fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <[email protected]>
  21  *  Edjard Mota <[email protected]>
  22  *  Ilias Biris <[email protected]>
  23  *  Mauricio Lin <[email protected]>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <[email protected]>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <[email protected]>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/generic-radix-tree.h>
  63 #include <linux/string.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/namei.h>
  66 #include <linux/mnt_namespace.h>
  67 #include <linux/mm.h>
  68 #include <linux/swap.h>
  69 #include <linux/rcupdate.h>
  70 #include <linux/kallsyms.h>
  71 #include <linux/stacktrace.h>
  72 #include <linux/resource.h>
  73 #include <linux/module.h>
  74 #include <linux/mount.h>
  75 #include <linux/security.h>
  76 #include <linux/ptrace.h>
  77 #include <linux/printk.h>
  78 #include <linux/cache.h>
  79 #include <linux/cgroup.h>
  80 #include <linux/cpuset.h>
  81 #include <linux/audit.h>
  82 #include <linux/poll.h>
  83 #include <linux/nsproxy.h>
  84 #include <linux/oom.h>
  85 #include <linux/elf.h>
  86 #include <linux/pid_namespace.h>
  87 #include <linux/user_namespace.h>
  88 #include <linux/fs_parser.h>
  89 #include <linux/fs_struct.h>
  90 #include <linux/slab.h>
  91 #include <linux/sched/autogroup.h>
  92 #include <linux/sched/mm.h>
  93 #include <linux/sched/coredump.h>
  94 #include <linux/sched/debug.h>
  95 #include <linux/sched/stat.h>
  96 #include <linux/posix-timers.h>
  97 #include <linux/time_namespace.h>
  98 #include <linux/resctrl.h>
  99 #include <linux/cn_proc.h>
 100 #include <linux/ksm.h>
 101 #include <uapi/linux/lsm.h>
 102 #include <trace/events/oom.h>
 103 #include "internal.h"
 104 #include "fd.h"
 105
 106 #include "../../lib/kstrtox.h"
 107
 108 /* NOTE:
 109  *      Implementing inode permission operations in /proc is almost
 110  *      certainly an error.  Permission checks need to happen during
 111  *      each system call not at open time.  The reason is that most of
 112  *      what we wish to check for permissions in /proc varies at runtime.
 113  *
 114  *      The classic example of a problem is opening file descriptors
 115  *      in /proc for a task before it execs a suid executable.
 116  */
 117
 118 static u8 nlink_tid __ro_after_init;
 119 static u8 nlink_tgid __ro_after_init;
 120
 121 enum proc_mem_force {
 122         PROC_MEM_FORCE_ALWAYS,
 123         PROC_MEM_FORCE_PTRACE,
 124         PROC_MEM_FORCE_NEVER
 125 };
 126
 127 static enum proc_mem_force proc_mem_force_override __ro_after_init =
 128         IS_ENABLED(CONFIG_PROC_MEM_NO_FORCE) ? PROC_MEM_FORCE_NEVER :
 129         IS_ENABLED(CONFIG_PROC_MEM_FORCE_PTRACE) ? PROC_MEM_FORCE_PTRACE :
 130         PROC_MEM_FORCE_ALWAYS;
 131
 132 static const struct constant_table proc_mem_force_table[] __initconst = {
 133         { "always", PROC_MEM_FORCE_ALWAYS },
 134         { "ptrace", PROC_MEM_FORCE_PTRACE },
 135         { "never", PROC_MEM_FORCE_NEVER },
 136         { }
 137 };
 138
 139 static int __init early_proc_mem_force_override(char *buf)
 140 {
 141         if (!buf)
 142                 return -EINVAL;
 143
 144         /*
 145          * lookup_constant() defaults to proc_mem_force_override to preseve
 146          * the initial Kconfig choice in case an invalid param gets passed.
 147          */
 148         proc_mem_force_override = lookup_constant(proc_mem_force_table,
 149                                                   buf, proc_mem_force_override);
 150
 151         return 0;
 152 }
 153 early_param("proc_mem.force_override", early_proc_mem_force_override);
 154
 155 struct pid_entry {
 156         const char *name;
 157         unsigned int len;
 158         umode_t mode;
 159         const struct inode_operations *iop;
 160         const struct file_operations *fop;
 161         union proc_op op;
 162 };
 163
 164 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 165         .name = (NAME),                                 \
 166         .len  = sizeof(NAME) - 1,                       \
 167         .mode = MODE,                                   \
 168         .iop  = IOP,                                    \
 169         .fop  = FOP,                                    \
 170         .op   = OP,                                     \
 171 }
 172
 173 #define DIR(NAME, MODE, iops, fops)     \
 174         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 175 #define LNK(NAME, get_link)                                     \
 176         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 177                 &proc_pid_link_inode_operations, NULL,          \
 178                 { .proc_get_link = get_link } )
 179 #define REG(NAME, MODE, fops)                           \
 180         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 181 #define ONE(NAME, MODE, show)                           \
 182         NOD(NAME, (S_IFREG|(MODE)),                     \
 183                 NULL, &proc_single_file_operations,     \
 184                 { .proc_show = show } )
 185 #define ATTR(LSMID, NAME, MODE)                         \
 186         NOD(NAME, (S_IFREG|(MODE)),                     \
 187                 NULL, &proc_pid_attr_operations,        \
 188                 { .lsmid = LSMID })
 189
 190 /*
 191  * Count the number of hardlinks for the pid_entry table, excluding the .
 192  * and .. links.
 193  */
 194 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 195         unsigned int n)
 196 {
 197         unsigned int i;
 198         unsigned int count;
 199
 200         count = 2;
 201         for (i = 0; i < n; ++i) {
 202                 if (S_ISDIR(entries[i].mode))
 203                         ++count;
 204         }
 205
 206         return count;
 207 }
 208
 209 static int get_task_root(struct task_struct *task, struct path *root)
 210 {
 211         int result = -ENOENT;
 212
 213         task_lock(task);
 214         if (task->fs) {
 215                 get_fs_root(task->fs, root);
 216                 result = 0;
 217         }
 218         task_unlock(task);
 219         return result;
 220 }
 221
 222 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 223 {
 224         struct task_struct *task = get_proc_task(d_inode(dentry));
 225         int result = -ENOENT;
 226
 227         if (task) {
 228                 task_lock(task);
 229                 if (task->fs) {
 230                         get_fs_pwd(task->fs, path);
 231                         result = 0;
 232                 }
 233                 task_unlock(task);
 234                 put_task_struct(task);
 235         }
 236         return result;
 237 }
 238
 239 static int proc_root_link(struct dentry *dentry, struct path *path)
 240 {
 241         struct task_struct *task = get_proc_task(d_inode(dentry));
 242         int result = -ENOENT;
 243
 244         if (task) {
 245                 result = get_task_root(task, path);
 246                 put_task_struct(task);
 247         }
 248         return result;
 249 }
 250
 251 /*
 252  * If the user used setproctitle(), we just get the string from
 253  * user space at arg_start, and limit it to a maximum of one page.
 254  */
 255 static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
 256                                 size_t count, unsigned long pos,
 257                                 unsigned long arg_start)
 258 {
 259         char *page;
 260         int ret, got;
 261
 262         if (pos >= PAGE_SIZE)
 263                 return 0;
 264
 265         page = (char *)__get_free_page(GFP_KERNEL);
 266         if (!page)
 267                 return -ENOMEM;
 268
 269         ret = 0;
 270         got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
 271         if (got > 0) {
 272                 int len = strnlen(page, got);
 273
 274                 /* Include the NUL character if it was found */
 275                 if (len < got)
 276                         len++;
 277
 278                 if (len > pos) {
 279                         len -= pos;
 280                         if (len > count)
 281                                 len = count;
 282                         len -= copy_to_user(buf, page+pos, len);
 283                         if (!len)
 284                                 len = -EFAULT;
 285                         ret = len;
 286                 }
 287         }
 288         free_page((unsigned long)page);
 289         return ret;
 290 }
 291
 292 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 293                               size_t count, loff_t *ppos)
 294 {
 295         unsigned long arg_start, arg_end, env_start, env_end;
 296         unsigned long pos, len;
 297         char *page, c;
 298
 299         /* Check if process spawned far enough to have cmdline. */
 300         if (!mm->env_end)
 301                 return 0;
 302
 303         spin_lock(&mm->arg_lock);
 304         arg_start = mm->arg_start;
 305         arg_end = mm->arg_end;
 306         env_start = mm->env_start;
 307         env_end = mm->env_end;
 308         spin_unlock(&mm->arg_lock);
 309
 310         if (arg_start >= arg_end)
 311                 return 0;
 312
 313         /*
 314          * We allow setproctitle() to overwrite the argument
 315          * strings, and overflow past the original end. But
 316          * only when it overflows into the environment area.
 317          */
 318         if (env_start != arg_end || env_end < env_start)
 319                 env_start = env_end = arg_end;
 320         len = env_end - arg_start;
 321
 322         /* We're not going to care if "*ppos" has high bits set */
 323         pos = *ppos;
 324         if (pos >= len)
 325                 return 0;
 326         if (count > len - pos)
 327                 count = len - pos;
 328         if (!count)
 329                 return 0;
 330
 331         /*
 332          * Magical special case: if the argv[] end byte is not
 333          * zero, the user has overwritten it with setproctitle(3).
 334          *
 335          * Possible future enhancement: do this only once when
 336          * pos is 0, and set a flag in the 'struct file'.
 337          */
 338         if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
 339                 return get_mm_proctitle(mm, buf, count, pos, arg_start);
 340
 341         /*
 342          * For the non-setproctitle() case we limit things strictly
 343          * to the [arg_start, arg_end[ range.
 344          */
 345         pos += arg_start;
 346         if (pos < arg_start || pos >= arg_end)
 347                 return 0;
 348         if (count > arg_end - pos)
 349                 count = arg_end - pos;
 350
 351         page = (char *)__get_free_page(GFP_KERNEL);
 352         if (!page)
 353                 return -ENOMEM;
 354
 355         len = 0;
 356         while (count) {
 357                 int got;
 358                 size_t size = min_t(size_t, PAGE_SIZE, count);
 359
 360                 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
 361                 if (got <= 0)
 362                         break;
 363                 got -= copy_to_user(buf, page, got);
 364                 if (unlikely(!got)) {
 365                         if (!len)
 366                                 len = -EFAULT;
 367                         break;
 368                 }
 369                 pos += got;
 370                 buf += got;
 371                 len += got;
 372                 count -= got;
 373         }
 374
 375         free_page((unsigned long)page);
 376         return len;
 377 }
 378
 379 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 380                                 size_t count, loff_t *pos)
 381 {
 382         struct mm_struct *mm;
 383         ssize_t ret;
 384
 385         mm = get_task_mm(tsk);
 386         if (!mm)
 387                 return 0;
 388
 389         ret = get_mm_cmdline(mm, buf, count, pos);
 390         mmput(mm);
 391         return ret;
 392 }
 393
 394 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 395                                      size_t count, loff_t *pos)
 396 {
 397         struct task_struct *tsk;
 398         ssize_t ret;
 399
 400         BUG_ON(*pos < 0);
 401
 402         tsk = get_proc_task(file_inode(file));
 403         if (!tsk)
 404                 return -ESRCH;
 405         ret = get_task_cmdline(tsk, buf, count, pos);
 406         put_task_struct(tsk);
 407         if (ret > 0)
 408                 *pos += ret;
 409         return ret;
 410 }
 411
 412 static const struct file_operations proc_pid_cmdline_ops = {
 413         .read   = proc_pid_cmdline_read,
 414         .llseek = generic_file_llseek,
 415 };
 416
 417 #ifdef CONFIG_KALLSYMS
 418 /*
 419  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 420  * Returns the resolved symbol.  If that fails, simply return the address.
 421  */
 422 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 423                           struct pid *pid, struct task_struct *task)
 424 {
 425         unsigned long wchan;
 426         char symname[KSYM_NAME_LEN];
 427
 428         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 429                 goto print0;
 430
 431         wchan = get_wchan(task);
 432         if (wchan && !lookup_symbol_name(wchan, symname)) {
 433                 seq_puts(m, symname);
 434                 return 0;
 435         }
 436
 437 print0:
 438         seq_putc(m, '0');
 439         return 0;
 440 }
 441 #endif /* CONFIG_KALLSYMS */
 442
 443 static int lock_trace(struct task_struct *task)
 444 {
 445         int err = down_read_killable(&task->signal->exec_update_lock);
 446         if (err)
 447                 return err;
 448         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 449                 up_read(&task->signal->exec_update_lock);
 450                 return -EPERM;
 451         }
 452         return 0;
 453 }
 454
 455 static void unlock_trace(struct task_struct *task)
 456 {
 457         up_read(&task->signal->exec_update_lock);
 458 }
 459
 460 #ifdef CONFIG_STACKTRACE
 461
 462 #define MAX_STACK_TRACE_DEPTH   64
 463
 464 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 465                           struct pid *pid, struct task_struct *task)
 466 {
 467         unsigned long *entries;
 468         int err;
 469
 470         /*
 471          * The ability to racily run the kernel stack unwinder on a running task
 472          * and then observe the unwinder output is scary; while it is useful for
 473          * debugging kernel issues, it can also allow an attacker to leak kernel
 474          * stack contents.
 475          * Doing this in a manner that is at least safe from races would require
 476          * some work to ensure that the remote task can not be scheduled; and
 477          * even then, this would still expose the unwinder as local attack
 478          * surface.
 479          * Therefore, this interface is restricted to root.
