Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/kernel/printk.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 *
   7 * Modified to make sys_syslog() more flexible: added commands to
   8 * return the last 4k of kernel messages, regardless of whether
   9 * they've been read or not.  Added option to suppress kernel printk's
  10 * to the console.  Added hook for sending the console messages
  11 * elsewhere, in preparation for a serial line console (someday).
  12 * Ted Ts'o, 2/11/93.
  13 * Modified for sysctl support, 1/8/97, Chris Horn.
  14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  15 *     manfred@colorfullife.com
  16 * Rewrote bits to get rid of console_lock
  17 *	01Mar01 Andrew Morton
  18 */
  19
  20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22#include <linux/kernel.h>
  23#include <linux/mm.h>
  24#include <linux/tty.h>
  25#include <linux/tty_driver.h>
  26#include <linux/console.h>
  27#include <linux/init.h>
  28#include <linux/jiffies.h>
  29#include <linux/nmi.h>
  30#include <linux/module.h>
  31#include <linux/moduleparam.h>
  32#include <linux/delay.h>
  33#include <linux/smp.h>
  34#include <linux/security.h>
  35#include <linux/memblock.h>
  36#include <linux/syscalls.h>
  37#include <linux/crash_core.h>
  38#include <linux/ratelimit.h>
  39#include <linux/kmsg_dump.h>
  40#include <linux/syslog.h>
  41#include <linux/cpu.h>
  42#include <linux/rculist.h>
  43#include <linux/poll.h>
  44#include <linux/irq_work.h>
  45#include <linux/ctype.h>
  46#include <linux/uio.h>
  47#include <linux/sched/clock.h>
  48#include <linux/sched/debug.h>
  49#include <linux/sched/task_stack.h>
  50
  51#include <linux/uaccess.h>
  52#include <asm/sections.h>
  53
  54#include <trace/events/initcall.h>
  55#define CREATE_TRACE_POINTS
  56#include <trace/events/printk.h>
  57
  58#include "printk_ringbuffer.h"
  59#include "console_cmdline.h"
  60#include "braille.h"
  61#include "internal.h"
  62
  63int console_printk[4] = {
  64	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  65	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  66	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  67	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  68};
  69EXPORT_SYMBOL_GPL(console_printk);
  70
  71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
  72EXPORT_SYMBOL(ignore_console_lock_warning);
  73
 
 
  74/*
  75 * Low level drivers may need that to know if they can schedule in
  76 * their unblank() callback or not. So let's export it.
  77 */
  78int oops_in_progress;
  79EXPORT_SYMBOL(oops_in_progress);
  80
  81/*
  82 * console_mutex protects console_list updates and console->flags updates.
  83 * The flags are synchronized only for consoles that are registered, i.e.
  84 * accessible via the console list.
  85 */
  86static DEFINE_MUTEX(console_mutex);
  87
  88/*
  89 * console_sem protects updates to console->seq and console_suspended,
  90 * and also provides serialization for console printing.
  91 */
  92static DEFINE_SEMAPHORE(console_sem);
  93HLIST_HEAD(console_list);
  94EXPORT_SYMBOL_GPL(console_list);
  95DEFINE_STATIC_SRCU(console_srcu);
  96
  97/*
  98 * System may need to suppress printk message under certain
  99 * circumstances, like after kernel panic happens.
 100 */
 101int __read_mostly suppress_printk;
 102
 103/*
 104 * During panic, heavy printk by other CPUs can delay the
 105 * panic and risk deadlock on console resources.
 106 */
 107static int __read_mostly suppress_panic_printk;
 108
 109#ifdef CONFIG_LOCKDEP
 110static struct lockdep_map console_lock_dep_map = {
 111	.name = "console_lock"
 112};
 113
 114void lockdep_assert_console_list_lock_held(void)
 115{
 116	lockdep_assert_held(&console_mutex);
 117}
 118EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
 119#endif
 120
 121#ifdef CONFIG_DEBUG_LOCK_ALLOC
 122bool console_srcu_read_lock_is_held(void)
 123{
 124	return srcu_read_lock_held(&console_srcu);
 125}
 126EXPORT_SYMBOL(console_srcu_read_lock_is_held);
 127#endif
 128
 129enum devkmsg_log_bits {
 130	__DEVKMSG_LOG_BIT_ON = 0,
 131	__DEVKMSG_LOG_BIT_OFF,
 132	__DEVKMSG_LOG_BIT_LOCK,
 133};
 134
 135enum devkmsg_log_masks {
 136	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
 137	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 138	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 139};
 140
 141/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 142#define DEVKMSG_LOG_MASK_DEFAULT	0
 143
 144static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 145
 146static int __control_devkmsg(char *str)
 147{
 148	size_t len;
 149
 150	if (!str)
 151		return -EINVAL;
 152
 153	len = str_has_prefix(str, "on");
 154	if (len) {
 155		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 156		return len;
 157	}
 158
 159	len = str_has_prefix(str, "off");
 160	if (len) {
 161		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 162		return len;
 163	}
 164
 165	len = str_has_prefix(str, "ratelimit");
 166	if (len) {
 167		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 168		return len;
 169	}
 170
 171	return -EINVAL;
 172}
 173
 174static int __init control_devkmsg(char *str)
 175{
 176	if (__control_devkmsg(str) < 0) {
 177		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
 178		return 1;
 179	}
 180
 181	/*
 182	 * Set sysctl string accordingly:
 183	 */
 184	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 185		strcpy(devkmsg_log_str, "on");
 186	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 187		strcpy(devkmsg_log_str, "off");
 188	/* else "ratelimit" which is set by default. */
 189
 190	/*
 191	 * Sysctl cannot change it anymore. The kernel command line setting of
 192	 * this parameter is to force the setting to be permanent throughout the
 193	 * runtime of the system. This is a precation measure against userspace
 194	 * trying to be a smarta** and attempting to change it up on us.
 195	 */
 196	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 197
 198	return 1;
 199}
 200__setup("printk.devkmsg=", control_devkmsg);
 201
 202char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 203#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
 204int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 205			      void *buffer, size_t *lenp, loff_t *ppos)
 206{
 207	char old_str[DEVKMSG_STR_MAX_SIZE];
 208	unsigned int old;
 209	int err;
 210
 211	if (write) {
 212		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 213			return -EINVAL;
 214
 215		old = devkmsg_log;
 216		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 217	}
 218
 219	err = proc_dostring(table, write, buffer, lenp, ppos);
 220	if (err)
 221		return err;
 222
 223	if (write) {
 224		err = __control_devkmsg(devkmsg_log_str);
 225
 226		/*
 227		 * Do not accept an unknown string OR a known string with
 228		 * trailing crap...
 229		 */
 230		if (err < 0 || (err + 1 != *lenp)) {
 231
 232			/* ... and restore old setting. */
 233			devkmsg_log = old;
 234			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 235
 236			return -EINVAL;
 237		}
 238	}
 239
 240	return 0;
 241}
 242#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
 243
 244/**
 245 * console_list_lock - Lock the console list
 246 *
 247 * For console list or console->flags updates
 248 */
 249void console_list_lock(void)
 250{
 251	/*
 252	 * In unregister_console() and console_force_preferred_locked(),
 253	 * synchronize_srcu() is called with the console_list_lock held.
 254	 * Therefore it is not allowed that the console_list_lock is taken
 255	 * with the srcu_lock held.
 256	 *
 257	 * Detecting if this context is really in the read-side critical
 258	 * section is only possible if the appropriate debug options are
 259	 * enabled.
 260	 */
 261	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
 262		     srcu_read_lock_held(&console_srcu));
 263
 264	mutex_lock(&console_mutex);
 265}
 266EXPORT_SYMBOL(console_list_lock);
 267
 268/**
 269 * console_list_unlock - Unlock the console list
 270 *
 271 * Counterpart to console_list_lock()
 272 */
 273void console_list_unlock(void)
 274{
 275	mutex_unlock(&console_mutex);
 276}
 277EXPORT_SYMBOL(console_list_unlock);
 278
 279/**
 280 * console_srcu_read_lock - Register a new reader for the
 281 *	SRCU-protected console list
 282 *
 283 * Use for_each_console_srcu() to iterate the console list
 284 *
 285 * Context: Any context.
 286 * Return: A cookie to pass to console_srcu_read_unlock().
 287 */
 288int console_srcu_read_lock(void)
 289{
 290	return srcu_read_lock_nmisafe(&console_srcu);
 291}
 292EXPORT_SYMBOL(console_srcu_read_lock);
 293
 294/**
 295 * console_srcu_read_unlock - Unregister an old reader from
 296 *	the SRCU-protected console list
 297 * @cookie: cookie returned from console_srcu_read_lock()
 298 *
 299 * Counterpart to console_srcu_read_lock()
 300 */
 301void console_srcu_read_unlock(int cookie)
 302{
 303	srcu_read_unlock_nmisafe(&console_srcu, cookie);
 304}
 305EXPORT_SYMBOL(console_srcu_read_unlock);
 306
 307/*
 308 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 309 * macros instead of functions so that _RET_IP_ contains useful information.
 310 */
 311#define down_console_sem() do { \
 312	down(&console_sem);\
 313	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 314} while (0)
 315
 316static int __down_trylock_console_sem(unsigned long ip)
 317{
 318	int lock_failed;
 319	unsigned long flags;
 320
 321	/*
 322	 * Here and in __up_console_sem() we need to be in safe mode,
 323	 * because spindump/WARN/etc from under console ->lock will
 324	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 325	 */
 326	printk_safe_enter_irqsave(flags);
 327	lock_failed = down_trylock(&console_sem);
 328	printk_safe_exit_irqrestore(flags);
 329
 330	if (lock_failed)
 331		return 1;
 332	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 333	return 0;
 334}
 335#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 336
 337static void __up_console_sem(unsigned long ip)
 338{
 339	unsigned long flags;
 340
 341	mutex_release(&console_lock_dep_map, ip);
 342
 343	printk_safe_enter_irqsave(flags);
 344	up(&console_sem);
 345	printk_safe_exit_irqrestore(flags);
 346}
 347#define up_console_sem() __up_console_sem(_RET_IP_)
 348
 349static bool panic_in_progress(void)
 350{
 351	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
 352}
 353
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 354/*
 355 * This is used for debugging the mess that is the VT code by
 356 * keeping track if we have the console semaphore held. It's
 357 * definitely not the perfect debug tool (we don't know if _WE_
 358 * hold it and are racing, but it helps tracking those weird code
 359 * paths in the console code where we end up in places I want
 360 * locked without the console semaphore held).
 361 */
 362static int console_locked, console_suspended;
 363
 364/*
 365 *	Array of consoles built from command line options (console=)
 366 */
 367
 368#define MAX_CMDLINECONSOLES 8
 369
 370static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 371
 372static int preferred_console = -1;
 373int console_set_on_cmdline;
 374EXPORT_SYMBOL(console_set_on_cmdline);
 375
 376/* Flag: console code may call schedule() */
 377static int console_may_schedule;
 378
 379enum con_msg_format_flags {
 380	MSG_FORMAT_DEFAULT	= 0,
 381	MSG_FORMAT_SYSLOG	= (1 << 0),
 382};
 383
 384static int console_msg_format = MSG_FORMAT_DEFAULT;
 385
 386/*
 387 * The printk log buffer consists of a sequenced collection of records, each
 388 * containing variable length message text. Every record also contains its
 389 * own meta-data (@info).
 390 *
 391 * Every record meta-data carries the timestamp in microseconds, as well as
 392 * the standard userspace syslog level and syslog facility. The usual kernel
 393 * messages use LOG_KERN; userspace-injected messages always carry a matching
 394 * syslog facility, by default LOG_USER. The origin of every message can be
 395 * reliably determined that way.
 396 *
 397 * The human readable log message of a record is available in @text, the
 398 * length of the message text in @text_len. The stored message is not
 399 * terminated.
 400 *
 401 * Optionally, a record can carry a dictionary of properties (key/value
 402 * pairs), to provide userspace with a machine-readable message context.
 403 *
 404 * Examples for well-defined, commonly used property names are:
 405 *   DEVICE=b12:8               device identifier
 406 *                                b12:8         block dev_t
 407 *                                c127:3        char dev_t
 408 *                                n8            netdev ifindex
 409 *                                +sound:card0  subsystem:devname
 410 *   SUBSYSTEM=pci              driver-core subsystem name
 411 *
 412 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
 413 * and values are terminated by a '\0' character.
 414 *
 415 * Example of record values:
 416 *   record.text_buf                = "it's a line" (unterminated)
 417 *   record.info.seq                = 56
 418 *   record.info.ts_nsec            = 36863
 419 *   record.info.text_len           = 11
 420 *   record.info.facility           = 0 (LOG_KERN)
 421 *   record.info.flags              = 0
 422 *   record.info.level              = 3 (LOG_ERR)
 423 *   record.info.caller_id          = 299 (task 299)
 424 *   record.info.dev_info.subsystem = "pci" (terminated)
 425 *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
 426 *
 427 * The 'struct printk_info' buffer must never be directly exported to
 428 * userspace, it is a kernel-private implementation detail that might
 429 * need to be changed in the future, when the requirements change.
 430 *
 431 * /dev/kmsg exports the structured data in the following line format:
 432 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 433 *
 434 * Users of the export format should ignore possible additional values
 435 * separated by ',', and find the message after the ';' character.
 436 *
 437 * The optional key/value pairs are attached as continuation lines starting
 438 * with a space character and terminated by a newline. All possible
 439 * non-prinatable characters are escaped in the "\xff" notation.
 440 */
 441
 442/* syslog_lock protects syslog_* variables and write access to clear_seq. */
 443static DEFINE_MUTEX(syslog_lock);
 444
 445#ifdef CONFIG_PRINTK
 446DECLARE_WAIT_QUEUE_HEAD(log_wait);
 447/* All 3 protected by @syslog_lock. */
 448/* the next printk record to read by syslog(READ) or /proc/kmsg */
 449static u64 syslog_seq;
 450static size_t syslog_partial;
 451static bool syslog_time;
 452
 453struct latched_seq {
 454	seqcount_latch_t	latch;
 455	u64			val[2];
 456};
 457
 458/*
 459 * The next printk record to read after the last 'clear' command. There are
 460 * two copies (updated with seqcount_latch) so that reads can locklessly
 461 * access a valid value. Writers are synchronized by @syslog_lock.
 462 */
 463static struct latched_seq clear_seq = {
 464	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
 465	.val[0]		= 0,
 466	.val[1]		= 0,
 467};
 468
 469#ifdef CONFIG_PRINTK_CALLER
 470#define PREFIX_MAX		48
 471#else
 472#define PREFIX_MAX		32
 473#endif
 474
 475/* the maximum size of a formatted record (i.e. with prefix added per line) */
 476#define CONSOLE_LOG_MAX		1024
 477
 478/* the maximum size for a dropped text message */
 479#define DROPPED_TEXT_MAX	64
 480
 481/* the maximum size allowed to be reserved for a record */
 482#define LOG_LINE_MAX		(CONSOLE_LOG_MAX - PREFIX_MAX)
 483
 484#define LOG_LEVEL(v)		((v) & 0x07)
 485#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 486
 487/* record buffer */
 488#define LOG_ALIGN __alignof__(unsigned long)
 489#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 490#define LOG_BUF_LEN_MAX (u32)(1 << 31)
 491static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 492static char *log_buf = __log_buf;
 493static u32 log_buf_len = __LOG_BUF_LEN;
 494
 495/*
 496 * Define the average message size. This only affects the number of
 497 * descriptors that will be available. Underestimating is better than
 498 * overestimating (too many available descriptors is better than not enough).
 499 */
 500#define PRB_AVGBITS 5	/* 32 character average length */
 501
 502#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
 503#error CONFIG_LOG_BUF_SHIFT value too small.
 504#endif
 505_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
 506		 PRB_AVGBITS, &__log_buf[0]);
 507
 508static struct printk_ringbuffer printk_rb_dynamic;
 509
 510static struct printk_ringbuffer *prb = &printk_rb_static;
 511
 512/*
 513 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
 514 * per_cpu_areas are initialised. This variable is set to true when
 515 * it's safe to access per-CPU data.
 516 */
 517static bool __printk_percpu_data_ready __ro_after_init;
 518
 519bool printk_percpu_data_ready(void)
 520{
 521	return __printk_percpu_data_ready;
 522}
 523
 524/* Must be called under syslog_lock. */
 525static void latched_seq_write(struct latched_seq *ls, u64 val)
 526{
 527	raw_write_seqcount_latch(&ls->latch);
 528	ls->val[0] = val;
 529	raw_write_seqcount_latch(&ls->latch);
 530	ls->val[1] = val;
 531}
 532
 533/* Can be called from any context. */
 534static u64 latched_seq_read_nolock(struct latched_seq *ls)
 535{
 536	unsigned int seq;
 537	unsigned int idx;
 538	u64 val;
 539
 540	do {
 541		seq = raw_read_seqcount_latch(&ls->latch);
 542		idx = seq & 0x1;
 543		val = ls->val[idx];
 544	} while (read_seqcount_latch_retry(&ls->latch, seq));
 545
 546	return val;
 547}
 548
 549/* Return log buffer address */
 550char *log_buf_addr_get(void)
 551{
 552	return log_buf;
 553}
 554
 555/* Return log buffer size */
 556u32 log_buf_len_get(void)
 557{
 558	return log_buf_len;
 559}
 560
 561/*
 562 * Define how much of the log buffer we could take at maximum. The value
 563 * must be greater than two. Note that only half of the buffer is available
 564 * when the index points to the middle.
 565 */
 566#define MAX_LOG_TAKE_PART 4
 567static const char trunc_msg[] = "<truncated>";
 568
 569static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
 570{
 571	/*
 572	 * The message should not take the whole buffer. Otherwise, it might
 573	 * get removed too soon.
 574	 */
 575	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 576
 577	if (*text_len > max_text_len)
 578		*text_len = max_text_len;
 579
 580	/* enable the warning message (if there is room) */
 581	*trunc_msg_len = strlen(trunc_msg);
 582	if (*text_len >= *trunc_msg_len)
 583		*text_len -= *trunc_msg_len;
 584	else
 585		*trunc_msg_len = 0;
 586}
 587
 588int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 589
 590static int syslog_action_restricted(int type)
 591{
 592	if (dmesg_restrict)
 593		return 1;
 594	/*
 595	 * Unless restricted, we allow "read all" and "get buffer size"
 596	 * for everybody.
 597	 */
 598	return type != SYSLOG_ACTION_READ_ALL &&
 599	       type != SYSLOG_ACTION_SIZE_BUFFER;
 600}
 601
 602static int check_syslog_permissions(int type, int source)
 603{
 604	/*
 605	 * If this is from /proc/kmsg and we've already opened it, then we've
 606	 * already done the capabilities checks at open time.
 607	 */
 608	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 609		goto ok;
 610
 611	if (syslog_action_restricted(type)) {
 612		if (capable(CAP_SYSLOG))
 613			goto ok;
 614		/*
 615		 * For historical reasons, accept CAP_SYS_ADMIN too, with
 616		 * a warning.
 617		 */
 618		if (capable(CAP_SYS_ADMIN)) {
 619			pr_warn_once("%s (%d): Attempt to access syslog with "
 620				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
 621				     "(deprecated).\n",
 622				 current->comm, task_pid_nr(current));
 623			goto ok;
 624		}
 625		return -EPERM;
 626	}
 627ok:
 628	return security_syslog(type);
 629}
 630
 631static void append_char(char **pp, char *e, char c)
 632{
 633	if (*pp < e)
 634		*(*pp)++ = c;
 635}
 636
 637static ssize_t info_print_ext_header(char *buf, size_t size,
 638				     struct printk_info *info)
 639{
 640	u64 ts_usec = info->ts_nsec;
 641	char caller[20];
 642#ifdef CONFIG_PRINTK_CALLER
 643	u32 id = info->caller_id;
 644
 645	snprintf(caller, sizeof(caller), ",caller=%c%u",
 646		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
 647#else
 648	caller[0] = '\0';
 649#endif
 650
 651	do_div(ts_usec, 1000);
 652
 653	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
 654			 (info->facility << 3) | info->level, info->seq,
 655			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
 656}
 657
 658static ssize_t msg_add_ext_text(char *buf, size_t size,
 659				const char *text, size_t text_len,
 660				unsigned char endc)
 661{
 662	char *p = buf, *e = buf + size;
 663	size_t i;
 664
 665	/* escape non-printable characters */
 666	for (i = 0; i < text_len; i++) {
 667		unsigned char c = text[i];
 668
 669		if (c < ' ' || c >= 127 || c == '\\')
 670			p += scnprintf(p, e - p, "\\x%02x", c);
 671		else
 672			append_char(&p, e, c);
 673	}
 674	append_char(&p, e, endc);
 675
 676	return p - buf;
 677}
 678
 679static ssize_t msg_add_dict_text(char *buf, size_t size,
 680				 const char *key, const char *val)
 681{
 682	size_t val_len = strlen(val);
 683	ssize_t len;
 684
 685	if (!val_len)
 686		return 0;
 687
 688	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
 689	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
 690	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
 691
 692	return len;
 693}
 694
 695static ssize_t msg_print_ext_body(char *buf, size_t size,
 696				  char *text, size_t text_len,
 697				  struct dev_printk_info *dev_info)
 698{
 699	ssize_t len;
 700
 701	len = msg_add_ext_text(buf, size, text, text_len, '\n');
 702
 703	if (!dev_info)
 704		goto out;
 705
 706	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
 707				 dev_info->subsystem);
 708	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
 709				 dev_info->device);
 710out:
 711	return len;
 712}
 713
 714/* /dev/kmsg - userspace message inject/listen interface */
 715struct devkmsg_user {
 716	atomic64_t seq;
 717	struct ratelimit_state rs;
 718	struct mutex lock;
 719	char buf[CONSOLE_EXT_LOG_MAX];
 720
 721	struct printk_info info;
 722	char text_buf[CONSOLE_EXT_LOG_MAX];
 723	struct printk_record record;
 724};
 725
 726static __printf(3, 4) __cold
 727int devkmsg_emit(int facility, int level, const char *fmt, ...)
 728{
 729	va_list args;
 730	int r;
 731
 732	va_start(args, fmt);
 733	r = vprintk_emit(facility, level, NULL, fmt, args);
 734	va_end(args);
 735
 736	return r;
 737}
 738
 739static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 740{
 741	char *buf, *line;
 742	int level = default_message_loglevel;
 743	int facility = 1;	/* LOG_USER */
 744	struct file *file = iocb->ki_filp;
 745	struct devkmsg_user *user = file->private_data;
 746	size_t len = iov_iter_count(from);
 747	ssize_t ret = len;
 748
 749	if (!user || len > LOG_LINE_MAX)
 750		return -EINVAL;
 751
 752	/* Ignore when user logging is disabled. */
 753	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 754		return len;
 755
 756	/* Ratelimit when not explicitly enabled. */
 757	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 758		if (!___ratelimit(&user->rs, current->comm))
 759			return ret;
 760	}
 761
 762	buf = kmalloc(len+1, GFP_KERNEL);
 763	if (buf == NULL)
 764		return -ENOMEM;
 765
 766	buf[len] = '\0';
 767	if (!copy_from_iter_full(buf, len, from)) {
 768		kfree(buf);
 769		return -EFAULT;
 770	}
 771
 772	/*
 773	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 774	 * the decimal value represents 32bit, the lower 3 bit are the log
 775	 * level, the rest are the log facility.
 776	 *
 777	 * If no prefix or no userspace facility is specified, we
 778	 * enforce LOG_USER, to be able to reliably distinguish
 779	 * kernel-generated messages from userspace-injected ones.
 780	 */
 781	line = buf;
 782	if (line[0] == '<') {
 783		char *endp = NULL;
 784		unsigned int u;
 785
 786		u = simple_strtoul(line + 1, &endp, 10);
 787		if (endp && endp[0] == '>') {
 788			level = LOG_LEVEL(u);
 789			if (LOG_FACILITY(u) != 0)
 790				facility = LOG_FACILITY(u);
 791			endp++;
 792			line = endp;
 793		}
 794	}
 795
 796	devkmsg_emit(facility, level, "%s", line);
 797	kfree(buf);
 798	return ret;
 799}
 800
 801static ssize_t devkmsg_read(struct file *file, char __user *buf,
 802			    size_t count, loff_t *ppos)
 803{
 804	struct devkmsg_user *user = file->private_data;
 805	struct printk_record *r = &user->record;
 806	size_t len;
 
