];
1475         ssize_t len;
1476
1477         task = get_proc_task(file_inode(file));
1478         if (!task)
1479                 return -ESRCH;
1480         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1481         put_task_struct(task);
1482         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1483 }
1484
1485 static const struct file_operations proc_fail_nth_operations = {
1486         .read           = proc_fail_nth_read,
1487         .write          = proc_fail_nth_write,
1488 };
1489 #endif
1490
1491
1492 #ifdef CONFIG_SCHED_DEBUG
1493 /*
1494  * Print out various scheduling related per-task fields:
1495  */
1496 static int sched_show(struct seq_file *m, void *v)
1497 {
1498         struct inode *inode = m->private;
1499         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1500         struct task_struct *p;
1501
1502         p = get_proc_task(inode);
1503         if (!p)
1504                 return -ESRCH;
1505         proc_sched_show_task(p, ns, m);
1506
1507         put_task_struct(p);
1508
1509         return 0;
1510 }
1511
1512 static ssize_t
1513 sched_write(struct file *file, const char __user *buf,
1514             size_t count, loff_t *offset)
1515 {
1516         struct inode *inode = file_inode(file);
1517         struct task_struct *p;
1518
1519         p = get_proc_task(inode);
1520         if (!p)
1521                 return -ESRCH;
1522         proc_sched_set_task(p);
1523
1524         put_task_struct(p);
1525
1526         return count;
1527 }
1528
1529 static int sched_open(struct inode *inode, struct file *filp)
1530 {
1531         return single_open(filp, sched_show, inode);
1532 }
1533
1534 static const struct file_operations proc_pid_sched_operations = {
1535         .open           = sched_open,
1536         .read           = seq_read,
1537         .write          = sched_write,
1538         .llseek         = seq_lseek,
1539         .release        = single_release,
1540 };
1541
1542 #endif
1543
1544 #ifdef CONFIG_SCHED_AUTOGROUP
1545 /*
1546  * Print out autogroup related information:
1547  */
1548 static int sched_autogroup_show(struct seq_file *m, void *v)
1549 {
1550         struct inode *inode = m->private;
1551         struct task_struct *p;
1552
1553         p = get_proc_task(inode);
1554         if (!p)
1555                 return -ESRCH;
1556         proc_sched_autogroup_show_task(p, m);
1557
1558         put_task_struct(p);
1559
1560         return 0;
1561 }
1562
1563 static ssize_t
1564 sched_autogroup_write(struct file *file, const char __user *buf,
1565             size_t count, loff_t *offset)
1566 {
1567         struct inode *inode = file_inode(file);
1568         struct task_struct *p;
1569         char buffer[PROC_NUMBUF] = {};
1570         int nice;
1571         int err;
1572
1573         if (count > sizeof(buffer) - 1)
1574                 count = sizeof(buffer) - 1;
1575         if (copy_from_user(buffer, buf, count))
1576                 return -EFAULT;
1577
1578         err = kstrtoint(strstrip(buffer), 0, &nice);
1579         if (err < 0)
1580                 return err;
1581
1582         p = get_proc_task(inode);
1583         if (!p)
1584                 return -ESRCH;
1585
1586         err = proc_sched_autogroup_set_nice(p, nice);
1587         if (err)
1588                 count = err;
1589
1590         put_task_struct(p);
1591
1592         return count;
1593 }
1594
1595 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1596 {
1597         int ret;
1598
1599         ret = single_open(filp, sched_autogroup_show, NULL);
1600         if (!ret) {
1601                 struct seq_file *m = filp->private_data;
1602
1603                 m->private = inode;
1604         }
1605         return ret;
1606 }
1607
1608 static const struct file_operations proc_pid_sched_autogroup_operations = {
1609         .open           = sched_autogroup_open,
1610         .read           = seq_read,
1611         .write          = sched_autogroup_write,
1612         .llseek         = seq_lseek,
1613         .release        = single_release,
1614 };
1615
1616 #endif /* CONFIG_SCHED_AUTOGROUP */
1617
1618 #ifdef CONFIG_TIME_NS
1619 static int timens_offsets_show(struct seq_file *m, void *v)
1620 {
1621         struct task_struct *p;
1622
1623         p = get_proc_task(file_inode(m->file));
1624         if (!p)
1625                 return -ESRCH;
1626         proc_timens_show_offsets(p, m);
1627
1628         put_task_struct(p);
1629
1630         return 0;
1631 }
1632
1633 static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1634                                     size_t count, loff_t *ppos)
1635 {
1636         struct inode *inode = file_inode(file);
1637         struct proc_timens_offset offsets[2];
1638         char *kbuf = NULL, *pos, *next_line;
1639         struct task_struct *p;
1640         int ret, noffsets;
1641
1642         /* Only allow < page size writes at the beginning of the file */
1643         if ((*ppos != 0) || (count >= PAGE_SIZE))
1644                 return -EINVAL;
1645
1646         /* Slurp in the user data */
1647         kbuf = memdup_user_nul(buf, count);
1648         if (IS_ERR(kbuf))
1649                 return PTR_ERR(kbuf);
1650
1651         /* Parse the user data */
1652         ret = -EINVAL;
1653         noffsets = 0;
1654         for (pos = kbuf; pos; pos = next_line) {
1655                 struct proc_timens_offset *off = &offsets[noffsets];
1656                 char clock[10];
1657                 int err;
1658
1659                 /* Find the end of line and ensure we don't look past it */
1660                 next_line = strchr(pos, '\n');
1661                 if (next_line) {
1662                         *next_line = '\0';
1663                         next_line++;
1664                         if (*next_line == '\0')
1665                                 next_line = NULL;
1666                 }
1667
1668                 err = sscanf(pos, "%9s %lld %lu", clock,
1669                                 &off->val.tv_sec, &off->val.tv_nsec);
1670                 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1671                         goto out;
1672
1673                 clock[sizeof(clock) - 1] = 0;
1674                 if (strcmp(clock, "monotonic") == 0 ||
1675                     strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1676                         off->clockid = CLOCK_MONOTONIC;
1677                 else if (strcmp(clock, "boottime") == 0 ||
1678                          strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1679                         off->clockid = CLOCK_BOOTTIME;
1680                 else
1681                         goto out;
1682
1683                 noffsets++;
1684                 if (noffsets == ARRAY_SIZE(offsets)) {
1685                         if (next_line)
1686                                 count = next_line - kbuf;
1687                         break;
1688                 }
1689         }
1690
1691         ret = -ESRCH;
1692         p = get_proc_task(inode);
1693         if (!p)
1694                 goto out;
1695         ret = proc_timens_set_offset(file, p, offsets, noffsets);
1696         put_task_struct(p);
1697         if (ret)
1698                 goto out;
1699
1700         ret = count;
1701 out:
1702         kfree(kbuf);
1703         return ret;
1704 }
1705
1706 static int timens_offsets_open(struct inode *inode, struct file *filp)
1707 {
1708         return single_open(filp, timens_offsets_show, inode);
1709 }
1710
1711 static const struct file_operations proc_timens_offsets_operations = {
1712         .open           = timens_offsets_open,
1713         .read           = seq_read,
1714         .write          = timens_offsets_write,
1715         .llseek         = seq_lseek,
1716         .release        = single_release,
1717 };
1718 #endif /* CONFIG_TIME_NS */
1719
1720 static ssize_t comm_write(struct file *file, const char __user *buf,
1721                                 size_t count, loff_t *offset)
1722 {
1723         struct inode *inode = file_inode(file);
1724         struct task_struct *p;
1725         char buffer[TASK_COMM_LEN] = {};
1726         const size_t maxlen = sizeof(buffer) - 1;
1727
1728         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1729                 return -EFAULT;
1730
1731         p = get_proc_task(inode);
1732         if (!p)
1733                 return -ESRCH;
1734
1735         if (same_thread_group(current, p)) {
1736                 set_task_comm(p, buffer);
1737                 proc_comm_connector(p);
1738         }
1739         else
1740                 count = -EINVAL;
1741
1742         put_task_struct(p);
1743
1744         return count;
1745 }
1746
1747 static int comm_show(struct seq_file *m, void *v)
1748 {
1749         struct inode *inode = m->private;
1750         struct task_struct *p;
1751
1752         p = get_proc_task(inode);
1753         if (!p)
1754                 return -ESRCH;
1755
1756         proc_task_name(m, p, false);
1757         seq_putc(m, '\n');
1758
1759         put_task_struct(p);
1760
1761         return 0;
1762 }
1763
1764 static int comm_open(struct inode *inode, struct file *filp)
1765 {
1766         return single_open(filp, comm_show, inode);
1767 }
1768
1769 static const struct file_operations proc_pid_set_comm_operations = {
1770         .open           = comm_open,
1771         .read           = seq_read,
1772         .write          = comm_write,
1773         .llseek         = seq_lseek,
1774         .release        = single_release,
1775 };
1776
1777 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1778 {
1779         struct task_struct *task;
1780         struct file *exe_file;
1781
1782         task = get_proc_task(d_inode(dentry));
1783         if (!task)
1784                 return -ENOENT;
1785         exe_file = get_task_exe_file(task);
1786         put_task_struct(task);
1787         if (exe_file) {
1788                 *exe_path = exe_file->f_path;
1789                 path_get(&exe_file->f_path);
1790                 fput(exe_file);
1791                 return 0;
1792         } else
1793                 return -ENOENT;
1794 }
1795
1796 static const char *proc_pid_get_link(struct dentry *dentry,
1797                                      struct inode *inode,
1798                                      struct delayed_call *done)
1799 {
1800         struct path path;
1801         int error = -EACCES;
1802
1803         if (!dentry)
1804                 return ERR_PTR(-ECHILD);
1805
1806         /* Are we allowed to snoop on the tasks file descriptors? */
1807         if (!proc_fd_access_allowed(inode))
1808                 goto out;
1809
1810         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1811         if (error)
1812                 goto out;
1813
1814         error = nd_jump_link(&path);
1815 out:
1816         return ERR_PTR(error);
1817 }
1818
1819 static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1820 {
1821         char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1822         char *pathname;
1823         int len;
1824
1825         if (!tmp)
1826                 return -ENOMEM;
1827
1828         pathname = d_path(path, tmp, PATH_MAX);
1829         len = PTR_ERR(pathname);
1830         if (IS_ERR(pathname))
1831                 goto out;
1832         len = tmp + PATH_MAX - 1 - pathname;
1833
1834         if (len > buflen)
1835                 len = buflen;
1836         if (copy_to_user(buffer, pathname, len))
1837                 len = -EFAULT;
1838  out:
1839         kfree(tmp);
1840         return len;
1841 }
1842
1843 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1844 {
1845         int error = -EACCES;
1846         struct inode *inode = d_inode(dentry);
1847         struct path path;
1848
1849         /* Are we allowed to snoop on the tasks file descriptors? */
1850         if (!proc_fd_access_allowed(inode))
1851                 goto out;
1852
1853         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1854         if (error)
1855                 goto out;
1856
1857         error = do_proc_readlink(&path, buffer, buflen);
1858         path_put(&path);
1859 out:
1860         return error;
1861 }
1862
1863 const struct inode_operations proc_pid_link_inode_operations = {
1864         .readlink       = proc_pid_readlink,
1865         .get_link       = proc_pid_get_link,
1866         .setattr        = proc_setattr,
1867 };
1868
1869
1870 /* building an inode */
1871
1872 void task_dump_owner(struct task_struct *task, umode_t mode,
1873                      kuid_t *ruid, kgid_t *rgid)
1874 {
1875         /* Depending on the state of dumpable compute who should own a
1876          * proc file for a task.