 480          */
 481         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 482                 return -EACCES;
 483
 484         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 485                                 GFP_KERNEL);
 486         if (!entries)
 487                 return -ENOMEM;
 488
 489         err = lock_trace(task);
 490         if (!err) {
 491                 unsigned int i, nr_entries;
 492
 493                 nr_entries = stack_trace_save_tsk(task, entries,
 494                                                   MAX_STACK_TRACE_DEPTH, 0);
 495
 496                 for (i = 0; i < nr_entries; i++) {
 497                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 498                 }
 499
 500                 unlock_trace(task);
 501         }
 502         kfree(entries);
 503
 504         return err;
 505 }
 506 #endif
 507
 508 #ifdef CONFIG_SCHED_INFO
 509 /*
 510  * Provides /proc/PID/schedstat
 511  */
 512 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 513                               struct pid *pid, struct task_struct *task)
 514 {
 515         if (unlikely(!sched_info_on()))
 516                 seq_puts(m, "0 0 0\n");
 517         else
 518                 seq_printf(m, "%llu %llu %lu\n",
 519                    (unsigned long long)task->se.sum_exec_runtime,
 520                    (unsigned long long)task->sched_info.run_delay,
 521                    task->sched_info.pcount);
 522
 523         return 0;
 524 }
 525 #endif
 526
 527 #ifdef CONFIG_LATENCYTOP
 528 static int lstats_show_proc(struct seq_file *m, void *v)
 529 {
 530         int i;
 531         struct inode *inode = m->private;
 532         struct task_struct *task = get_proc_task(inode);
 533
 534         if (!task)
 535                 return -ESRCH;
 536         seq_puts(m, "Latency Top version : v0.1\n");
 537         for (i = 0; i < LT_SAVECOUNT; i++) {
 538                 struct latency_record *lr = &task->latency_record[i];
 539                 if (lr->backtrace[0]) {
 540                         int q;
 541                         seq_printf(m, "%i %li %li",
 542                                    lr->count, lr->time, lr->max);
 543                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 544                                 unsigned long bt = lr->backtrace[q];
 545
 546                                 if (!bt)
 547                                         break;
 548                                 seq_printf(m, " %ps", (void *)bt);
 549                         }
 550                         seq_putc(m, '\n');
 551                 }
 552
 553         }
 554         put_task_struct(task);
 555         return 0;
 556 }
 557
 558 static int lstats_open(struct inode *inode, struct file *file)
 559 {
 560         return single_open(file, lstats_show_proc, inode);
 561 }
 562
 563 static ssize_t lstats_write(struct file *file, const char __user *buf,
 564                             size_t count, loff_t *offs)
 565 {
 566         struct task_struct *task = get_proc_task(file_inode(file));
 567
 568         if (!task)
 569                 return -ESRCH;
 570         clear_tsk_latency_tracing(task);
 571         put_task_struct(task);
 572
 573         return count;
 574 }
 575
 576 static const struct file_operations proc_lstats_operations = {
 577         .open           = lstats_open,
 578         .read           = seq_read,
 579         .write          = lstats_write,
 580         .llseek         = seq_lseek,
 581         .release        = single_release,
 582 };
 583
 584 #endif
 585
 586 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 587                           struct pid *pid, struct task_struct *task)
 588 {
 589         unsigned long totalpages = totalram_pages() + total_swap_pages;
 590         unsigned long points = 0;
 591         long badness;
 592
 593         badness = oom_badness(task, totalpages);
 594         /*
 595          * Special case OOM_SCORE_ADJ_MIN for all others scale the
 596          * badness value into [0, 2000] range which we have been
 597          * exporting for a long time so userspace might depend on it.
 598          */
 599         if (badness != LONG_MIN)
 600                 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
 601
 602         seq_printf(m, "%lu\n", points);
 603
 604         return 0;
 605 }
 606
 607 struct limit_names {
 608         const char *name;
 609         const char *unit;
 610 };
 611
 612 static const struct limit_names lnames[RLIM_NLIMITS] = {
 613         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 614         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 615         [RLIMIT_DATA] = {"Max data size", "bytes"},
 616         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 617         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 618         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 619         [RLIMIT_NPROC] = {"Max processes", "processes"},
 620         [RLIMIT_NOFILE] = {"Max open files", "files"},
 621         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 622         [RLIMIT_AS] = {"Max address space", "bytes"},
 623         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 624         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 625         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 626         [RLIMIT_NICE] = {"Max nice priority", NULL},
 627         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 628         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 629 };
 630
 631 /* Display limits for a process */
 632 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 633                            struct pid *pid, struct task_struct *task)
 634 {
 635         unsigned int i;
 636         unsigned long flags;
 637
 638         struct rlimit rlim[RLIM_NLIMITS];
 639
 640         if (!lock_task_sighand(task, &flags))
 641                 return 0;
 642         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 643         unlock_task_sighand(task, &flags);
 644
 645         /*
 646          * print the file header
 647          */
 648         seq_puts(m, "Limit                     "
 649                 "Soft Limit           "
 650                 "Hard Limit           "
 651                 "Units     \n");
 652
 653         for (i = 0; i < RLIM_NLIMITS; i++) {
 654                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 655                         seq_printf(m, "%-25s %-20s ",
 656                                    lnames[i].name, "unlimited");
 657                 else
 658                         seq_printf(m, "%-25s %-20lu ",
 659                                    lnames[i].name, rlim[i].rlim_cur);
 660
 661                 if (rlim[i].rlim_max == RLIM_INFINITY)
 662                         seq_printf(m, "%-20s ", "unlimited");
 663                 else
 664                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 665
 666                 if (lnames[i].unit)
 667                         seq_printf(m, "%-10s\n", lnames[i].unit);
 668                 else
 669                         seq_putc(m, '\n');
 670         }
 671
 672         return 0;
 673 }
 674
 675 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 676 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 677                             struct pid *pid, struct task_struct *task)
 678 {
 679         struct syscall_info info;
 680         u64 *args = &info.data.args[0];
 681         int res;
 682
 683         res = lock_trace(task);
 684         if (res)
 685                 return res;
 686
 687         if (task_current_syscall(task, &info))
 688                 seq_puts(m, "running\n");
 689         else if (info.data.nr < 0)
 690                 seq_printf(m, "%d 0x%llx 0x%llx\n",
 691                            info.data.nr, info.sp, info.data.instruction_pointer);
 692         else
 693                 seq_printf(m,
 694                        "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
 695                        info.data.nr,
 696                        args[0], args[1], args[2], args[3], args[4], args[5],
 697                        info.sp, info.data.instruction_pointer);
 698         unlock_trace(task);
 699
 700         return 0;
 701 }
 702 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 703
 704 /************************************************************************/
 705 /*                       Here the fs part begins                        */
 706 /************************************************************************/
 707
 708 /* permission checks */
 709 static bool proc_fd_access_allowed(struct inode *inode)
 710 {
 711         struct task_struct *task;
 712         bool allowed = false;
 713         /* Allow access to a task's file descriptors if it is us or we
 714          * may use ptrace attach to the process and find out that
 715          * information.
 716          */
 717         task = get_proc_task(inode);
 718         if (task) {
 719                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 720                 put_task_struct(task);
 721         }
 722         return allowed;
 723 }
 724
 725 int proc_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 726                  struct iattr *attr)
 727 {
 728         int error;
 729         struct inode *inode = d_inode(dentry);
 730
 731         if (attr->ia_valid & ATTR_MODE)
 732                 return -EPERM;
 733
 734         error = setattr_prepare(&nop_mnt_idmap, dentry, attr);
 735         if (error)
 736                 return error;
 737
 738         setattr_copy(&nop_mnt_idmap, inode, attr);
 739         return 0;
 740 }
 741
 742 /*
 743  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 744  * or euid/egid (for hide_pid_min=2)?
 745  */
 746 static bool has_pid_permissions(struct proc_fs_info *fs_info,
 747                                  struct task_struct *task,
 748                                  enum proc_hidepid hide_pid_min)
 749 {
 750         /*
 751          * If 'hidpid' mount option is set force a ptrace check,
 752          * we indicate that we are using a filesystem syscall
 753          * by passing PTRACE_MODE_READ_FSCREDS
 754          */
 755         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE)
 756                 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 757
 758         if (fs_info->hide_pid < hide_pid_min)
 759                 return true;
 760         if (in_group_p(fs_info->pid_gid))
 761                 return true;
 762         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 763 }
 764
 765
 766 static int proc_pid_permission(struct mnt_idmap *idmap,
 767                                struct inode *inode, int mask)
 768 {
 769         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
 770         struct task_struct *task;
 771         bool has_perms;
 772
 773         task = get_proc_task(inode);
 774         if (!task)
 775                 return -ESRCH;
 776         has_perms = has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS);
 777         put_task_struct(task);
 778
 779         if (!has_perms) {
 780                 if (fs_info->hide_pid == HIDEPID_INVISIBLE) {
 781                         /*
 782                          * Let's make getdents(), stat(), and open()
 783                          * consistent with each other.  If a process
 784                          * may not stat() a file, it shouldn't be seen
 785                          * in procfs at all.
 786                          */
 787                         return -ENOENT;
 788                 }
 789
 790                 return -EPERM;
 791         }
 792         return generic_permission(&nop_mnt_idmap, inode, mask);
 793 }
 794
 795
 796
 797 static const struct inode_operations proc_def_inode_operations = {
 798         .setattr        = proc_setattr,
 799 };
 800
 801 static int proc_single_show(struct seq_file *m, void *v)
 802 {
 803         struct inode *inode = m->private;
 804         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
 805         struct pid *pid = proc_pid(inode);
 806         struct task_struct *task;
 807         int ret;
 808
 809         task = get_pid_task(pid, PIDTYPE_PID);
 810         if (!task)
 811                 return -ESRCH;
 812
 813         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 814
 815         put_task_struct(task);
 816         return ret;
 817 }
 818
 819 static int proc_single_open(struct inode *inode, struct file *filp)
 820 {
 821         return single_open(filp, proc_single_show, inode);
 822 }
 823
 824 static const struct file_operations proc_single_file_operations = {
 825         .open           = proc_single_open,
 826         .read           = seq_read,
 827         .llseek         = seq_lseek,
 828         .release        = single_release,
 829 };
 830
 831
 832 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 833 {
 834         struct task_struct *task = get_proc_task(inode);
 835         struct mm_struct *mm = ERR_PTR(-ESRCH);
 836
 837         if (task) {
 838                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 839                 put_task_struct(task);
 840
 841                 if (!IS_ERR_OR_NULL(mm)) {
 842                         /* ensure this mm_struct can't be freed */
 843                         mmgrab(mm);
 844                         /* but do not pin its memory */
 845                         mmput(mm);
 846                 }
 847         }
 848
 849         return mm;
 850 }
 851
 852 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 853 {
 854         struct mm_struct *mm = proc_mem_open(inode, mode);
 855
 856         if (IS_ERR(mm))
 857                 return PTR_ERR(mm);
 858
 859         file->private_data = mm;
 860         return 0;
 861 }
 862
 863 static int mem_open(struct inode *inode, struct file *file)
 864 {
 865         if (WARN_ON_ONCE(!(file->f_op->fop_flags & FOP_UNSIGNED_OFFSET)))
 866                 return -EINVAL;
 867         return __mem_open(inode, file, PTRACE_MODE_ATTACH);
 868 }
 869
 870 static bool proc_mem_foll_force(struct file *file, struct mm_struct *mm)
 871 {
 872         struct task_struct *task;
 873         bool ptrace_active = false;
 874
 875         switch (proc_mem_force_override) {
 876         case PROC_MEM_FORCE_NEVER:
 877                 return false;
 878         case PROC_MEM_FORCE_PTRACE:
 879                 task = get_proc_task(file_inode(file));
 880                 if (task) {
 881                         ptrace_active = READ_ONCE(task->ptrace) &&
 882                                         READ_ONCE(task->mm) == mm &&
 883                                         READ_ONCE(task->parent) == current;
 884                         put_task_struct(task);
 885                 }
 886                 return ptrace_active;
 887         default:
 888                 return true;
 889         }
 890 }
 891
 892 static ssize_t mem_rw(struct file *file, char __user *buf,
 893                         size_t count, loff_t *ppos, int write)
 894 {
 895         struct mm_struct *mm = file->private_data;
 896         unsigned long addr = *ppos;
 897         ssize_t copied;
 898         char *page;
 899         unsigned int flags;
 900
 901         if (!mm)
 902                 return 0;
 903
 904         page = (char *)__get_free_page(GFP_KERNEL);
 905         if (!page)
 906                 return -ENOMEM;
 907
 908         copied = 0;
 909         if (!mmget_not_zero(mm))
 910                 goto free;
 911
 912         flags = write ? FOLL_WRITE : 0;
 913         if (proc_mem_foll_force(file, mm))
 914                 flags |= FOLL_FORCE;
 915
 916         while (count > 0) {
 917                 size_t this_len = min_t(size_t, count, PAGE_SIZE);
 918
 919                 if (write && copy_from_user(page, buf, this_len)) {
 920                         copied = -EFAULT;
 921                         break;
 922                 }
 923
 924                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 925                 if (!this_len) {
 926                         if (!copied)
 927                                 copied = -EIO;
 928                         break;
 929                 }
 930
 931                 if (!write && copy_to_user(buf, page, this_len)) {
 932                         copied = -EFAULT;
 933                         break;
 934                 }
 935
 936                 buf += this_len;
 937                 addr += this_len;
 938                 copied += this_len;
 939                 count -= this_len;
 940         }
 941         *ppos = addr;
 942
 943         mmput(mm);
 944 free:
 945         free_page((unsigned long) page);
 946         return copied;
 947 }
 948
 949 static ssize_t mem_read(struct file *file, char __user *buf,
 950                         size_t count, loff_t *ppos)
 951 {
 952         return mem_rw(file, buf, count, ppos, 0);
 953 }
 954
 955 static ssize_t mem_write(struct file *file, const char __user *buf,
 956                          size_t count, loff_t *ppos)
 957 {
 958         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 959 }
 960
 961 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 962 {
 963         switch (orig) {
 964         case 0:
 965                 file->f_pos = offset;
 966                 break;
 967         case 1:
 968                 file->f_pos += offset;
 969                 break;
 970         default:
 971                 return -EINVAL;
 972         }
 973         force_successful_syscall_return();
 974         return file->f_pos;
 975 }
 976
 977 static int mem_release(struct inode *inode, struct file *file)
 978 {
 979         struct mm_struct *mm = file->private_data;
 980         if (mm)
 981                 mmdrop(mm);
 982         return 0;
 983 }
 984
 985 static const struct file_operations proc_mem_operations = {
 986         .llseek         = mem_lseek,
 987         .read           = mem_read,
 988         .write          = mem_write,
 989         .open           = mem_open,
 990         .release        = mem_release,
 991         .fop_flags      = FOP_UNSIGNED_OFFSET,
 992 };
 993
 994 static int environ_open(struct inode *inode, struct file *file)
 995 {
 996         return __mem_open(inode, file, PTRACE_MODE_READ);
 997 }
 998
 999 static ssize_t environ_read(struct file *file, char __user *buf,
1000                         size_t count, loff_t *ppos)
1001 {
1002         char *page;
1003         unsigned long src = *ppos;
1004         int ret = 0;
1005         struct mm_struct *mm = file->private_data;
1006         unsigned long env_start, env_end;
1007
1008         /* Ensure the process spawned far enough to have an environment. */
1009         if (!mm || !mm->env_end)
1010                 return 0;
1011
1012         page = (char *)__get_free_page(GFP_KERNEL);
1013         if (!page)
1014                 return -ENOMEM;
1015
1016         ret = 0;
1017         if (!mmget_not_zero(mm))
1018                 goto free;
1019
1020         spin_lock(&mm->arg_lock);
1021         env_start = mm->env_start;
1022         env_end = mm->env_end;
1023         spin_unlock(&mm->arg_lock);
1024
1025         while (count > 0) {
1026                 size_t this_len, max_len;
1027                 int retval;
1028
1029                 if (src >= (env_end - env_start))
1030                         break;
1031
1032                 this_len = env_end - (env_start + src);
1033
1034                 max_len = min_t(size_t, PAGE_SIZE, count);
1035                 this_len = min(max_len, this_len);
1036
1037                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
1038
1039                 if (retval <= 0) {
1040                         ret = retval;
1041                         break;
1042                 }
1043
1044                 if (copy_to_user(buf, page, retval)) {
1045                         ret = -EFAULT;
1046                         break;
1047                 }
1048
1049                 ret += retval;
1050                 src += retval;
1051                 buf += retval;
1052                 count -= retval;
1053         }
1054         *ppos = src;
1055         mmput(mm);
1056
1057 free:
1058         free_page((unsigned long) page);
1059         return ret;
1060 }
1061
1062 static const struct file_operations proc_environ_operations = {
1063         .open           = environ_open,
1064         .read           = environ_read,
1065         .llseek         = generic_file_llseek,
1066         .release        = mem_release,
1067 };
1068
1069 static int auxv_open(struct inode *inode, struct file *file)
1070 {
1071         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1072 }
1073
1074 static ssize_t auxv_read(struct file *file, char __user *buf,
1075                         size_t count, loff_t *ppos)
1076 {
1077         struct mm_struct *mm = file->private_data;
1078         unsigned int nwords = 0;
1079
1080         if (!mm)
1081                 return 0;
1082         do {
1083                 nwords += 2;
1084         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1085         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1086                                        nwords * sizeof(mm->saved_auxv[0]));
1087 }
1088
1089 static const struct file_operations proc_auxv_operations = {
1090         .open           = auxv_open,
1091         .read           = auxv_read,
1092         .llseek         = generic_file_llseek,
1093         .release        = mem_release,
1094 };
1095
1096 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1097                             loff_t *ppos)
1098 {
1099         struct task_struct *task = get_proc_task(file_inode(file));
1100         char buffer[PROC_NUMBUF];
1101         int oom_adj = OOM_ADJUST_MIN;
1102         size_t len;
1103
1104         if (!task)
1105                 return -ESRCH;
1106         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1107                 oom_adj = OOM_ADJUST_MAX;
1108         else
1109                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1110                           OOM_SCORE_ADJ_MAX;
1111         put_task_struct(task);
1112         if (oom_adj > OOM_ADJUST_MAX)
1113                 oom_adj = OOM_ADJUST_MAX;
1114         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1115         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1116 }
1117
1118 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1119 {
1120         struct mm_struct *mm = NULL;
1121         struct task_struct *task;
1122         int err = 0;
1123
1124         task = get_proc_task(file_inode(file));
1125         if (!task)
1126                 return -ESRCH;
1127
1128         mutex_lock(&oom_adj_mutex);
1129         if (legacy) {
1130                 if (oom_adj < task->signal->oom_score_adj &&
1131                                 !capable(CAP_SYS_RESOURCE)) {
1132                         err = -EACCES;
1133                         goto err_unlock;
1134                 }
1135                 /*
1136                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1137                  * /proc/pid/oom_score_adj instead.