 
 807	ssize_t ret;
 808
 809	if (!user)
 810		return -EBADF;
 811
 812	ret = mutex_lock_interruptible(&user->lock);
 813	if (ret)
 814		return ret;
 815
 816	if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
 817		if (file->f_flags & O_NONBLOCK) {
 818			ret = -EAGAIN;
 819			goto out;
 820		}
 821
 822		/*
 823		 * Guarantee this task is visible on the waitqueue before
 824		 * checking the wake condition.
 825		 *
 826		 * The full memory barrier within set_current_state() of
 827		 * prepare_to_wait_event() pairs with the full memory barrier
 828		 * within wq_has_sleeper().
 829		 *
 830		 * This pairs with __wake_up_klogd:A.
 831		 */
 832		ret = wait_event_interruptible(log_wait,
 833				prb_read_valid(prb,
 834					atomic64_read(&user->seq), r)); /* LMM(devkmsg_read:A) */
 835		if (ret)
 836			goto out;
 837	}
 838
 839	if (r->info->seq != atomic64_read(&user->seq)) {
 840		/* our last seen message is gone, return error and reset */
 841		atomic64_set(&user->seq, r->info->seq);
 842		ret = -EPIPE;
 843		goto out;
 844	}
 845
 846	len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
 847	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
 848				  &r->text_buf[0], r->info->text_len,
 849				  &r->info->dev_info);
 850
 851	atomic64_set(&user->seq, r->info->seq + 1);
 852
 853	if (len > count) {
 854		ret = -EINVAL;
 855		goto out;
 856	}
 857
 858	if (copy_to_user(buf, user->buf, len)) {
 859		ret = -EFAULT;
 860		goto out;
 861	}
 862	ret = len;
 863out:
 864	mutex_unlock(&user->lock);
 865	return ret;
 866}
 867
 868/*
 869 * Be careful when modifying this function!!!
 870 *
 871 * Only few operations are supported because the device works only with the
 872 * entire variable length messages (records). Non-standard values are
 873 * returned in the other cases and has been this way for quite some time.
 874 * User space applications might depend on this behavior.
 875 */
 876static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 877{
 878	struct devkmsg_user *user = file->private_data;
 879	loff_t ret = 0;
 880
 881	if (!user)
 882		return -EBADF;
 883	if (offset)
 884		return -ESPIPE;
 885
 886	switch (whence) {
 887	case SEEK_SET:
 888		/* the first record */
 889		atomic64_set(&user->seq, prb_first_valid_seq(prb));
 890		break;
 891	case SEEK_DATA:
 892		/*
 893		 * The first record after the last SYSLOG_ACTION_CLEAR,
 894		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 895		 * changes no global state, and does not clear anything.
 896		 */
 897		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
 898		break;
 899	case SEEK_END:
 900		/* after the last record */
 901		atomic64_set(&user->seq, prb_next_seq(prb));
 902		break;
 903	default:
 904		ret = -EINVAL;
 905	}
 906	return ret;
 907}
 908
 909static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 910{
 911	struct devkmsg_user *user = file->private_data;
 912	struct printk_info info;
 913	__poll_t ret = 0;
 914
 915	if (!user)
 916		return EPOLLERR|EPOLLNVAL;
 917
 918	poll_wait(file, &log_wait, wait);
 919
 920	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
 921		/* return error when data has vanished underneath us */
 922		if (info.seq != atomic64_read(&user->seq))
 923			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 924		else
 925			ret = EPOLLIN|EPOLLRDNORM;
 926	}
 927
 928	return ret;
 929}
 930
 931static int devkmsg_open(struct inode *inode, struct file *file)
 932{
 933	struct devkmsg_user *user;
 934	int err;
 935
 936	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 937		return -EPERM;
 938
 939	/* write-only does not need any file context */
 940	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 941		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 942					       SYSLOG_FROM_READER);
 943		if (err)
 944			return err;
 945	}
 946
 947	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 948	if (!user)
 949		return -ENOMEM;
 950
 951	ratelimit_default_init(&user->rs);
 952	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 953
 954	mutex_init(&user->lock);
 955
 956	prb_rec_init_rd(&user->record, &user->info,
 957			&user->text_buf[0], sizeof(user->text_buf));
 958
 959	atomic64_set(&user->seq, prb_first_valid_seq(prb));
 960
 961	file->private_data = user;
 962	return 0;
 963}
 964
 965static int devkmsg_release(struct inode *inode, struct file *file)
 966{
 967	struct devkmsg_user *user = file->private_data;
 968
 969	if (!user)
 970		return 0;
 971
 972	ratelimit_state_exit(&user->rs);
 973
 974	mutex_destroy(&user->lock);
 975	kvfree(user);
 976	return 0;
 977}
 978
 979const struct file_operations kmsg_fops = {
 980	.open = devkmsg_open,
 981	.read = devkmsg_read,
 982	.write_iter = devkmsg_write,
 983	.llseek = devkmsg_llseek,
 984	.poll = devkmsg_poll,
 985	.release = devkmsg_release,
 986};
 987
 988#ifdef CONFIG_CRASH_CORE
 989/*
 990 * This appends the listed symbols to /proc/vmcore
 991 *
 992 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 993 * obtain access to symbols that are otherwise very difficult to locate.  These
 994 * symbols are specifically used so that utilities can access and extract the
 995 * dmesg log from a vmcore file after a crash.
 996 */
 997void log_buf_vmcoreinfo_setup(void)
 998{
 999	struct dev_printk_info *dev_info = NULL;
1000
1001	VMCOREINFO_SYMBOL(prb);
1002	VMCOREINFO_SYMBOL(printk_rb_static);
1003	VMCOREINFO_SYMBOL(clear_seq);
1004
1005	/*
1006	 * Export struct size and field offsets. User space tools can
1007	 * parse it and detect any changes to structure down the line.
1008	 */
1009
1010	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
1011	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
1012	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
1013	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
1014
1015	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
1016	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
1017	VMCOREINFO_OFFSET(prb_desc_ring, descs);
1018	VMCOREINFO_OFFSET(prb_desc_ring, infos);
1019	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
1020	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
1021
1022	VMCOREINFO_STRUCT_SIZE(prb_desc);
1023	VMCOREINFO_OFFSET(prb_desc, state_var);
1024	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
1025
1026	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
1027	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1028	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1029
1030	VMCOREINFO_STRUCT_SIZE(printk_info);
1031	VMCOREINFO_OFFSET(printk_info, seq);
1032	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1033	VMCOREINFO_OFFSET(printk_info, text_len);
1034	VMCOREINFO_OFFSET(printk_info, caller_id);
1035	VMCOREINFO_OFFSET(printk_info, dev_info);
1036
1037	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1038	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1039	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1040	VMCOREINFO_OFFSET(dev_printk_info, device);
1041	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1042
1043	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1044	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1045	VMCOREINFO_OFFSET(prb_data_ring, data);
1046	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1047	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1048
1049	VMCOREINFO_SIZE(atomic_long_t);
1050	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1051
1052	VMCOREINFO_STRUCT_SIZE(latched_seq);
1053	VMCOREINFO_OFFSET(latched_seq, val);
1054}
1055#endif
1056
1057/* requested log_buf_len from kernel cmdline */
1058static unsigned long __initdata new_log_buf_len;
1059
1060/* we practice scaling the ring buffer by powers of 2 */
1061static void __init log_buf_len_update(u64 size)
1062{
1063	if (size > (u64)LOG_BUF_LEN_MAX) {
1064		size = (u64)LOG_BUF_LEN_MAX;
1065		pr_err("log_buf over 2G is not supported.\n");
1066	}
1067
1068	if (size)
1069		size = roundup_pow_of_two(size);
1070	if (size > log_buf_len)
1071		new_log_buf_len = (unsigned long)size;
1072}
1073
1074/* save requested log_buf_len since it's too early to process it */
1075static int __init log_buf_len_setup(char *str)
1076{
1077	u64 size;
1078
1079	if (!str)
1080		return -EINVAL;
1081
1082	size = memparse(str, &str);
1083
1084	log_buf_len_update(size);
1085
1086	return 0;
1087}
1088early_param("log_buf_len", log_buf_len_setup);
1089
1090#ifdef CONFIG_SMP
1091#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1092
1093static void __init log_buf_add_cpu(void)
1094{
1095	unsigned int cpu_extra;
1096
1097	/*
1098	 * archs should set up cpu_possible_bits properly with
1099	 * set_cpu_possible() after setup_arch() but just in
1100	 * case lets ensure this is valid.
1101	 */
1102	if (num_possible_cpus() == 1)
1103		return;
1104
1105	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1106
1107	/* by default this will only continue through for large > 64 CPUs */
1108	if (cpu_extra <= __LOG_BUF_LEN / 2)
1109		return;
1110
1111	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1112		__LOG_CPU_MAX_BUF_LEN);
1113	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1114		cpu_extra);
1115	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1116
1117	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1118}
1119#else /* !CONFIG_SMP */
1120static inline void log_buf_add_cpu(void) {}
1121#endif /* CONFIG_SMP */
1122
1123static void __init set_percpu_data_ready(void)
1124{
1125	__printk_percpu_data_ready = true;
1126}
1127
1128static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1129				     struct printk_record *r)
1130{
1131	struct prb_reserved_entry e;
1132	struct printk_record dest_r;
1133
1134	prb_rec_init_wr(&dest_r, r->info->text_len);
1135
1136	if (!prb_reserve(&e, rb, &dest_r))
1137		return 0;
1138
1139	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1140	dest_r.info->text_len = r->info->text_len;
1141	dest_r.info->facility = r->info->facility;
1142	dest_r.info->level = r->info->level;
1143	dest_r.info->flags = r->info->flags;
1144	dest_r.info->ts_nsec = r->info->ts_nsec;
1145	dest_r.info->caller_id = r->info->caller_id;
1146	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1147
1148	prb_final_commit(&e);
1149
1150	return prb_record_text_space(&e);
1151}
1152
1153static char setup_text_buf[LOG_LINE_MAX] __initdata;
1154
1155void __init setup_log_buf(int early)
1156{
1157	struct printk_info *new_infos;
1158	unsigned int new_descs_count;
1159	struct prb_desc *new_descs;
1160	struct printk_info info;
1161	struct printk_record r;
1162	unsigned int text_size;
1163	size_t new_descs_size;
1164	size_t new_infos_size;
1165	unsigned long flags;
1166	char *new_log_buf;
1167	unsigned int free;
1168	u64 seq;
1169
1170	/*
1171	 * Some archs call setup_log_buf() multiple times - first is very
1172	 * early, e.g. from setup_arch(), and second - when percpu_areas
1173	 * are initialised.
1174	 */
1175	if (!early)
1176		set_percpu_data_ready();
1177
1178	if (log_buf != __log_buf)
1179		return;
1180
1181	if (!early && !new_log_buf_len)
1182		log_buf_add_cpu();
1183
1184	if (!new_log_buf_len)
1185		return;
1186
1187	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1188	if (new_descs_count == 0) {
1189		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1190		return;
1191	}
1192
1193	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1194	if (unlikely(!new_log_buf)) {
1195		pr_err("log_buf_len: %lu text bytes not available\n",
1196		       new_log_buf_len);
1197		return;
1198	}
1199
1200	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1201	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1202	if (unlikely(!new_descs)) {
1203		pr_err("log_buf_len: %zu desc bytes not available\n",
1204		       new_descs_size);
1205		goto err_free_log_buf;
1206	}
1207
1208	new_infos_size = new_descs_count * sizeof(struct printk_info);
1209	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1210	if (unlikely(!new_infos)) {
1211		pr_err("log_buf_len: %zu info bytes not available\n",
1212		       new_infos_size);
1213		goto err_free_descs;
1214	}
1215
1216	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1217
1218	prb_init(&printk_rb_dynamic,
1219		 new_log_buf, ilog2(new_log_buf_len),
1220		 new_descs, ilog2(new_descs_count),
1221		 new_infos);
1222
1223	local_irq_save(flags);
1224
1225	log_buf_len = new_log_buf_len;
1226	log_buf = new_log_buf;
1227	new_log_buf_len = 0;
1228
1229	free = __LOG_BUF_LEN;
1230	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1231		text_size = add_to_rb(&printk_rb_dynamic, &r);
1232		if (text_size > free)
1233			free = 0;
1234		else
1235			free -= text_size;
1236	}
1237
1238	prb = &printk_rb_dynamic;
1239
1240	local_irq_restore(flags);
1241
1242	/*
1243	 * Copy any remaining messages that might have appeared from
1244	 * NMI context after copying but before switching to the
1245	 * dynamic buffer.
1246	 */
1247	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1248		text_size = add_to_rb(&printk_rb_dynamic, &r);
1249		if (text_size > free)
1250			free = 0;
1251		else
1252			free -= text_size;
1253	}
1254
1255	if (seq != prb_next_seq(&printk_rb_static)) {
1256		pr_err("dropped %llu messages\n",
1257		       prb_next_seq(&printk_rb_static) - seq);
1258	}
1259
1260	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1261	pr_info("early log buf free: %u(%u%%)\n",
1262		free, (free * 100) / __LOG_BUF_LEN);
1263	return;
1264
1265err_free_descs:
1266	memblock_free(new_descs, new_descs_size);
1267err_free_log_buf:
1268	memblock_free(new_log_buf, new_log_buf_len);
1269}
1270
1271static bool __read_mostly ignore_loglevel;
1272
1273static int __init ignore_loglevel_setup(char *str)
1274{
1275	ignore_loglevel = true;
1276	pr_info("debug: ignoring loglevel setting.\n");
1277
1278	return 0;
1279}
1280
1281early_param("ignore_loglevel", ignore_loglevel_setup);
1282module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1283MODULE_PARM_DESC(ignore_loglevel,
1284		 "ignore loglevel setting (prints all kernel messages to the console)");
1285
1286static bool suppress_message_printing(int level)
1287{
1288	return (level >= console_loglevel && !ignore_loglevel);
1289}
1290
1291#ifdef CONFIG_BOOT_PRINTK_DELAY
1292
1293static int boot_delay; /* msecs delay after each printk during bootup */
1294static unsigned long long loops_per_msec;	/* based on boot_delay */
1295
1296static int __init boot_delay_setup(char *str)
1297{
1298	unsigned long lpj;
1299
1300	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1301	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1302
1303	get_option(&str, &boot_delay);
1304	if (boot_delay > 10 * 1000)
1305		boot_delay = 0;
1306
1307	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1308		"HZ: %d, loops_per_msec: %llu\n",
1309		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1310	return 0;
1311}
1312early_param("boot_delay", boot_delay_setup);
1313
1314static void boot_delay_msec(int level)
1315{
1316	unsigned long long k;
1317	unsigned long timeout;
1318
1319	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1320		|| suppress_message_printing(level)) {
1321		return;
1322	}
1323
1324	k = (unsigned long long)loops_per_msec * boot_delay;
1325
1326	timeout = jiffies + msecs_to_jiffies(boot_delay);
1327	while (k) {
1328		k--;
1329		cpu_relax();
1330		/*
1331		 * use (volatile) jiffies to prevent
1332		 * compiler reduction; loop termination via jiffies
1333		 * is secondary and may or may not happen.
1334		 */
1335		if (time_after(jiffies, timeout))
1336			break;
1337		touch_nmi_watchdog();
1338	}
1339}
1340#else
1341static inline void boot_delay_msec(int level)
1342{
1343}
1344#endif
1345
1346static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1347module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1348
1349static size_t print_syslog(unsigned int level, char *buf)
1350{
1351	return sprintf(buf, "<%u>", level);
1352}
1353
1354static size_t print_time(u64 ts, char *buf)
1355{
1356	unsigned long rem_nsec = do_div(ts, 1000000000);
1357
1358	return sprintf(buf, "[%5lu.%06lu]",
1359		       (unsigned long)ts, rem_nsec / 1000);
1360}
1361
1362#ifdef CONFIG_PRINTK_CALLER
1363static size_t print_caller(u32 id, char *buf)
1364{
1365	char caller[12];
1366
1367	snprintf(caller, sizeof(caller), "%c%u",
1368		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1369	return sprintf(buf, "[%6s]", caller);
1370}
1371#else
1372#define print_caller(id, buf) 0
1373#endif
1374
1375static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1376				bool time, char *buf)
1377{
1378	size_t len = 0;
1379
1380	if (syslog)
1381		len = print_syslog((info->facility << 3) | info->level, buf);
1382
1383	if (time)
1384		len += print_time(info->ts_nsec, buf + len);
1385
1386	len += print_caller(info->caller_id, buf + len);
1387
1388	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1389		buf[len++] = ' ';
1390		buf[len] = '\0';
1391	}
1392
1393	return len;
1394}
1395
1396/*
1397 * Prepare the record for printing. The text is shifted within the given
1398 * buffer to avoid a need for another one. The following operations are
1399 * done:
1400 *
1401 *   - Add prefix for each line.
1402 *   - Drop truncated lines that no longer fit into the buffer.
1403 *   - Add the trailing newline that has been removed in vprintk_store().
1404 *   - Add a string terminator.
1405 *
1406 * Since the produced string is always terminated, the maximum possible
1407 * return value is @r->text_buf_size - 1;
1408 *
1409 * Return: The length of the updated/prepared text, including the added
1410 * prefixes and the newline. The terminator is not counted. The dropped
1411 * line(s) are not counted.
1412 */
1413static size_t record_print_text(struct printk_record *r, bool syslog,
1414				bool time)
1415{
1416	size_t text_len = r->info->text_len;
1417	size_t buf_size = r->text_buf_size;
1418	char *text = r->text_buf;
1419	char prefix[PREFIX_MAX];
1420	bool truncated = false;
1421	size_t prefix_len;
1422	size_t line_len;
1423	size_t len = 0;
1424	char *next;
1425
1426	/*
1427	 * If the message was truncated because the buffer was not large
1428	 * enough, treat the available text as if it were the full text.
1429	 */
1430	if (text_len > buf_size)
1431		text_len = buf_size;
1432
1433	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1434
1435	/*
1436	 * @text_len: bytes of unprocessed text
1437	 * @line_len: bytes of current line _without_ newline
1438	 * @text:     pointer to beginning of current line
1439	 * @len:      number of bytes prepared in r->text_buf
1440	 */
1441	for (;;) {
1442		next = memchr(text, '\n', text_len);
1443		if (next) {
1444			line_len = next - text;
1445		} else {
1446			/* Drop truncated line(s). */
1447			if (truncated)
1448				break;
1449			line_len = text_len;
1450		}
1451
1452		/*
1453		 * Truncate the text if there is not enough space to add the
1454		 * prefix and a trailing newline and a terminator.
1455		 */
1456		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1457			/* Drop even the current line if no space. */
1458			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1459				break;
1460
1461			text_len = buf_size - len - prefix_len - 1 - 1;
1462			truncated = true;
1463		}
1464
1465		memmove(text + prefix_len, text, text_len);
1466		memcpy(text, prefix, prefix_len);
1467
1468		/*
1469		 * Increment the prepared length to include the text and
1470		 * prefix that were just moved+copied. Also increment for the
1471		 * newline at the end of this line. If this is the last line,
1472		 * there is no newline, but it will be added immediately below.
1473		 */
1474		len += prefix_len + line_len + 1;
1475		if (text_len == line_len) {
1476			/*
1477			 * This is the last line. Add the trailing newline
1478			 * removed in vprintk_store().
1479			 */
1480			text[prefix_len + line_len] = '\n';
1481			break;
1482		}
1483
1484		/*
1485		 * Advance beyond the added prefix and the related line with
1486		 * its newline.
1487		 */
1488		text += prefix_len + line_len + 1;
1489
1490		/*
1491		 * The remaining text has only decreased by the line with its
1492		 * newline.
1493		 *
1494		 * Note that @text_len can become zero. It happens when @text
1495		 * ended with a newline (either due to truncation or the
1496		 * original string ending with "\n\n"). The loop is correctly
1497		 * repeated and (if not truncated) an empty line with a prefix
1498		 * will be prepared.
1499		 */
1500		text_len -= line_len + 1;
1501	}
1502
1503	/*
1504	 * If a buffer was provided, it will be terminated. Space for the
1505	 * string terminator is guaranteed to be available. The terminator is
1506	 * not counted in the return value.
1507	 */
1508	if (buf_size > 0)
1509		r->text_buf[len] = 0;
1510
1511	return len;
1512}
1513
1514static size_t get_record_print_text_size(struct printk_info *info,
1515					 unsigned int line_count,
1516					 bool syslog, bool time)
1517{
1518	char prefix[PREFIX_MAX];
1519	size_t prefix_len;
1520
1521	prefix_len = info_print_prefix(info, syslog, time, prefix);
1522
1523	/*
1524	 * Each line will be preceded with a prefix. The intermediate
1525	 * newlines are already within the text, but a final trailing
1526	 * newline will be added.
1527	 */
1528	return ((prefix_len * line_count) + info->text_len + 1);
1529}
1530
1531/*
1532 * Beginning with @start_seq, find the first record where it and all following
1533 * records up to (but not including) @max_seq fit into @size.
1534 *
1535 * @max_seq is simply an upper bound and does not need to exist. If the caller
1536 * does not require an upper bound, -1 can be used for @max_seq.
1537 */
1538static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1539				  bool syslog, bool time)
1540{
1541	struct printk_info info;
1542	unsigned int line_count;
1543	size_t len = 0;
1544	u64 seq;
1545
1546	/* Determine the size of the records up to @max_seq. */
1547	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1548		if (info.seq >= max_seq)
1549			break;
1550		len += get_record_print_text_size(&info, line_count, syslog, time);
1551	}
1552
1553	/*
1554	 * Adjust the upper bound for the next loop to avoid subtracting
1555	 * lengths that were never added.
1556	 */
1557	if (seq < max_seq)
1558		max_seq = seq;
1559
1560	/*
1561	 * Move first record forward until length fits into the buffer. Ignore
1562	 * newest messages that were not counted in the above cycle. Messages
1563	 * might appear and get lost in the meantime. This is a best effort
1564	 * that prevents an infinite loop that could occur with a retry.
1565	 */
1566	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1567		if (len <= size || info.seq >= max_seq)
1568			break;
1569		len -= get_record_print_text_size(&info, line_count, syslog, time);
1570	}
1571
1572	return seq;
1573}
1574
1575/* The caller is responsible for making sure @size is greater than 0. */
1576static int syslog_print(char __user *buf, int size)
1577{
1578	struct printk_info info;
1579	struct printk_record r;
1580	char *text;
1581	int len = 0;
1582	u64 seq;
1583
1584	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1585	if (!text)
1586		return -ENOMEM;
1587
1588	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1589
1590	mutex_lock(&syslog_lock);
1591
1592	/*
1593	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1594	 * change while waiting.
1595	 */
1596	do {
1597		seq = syslog_seq;
1598
1599		mutex_unlock(&syslog_lock);
1600		/*
1601		 * Guarantee this task is visible on the waitqueue before
1602		 * checking the wake condition.
1603		 *
1604		 * The full memory barrier within set_current_state() of
1605		 * prepare_to_wait_event() pairs with the full memory barrier
1606		 * within wq_has_sleeper().
1607		 *
1608		 * This pairs with __wake_up_klogd:A.
1609		 */
1610		len = wait_event_interruptible(log_wait,
1611				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1612		mutex_lock(&syslog_lock);
1613
1614		if (len)
1615			goto out;
1616	} while (syslog_seq != seq);
1617
1618	/*
1619	 * Copy records that fit into the buffer. The above cycle makes sure
1620	 * that the first record is always available.
1621	 */
1622	do {
1623		size_t n;
1624		size_t skip;
1625		int err;
1626
1627		if (!prb_read_valid(prb, syslog_seq, &r))
1628			break;
1629
1630		if (r.info->seq != syslog_seq) {
1631			/* message is gone, move to next valid one */
1632			syslog_seq = r.info->seq;
1633			syslog_partial = 0;
1634		}
1635
1636		/*
1637		 * To keep reading/counting partial line consistent,
1638		 * use printk_time value as of the beginning of a line.
1639		 */
1640		if (!syslog_partial)
1641			syslog_time = printk_time;
1642
1643		skip = syslog_partial;
1644		n = record_print_text(&r, true, syslog_time);
1645		if (n - syslog_partial <= size) {
1646			/* message fits into buffer, move forward */
1647			syslog_seq = r.info->seq + 1;
1648			n -= syslog_partial;
1649			syslog_partial = 0;
1650		} else if (!len){
1651			/* partial read(), remember position */
1652			n = size;
1653			syslog_partial += n;
1654		} else
1655			n = 0;
1656
1657		if (!n)
1658			break;
1659
1660		mutex_unlock(&syslog_lock);
1661		err = copy_to_user(buf, text + skip, n);
1662		mutex_lock(&syslog_lock);
1663
1664		if (err) {
1665			if (!len)
1666				len = -EFAULT;
1667			break;
1668		}
1669
1670		len += n;
1671		size -= n;
1672		buf += n;
1673	} while (size);
1674out:
1675	mutex_unlock(&syslog_lock);
1676	kfree(text);
1677	return len;
1678}
1679
1680static int syslog_print_all(char __user *buf, int size, bool clear)
1681{
1682	struct printk_info info;
1683	struct printk_record r;
1684	char *text;
1685	int len = 0;
1686	u64 seq;
1687	bool time;
1688
1689	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1690	if (!text)
1691		return -ENOMEM;
1692
1693	time = printk_time;
1694	/*
1695	 * Find first record that fits, including all following records,
1696	 * into the user-provided buffer for this dump.
1697	 */
1698	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1699				     size, true, time);
1700
1701	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1702
1703	len = 0;
1704	prb_for_each_record(seq, prb, seq, &r) {
1705		int textlen;
1706
1707		textlen = record_print_text(&r, true, time);
1708
1709		if (len + textlen > size) {
1710			seq--;
1711			break;
1712		}
1713
1714		if (copy_to_user(buf + len, text, textlen))
1715			len = -EFAULT;
1716		else
1717			len += textlen;
1718
1719		if (len < 0)
1720			break;
1721	}
1722
1723	if (clear) {
1724		mutex_lock(&syslog_lock);
1725		latched_seq_write(&clear_seq, seq);
1726		mutex_unlock(&syslog_lock);
1727	}
1728
1729	kfree(text);
1730	return len;
1731}
1732
1733static void syslog_clear(void)
1734{
1735	mutex_lock(&syslog_lock);
1736	latched_seq_write(&clear_seq, prb_next_seq(prb));
1737	mutex_unlock(&syslog_lock);
1738}
1739
1740int do_syslog(int type, char __user *buf, int len, int source)
1741{
1742	struct printk_info info;
1743	bool clear = false;
1744	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1745	int error;
1746
1747	error = check_syslog_permissions(type, source);
1748	if (error)
1749		return error;
1750
1751	switch (type) {
1752	case SYSLOG_ACTION_CLOSE:	/* Close log */
1753		break;
1754	case SYSLOG_ACTION_OPEN:	/* Open log */
1755		break;
1756	case SYSLOG_ACTION_READ:	/* Read from log */
1757		if (!buf || len < 0)
1758			return -EINVAL;
1759		if (!len)
1760			return 0;
1761		if (!access_ok(buf, len))
1762			return -EFAULT;
1763		error = syslog_print(buf, len);
1764		break;
1765	/* Read/clear last kernel messages */
1766	case SYSLOG_ACTION_READ_CLEAR:
1767		clear = true;
1768		fallthrough;
1769	/* Read last kernel messages */
1770	case SYSLOG_ACTION_READ_ALL:
1771		if (!buf || len < 0)
1772			return -EINVAL;
1773		if (!len)
1774			return 0;
1775		if (!access_ok(buf, len))
1776			return -EFAULT;
1777		error = syslog_print_all(buf, len, clear);
1778		break;
1779	/* Clear ring buffer */
1780	case SYSLOG_ACTION_CLEAR:
1781		syslog_clear();
1782		break;
1783	/* Disable logging to console */
1784	case SYSLOG_ACTION_CONSOLE_OFF:
1785		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1786			saved_console_loglevel = console_loglevel;
1787		console_loglevel = minimum_console_loglevel;
1788		break;
1789	/* Enable logging to console */
1790	case SYSLOG_ACTION_CONSOLE_ON:
1791		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1792			console_loglevel = saved_console_loglevel;
1793			saved_console_loglevel = LOGLEVEL_DEFAULT;
1794		}
1795		break;
1796	/* Set level of messages printed to console */
1797	case SYSLOG_ACTION_CONSOLE_LEVEL:
1798		if (len < 1 || len > 8)
1799			return -EINVAL;
1800		if (len < minimum_console_loglevel)
1801			len = minimum_console_loglevel;
1802		console_loglevel = len;
1803		/* Implicitly re-enable logging to console */
1804		saved_console_loglevel = LOGLEVEL_DEFAULT;
1805		break;
1806	/* Number of chars in the log buffer */
1807	case SYSLOG_ACTION_SIZE_UNREAD:
1808		mutex_lock(&syslog_lock);
1809		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1810			/* No unread messages. */
1811			mutex_unlock(&syslog_lock);
1812			return 0;
1813		}
1814		if (info.seq != syslog_seq) {
1815			/* messages are gone, move to first one */
1816			syslog_seq = info.seq;
1817			syslog_partial = 0;
1818		}
1819		if (source == SYSLOG_FROM_PROC) {
1820			/*
1821			 * Short-cut for poll(/"proc/kmsg") which simply checks
1822			 * for pending data, not the size; return the count of
1823			 * records, not the length.
1824			 */
1825			error = prb_next_seq(prb) - syslog_seq;
1826		} else {
1827			bool time = syslog_partial ? syslog_time : printk_time;
1828			unsigned int line_count;
1829			u64 seq;
1830
1831			prb_for_each_info(syslog_seq, prb, seq, &info,
1832					  &line_count) {
1833				error += get_record_print_text_size(&info, line_count,
1834								    true, time);
1835				time = printk_time;
1836			}
1837			error -= syslog_partial;
1838		}
1839		mutex_unlock(&syslog_lock);
1840		break;
1841	/* Size of the log buffer */
1842	case SYSLOG_ACTION_SIZE_BUFFER:
1843		error = log_buf_len;
1844		break;
1845	default:
1846		error = -EINVAL;
1847		break;
1848	}
1849
1850	return error;
1851}
1852
1853SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1854{
1855	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1856}
1857
1858/*
1859 * Special console_lock variants that help to reduce the risk of soft-lockups.
1860 * They allow to pass console_lock to another printk() call using a busy wait.
1861 */
1862
1863#ifdef CONFIG_LOCKDEP
1864static struct lockdep_map console_owner_dep_map = {
1865	.name = "console_owner"
1866};
1867#endif
1868
1869static DEFINE_RAW_SPINLOCK(console_owner_lock);
1870static struct task_struct *console_owner;
1871static bool console_waiter;
1872
1873/**
1874 * console_lock_spinning_enable - mark beginning of code where another
1875 *	thread might safely busy wait
1876 *
1877 * This basically converts console_lock into a spinlock. This marks
1878 * the section where the console_lock owner can not sleep, because
1879 * there may be a waiter spinning (like a spinlock). Also it must be
1880 * ready to hand over the lock at the end of the section.
1881 */
1882static void console_lock_spinning_enable(void)
1883{
 