1877          */
1878         const struct cred *cred;
1879         kuid_t uid;
1880         kgid_t gid;
1881
1882         if (unlikely(task->flags & PF_KTHREAD)) {
1883                 *ruid = GLOBAL_ROOT_UID;
1884                 *rgid = GLOBAL_ROOT_GID;
1885                 return;
1886         }
1887
1888         /* Default to the tasks effective ownership */
1889         rcu_read_lock();
1890         cred = __task_cred(task);
1891         uid = cred->euid;
1892         gid = cred->egid;
1893         rcu_read_unlock();
1894
1895         /*
1896          * Before the /proc/pid/status file was created the only way to read
1897          * the effective uid of a /process was to stat /proc/pid.  Reading
1898          * /proc/pid/status is slow enough that procps and other packages
1899          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1900          * made this apply to all per process world readable and executable
1901          * directories.
1902          */
1903         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1904                 struct mm_struct *mm;
1905                 task_lock(task);
1906                 mm = task->mm;
1907                 /* Make non-dumpable tasks owned by some root */
1908                 if (mm) {
1909                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1910                                 struct user_namespace *user_ns = mm->user_ns;
1911
1912                                 uid = make_kuid(user_ns, 0);
1913                                 if (!uid_valid(uid))
1914                                         uid = GLOBAL_ROOT_UID;
1915
1916                                 gid = make_kgid(user_ns, 0);
1917                                 if (!gid_valid(gid))
1918                                         gid = GLOBAL_ROOT_GID;
1919                         }
1920                 } else {
1921                         uid = GLOBAL_ROOT_UID;
1922                         gid = GLOBAL_ROOT_GID;
1923                 }
1924                 task_unlock(task);
1925         }
1926         *ruid = uid;
1927         *rgid = gid;
1928 }
1929
1930 void proc_pid_evict_inode(struct proc_inode *ei)
1931 {
1932         struct pid *pid = ei->pid;
1933
1934         if (S_ISDIR(ei->vfs_inode.i_mode)) {
1935                 spin_lock(&pid->lock);
1936                 hlist_del_init_rcu(&ei->sibling_inodes);
1937                 spin_unlock(&pid->lock);
1938         }
1939 }
1940
1941 struct inode *proc_pid_make_inode(struct super_block *sb,
1942                                   struct task_struct *task, umode_t mode)
1943 {
1944         struct inode * inode;
1945         struct proc_inode *ei;
1946         struct pid *pid;
1947
1948         /* We need a new inode */
1949
1950         inode = new_inode(sb);
1951         if (!inode)
1952                 goto out;
1953
1954         /* Common stuff */
1955         ei = PROC_I(inode);
1956         inode->i_mode = mode;
1957         inode->i_ino = get_next_ino();
1958         simple_inode_init_ts(inode);
1959         inode->i_op = &proc_def_inode_operations;
1960
1961         /*
1962          * grab the reference to task.
1963          */
1964         pid = get_task_pid(task, PIDTYPE_PID);
1965         if (!pid)
1966                 goto out_unlock;
1967
1968         /* Let the pid remember us for quick removal */
1969         ei->pid = pid;
1970
1971         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1972         security_task_to_inode(task, inode);
1973
1974 out:
1975         return inode;
1976
1977 out_unlock:
1978         iput(inode);
1979         return NULL;
1980 }
1981
1982 /*
1983  * Generating an inode and adding it into @pid->inodes, so that task will
1984  * invalidate inode's dentry before being released.
1985  *
1986  * This helper is used for creating dir-type entries under '/proc' and
1987  * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1988  * can be released by invalidating '/proc/<tgid>' dentry.
1989  * In theory, dentries under '/proc/<tgid>/task' can also be released by
1990  * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1991  * thread exiting situation: Any one of threads should invalidate its
1992  * '/proc/<tgid>/task/<pid>' dentry before released.
1993  */
1994 static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1995                                 struct task_struct *task, umode_t mode)
1996 {
1997         struct inode *inode;
1998         struct proc_inode *ei;
1999         struct pid *pid;
2000
2001         inode = proc_pid_make_inode(sb, task, mode);
2002         if (!inode)
2003                 return NULL;
2004
2005         /* Let proc_flush_pid find this directory inode */
2006         ei = PROC_I(inode);
2007         pid = ei->pid;
2008         spin_lock(&pid->lock);
2009         hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
2010         spin_unlock(&pid->lock);
2011
2012         return inode;
2013 }
2014
2015 int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
2016                 struct kstat *stat, u32 request_mask, unsigned int query_flags)
2017 {
2018         struct inode *inode = d_inode(path->dentry);
2019         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
2020         struct task_struct *task;
2021
2022         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2023
2024         stat->uid = GLOBAL_ROOT_UID;
2025         stat->gid = GLOBAL_ROOT_GID;
2026         rcu_read_lock();
2027         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2028         if (task) {
2029                 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2030                         rcu_read_unlock();
2031                         /*
2032                          * This doesn't prevent learning whether PID exists,
2033                          * it only makes getattr() consistent with readdir().
2034                          */
2035                         return -ENOENT;
2036                 }
2037                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2038         }
2039         rcu_read_unlock();
2040         return 0;
2041 }
2042
2043 /* dentry stuff */
2044
2045 /*
2046  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2047  */
2048 void pid_update_inode(struct task_struct *task, struct inode *inode)
2049 {
2050         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
2051
2052         inode->i_mode &= ~(S_ISUID | S_ISGID);
2053         security_task_to_inode(task, inode);
2054 }
2055
2056 /*
2057  * Rewrite the inode's ownerships here because the owning task may have
2058  * performed a setuid(), etc.
2059  *
2060  */
2061 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2062 {
2063         struct inode *inode;
2064         struct task_struct *task;
2065         int ret = 0;
2066
2067         rcu_read_lock();
2068         inode = d_inode_rcu(dentry);
2069         if (!inode)
2070                 goto out;
2071         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2072
2073         if (task) {
2074                 pid_update_inode(task, inode);
2075                 ret = 1;
2076         }
2077 out:
2078         rcu_read_unlock();
2079         return ret;
2080 }
2081
2082 static inline bool proc_inode_is_dead(struct inode *inode)
2083 {
2084         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2085 }
2086
2087 int pid_delete_dentry(const struct dentry *dentry)
2088 {
2089         /* Is the task we represent dead?
2090          * If so, then don't put the dentry on the lru list,
2091          * kill it immediately.
2092          */
2093         return proc_inode_is_dead(d_inode(dentry));
2094 }
2095
2096 const struct dentry_operations pid_dentry_operations =
2097 {
2098         .d_revalidate   = pid_revalidate,
2099         .d_delete       = pid_delete_dentry,
2100 };
2101
2102 /* Lookups */
2103
2104 /*
2105  * Fill a directory entry.
2106  *
2107  * If possible create the dcache entry and derive our inode number and
2108  * file type from dcache entry.
2109  *
2110  * Since all of the proc inode numbers are dynamically generated, the inode
2111  * numbers do not exist until the inode is cache.  This means creating
2112  * the dcache entry in readdir is necessary to keep the inode numbers
2113  * reported by readdir in sync with the inode numbers reported
2114  * by stat.