1138                  */
1139                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1140                           current->comm, task_pid_nr(current), task_pid_nr(task),
1141                           task_pid_nr(task));
1142         } else {
1143                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1144                                 !capable(CAP_SYS_RESOURCE)) {
1145                         err = -EACCES;
1146                         goto err_unlock;
1147                 }
1148         }
1149
1150         /*
1151          * Make sure we will check other processes sharing the mm if this is
1152          * not vfrok which wants its own oom_score_adj.
1153          * pin the mm so it doesn't go away and get reused after task_unlock
1154          */
1155         if (!task->vfork_done) {
1156                 struct task_struct *p = find_lock_task_mm(task);
1157
1158                 if (p) {
1159                         if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1160                                 mm = p->mm;
1161                                 mmgrab(mm);
1162                         }
1163                         task_unlock(p);
1164                 }
1165         }
1166
1167         task->signal->oom_score_adj = oom_adj;
1168         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1169                 task->signal->oom_score_adj_min = (short)oom_adj;
1170         trace_oom_score_adj_update(task);
1171
1172         if (mm) {
1173                 struct task_struct *p;
1174
1175                 rcu_read_lock();
1176                 for_each_process(p) {
1177                         if (same_thread_group(task, p))
1178                                 continue;
1179
1180                         /* do not touch kernel threads or the global init */
1181                         if (p->flags & PF_KTHREAD || is_global_init(p))
1182                                 continue;
1183
1184                         task_lock(p);
1185                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1186                                 p->signal->oom_score_adj = oom_adj;
1187                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1188                                         p->signal->oom_score_adj_min = (short)oom_adj;
1189                         }
1190                         task_unlock(p);
1191                 }
1192                 rcu_read_unlock();
1193                 mmdrop(mm);
1194         }
1195 err_unlock:
1196         mutex_unlock(&oom_adj_mutex);
1197         put_task_struct(task);
1198         return err;
1199 }
1200
1201 /*
1202  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1203  * kernels.  The effective policy is defined by oom_score_adj, which has a
1204  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1205  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1206  * Processes that become oom disabled via oom_adj will still be oom disabled
1207  * with this implementation.
1208  *
1209  * oom_adj cannot be removed since existing userspace binaries use it.
1210  */
1211 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1212                              size_t count, loff_t *ppos)
1213 {
1214         char buffer[PROC_NUMBUF] = {};
1215         int oom_adj;
1216         int err;
1217
1218         if (count > sizeof(buffer) - 1)
1219                 count = sizeof(buffer) - 1;
1220         if (copy_from_user(buffer, buf, count)) {
1221                 err = -EFAULT;
1222                 goto out;
1223         }
1224
1225         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1226         if (err)
1227                 goto out;
1228         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1229              oom_adj != OOM_DISABLE) {
1230                 err = -EINVAL;
1231                 goto out;
1232         }
1233
1234         /*
1235          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1236          * value is always attainable.
1237          */
1238         if (oom_adj == OOM_ADJUST_MAX)
1239                 oom_adj = OOM_SCORE_ADJ_MAX;
1240         else
1241                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1242
1243         err = __set_oom_adj(file, oom_adj, true);
1244 out:
1245         return err < 0 ? err : count;
1246 }
1247
1248 static const struct file_operations proc_oom_adj_operations = {
1249         .read           = oom_adj_read,
1250         .write          = oom_adj_write,
1251         .llseek         = generic_file_llseek,
1252 };
1253
1254 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1255                                         size_t count, loff_t *ppos)
1256 {
1257         struct task_struct *task = get_proc_task(file_inode(file));
1258         char buffer[PROC_NUMBUF];
1259         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1260         size_t len;
1261
1262         if (!task)
1263                 return -ESRCH;
1264         oom_score_adj = task->signal->oom_score_adj;
1265         put_task_struct(task);
1266         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1267         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1268 }
1269
1270 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1271                                         size_t count, loff_t *ppos)
1272 {
1273         char buffer[PROC_NUMBUF] = {};
1274         int oom_score_adj;
1275         int err;
1276
1277         if (count > sizeof(buffer) - 1)
1278                 count = sizeof(buffer) - 1;
1279         if (copy_from_user(buffer, buf, count)) {
1280                 err = -EFAULT;
1281                 goto out;
1282         }
1283
1284         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1285         if (err)
1286                 goto out;
1287         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1288                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1289                 err = -EINVAL;
1290                 goto out;
1291         }
1292
1293         err = __set_oom_adj(file, oom_score_adj, false);
1294 out:
1295         return err < 0 ? err : count;
1296 }
1297
1298 static const struct file_operations proc_oom_score_adj_operations = {
1299         .read           = oom_score_adj_read,
1300         .write          = oom_score_adj_write,
1301         .llseek         = default_llseek,
1302 };
1303
1304 #ifdef CONFIG_AUDIT
1305 #define TMPBUFLEN 11
1306 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1307                                   size_t count, loff_t *ppos)
1308 {
1309         struct inode * inode = file_inode(file);
1310         struct task_struct *task = get_proc_task(inode);
1311         ssize_t length;
1312         char tmpbuf[TMPBUFLEN];
1313
1314         if (!task)
1315                 return -ESRCH;
1316         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1317                            from_kuid(file->f_cred->user_ns,
1318                                      audit_get_loginuid(task)));
1319         put_task_struct(task);
1320         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1321 }
1322
1323 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1324                                    size_t count, loff_t *ppos)
1325 {
1326         struct inode * inode = file_inode(file);
1327         uid_t loginuid;
1328         kuid_t kloginuid;
1329         int rv;
1330
1331         /* Don't let kthreads write their own loginuid */
1332         if (current->flags & PF_KTHREAD)
1333                 return -EPERM;
1334
1335         rcu_read_lock();
1336         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1337                 rcu_read_unlock();
1338                 return -EPERM;
1339         }
1340         rcu_read_unlock();
1341
1342         if (*ppos != 0) {
1343                 /* No partial writes. */
1344                 return -EINVAL;
1345         }
1346
1347         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1348         if (rv < 0)
1349                 return rv;
1350
1351         /* is userspace tring to explicitly UNSET the loginuid? */
1352         if (loginuid == AUDIT_UID_UNSET) {
1353                 kloginuid = INVALID_UID;
1354         } else {
1355                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1356                 if (!uid_valid(kloginuid))
1357                         return -EINVAL;
1358         }
1359
1360         rv = audit_set_loginuid(kloginuid);
1361         if (rv < 0)
1362                 return rv;
1363         return count;
1364 }
1365
1366 static const struct file_operations proc_loginuid_operations = {
1367         .read           = proc_loginuid_read,
1368         .write          = proc_loginuid_write,
1369         .llseek         = generic_file_llseek,
1370 };
1371
1372 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1373                                   size_t count, loff_t *ppos)
1374 {
1375         struct inode * inode = file_inode(file);
1376         struct task_struct *task = get_proc_task(inode);
1377         ssize_t length;
1378         char tmpbuf[TMPBUFLEN];
1379
1380         if (!task)
1381                 return -ESRCH;
1382         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1383                                 audit_get_sessionid(task));
1384         put_task_struct(task);
1385         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1386 }
1387
1388 static const struct file_operations proc_sessionid_operations = {
1389         .read           = proc_sessionid_read,
1390         .llseek         = generic_file_llseek,
1391 };
1392 #endif
1393
1394 #ifdef CONFIG_FAULT_INJECTION
1395 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1396                                       size_t count, loff_t *ppos)
1397 {
1398         struct task_struct *task = get_proc_task(file_inode(file));
1399         char buffer[PROC_NUMBUF];
1400         size_t len;
1401         int make_it_fail;
1402
1403         if (!task)
1404                 return -ESRCH;
1405         make_it_fail = task->make_it_fail;
1406         put_task_struct(task);
1407
1408         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1409
1410         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1411 }
1412
1413 static ssize_t proc_fault_inject_write(struct file * file,
1414                         const char __user * buf, size_t count, loff_t *ppos)
1415 {
1416         struct task_struct *task;
1417         char buffer[PROC_NUMBUF] = {};
1418         int make_it_fail;
1419         int rv;
1420
1421         if (!capable(CAP_SYS_RESOURCE))
1422                 return -EPERM;
1423
1424         if (count > sizeof(buffer) - 1)
1425                 count = sizeof(buffer) - 1;
1426         if (copy_from_user(buffer, buf, count))
1427                 return -EFAULT;
1428         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1429         if (rv < 0)
1430                 return rv;
1431         if (make_it_fail < 0 || make_it_fail > 1)
1432                 return -EINVAL;
1433
1434         task = get_proc_task(file_inode(file));
1435         if (!task)
1436                 return -ESRCH;
1437         task->make_it_fail = make_it_fail;
1438         put_task_struct(task);
1439
1440         return count;
1441 }
1442
1443 static const struct file_operations proc_fault_inject_operations = {
1444         .read           = proc_fault_inject_read,
1445         .write          = proc_fault_inject_write,
1446         .llseek         = generic_file_llseek,
1447 };
1448
1449 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1450                                    size_t count, loff_t *ppos)
1451 {
1452         struct task_struct *task;
1453         int err;
1454         unsigned int n;
1455
1456         err = kstrtouint_from_user(buf, count, 0, &n);
1457         if (err)
1458                 return err;
1459
1460         task = get_proc_task(file_inode(file));
1461         if (!task)
1462                 return -ESRCH;
1463         task->fail_nth = n;
1464         put_task_struct(task);
1465
1466         return count;
1467 }
1468
1469 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1470                                   size_t count, loff_t *ppos)
1471 {
1472         struct task_struct *task;
1473         char numbuf[PROC_NUMBUF];
1474         ssize_t len;
1475
1476         task = get_proc_task(file_inode(file));
1477         if (!task)
1478                 return -ESRCH;
1479         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1480         put_task_struct(task);
1481         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1482 }
1483
1484 static const struct file_operations proc_fail_nth_operations = {
1485         .read           = proc_fail_nth_read,
1486         .write          = proc_fail_nth_write,
1487 };
1488 #endif
1489
1490
1491 #ifdef CONFIG_SCHED_DEBUG
1492 /*
1493  * Print out various scheduling related per-task fields:
1494  */
1495 static int sched_show(struct seq_file *m, void *v)
1496 {
1497         struct inode *inode = m->private;
1498         struct pid_namespace *ns = proc_pid_ns(inode->i_sb);
1499         struct task_struct *p;
1500
1501         p = get_proc_task(inode);
1502         if (!p)
1503                 return -ESRCH;
1504         proc_sched_show_task(p, ns, m);
1505
1506         put_task_struct(p);
1507
1508         return 0;
1509 }
1510
1511 static ssize_t
1512 sched_write(struct file *file, const char __user *buf,
1513             size_t count, loff_t *offset)
1514 {
1515         struct inode *inode = file_inode(file);
1516         struct task_struct *p;
1517
1518         p = get_proc_task(inode);
1519         if (!p)
1520                 return -ESRCH;
1521         proc_sched_set_task(p);
1522
1523         put_task_struct(p);
1524
1525         return count;
1526 }
1527
1528 static int sched_open(struct inode *inode, struct file *filp)
1529 {
1530         return single_open(filp, sched_show, inode);
1531 }
1532
1533 static const struct file_operations proc_pid_sched_operations = {
1534         .open           = sched_open,
1535         .read           = seq_read,
1536         .write          = sched_write,
1537         .llseek         = seq_lseek,
1538         .release        = single_release,
1539 };
1540
1541 #endif
1542
1543 #ifdef CONFIG_SCHED_AUTOGROUP
1544 /*
1545  * Print out autogroup related information:
1546  */
1547 static int sched_autogroup_show(struct seq_file *m, void *v)
1548 {
1549         struct inode *inode = m->private;
1550         struct task_struct *p;
1551
1552         p = get_proc_task(inode);
1553         if (!p)
1554                 return -ESRCH;
1555         proc_sched_autogroup_show_task(p, m);
1556
1557         put_task_struct(p);
1558
1559         return 0;
1560 }
1561
1562 static ssize_t
1563 sched_autogroup_write(struct file *file, const char __user *buf,
1564             size_t count, loff_t *offset)
1565 {
1566         struct inode *inode = file_inode(file);
1567         struct task_struct *p;
1568         char buffer[PROC_NUMBUF] = {};
1569         int nice;
1570         int err;
1571
1572         if (count > sizeof(buffer) - 1)
1573                 count = sizeof(buffer) - 1;
1574         if (copy_from_user(buffer, buf, count))
1575                 return -EFAULT;
1576
1577         err = kstrtoint(strstrip(buffer), 0, &nice);
1578         if (err < 0)
1579                 return err;
1580
1581         p = get_proc_task(inode);
1582         if (!p)
1583                 return -ESRCH;
1584
1585         err = proc_sched_autogroup_set_nice(p, nice);
1586         if (err)
1587                 count = err;
1588
1589         put_task_struct(p);
1590
1591         return count;
1592 }
1593
1594 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1595 {
1596         int ret;
1597
1598         ret = single_open(filp, sched_autogroup_show, NULL);