 
 
 
 
 
 
 
 
 
 
 
1884	raw_spin_lock(&console_owner_lock);
1885	console_owner = current;
1886	raw_spin_unlock(&console_owner_lock);
1887
 
1888	/* The waiter may spin on us after setting console_owner */
1889	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1890}
1891
1892/**
1893 * console_lock_spinning_disable_and_check - mark end of code where another
1894 *	thread was able to busy wait and check if there is a waiter
1895 * @cookie: cookie returned from console_srcu_read_lock()
1896 *
1897 * This is called at the end of the section where spinning is allowed.
1898 * It has two functions. First, it is a signal that it is no longer
1899 * safe to start busy waiting for the lock. Second, it checks if
1900 * there is a busy waiter and passes the lock rights to her.
1901 *
1902 * Important: Callers lose both the console_lock and the SRCU read lock if
1903 *	there was a busy waiter. They must not touch items synchronized by
1904 *	console_lock or SRCU read lock in this case.
1905 *
1906 * Return: 1 if the lock rights were passed, 0 otherwise.
1907 */
1908static int console_lock_spinning_disable_and_check(int cookie)
1909{
1910	int waiter;
1911
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1912	raw_spin_lock(&console_owner_lock);
1913	waiter = READ_ONCE(console_waiter);
1914	console_owner = NULL;
1915	raw_spin_unlock(&console_owner_lock);
1916
1917	if (!waiter) {
1918		spin_release(&console_owner_dep_map, _THIS_IP_);
1919		return 0;
1920	}
1921
1922	/* The waiter is now free to continue */
1923	WRITE_ONCE(console_waiter, false);
1924
1925	spin_release(&console_owner_dep_map, _THIS_IP_);
1926
1927	/*
1928	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1929	 * releasing the console_lock.
1930	 */
1931	console_srcu_read_unlock(cookie);
1932
1933	/*
1934	 * Hand off console_lock to waiter. The waiter will perform
1935	 * the up(). After this, the waiter is the console_lock owner.
1936	 */
1937	mutex_release(&console_lock_dep_map, _THIS_IP_);
1938	return 1;
1939}
1940
1941/**
1942 * console_trylock_spinning - try to get console_lock by busy waiting
1943 *
1944 * This allows to busy wait for the console_lock when the current
1945 * owner is running in specially marked sections. It means that
1946 * the current owner is running and cannot reschedule until it
1947 * is ready to lose the lock.
1948 *
1949 * Return: 1 if we got the lock, 0 othrewise
1950 */
1951static int console_trylock_spinning(void)
1952{
1953	struct task_struct *owner = NULL;
1954	bool waiter;
1955	bool spin = false;
1956	unsigned long flags;
1957
1958	if (console_trylock())
1959		return 1;
1960
1961	/*
1962	 * It's unsafe to spin once a panic has begun. If we are the
1963	 * panic CPU, we may have already halted the owner of the
1964	 * console_sem. If we are not the panic CPU, then we should
1965	 * avoid taking console_sem, so the panic CPU has a better
1966	 * chance of cleanly acquiring it later.
1967	 */
1968	if (panic_in_progress())
1969		return 0;
1970
1971	printk_safe_enter_irqsave(flags);
1972
1973	raw_spin_lock(&console_owner_lock);
1974	owner = READ_ONCE(console_owner);
1975	waiter = READ_ONCE(console_waiter);
1976	if (!waiter && owner && owner != current) {
1977		WRITE_ONCE(console_waiter, true);
1978		spin = true;
1979	}
1980	raw_spin_unlock(&console_owner_lock);
1981
1982	/*
1983	 * If there is an active printk() writing to the
1984	 * consoles, instead of having it write our data too,
1985	 * see if we can offload that load from the active
1986	 * printer, and do some printing ourselves.
1987	 * Go into a spin only if there isn't already a waiter
1988	 * spinning, and there is an active printer, and
1989	 * that active printer isn't us (recursive printk?).
1990	 */
1991	if (!spin) {
1992		printk_safe_exit_irqrestore(flags);
1993		return 0;
1994	}
1995
1996	/* We spin waiting for the owner to release us */
1997	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1998	/* Owner will clear console_waiter on hand off */
1999	while (READ_ONCE(console_waiter))
2000		cpu_relax();
2001	spin_release(&console_owner_dep_map, _THIS_IP_);
2002
2003	printk_safe_exit_irqrestore(flags);
2004	/*
2005	 * The owner passed the console lock to us.
2006	 * Since we did not spin on console lock, annotate
2007	 * this as a trylock. Otherwise lockdep will
2008	 * complain.
2009	 */
2010	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2011
2012	return 1;
2013}
2014
2015/*
2016 * Call the specified console driver, asking it to write out the specified
2017 * text and length. If @dropped_text is non-NULL and any records have been
2018 * dropped, a dropped message will be written out first.
2019 */
2020static void call_console_driver(struct console *con, const char *text, size_t len,
2021				char *dropped_text)
2022{
2023	size_t dropped_len;
2024
2025	if (con->dropped && dropped_text) {
2026		dropped_len = snprintf(dropped_text, DROPPED_TEXT_MAX,
2027				       "** %lu printk messages dropped **\n",
2028				       con->dropped);
2029		con->dropped = 0;
2030		con->write(con, dropped_text, dropped_len);
2031	}
2032
2033	con->write(con, text, len);
2034}
2035
2036/*
2037 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2038 * additional NMI context per CPU is also separately tracked. Until per-CPU
2039 * is available, a separate "early tracking" is performed.
2040 */
2041static DEFINE_PER_CPU(u8, printk_count);
2042static u8 printk_count_early;
2043#ifdef CONFIG_HAVE_NMI
2044static DEFINE_PER_CPU(u8, printk_count_nmi);
2045static u8 printk_count_nmi_early;
2046#endif
2047
2048/*
2049 * Recursion is limited to keep the output sane. printk() should not require
2050 * more than 1 level of recursion (allowing, for example, printk() to trigger
2051 * a WARN), but a higher value is used in case some printk-internal errors
2052 * exist, such as the ringbuffer validation checks failing.
2053 */
2054#define PRINTK_MAX_RECURSION 3
2055
2056/*
2057 * Return a pointer to the dedicated counter for the CPU+context of the
2058 * caller.
2059 */
2060static u8 *__printk_recursion_counter(void)
2061{
2062#ifdef CONFIG_HAVE_NMI
2063	if (in_nmi()) {
2064		if (printk_percpu_data_ready())
2065			return this_cpu_ptr(&printk_count_nmi);
2066		return &printk_count_nmi_early;
2067	}
2068#endif
2069	if (printk_percpu_data_ready())
2070		return this_cpu_ptr(&printk_count);
2071	return &printk_count_early;
2072}
2073
2074/*
2075 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2076 * The caller must check the boolean return value to see if the recursion is
2077 * allowed. On failure, interrupts are not disabled.
2078 *
2079 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2080 * that is passed to printk_exit_irqrestore().
2081 */
2082#define printk_enter_irqsave(recursion_ptr, flags)	\
2083({							\
2084	bool success = true;				\
2085							\
2086	typecheck(u8 *, recursion_ptr);			\
2087	local_irq_save(flags);				\
2088	(recursion_ptr) = __printk_recursion_counter();	\
2089	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2090		local_irq_restore(flags);		\
2091		success = false;			\
2092	} else {					\
2093		(*(recursion_ptr))++;			\
2094	}						\
2095	success;					\
2096})
2097
2098/* Exit recursion tracking, restoring interrupts. */
2099#define printk_exit_irqrestore(recursion_ptr, flags)	\
2100	do {						\
2101		typecheck(u8 *, recursion_ptr);		\
2102		(*(recursion_ptr))--;			\
2103		local_irq_restore(flags);		\
2104	} while (0)
2105
2106int printk_delay_msec __read_mostly;
2107
2108static inline void printk_delay(int level)
2109{
2110	boot_delay_msec(level);
2111
2112	if (unlikely(printk_delay_msec)) {
2113		int m = printk_delay_msec;
2114
2115		while (m--) {
2116			mdelay(1);
2117			touch_nmi_watchdog();
2118		}
2119	}
2120}
2121
2122static inline u32 printk_caller_id(void)
2123{
2124	return in_task() ? task_pid_nr(current) :
2125		0x80000000 + smp_processor_id();
2126}
2127
2128/**
2129 * printk_parse_prefix - Parse level and control flags.
2130 *
2131 * @text:     The terminated text message.
2132 * @level:    A pointer to the current level value, will be updated.
2133 * @flags:    A pointer to the current printk_info flags, will be updated.
2134 *
2135 * @level may be NULL if the caller is not interested in the parsed value.
2136 * Otherwise the variable pointed to by @level must be set to
2137 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2138 *
2139 * @flags may be NULL if the caller is not interested in the parsed value.
2140 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2141 * value.
2142 *
2143 * Return: The length of the parsed level and control flags.
2144 */
2145u16 printk_parse_prefix(const char *text, int *level,
2146			enum printk_info_flags *flags)
2147{
2148	u16 prefix_len = 0;
2149	int kern_level;
2150
2151	while (*text) {
2152		kern_level = printk_get_level(text);
2153		if (!kern_level)
2154			break;
2155
2156		switch (kern_level) {
2157		case '0' ... '7':
2158			if (level && *level == LOGLEVEL_DEFAULT)
2159				*level = kern_level - '0';
2160			break;
2161		case 'c':	/* KERN_CONT */
2162			if (flags)
2163				*flags |= LOG_CONT;
2164		}
2165
2166		prefix_len += 2;
2167		text += 2;
2168	}
2169
2170	return prefix_len;
2171}
2172
2173__printf(5, 0)
2174static u16 printk_sprint(char *text, u16 size, int facility,
2175			 enum printk_info_flags *flags, const char *fmt,
2176			 va_list args)
2177{
2178	u16 text_len;
2179
2180	text_len = vscnprintf(text, size, fmt, args);
2181
2182	/* Mark and strip a trailing newline. */
2183	if (text_len && text[text_len - 1] == '\n') {
2184		text_len--;
2185		*flags |= LOG_NEWLINE;
2186	}
2187
2188	/* Strip log level and control flags. */
2189	if (facility == 0) {
2190		u16 prefix_len;
2191
2192		prefix_len = printk_parse_prefix(text, NULL, NULL);
2193		if (prefix_len) {
2194			text_len -= prefix_len;
2195			memmove(text, text + prefix_len, text_len);
2196		}
2197	}
2198
2199	trace_console_rcuidle(text, text_len);
2200
2201	return text_len;
2202}
2203
2204__printf(4, 0)
2205int vprintk_store(int facility, int level,
2206		  const struct dev_printk_info *dev_info,
2207		  const char *fmt, va_list args)
2208{
2209	struct prb_reserved_entry e;
2210	enum printk_info_flags flags = 0;
2211	struct printk_record r;
2212	unsigned long irqflags;
2213	u16 trunc_msg_len = 0;
2214	char prefix_buf[8];
2215	u8 *recursion_ptr;
2216	u16 reserve_size;
2217	va_list args2;
2218	u32 caller_id;
2219	u16 text_len;
2220	int ret = 0;
2221	u64 ts_nsec;
2222
2223	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2224		return 0;
2225
2226	/*
2227	 * Since the duration of printk() can vary depending on the message
2228	 * and state of the ringbuffer, grab the timestamp now so that it is
2229	 * close to the call of printk(). This provides a more deterministic
2230	 * timestamp with respect to the caller.
2231	 */
2232	ts_nsec = local_clock();
2233
2234	caller_id = printk_caller_id();
2235
2236	/*
2237	 * The sprintf needs to come first since the syslog prefix might be
2238	 * passed in as a parameter. An extra byte must be reserved so that
2239	 * later the vscnprintf() into the reserved buffer has room for the
2240	 * terminating '\0', which is not counted by vsnprintf().
2241	 */
2242	va_copy(args2, args);
2243	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2244	va_end(args2);
2245
2246	if (reserve_size > LOG_LINE_MAX)
2247		reserve_size = LOG_LINE_MAX;
2248
2249	/* Extract log level or control flags. */
2250	if (facility == 0)
2251		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2252
2253	if (level == LOGLEVEL_DEFAULT)
2254		level = default_message_loglevel;
2255
2256	if (dev_info)
2257		flags |= LOG_NEWLINE;
2258
2259	if (flags & LOG_CONT) {
2260		prb_rec_init_wr(&r, reserve_size);
2261		if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2262			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2263						 facility, &flags, fmt, args);
2264			r.info->text_len += text_len;
2265
2266			if (flags & LOG_NEWLINE) {
2267				r.info->flags |= LOG_NEWLINE;
2268				prb_final_commit(&e);
2269			} else {
2270				prb_commit(&e);
2271			}
2272
2273			ret = text_len;
2274			goto out;
2275		}
2276	}
2277
2278	/*
2279	 * Explicitly initialize the record before every prb_reserve() call.
2280	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2281	 * structure when they fail.
2282	 */
2283	prb_rec_init_wr(&r, reserve_size);
2284	if (!prb_reserve(&e, prb, &r)) {
2285		/* truncate the message if it is too long for empty buffer */
2286		truncate_msg(&reserve_size, &trunc_msg_len);
2287
2288		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2289		if (!prb_reserve(&e, prb, &r))
2290			goto out;
2291	}
2292
2293	/* fill message */
2294	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2295	if (trunc_msg_len)
2296		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2297	r.info->text_len = text_len + trunc_msg_len;
2298	r.info->facility = facility;
2299	r.info->level = level & 7;
2300	r.info->flags = flags & 0x1f;
2301	r.info->ts_nsec = ts_nsec;
2302	r.info->caller_id = caller_id;
2303	if (dev_info)
2304		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2305
2306	/* A message without a trailing newline can be continued. */
2307	if (!(flags & LOG_NEWLINE))
2308		prb_commit(&e);
2309	else
2310		prb_final_commit(&e);
2311
2312	ret = text_len + trunc_msg_len;
2313out:
2314	printk_exit_irqrestore(recursion_ptr, irqflags);
2315	return ret;
2316}
2317
2318asmlinkage int vprintk_emit(int facility, int level,
2319			    const struct dev_printk_info *dev_info,
2320			    const char *fmt, va_list args)
2321{
2322	int printed_len;
2323	bool in_sched = false;
2324
2325	/* Suppress unimportant messages after panic happens */
2326	if (unlikely(suppress_printk))
2327		return 0;
2328
2329	if (unlikely(suppress_panic_printk) &&
2330	    atomic_read(&panic_cpu) != raw_smp_processor_id())
 
 
 
 
2331		return 0;
2332
2333	if (level == LOGLEVEL_SCHED) {
2334		level = LOGLEVEL_DEFAULT;
2335		in_sched = true;
2336	}
2337
2338	printk_delay(level);
2339
2340	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2341
2342	/* If called from the scheduler, we can not call up(). */
2343	if (!in_sched) {
2344		/*
2345		 * The caller may be holding system-critical or
2346		 * timing-sensitive locks. Disable preemption during
2347		 * printing of all remaining records to all consoles so that
2348		 * this context can return as soon as possible. Hopefully
2349		 * another printk() caller will take over the printing.
2350		 */
2351		preempt_disable();
2352		/*
2353		 * Try to acquire and then immediately release the console
2354		 * semaphore. The release will print out buffers. With the
2355		 * spinning variant, this context tries to take over the
2356		 * printing from another printing context.
2357		 */
2358		if (console_trylock_spinning())
2359			console_unlock();
2360		preempt_enable();
2361	}
2362
2363	wake_up_klogd();
 
 
 
 
2364	return printed_len;
2365}
2366EXPORT_SYMBOL(vprintk_emit);
2367
2368int vprintk_default(const char *fmt, va_list args)
2369{
2370	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2371}
2372EXPORT_SYMBOL_GPL(vprintk_default);
2373
2374asmlinkage __visible int _printk(const char *fmt, ...)
2375{
2376	va_list args;
2377	int r;
2378
2379	va_start(args, fmt);
2380	r = vprintk(fmt, args);
2381	va_end(args);
2382
2383	return r;
2384}
2385EXPORT_SYMBOL(_printk);
2386
2387static bool pr_flush(int timeout_ms, bool reset_on_progress);
2388static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2389
2390#else /* CONFIG_PRINTK */
2391
2392#define CONSOLE_LOG_MAX		0
2393#define DROPPED_TEXT_MAX	0
2394#define printk_time		false
2395
2396#define prb_read_valid(rb, seq, r)	false
2397#define prb_first_valid_seq(rb)		0
2398#define prb_next_seq(rb)		0
2399
2400static u64 syslog_seq;
2401
2402static size_t record_print_text(const struct printk_record *r,
2403				bool syslog, bool time)
2404{
2405	return 0;
2406}
2407static ssize_t info_print_ext_header(char *buf, size_t size,
2408				     struct printk_info *info)
2409{
2410	return 0;
2411}
2412static ssize_t msg_print_ext_body(char *buf, size_t size,
2413				  char *text, size_t text_len,
2414				  struct dev_printk_info *dev_info) { return 0; }
2415static void console_lock_spinning_enable(void) { }
2416static int console_lock_spinning_disable_and_check(int cookie) { return 0; }
2417static void call_console_driver(struct console *con, const char *text, size_t len,
2418				char *dropped_text)
2419{
2420}
2421static bool suppress_message_printing(int level) { return false; }
2422static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2423static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2424
2425#endif /* CONFIG_PRINTK */
2426
2427#ifdef CONFIG_EARLY_PRINTK
2428struct console *early_console;
2429
2430asmlinkage __visible void early_printk(const char *fmt, ...)
2431{
2432	va_list ap;
2433	char buf[512];
2434	int n;
2435
2436	if (!early_console)
2437		return;
2438
2439	va_start(ap, fmt);
2440	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2441	va_end(ap);
2442
2443	early_console->write(early_console, buf, n);
2444}
2445#endif
2446
2447static void set_user_specified(struct console_cmdline *c, bool user_specified)
2448{
2449	if (!user_specified)
2450		return;
2451
2452	/*
2453	 * @c console was defined by the user on the command line.
2454	 * Do not clear when added twice also by SPCR or the device tree.
2455	 */
2456	c->user_specified = true;
2457	/* At least one console defined by the user on the command line. */
2458	console_set_on_cmdline = 1;
2459}
2460
2461static int __add_preferred_console(char *name, int idx, char *options,
2462				   char *brl_options, bool user_specified)
2463{
2464	struct console_cmdline *c;
2465	int i;
2466
2467	/*
 
 
 
 
 
 
 
 
2468	 *	See if this tty is not yet registered, and
2469	 *	if we have a slot free.
2470	 */
2471	for (i = 0, c = console_cmdline;
2472	     i < MAX_CMDLINECONSOLES && c->name[0];
2473	     i++, c++) {
2474		if (strcmp(c->name, name) == 0 && c->index == idx) {
2475			if (!brl_options)
2476				preferred_console = i;
2477			set_user_specified(c, user_specified);
2478			return 0;
2479		}
2480	}
2481	if (i == MAX_CMDLINECONSOLES)
2482		return -E2BIG;
2483	if (!brl_options)
2484		preferred_console = i;
2485	strscpy(c->name, name, sizeof(c->name));
2486	c->options = options;
2487	set_user_specified(c, user_specified);
2488	braille_set_options(c, brl_options);
2489
2490	c->index = idx;
2491	return 0;
2492}
2493
2494static int __init console_msg_format_setup(char *str)
2495{
2496	if (!strcmp(str, "syslog"))
2497		console_msg_format = MSG_FORMAT_SYSLOG;
2498	if (!strcmp(str, "default"))
2499		console_msg_format = MSG_FORMAT_DEFAULT;
2500	return 1;
2501}
2502__setup("console_msg_format=", console_msg_format_setup);
2503
2504/*
2505 * Set up a console.  Called via do_early_param() in init/main.c
2506 * for each "console=" parameter in the boot command line.
2507 */
2508static int __init console_setup(char *str)
2509{
2510	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2511	char *s, *options, *brl_options = NULL;
2512	int idx;
2513
2514	/*
2515	 * console="" or console=null have been suggested as a way to
2516	 * disable console output. Use ttynull that has been created
2517	 * for exactly this purpose.
2518	 */
2519	if (str[0] == 0 || strcmp(str, "null") == 0) {
2520		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2521		return 1;
2522	}
2523
2524	if (_braille_console_setup(&str, &brl_options))
2525		return 1;
2526
2527	/*
2528	 * Decode str into name, index, options.
2529	 */
2530	if (str[0] >= '0' && str[0] <= '9') {
2531		strcpy(buf, "ttyS");
2532		strncpy(buf + 4, str, sizeof(buf) - 5);
2533	} else {
2534		strncpy(buf, str, sizeof(buf) - 1);
2535	}
2536	buf[sizeof(buf) - 1] = 0;
2537	options = strchr(str, ',');
2538	if (options)
2539		*(options++) = 0;
2540#ifdef __sparc__
2541	if (!strcmp(str, "ttya"))
2542		strcpy(buf, "ttyS0");
2543	if (!strcmp(str, "ttyb"))
2544		strcpy(buf, "ttyS1");
2545#endif
2546	for (s = buf; *s; s++)
2547		if (isdigit(*s) || *s == ',')
2548			break;
2549	idx = simple_strtoul(s, NULL, 10);
2550	*s = 0;
2551
2552	__add_preferred_console(buf, idx, options, brl_options, true);
2553	return 1;
2554}
2555__setup("console=", console_setup);
2556
2557/**
2558 * add_preferred_console - add a device to the list of preferred consoles.
2559 * @name: device name
2560 * @idx: device index
2561 * @options: options for this console
2562 *
2563 * The last preferred console added will be used for kernel messages
2564 * and stdin/out/err for init.  Normally this is used by console_setup
2565 * above to handle user-supplied console arguments; however it can also
2566 * be used by arch-specific code either to override the user or more
2567 * commonly to provide a default console (ie from PROM variables) when
2568 * the user has not supplied one.
2569 */
2570int add_preferred_console(char *name, int idx, char *options)
2571{
2572	return __add_preferred_console(name, idx, options, NULL, false);
2573}
2574
2575bool console_suspend_enabled = true;
2576EXPORT_SYMBOL(console_suspend_enabled);
2577
2578static int __init console_suspend_disable(char *str)
2579{
2580	console_suspend_enabled = false;
2581	return 1;
2582}
2583__setup("no_console_suspend", console_suspend_disable);
2584module_param_named(console_suspend, console_suspend_enabled,
2585		bool, S_IRUGO | S_IWUSR);
2586MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2587	" and hibernate operations");
2588
2589static bool printk_console_no_auto_verbose;
2590
2591void console_verbose(void)
2592{
2593	if (console_loglevel && !printk_console_no_auto_verbose)
2594		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2595}
2596EXPORT_SYMBOL_GPL(console_verbose);
2597
2598module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2599MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2600
2601/**
2602 * suspend_console - suspend the console subsystem
2603 *
2604 * This disables printk() while we go into suspend states
2605 */
2606void suspend_console(void)
2607{
 
 
2608	if (!console_suspend_enabled)
2609		return;
2610	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2611	pr_flush(1000, true);
2612	console_lock();
2613	console_suspended = 1;
2614	up_console_sem();
 
 
 
 
 
 
 
 
 
 
2615}
2616
2617void resume_console(void)
2618{
 
 
2619	if (!console_suspend_enabled)
2620		return;
2621	down_console_sem();
2622	console_suspended = 0;
2623	console_unlock();
 
 
 
 
 
 
 
 
 
 
2624	pr_flush(1000, true);
2625}
2626
2627/**
2628 * console_cpu_notify - print deferred console messages after CPU hotplug
2629 * @cpu: unused
2630 *
2631 * If printk() is called from a CPU that is not online yet, the messages
2632 * will be printed on the console only if there are CON_ANYTIME consoles.
2633 * This function is called when a new CPU comes online (or fails to come
2634 * up) or goes offline.
2635 */
2636static int console_cpu_notify(unsigned int cpu)
2637{
2638	if (!cpuhp_tasks_frozen) {
2639		/* If trylock fails, someone else is doing the printing */
2640		if (console_trylock())
2641			console_unlock();
2642	}
2643	return 0;
2644}
2645
2646/**
2647 * console_lock - block the console subsystem from printing
2648 *
2649 * Acquires a lock which guarantees that no consoles will
2650 * be in or enter their write() callback.
2651 *
2652 * Can sleep, returns nothing.
2653 */
2654void console_lock(void)
2655{
2656	might_sleep();
2657
 
 
 
 
2658	down_console_sem();
2659	if (console_suspended)
2660		return;
2661	console_locked = 1;
2662	console_may_schedule = 1;
2663}
2664EXPORT_SYMBOL(console_lock);
2665
2666/**
2667 * console_trylock - try to block the console subsystem from printing
2668 *
2669 * Try to acquire a lock which guarantees that no consoles will
2670 * be in or enter their write() callback.
2671 *
2672 * returns 1 on success, and 0 on failure to acquire the lock.
2673 */
2674int console_trylock(void)
2675{
2676	if (down_trylock_console_sem())
 
2677		return 0;
2678	if (console_suspended) {
2679		up_console_sem();
2680		return 0;
2681	}
2682	console_locked = 1;
2683	console_may_schedule = 0;
2684	return 1;
2685}
2686EXPORT_SYMBOL(console_trylock);
2687
2688int is_console_locked(void)
2689{
2690	return console_locked;
2691}
2692EXPORT_SYMBOL(is_console_locked);
2693
2694/*
2695 * Return true when this CPU should unlock console_sem without pushing all
2696 * messages to the console. This reduces the chance that the console is
2697 * locked when the panic CPU tries to use it.
2698 */
2699static bool abandon_console_lock_in_panic(void)
2700{
2701	if (!panic_in_progress())
2702		return false;
2703
2704	/*
2705	 * We can use raw_smp_processor_id() here because it is impossible for
2706	 * the task to be migrated to the panic_cpu, or away from it. If
2707	 * panic_cpu has already been set, and we're not currently executing on
2708	 * that CPU, then we never will be.
2709	 */
2710	return atomic_read(&panic_cpu) != raw_smp_processor_id();
2711}
2712
2713/*
2714 * Check if the given console is currently capable and allowed to print
2715 * records.
2716 *
2717 * Requires the console_srcu_read_lock.
2718 */
2719static inline bool console_is_usable(struct console *con)
2720{
2721	short flags = console_srcu_read_flags(con);
2722
2723	if (!(flags & CON_ENABLED))
2724		return false;
2725
 
 
 
2726	if (!con->write)
2727		return false;
2728
2729	/*
2730	 * Console drivers may assume that per-cpu resources have been
2731	 * allocated. So unless they're explicitly marked as being able to
2732	 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2733	 */
2734	if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2735		return false;
2736
2737	return true;
2738}
2739
2740static void __console_unlock(void)
2741{
2742	console_locked = 0;
2743	up_console_sem();
2744}
2745
 
 
2746/*
2747 * Print one record for the given console. The record printed is whatever
2748 * record is the next available record for the given console.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2749 *
2750 * @text is a buffer of size CONSOLE_LOG_MAX.
 