2115  */
2116 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2117         const char *name, unsigned int len,
2118         instantiate_t instantiate, struct task_struct *task, const void *ptr)
2119 {
2120         struct dentry *child, *dir = file->f_path.dentry;
2121         struct qstr qname = QSTR_INIT(name, len);
2122         struct inode *inode;
2123         unsigned type = DT_UNKNOWN;
2124         ino_t ino = 1;
2125
2126         child = d_hash_and_lookup(dir, &qname);
2127         if (!child) {
2128                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2129                 child = d_alloc_parallel(dir, &qname, &wq);
2130                 if (IS_ERR(child))
2131                         goto end_instantiate;
2132                 if (d_in_lookup(child)) {
2133                         struct dentry *res;
2134                         res = instantiate(child, task, ptr);
2135                         d_lookup_done(child);
2136                         if (unlikely(res)) {
2137                                 dput(child);
2138                                 child = res;
2139                                 if (IS_ERR(child))
2140                                         goto end_instantiate;
2141                         }
2142                 }
2143         }
2144         inode = d_inode(child);
2145         ino = inode->i_ino;
2146         type = inode->i_mode >> 12;
2147         dput(child);
2148 end_instantiate:
2149         return dir_emit(ctx, name, len, ino, type);
2150 }
2151
2152 /*
2153  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2154  * which represent vma start and end addresses.
2155  */
2156 static int dname_to_vma_addr(struct dentry *dentry,
2157                              unsigned long *start, unsigned long *end)
2158 {
2159         const char *str = dentry->d_name.name;
2160         unsigned long long sval, eval;
2161         unsigned int len;
2162
2163         if (str[0] == '0' && str[1] != '-')
2164                 return -EINVAL;
2165         len = _parse_integer(str, 16, &sval);
2166         if (len & KSTRTOX_OVERFLOW)
2167                 return -EINVAL;
2168         if (sval != (unsigned long)sval)
2169                 return -EINVAL;
2170         str += len;
2171
2172         if (*str != '-')
2173                 return -EINVAL;
2174         str++;
2175
2176         if (str[0] == '0' && str[1])
2177                 return -EINVAL;
2178         len = _parse_integer(str, 16, &eval);
2179         if (len & KSTRTOX_OVERFLOW)
2180                 return -EINVAL;
2181         if (eval != (unsigned long)eval)
2182                 return -EINVAL;
2183         str += len;
2184
2185         if (*str != '\0')
2186                 return -EINVAL;
2187
2188         *start = sval;
2189         *end = eval;
2190
2191         return 0;
2192 }
2193
2194 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2195 {
2196         unsigned long vm_start, vm_end;
2197         bool exact_vma_exists = false;
2198         struct mm_struct *mm = NULL;
2199         struct task_struct *task;
2200         struct inode *inode;
2201         int status = 0;
2202
2203         if (flags & LOOKUP_RCU)
2204                 return -ECHILD;
2205
2206         inode = d_inode(dentry);
2207         task = get_proc_task(inode);
2208         if (!task)
2209                 goto out_notask;
2210
2211         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2212         if (IS_ERR(mm))
2213                 goto out;
2214
2215         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2216                 status = mmap_read_lock_killable(mm);
2217                 if (!status) {
2218                         exact_vma_exists = !!find_exact_vma(mm, vm_start,
2219                                                             vm_end);
2220                         mmap_read_unlock(mm);
2221                 }
2222         }
2223
2224         mmput(mm);
2225
2226         if (exact_vma_exists) {
2227                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2228
2229                 security_task_to_inode(task, inode);
2230                 status = 1;
2231         }
2232
2233 out:
2234         put_task_struct(task);
2235
2236 out_notask:
2237         return status;
2238 }
2239
2240 static const struct dentry_operations tid_map_files_dentry_operations = {
2241         .d_revalidate   = map_files_d_revalidate,
2242         .d_delete       = pid_delete_dentry,
2243 };
2244
2245 static int map_files_get_link(struct dentry *dentry, struct path *path)
2246 {
2247         unsigned long vm_start, vm_end;
2248         struct vm_area_struct *vma;
2249         struct task_struct *task;
2250         struct mm_struct *mm;
2251         int rc;
2252
2253         rc = -ENOENT;
2254         task = get_proc_task(d_inode(dentry));
2255         if (!task)
2256                 goto out;
2257
2258         mm = get_task_mm(task);
2259         put_task_struct(task);
2260         if (!mm)
2261                 goto out;
2262
2263         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2264         if (rc)
2265                 goto out_mmput;
2266
2267         rc = mmap_read_lock_killable(mm);
2268         if (rc)
2269                 goto out_mmput;
2270
2271         rc = -ENOENT;
2272         vma = find_exact_vma(mm, vm_start, vm_end);
2273         if (vma && vma->vm_file) {
2274                 *path = *file_user_path(vma->vm_file);
2275                 path_get(path);
2276                 rc = 0;
2277         }
2278         mmap_read_unlock(mm);
2279
2280 out_mmput:
2281         mmput(mm);
2282 out:
2283         return rc;
2284 }
2285
2286 struct map_files_info {
2287         unsigned long   start;
2288         unsigned long   end;
2289         fmode_t         mode;
2290 };
2291
2292 /*
2293  * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2294  * to concerns about how the symlinks may be used to bypass permissions on
2295  * ancestor directories in the path to the file in question.
2296  */
2297 static const char *
2298 proc_map_files_get_link(struct dentry *dentry,
2299                         struct inode *inode,
2300                         struct delayed_call *done)
2301 {
2302         if (!checkpoint_restore_ns_capable(&init_user_ns))
2303                 return ERR_PTR(-EPERM);
2304
2305         return proc_pid_get_link(dentry, inode, done);
2306 }
2307
2308 /*
2309  * Identical to proc_pid_link_inode_operations except for get_link()
2310  */
2311 static const struct inode_operations proc_map_files_link_inode_operations = {
2312         .readlink       = proc_pid_readlink,
2313         .get_link       = proc_map_files_get_link,
2314         .setattr        = proc_setattr,
2315 };
2316
2317 static struct dentry *
2318 proc_map_files_instantiate(struct dentry *dentry,
2319                            struct task_struct *task, const void *ptr)
2320 {
2321         fmode_t mode = (fmode_t)(unsigned long)ptr;
2322         struct proc_inode *ei;
2323         struct inode *inode;
2324
2325         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2326                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2327                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2328         if (!inode)
2329                 return ERR_PTR(-ENOENT);
2330
2331         ei = PROC_I(inode);
2332         ei->op.proc_get_link = map_files_get_link;
2333
2334         inode->i_op = &proc_map_files_link_inode_operations;
2335         inode->i_size = 64;
2336
2337         return proc_splice_unmountable(inode, dentry,
2338                                        &tid_map_files_dentry_operations);
2339 }
2340
2341 static struct dentry *proc_map_files_lookup(struct inode *dir,
2342                 struct dentry *dentry, unsigned int flags)
2343 {
2344         unsigned long vm_start, vm_end;
2345         struct vm_area_struct *vma;
2346         struct task_struct *task;
2347         struct dentry *result;
2348         struct mm_struct *mm;
2349
2350         result = ERR_PTR(-ENOENT);
2351         task = get_proc_task(dir);
2352         if (!task)
2353                 goto out;
2354
2355         result = ERR_PTR(-EACCES);
2356         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2357                 goto out_put_task;
2358
2359         result = ERR_PTR(-ENOENT);
2360         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2361                 goto out_put_task;
2362
2363         mm = get_task_mm(task);
2364         if (!mm)
2365                 goto out_put_task;
2366
2367         result = ERR_PTR(-EINTR);
2368         if (mmap_read_lock_killable(mm))
2369                 goto out_put_mm;
2370
2371         result = ERR_PTR(-ENOENT);
2372         vma = find_exact_vma(mm, vm_start, vm_end);
2373         if (!vma)
2374                 goto out_no_vma;
2375
2376         if (vma->vm_file)
2377                 result = proc_map_files_instantiate(dentry, task,
2378                                 (void *)(unsigned long)vma->vm_file->f_mode);
2379
2380 out_no_vma:
2381         mmap_read_unlock(mm);
2382 out_put_mm:
2383         mmput(mm);
2384 out_put_task:
2385         put_task_struct(task);
2386 out:
2387         return result;
2388 }
2389
2390 static const struct inode_operations proc_map_files_inode_operations = {
2391         .lookup         = proc_map_files_lookup,
2392         .permission     = proc_fd_permission,
2393         .setattr        = proc_setattr,
2394 };
2395
2396 static int
2397 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2398 {
2399         struct vm_area_struct *vma;
2400         struct task_struct *task;
2401         struct mm_struct *mm;
2402         unsigned long nr_files, pos, i;
2403         GENRADIX(struct map_files_info) fa;
2404         struct map_files_info *p;
2405         int ret;
2406         struct vma_iterator vmi;
2407
2408         genradix_init(&fa);
2409
2410         ret = -ENOENT;
2411         task = get_proc_task(file_inode(file));
2412         if (!task)
2413                 goto out;
2414
2415         ret = -EACCES;
2416         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2417                 goto out_put_task;
2418
2419         ret = 0;
2420         if (!dir_emit_dots(file, ctx))
2421                 goto out_put_task;
2422
2423         mm = get_task_mm(task);
2424         if (!mm)
2425                 goto out_put_task;
2426
2427         ret = mmap_read_lock_killable(mm);
2428         if (ret) {
2429                 mmput(mm);
2430                 goto out_put_task;
2431         }
2432
2433         nr_files = 0;
2434
2435         /*
2436          * We need two passes here:
2437          *
2438          *  1) Collect vmas of mapped files with mmap_lock taken
2439          *  2) Release mmap_lock and instantiate entries
2440          *
2441          * otherwise we get lockdep complained, since filldir()
2442          * routine might require mmap_lock taken in might_fault().