1599         if (!ret) {
1600                 struct seq_file *m = filp->private_data;
1601
1602                 m->private = inode;
1603         }
1604         return ret;
1605 }
1606
1607 static const struct file_operations proc_pid_sched_autogroup_operations = {
1608         .open           = sched_autogroup_open,
1609         .read           = seq_read,
1610         .write          = sched_autogroup_write,
1611         .llseek         = seq_lseek,
1612         .release        = single_release,
1613 };
1614
1615 #endif /* CONFIG_SCHED_AUTOGROUP */
1616
1617 #ifdef CONFIG_TIME_NS
1618 static int timens_offsets_show(struct seq_file *m, void *v)
1619 {
1620         struct task_struct *p;
1621
1622         p = get_proc_task(file_inode(m->file));
1623         if (!p)
1624                 return -ESRCH;
1625         proc_timens_show_offsets(p, m);
1626
1627         put_task_struct(p);
1628
1629         return 0;
1630 }
1631
1632 static ssize_t timens_offsets_write(struct file *file, const char __user *buf,
1633                                     size_t count, loff_t *ppos)
1634 {
1635         struct inode *inode = file_inode(file);
1636         struct proc_timens_offset offsets[2];
1637         char *kbuf = NULL, *pos, *next_line;
1638         struct task_struct *p;
1639         int ret, noffsets;
1640
1641         /* Only allow < page size writes at the beginning of the file */
1642         if ((*ppos != 0) || (count >= PAGE_SIZE))
1643                 return -EINVAL;
1644
1645         /* Slurp in the user data */
1646         kbuf = memdup_user_nul(buf, count);
1647         if (IS_ERR(kbuf))
1648                 return PTR_ERR(kbuf);
1649
1650         /* Parse the user data */
1651         ret = -EINVAL;
1652         noffsets = 0;
1653         for (pos = kbuf; pos; pos = next_line) {
1654                 struct proc_timens_offset *off = &offsets[noffsets];
1655                 char clock[10];
1656                 int err;
1657
1658                 /* Find the end of line and ensure we don't look past it */
1659                 next_line = strchr(pos, '\n');
1660                 if (next_line) {
1661                         *next_line = '\0';
1662                         next_line++;
1663                         if (*next_line == '\0')
1664                                 next_line = NULL;
1665                 }
1666
1667                 err = sscanf(pos, "%9s %lld %lu", clock,
1668                                 &off->val.tv_sec, &off->val.tv_nsec);
1669                 if (err != 3 || off->val.tv_nsec >= NSEC_PER_SEC)
1670                         goto out;
1671
1672                 clock[sizeof(clock) - 1] = 0;
1673                 if (strcmp(clock, "monotonic") == 0 ||
1674                     strcmp(clock, __stringify(CLOCK_MONOTONIC)) == 0)
1675                         off->clockid = CLOCK_MONOTONIC;
1676                 else if (strcmp(clock, "boottime") == 0 ||
1677                          strcmp(clock, __stringify(CLOCK_BOOTTIME)) == 0)
1678                         off->clockid = CLOCK_BOOTTIME;
1679                 else
1680                         goto out;
1681
1682                 noffsets++;
1683                 if (noffsets == ARRAY_SIZE(offsets)) {
1684                         if (next_line)
1685                                 count = next_line - kbuf;
1686                         break;
1687                 }
1688         }
1689
1690         ret = -ESRCH;
1691         p = get_proc_task(inode);
1692         if (!p)
1693                 goto out;
1694         ret = proc_timens_set_offset(file, p, offsets, noffsets);
1695         put_task_struct(p);
1696         if (ret)
1697                 goto out;
1698
1699         ret = count;
1700 out:
1701         kfree(kbuf);
1702         return ret;
1703 }
1704
1705 static int timens_offsets_open(struct inode *inode, struct file *filp)
1706 {
1707         return single_open(filp, timens_offsets_show, inode);
1708 }
1709
1710 static const struct file_operations proc_timens_offsets_operations = {
1711         .open           = timens_offsets_open,
1712         .read           = seq_read,
1713         .write          = timens_offsets_write,
1714         .llseek         = seq_lseek,
1715         .release        = single_release,
1716 };
1717 #endif /* CONFIG_TIME_NS */
1718
1719 static ssize_t comm_write(struct file *file, const char __user *buf,
1720                                 size_t count, loff_t *offset)
1721 {
1722         struct inode *inode = file_inode(file);
1723         struct task_struct *p;
1724         char buffer[TASK_COMM_LEN] = {};
1725         const size_t maxlen = sizeof(buffer) - 1;
1726
1727         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1728                 return -EFAULT;
1729
1730         p = get_proc_task(inode);
1731         if (!p)
1732                 return -ESRCH;
1733
1734         if (same_thread_group(current, p)) {
1735                 set_task_comm(p, buffer);
1736                 proc_comm_connector(p);
1737         }
1738         else
1739                 count = -EINVAL;
1740
1741         put_task_struct(p);
1742
1743         return count;
1744 }
1745
1746 static int comm_show(struct seq_file *m, void *v)
1747 {
1748         struct inode *inode = m->private;
1749         struct task_struct *p;
1750
1751         p = get_proc_task(inode);
1752         if (!p)
1753                 return -ESRCH;
1754
1755         proc_task_name(m, p, false);
1756         seq_putc(m, '\n');
1757
1758         put_task_struct(p);
1759
1760         return 0;
1761 }
1762
1763 static int comm_open(struct inode *inode, struct file *filp)
1764 {
1765         return single_open(filp, comm_show, inode);
1766 }
1767
1768 static const struct file_operations proc_pid_set_comm_operations = {
1769         .open           = comm_open,
1770         .read           = seq_read,
1771         .write          = comm_write,
1772         .llseek         = seq_lseek,
1773         .release        = single_release,
1774 };
1775
1776 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1777 {
1778         struct task_struct *task;
1779         struct file *exe_file;
1780
1781         task = get_proc_task(d_inode(dentry));
1782         if (!task)
1783                 return -ENOENT;
1784         exe_file = get_task_exe_file(task);
1785         put_task_struct(task);
1786         if (exe_file) {
1787                 *exe_path = exe_file->f_path;
1788                 path_get(&exe_file->f_path);
1789                 fput(exe_file);
1790                 return 0;
1791         } else
1792                 return -ENOENT;
1793 }
1794
1795 static const char *proc_pid_get_link(struct dentry *dentry,
1796                                      struct inode *inode,
1797                                      struct delayed_call *done)
1798 {
1799         struct path path;
1800         int error = -EACCES;
1801
1802         if (!dentry)
1803                 return ERR_PTR(-ECHILD);
1804
1805         /* Are we allowed to snoop on the tasks file descriptors? */
1806         if (!proc_fd_access_allowed(inode))
1807                 goto out;
1808
1809         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1810         if (error)
1811                 goto out;
1812
1813         error = nd_jump_link(&path);
1814 out:
1815         return ERR_PTR(error);
1816 }
1817
1818 static int do_proc_readlink(const struct path *path, char __user *buffer, int buflen)
1819 {
1820         char *tmp = kmalloc(PATH_MAX, GFP_KERNEL);
1821         char *pathname;
1822         int len;
1823
1824         if (!tmp)
1825                 return -ENOMEM;
1826
1827         pathname = d_path(path, tmp, PATH_MAX);
1828         len = PTR_ERR(pathname);
1829         if (IS_ERR(pathname))
1830                 goto out;
1831         len = tmp + PATH_MAX - 1 - pathname;
1832
1833         if (len > buflen)
1834                 len = buflen;
1835         if (copy_to_user(buffer, pathname, len))
1836                 len = -EFAULT;
1837  out:
1838         kfree(tmp);
1839         return len;
1840 }
1841
1842 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1843 {
1844         int error = -EACCES;
1845         struct inode *inode = d_inode(dentry);
1846         struct path path;
1847
1848         /* Are we allowed to snoop on the tasks file descriptors? */
1849         if (!proc_fd_access_allowed(inode))
1850                 goto out;
1851
1852         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1853         if (error)
1854                 goto out;
1855
1856         error = do_proc_readlink(&path, buffer, buflen);
1857         path_put(&path);
1858 out:
1859         return error;
1860 }
1861
1862 const struct inode_operations proc_pid_link_inode_operations = {
1863         .readlink       = proc_pid_readlink,
1864         .get_link       = proc_pid_get_link,
1865         .setattr        = proc_setattr,
1866 };
1867
1868
1869 /* building an inode */
1870
1871 void task_dump_owner(struct task_struct *task, umode_t mode,
1872                      kuid_t *ruid, kgid_t *rgid)
1873 {
1874         /* Depending on the state of dumpable compute who should own a
1875          * proc file for a task.
1876          */
1877         const struct cred *cred;
1878         kuid_t uid;
1879         kgid_t gid;
1880
1881         if (unlikely(task->flags & PF_KTHREAD)) {
1882                 *ruid = GLOBAL_ROOT_UID;
1883                 *rgid = GLOBAL_ROOT_GID;
1884                 return;
1885         }
1886
1887         /* Default to the tasks effective ownership */
1888         rcu_read_lock();
1889         cred = __task_cred(task);
1890         uid = cred->euid;
1891         gid = cred->egid;
1892         rcu_read_unlock();
1893
1894         /*
1895          * Before the /proc/pid/status file was created the only way to read
1896          * the effective uid of a /process was to stat /proc/pid.  Reading
1897          * /proc/pid/status is slow enough that procps and other packages
1898          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1899          * made this apply to all per process world readable and executable
1900          * directories.
1901          */
1902         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1903                 struct mm_struct *mm;
1904                 task_lock(task);
1905                 mm = task->mm;
1906                 /* Make non-dumpable tasks owned by some root */
1907                 if (mm) {
1908                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1909                                 struct user_namespace *user_ns = mm->user_ns;
1910
1911                                 uid = make_kuid(user_ns, 0);
1912                                 if (!uid_valid(uid))
1913                                         uid = GLOBAL_ROOT_UID;
1914
1915                                 gid = make_kgid(user_ns, 0);
1916                                 if (!gid_valid(gid))
1917                                         gid = GLOBAL_ROOT_GID;
1918                         }
1919                 } else {
1920                         uid = GLOBAL_ROOT_UID;
1921                         gid = GLOBAL_ROOT_GID;
1922                 }
1923                 task_unlock(task);
1924         }
1925         *ruid = uid;
1926         *rgid = gid;
1927 }
1928
1929 void proc_pid_evict_inode(struct proc_inode *ei)
1930 {
1931         struct pid *pid = ei->pid;
1932
1933         if (S_ISDIR(ei->vfs_inode.i_mode)) {
1934                 spin_lock(&pid->lock);
1935                 hlist_del_init_rcu(&ei->sibling_inodes);
1936                 spin_unlock(&pid->lock);
1937         }
1938 }
1939
1940 struct inode *proc_pid_make_inode(struct super_block *sb,
1941                                   struct task_struct *task, umode_t mode)
1942 {
1943         struct inode * inode;
1944         struct proc_inode *ei;
1945         struct pid *pid;
1946
1947         /* We need a new inode */
1948
1949         inode = new_inode(sb);
1950         if (!inode)
1951                 goto out;
1952
1953         /* Common stuff */
1954         ei = PROC_I(inode);
1955         inode->i_mode = mode;
1956         inode->i_ino = get_next_ino();
1957         simple_inode_init_ts(inode);
1958         inode->i_op = &proc_def_inode_operations;
1959
1960         /*
1961          * grab the reference to task.
1962          */
1963         pid = get_task_pid(task, PIDTYPE_PID);
1964         if (!pid)
1965                 goto out_unlock;
1966
1967         /* Let the pid remember us for quick removal */
1968         ei->pid = pid;
1969
1970         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1971         security_task_to_inode(task, inode);
1972
1973 out:
1974         return inode;
1975
1976 out_unlock:
1977         iput(inode);
1978         return NULL;
1979 }
1980
1981 /*
1982  * Generating an inode and adding it into @pid->inodes, so that task will
1983  * invalidate inode's dentry before being released.
1984  *
1985  * This helper is used for creating dir-type entries under '/proc' and
1986  * '/proc/<tgid>/task'. Other entries(eg. fd, stat) under '/proc/<tgid>'
1987  * can be released by invalidating '/proc/<tgid>' dentry.
1988  * In theory, dentries under '/proc/<tgid>/task' can also be released by
1989  * invalidating '/proc/<tgid>' dentry, we reserve it to handle single
1990  * thread exiting situation: Any one of threads should invalidate its
1991  * '/proc/<tgid>/task/<pid>' dentry before released.
1992  */
1993 static struct inode *proc_pid_make_base_inode(struct super_block *sb,
1994                                 struct task_struct *task, umode_t mode)
1995 {
1996         struct inode *inode;
1997         struct proc_inode *ei;
1998         struct pid *pid;
1999
2000         inode = proc_pid_make_inode(sb, task, mode);
2001         if (!inode)
2002                 return NULL;
2003
2004         /* Let proc_flush_pid find this directory inode */
2005         ei = PROC_I(inode);
2006         pid = ei->pid;
2007         spin_lock(&pid->lock);
2008         hlist_add_head_rcu(&ei->sibling_inodes, &pid->inodes);
2009         spin_unlock(&pid->lock);
2010
2011         return inode;
2012 }
2013
2014 int pid_getattr(struct mnt_idmap *idmap, const struct path *path,
2015                 struct kstat *stat, u32 request_mask, unsigned int query_flags)
2016 {
2017         struct inode *inode = d_inode(path->dentry);
2018         struct proc_fs_info *fs_info = proc_sb_info(inode->i_sb);
2019         struct task_struct *task;
2020
2021         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
2022
2023         stat->uid = GLOBAL_ROOT_UID;
2024         stat->gid = GLOBAL_ROOT_GID;
2025         rcu_read_lock();
2026         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2027         if (task) {
2028                 if (!has_pid_permissions(fs_info, task, HIDEPID_INVISIBLE)) {
2029                         rcu_read_unlock();
2030                         /*
2031                          * This doesn't prevent learning whether PID exists,
2032                          * it only makes getattr() consistent with readdir().
2033                          */
2034                         return -ENOENT;
2035                 }
2036                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
2037         }
2038         rcu_read_unlock();
2039         return 0;
2040 }
2041
2042 /* dentry stuff */
2043
2044 /*
2045  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
2046  */
2047 void pid_update_inode(struct task_struct *task, struct inode *inode)
2048 {
2049         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
2050
2051         inode->i_mode &= ~(S_ISUID | S_ISGID);
2052         security_task_to_inode(task, inode);
2053 }
2054
2055 /*
2056  * Rewrite the inode's ownerships here because the owning task may have
2057  * performed a setuid(), etc.