2751 *
2752 * If extended messages should be printed, @ext_text is a buffer of size
2753 * CONSOLE_EXT_LOG_MAX. Otherwise @ext_text must be NULL.
2754 *
2755 * If dropped messages should be printed, @dropped_text is a buffer of size
2756 * DROPPED_TEXT_MAX. Otherwise @dropped_text must be NULL.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2757 *
2758 * @handover will be set to true if a printk waiter has taken over the
2759 * console_lock, in which case the caller is no longer holding both the
2760 * console_lock and the SRCU read lock. Otherwise it is set to false.
2761 *
2762 * @cookie is the cookie from the SRCU read lock.
2763 *
2764 * Returns false if the given console has no next record to print, otherwise
2765 * true.
2766 *
2767 * Requires the console_lock and the SRCU read lock.
2768 */
2769static bool console_emit_next_record(struct console *con, char *text, char *ext_text,
2770				     char *dropped_text, bool *handover, int cookie)
2771{
2772	static int panic_console_dropped;
2773	struct printk_info info;
2774	struct printk_record r;
 
 
2775	unsigned long flags;
2776	char *write_text;
2777	size_t len;
2778
2779	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
2780
2781	*handover = false;
2782
2783	if (!prb_read_valid(prb, con->seq, &r))
2784		return false;
2785
2786	if (con->seq != r.info->seq) {
2787		con->dropped += r.info->seq - con->seq;
2788		con->seq = r.info->seq;
2789		if (panic_in_progress() && panic_console_dropped++ > 10) {
2790			suppress_panic_printk = 1;
2791			pr_warn_once("Too many dropped messages. Suppress messages on non-panic CPUs to prevent livelock.\n");
2792		}
2793	}
2794
2795	/* Skip record that has level above the console loglevel. */
2796	if (suppress_message_printing(r.info->level)) {
2797		con->seq++;
2798		goto skip;
2799	}
2800
2801	if (ext_text) {
2802		write_text = ext_text;
2803		len = info_print_ext_header(ext_text, CONSOLE_EXT_LOG_MAX, r.info);
2804		len += msg_print_ext_body(ext_text + len, CONSOLE_EXT_LOG_MAX - len,
2805					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2806	} else {
2807		write_text = text;
2808		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2809	}
2810
2811	/*
2812	 * While actively printing out messages, if another printk()
2813	 * were to occur on another CPU, it may wait for this one to
2814	 * finish. This task can not be preempted if there is a
2815	 * waiter waiting to take over.
2816	 *
2817	 * Interrupts are disabled because the hand over to a waiter
2818	 * must not be interrupted until the hand over is completed
2819	 * (@console_waiter is cleared).
2820	 */
2821	printk_safe_enter_irqsave(flags);
2822	console_lock_spinning_enable();
2823
2824	stop_critical_timings();	/* don't trace print latency */
2825	call_console_driver(con, write_text, len, dropped_text);
 
 
 
 
2826	start_critical_timings();
2827
2828	con->seq++;
2829
2830	*handover = console_lock_spinning_disable_and_check(cookie);
2831	printk_safe_exit_irqrestore(flags);
2832skip:
2833	return true;
2834}
2835
 
 
 
 
 
 
 
 
 
 
2836/*
2837 * Print out all remaining records to all consoles.
2838 *
2839 * @do_cond_resched is set by the caller. It can be true only in schedulable
2840 * context.
2841 *
2842 * @next_seq is set to the sequence number after the last available record.
2843 * The value is valid only when this function returns true. It means that all
2844 * usable consoles are completely flushed.
2845 *
2846 * @handover will be set to true if a printk waiter has taken over the
2847 * console_lock, in which case the caller is no longer holding the
2848 * console_lock. Otherwise it is set to false.
2849 *
2850 * Returns true when there was at least one usable console and all messages
2851 * were flushed to all usable consoles. A returned false informs the caller
2852 * that everything was not flushed (either there were no usable consoles or
2853 * another context has taken over printing or it is a panic situation and this
2854 * is not the panic CPU). Regardless the reason, the caller should assume it
2855 * is not useful to immediately try again.
2856 *
2857 * Requires the console_lock.
2858 */
2859static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2860{
2861	static char dropped_text[DROPPED_TEXT_MAX];
2862	static char ext_text[CONSOLE_EXT_LOG_MAX];
2863	static char text[CONSOLE_LOG_MAX];
2864	bool any_usable = false;
2865	struct console *con;
2866	bool any_progress;
2867	int cookie;
2868
2869	*next_seq = 0;
2870	*handover = false;
2871
2872	do {
2873		any_progress = false;
2874
2875		cookie = console_srcu_read_lock();
2876		for_each_console_srcu(con) {
2877			bool progress;
2878
2879			if (!console_is_usable(con))
2880				continue;
2881			any_usable = true;
2882
2883			if (console_srcu_read_flags(con) & CON_EXTENDED) {
2884				/* Extended consoles do not print "dropped messages". */
2885				progress = console_emit_next_record(con, &text[0],
2886								    &ext_text[0], NULL,
2887								    handover, cookie);
2888			} else {
2889				progress = console_emit_next_record(con, &text[0],
2890								    NULL, &dropped_text[0],
2891								    handover, cookie);
2892			}
2893
2894			/*
2895			 * If a handover has occurred, the SRCU read lock
2896			 * is already released.
2897			 */
2898			if (*handover)
2899				return false;
2900
2901			/* Track the next of the highest seq flushed. */
2902			if (con->seq > *next_seq)
2903				*next_seq = con->seq;
2904
2905			if (!progress)
2906				continue;
2907			any_progress = true;
2908
2909			/* Allow panic_cpu to take over the consoles safely. */
2910			if (abandon_console_lock_in_panic())
2911				goto abandon;
2912
2913			if (do_cond_resched)
2914				cond_resched();
2915		}
2916		console_srcu_read_unlock(cookie);
2917	} while (any_progress);
2918
2919	return any_usable;
2920
2921abandon:
2922	console_srcu_read_unlock(cookie);
2923	return false;
2924}
2925
2926/**
2927 * console_unlock - unblock the console subsystem from printing
2928 *
2929 * Releases the console_lock which the caller holds to block printing of
2930 * the console subsystem.
2931 *
2932 * While the console_lock was held, console output may have been buffered
2933 * by printk().  If this is the case, console_unlock(); emits
2934 * the output prior to releasing the lock.
2935 *
2936 * console_unlock(); may be called from any context.
2937 */
2938void console_unlock(void)
2939{
2940	bool do_cond_resched;
2941	bool handover;
2942	bool flushed;
2943	u64 next_seq;
2944
2945	if (console_suspended) {
2946		up_console_sem();
2947		return;
2948	}
2949
2950	/*
2951	 * Console drivers are called with interrupts disabled, so
2952	 * @console_may_schedule should be cleared before; however, we may
2953	 * end up dumping a lot of lines, for example, if called from
2954	 * console registration path, and should invoke cond_resched()
2955	 * between lines if allowable.  Not doing so can cause a very long
2956	 * scheduling stall on a slow console leading to RCU stall and
2957	 * softlockup warnings which exacerbate the issue with more
2958	 * messages practically incapacitating the system. Therefore, create
2959	 * a local to use for the printing loop.
2960	 */
2961	do_cond_resched = console_may_schedule;
2962
2963	do {
2964		console_may_schedule = 0;
2965
2966		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
2967		if (!handover)
2968			__console_unlock();
2969
2970		/*
2971		 * Abort if there was a failure to flush all messages to all
2972		 * usable consoles. Either it is not possible to flush (in
2973		 * which case it would be an infinite loop of retrying) or
2974		 * another context has taken over printing.
2975		 */
2976		if (!flushed)
2977			break;
2978
2979		/*
2980		 * Some context may have added new records after
2981		 * console_flush_all() but before unlocking the console.
2982		 * Re-check if there is a new record to flush. If the trylock
2983		 * fails, another context is already handling the printing.
2984		 */
2985	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
2986}
2987EXPORT_SYMBOL(console_unlock);
2988
2989/**
2990 * console_conditional_schedule - yield the CPU if required
2991 *
2992 * If the console code is currently allowed to sleep, and
2993 * if this CPU should yield the CPU to another task, do
2994 * so here.
2995 *
2996 * Must be called within console_lock();.
2997 */
2998void __sched console_conditional_schedule(void)
2999{
3000	if (console_may_schedule)
3001		cond_resched();
3002}
3003EXPORT_SYMBOL(console_conditional_schedule);
3004
3005void console_unblank(void)
3006{
 
3007	struct console *c;
3008	int cookie;
3009
3010	/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3011	 * Stop console printing because the unblank() callback may
3012	 * assume the console is not within its write() callback.
3013	 *
3014	 * If @oops_in_progress is set, this may be an atomic context.
3015	 * In that case, attempt a trylock as best-effort.
3016	 */
3017	if (oops_in_progress) {
 
 
 
 
 
 
 
 
 
 
3018		if (down_trylock_console_sem() != 0)
3019			return;
3020	} else
3021		console_lock();
3022
3023	console_locked = 1;
3024	console_may_schedule = 0;
3025
3026	cookie = console_srcu_read_lock();
3027	for_each_console_srcu(c) {
3028		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3029			c->unblank();
3030	}
3031	console_srcu_read_unlock(cookie);
3032
3033	console_unlock();
3034
3035	if (!oops_in_progress)
3036		pr_flush(1000, true);
3037}
3038
3039/**
3040 * console_flush_on_panic - flush console content on panic
3041 * @mode: flush all messages in buffer or just the pending ones
3042 *
3043 * Immediately output all pending messages no matter what.
3044 */
3045void console_flush_on_panic(enum con_flush_mode mode)
3046{
 
 
 
3047	/*
3048	 * If someone else is holding the console lock, trylock will fail
3049	 * and may_schedule may be set.  Ignore and proceed to unlock so
3050	 * that messages are flushed out.  As this can be called from any
3051	 * context and we don't want to get preempted while flushing,
3052	 * ensure may_schedule is cleared.
 
 
 
 
 
 
 
 
3053	 */
3054	console_trylock();
3055	console_may_schedule = 0;
3056
3057	if (mode == CONSOLE_REPLAY_ALL) {
3058		struct console *c;
 
3059		int cookie;
3060		u64 seq;
3061
3062		seq = prb_first_valid_seq(prb);
3063
3064		cookie = console_srcu_read_lock();
3065		for_each_console_srcu(c) {
3066			/*
3067			 * If the above console_trylock() failed, this is an
3068			 * unsynchronized assignment. But in that case, the
3069			 * kernel is in "hope and pray" mode anyway.
3070			 */
3071			c->seq = seq;
 
 
 
 
 
3072		}
3073		console_srcu_read_unlock(cookie);
3074	}
3075	console_unlock();
 
3076}
3077
3078/*
3079 * Return the console tty driver structure and its associated index
3080 */
3081struct tty_driver *console_device(int *index)
3082{
3083	struct console *c;
3084	struct tty_driver *driver = NULL;
3085	int cookie;
3086
3087	/*
3088	 * Take console_lock to serialize device() callback with
3089	 * other console operations. For example, fg_console is
3090	 * modified under console_lock when switching vt.
3091	 */
3092	console_lock();
3093
3094	cookie = console_srcu_read_lock();
3095	for_each_console_srcu(c) {
3096		if (!c->device)
3097			continue;
3098		driver = c->device(c, index);
3099		if (driver)
3100			break;
3101	}
3102	console_srcu_read_unlock(cookie);
3103
3104	console_unlock();
3105	return driver;
3106}
3107
3108/*
3109 * Prevent further output on the passed console device so that (for example)
3110 * serial drivers can disable console output before suspending a port, and can
3111 * re-enable output afterwards.
3112 */
3113void console_stop(struct console *console)
3114{
3115	__pr_flush(console, 1000, true);
3116	console_list_lock();
3117	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3118	console_list_unlock();
3119
3120	/*
3121	 * Ensure that all SRCU list walks have completed. All contexts must
3122	 * be able to see that this console is disabled so that (for example)
3123	 * the caller can suspend the port without risk of another context
3124	 * using the port.
3125	 */
3126	synchronize_srcu(&console_srcu);
3127}
3128EXPORT_SYMBOL(console_stop);
3129
3130void console_start(struct console *console)
3131{
3132	console_list_lock();
3133	console_srcu_write_flags(console, console->flags | CON_ENABLED);
3134	console_list_unlock();
3135	__pr_flush(console, 1000, true);
3136}
3137EXPORT_SYMBOL(console_start);
3138
3139static int __read_mostly keep_bootcon;
3140
3141static int __init keep_bootcon_setup(char *str)
3142{
3143	keep_bootcon = 1;
3144	pr_info("debug: skip boot console de-registration.\n");
3145
3146	return 0;
3147}
3148
3149early_param("keep_bootcon", keep_bootcon_setup);
3150
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3151/*
3152 * This is called by register_console() to try to match
3153 * the newly registered console with any of the ones selected
3154 * by either the command line or add_preferred_console() and
3155 * setup/enable it.
3156 *
3157 * Care need to be taken with consoles that are statically
3158 * enabled such as netconsole
3159 */
3160static int try_enable_preferred_console(struct console *newcon,
3161					bool user_specified)
3162{
3163	struct console_cmdline *c;
3164	int i, err;
3165
3166	for (i = 0, c = console_cmdline;
3167	     i < MAX_CMDLINECONSOLES && c->name[0];
3168	     i++, c++) {
3169		if (c->user_specified != user_specified)
3170			continue;
3171		if (!newcon->match ||
3172		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3173			/* default matching */
3174			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3175			if (strcmp(c->name, newcon->name) != 0)
3176				continue;
3177			if (newcon->index >= 0 &&
3178			    newcon->index != c->index)
3179				continue;
3180			if (newcon->index < 0)
3181				newcon->index = c->index;
3182
3183			if (_braille_register_console(newcon, c))
3184				return 0;
3185
3186			if (newcon->setup &&
3187			    (err = newcon->setup(newcon, c->options)) != 0)
3188				return err;
3189		}
3190		newcon->flags |= CON_ENABLED;
3191		if (i == preferred_console)
3192			newcon->flags |= CON_CONSDEV;
3193		return 0;
3194	}
3195
3196	/*
3197	 * Some consoles, such as pstore and netconsole, can be enabled even
3198	 * without matching. Accept the pre-enabled consoles only when match()
3199	 * and setup() had a chance to be called.
3200	 */
3201	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3202		return 0;
3203
3204	return -ENOENT;
3205}
3206
3207/* Try to enable the console unconditionally */
3208static void try_enable_default_console(struct console *newcon)
3209{
3210	if (newcon->index < 0)
3211		newcon->index = 0;
3212
3213	if (newcon->setup && newcon->setup(newcon, NULL) != 0)
3214		return;
3215
3216	newcon->flags |= CON_ENABLED;
3217
3218	if (newcon->device)
3219		newcon->flags |= CON_CONSDEV;
3220}
3221
3222#define con_printk(lvl, con, fmt, ...)			\
3223	printk(lvl pr_fmt("%sconsole [%s%d] " fmt),	\
3224	       (con->flags & CON_BOOT) ? "boot" : "",	\
3225	       con->name, con->index, ##__VA_ARGS__)
3226
3227static void console_init_seq(struct console *newcon, bool bootcon_registered)
3228{
3229	struct console *con;
3230	bool handover;
3231
3232	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3233		/* Get a consistent copy of @syslog_seq. */
3234		mutex_lock(&syslog_lock);
3235		newcon->seq = syslog_seq;
3236		mutex_unlock(&syslog_lock);
3237	} else {
3238		/* Begin with next message added to ringbuffer. */
3239		newcon->seq = prb_next_seq(prb);
3240
3241		/*
3242		 * If any enabled boot consoles are due to be unregistered
3243		 * shortly, some may not be caught up and may be the same
3244		 * device as @newcon. Since it is not known which boot console
3245		 * is the same device, flush all consoles and, if necessary,
3246		 * start with the message of the enabled boot console that is
3247		 * the furthest behind.
3248		 */
3249		if (bootcon_registered && !keep_bootcon) {
3250			/*
3251			 * Hold the console_lock to stop console printing and
3252			 * guarantee safe access to console->seq.
3253			 */
3254			console_lock();
3255
3256			/*
3257			 * Flush all consoles and set the console to start at
3258			 * the next unprinted sequence number.
3259			 */
3260			if (!console_flush_all(true, &newcon->seq, &handover)) {
3261				/*
3262				 * Flushing failed. Just choose the lowest
3263				 * sequence of the enabled boot consoles.
3264				 */
3265
3266				/*
3267				 * If there was a handover, this context no
3268				 * longer holds the console_lock.
3269				 */
3270				if (handover)
3271					console_lock();
3272
3273				newcon->seq = prb_next_seq(prb);
3274				for_each_console(con) {
3275					if ((con->flags & CON_BOOT) &&
3276					    (con->flags & CON_ENABLED) &&
3277					    con->seq < newcon->seq) {
3278						newcon->seq = con->seq;
3279					}
3280				}
3281			}
3282
3283			console_unlock();
3284		}
3285	}
3286}
3287
3288#define console_first()				\
3289	hlist_entry(console_list.first, struct console, node)
3290
3291static int unregister_console_locked(struct console *console);
3292
3293/*
3294 * The console driver calls this routine during kernel initialization
3295 * to register the console printing procedure with printk() and to
3296 * print any messages that were printed by the kernel before the
3297 * console driver was initialized.
3298 *
3299 * This can happen pretty early during the boot process (because of
3300 * early_printk) - sometimes before setup_arch() completes - be careful
3301 * of what kernel features are used - they may not be initialised yet.
3302 *
3303 * There are two types of consoles - bootconsoles (early_printk) and
3304 * "real" consoles (everything which is not a bootconsole) which are
3305 * handled differently.
3306 *  - Any number of bootconsoles can be registered at any time.
3307 *  - As soon as a "real" console is registered, all bootconsoles
3308 *    will be unregistered automatically.
3309 *  - Once a "real" console is registered, any attempt to register a
3310 *    bootconsoles will be rejected
3311 */
3312void register_console(struct console *newcon)
3313{
3314	struct console *con;
3315	bool bootcon_registered = false;
3316	bool realcon_registered = false;
3317	int err;
3318
3319	console_list_lock();
3320
3321	for_each_console(con) {
3322		if (WARN(con == newcon, "console '%s%d' already registered\n",
3323					 con->name, con->index)) {
3324			goto unlock;
3325		}
3326
3327		if (con->flags & CON_BOOT)
3328			bootcon_registered = true;
3329		else
3330			realcon_registered = true;
3331	}
3332
3333	/* Do not register boot consoles when there already is a real one. */
3334	if ((newcon->flags & CON_BOOT) && realcon_registered) {
3335		pr_info("Too late to register bootconsole %s%d\n",
3336			newcon->name, newcon->index);
3337		goto unlock;
3338	}
3339
 