2443          */
2444
2445         pos = 2;
2446         vma_iter_init(&vmi, mm, 0);
2447         for_each_vma(vmi, vma) {
2448                 if (!vma->vm_file)
2449                         continue;
2450                 if (++pos <= ctx->pos)
2451                         continue;
2452
2453                 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2454                 if (!p) {
2455                         ret = -ENOMEM;
2456                         mmap_read_unlock(mm);
2457                         mmput(mm);
2458                         goto out_put_task;
2459                 }
2460
2461                 p->start = vma->vm_start;
2462                 p->end = vma->vm_end;
2463                 p->mode = vma->vm_file->f_mode;
2464         }
2465         mmap_read_unlock(mm);
2466         mmput(mm);
2467
2468         for (i = 0; i < nr_files; i++) {
2469                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2470                 unsigned int len;
2471
2472                 p = genradix_ptr(&fa, i);
2473                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2474                 if (!proc_fill_cache(file, ctx,
2475                                       buf, len,
2476                                       proc_map_files_instantiate,
2477                                       task,
2478                                       (void *)(unsigned long)p->mode))
2479                         break;
2480                 ctx->pos++;
2481         }
2482
2483 out_put_task:
2484         put_task_struct(task);
2485 out:
2486         genradix_free(&fa);
2487         return ret;
2488 }
2489
2490 static const struct file_operations proc_map_files_operations = {
2491         .read           = generic_read_dir,
2492         .iterate_shared = proc_map_files_readdir,
2493         .llseek         = generic_file_llseek,
2494 };
2495
2496 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2497 struct timers_private {
2498         struct pid *pid;
2499         struct task_struct *task;
2500         struct sighand_struct *sighand;
2501         struct pid_namespace *ns;
2502         unsigned long flags;
2503 };
2504
2505 static void *timers_start(struct seq_file *m, loff_t *pos)
2506 {
2507         struct timers_private *tp = m->private;
2508
2509         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2510         if (!tp->task)
2511                 return ERR_PTR(-ESRCH);
2512
2513         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2514         if (!tp->sighand)
2515                 return ERR_PTR(-ESRCH);
2516
2517         return seq_hlist_start(&tp->task->signal->posix_timers, *pos);
2518 }
2519
2520 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2521 {
2522         struct timers_private *tp = m->private;
2523         return seq_hlist_next(v, &tp->task->signal->posix_timers, pos);
2524 }
2525
2526 static void timers_stop(struct seq_file *m, void *v)
2527 {
2528         struct timers_private *tp = m->private;
2529
2530         if (tp->sighand) {
2531                 unlock_task_sighand(tp->task, &tp->flags);
2532                 tp->sighand = NULL;
2533         }
2534
2535         if (tp->task) {
2536                 put_task_struct(tp->task);
2537                 tp->task = NULL;
2538         }
2539 }
2540
2541 static int show_timer(struct seq_file *m, void *v)
2542 {
2543         struct k_itimer *timer;
2544         struct timers_private *tp = m->private;
2545         int notify;
2546         static const char * const nstr[] = {
2547                 [SIGEV_SIGNAL] = "signal",
2548                 [SIGEV_NONE] = "none",
2549                 [SIGEV_THREAD] = "thread",
2550         };
2551
2552         timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2553         notify = timer->it_sigev_notify;
2554
2555         seq_printf(m, "ID: %d\n", timer->it_id);
2556         seq_printf(m, "signal: %d/%px\n",
2557                    timer->sigq.info.si_signo,
2558                    timer->sigq.info.si_value.sival_ptr);
2559         seq_printf(m, "notify: %s/%s.%d\n",
2560                    nstr[notify & ~SIGEV_THREAD_ID],
2561                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2562                    pid_nr_ns(timer->it_pid, tp->ns));
2563         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2564
2565         return 0;
2566 }
2567
2568 static const struct seq_operations proc_timers_seq_ops = {
2569         .start  = timers_start,
2570         .next   = timers_next,
2571         .stop   = timers_stop,
2572         .show   = show_timer,
2573 };
2574
2575 static int proc_timers_open(struct inode *inode, struct file *file)
2576 {
2577         struct timers_private *tp;
2578
2579         tp = __seq_open_private(file, &proc_timers_seq_ops,
2580                         sizeof(struct timers_private));
2581         if (!tp)
2582                 return -ENOMEM;
2583
2584         tp->pid = proc_pid(inode);
2585         tp->ns = proc_pid_ns(inode->i_sb);
2586         return 0;
2587 }
2588
2589 static const struct file_operations proc_timers_operations = {
2590         .open           = proc_timers_open,
2591         .read           = seq_read,
2592         .llseek         = seq_lseek,
2593         .release        = seq_release_private,
2594 };
2595 #endif
2596
2597 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2598                                         size_t count, loff_t *offset)
2599 {
2600         struct inode *inode = file_inode(file);
2601         struct task_struct *p;
2602         u64 slack_ns;
2603         int err;
2604
2605         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2606         if (err < 0)
2607                 return err;
2608
2609         p = get_proc_task(inode);
2610         if (!p)
2611                 return -ESRCH;
2612
2613         if (p != current) {
2614                 rcu_read_lock();
2615                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2616                         rcu_read_unlock();
2617                         count = -EPERM;
2618                         goto out;
2619                 }
2620                 rcu_read_unlock();
2621
2622                 err = security_task_setscheduler(p);
2623                 if (err) {
2624                         count = err;
2625                         goto out;
2626                 }
2627         }
2628
2629         task_lock(p);
2630         if (rt_or_dl_task_policy(p))
2631                 slack_ns = 0;
2632         else if (slack_ns == 0)
2633                 slack_ns = p->default_timer_slack_ns;
2634         p->timer_slack_ns = slack_ns;
2635         task_unlock(p);
2636
2637 out:
2638         put_task_struct(p);
2639
2640         return count;
2641 }
2642
2643 static int timerslack_ns_show(struct seq_file *m, void *v)
2644 {
2645         struct inode *inode = m->private;
2646         struct task_struct *p;
2647         int err = 0;
2648
2649         p = get_proc_task(inode);
2650         if (!p)
2651                 return -ESRCH;
2652
2653         if (p != current) {
2654                 rcu_read_lock();
2655                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2656                         rcu_read_unlock();
2657                         err = -EPERM;
2658                         goto out;
2659                 }
2660                 rcu_read_unlock();
2661
2662                 err = security_task_getscheduler(p);
2663                 if (err)
2664                         goto out;
2665         }
2666
2667         task_lock(p);
2668         seq_printf(m, "%llu\n", p->timer_slack_ns);
2669         task_unlock(p);
2670
2671 out:
2672         put_task_struct(p);
2673
2674         return err;
2675 }
2676
2677 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2678 {
2679         return single_open(filp, timerslack_ns_show, inode);
2680 }
2681
2682 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2683         .open           = timerslack_ns_open,
2684         .read           = seq_read,
2685         .write          = timerslack_ns_write,
2686         .llseek         = seq_lseek,
2687         .release        = single_release,
2688 };
2689
2690 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2691         struct task_struct *task, const void *ptr)
2692 {
2693         const struct pid_entry *p = ptr;
2694         struct inode *inode;
2695         struct proc_inode *ei;
2696
2697         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2698         if (!inode)
2699                 return ERR_PTR(-ENOENT);
2700
2701         ei = PROC_I(inode);
2702         if (S_ISDIR(inode->i_mode))
2703                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2704         if (p->iop)
2705                 inode->i_op = p->iop;
2706         if (p->fop)
2707                 inode->i_fop = p->fop;
2708         ei->op = p->op;
2709         pid_update_inode(task, inode);
2710         d_set_d_op(dentry, &pid_dentry_operations);
2711         return d_splice_alias(inode, dentry);
2712 }
2713
2714 static struct dentry *proc_pident_lookup(struct inode *dir,
2715                                          struct dentry *dentry,
2716                                          const struct pid_entry *p,
2717                                          const struct pid_entry *end)
2718 {
2719         struct task_struct *task = get_proc_task(dir);
2720         struct dentry *res = ERR_PTR(-ENOENT);
2721
2722         if (!task)
2723                 goto out_no_task;
2724
2725         /*
2726          * Yes, it does not scale. And it should not. Don't add
2727          * new entries into /proc/<tgid>/ without very good reasons.