2058  *
2059  */
2060 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
2061 {
2062         struct inode *inode;
2063         struct task_struct *task;
2064         int ret = 0;
2065
2066         rcu_read_lock();
2067         inode = d_inode_rcu(dentry);
2068         if (!inode)
2069                 goto out;
2070         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2071
2072         if (task) {
2073                 pid_update_inode(task, inode);
2074                 ret = 1;
2075         }
2076 out:
2077         rcu_read_unlock();
2078         return ret;
2079 }
2080
2081 static inline bool proc_inode_is_dead(struct inode *inode)
2082 {
2083         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
2084 }
2085
2086 int pid_delete_dentry(const struct dentry *dentry)
2087 {
2088         /* Is the task we represent dead?
2089          * If so, then don't put the dentry on the lru list,
2090          * kill it immediately.
2091          */
2092         return proc_inode_is_dead(d_inode(dentry));
2093 }
2094
2095 const struct dentry_operations pid_dentry_operations =
2096 {
2097         .d_revalidate   = pid_revalidate,
2098         .d_delete       = pid_delete_dentry,
2099 };
2100
2101 /* Lookups */
2102
2103 /*
2104  * Fill a directory entry.
2105  *
2106  * If possible create the dcache entry and derive our inode number and
2107  * file type from dcache entry.
2108  *
2109  * Since all of the proc inode numbers are dynamically generated, the inode
2110  * numbers do not exist until the inode is cache.  This means creating
2111  * the dcache entry in readdir is necessary to keep the inode numbers
2112  * reported by readdir in sync with the inode numbers reported
2113  * by stat.
2114  */
2115 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
2116         const char *name, unsigned int len,
2117         instantiate_t instantiate, struct task_struct *task, const void *ptr)
2118 {
2119         struct dentry *child, *dir = file->f_path.dentry;
2120         struct qstr qname = QSTR_INIT(name, len);
2121         struct inode *inode;
2122         unsigned type = DT_UNKNOWN;
2123         ino_t ino = 1;
2124
2125         child = d_hash_and_lookup(dir, &qname);
2126         if (!child) {
2127                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2128                 child = d_alloc_parallel(dir, &qname, &wq);
2129                 if (IS_ERR(child))
2130                         goto end_instantiate;
2131                 if (d_in_lookup(child)) {
2132                         struct dentry *res;
2133                         res = instantiate(child, task, ptr);
2134                         d_lookup_done(child);
2135                         if (unlikely(res)) {
2136                                 dput(child);
2137                                 child = res;
2138                                 if (IS_ERR(child))
2139                                         goto end_instantiate;
2140                         }
2141                 }
2142         }
2143         inode = d_inode(child);
2144         ino = inode->i_ino;
2145         type = inode->i_mode >> 12;
2146         dput(child);
2147 end_instantiate:
2148         return dir_emit(ctx, name, len, ino, type);
2149 }
2150
2151 /*
2152  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2153  * which represent vma start and end addresses.
2154  */
2155 static int dname_to_vma_addr(struct dentry *dentry,
2156                              unsigned long *start, unsigned long *end)
2157 {
2158         const char *str = dentry->d_name.name;
2159         unsigned long long sval, eval;
2160         unsigned int len;
2161
2162         if (str[0] == '0' && str[1] != '-')
2163                 return -EINVAL;
2164         len = _parse_integer(str, 16, &sval);
2165         if (len & KSTRTOX_OVERFLOW)
2166                 return -EINVAL;
2167         if (sval != (unsigned long)sval)
2168                 return -EINVAL;
2169         str += len;
2170
2171         if (*str != '-')
2172                 return -EINVAL;
2173         str++;
2174
2175         if (str[0] == '0' && str[1])
2176                 return -EINVAL;
2177         len = _parse_integer(str, 16, &eval);
2178         if (len & KSTRTOX_OVERFLOW)
2179                 return -EINVAL;
2180         if (eval != (unsigned long)eval)
2181                 return -EINVAL;
2182         str += len;
2183
2184         if (*str != '\0')
2185                 return -EINVAL;
2186
2187         *start = sval;
2188         *end = eval;
2189
2190         return 0;
2191 }
2192
2193 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
2194 {
2195         unsigned long vm_start, vm_end;
2196         bool exact_vma_exists = false;
2197         struct mm_struct *mm = NULL;
2198         struct task_struct *task;
2199         struct inode *inode;
2200         int status = 0;
2201
2202         if (flags & LOOKUP_RCU)
2203                 return -ECHILD;
2204
2205         inode = d_inode(dentry);
2206         task = get_proc_task(inode);
2207         if (!task)
2208                 goto out_notask;
2209
2210         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
2211         if (IS_ERR_OR_NULL(mm))
2212                 goto out;
2213
2214         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2215                 status = mmap_read_lock_killable(mm);
2216                 if (!status) {
2217                         exact_vma_exists = !!find_exact_vma(mm, vm_start,
2218                                                             vm_end);
2219                         mmap_read_unlock(mm);
2220                 }
2221         }
2222
2223         mmput(mm);
2224
2225         if (exact_vma_exists) {
2226                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2227
2228                 security_task_to_inode(task, inode);
2229                 status = 1;
2230         }
2231
2232 out:
2233         put_task_struct(task);
2234
2235 out_notask:
2236         return status;
2237 }
2238
2239 static const struct dentry_operations tid_map_files_dentry_operations = {
2240         .d_revalidate   = map_files_d_revalidate,
2241         .d_delete       = pid_delete_dentry,
2242 };
2243
2244 static int map_files_get_link(struct dentry *dentry, struct path *path)
2245 {
2246         unsigned long vm_start, vm_end;
2247         struct vm_area_struct *vma;
2248         struct task_struct *task;
2249         struct mm_struct *mm;
2250         int rc;
2251
2252         rc = -ENOENT;
2253         task = get_proc_task(d_inode(dentry));
2254         if (!task)
2255                 goto out;
2256
2257         mm = get_task_mm(task);
2258         put_task_struct(task);
2259         if (!mm)
2260                 goto out;
2261
2262         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2263         if (rc)
2264                 goto out_mmput;
2265
2266         rc = mmap_read_lock_killable(mm);
2267         if (rc)
2268                 goto out_mmput;
2269
2270         rc = -ENOENT;
2271         vma = find_exact_vma(mm, vm_start, vm_end);
2272         if (vma && vma->vm_file) {
2273                 *path = *file_user_path(vma->vm_file);
2274                 path_get(path);
2275                 rc = 0;
2276         }
2277         mmap_read_unlock(mm);
2278
2279 out_mmput:
2280         mmput(mm);
2281 out:
2282         return rc;
2283 }
2284
2285 struct map_files_info {
2286         unsigned long   start;
2287         unsigned long   end;
2288         fmode_t         mode;
2289 };
2290
2291 /*
2292  * Only allow CAP_SYS_ADMIN and CAP_CHECKPOINT_RESTORE to follow the links, due
2293  * to concerns about how the symlinks may be used to bypass permissions on
2294  * ancestor directories in the path to the file in question.
2295  */
2296 static const char *
2297 proc_map_files_get_link(struct dentry *dentry,
2298                         struct inode *inode,
2299                         struct delayed_call *done)
2300 {
2301         if (!checkpoint_restore_ns_capable(&init_user_ns))
2302                 return ERR_PTR(-EPERM);
2303
2304         return proc_pid_get_link(dentry, inode, done);
2305 }
2306
2307 /*
2308  * Identical to proc_pid_link_inode_operations except for get_link()
2309  */
2310 static const struct inode_operations proc_map_files_link_inode_operations = {
2311         .readlink       = proc_pid_readlink,
2312         .get_link       = proc_map_files_get_link,
2313         .setattr        = proc_setattr,
2314 };
2315
2316 static struct dentry *
2317 proc_map_files_instantiate(struct dentry *dentry,
2318                            struct task_struct *task, const void *ptr)
2319 {
2320         fmode_t mode = (fmode_t)(unsigned long)ptr;
2321         struct proc_inode *ei;
2322         struct inode *inode;
2323
2324         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2325                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2326                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2327         if (!inode)
2328                 return ERR_PTR(-ENOENT);
2329
2330         ei = PROC_I(inode);
2331         ei->op.proc_get_link = map_files_get_link;
2332
2333         inode->i_op = &proc_map_files_link_inode_operations;
2334         inode->i_size = 64;
2335
2336         return proc_splice_unmountable(inode, dentry,
2337                                        &tid_map_files_dentry_operations);
2338 }
2339
2340 static struct dentry *proc_map_files_lookup(struct inode *dir,
2341                 struct dentry *dentry, unsigned int flags)
2342 {
2343         unsigned long vm_start, vm_end;
2344         struct vm_area_struct *vma;
2345         struct task_struct *task;
2346         struct dentry *result;
2347         struct mm_struct *mm;
2348
2349         result = ERR_PTR(-ENOENT);
2350         task = get_proc_task(dir);
2351         if (!task)
2352                 goto out;
2353
2354         result = ERR_PTR(-EACCES);
2355         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2356                 goto out_put_task;
2357
2358         result = ERR_PTR(-ENOENT);
2359         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2360                 goto out_put_task;
2361
2362         mm = get_task_mm(task);
2363         if (!mm)
2364                 goto out_put_task;
2365
2366         result = ERR_PTR(-EINTR);
2367         if (mmap_read_lock_killable(mm))
2368                 goto out_put_mm;
2369
2370         result = ERR_PTR(-ENOENT);
2371         vma = find_exact_vma(mm, vm_start, vm_end);
2372         if (!vma)
2373                 goto out_no_vma;
2374
2375         if (vma->vm_file)
2376                 result = proc_map_files_instantiate(dentry, task,
2377                                 (void *)(unsigned long)vma->vm_file->f_mode);
2378
2379 out_no_vma:
2380         mmap_read_unlock(mm);
2381 out_put_mm:
2382         mmput(mm);
2383 out_put_task:
2384         put_task_struct(task);
2385 out:
2386         return result;
2387 }
2388
2389 static const struct inode_operations proc_map_files_inode_operations = {
2390         .lookup         = proc_map_files_lookup,
2391         .permission     = proc_fd_permission,
2392         .setattr        = proc_setattr,
2393 };
2394
2395 static int
2396 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2397 {
2398         struct vm_area_struct *vma;
2399         struct task_struct *task;
2400         struct mm_struct *mm;
2401         unsigned long nr_files, pos, i;
2402         GENRADIX(struct map_files_info) fa;
2403         struct map_files_info *p;
2404         int ret;
2405         struct vma_iterator vmi;
2406
2407         genradix_init(&fa);
2408
2409         ret = -ENOENT;
2410         task = get_proc_task(file_inode(file));
2411         if (!task)
2412                 goto out;
2413
2414         ret = -EACCES;
2415         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2416                 goto out_put_task;
2417
2418         ret = 0;
2419         if (!dir_emit_dots(file, ctx))
2420                 goto out_put_task;
2421
2422         mm = get_task_mm(task);
2423         if (!mm)
2424                 goto out_put_task;
2425
2426         ret = mmap_read_lock_killable(mm);
2427         if (ret) {
2428                 mmput(mm);
2429                 goto out_put_task;
2430         }
2431
2432         nr_files = 0;
2433
2434         /*
2435          * We need two passes here:
2436          *
2437          *  1) Collect vmas of mapped files with mmap_lock taken
2438          *  2) Release mmap_lock and instantiate entries
2439          *
2440          * otherwise we get lockdep complained, since filldir()
2441          * routine might require mmap_lock taken in might_fault().