 
 
 
 
 
 
 
 
3340	/*
3341	 * See if we want to enable this console driver by default.
3342	 *
3343	 * Nope when a console is preferred by the command line, device
3344	 * tree, or SPCR.
3345	 *
3346	 * The first real console with tty binding (driver) wins. More
3347	 * consoles might get enabled before the right one is found.
3348	 *
3349	 * Note that a console with tty binding will have CON_CONSDEV
3350	 * flag set and will be first in the list.
3351	 */
3352	if (preferred_console < 0) {
3353		if (hlist_empty(&console_list) || !console_first()->device ||
3354		    console_first()->flags & CON_BOOT) {
3355			try_enable_default_console(newcon);
3356		}
3357	}
3358
3359	/* See if this console matches one we selected on the command line */
3360	err = try_enable_preferred_console(newcon, true);
3361
3362	/* If not, try to match against the platform default(s) */
3363	if (err == -ENOENT)
3364		err = try_enable_preferred_console(newcon, false);
3365
3366	/* printk() messages are not printed to the Braille console. */
3367	if (err || newcon->flags & CON_BRL)
 
 
3368		goto unlock;
 
3369
3370	/*
3371	 * If we have a bootconsole, and are switching to a real console,
3372	 * don't print everything out again, since when the boot console, and
3373	 * the real console are the same physical device, it's annoying to
3374	 * see the beginning boot messages twice
3375	 */
3376	if (bootcon_registered &&
3377	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3378		newcon->flags &= ~CON_PRINTBUFFER;
3379	}
3380
3381	newcon->dropped = 0;
3382	console_init_seq(newcon, bootcon_registered);
3383
 
 
 
3384	/*
3385	 * Put this console in the list - keep the
3386	 * preferred driver at the head of the list.
3387	 */
3388	if (hlist_empty(&console_list)) {
3389		/* Ensure CON_CONSDEV is always set for the head. */
3390		newcon->flags |= CON_CONSDEV;
3391		hlist_add_head_rcu(&newcon->node, &console_list);
3392
3393	} else if (newcon->flags & CON_CONSDEV) {
3394		/* Only the new head can have CON_CONSDEV set. */
3395		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3396		hlist_add_head_rcu(&newcon->node, &console_list);
3397
3398	} else {
3399		hlist_add_behind_rcu(&newcon->node, console_list.first);
3400	}
3401
3402	/*
3403	 * No need to synchronize SRCU here! The caller does not rely
3404	 * on all contexts being able to see the new console before
3405	 * register_console() completes.
3406	 */
3407
3408	console_sysfs_notify();
3409
3410	/*
3411	 * By unregistering the bootconsoles after we enable the real console
3412	 * we get the "console xxx enabled" message on all the consoles -
3413	 * boot consoles, real consoles, etc - this is to ensure that end
3414	 * users know there might be something in the kernel's log buffer that
3415	 * went to the bootconsole (that they do not see on the real console)
3416	 */
3417	con_printk(KERN_INFO, newcon, "enabled\n");
3418	if (bootcon_registered &&
3419	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3420	    !keep_bootcon) {
3421		struct hlist_node *tmp;
3422
3423		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3424			if (con->flags & CON_BOOT)
3425				unregister_console_locked(con);
3426		}
3427	}
3428unlock:
3429	console_list_unlock();
3430}
3431EXPORT_SYMBOL(register_console);
3432
3433/* Must be called under console_list_lock(). */
3434static int unregister_console_locked(struct console *console)
3435{
3436	int res;
3437
3438	lockdep_assert_console_list_lock_held();
3439
3440	con_printk(KERN_INFO, console, "disabled\n");
3441
3442	res = _braille_unregister_console(console);
3443	if (res < 0)
3444		return res;
3445	if (res > 0)
3446		return 0;
3447
3448	/* Disable it unconditionally */
3449	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3450
3451	if (!console_is_registered_locked(console))
3452		return -ENODEV;
3453
3454	hlist_del_init_rcu(&console->node);
3455
3456	/*
3457	 * <HISTORICAL>
3458	 * If this isn't the last console and it has CON_CONSDEV set, we
3459	 * need to set it on the next preferred console.
3460	 * </HISTORICAL>
3461	 *
3462	 * The above makes no sense as there is no guarantee that the next
3463	 * console has any device attached. Oh well....
3464	 */
3465	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3466		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3467
3468	/*
3469	 * Ensure that all SRCU list walks have completed. All contexts
3470	 * must not be able to see this console in the list so that any
3471	 * exit/cleanup routines can be performed safely.
3472	 */
3473	synchronize_srcu(&console_srcu);
3474
 
 
 
3475	console_sysfs_notify();
3476
3477	if (console->exit)
3478		res = console->exit(console);
3479
3480	return res;
3481}
3482
3483int unregister_console(struct console *console)
3484{
3485	int res;
3486
3487	console_list_lock();
3488	res = unregister_console_locked(console);
3489	console_list_unlock();
3490	return res;
3491}
3492EXPORT_SYMBOL(unregister_console);
3493
3494/**
3495 * console_force_preferred_locked - force a registered console preferred
3496 * @con: The registered console to force preferred.
3497 *
3498 * Must be called under console_list_lock().
3499 */
3500void console_force_preferred_locked(struct console *con)
3501{
3502	struct console *cur_pref_con;
3503
3504	if (!console_is_registered_locked(con))
3505		return;
3506
3507	cur_pref_con = console_first();
3508
3509	/* Already preferred? */
3510	if (cur_pref_con == con)
3511		return;
3512
3513	/*
3514	 * Delete, but do not re-initialize the entry. This allows the console
3515	 * to continue to appear registered (via any hlist_unhashed_lockless()
3516	 * checks), even though it was briefly removed from the console list.
3517	 */
3518	hlist_del_rcu(&con->node);
3519
3520	/*
3521	 * Ensure that all SRCU list walks have completed so that the console
3522	 * can be added to the beginning of the console list and its forward
3523	 * list pointer can be re-initialized.
3524	 */
3525	synchronize_srcu(&console_srcu);
3526
3527	con->flags |= CON_CONSDEV;
3528	WARN_ON(!con->device);
3529
3530	/* Only the new head can have CON_CONSDEV set. */
3531	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3532	hlist_add_head_rcu(&con->node, &console_list);
3533}
3534EXPORT_SYMBOL(console_force_preferred_locked);
3535
3536/*
3537 * Initialize the console device. This is called *early*, so
3538 * we can't necessarily depend on lots of kernel help here.
3539 * Just do some early initializations, and do the complex setup
3540 * later.
3541 */
3542void __init console_init(void)
3543{
3544	int ret;
3545	initcall_t call;
3546	initcall_entry_t *ce;
3547
3548	/* Setup the default TTY line discipline. */
3549	n_tty_init();
3550
3551	/*
3552	 * set up the console device so that later boot sequences can
3553	 * inform about problems etc..
3554	 */
3555	ce = __con_initcall_start;
3556	trace_initcall_level("console");
3557	while (ce < __con_initcall_end) {
3558		call = initcall_from_entry(ce);
3559		trace_initcall_start(call);
3560		ret = call();
3561		trace_initcall_finish(call, ret);
3562		ce++;
3563	}
3564}
3565
3566/*
3567 * Some boot consoles access data that is in the init section and which will
3568 * be discarded after the initcalls have been run. To make sure that no code
3569 * will access this data, unregister the boot consoles in a late initcall.
3570 *
3571 * If for some reason, such as deferred probe or the driver being a loadable
3572 * module, the real console hasn't registered yet at this point, there will
3573 * be a brief interval in which no messages are logged to the console, which
3574 * makes it difficult to diagnose problems that occur during this time.
3575 *
3576 * To mitigate this problem somewhat, only unregister consoles whose memory
3577 * intersects with the init section. Note that all other boot consoles will
3578 * get unregistered when the real preferred console is registered.
3579 */
3580static int __init printk_late_init(void)
3581{
3582	struct hlist_node *tmp;
3583	struct console *con;
3584	int ret;
3585
3586	console_list_lock();
3587	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3588		if (!(con->flags & CON_BOOT))
3589			continue;
3590
3591		/* Check addresses that might be used for enabled consoles. */
3592		if (init_section_intersects(con, sizeof(*con)) ||
3593		    init_section_contains(con->write, 0) ||
3594		    init_section_contains(con->read, 0) ||
3595		    init_section_contains(con->device, 0) ||
3596		    init_section_contains(con->unblank, 0) ||
3597		    init_section_contains(con->data, 0)) {
3598			/*
3599			 * Please, consider moving the reported consoles out
3600			 * of the init section.
3601			 */
3602			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3603				con->name, con->index);
3604			unregister_console_locked(con);
3605		}
3606	}
3607	console_list_unlock();
3608
3609	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3610					console_cpu_notify);
3611	WARN_ON(ret < 0);
3612	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3613					console_cpu_notify, NULL);
3614	WARN_ON(ret < 0);
3615	printk_sysctl_init();
3616	return 0;
3617}
3618late_initcall(printk_late_init);
3619
3620#if defined CONFIG_PRINTK
3621/* If @con is specified, only wait for that console. Otherwise wait for all. */
3622static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3623{
3624	int remaining = timeout_ms;
 
3625	struct console *c;
3626	u64 last_diff = 0;
3627	u64 printk_seq;
 
3628	int cookie;
3629	u64 diff;
3630	u64 seq;
3631
3632	might_sleep();
3633
3634	seq = prb_next_seq(prb);
 
 
 
 
3635
3636	for (;;) {
 
 
 
3637		diff = 0;
3638
3639		/*
3640		 * Hold the console_lock to guarantee safe access to
3641		 * console->seq and to prevent changes to @console_suspended
3642		 * until all consoles have been processed.
 
3643		 */
3644		console_lock();
3645
3646		cookie = console_srcu_read_lock();
3647		for_each_console_srcu(c) {
3648			if (con && con != c)
3649				continue;
 
 
 
 
 
 
 
 
3650			if (!console_is_usable(c))
3651				continue;
3652			printk_seq = c->seq;
 
 
 
 
 
 
3653			if (printk_seq < seq)
3654				diff += seq - printk_seq;
3655		}
3656		console_srcu_read_unlock(cookie);
3657
3658		/*
3659		 * If consoles are suspended, it cannot be expected that they
3660		 * make forward progress, so timeout immediately. @diff is
3661		 * still used to return a valid flush status.
3662		 */
3663		if (console_suspended)
3664			remaining = 0;
3665		else if (diff != last_diff && reset_on_progress)
3666			remaining = timeout_ms;
3667
3668		console_unlock();
3669
3670		if (diff == 0 || remaining == 0)
 
3671			break;
3672
3673		if (remaining < 0) {
3674			/* no timeout limit */
3675			msleep(100);
3676		} else if (remaining < 100) {
3677			msleep(remaining);
3678			remaining = 0;
3679		} else {
3680			msleep(100);
3681			remaining -= 100;
3682		}
3683
3684		last_diff = diff;
3685	}
3686
3687	return (diff == 0);
3688}
3689
3690/**
3691 * pr_flush() - Wait for printing threads to catch up.
3692 *
3693 * @timeout_ms:        The maximum time (in ms) to wait.
3694 * @reset_on_progress: Reset the timeout if forward progress is seen.
3695 *
3696 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3697 * represents infinite waiting.
3698 *
3699 * If @reset_on_progress is true, the timeout will be reset whenever any
3700 * printer has been seen to make some forward progress.
3701 *
3702 * Context: Process context. May sleep while acquiring console lock.
3703 * Return: true if all enabled printers are caught up.
3704 */
3705static bool pr_flush(int timeout_ms, bool reset_on_progress)
3706{
3707	return __pr_flush(NULL, timeout_ms, reset_on_progress);
3708}
3709
3710/*
3711 * Delayed printk version, for scheduler-internal messages:
3712 */
3713#define PRINTK_PENDING_WAKEUP	0x01
3714#define PRINTK_PENDING_OUTPUT	0x02
3715
3716static DEFINE_PER_CPU(int, printk_pending);
3717
3718static void wake_up_klogd_work_func(struct irq_work *irq_work)
3719{
3720	int pending = this_cpu_xchg(printk_pending, 0);
3721
3722	if (pending & PRINTK_PENDING_OUTPUT) {
3723		/* If trylock fails, someone else is doing the printing */
3724		if (console_trylock())
3725			console_unlock();
3726	}
3727
3728	if (pending & PRINTK_PENDING_WAKEUP)
3729		wake_up_interruptible(&log_wait);
3730}
3731
3732static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3733	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3734
3735static void __wake_up_klogd(int val)
3736{
3737	if (!printk_percpu_data_ready())
3738		return;
3739
3740	preempt_disable();
3741	/*
3742	 * Guarantee any new records can be seen by tasks preparing to wait
3743	 * before this context checks if the wait queue is empty.
3744	 *
3745	 * The full memory barrier within wq_has_sleeper() pairs with the full
3746	 * memory barrier within set_current_state() of
3747	 * prepare_to_wait_event(), which is called after ___wait_event() adds
3748	 * the waiter but before it has checked the wait condition.
3749	 *
3750	 * This pairs with devkmsg_read:A and syslog_print:A.
3751	 */
3752	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3753	    (val & PRINTK_PENDING_OUTPUT)) {
3754		this_cpu_or(printk_pending, val);
3755		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3756	}
3757	preempt_enable();
3758}
3759
 
 
 
 
 
 
 
 
 
 
3760void wake_up_klogd(void)
3761{
3762	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
3763}
3764
 
 
 
 
 
 
 
 
 
 
 