2728          */
2729         for (; p < end; p++) {
2730                 if (p->len != dentry->d_name.len)
2731                         continue;
2732                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2733                         res = proc_pident_instantiate(dentry, task, p);
2734                         break;
2735                 }
2736         }
2737         put_task_struct(task);
2738 out_no_task:
2739         return res;
2740 }
2741
2742 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2743                 const struct pid_entry *ents, unsigned int nents)
2744 {
2745         struct task_struct *task = get_proc_task(file_inode(file));
2746         const struct pid_entry *p;
2747
2748         if (!task)
2749                 return -ENOENT;
2750
2751         if (!dir_emit_dots(file, ctx))
2752                 goto out;
2753
2754         if (ctx->pos >= nents + 2)
2755                 goto out;
2756
2757         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2758                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2759                                 proc_pident_instantiate, task, p))
2760                         break;
2761                 ctx->pos++;
2762         }
2763 out:
2764         put_task_struct(task);
2765         return 0;
2766 }
2767
2768 #ifdef CONFIG_SECURITY
2769 static int proc_pid_attr_open(struct inode *inode, struct file *file)
2770 {
2771         file->private_data = NULL;
2772         __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2773         return 0;
2774 }
2775
2776 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2777                                   size_t count, loff_t *ppos)
2778 {
2779         struct inode * inode = file_inode(file);
2780         char *p = NULL;
2781         ssize_t length;
2782         struct task_struct *task = get_proc_task(inode);
2783
2784         if (!task)
2785                 return -ESRCH;
2786
2787         length = security_getprocattr(task, PROC_I(inode)->op.lsmid,
2788                                       file->f_path.dentry->d_name.name,
2789                                       &p);
2790         put_task_struct(task);
2791         if (length > 0)
2792                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2793         kfree(p);
2794         return length;
2795 }
2796
2797 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2798                                    size_t count, loff_t *ppos)
2799 {
2800         struct inode * inode = file_inode(file);
2801         struct task_struct *task;
2802         void *page;
2803         int rv;
2804
2805         /* A task may only write when it was the opener. */
2806         if (file->private_data != current->mm)
2807                 return -EPERM;
2808
2809         rcu_read_lock();
2810         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2811         if (!task) {
2812                 rcu_read_unlock();
2813                 return -ESRCH;
2814         }
2815         /* A task may only write its own attributes. */
2816         if (current != task) {
2817                 rcu_read_unlock();
2818                 return -EACCES;
2819         }
2820         /* Prevent changes to overridden credentials. */
2821         if (current_cred() != current_real_cred()) {
2822                 rcu_read_unlock();
2823                 return -EBUSY;
2824         }
2825         rcu_read_unlock();
2826
2827         if (count > PAGE_SIZE)
2828                 count = PAGE_SIZE;
2829
2830         /* No partial writes. */
2831         if (*ppos != 0)
2832                 return -EINVAL;
2833
2834         page = memdup_user(buf, count);
2835         if (IS_ERR(page)) {
2836                 rv = PTR_ERR(page);
2837                 goto out;
2838         }
2839
2840         /* Guard against adverse ptrace interaction */
2841         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2842         if (rv < 0)
2843                 goto out_free;
2844
2845         rv = security_setprocattr(PROC_I(inode)->op.lsmid,
2846                                   file->f_path.dentry->d_name.name, page,
2847                                   count);
2848         mutex_unlock(&current->signal->cred_guard_mutex);
2849 out_free:
2850         kfree(page);
2851 out:
2852         return rv;
2853 }
2854
2855 static const struct file_operations proc_pid_attr_operations = {
2856         .open           = proc_pid_attr_open,
2857         .read           = proc_pid_attr_read,
2858         .write          = proc_pid_attr_write,
2859         .llseek         = generic_file_llseek,
2860         .release        = mem_release,
2861 };
2862
2863 #define LSM_DIR_OPS(LSM) \
2864 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2865                              struct dir_context *ctx) \
2866 { \
2867         return proc_pident_readdir(filp, ctx, \
2868                                    LSM##_attr_dir_stuff, \
2869                                    ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2870 } \
2871 \
2872 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2873         .read           = generic_read_dir, \
2874         .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2875         .llseek         = default_llseek, \
2876 }; \
2877 \
2878 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2879                                 struct dentry *dentry, unsigned int flags) \
2880 { \
2881         return proc_pident_lookup(dir, dentry, \
2882                                   LSM##_attr_dir_stuff, \
2883                                   LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2884 } \
2885 \
2886 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2887         .lookup         = proc_##LSM##_attr_dir_lookup, \
2888         .getattr        = pid_getattr, \
2889         .setattr        = proc_setattr, \
2890 }
2891
2892 #ifdef CONFIG_SECURITY_SMACK
2893 static const struct pid_entry smack_attr_dir_stuff[] = {
2894         ATTR(LSM_ID_SMACK, "current",   0666),
2895 };
2896 LSM_DIR_OPS(smack);
2897 #endif
2898
2899 #ifdef CONFIG_SECURITY_APPARMOR
2900 static const struct pid_entry apparmor_attr_dir_stuff[] = {
2901         ATTR(LSM_ID_APPARMOR, "current",        0666),
2902         ATTR(LSM_ID_APPARMOR, "prev",           0444),
2903         ATTR(LSM_ID_APPARMOR, "exec",           0666),
2904 };
2905 LSM_DIR_OPS(apparmor);
2906 #endif
2907
2908 static const struct pid_entry attr_dir_stuff[] = {
2909         ATTR(LSM_ID_UNDEF, "current",   0666),
2910         ATTR(LSM_ID_UNDEF, "prev",              0444),
2911         ATTR(LSM_ID_UNDEF, "exec",              0666),
2912         ATTR(LSM_ID_UNDEF, "fscreate",  0666),
2913         ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2914         ATTR(LSM_ID_UNDEF, "sockcreate",        0666),
2915 #ifdef CONFIG_SECURITY_SMACK
2916         DIR("smack",                    0555,
2917             proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2918 #endif
2919 #ifdef CONFIG_SECURITY_APPARMOR
2920         DIR("apparmor",                 0555,
2921             proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2922 #endif
2923 };
2924
2925 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2926 {
2927         return proc_pident_readdir(file, ctx,
2928                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2929 }
2930
2931 static const struct file_operations proc_attr_dir_operations = {
2932         .read           = generic_read_dir,
2933         .iterate_shared = proc_attr_dir_readdir,
2934         .llseek         = generic_file_llseek,
2935 };
2936
2937 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2938                                 struct dentry *dentry, unsigned int flags)
2939 {
2940         return proc_pident_lookup(dir, dentry,
2941                                   attr_dir_stuff,
2942                                   attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2943 }
2944
2945 static const struct inode_operations proc_attr_dir_inode_operations = {
2946         .lookup         = proc_attr_dir_lookup,
2947         .getattr        = pid_getattr,
2948         .setattr        = proc_setattr,
2949 };
2950
2951 #endif
2952
2953 #ifdef CONFIG_ELF_CORE
2954 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2955                                          size_t count, loff_t *ppos)
2956 {
2957         struct task_struct *task = get_proc_task(file_inode(file));
2958         struct mm_struct *mm;
2959         char buffer[PROC_NUMBUF];
2960         size_t len;
2961         int ret;
2962
2963         if (!task)
2964                 return -ESRCH;
2965
2966         ret = 0;
2967         mm = get_task_mm(task);
2968         if (mm) {
2969                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2970                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2971                                 MMF_DUMP_FILTER_SHIFT));
2972                 mmput(mm);
2973                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2974         }
2975
2976         put_task_struct(task);
2977
2978         return ret;
2979 }
2980
2981 static ssize_t proc_coredump_filter_write(struct file *file,
2982                                           const char __user *buf,
2983                                           size_t count,
2984                                           loff_t *ppos)
2985 {
2986         struct task_struct *task;
2987         struct mm_struct *mm;
2988         unsigned int val;
2989         int ret;
2990         int i;
2991         unsigned long mask;
2992
2993         ret = kstrtouint_from_user(buf, count, 0, &val);
2994         if (ret < 0)
2995                 return ret;
2996
2997         ret = -ESRCH;
2998         task = get_proc_task(file_inode(file));
2999         if (!task)
3000                 goto out_no_task;
3001
3002         mm = get_task_mm(task);
3003         if (!mm)
3004                 goto out_no_mm;
3005         ret = 0;
3006
3007         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
3008                 if (val & mask)
3009                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3010                 else
3011                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3012         }
3013
3014         mmput(mm);
3015  out_no_mm:
3016         put_task_struct(task);
3017  out_no_task:
3018         if (ret < 0)
3019                 return ret;
3020         return count;
3021 }
3022
3023 static const struct file_operations proc_coredump_filter_operations = {
3024         .read           = proc_coredump_filter_read,
3025         .write          = proc_coredump_filter_write,
3026         .llseek         = generic_file_llseek,
3027 };
3028 #endif
3029
3030 #ifdef CONFIG_TASK_IO_ACCOUNTING
3031 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3032 {
3033         struct task_io_accounting acct;
3034         int result;
3035
3036         result = down_read_killable(&task->signal->exec_update_lock);
3037         if (result)
3038                 return result;
3039
3040         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3041                 result = -EACCES;
3042                 goto out_unlock;
3043         }
3044
3045         if (whole) {
3046                 struct signal_struct *sig = task->signal;
3047                 struct task_struct *t;
3048                 unsigned int seq = 1;
3049                 unsigned long flags;
3050
3051                 rcu_read_lock();
3052                 do {
3053                         seq++; /* 2 on the 1st/lockless path, otherwise odd */
3054                         flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
3055
3056                         acct = sig->ioac;
3057                         __for_each_thread(sig, t)
3058                                 task_io_accounting_add(&acct, &t->ioac);
3059
3060                 } while (need_seqretry(&sig->stats_lock, seq));
3061                 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
3062                 rcu_read_unlock();
3063         } else {
3064                 acct = task->ioac;
3065         }
3066
3067         seq_printf(m,
3068                    "rchar: %llu\n"
3069                    "wchar: %llu\n"
3070                    "syscr: %llu\n"
3071                    "syscw: %llu\n"
3072                    "read_bytes: %llu\n"
3073                    "write_bytes: %llu\n"
3074                    "cancelled_write_bytes: %llu\n",
3075                    (unsigned long long)acct.rchar,
3076                    (unsigned long long)acct.wchar,
3077                    (unsigned long long)acct.syscr,
3078                    (unsigned long long)acct.syscw,
3079                    (unsigned long long)acct.read_bytes,
3080                    (unsigned long long)acct.write_bytes,
3081                    (unsigned long long)acct.cancelled_write_bytes);
3082         result = 0;
3083
3084 out_unlock:
3085         up_read(&task->signal->exec_update_lock);
3086         return result;
3087 }
3088
3089 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3090                                   struct pid *pid, struct task_struct *task)
3091 {
3092         return do_io_accounting(task, m, 0);
3093 }
3094
3095 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3096                                    struct pid *pid, struct task_struct *task)
3097 {
3098         return do_io_accounting(task, m, 1);
3099 }
3100 #endif /* CONFIG_TASK_IO_ACCOUNTING */
3101