2442          */
2443
2444         pos = 2;
2445         vma_iter_init(&vmi, mm, 0);
2446         for_each_vma(vmi, vma) {
2447                 if (!vma->vm_file)
2448                         continue;
2449                 if (++pos <= ctx->pos)
2450                         continue;
2451
2452                 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2453                 if (!p) {
2454                         ret = -ENOMEM;
2455                         mmap_read_unlock(mm);
2456                         mmput(mm);
2457                         goto out_put_task;
2458                 }
2459
2460                 p->start = vma->vm_start;
2461                 p->end = vma->vm_end;
2462                 p->mode = vma->vm_file->f_mode;
2463         }
2464         mmap_read_unlock(mm);
2465         mmput(mm);
2466
2467         for (i = 0; i < nr_files; i++) {
2468                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2469                 unsigned int len;
2470
2471                 p = genradix_ptr(&fa, i);
2472                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2473                 if (!proc_fill_cache(file, ctx,
2474                                       buf, len,
2475                                       proc_map_files_instantiate,
2476                                       task,
2477                                       (void *)(unsigned long)p->mode))
2478                         break;
2479                 ctx->pos++;
2480         }
2481
2482 out_put_task:
2483         put_task_struct(task);
2484 out:
2485         genradix_free(&fa);
2486         return ret;
2487 }
2488
2489 static const struct file_operations proc_map_files_operations = {
2490         .read           = generic_read_dir,
2491         .iterate_shared = proc_map_files_readdir,
2492         .llseek         = generic_file_llseek,
2493 };
2494
2495 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2496 struct timers_private {
2497         struct pid *pid;
2498         struct task_struct *task;
2499         struct sighand_struct *sighand;
2500         struct pid_namespace *ns;
2501         unsigned long flags;
2502 };
2503
2504 static void *timers_start(struct seq_file *m, loff_t *pos)
2505 {
2506         struct timers_private *tp = m->private;
2507
2508         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2509         if (!tp->task)
2510                 return ERR_PTR(-ESRCH);
2511
2512         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2513         if (!tp->sighand)
2514                 return ERR_PTR(-ESRCH);
2515
2516         return seq_hlist_start(&tp->task->signal->posix_timers, *pos);
2517 }
2518
2519 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2520 {
2521         struct timers_private *tp = m->private;
2522         return seq_hlist_next(v, &tp->task->signal->posix_timers, pos);
2523 }
2524
2525 static void timers_stop(struct seq_file *m, void *v)
2526 {
2527         struct timers_private *tp = m->private;
2528
2529         if (tp->sighand) {
2530                 unlock_task_sighand(tp->task, &tp->flags);
2531                 tp->sighand = NULL;
2532         }
2533
2534         if (tp->task) {
2535                 put_task_struct(tp->task);
2536                 tp->task = NULL;
2537         }
2538 }
2539
2540 static int show_timer(struct seq_file *m, void *v)
2541 {
2542         struct k_itimer *timer;
2543         struct timers_private *tp = m->private;
2544         int notify;
2545         static const char * const nstr[] = {
2546                 [SIGEV_SIGNAL] = "signal",
2547                 [SIGEV_NONE] = "none",
2548                 [SIGEV_THREAD] = "thread",
2549         };
2550
2551         timer = hlist_entry((struct hlist_node *)v, struct k_itimer, list);
2552         notify = timer->it_sigev_notify;
2553
2554         seq_printf(m, "ID: %d\n", timer->it_id);
2555         seq_printf(m, "signal: %d/%px\n",
2556                    timer->sigq->info.si_signo,
2557                    timer->sigq->info.si_value.sival_ptr);
2558         seq_printf(m, "notify: %s/%s.%d\n",
2559                    nstr[notify & ~SIGEV_THREAD_ID],
2560                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2561                    pid_nr_ns(timer->it_pid, tp->ns));
2562         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2563
2564         return 0;
2565 }
2566
2567 static const struct seq_operations proc_timers_seq_ops = {
2568         .start  = timers_start,
2569         .next   = timers_next,
2570         .stop   = timers_stop,
2571         .show   = show_timer,
2572 };
2573
2574 static int proc_timers_open(struct inode *inode, struct file *file)
2575 {
2576         struct timers_private *tp;
2577
2578         tp = __seq_open_private(file, &proc_timers_seq_ops,
2579                         sizeof(struct timers_private));
2580         if (!tp)
2581                 return -ENOMEM;
2582
2583         tp->pid = proc_pid(inode);
2584         tp->ns = proc_pid_ns(inode->i_sb);
2585         return 0;
2586 }
2587
2588 static const struct file_operations proc_timers_operations = {
2589         .open           = proc_timers_open,
2590         .read           = seq_read,
2591         .llseek         = seq_lseek,
2592         .release        = seq_release_private,
2593 };
2594 #endif
2595
2596 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2597                                         size_t count, loff_t *offset)
2598 {
2599         struct inode *inode = file_inode(file);
2600         struct task_struct *p;
2601         u64 slack_ns;
2602         int err;
2603
2604         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2605         if (err < 0)
2606                 return err;
2607
2608         p = get_proc_task(inode);
2609         if (!p)
2610                 return -ESRCH;
2611
2612         if (p != current) {
2613                 rcu_read_lock();
2614                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2615                         rcu_read_unlock();
2616                         count = -EPERM;
2617                         goto out;
2618                 }
2619                 rcu_read_unlock();
2620
2621                 err = security_task_setscheduler(p);
2622                 if (err) {
2623                         count = err;
2624                         goto out;
2625                 }
2626         }
2627
2628         task_lock(p);
2629         if (rt_or_dl_task_policy(p))
2630                 slack_ns = 0;
2631         else if (slack_ns == 0)
2632                 slack_ns = p->default_timer_slack_ns;
2633         p->timer_slack_ns = slack_ns;
2634         task_unlock(p);
2635
2636 out:
2637         put_task_struct(p);
2638
2639         return count;
2640 }
2641
2642 static int timerslack_ns_show(struct seq_file *m, void *v)
2643 {
2644         struct inode *inode = m->private;
2645         struct task_struct *p;
2646         int err = 0;
2647
2648         p = get_proc_task(inode);
2649         if (!p)
2650                 return -ESRCH;
2651
2652         if (p != current) {
2653                 rcu_read_lock();
2654                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2655                         rcu_read_unlock();
2656                         err = -EPERM;
2657                         goto out;
2658                 }
2659                 rcu_read_unlock();
2660
2661                 err = security_task_getscheduler(p);
2662                 if (err)
2663                         goto out;
2664         }
2665
2666         task_lock(p);
2667         seq_printf(m, "%llu\n", p->timer_slack_ns);
2668         task_unlock(p);
2669
2670 out:
2671         put_task_struct(p);
2672
2673         return err;
2674 }
2675
2676 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2677 {
2678         return single_open(filp, timerslack_ns_show, inode);
2679 }
2680
2681 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2682         .open           = timerslack_ns_open,
2683         .read           = seq_read,
2684         .write          = timerslack_ns_write,
2685         .llseek         = seq_lseek,
2686         .release        = single_release,
2687 };
2688
2689 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2690         struct task_struct *task, const void *ptr)
2691 {
2692         const struct pid_entry *p = ptr;
2693         struct inode *inode;
2694         struct proc_inode *ei;
2695
2696         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2697         if (!inode)
2698                 return ERR_PTR(-ENOENT);
2699
2700         ei = PROC_I(inode);
2701         if (S_ISDIR(inode->i_mode))
2702                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2703         if (p->iop)
2704                 inode->i_op = p->iop;
2705         if (p->fop)
2706                 inode->i_fop = p->fop;
2707         ei->op = p->op;
2708         pid_update_inode(task, inode);
2709         d_set_d_op(dentry, &pid_dentry_operations);
2710         return d_splice_alias(inode, dentry);
2711 }
2712
2713 static struct dentry *proc_pident_lookup(struct inode *dir,
2714                                          struct dentry *dentry,
2715                                          const struct pid_entry *p,
2716                                          const struct pid_entry *end)
2717 {
2718         struct task_struct *task = get_proc_task(dir);
2719         struct dentry *res = ERR_PTR(-ENOENT);
2720
2721         if (!task)
2722                 goto out_no_task;
2723
2724         /*
2725          * Yes, it does not scale. And it should not. Don't add
2726          * new entries into /proc/<tgid>/ without very good reasons.
2727          */
2728         for (; p < end; p++) {
2729                 if (p->len != dentry->d_name.len)
2730                         continue;
2731                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2732                         res = proc_pident_instantiate(dentry, task, p);
2733                         break;
2734                 }
2735         }
2736         put_task_struct(task);
2737 out_no_task:
2738         return res;
2739 }
2740
2741 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2742                 const struct pid_entry *ents, unsigned int nents)
2743 {
2744         struct task_struct *task = get_proc_task(file_inode(file));
2745         const struct pid_entry *p;
2746
2747         if (!task)
2748                 return -ENOENT;
2749
2750         if (!dir_emit_dots(file, ctx))
2751                 goto out;
2752
2753         if (ctx->pos >= nents + 2)
2754                 goto out;
2755
2756         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2757                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2758                                 proc_pident_instantiate, task, p))
2759                         break;
2760                 ctx->pos++;
2761         }
2762 out:
2763         put_task_struct(task);
2764         return 0;
2765 }
2766
2767 #ifdef CONFIG_SECURITY
2768 static int proc_pid_attr_open(struct inode *inode, struct file *file)
2769 {
2770         file->private_data = NULL;
2771         __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2772         return 0;
2773 }
2774
2775 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2776                                   size_t count, loff_t *ppos)
2777 {
2778         struct inode * inode = file_inode(file);
2779         char *p = NULL;
2780         ssize_t length;
2781         struct task_struct *task = get_proc_task(inode);
2782
2783         if (!task)
2784                 return -ESRCH;
2785
2786         length = security_getprocattr(task, PROC_I(inode)->op.lsmid,
2787                                       file->f_path.dentry->d_name.name,
2788                                       &p);
2789         put_task_struct(task);
2790         if (length > 0)
2791                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2792         kfree(p);
2793         return length;
2794 }
2795
2796 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2797                                    size_t count, loff_t *ppos)
2798 {
2799         struct inode * inode = file_inode(file);
2800         struct task_struct *task;
2801         void *page;
2802         int rv;
2803
2804         /* A task may only write when it was the opener. */
2805         if (file->private_data != current->mm)
2806                 return -EPERM;
2807
2808         rcu_read_lock();
2809         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2810         if (!task) {
2811                 rcu_read_unlock();
2812                 return -ESRCH;
2813         }
2814         /* A task may only write its own attributes. */
2815         if (current != task) {
2816                 rcu_read_unlock();
2817                 return -EACCES;
2818         }
2819         /* Prevent changes to overridden credentials. */
2820         if (current_cred() != current_real_cred()) {
2821                 rcu_read_unlock();
2822                 return -EBUSY;
2823         }
2824         rcu_read_unlock();
2825
2826         if (count > PAGE_SIZE)
2827                 count = PAGE_SIZE;
2828
2829         /* No partial writes. */
2830         if (*ppos != 0)
2831                 return -EINVAL;
2832
2833         page = memdup_user(buf, count);
2834         if (IS_ERR(page)) {
2835                 rv = PTR_ERR(page);
2836                 goto out;
2837         }
2838
2839         /* Guard against adverse ptrace interaction */
2840         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2841         if (rv < 0)
2842                 goto out_free;
2843
2844         rv = security_setprocattr(PROC_I(inode)->op.lsmid,
2845                                   file->f_path.dentry->d_name.name, page,
2846                                   count);
2847         mutex_unlock(&current->signal->cred_guard_mutex);
2848 out_free:
2849         kfree(page);
2850 out:
2851         return rv;
2852 }
2853
2854 static const struct file_operations proc_pid_attr_operations = {
2855         .open           = proc_pid_attr_open,
2856         .read           = proc_pid_attr_read,
2857         .write          = proc_pid_attr_write,
2858         .llseek         = generic_file_llseek,
2859         .release        = mem_release,
2860 };
2861
2862 #define LSM_DIR_OPS(LSM) \
2863 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2864                              struct dir_context *ctx) \
2865 { \
2866         return proc_pident_readdir(filp, ctx, \
2867                                    LSM##_attr_dir_stuff, \
2868                                    ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2869 } \
2870 \
2871 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2872         .read           = generic_read_dir, \
2873         .iterate_shared = proc_##LSM##_attr_dir_iterate, \
2874         .llseek         = default_llseek, \
2875 }; \
2876 \
2877 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2878                                 struct dentry *dentry, unsigned int flags) \
2879 { \
2880         return proc_pident_lookup(dir, dentry, \
2881                                   LSM##_attr_dir_stuff, \
2882                                   LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2883 } \
2884 \
2885 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2886         .lookup         = proc_##LSM##_attr_dir_lookup, \
2887         .getattr        = pid_getattr, \
2888         .setattr        = proc_setattr, \
2889 }
2890
2891 #ifdef CONFIG_SECURITY_SMACK
2892 static const struct pid_entry smack_attr_dir_stuff[] = {
2893         ATTR(LSM_ID_SMACK, "current",   0666),
2894 };
2895 LSM_DIR_OPS(smack);
2896 #endif
2897
2898 #ifdef CONFIG_SECURITY_APPARMOR
2899 static const struct pid_entry apparmor_attr_dir_stuff[] = {
2900         ATTR(LSM_ID_APPARMOR, "current",        0666),
2901         ATTR(LSM_ID_APPARMOR, "prev",           0444),
2902         ATTR(LSM_ID_APPARMOR, "exec",           0666),
2903 };
2904 LSM_DIR_OPS(apparmor);
2905 #endif
2906
2907 static const struct pid_entry attr_dir_stuff[] = {
2908         ATTR(LSM_ID_UNDEF, "current",   0666),
2909         ATTR(LSM_ID_UNDEF, "prev",              0444),
2910         ATTR(LSM_ID_UNDEF, "exec",              0666),
2911         ATTR(LSM_ID_UNDEF, "fscreate",  0666),
2912         ATTR(LSM_ID_UNDEF, "keycreate", 0666),
2913         ATTR(LSM_ID_UNDEF, "sockcreate",        0666),
2914 #ifdef CONFIG_SECURITY_SMACK
2915         DIR("smack",                    0555,
2916             proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2917 #endif
2918 #ifdef CONFIG_SECURITY_APPARMOR
2919         DIR("apparmor",                 0555,
2920             proc_apparmor_attr_dir_inode_ops, proc_apparmor_attr_dir_ops),
2921 #endif
2922 };
2923
2924 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2925 {
2926         return proc_pident_readdir(file, ctx,
2927                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2928 }
2929
2930 static const struct file_operations proc_attr_dir_operations = {
2931         .read           = generic_read_dir,
2932         .iterate_shared = proc_attr_dir_readdir,
2933         .llseek         = generic_file_llseek,
2934 };
2935
2936 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2937                                 struct dentry *dentry, unsigned int flags)
2938 {
2939         return proc_pident_lookup(dir, dentry,
2940                                   attr_dir_stuff,
2941                                   attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2942 }
2943
2944 static const struct inode_operations proc_attr_dir_inode_operations = {
2945         .lookup         = proc_attr_dir_lookup,
2946         .getattr        = pid_getattr,
2947         .setattr        = proc_setattr,
2948 };
2949
2950 #endif
2951
2952 #ifdef CONFIG_ELF_CORE
2953 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2954                                          size_t count, loff_t *ppos)
2955 {
2956         struct task_struct *task = get_proc_task(file_inode(file));
2957         struct mm_struct *mm;
2958         char buffer[PROC_NUMBUF];
2959         size_t len;
2960         int ret;
2961
2962         if (!task)
2963                 return -ESRCH;
2964
2965         ret = 0;
2966         mm = get_task_mm(task);
2967         if (mm) {
2968                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2969                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2970                                 MMF_DUMP_FILTER_SHIFT));
2971                 mmput(mm);
2972                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2973         }
2974
2975         put_task_struct(task);
2976
2977         return ret;
2978 }
2979
2980 static ssize_t proc_coredump_filter_write(struct file *file,
2981                                           const char __user *buf,
2982                                           size_t count,
2983                                           loff_t *ppos)
2984 {
2985         struct task_struct *task;
2986         struct mm_struct *mm;
2987         unsigned int val;
2988         int ret;
2989         int i;
2990         unsigned long mask;
2991
2992         ret = kstrtouint_from_user(buf, count, 0, &val);
2993         if (ret < 0)
2994                 return ret;
2995
2996         ret = -ESRCH;
2997         task = get_proc_task(file_inode(file));
2998         if (!task)
2999                 goto out_no_task;
3000
3001         mm = get_task_mm(task);
3002         if (!mm)
3003                 goto out_no_mm;
3004         ret = 0;
3005
3006         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
3007                 if (val & mask)
3008                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3009                 else
3010                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
3011         }
3012
3013         mmput(mm);
3014  out_no_mm:
3015         put_task_struct(task);
3016  out_no_task:
3017         if (ret < 0)
3018                 return ret;
3019         return count;
3020 }
3021
3022 static const struct file_operations proc_coredump_filter_operations = {
3023         .read           = proc_coredump_filter_read,
3024         .write          = proc_coredump_filter_write,
3025         .llseek         = generic_file_llseek,
3026 };
3027 #endif
3028
3029 #ifdef CONFIG_TASK_IO_ACCOUNTING
3030 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
3031 {
3032         struct task_io_accounting acct;
3033         int result;
3034
3035         result = down_read_killable(&task->signal->exec_update_lock);
3036         if (result)
3037                 return result;
3038
3039         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
3040                 result = -EACCES;
3041                 goto out_unlock;
3042         }
3043
3044         if (whole) {
3045                 struct signal_struct *sig = task->signal;
3046                 struct task_struct *t;
3047                 unsigned int seq = 1;
3048                 unsigned long flags;
3049
3050                 rcu_read_lock();
3051                 do {
3052                         seq++; /* 2 on the 1st/lockless path, otherwise odd */
3053                         flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
3054
3055                         acct = sig->ioac;