 
3765void defer_console_output(void)
3766{
3767	/*
3768	 * New messages may have been added directly to the ringbuffer
3769	 * using vprintk_store(), so wake any waiters as well.
3770	 */
3771	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3772}
3773
3774void printk_trigger_flush(void)
3775{
3776	defer_console_output();
3777}
3778
3779int vprintk_deferred(const char *fmt, va_list args)
3780{
3781	int r;
3782
3783	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3784	defer_console_output();
3785
3786	return r;
3787}
3788
3789int _printk_deferred(const char *fmt, ...)
3790{
3791	va_list args;
3792	int r;
3793
3794	va_start(args, fmt);
3795	r = vprintk_deferred(fmt, args);
3796	va_end(args);
3797
3798	return r;
3799}
3800
3801/*
3802 * printk rate limiting, lifted from the networking subsystem.
3803 *
3804 * This enforces a rate limit: not more than 10 kernel messages
3805 * every 5s to make a denial-of-service attack impossible.
3806 */
3807DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3808
3809int __printk_ratelimit(const char *func)
3810{
3811	return ___ratelimit(&printk_ratelimit_state, func);
3812}
3813EXPORT_SYMBOL(__printk_ratelimit);
3814
3815/**
3816 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3817 * @caller_jiffies: pointer to caller's state
3818 * @interval_msecs: minimum interval between prints
3819 *
3820 * printk_timed_ratelimit() returns true if more than @interval_msecs
3821 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3822 * returned true.
3823 */
3824bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3825			unsigned int interval_msecs)
3826{
3827	unsigned long elapsed = jiffies - *caller_jiffies;
3828
3829	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3830		return false;
3831
3832	*caller_jiffies = jiffies;
3833	return true;
3834}
3835EXPORT_SYMBOL(printk_timed_ratelimit);
3836
3837static DEFINE_SPINLOCK(dump_list_lock);
3838static LIST_HEAD(dump_list);
3839
3840/**
3841 * kmsg_dump_register - register a kernel log dumper.
3842 * @dumper: pointer to the kmsg_dumper structure
3843 *
3844 * Adds a kernel log dumper to the system. The dump callback in the
3845 * structure will be called when the kernel oopses or panics and must be
3846 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3847 */
3848int kmsg_dump_register(struct kmsg_dumper *dumper)
3849{
3850	unsigned long flags;
3851	int err = -EBUSY;
3852
3853	/* The dump callback needs to be set */
3854	if (!dumper->dump)
3855		return -EINVAL;
3856
3857	spin_lock_irqsave(&dump_list_lock, flags);
3858	/* Don't allow registering multiple times */
3859	if (!dumper->registered) {
3860		dumper->registered = 1;
3861		list_add_tail_rcu(&dumper->list, &dump_list);
3862		err = 0;
3863	}
3864	spin_unlock_irqrestore(&dump_list_lock, flags);
3865
3866	return err;
3867}
3868EXPORT_SYMBOL_GPL(kmsg_dump_register);
3869
3870/**
3871 * kmsg_dump_unregister - unregister a kmsg dumper.
3872 * @dumper: pointer to the kmsg_dumper structure
3873 *
3874 * Removes a dump device from the system. Returns zero on success and
3875 * %-EINVAL otherwise.
3876 */
3877int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3878{
3879	unsigned long flags;
3880	int err = -EINVAL;
3881
3882	spin_lock_irqsave(&dump_list_lock, flags);
3883	if (dumper->registered) {
3884		dumper->registered = 0;
3885		list_del_rcu(&dumper->list);
3886		err = 0;
3887	}
3888	spin_unlock_irqrestore(&dump_list_lock, flags);
3889	synchronize_rcu();
3890
3891	return err;
3892}
3893EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3894
3895static bool always_kmsg_dump;
3896module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3897
3898const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3899{
3900	switch (reason) {
3901	case KMSG_DUMP_PANIC:
3902		return "Panic";
3903	case KMSG_DUMP_OOPS:
3904		return "Oops";
3905	case KMSG_DUMP_EMERG:
3906		return "Emergency";
3907	case KMSG_DUMP_SHUTDOWN:
3908		return "Shutdown";
3909	default:
3910		return "Unknown";
3911	}
3912}
3913EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3914
3915/**
3916 * kmsg_dump - dump kernel log to kernel message dumpers.
3917 * @reason: the reason (oops, panic etc) for dumping
3918 *
3919 * Call each of the registered dumper's dump() callback, which can
3920 * retrieve the kmsg records with kmsg_dump_get_line() or
3921 * kmsg_dump_get_buffer().
3922 */
3923void kmsg_dump(enum kmsg_dump_reason reason)
3924{
3925	struct kmsg_dumper *dumper;
3926
3927	rcu_read_lock();
3928	list_for_each_entry_rcu(dumper, &dump_list, list) {
3929		enum kmsg_dump_reason max_reason = dumper->max_reason;
3930
3931		/*
3932		 * If client has not provided a specific max_reason, default
3933		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3934		 */
3935		if (max_reason == KMSG_DUMP_UNDEF) {
3936			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3937							KMSG_DUMP_OOPS;
3938		}
3939		if (reason > max_reason)
3940			continue;
3941
3942		/* invoke dumper which will iterate over records */
3943		dumper->dump(dumper, reason);
3944	}
3945	rcu_read_unlock();
3946}
3947
3948/**
3949 * kmsg_dump_get_line - retrieve one kmsg log line
3950 * @iter: kmsg dump iterator
3951 * @syslog: include the "<4>" prefixes
3952 * @line: buffer to copy the line to
3953 * @size: maximum size of the buffer
3954 * @len: length of line placed into buffer
3955 *
3956 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3957 * record, and copy one record into the provided buffer.
3958 *
3959 * Consecutive calls will return the next available record moving
3960 * towards the end of the buffer with the youngest messages.
3961 *
3962 * A return value of FALSE indicates that there are no more records to
3963 * read.
3964 */
3965bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3966			char *line, size_t size, size_t *len)
3967{
3968	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3969	struct printk_info info;
3970	unsigned int line_count;
3971	struct printk_record r;
3972	size_t l = 0;
3973	bool ret = false;
3974
3975	if (iter->cur_seq < min_seq)
3976		iter->cur_seq = min_seq;
3977
3978	prb_rec_init_rd(&r, &info, line, size);
3979
3980	/* Read text or count text lines? */
3981	if (line) {
3982		if (!prb_read_valid(prb, iter->cur_seq, &r))
3983			goto out;
3984		l = record_print_text(&r, syslog, printk_time);
3985	} else {
3986		if (!prb_read_valid_info(prb, iter->cur_seq,
3987					 &info, &line_count)) {
3988			goto out;
3989		}
3990		l = get_record_print_text_size(&info, line_count, syslog,
3991					       printk_time);
3992
3993	}
3994
3995	iter->cur_seq = r.info->seq + 1;
3996	ret = true;
3997out:
3998	if (len)
3999		*len = l;
4000	return ret;
4001}
4002EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4003
4004/**
4005 * kmsg_dump_get_buffer - copy kmsg log lines
4006 * @iter: kmsg dump iterator
4007 * @syslog: include the "<4>" prefixes
4008 * @buf: buffer to copy the line to
4009 * @size: maximum size of the buffer
4010 * @len_out: length of line placed into buffer
4011 *
4012 * Start at the end of the kmsg buffer and fill the provided buffer
4013 * with as many of the *youngest* kmsg records that fit into it.
4014 * If the buffer is large enough, all available kmsg records will be
4015 * copied with a single call.
4016 *
4017 * Consecutive calls will fill the buffer with the next block of
4018 * available older records, not including the earlier retrieved ones.
4019 *
4020 * A return value of FALSE indicates that there are no more records to
4021 * read.
4022 */
4023bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4024			  char *buf, size_t size, size_t *len_out)
4025{
4026	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4027	struct printk_info info;
4028	struct printk_record r;
4029	u64 seq;
4030	u64 next_seq;
4031	size_t len = 0;
4032	bool ret = false;
4033	bool time = printk_time;
4034
4035	if (!buf || !size)
4036		goto out;
4037
4038	if (iter->cur_seq < min_seq)
4039		iter->cur_seq = min_seq;
4040
4041	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4042		if (info.seq != iter->cur_seq) {
4043			/* messages are gone, move to first available one */
4044			iter->cur_seq = info.seq;
4045		}
4046	}
4047
4048	/* last entry */
4049	if (iter->cur_seq >= iter->next_seq)
4050		goto out;
4051
4052	/*
4053	 * Find first record that fits, including all following records,
4054	 * into the user-provided buffer for this dump. Pass in size-1
4055	 * because this function (by way of record_print_text()) will
4056	 * not write more than size-1 bytes of text into @buf.
4057	 */
4058	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4059				     size - 1, syslog, time);
4060
4061	/*
4062	 * Next kmsg_dump_get_buffer() invocation will dump block of
4063	 * older records stored right before this one.
4064	 */
4065	next_seq = seq;
4066
4067	prb_rec_init_rd(&r, &info, buf, size);
4068
4069	len = 0;
4070	prb_for_each_record(seq, prb, seq, &r) {
4071		if (r.info->seq >= iter->next_seq)
4072			break;
4073
4074		len += record_print_text(&r, syslog, time);
4075
4076		/* Adjust record to store to remaining buffer space. */
4077		prb_rec_init_rd(&r, &info, buf + len, size - len);
4078	}
4079
4080	iter->next_seq = next_seq;
4081	ret = true;
4082out:
4083	if (len_out)
4084		*len_out = len;
4085	return ret;
4086}
4087EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4088
4089/**
4090 * kmsg_dump_rewind - reset the iterator
4091 * @iter: kmsg dump iterator
4092 *
4093 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4094 * kmsg_dump_get_buffer() can be called again and used multiple
4095 * times within the same dumper.dump() callback.
4096 */
4097void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4098{
4099	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4100	iter->next_seq = prb_next_seq(prb);
4101}
4102EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4103
4104#endif
4105
4106#ifdef CONFIG_SMP
4107static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4108static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4109
4110/**
4111 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4112 *                            spinning lock is not owned by any CPU.
4113 *
4114 * Context: Any context.
4115 */
4116void __printk_cpu_sync_wait(void)
4117{
4118	do {
4119		cpu_relax();
4120	} while (atomic_read(&printk_cpu_sync_owner) != -1);
4121}
4122EXPORT_SYMBOL(__printk_cpu_sync_wait);
4123
4124/**
4125 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4126 *                               spinning lock.
4127 *
4128 * If no processor has the lock, the calling processor takes the lock and
4129 * becomes the owner. If the calling processor is already the owner of the
4130 * lock, this function succeeds immediately.
4131 *
4132 * Context: Any context. Expects interrupts to be disabled.
4133 * Return: 1 on success, otherwise 0.
4134 */
4135int __printk_cpu_sync_try_get(void)
4136{
4137	int cpu;
4138	int old;
4139
4140	cpu = smp_processor_id();
4141
4142	/*
4143	 * Guarantee loads and stores from this CPU when it is the lock owner
4144	 * are _not_ visible to the previous lock owner. This pairs with
4145	 * __printk_cpu_sync_put:B.
4146	 *
4147	 * Memory barrier involvement:
4148	 *
4149	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4150	 * then __printk_cpu_sync_put:A can never read from
4151	 * __printk_cpu_sync_try_get:B.
4152	 *
4153	 * Relies on:
4154	 *
4155	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4156	 * of the previous CPU
4157	 *    matching
4158	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4159	 * __printk_cpu_sync_try_get:B of this CPU
4160	 */
4161	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4162				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4163	if (old == -1) {
4164		/*
4165		 * This CPU is now the owner and begins loading/storing
4166		 * data: LMM(__printk_cpu_sync_try_get:B)
4167		 */
4168		return 1;
4169
4170	} else if (old == cpu) {
4171		/* This CPU is already the owner. */
4172		atomic_inc(&printk_cpu_sync_nested);
4173		return 1;
4174	}
4175
4176	return 0;
4177}
4178EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4179
4180/**
4181 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4182 *
4183 * The calling processor must be the owner of the lock.
4184 *
4185 * Context: Any context. Expects interrupts to be disabled.
4186 */
4187void __printk_cpu_sync_put(void)
4188{
4189	if (atomic_read(&printk_cpu_sync_nested)) {
4190		atomic_dec(&printk_cpu_sync_nested);
4191		return;
4192	}
4193
4194	/*
4195	 * This CPU is finished loading/storing data:
4196	 * LMM(__printk_cpu_sync_put:A)
4197	 */
4198
4199	/*
4200	 * Guarantee loads and stores from this CPU when it was the
4201	 * lock owner are visible to the next lock owner. This pairs
4202	 * with __printk_cpu_sync_try_get:A.
4203	 *
4204	 * Memory barrier involvement:
4205	 *
4206	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4207	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4208	 *
4209	 * Relies on:
4210	 *
4211	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4212	 * of this CPU
4213	 *    matching
4214	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4215	 * __printk_cpu_sync_try_get:B of the next CPU
4216	 */
4217	atomic_set_release(&printk_cpu_sync_owner,
4218			   -1); /* LMM(__printk_cpu_sync_put:B) */
4219}
4220EXPORT_SYMBOL(__printk_cpu_sync_put);
4221#endif /* CONFIG_SMP */
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/kernel/printk.c
   4 *
   5 *  Copyright (C) 1991, 1992  Linus Torvalds
   6 *
   7 * Modified to make sys_syslog() more flexible: added commands to
   8 * return the last 4k of kernel messages, regardless of whether
   9 * they've been read or not.  Added option to suppress kernel printk's
  10 * to the console.  Added hook for sending the console messages
  11 * elsewhere, in preparation for a serial line console (someday).
  12 * Ted Ts'o, 2/11/93.
  13 * Modified for sysctl support, 1/8/97, Chris Horn.
  14 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  15 *     manfred@colorfullife.com
  16 * Rewrote bits to get rid of console_lock
  17 *	01Mar01 Andrew Morton
  18 */
  19
  20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  21
  22#include <linux/kernel.h>
  23#include <linux/mm.h>
  24#include <linux/tty.h>
  25#include <linux/tty_driver.h>
  26#include <linux/console.h>
  27#include <linux/init.h>
  28#include <linux/jiffies.h>
  29#include <linux/nmi.h>
  30#include <linux/module.h>
  31#include <linux/moduleparam.h>
  32#include <linux/delay.h>
  33#include <linux/smp.h>
  34#include <linux/security.h>
  35#include <linux/memblock.h>
  36#include <linux/syscalls.h>
  37#include <linux/vmcore_info.h>
  38#include <linux/ratelimit.h>
  39#include <linux/kmsg_dump.h>
  40#include <linux/syslog.h>
  41#include <linux/cpu.h>
  42#include <linux/rculist.h>
  43#include <linux/poll.h>
  44#include <linux/irq_work.h>
  45#include <linux/ctype.h>
  46#include <linux/uio.h>
  47#include <linux/sched/clock.h>
  48#include <linux/sched/debug.h>
  49#include <linux/sched/task_stack.h>
  50
  51#include <linux/uaccess.h>
  52#include <asm/sections.h>
  53
  54#include <trace/events/initcall.h>
  55#define CREATE_TRACE_POINTS
  56#include <trace/events/printk.h>
  57
  58#include "printk_ringbuffer.h"
  59#include "console_cmdline.h"
  60#include "braille.h"
  61#include "internal.h"
  62
  63int console_printk[4] = {
  64	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  65	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  66	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  67	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  68};
  69EXPORT_SYMBOL_GPL(console_printk);
  70
  71atomic_t ignore_console_lock_warning __read_mostly = ATOMIC_INIT(0);
  72EXPORT_SYMBOL(ignore_console_lock_warning);
  73
  74EXPORT_TRACEPOINT_SYMBOL_GPL(console);
  75
  76/*
  77 * Low level drivers may need that to know if they can schedule in
  78 * their unblank() callback or not. So let's export it.
  79 */
  80int oops_in_progress;
  81EXPORT_SYMBOL(oops_in_progress);
  82
  83/*
  84 * console_mutex protects console_list updates and console->flags updates.
  85 * The flags are synchronized only for consoles that are registered, i.e.
  86 * accessible via the console list.
  87 */
  88static DEFINE_MUTEX(console_mutex);
  89
  90/*
  91 * console_sem protects updates to console->seq
  92 * and also provides serialization for console printing.
  93 */
  94static DEFINE_SEMAPHORE(console_sem, 1);
  95HLIST_HEAD(console_list);