3102 #ifdef CONFIG_USER_NS
3103 static int proc_id_map_open(struct inode *inode, struct file *file,
3104         const struct seq_operations *seq_ops)
3105 {
3106         struct user_namespace *ns = NULL;
3107         struct task_struct *task;
3108         struct seq_file *seq;
3109         int ret = -EINVAL;
3110
3111         task = get_proc_task(inode);
3112         if (task) {
3113                 rcu_read_lock();
3114                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3115                 rcu_read_unlock();
3116                 put_task_struct(task);
3117         }
3118         if (!ns)
3119                 goto err;
3120
3121         ret = seq_open(file, seq_ops);
3122         if (ret)
3123                 goto err_put_ns;
3124
3125         seq = file->private_data;
3126         seq->private = ns;
3127
3128         return 0;
3129 err_put_ns:
3130         put_user_ns(ns);
3131 err:
3132         return ret;
3133 }
3134
3135 static int proc_id_map_release(struct inode *inode, struct file *file)
3136 {
3137         struct seq_file *seq = file->private_data;
3138         struct user_namespace *ns = seq->private;
3139         put_user_ns(ns);
3140         return seq_release(inode, file);
3141 }
3142
3143 static int proc_uid_map_open(struct inode *inode, struct file *file)
3144 {
3145         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3146 }
3147
3148 static int proc_gid_map_open(struct inode *inode, struct file *file)
3149 {
3150         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3151 }
3152
3153 static int proc_projid_map_open(struct inode *inode, struct file *file)
3154 {
3155         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3156 }
3157
3158 static const struct file_operations proc_uid_map_operations = {
3159         .open           = proc_uid_map_open,
3160         .write          = proc_uid_map_write,
3161         .read           = seq_read,
3162         .llseek         = seq_lseek,
3163         .release        = proc_id_map_release,
3164 };
3165
3166 static const struct file_operations proc_gid_map_operations = {
3167         .open           = proc_gid_map_open,
3168         .write          = proc_gid_map_write,
3169         .read           = seq_read,
3170         .llseek         = seq_lseek,
3171         .release        = proc_id_map_release,
3172 };
3173
3174 static const struct file_operations proc_projid_map_operations = {
3175         .open           = proc_projid_map_open,
3176         .write          = proc_projid_map_write,
3177         .read           = seq_read,
3178         .llseek         = seq_lseek,
3179         .release        = proc_id_map_release,
3180 };
3181
3182 static int proc_setgroups_open(struct inode *inode, struct file *file)
3183 {
3184         struct user_namespace *ns = NULL;
3185         struct task_struct *task;
3186         int ret;
3187
3188         ret = -ESRCH;
3189         task = get_proc_task(inode);
3190         if (task) {
3191                 rcu_read_lock();
3192                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3193                 rcu_read_unlock();
3194                 put_task_struct(task);
3195         }
3196         if (!ns)
3197                 goto err;
3198
3199         if (file->f_mode & FMODE_WRITE) {
3200                 ret = -EACCES;
3201                 if (!ns_capable(ns, CAP_SYS_ADMIN))
3202                         goto err_put_ns;
3203         }
3204
3205         ret = single_open(file, &proc_setgroups_show, ns);
3206         if (ret)
3207                 goto err_put_ns;
3208
3209         return 0;
3210 err_put_ns:
3211         put_user_ns(ns);
3212 err:
3213         return ret;
3214 }
3215
3216 static int proc_setgroups_release(struct inode *inode, struct file *file)
3217 {
3218         struct seq_file *seq = file->private_data;
3219         struct user_namespace *ns = seq->private;
3220         int ret = single_release(inode, file);
3221         put_user_ns(ns);
3222         return ret;
3223 }
3224
3225 static const struct file_operations proc_setgroups_operations = {
3226         .open           = proc_setgroups_open,
3227         .write          = proc_setgroups_write,
3228         .read           = seq_read,
3229         .llseek         = seq_lseek,
3230         .release        = proc_setgroups_release,
3231 };
3232 #endif /* CONFIG_USER_NS */
3233
3234 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3235                                 struct pid *pid, struct task_struct *task)
3236 {
3237         int err = lock_trace(task);
3238         if (!err) {
3239                 seq_printf(m, "%08x\n", task->personality);
3240                 unlock_trace(task);
3241         }
3242         return err;
3243 }
3244
3245 #ifdef CONFIG_LIVEPATCH
3246 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3247                                 struct pid *pid, struct task_struct *task)
3248 {
3249         seq_printf(m, "%d\n", task->patch_state);
3250         return 0;
3251 }
3252 #endif /* CONFIG_LIVEPATCH */
3253
3254 #ifdef CONFIG_KSM
3255 static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3256                                 struct pid *pid, struct task_struct *task)
3257 {
3258         struct mm_struct *mm;
3259
3260         mm = get_task_mm(task);
3261         if (mm) {
3262                 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3263                 mmput(mm);
3264         }
3265
3266         return 0;
3267 }
3268 static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3269                                 struct pid *pid, struct task_struct *task)
3270 {
3271         struct mm_struct *mm;
3272
3273         mm = get_task_mm(task);
3274         if (mm) {
3275                 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3276                 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3277                 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3278                 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3279                 mmput(mm);
3280         }
3281
3282         return 0;
3283 }
3284 #endif /* CONFIG_KSM */
3285
3286 #ifdef CONFIG_STACKLEAK_METRICS
3287 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3288                                 struct pid *pid, struct task_struct *task)
3289 {
3290         unsigned long prev_depth = THREAD_SIZE -
3291                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3292         unsigned long depth = THREAD_SIZE -
3293                                 (task->lowest_stack & (THREAD_SIZE - 1));
3294
3295         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3296                                                         prev_depth, depth);
3297         return 0;
3298 }
3299 #endif /* CONFIG_STACKLEAK_METRICS */
3300
3301 /*
3302  * Thread groups
3303  */
3304 static const struct file_operations proc_task_operations;
3305 static const struct inode_operations proc_task_inode_operations;
3306
3307 static const struct pid_entry tgid_base_stuff[] = {
3308         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3309         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3310         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3311         DIR("fdinfo",     S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3312         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3313 #ifdef CONFIG_NET
3314         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3315 #endif
3316         REG("environ",    S_IRUSR, proc_environ_operations),
3317         REG("auxv",       S_IRUSR, proc_auxv_operations),
3318         ONE("status",     S_IRUGO, proc_pid_status),
3319         ONE("personality", S_IRUSR, proc_pid_personality),
3320         ONE("limits",     S_IRUGO, proc_pid_limits),
3321 #ifdef CONFIG_SCHED_DEBUG
3322         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3323 #endif
3324 #ifdef CONFIG_SCHED_AUTOGROUP
3325         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3326 #endif
3327 #ifdef CONFIG_TIME_NS
3328         REG("timens_offsets",  S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3329 #endif
3330         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3331 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3332         ONE("syscall",    S_IRUSR, proc_pid_syscall),
3333 #endif
3334         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3335         ONE("stat",       S_IRUGO, proc_tgid_stat),
3336         ONE("statm",      S_IRUGO, proc_pid_statm),
3337         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3338 #ifdef CONFIG_NUMA
3339         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3340 #endif
3341         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3342         LNK("cwd",        proc_cwd_link),
3343         LNK("root",       proc_root_link),
3344         LNK("exe",        proc_exe_link),
3345         REG("mounts",     S_IRUGO, proc_mounts_operations),
3346         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3347         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3348 #ifdef CONFIG_PROC_PAGE_MONITOR
3349         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3350         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3351         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3352         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3353 #endif
3354 #ifdef CONFIG_SECURITY
3355         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3356 #endif
3357 #ifdef CONFIG_KALLSYMS
3358         ONE("wchan",      S_IRUGO, proc_pid_wchan),
3359 #endif
3360 #ifdef CONFIG_STACKTRACE
3361         ONE("stack",      S_IRUSR, proc_pid_stack),
3362 #endif
3363 #ifdef CONFIG_SCHED_INFO
3364         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3365 #endif
3366 #ifdef CONFIG_LATENCYTOP
3367         REG("latency",  S_IRUGO, proc_lstats_operations),
3368 #endif
3369 #ifdef CONFIG_PROC_PID_CPUSET
3370         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3371 #endif
3372 #ifdef CONFIG_CGROUPS
3373         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3374 #endif
3375 #ifdef CONFIG_PROC_CPU_RESCTRL
3376         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3377 #endif
3378         ONE("oom_score",  S_IRUGO, proc_oom_score),
3379         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3380         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3381 #ifdef CONFIG_AUDIT
3382         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3383         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3384 #endif
3385 #ifdef CONFIG_FAULT_INJECTION
3386         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3387         REG("fail-nth", 0644, proc_fail_nth_operations),
3388 #endif
3389 #ifdef CONFIG_ELF_CORE
3390         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3391 #endif
3392 #ifdef CONFIG_TASK_IO_ACCOUNTING
3393         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3394 #endif
3395 #ifdef CONFIG_USER_NS
3396         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3397         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3398         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3399         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3400 #endif
3401 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3402         REG("timers",     S_IRUGO, proc_timers_operations),
3403 #endif
3404         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3405 #ifdef CONFIG_LIVEPATCH
3406         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3407 #endif
3408 #ifdef CONFIG_STACKLEAK_METRICS
3409         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3410 #endif
3411 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3412         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3413 #endif
3414 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3415         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3416 #endif
3417 #ifdef CONFIG_KSM
3418         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3419         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3420 #endif
3421 };
3422
3423 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3424 {
3425         return proc_pident_readdir(file, ctx,
3426                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3427 }
3428
3429 static const struct file_operations proc_tgid_base_operations = {
3430         .read           = generic_read_dir,
3431         .iterate_shared = proc_tgid_base_readdir,
3432         .llseek         = generic_file_llseek,
3433 };
3434
3435 struct pid *tgid_pidfd_to_pid(const struct file *file)
3436 {
3437         if (file->f_op != &proc_tgid_base_operations)
3438                 return ERR_PTR(-EBADF);
3439
3440         return proc_pid(file_inode(file));
3441 }
3442
3443 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3444 {
3445         return proc_pident_lookup(dir, dentry,
3446                                   tgid_base_stuff,
3447                                   tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3448 }
3449
3450 static const struct inode_operations proc_tgid_base_inode_operations = {
3451         .lookup         = proc_tgid_base_lookup,
3452         .getattr        = pid_getattr,
3453         .setattr        = proc_setattr,
3454         .permission     = proc_pid_permission,
3455 };
3456
3457 /**
3458  * proc_flush_pid -  Remove dcache entries for @pid from the /proc dcache.