3056                         __for_each_thread(sig, t)
3057                                 task_io_accounting_add(&acct, &t->ioac);
3058
3059                 } while (need_seqretry(&sig->stats_lock, seq));
3060                 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
3061                 rcu_read_unlock();
3062         } else {
3063                 acct = task->ioac;
3064         }
3065
3066         seq_printf(m,
3067                    "rchar: %llu\n"
3068                    "wchar: %llu\n"
3069                    "syscr: %llu\n"
3070                    "syscw: %llu\n"
3071                    "read_bytes: %llu\n"
3072                    "write_bytes: %llu\n"
3073                    "cancelled_write_bytes: %llu\n",
3074                    (unsigned long long)acct.rchar,
3075                    (unsigned long long)acct.wchar,
3076                    (unsigned long long)acct.syscr,
3077                    (unsigned long long)acct.syscw,
3078                    (unsigned long long)acct.read_bytes,
3079                    (unsigned long long)acct.write_bytes,
3080                    (unsigned long long)acct.cancelled_write_bytes);
3081         result = 0;
3082
3083 out_unlock:
3084         up_read(&task->signal->exec_update_lock);
3085         return result;
3086 }
3087
3088 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3089                                   struct pid *pid, struct task_struct *task)
3090 {
3091         return do_io_accounting(task, m, 0);
3092 }
3093
3094 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
3095                                    struct pid *pid, struct task_struct *task)
3096 {
3097         return do_io_accounting(task, m, 1);
3098 }
3099 #endif /* CONFIG_TASK_IO_ACCOUNTING */
3100
3101 #ifdef CONFIG_USER_NS
3102 static int proc_id_map_open(struct inode *inode, struct file *file,
3103         const struct seq_operations *seq_ops)
3104 {
3105         struct user_namespace *ns = NULL;
3106         struct task_struct *task;
3107         struct seq_file *seq;
3108         int ret = -EINVAL;
3109
3110         task = get_proc_task(inode);
3111         if (task) {
3112                 rcu_read_lock();
3113                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3114                 rcu_read_unlock();
3115                 put_task_struct(task);
3116         }
3117         if (!ns)
3118                 goto err;
3119
3120         ret = seq_open(file, seq_ops);
3121         if (ret)
3122                 goto err_put_ns;
3123
3124         seq = file->private_data;
3125         seq->private = ns;
3126
3127         return 0;
3128 err_put_ns:
3129         put_user_ns(ns);
3130 err:
3131         return ret;
3132 }
3133
3134 static int proc_id_map_release(struct inode *inode, struct file *file)
3135 {
3136         struct seq_file *seq = file->private_data;
3137         struct user_namespace *ns = seq->private;
3138         put_user_ns(ns);
3139         return seq_release(inode, file);
3140 }
3141
3142 static int proc_uid_map_open(struct inode *inode, struct file *file)
3143 {
3144         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
3145 }
3146
3147 static int proc_gid_map_open(struct inode *inode, struct file *file)
3148 {
3149         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
3150 }
3151
3152 static int proc_projid_map_open(struct inode *inode, struct file *file)
3153 {
3154         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
3155 }
3156
3157 static const struct file_operations proc_uid_map_operations = {
3158         .open           = proc_uid_map_open,
3159         .write          = proc_uid_map_write,
3160         .read           = seq_read,
3161         .llseek         = seq_lseek,
3162         .release        = proc_id_map_release,
3163 };
3164
3165 static const struct file_operations proc_gid_map_operations = {
3166         .open           = proc_gid_map_open,
3167         .write          = proc_gid_map_write,
3168         .read           = seq_read,
3169         .llseek         = seq_lseek,
3170         .release        = proc_id_map_release,
3171 };
3172
3173 static const struct file_operations proc_projid_map_operations = {
3174         .open           = proc_projid_map_open,
3175         .write          = proc_projid_map_write,
3176         .read           = seq_read,
3177         .llseek         = seq_lseek,
3178         .release        = proc_id_map_release,
3179 };
3180
3181 static int proc_setgroups_open(struct inode *inode, struct file *file)
3182 {
3183         struct user_namespace *ns = NULL;
3184         struct task_struct *task;
3185         int ret;
3186
3187         ret = -ESRCH;
3188         task = get_proc_task(inode);
3189         if (task) {
3190                 rcu_read_lock();
3191                 ns = get_user_ns(task_cred_xxx(task, user_ns));
3192                 rcu_read_unlock();
3193                 put_task_struct(task);
3194         }
3195         if (!ns)
3196                 goto err;
3197
3198         if (file->f_mode & FMODE_WRITE) {
3199                 ret = -EACCES;
3200                 if (!ns_capable(ns, CAP_SYS_ADMIN))
3201                         goto err_put_ns;
3202         }
3203
3204         ret = single_open(file, &proc_setgroups_show, ns);
3205         if (ret)
3206                 goto err_put_ns;
3207
3208         return 0;
3209 err_put_ns:
3210         put_user_ns(ns);
3211 err:
3212         return ret;
3213 }
3214
3215 static int proc_setgroups_release(struct inode *inode, struct file *file)
3216 {
3217         struct seq_file *seq = file->private_data;
3218         struct user_namespace *ns = seq->private;
3219         int ret = single_release(inode, file);
3220         put_user_ns(ns);
3221         return ret;
3222 }
3223
3224 static const struct file_operations proc_setgroups_operations = {
3225         .open           = proc_setgroups_open,
3226         .write          = proc_setgroups_write,
3227         .read           = seq_read,
3228         .llseek         = seq_lseek,
3229         .release        = proc_setgroups_release,
3230 };
3231 #endif /* CONFIG_USER_NS */
3232
3233 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
3234                                 struct pid *pid, struct task_struct *task)
3235 {
3236         int err = lock_trace(task);
3237         if (!err) {
3238                 seq_printf(m, "%08x\n", task->personality);
3239                 unlock_trace(task);
3240         }
3241         return err;
3242 }
3243
3244 #ifdef CONFIG_LIVEPATCH
3245 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
3246                                 struct pid *pid, struct task_struct *task)
3247 {
3248         seq_printf(m, "%d\n", task->patch_state);
3249         return 0;
3250 }
3251 #endif /* CONFIG_LIVEPATCH */
3252
3253 #ifdef CONFIG_KSM
3254 static int proc_pid_ksm_merging_pages(struct seq_file *m, struct pid_namespace *ns,
3255                                 struct pid *pid, struct task_struct *task)
3256 {
3257         struct mm_struct *mm;
3258
3259         mm = get_task_mm(task);
3260         if (mm) {
3261                 seq_printf(m, "%lu\n", mm->ksm_merging_pages);
3262                 mmput(mm);
3263         }
3264
3265         return 0;
3266 }
3267 static int proc_pid_ksm_stat(struct seq_file *m, struct pid_namespace *ns,
3268                                 struct pid *pid, struct task_struct *task)
3269 {
3270         struct mm_struct *mm;
3271
3272         mm = get_task_mm(task);
3273         if (mm) {
3274                 seq_printf(m, "ksm_rmap_items %lu\n", mm->ksm_rmap_items);
3275                 seq_printf(m, "ksm_zero_pages %ld\n", mm_ksm_zero_pages(mm));
3276                 seq_printf(m, "ksm_merging_pages %lu\n", mm->ksm_merging_pages);
3277                 seq_printf(m, "ksm_process_profit %ld\n", ksm_process_profit(mm));
3278                 mmput(mm);
3279         }
3280
3281         return 0;
3282 }
3283 #endif /* CONFIG_KSM */
3284
3285 #ifdef CONFIG_STACKLEAK_METRICS
3286 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3287                                 struct pid *pid, struct task_struct *task)
3288 {
3289         unsigned long prev_depth = THREAD_SIZE -
3290                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3291         unsigned long depth = THREAD_SIZE -
3292                                 (task->lowest_stack & (THREAD_SIZE - 1));
3293
3294         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3295                                                         prev_depth, depth);
3296         return 0;
3297 }
3298 #endif /* CONFIG_STACKLEAK_METRICS */
3299
3300 /*
3301  * Thread groups
3302  */
3303 static const struct file_operations proc_task_operations;
3304 static const struct inode_operations proc_task_inode_operations;
3305
3306 static const struct pid_entry tgid_base_stuff[] = {
3307         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3308         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3309         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3310         DIR("fdinfo",     S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3311         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3312 #ifdef CONFIG_NET
3313         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3314 #endif
3315         REG("environ",    S_IRUSR, proc_environ_operations),
3316         REG("auxv",       S_IRUSR, proc_auxv_operations),
3317         ONE("status",     S_IRUGO, proc_pid_status),
3318         ONE("personality", S_IRUSR, proc_pid_personality),
3319         ONE("limits",     S_IRUGO, proc_pid_limits),
3320 #ifdef CONFIG_SCHED_DEBUG
3321         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3322 #endif
3323 #ifdef CONFIG_SCHED_AUTOGROUP
3324         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3325 #endif
3326 #ifdef CONFIG_TIME_NS
3327         REG("timens_offsets",  S_IRUGO|S_IWUSR, proc_timens_offsets_operations),
3328 #endif
3329         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3330 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3331         ONE("syscall",    S_IRUSR, proc_pid_syscall),
3332 #endif
3333         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3334         ONE("stat",       S_IRUGO, proc_tgid_stat),
3335         ONE("statm",      S_IRUGO, proc_pid_statm),
3336         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3337 #ifdef CONFIG_NUMA
3338         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3339 #endif
3340         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3341         LNK("cwd",        proc_cwd_link),
3342         LNK("root",       proc_root_link),
3343         LNK("exe",        proc_exe_link),
3344         REG("mounts",     S_IRUGO, proc_mounts_operations),
3345         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3346         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3347 #ifdef CONFIG_PROC_PAGE_MONITOR
3348         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3349         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3350         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3351         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3352 #endif
3353 #ifdef CONFIG_SECURITY
3354         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3355 #endif
3356 #ifdef CONFIG_KALLSYMS
3357         ONE("wchan",      S_IRUGO, proc_pid_wchan),
3358 #endif
3359 #ifdef CONFIG_STACKTRACE
3360         ONE("stack",      S_IRUSR, proc_pid_stack),
3361 #endif
3362 #ifdef CONFIG_SCHED_INFO
3363         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3364 #endif
3365 #ifdef CONFIG_LATENCYTOP
3366         REG("latency",  S_IRUGO, proc_lstats_operations),
3367 #endif
3368 #ifdef CONFIG_PROC_PID_CPUSET
3369         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3370 #endif
3371 #ifdef CONFIG_CGROUPS
3372         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3373 #endif
3374 #ifdef CONFIG_PROC_CPU_RESCTRL
3375         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3376 #endif
3377         ONE("oom_score",  S_IRUGO, proc_oom_score),
3378         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3379         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3380 #ifdef CONFIG_AUDIT
3381         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3382         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3383 #endif
3384 #ifdef CONFIG_FAULT_INJECTION
3385         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3386         REG("fail-nth", 0644, proc_fail_nth_operations),
3387 #endif
3388 #ifdef CONFIG_ELF_CORE
3389         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3390 #endif
3391 #ifdef CONFIG_TASK_IO_ACCOUNTING
3392         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3393 #endif
3394 #ifdef CONFIG_USER_NS
3395         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3396         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3397         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3398         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3399 #endif
3400 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3401         REG("timers",     S_IRUGO, proc_timers_operations),
3402 #endif
3403         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3404 #ifdef CONFIG_LIVEPATCH
3405         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3406 #endif
3407 #ifdef CONFIG_STACKLEAK_METRICS
3408         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3409 #endif
3410 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3411         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3412 #endif
3413 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3414         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3415 #endif
3416 #ifdef CONFIG_KSM
3417         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3418         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3419 #endif
3420 };
3421
3422 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3423 {
3424         return proc_pident_readdir(file, ctx,
3425                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3426 }
3427
3428 static const struct file_operations proc_tgid_base_operations = {
3429         .read           = generic_read_dir,
3430         .iterate_shared = proc_tgid_base_readdir,
3431         .llseek         = generic_file_llseek,
3432 };
3433
3434 struct pid *tgid_pidfd_to_pid(const struct file *file)
3435 {
3436         if (file->f_op != &proc_tgid_base_operations)
3437                 return ERR_PTR(-EBADF);
3438
3439         return proc_pid(file_inode(file));
3440 }
3441
3442 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3443 {
3444         return proc_pident_lookup(dir, dentry,
3445                                   tgid_base_stuff,
3446                                   tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3447 }
3448
3449 static const struct inode_operations proc_tgid_base_inode_operations = {
3450         .lookup         = proc_tgid_base_lookup,
3451         .getattr        = pid_getattr,
3452         .setattr        = proc_setattr,
3453         .permission     = proc_pid_permission,
3454 };
3455
3456 /**
3457  * proc_flush_pid -  Remove dcache entries for @pid from the /proc dcache.
3458  * @pid: pid that should be flushed.
3459  *
3460  * This function walks a list of inodes (that belong to any proc
3461  * filesystem) that are attached to the pid and flushes them from
3462  * the dentry cache.
3463  *
3464  * It is safe and reasonable to cache /proc entries for a task until
3465  * that task exits.  After that they just clog up the dcache with
3466  * useless entries, possibly causing useful dcache entries to be
3467  * flushed instead.  This routine is provided to flush those useless
3468  * dcache entries when a process is reaped.
3469  *
3470  * NOTE: This routine is just an optimization so it does not guarantee
3471  *       that no dcache entries will exist after a process is reaped
3472  *       it just makes it very unlikely that any will persist.
3473  */
3474
3475 void proc_flush_pid(struct pid *pid)
3476 {
3477         proc_invalidate_siblings_dcache(&pid->inodes, &pid->lock);
3478 }
3479
3480 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3481                                    struct task_struct *task, const void *ptr)
3482 {
3483         struct inode *inode;
3484
3485         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3486                                          S_IFDIR | S_IRUGO | S_IXUGO);
3487         if (!inode)
3488                 return ERR_PTR(-ENOENT);
3489
3490         inode->i_op = &proc_tgid_base_inode_operations;
3491         inode->i_fop = &proc_tgid_base_operations;
3492         inode->i_flags|=S_IMMUTABLE;
3493
3494         set_nlink(inode, nlink_tgid);
3495         pid_update_inode(task, inode);
3496
3497         d_set_d_op(dentry, &pid_dentry_operations);
3498         return d_splice_alias(inode, dentry);
3499 }
3500
3501 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3502 {
3503         struct task_struct *task;
3504         unsigned tgid;
3505         struct proc_fs_info *fs_info;
3506         struct pid_namespace *ns;
3507         struct dentry *result = ERR_PTR(-ENOENT);
3508
3509         tgid = name_to_int(&dentry->d_name);
3510         if (tgid == ~0U)
3511                 goto out;
3512
3513         fs_info = proc_sb_info(dentry->d_sb);
3514         ns = fs_info->pid_ns;
3515         rcu_read_lock();
3516         task = find_task_by_pid_ns(tgid, ns);
3517         if (task)
3518                 get_task_struct(task);
3519         rcu_read_unlock();
3520         if (!task)
3521                 goto out;
3522
3523         /* Limit procfs to only ptraceable tasks */
3524         if (fs_info->hide_pid == HIDEPID_NOT_PTRACEABLE) {
3525                 if (!has_pid_permissions(fs_info, task, HIDEPID_NO_ACCESS))
3526                         goto out_put_task;
3527         }
3528
3529         result = proc_pid_instantiate(dentry, task, NULL);
3530 out_put_task:
3531         put_task_struct(task);
3532 out:
3533         return result;
3534 }
3535
3536 /*
3537  * Find the first task with tgid >= tgid
3538  *
3539  */
3540 struct tgid_iter {
3541         unsigned int tgid;
3542         struct task_struct *task;
3543 };
3544 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3545 {
3546         struct pid *pid;
3547
3548         if (iter.task)
3549                 put_task_struct(iter.task);
3550         rcu_read_lock();
3551 retry:
3552         iter.task = NULL;
3553         pid = find_ge_pid(iter.tgid, ns);
3554         if (pid) {
3555                 iter.tgid = pid_nr_ns(pid, ns);
3556                 iter.task = pid_task(pid, PIDTYPE_TGID);
3557                 if (!iter.task) {
3558                         iter.tgid += 1;
3559                         goto retry;
3560                 }
3561                 get_task_struct(iter.task);
3562         }
3563         rcu_read_unlock();
3564         return iter;
3565 }
3566
3567 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3568
3569 /* for the /proc/ directory itself, after non-process stuff has been done */
3570 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3571 {
3572         struct tgid_iter iter;
3573         struct proc_fs_info *fs_info = proc_sb_info(file_inode(file)->i_sb);
3574         struct pid_namespace *ns = proc_pid_ns(file_inode(file)->i_sb);
3575         loff_t pos = ctx->pos;
3576
3577         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3578                 return 0;
3579
3580         if (pos == TGID_OFFSET - 2) {
3581                 struct inode *inode = d_inode(fs_info->proc_self);
3582                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3583                         return 0;
3584                 ctx->pos = pos = pos + 1;
3585         }
3586         if (pos == TGID_OFFSET - 1) {
3587                 struct inode *inode = d_inode(fs_info->proc_thread_self);
3588                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3589                         return 0;
3590                 ctx->pos = pos = pos + 1;
3591         }
3592         iter.tgid = pos - TGID_OFFSET;
3593         iter.task = NULL;
3594         for (iter = next_tgid(ns, iter);
3595              iter.task;
3596              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3597                 char name[10 + 1];
3598                 unsigned int len;
3599
3600                 cond_resched();
3601                 if (!has_pid_permissions(fs_info, iter.task, HIDEPID_INVISIBLE))
3602                         continue;
3603
3604                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3605                 ctx->pos = iter.tgid + TGID_OFFSET;
3606                 if (!proc_fill_cache(file, ctx, name, len,
3607                                      proc_pid_instantiate, iter.task, NULL)) {
3608                         put_task_struct(iter.task);
3609                         return 0;
3610                 }
3611         }
3612         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3613         return 0;
3614 }
3615
3616 /*
3617  * proc_tid_comm_permission is a special permission function exclusively
3618  * used for the node /proc/<pid>/task/<tid>/comm.