  96EXPORT_SYMBOL_GPL(console_list);
  97DEFINE_STATIC_SRCU(console_srcu);
  98
  99/*
 100 * System may need to suppress printk message under certain
 101 * circumstances, like after kernel panic happens.
 102 */
 103int __read_mostly suppress_printk;
 104
 
 
 
 
 
 
 105#ifdef CONFIG_LOCKDEP
 106static struct lockdep_map console_lock_dep_map = {
 107	.name = "console_lock"
 108};
 109
 110void lockdep_assert_console_list_lock_held(void)
 111{
 112	lockdep_assert_held(&console_mutex);
 113}
 114EXPORT_SYMBOL(lockdep_assert_console_list_lock_held);
 115#endif
 116
 117#ifdef CONFIG_DEBUG_LOCK_ALLOC
 118bool console_srcu_read_lock_is_held(void)
 119{
 120	return srcu_read_lock_held(&console_srcu);
 121}
 122EXPORT_SYMBOL(console_srcu_read_lock_is_held);
 123#endif
 124
 125enum devkmsg_log_bits {
 126	__DEVKMSG_LOG_BIT_ON = 0,
 127	__DEVKMSG_LOG_BIT_OFF,
 128	__DEVKMSG_LOG_BIT_LOCK,
 129};
 130
 131enum devkmsg_log_masks {
 132	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
 133	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 134	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 135};
 136
 137/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 138#define DEVKMSG_LOG_MASK_DEFAULT	0
 139
 140static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 141
 142static int __control_devkmsg(char *str)
 143{
 144	size_t len;
 145
 146	if (!str)
 147		return -EINVAL;
 148
 149	len = str_has_prefix(str, "on");
 150	if (len) {
 151		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 152		return len;
 153	}
 154
 155	len = str_has_prefix(str, "off");
 156	if (len) {
 157		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 158		return len;
 159	}
 160
 161	len = str_has_prefix(str, "ratelimit");
 162	if (len) {
 163		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 164		return len;
 165	}
 166
 167	return -EINVAL;
 168}
 169
 170static int __init control_devkmsg(char *str)
 171{
 172	if (__control_devkmsg(str) < 0) {
 173		pr_warn("printk.devkmsg: bad option string '%s'\n", str);
 174		return 1;
 175	}
 176
 177	/*
 178	 * Set sysctl string accordingly:
 179	 */
 180	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 181		strcpy(devkmsg_log_str, "on");
 182	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 183		strcpy(devkmsg_log_str, "off");
 184	/* else "ratelimit" which is set by default. */
 185
 186	/*
 187	 * Sysctl cannot change it anymore. The kernel command line setting of
 188	 * this parameter is to force the setting to be permanent throughout the
 189	 * runtime of the system. This is a precation measure against userspace
 190	 * trying to be a smarta** and attempting to change it up on us.
 191	 */
 192	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 193
 194	return 1;
 195}
 196__setup("printk.devkmsg=", control_devkmsg);
 197
 198char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 199#if defined(CONFIG_PRINTK) && defined(CONFIG_SYSCTL)
 200int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 201			      void *buffer, size_t *lenp, loff_t *ppos)
 202{
 203	char old_str[DEVKMSG_STR_MAX_SIZE];
 204	unsigned int old;
 205	int err;
 206
 207	if (write) {
 208		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 209			return -EINVAL;
 210
 211		old = devkmsg_log;
 212		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 213	}
 214
 215	err = proc_dostring(table, write, buffer, lenp, ppos);
 216	if (err)
 217		return err;
 218
 219	if (write) {
 220		err = __control_devkmsg(devkmsg_log_str);
 221
 222		/*
 223		 * Do not accept an unknown string OR a known string with
 224		 * trailing crap...
 225		 */
 226		if (err < 0 || (err + 1 != *lenp)) {
 227
 228			/* ... and restore old setting. */
 229			devkmsg_log = old;
 230			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 231
 232			return -EINVAL;
 233		}
 234	}
 235
 236	return 0;
 237}
 238#endif /* CONFIG_PRINTK && CONFIG_SYSCTL */
 239
 240/**
 241 * console_list_lock - Lock the console list
 242 *
 243 * For console list or console->flags updates
 244 */
 245void console_list_lock(void)
 246{
 247	/*
 248	 * In unregister_console() and console_force_preferred_locked(),
 249	 * synchronize_srcu() is called with the console_list_lock held.
 250	 * Therefore it is not allowed that the console_list_lock is taken
 251	 * with the srcu_lock held.
 252	 *
 253	 * Detecting if this context is really in the read-side critical
 254	 * section is only possible if the appropriate debug options are
 255	 * enabled.
 256	 */
 257	WARN_ON_ONCE(debug_lockdep_rcu_enabled() &&
 258		     srcu_read_lock_held(&console_srcu));
 259
 260	mutex_lock(&console_mutex);
 261}
 262EXPORT_SYMBOL(console_list_lock);
 263
 264/**
 265 * console_list_unlock - Unlock the console list
 266 *
 267 * Counterpart to console_list_lock()
 268 */
 269void console_list_unlock(void)
 270{
 271	mutex_unlock(&console_mutex);
 272}
 273EXPORT_SYMBOL(console_list_unlock);
 274
 275/**
 276 * console_srcu_read_lock - Register a new reader for the
 277 *	SRCU-protected console list
 278 *
 279 * Use for_each_console_srcu() to iterate the console list
 280 *
 281 * Context: Any context.
 282 * Return: A cookie to pass to console_srcu_read_unlock().
 283 */
 284int console_srcu_read_lock(void)
 285{
 286	return srcu_read_lock_nmisafe(&console_srcu);
 287}
 288EXPORT_SYMBOL(console_srcu_read_lock);
 289
 290/**
 291 * console_srcu_read_unlock - Unregister an old reader from
 292 *	the SRCU-protected console list
 293 * @cookie: cookie returned from console_srcu_read_lock()
 294 *
 295 * Counterpart to console_srcu_read_lock()
 296 */
 297void console_srcu_read_unlock(int cookie)
 298{
 299	srcu_read_unlock_nmisafe(&console_srcu, cookie);
 300}
 301EXPORT_SYMBOL(console_srcu_read_unlock);
 302
 303/*
 304 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 305 * macros instead of functions so that _RET_IP_ contains useful information.
 306 */
 307#define down_console_sem() do { \
 308	down(&console_sem);\
 309	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 310} while (0)
 311
 312static int __down_trylock_console_sem(unsigned long ip)
 313{
 314	int lock_failed;
 315	unsigned long flags;
 316
 317	/*
 318	 * Here and in __up_console_sem() we need to be in safe mode,
 319	 * because spindump/WARN/etc from under console ->lock will
 320	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 321	 */
 322	printk_safe_enter_irqsave(flags);
 323	lock_failed = down_trylock(&console_sem);
 324	printk_safe_exit_irqrestore(flags);
 325
 326	if (lock_failed)
 327		return 1;
 328	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 329	return 0;
 330}
 331#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 332
 333static void __up_console_sem(unsigned long ip)
 334{
 335	unsigned long flags;
 336
 337	mutex_release(&console_lock_dep_map, ip);
 338
 339	printk_safe_enter_irqsave(flags);
 340	up(&console_sem);
 341	printk_safe_exit_irqrestore(flags);
 342}
 343#define up_console_sem() __up_console_sem(_RET_IP_)
 344
 345static bool panic_in_progress(void)
 346{
 347	return unlikely(atomic_read(&panic_cpu) != PANIC_CPU_INVALID);
 348}
 349
 350/* Return true if a panic is in progress on the current CPU. */
 351bool this_cpu_in_panic(void)
 352{
 353	/*
 354	 * We can use raw_smp_processor_id() here because it is impossible for
 355	 * the task to be migrated to the panic_cpu, or away from it. If
 356	 * panic_cpu has already been set, and we're not currently executing on
 357	 * that CPU, then we never will be.
 358	 */
 359	return unlikely(atomic_read(&panic_cpu) == raw_smp_processor_id());
 360}
 361
 362/*
 363 * Return true if a panic is in progress on a remote CPU.
 364 *
 365 * On true, the local CPU should immediately release any printing resources
 366 * that may be needed by the panic CPU.
 367 */
 368bool other_cpu_in_panic(void)
 369{
 370	return (panic_in_progress() && !this_cpu_in_panic());
 371}
 372
 373/*
 374 * This is used for debugging the mess that is the VT code by
 375 * keeping track if we have the console semaphore held. It's
 376 * definitely not the perfect debug tool (we don't know if _WE_
 377 * hold it and are racing, but it helps tracking those weird code
 378 * paths in the console code where we end up in places I want
 379 * locked without the console semaphore held).
 380 */
 381static int console_locked;
 382
 383/*
 384 *	Array of consoles built from command line options (console=)
 385 */
 386
 387#define MAX_CMDLINECONSOLES 8
 388
 389static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 390
 391static int preferred_console = -1;
 392int console_set_on_cmdline;
 393EXPORT_SYMBOL(console_set_on_cmdline);
 394
 395/* Flag: console code may call schedule() */
 396static int console_may_schedule;
 397
 398enum con_msg_format_flags {
 399	MSG_FORMAT_DEFAULT	= 0,
 400	MSG_FORMAT_SYSLOG	= (1 << 0),
 401};
 402
 403static int console_msg_format = MSG_FORMAT_DEFAULT;
 404
 405/*
 406 * The printk log buffer consists of a sequenced collection of records, each
 407 * containing variable length message text. Every record also contains its
 408 * own meta-data (@info).
 409 *
 410 * Every record meta-data carries the timestamp in microseconds, as well as
 411 * the standard userspace syslog level and syslog facility. The usual kernel
 412 * messages use LOG_KERN; userspace-injected messages always carry a matching
 413 * syslog facility, by default LOG_USER. The origin of every message can be
 414 * reliably determined that way.
 415 *
 416 * The human readable log message of a record is available in @text, the
 417 * length of the message text in @text_len. The stored message is not
 418 * terminated.
 419 *
 420 * Optionally, a record can carry a dictionary of properties (key/value
 421 * pairs), to provide userspace with a machine-readable message context.
 422 *
 423 * Examples for well-defined, commonly used property names are:
 424 *   DEVICE=b12:8               device identifier
 425 *                                b12:8         block dev_t
 426 *                                c127:3        char dev_t
 427 *                                n8            netdev ifindex
 428 *                                +sound:card0  subsystem:devname
 429 *   SUBSYSTEM=pci              driver-core subsystem name
 430 *
 431 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
 432 * and values are terminated by a '\0' character.
 433 *
 434 * Example of record values:
 435 *   record.text_buf                = "it's a line" (unterminated)
 436 *   record.info.seq                = 56
 437 *   record.info.ts_nsec            = 36863
 438 *   record.info.text_len           = 11
 439 *   record.info.facility           = 0 (LOG_KERN)
 440 *   record.info.flags              = 0
 441 *   record.info.level              = 3 (LOG_ERR)
 442 *   record.info.caller_id          = 299 (task 299)
 443 *   record.info.dev_info.subsystem = "pci" (terminated)
 444 *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
 445 *
 446 * The 'struct printk_info' buffer must never be directly exported to
 447 * userspace, it is a kernel-private implementation detail that might
 448 * need to be changed in the future, when the requirements change.
 449 *
 450 * /dev/kmsg exports the structured data in the following line format:
 451 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 452 *
 453 * Users of the export format should ignore possible additional values
 454 * separated by ',', and find the message after the ';' character.
 455 *
 456 * The optional key/value pairs are attached as continuation lines starting
 457 * with a space character and terminated by a newline. All possible
 458 * non-prinatable characters are escaped in the "\xff" notation.
 459 */
 460
 461/* syslog_lock protects syslog_* variables and write access to clear_seq. */
 462static DEFINE_MUTEX(syslog_lock);
 463
 464#ifdef CONFIG_PRINTK
 465DECLARE_WAIT_QUEUE_HEAD(log_wait);
 466/* All 3 protected by @syslog_lock. */
 467/* the next printk record to read by syslog(READ) or /proc/kmsg */
 468static u64 syslog_seq;
 469static size_t syslog_partial;
 470static bool syslog_time;
 471
 472struct latched_seq {
 473	seqcount_latch_t	latch;
 474	u64			val[2];
 475};
 476
 477/*
 478 * The next printk record to read after the last 'clear' command. There are
 479 * two copies (updated with seqcount_latch) so that reads can locklessly
 480 * access a valid value. Writers are synchronized by @syslog_lock.
 481 */
 482static struct latched_seq clear_seq = {
 483	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
 484	.val[0]		= 0,
 485	.val[1]		= 0,
 486};
 487
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 488#define LOG_LEVEL(v)		((v) & 0x07)
 489#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 490
 491/* record buffer */
 492#define LOG_ALIGN __alignof__(unsigned long)
 493#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 494#define LOG_BUF_LEN_MAX (u32)(1 << 31)
 495static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 496static char *log_buf = __log_buf;
 497static u32 log_buf_len = __LOG_BUF_LEN;
 498
 499/*
 500 * Define the average message size. This only affects the number of
 501 * descriptors that will be available. Underestimating is better than
 502 * overestimating (too many available descriptors is better than not enough).
 503 */
 504#define PRB_AVGBITS 5	/* 32 character average length */
 505
 506#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
 507#error CONFIG_LOG_BUF_SHIFT value too small.
 508#endif
 509_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
 510		 PRB_AVGBITS, &__log_buf[0]);
 511
 512static struct printk_ringbuffer printk_rb_dynamic;
 513
 514struct printk_ringbuffer *prb = &printk_rb_static;
 515
 516/*
 517 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
 518 * per_cpu_areas are initialised. This variable is set to true when
 519 * it's safe to access per-CPU data.
 520 */
 521static bool __printk_percpu_data_ready __ro_after_init;
 522
 523bool printk_percpu_data_ready(void)
 524{
 525	return __printk_percpu_data_ready;
 526}
 527
 528/* Must be called under syslog_lock. */
 529static void latched_seq_write(struct latched_seq *ls, u64 val)
 530{
 531	raw_write_seqcount_latch(&ls->latch);
 532	ls->val[0] = val;
 533	raw_write_seqcount_latch(&ls->latch);
 534	ls->val[1] = val;
 535}
 536
 537/* Can be called from any context. */
 538static u64 latched_seq_read_nolock(struct latched_seq *ls)
 539{
 540	unsigned int seq;
 541	unsigned int idx;
 542	u64 val;
 543
 544	do {
 545		seq = raw_read_seqcount_latch(&ls->latch);
 546		idx = seq & 0x1;
 547		val = ls->val[idx];
 548	} while (raw_read_seqcount_latch_retry(&ls->latch, seq));
 549
 550	return val;
 551}
 552
 553/* Return log buffer address */
 554char *log_buf_addr_get(void)
 555{
 556	return log_buf;
 557}
 558
 559/* Return log buffer size */
 560u32 log_buf_len_get(void)
 561{
 562	return log_buf_len;
 563}
 564
 565/*
 566 * Define how much of the log buffer we could take at maximum. The value
 567 * must be greater than two. Note that only half of the buffer is available
 568 * when the index points to the middle.
 569 */
 570#define MAX_LOG_TAKE_PART 4
 571static const char trunc_msg[] = "<truncated>";
 572
 573static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
 574{
 575	/*
 576	 * The message should not take the whole buffer. Otherwise, it might
 577	 * get removed too soon.
 578	 */
 579	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 580
 581	if (*text_len > max_text_len)
 582		*text_len = max_text_len;
 583
 584	/* enable the warning message (if there is room) */
 585	*trunc_msg_len = strlen(trunc_msg);
 586	if (*text_len >= *trunc_msg_len)
 587		*text_len -= *trunc_msg_len;
 588	else
 589		*trunc_msg_len = 0;
 590}
 591
 592int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 593
 594static int syslog_action_restricted(int type)
 595{
 596	if (dmesg_restrict)
 597		return 1;
 598	/*
 599	 * Unless restricted, we allow "read all" and "get buffer size"
 600	 * for everybody.
 601	 */
 602	return type != SYSLOG_ACTION_READ_ALL &&
 603	       type != SYSLOG_ACTION_SIZE_BUFFER;
 604}
 605
 606static int check_syslog_permissions(int type, int source)
 607{
 608	/*
 609	 * If this is from /proc/kmsg and we've already opened it, then we've
 610	 * already done the capabilities checks at open time.
 611	 */
 612	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 613		goto ok;
 614
 615	if (syslog_action_restricted(type)) {
 616		if (capable(CAP_SYSLOG))
 617			goto ok;
 