3459  * @pid: pid that should be flushed.
3460  *
3461  * This function walks a list of inodes (that belong to any proc
3462  * filesystem) that are attached to the pid and flushes them from
3463  * the dentry cache.
3464  *
3465  * It is safe and reasonable to cache /proc entries for a task until
3466  * that task exits.  After that they just clog up the dcache with
3467  * useless entries, possibly causing useful dcache entries to be
3468  * flushed instead.  This routine is provided to flush those useless
3469  * dcache entries when a process is reaped.
3470  *
3471  * NOTE: This routine is just an optimization so it does not guarantee
3472  *       that no dcache entries will exist after a process is reaped
3473  *       it just makes it very unlikely that any will persist.
3474  */
3475
3476 void proc_flush_pid(struct pid *pid)
3477 {
3478         proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3479 }
3480
3481 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3482                                    struct task_struct *task, const void *ptr)
3483 {
3484         struct inode *inode;
3485
3486         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3487                                          S_IFDIR | S_IRUGO | S_IXUGO);
3488         if (!inode)
3489                 return ERR_PTR(-ENOENT);
3490
3491         inode->i_op = &proc_tgid_base_inode_operations;
3492         inode->i_fop = &proc_tgid_base_operations;
3493         inode->i_flags|=S_IMMUTABLE;
3494
3495         set_nlink(inode, nlink_tgid);
3496         pid_update_inode(task, inode);
3497
3498         d_set_d_op(dentry, &pid_dentry_operations);
3499         return d_splice_alias(inode, dentry);
3500 }
3501
3502 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3503 {
3504         struct task_struct *task;
3505         unsigned tgid;
3506         struct proc_fs_info *fs_info;
3507         struct pid_namespace *ns;
3508         struct dentry *result = ERR_PTR(-ENOENT);
3509
3510         tgid = name_to_int(&dentry->d_name);
3511         if (tgid == ~0U)
3512                 goto out;
3513
3514         fs_info = proc_sb_info(dentry->d_sb);
3515         ns = fs_info->pid_ns;
3516         rcu_read_lock();
3517         task = find_task_by_pid_ns(tgid, ns);
3518         if (task)
3519                 get_task_struct(task);
3520         rcu_read_unlock();
3521         if (!task)
3522                 goto out;
3523
3524         /* Limit procfs to only ptraceable tasks */
3525         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3526                 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3527                         goto out_put_task;
3528         }
3529
3530         result = proc_pid_instantiate(dentry, task, NULL);
3531 out_put_task:
3532         put_task_struct(task);
3533 out:
3534         return result;
3535 }
3536
3537 /*
3538  * Find the first task with tgid >= tgid
3539  *
3540  */
3541 struct tgid_iter {
3542         unsigned int tgid;
3543         struct task_struct *task;
3544 };
3545 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3546 {
3547         struct pid *pid;
3548
3549         if (iter.task)
3550                 put_task_struct(iter.task);
3551         rcu_read_lock();
3552 retry:
3553         iter.task = NULL;
3554         pid = find_ge_pid(iter.tgid, ns);
3555         if (pid) {
3556                 iter.tgid = pid_nr_ns(pid, ns);
3557                 iter.task = pid_task(pid, PIDTYPE_TGID);
3558                 if (!iter.task) {
3559                         iter.tgid += 1;
3560                         goto retry;
3561                 }
3562                 get_task_struct(iter.task);
3563         }
3564         rcu_read_unlock();
3565         return iter;
3566 }
3567
3568 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3569
3570 /* for the /proc/ directory itself, after non-process stuff has been done */
3571 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3572 {
3573         struct tgid_iter iter;
3574         struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3575         struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3576         loff_t pos = ctx->pos;
3577
3578         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3579                 return 0;
3580
3581         if (pos == TGID_OFFSET - 2) {
3582                 struct inode *inode = d_inode(fs_info->proc_self);
3583                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3584                         return 0;
3585                 ctx->pos = pos = pos + 1;
3586         }
3587         if (pos == TGID_OFFSET - 1) {
3588                 struct inode *inode = d_inode(fs_info->proc_thread_self);
3589                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3590                         return 0;
3591                 ctx->pos = pos = pos + 1;
3592         }
3593         iter.tgid = pos - TGID_OFFSET;
3594         iter.task = NULL;
3595         for (iter = next_tgid(ns, iter);
3596              iter.task;
3597              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3598                 char name[10 + 1];
3599                 unsigned int len;
3600
3601                 cond_resched();
3602                 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3603                         continue;
3604
3605                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3606                 ctx->pos = iter.tgid + TGID_OFFSET;
3607                 if (!proc_fill_cache(file, ctx, name, len,
3608                                      proc_pid_instantiate, iter.task, NULL)) {
3609                         put_task_struct(iter.task);
3610                         return 0;
3611                 }
3612         }
3613         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3614         return 0;
3615 }
3616
3617 /*
3618  * proc_tid_comm_permission is a special permission function exclusively
3619  * used for the node /proc/<pid>/task/<tid>/comm.
3620  * It bypasses generic permission checks in the case where a task of the same
3621  * task group attempts to access the node.
3622  * The rationale behind this is that glibc and bionic access this node for
3623  * cross thread naming (pthread_set/getname_np(!self)). However, if
3624  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3625  * which locks out the cross thread naming implementation.
3626  * This function makes sure that the node is always accessible for members of
3627  * same thread group.
3628  */
3629 static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3630                                     struct inode *inode, int mask)
3631 {
3632         bool is_same_tgroup;
3633         struct task_struct *task;
3634
3635         task = get_proc_task(inode);
3636         if (!task)
3637                 return -ESRCH;
3638         is_same_tgroup = same_thread_group(current, task);
3639         put_task_struct(task);
3640
3641         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3642                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3643                  * read or written by the members of the corresponding
3644                  * thread group.