3619  * It bypasses generic permission checks in the case where a task of the same
3620  * task group attempts to access the node.
3621  * The rationale behind this is that glibc and bionic access this node for
3622  * cross thread naming (pthread_set/getname_np(!self)). However, if
3623  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3624  * which locks out the cross thread naming implementation.
3625  * This function makes sure that the node is always accessible for members of
3626  * same thread group.
3627  */
3628 static int proc_tid_comm_permission(struct mnt_idmap *idmap,
3629                                     struct inode *inode, int mask)
3630 {
3631         bool is_same_tgroup;
3632         struct task_struct *task;
3633
3634         task = get_proc_task(inode);
3635         if (!task)
3636                 return -ESRCH;
3637         is_same_tgroup = same_thread_group(current, task);
3638         put_task_struct(task);
3639
3640         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3641                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3642                  * read or written by the members of the corresponding
3643                  * thread group.
3644                  */
3645                 return 0;
3646         }
3647
3648         return generic_permission(&nop_mnt_idmap, inode, mask);
3649 }
3650
3651 static const struct inode_operations proc_tid_comm_inode_operations = {
3652                 .setattr        = proc_setattr,
3653                 .permission     = proc_tid_comm_permission,
3654 };
3655
3656 /*
3657  * Tasks
3658  */
3659 static const struct pid_entry tid_base_stuff[] = {
3660         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3661         DIR("fdinfo",    S_IRUGO|S_IXUGO, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3662         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3663 #ifdef CONFIG_NET
3664         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3665 #endif
3666         REG("environ",   S_IRUSR, proc_environ_operations),
3667         REG("auxv",      S_IRUSR, proc_auxv_operations),
3668         ONE("status",    S_IRUGO, proc_pid_status),
3669         ONE("personality", S_IRUSR, proc_pid_personality),
3670         ONE("limits",    S_IRUGO, proc_pid_limits),
3671 #ifdef CONFIG_SCHED_DEBUG
3672         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3673 #endif
3674         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3675                          &proc_tid_comm_inode_operations,
3676                          &proc_pid_set_comm_operations, {}),
3677 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3678         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3679 #endif
3680         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3681         ONE("stat",      S_IRUGO, proc_tid_stat),
3682         ONE("statm",     S_IRUGO, proc_pid_statm),
3683         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3684 #ifdef CONFIG_PROC_CHILDREN
3685         REG("children",  S_IRUGO, proc_tid_children_operations),
3686 #endif
3687 #ifdef CONFIG_NUMA
3688         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3689 #endif
3690         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3691         LNK("cwd",       proc_cwd_link),
3692         LNK("root",      proc_root_link),
3693         LNK("exe",       proc_exe_link),
3694         REG("mounts",    S_IRUGO, proc_mounts_operations),
3695         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3696 #ifdef CONFIG_PROC_PAGE_MONITOR
3697         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3698         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3699         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3700         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3701 #endif
3702 #ifdef CONFIG_SECURITY
3703         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3704 #endif
3705 #ifdef CONFIG_KALLSYMS
3706         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3707 #endif
3708 #ifdef CONFIG_STACKTRACE
3709         ONE("stack",      S_IRUSR, proc_pid_stack),
3710 #endif
3711 #ifdef CONFIG_SCHED_INFO
3712         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3713 #endif
3714 #ifdef CONFIG_LATENCYTOP
3715         REG("latency",  S_IRUGO, proc_lstats_operations),
3716 #endif
3717 #ifdef CONFIG_PROC_PID_CPUSET
3718         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3719 #endif
3720 #ifdef CONFIG_CGROUPS
3721         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3722 #endif
3723 #ifdef CONFIG_PROC_CPU_RESCTRL
3724         ONE("cpu_resctrl_groups", S_IRUGO, proc_resctrl_show),
3725 #endif
3726         ONE("oom_score", S_IRUGO, proc_oom_score),
3727         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3728         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3729 #ifdef CONFIG_AUDIT
3730         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3731         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3732 #endif
3733 #ifdef CONFIG_FAULT_INJECTION
3734         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3735         REG("fail-nth", 0644, proc_fail_nth_operations),
3736 #endif
3737 #ifdef CONFIG_TASK_IO_ACCOUNTING
3738         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3739 #endif
3740 #ifdef CONFIG_USER_NS
3741         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3742         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3743         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3744         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3745 #endif
3746 #ifdef CONFIG_LIVEPATCH
3747         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3748 #endif
3749 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3750         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3751 #endif
3752 #ifdef CONFIG_SECCOMP_CACHE_DEBUG
3753         ONE("seccomp_cache", S_IRUSR, proc_pid_seccomp_cache),
3754 #endif
3755 #ifdef CONFIG_KSM
3756         ONE("ksm_merging_pages",  S_IRUSR, proc_pid_ksm_merging_pages),
3757         ONE("ksm_stat",  S_IRUSR, proc_pid_ksm_stat),
3758 #endif
3759 };
3760
3761 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3762 {
3763         return proc_pident_readdir(file, ctx,
3764                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3765 }
3766
3767 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3768 {
3769         return proc_pident_lookup(dir, dentry,
3770                                   tid_base_stuff,
3771                                   tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3772 }
3773
3774 static const struct file_operations proc_tid_base_operations = {
3775         .read           = generic_read_dir,
3776         .iterate_shared = proc_tid_base_readdir,
3777         .llseek         = generic_file_llseek,
3778 };
3779
3780 static const struct inode_operations proc_tid_base_inode_operations = {
3781         .lookup         = proc_tid_base_lookup,
3782         .getattr        = pid_getattr,
3783         .setattr        = proc_setattr,
3784 };
3785
3786 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3787         struct task_struct *task, const void *ptr)
3788 {
3789         struct inode *inode;
3790         inode = proc_pid_make_base_inode(dentry->d_sb, task,
3791                                          S_IFDIR | S_IRUGO | S_IXUGO);
3792         if (!inode)
3793                 return ERR_PTR(-ENOENT);
3794
3795         inode->i_op = &proc_tid_base_inode_operations;
3796         inode->i_fop = &proc_tid_base_operations;
3797         inode->i_flags |= S_IMMUTABLE;
3798
3799         set_nlink(inode, nlink_tid);
3800         pid_update_inode(task, inode);
3801
3802         d_set_d_op(dentry, &pid_dentry_operations);
3803         return d_splice_alias(inode, dentry);
3804 }
3805
3806 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3807 {
3808         struct task_struct *task;
3809         struct task_struct *leader = get_proc_task(dir);
3810         unsigned tid;
3811         struct proc_fs_info *fs_info;
3812         struct pid_namespace *ns;
3813         struct dentry *result = ERR_PTR(-ENOENT);
3814
3815         if (!leader)
3816                 goto out_no_task;
3817
3818         tid = name_to_int(&dentry->d_name);
3819         if (tid == ~0U)
3820                 goto out;
3821
3822         fs_info = proc_sb_info(dentry->d_sb);
3823         ns = fs_info->pid_ns;
3824         rcu_read_lock();
3825         task = find_task_by_pid_ns(tid, ns);
3826         if (task)
3827                 get_task_struct(task);
3828         rcu_read_unlock();
3829         if (!task)
3830                 goto out;
3831         if (!same_thread_group(leader, task))
3832                 goto out_drop_task;
3833
3834         result = proc_task_instantiate(dentry, task, NULL);
3835 out_drop_task:
3836         put_task_struct(task);
3837 out:
3838         put_task_struct(leader);
3839 out_no_task:
3840         return result;
3841 }
3842
3843 /*
3844  * Find the first tid of a thread group to return to user space.
3845  *
3846  * Usually this is just the thread group leader, but if the users
3847  * buffer was too small or there was a seek into the middle of the
3848  * directory we have more work todo.
3849  *
3850  * In the case of a short read we start with find_task_by_pid.
3851  *
3852  * In the case of a seek we start with the leader and walk nr
3853  * threads past it.
3854  */
3855 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3856                                         struct pid_namespace *ns)
3857 {
3858         struct task_struct *pos, *task;
3859         unsigned long nr = f_pos;
3860
3861         if (nr != f_pos)        /* 32bit overflow? */
3862                 return NULL;
3863
3864         rcu_read_lock();
3865         task = pid_task(pid, PIDTYPE_PID);
3866         if (!task)
3867                 goto fail;
3868
3869         /* Attempt to start with the tid of a thread */
3870         if (tid && nr) {
3871                 pos = find_task_by_pid_ns(tid, ns);
3872                 if (pos && same_thread_group(pos, task))
3873                         goto found;
3874         }
3875
3876         /* If nr exceeds the number of threads there is nothing todo */
3877         if (nr >= get_nr_threads(task))
3878                 goto fail;
3879
3880         /* If we haven't found our starting place yet start
3881          * with the leader and walk nr threads forward.
3882          */
3883         for_each_thread(task, pos) {
3884                 if (!nr--)
3885                         goto found;
3886         }
3887 fail:
3888         pos = NULL;
3889         goto out;
3890 found:
3891         get_task_struct(pos);
3892 out:
3893         rcu_read_unlock();
3894         return pos;
3895 }
3896
3897 /*
3898  * Find the next thread in the thread list.
3899  * Return NULL if there is an error or no next thread.
3900  *
3901  * The reference to the input task_struct is released.
3902  */
3903 static struct task_struct *next_tid(struct task_struct *start)
3904 {
3905         struct task_struct *pos = NULL;
3906         rcu_read_lock();
3907         if (pid_alive(start)) {
3908                 pos = __next_thread(start);
3909                 if (pos)
3910                         get_task_struct(pos);
3911         }
3912         rcu_read_unlock();
3913         put_task_struct(start);
3914         return pos;
3915 }
3916
3917 /* for the /proc/TGID/task/ directories */
3918 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3919 {
3920         struct inode *inode = file_inode(file);
3921         struct task_struct *task;
3922         struct pid_namespace *ns;
3923         int tid;
3924
3925         if (proc_inode_is_dead(inode))
3926                 return -ENOENT;
3927
3928         if (!dir_emit_dots(file, ctx))
3929                 return 0;
3930
3931         /* We cache the tgid value that the last readdir call couldn't
3932          * return and lseek resets it to 0.
3933          */
3934         ns = proc_pid_ns(inode->i_sb);
3935         tid = (int)(intptr_t)file->private_data;
3936         file->private_data = NULL;
3937         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3938              task;
3939              task = next_tid(task), ctx->pos++) {
3940                 char name[10 + 1];
3941                 unsigned int len;
3942
3943                 tid = task_pid_nr_ns(task, ns);
3944                 if (!tid)
3945                         continue;       /* The task has just exited. */
3946                 len = snprintf(name, sizeof(name), "%u", tid);
3947                 if (!proc_fill_cache(file, ctx, name, len,
3948                                 proc_task_instantiate, task, NULL)) {
3949                         /* returning this tgid failed, save it as the first
3950                          * pid for the next readir call */
3951                         file->private_data = (void *)(intptr_t)tid;
3952                         put_task_struct(task);
3953                         break;
3954                 }
3955         }
3956
3957         return 0;
3958 }
3959
3960 static int proc_task_getattr(struct mnt_idmap *idmap,
3961                              const struct path *path, struct kstat *stat,
3962                              u32 request_mask, unsigned int query_flags)
3963 {
3964         struct inode *inode = d_inode(path->dentry);
3965         struct task_struct *p = get_proc_task(inode);
3966         generic_fillattr(&nop_mnt_idmap, request_mask, inode, stat);
3967
3968         if (p) {
3969                 stat->nlink += get_nr_threads(p);
3970                 put_task_struct(p);
3971         }
3972
3973         return 0;
3974 }
3975
3976 /*
3977  * proc_task_readdir() set @file->private_data to a positive integer
3978  * value, so casting that to u64 is safe. generic_llseek_cookie() will
3979  * set @cookie to 0, so casting to an int is safe. The WARN_ON_ONCE() is
3980  * here to catch any unexpected change in behavior either in
3981  * proc_task_readdir() or generic_llseek_cookie().
3982  */
3983 static loff_t proc_dir_llseek(struct file *file, loff_t offset, int whence)
3984 {
3985         u64 cookie = (u64)(intptr_t)file->private_data;
3986         loff_t off;
3987
3988         off = generic_llseek_cookie(file, offset, whence, &cookie);
3989         WARN_ON_ONCE(cookie > INT_MAX);
3990         file->private_data = (void *)(intptr_t)cookie; /* serialized by f_pos_lock */
3991         return off;
3992 }
3993
3994 static const struct inode_operations proc_task_inode_operations = {
3995         .lookup         = proc_task_lookup,
3996         .getattr        = proc_task_getattr,
3997         .setattr        = proc_setattr,
3998         .permission     = proc_pid_permission,
3999 };
4000
4001 static const struct file_operations proc_task_operations = {
4002         .read           = generic_read_dir,
4003         .iterate_shared = proc_task_readdir,
4004         .llseek         = proc_dir_llseek,
4005 };
4006
4007 void __init set_proc_pid_nlink(void)
4008 {
4009         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
4010         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
4011 }