3645                  */
3646                 return 0;
3647         }
3648
3649         return generic_permission(&nop_mnt_idmap, inode, mask);
3650 }
3651
3652 static const struct inode_operations proc_tid_comm_inode_operations = {
3653                 .setattr        = proc_setattr,
3654                 .permission     = proc_tid_comm_permission,
3655 };
3656
3657 /*
3658  * Tasks
3659  */
3660 static const struct pid_entry tid_base_stuff[] = {
3661         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3662         DIR("fdinfo",    S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3663         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3664 #ifdef CONFIG_NET
3665         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3666 #endif
3667         REG("environ",   S_IRUSR, proc_environ_operations),
3668         REG("auxv",      S_IRUSR, proc_auxv_operations),
3669         ONE("status",    S_IRUGO, proc_pid_status),
3670         ONE("personality", S_IRUSR, proc_pid_personality),
3671         ONE("limits",    S_IRUGO, proc_pid_limits),
3672 #ifdef CONFIG_SCHED_DEBUG
3673         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3674 #endif
3675         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3676                          &proc_tid_comm_inode_operations,
3677                          &proc_pid_set_comm_operations, {}),
3678 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3679         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3680 #endif
3681         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3682         ONE("stat",      S_IRUGO, proc_tid_stat),
3683         ONE("statm",     S_IRUGO, proc_pid_statm),
3684         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3685 #ifdef CONFIG_PROC_CHILDREN
3686         REG("children",  S_IRUGO, proc_tid_children_operations),
3687 #endif
3688 #ifdef CONFIG_NUMA
3689         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3690 #endif
3691         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3692         LNK("cwd",       proc_cwd_link),
3693         LNK("root",      proc_root_link),
3694         LNK("exe",       proc_exe_link),
3695         REG("mounts",    S_IRUGO, proc_mounts_operations),
3696         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3697 #ifdef CONFIG_PROC_PAGE_MONITOR
3698         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3699         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3700         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3701         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3702 #endif
3703 #ifdef CONFIG_SECURITY
3704         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3705 #endif
3706 #ifdef CONFIG_KALLSYMS
3707         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3708 #endif
3709 #ifdef CONFIG_STACKTRACE
3710         ONE("stack",      S_IRUSR, proc_pid_stack),
3711 #endif
3712 #ifdef CONFIG_SCHED_INFO
3713         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3714 #endif
3715 #ifdef CONFIG_LATENCYTOP
3716         REG("latency",  S_IRUGO, proc_lstats_operations),
3717 #endif
3718 #ifdef CONFIG_PROC_PID_CPUSET
3719         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3720 #endif
3721 #ifdef CONFIG_CGROUPS
3722         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3723 #endif
3724 #ifdef CONFIG_PROC_CPU_RESCTRL
3725         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3726 #endif
3727         ONE("oom_score", S_IRUGO, proc_oom_score),
3728         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3729         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3730 #ifdef CONFIG_AUDIT
3731         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3732         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3733 #endif
3734 #ifdef CONFIG_FAULT_INJECTION
3735         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3736         REG("fail-nth", 0644, proc_fail_nth_operations),
3737 #endif
3738 #ifdef CONFIG_TASK_IO_ACCOUNTING
3739         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3740 #endif
3741 #ifdef CONFIG_USER_NS
3742         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3743         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3744         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3745         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3746 #endif
3747 #ifdef CONFIG_LIVEPATCH
3748         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3749 #endif
3750 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3751         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3752 #endif
3753 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3754         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3755 #endif
3756 #ifdef CONFIG_KSM
3757         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3758         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3759 #endif
3760 };
3761
3762 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3763 {
3764         return proc_pident_readdir(file, ctx,
3765                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3766 }
3767
3768 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3769 {
3770         return proc_pident_lookup(dir, dentry,
3771                                   tid_base_stuff,
3772                                   tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3773 }
3774
3775 static const struct file_operations proc_tid_base_operations = {
3776         .read           = generic_read_dir,
3777         .iterate_shared = proc_tid_base_readdir,
3778         .llseek         = generic_file_llseek,
3779 };
3780
3781 static const struct inode_operations proc_tid_base_inode_operations = {
3782         .lookup         = proc_tid_base_lookup,
3783         .getattr        = pid_getattr,
3784         .setattr        = proc_setattr,
3785 };
3786
3787 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3788         struct task_struct *task, const void *ptr)
3789 {
3790         struct inode *inode;
3791         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3792                                          S_IFDIR | S_IRUGO | S_IXUGO);
3793         if (!inode)
3794                 return ERR_PTR(-ENOENT);
3795
3796         inode->i_op = &proc_tid_base_inode_operations;
3797         inode->i_fop = &proc_tid_base_operations;
3798         inode->i_flags |= S_IMMUTABLE;
3799
3800         set_nlink(inode, nlink_tid);
3801         pid_update_inode(task, inode);
3802
3803         d_set_d_op(dentry, &pid_dentry_operations);
3804         return d_splice_alias(inode, dentry);
3805 }
3806
3807 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3808 {
3809         struct task_struct *task;
3810         struct task_struct *leader = get_proc_task(dir);
3811         unsigned tid;
3812         struct proc_fs_info *fs_info;
3813         struct pid_namespace *ns;
3814         struct dentry *result = ERR_PTR(-ENOENT);
3815
3816         if (!leader)
3817                 goto out_no_task;
3818
3819         tid = name_to_int(&dentry->d_name);
3820         if (tid == ~0U)
3821                 goto out;
3822
3823         fs_info = proc_sb_info(dentry->d_sb);
3824         ns = fs_info->pid_ns;
3825         rcu_read_lock();
3826         task = find_task_by_pid_ns(tid, ns);
3827         if (task)
3828                 get_task_struct(task);
3829         rcu_read_unlock();
3830         if (!task)
3831                 goto out;
3832         if (!same_thread_group(leader, task))
3833                 goto out_drop_task;
3834
3835         result = proc_task_instantiate(dentry, task, NULL);
3836 out_drop_task:
3837         put_task_struct(task);
3838 out:
3839         put_task_struct(leader);
3840 out_no_task:
3841         return result;
3842 }
3843
3844 /*
3845  * Find the first tid of a thread group to return to user space.
3846  *
3847  * Usually this is just the thread group leader, but if the users
3848  * buffer was too small or there was a seek into the middle of the
3849  * directory we have more work todo.
3850  *
3851  * In the case of a short read we start with find_task_by_pid.
3852  *
3853  * In the case of a seek we start with the leader and walk nr
3854  * threads past it.
3855  */
3856 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3857                                         struct pid_namespace *ns)
3858 {
3859         struct task_struct *pos, *task;
3860         unsigned long nr = f_pos;
3861
3862         if (nr != f_pos)        /* 32bit overflow? */
3863                 return NULL;
3864
3865         rcu_read_lock();
3866         task = pid_task(pid, PIDTYPE_PID);
3867         if (!task)
3868                 goto fail;
3869
3870         /* Attempt to start with the tid of a thread */
3871         if (tid && nr) {
3872                 pos = find_task_by_pid_ns(tid, ns);
3873                 if (pos && same_thread_group(pos, task))
3874                         goto found;
3875         }
3876
3877         /* If nr exceeds the number of threads there is nothing todo */
3878         if (nr >= get_nr_threads(task))
3879                 goto fail;
3880
3881         /* If we haven't found our starting place yet start
3882          * with the leader and walk nr threads forward.
3883          */
3884         for_each_thread(task, pos) {
3885                 if (!nr--)
3886                         goto found;
3887         }
3888 fail:
3889         pos = NULL;
3890         goto out;
3891 found:
3892         get_task_struct(pos);
3893 out:
3894         rcu_read_unlock();
3895         return pos;
3896 }
3897
3898 /*
3899  * Find the next thread in the thread list.
3900  * Return NULL if there is an error or no next thread.
3901  *
3902  * The reference to the input task_struct is released.
3903  */
3904 static struct task_struct *next_tid(struct task_struct *start)
3905 {
3906         struct task_struct *pos = NULL;
3907         rcu_read_lock();
3908         if (pid_alive(start)) {
3909                 pos = __next_thread(start);
3910                 if (pos)
3911                         get_task_struct(pos);
3912         }
3913         rcu_read_unlock();
3914         put_task_struct(start);
3915         return pos;
3916 }
3917
3918 /* for the /proc/TGID/task/ directories */
3919 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3920 {
3921         struct inode *inode = file_inode(file);
3922         struct task_struct *task;
3923         struct pid_namespace *ns;
3924         int tid;
3925
3926         if (proc_inode_is_dead(inode))
3927                 return -ENOENT;
3928
3929         if (!dir_emit_dots(file, ctx))
3930                 return 0;
3931
3932         /* We cache the tgid value that the last readdir call couldn't
3933          * return and lseek resets it to 0.
3934          */
3935         ns = proc_pid_ns(inode->i_sb);
3936         tid = (int)(intptr_t)file->private_data;
3937         file->private_data = NULL;
3938         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3939              task;
3940              task = next_tid(task), ctx->pos++) {
3941                 char name[10 + 1];
3942                 unsigned int len;
3943
3944                 tid = task_pid_nr_ns(task, ns);
3945                 if (!tid)
3946                         continue;       /* The task has just exited. */
3947                 len = snprintf(name, sizeof(name), "%u", tid);
3948                 if (!proc_fill_cache(file, ctx, name, len,
3949                                 proc_task_instantiate, task, NULL)) {
3950                         /* returning this tgid failed, save it as the first
3951                          * pid for the next readir call */
3952                         file->private_data = (void *)(intptr_t)tid;
3953                         put_task_struct(task);
3954                         break;
3955                 }
3956         }
3957
3958         return 0;
3959 }
3960
3961 static int proc_task_getattr(struct mnt_idmap *idmap,
3962                              const struct path *path, struct kstat *stat,
3963                              u32 request_mask, unsigned int query_flags)
3964 {
3965         struct inode *inode = d_inode(path->dentry);
3966         struct task_struct *p = get_proc_task(inode);
3967         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3968
3969         if (p) {
3970                 stat->nlink += get_nr_threads(p);
3971                 put_task_struct(p);
3972         }
3973
3974         return 0;
3975 }
3976
3977 /*
3978  * proc_task_readdir() set @file->private_data to a positive integer
3979  * value, so casting that to u64 is safe. generic_llseek_cookie() will
3980  * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3981  * here to catch any unexpected change in behavior either in
3982  * proc_task_readdir() or generic_llseek_cookie().
3983  */
3984 static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3985 {
3986         u64 cookie = (u64)(intptr_t)file->private_data;
3987         loff_t off;
3988
3989         off = generic_llseek_cookie(file, offset, whence, &cookie);
3990         WARN_ON_ONCE(cookie > INT_MAX);
3991         file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3992         return off;
3993 }
3994
3995 static const struct inode_operations proc_task_inode_operations = {
3996         .lookup         = proc_task_lookup,
3997         .getattr        = proc_task_getattr,
3998         .setattr        = proc_setattr,
3999         .permission     = proc_pid_permission,
4000 };
4001
4002 static const struct file_operations proc_task_operations = {
4003         .read           = generic_read_dir,
4004         .iterate_shared = proc_task_readdir,
4005         .llseek         = proc_dir_llseek,
4006 };
4007
4008 void __init set_proc_pid_nlink(void)
4009 {
4010         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
4011         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
4012 }