 
 
 
 
 
 
 
 
 
 
 618		return -EPERM;
 619	}
 620ok:
 621	return security_syslog(type);
 622}
 623
 624static void append_char(char **pp, char *e, char c)
 625{
 626	if (*pp < e)
 627		*(*pp)++ = c;
 628}
 629
 630static ssize_t info_print_ext_header(char *buf, size_t size,
 631				     struct printk_info *info)
 632{
 633	u64 ts_usec = info->ts_nsec;
 634	char caller[20];
 635#ifdef CONFIG_PRINTK_CALLER
 636	u32 id = info->caller_id;
 637
 638	snprintf(caller, sizeof(caller), ",caller=%c%u",
 639		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
 640#else
 641	caller[0] = '\0';
 642#endif
 643
 644	do_div(ts_usec, 1000);
 645
 646	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
 647			 (info->facility << 3) | info->level, info->seq,
 648			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
 649}
 650
 651static ssize_t msg_add_ext_text(char *buf, size_t size,
 652				const char *text, size_t text_len,
 653				unsigned char endc)
 654{
 655	char *p = buf, *e = buf + size;
 656	size_t i;
 657
 658	/* escape non-printable characters */
 659	for (i = 0; i < text_len; i++) {
 660		unsigned char c = text[i];
 661
 662		if (c < ' ' || c >= 127 || c == '\\')
 663			p += scnprintf(p, e - p, "\\x%02x", c);
 664		else
 665			append_char(&p, e, c);
 666	}
 667	append_char(&p, e, endc);
 668
 669	return p - buf;
 670}
 671
 672static ssize_t msg_add_dict_text(char *buf, size_t size,
 673				 const char *key, const char *val)
 674{
 675	size_t val_len = strlen(val);
 676	ssize_t len;
 677
 678	if (!val_len)
 679		return 0;
 680
 681	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
 682	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
 683	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
 684
 685	return len;
 686}
 687
 688static ssize_t msg_print_ext_body(char *buf, size_t size,
 689				  char *text, size_t text_len,
 690				  struct dev_printk_info *dev_info)
 691{
 692	ssize_t len;
 693
 694	len = msg_add_ext_text(buf, size, text, text_len, '\n');
 695
 696	if (!dev_info)
 697		goto out;
 698
 699	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
 700				 dev_info->subsystem);
 701	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
 702				 dev_info->device);
 703out:
 704	return len;
 705}
 706
 707/* /dev/kmsg - userspace message inject/listen interface */
 708struct devkmsg_user {
 709	atomic64_t seq;
 710	struct ratelimit_state rs;
 711	struct mutex lock;
 712	struct printk_buffers pbufs;
 
 
 
 
 713};
 714
 715static __printf(3, 4) __cold
 716int devkmsg_emit(int facility, int level, const char *fmt, ...)
 717{
 718	va_list args;
 719	int r;
 720
 721	va_start(args, fmt);
 722	r = vprintk_emit(facility, level, NULL, fmt, args);
 723	va_end(args);
 724
 725	return r;
 726}
 727
 728static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 729{
 730	char *buf, *line;
 731	int level = default_message_loglevel;
 732	int facility = 1;	/* LOG_USER */
 733	struct file *file = iocb->ki_filp;
 734	struct devkmsg_user *user = file->private_data;
 735	size_t len = iov_iter_count(from);
 736	ssize_t ret = len;
 737
 738	if (len > PRINTKRB_RECORD_MAX)
 739		return -EINVAL;
 740
 741	/* Ignore when user logging is disabled. */
 742	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 743		return len;
 744
 745	/* Ratelimit when not explicitly enabled. */
 746	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 747		if (!___ratelimit(&user->rs, current->comm))
 748			return ret;
 749	}
 750
 751	buf = kmalloc(len+1, GFP_KERNEL);
 752	if (buf == NULL)
 753		return -ENOMEM;
 754
 755	buf[len] = '\0';
 756	if (!copy_from_iter_full(buf, len, from)) {
 757		kfree(buf);
 758		return -EFAULT;
 759	}
 760
 761	/*
 762	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 763	 * the decimal value represents 32bit, the lower 3 bit are the log
 764	 * level, the rest are the log facility.
 765	 *
 766	 * If no prefix or no userspace facility is specified, we
 767	 * enforce LOG_USER, to be able to reliably distinguish
 768	 * kernel-generated messages from userspace-injected ones.
 769	 */
 770	line = buf;
 771	if (line[0] == '<') {
 772		char *endp = NULL;
 773		unsigned int u;
 774
 775		u = simple_strtoul(line + 1, &endp, 10);
 776		if (endp && endp[0] == '>') {
 777			level = LOG_LEVEL(u);
 778			if (LOG_FACILITY(u) != 0)
 779				facility = LOG_FACILITY(u);
 780			endp++;
 781			line = endp;
 782		}
 783	}
 784
 785	devkmsg_emit(facility, level, "%s", line);
 786	kfree(buf);
 787	return ret;
 788}
 789
 790static ssize_t devkmsg_read(struct file *file, char __user *buf,
 791			    size_t count, loff_t *ppos)
 792{
 793	struct devkmsg_user *user = file->private_data;
 794	char *outbuf = &user->pbufs.outbuf[0];
 795	struct printk_message pmsg = {
 796		.pbufs = &user->pbufs,
 797	};
 798	ssize_t ret;
 799
 
 
 
 800	ret = mutex_lock_interruptible(&user->lock);
 801	if (ret)
 802		return ret;
 803
 804	if (!printk_get_next_message(&pmsg, atomic64_read(&user->seq), true, false)) {
 805		if (file->f_flags & O_NONBLOCK) {
 806			ret = -EAGAIN;
 807			goto out;
 808		}
 809
 810		/*
 811		 * Guarantee this task is visible on the waitqueue before
 812		 * checking the wake condition.
 813		 *
 814		 * The full memory barrier within set_current_state() of
 815		 * prepare_to_wait_event() pairs with the full memory barrier
 816		 * within wq_has_sleeper().
 817		 *
 818		 * This pairs with __wake_up_klogd:A.
 819		 */
 820		ret = wait_event_interruptible(log_wait,
 821				printk_get_next_message(&pmsg, atomic64_read(&user->seq), true,
 822							false)); /* LMM(devkmsg_read:A) */
 823		if (ret)
 824			goto out;
 825	}
 826
 827	if (pmsg.dropped) {
 828		/* our last seen message is gone, return error and reset */
 829		atomic64_set(&user->seq, pmsg.seq);
 830		ret = -EPIPE;
 831		goto out;
 832	}
 833
 834	atomic64_set(&user->seq, pmsg.seq + 1);
 
 
 
 835
 836	if (pmsg.outbuf_len > count) {
 
 
 837		ret = -EINVAL;
 838		goto out;
 839	}
 840
 841	if (copy_to_user(buf, outbuf, pmsg.outbuf_len)) {
 842		ret = -EFAULT;
 843		goto out;
 844	}
 845	ret = pmsg.outbuf_len;
 846out:
 847	mutex_unlock(&user->lock);
 848	return ret;
 849}
 850
 851/*
 852 * Be careful when modifying this function!!!
 853 *
 854 * Only few operations are supported because the device works only with the
 855 * entire variable length messages (records). Non-standard values are
 856 * returned in the other cases and has been this way for quite some time.
 857 * User space applications might depend on this behavior.
 858 */
 859static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 860{
 861	struct devkmsg_user *user = file->private_data;
 862	loff_t ret = 0;
 863
 
 
 864	if (offset)
 865		return -ESPIPE;
 866
 867	switch (whence) {
 868	case SEEK_SET:
 869		/* the first record */
 870		atomic64_set(&user->seq, prb_first_valid_seq(prb));
 871		break;
 872	case SEEK_DATA:
 873		/*
 874		 * The first record after the last SYSLOG_ACTION_CLEAR,
 875		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 876		 * changes no global state, and does not clear anything.
 877		 */
 878		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
 879		break;
 880	case SEEK_END:
 881		/* after the last record */
 882		atomic64_set(&user->seq, prb_next_seq(prb));
 883		break;
 884	default:
 885		ret = -EINVAL;
 886	}
 887	return ret;
 888}
 889
 890static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 891{
 892	struct devkmsg_user *user = file->private_data;
 893	struct printk_info info;
 894	__poll_t ret = 0;
 895
 
 
 
 896	poll_wait(file, &log_wait, wait);
 897
 898	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
 899		/* return error when data has vanished underneath us */
 900		if (info.seq != atomic64_read(&user->seq))
 901			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 902		else
 903			ret = EPOLLIN|EPOLLRDNORM;
 904	}
 905
 906	return ret;
 907}
 908
 909static int devkmsg_open(struct inode *inode, struct file *file)
 910{
 911	struct devkmsg_user *user;
 912	int err;
 913
 914	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 915		return -EPERM;
 916
 917	/* write-only does not need any file context */
 918	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 919		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 920					       SYSLOG_FROM_READER);
 921		if (err)
 922			return err;
 923	}
 924
 925	user = kvmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 926	if (!user)
 927		return -ENOMEM;
 928
 929	ratelimit_default_init(&user->rs);
 930	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 931
 932	mutex_init(&user->lock);
 933
 
 
 
 934	atomic64_set(&user->seq, prb_first_valid_seq(prb));
 935
 936	file->private_data = user;
 937	return 0;
 938}
 939
 940static int devkmsg_release(struct inode *inode, struct file *file)
 941{
 942	struct devkmsg_user *user = file->private_data;
 943
 
 
 
 944	ratelimit_state_exit(&user->rs);
 945
 946	mutex_destroy(&user->lock);
 947	kvfree(user);
 948	return 0;
 949}
 950
 951const struct file_operations kmsg_fops = {
 952	.open = devkmsg_open,
 953	.read = devkmsg_read,
 954	.write_iter = devkmsg_write,
 955	.llseek = devkmsg_llseek,
 956	.poll = devkmsg_poll,
 957	.release = devkmsg_release,
 958};
 959
 960#ifdef CONFIG_VMCORE_INFO
 961/*
 962 * This appends the listed symbols to /proc/vmcore
 963 *
 964 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 965 * obtain access to symbols that are otherwise very difficult to locate.  These
 966 * symbols are specifically used so that utilities can access and extract the
 967 * dmesg log from a vmcore file after a crash.
 968 */
 969void log_buf_vmcoreinfo_setup(void)
 970{
 971	struct dev_printk_info *dev_info = NULL;
 972
 973	VMCOREINFO_SYMBOL(prb);
 974	VMCOREINFO_SYMBOL(printk_rb_static);
 975	VMCOREINFO_SYMBOL(clear_seq);
 976
 977	/*
 978	 * Export struct size and field offsets. User space tools can
 979	 * parse it and detect any changes to structure down the line.
 980	 */
 981
 982	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
 983	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
 984	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
 985	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
 986
 987	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
 988	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
 989	VMCOREINFO_OFFSET(prb_desc_ring, descs);
 990	VMCOREINFO_OFFSET(prb_desc_ring, infos);
 991	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
 992	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
 993
 994	VMCOREINFO_STRUCT_SIZE(prb_desc);
 995	VMCOREINFO_OFFSET(prb_desc, state_var);
 996	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
 997
 998	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
 999	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
1000	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
1001
1002	VMCOREINFO_STRUCT_SIZE(printk_info);
1003	VMCOREINFO_OFFSET(printk_info, seq);
1004	VMCOREINFO_OFFSET(printk_info, ts_nsec);
1005	VMCOREINFO_OFFSET(printk_info, text_len);
1006	VMCOREINFO_OFFSET(printk_info, caller_id);
1007	VMCOREINFO_OFFSET(printk_info, dev_info);
1008
1009	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
1010	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
1011	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
1012	VMCOREINFO_OFFSET(dev_printk_info, device);
1013	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
1014
1015	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
1016	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
1017	VMCOREINFO_OFFSET(prb_data_ring, data);
1018	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
1019	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
1020
1021	VMCOREINFO_SIZE(atomic_long_t);
1022	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
1023
1024	VMCOREINFO_STRUCT_SIZE(latched_seq);
1025	VMCOREINFO_OFFSET(latched_seq, val);
1026}
1027#endif
1028
1029/* requested log_buf_len from kernel cmdline */
1030static unsigned long __initdata new_log_buf_len;
1031
1032/* we practice scaling the ring buffer by powers of 2 */
1033static void __init log_buf_len_update(u64 size)
1034{
1035	if (size > (u64)LOG_BUF_LEN_MAX) {
1036		size = (u64)LOG_BUF_LEN_MAX;
1037		pr_err("log_buf over 2G is not supported.\n");
1038	}
1039
1040	if (size)
1041		size = roundup_pow_of_two(size);
1042	if (size > log_buf_len)
1043		new_log_buf_len = (unsigned long)size;
1044}
1045
1046/* save requested log_buf_len since it's too early to process it */
1047static int __init log_buf_len_setup(char *str)
1048{
1049	u64 size;
1050
1051	if (!str)
1052		return -EINVAL;
1053
1054	size = memparse(str, &str);
1055
1056	log_buf_len_update(size);
1057
1058	return 0;
1059}
1060early_param("log_buf_len", log_buf_len_setup);
1061
1062#ifdef CONFIG_SMP
1063#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1064
1065static void __init log_buf_add_cpu(void)
1066{
1067	unsigned int cpu_extra;
1068
1069	/*
1070	 * archs should set up cpu_possible_bits properly with
1071	 * set_cpu_possible() after setup_arch() but just in
1072	 * case lets ensure this is valid.
1073	 */
1074	if (num_possible_cpus() == 1)
1075		return;
1076
1077	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1078
1079	/* by default this will only continue through for large > 64 CPUs */
1080	if (cpu_extra <= __LOG_BUF_LEN / 2)
1081		return;
1082
1083	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1084		__LOG_CPU_MAX_BUF_LEN);
1085	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1086		cpu_extra);
1087	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1088
1089	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1090}
1091#else /* !CONFIG_SMP */
1092static inline void log_buf_add_cpu(void) {}
1093#endif /* CONFIG_SMP */
1094
1095static void __init set_percpu_data_ready(void)
1096{
1097	__printk_percpu_data_ready = true;
1098}
1099
1100static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1101				     struct printk_record *r)
1102{
1103	struct prb_reserved_entry e;
1104	struct printk_record dest_r;
1105
1106	prb_rec_init_wr(&dest_r, r->info->text_len);
1107
1108	if (!prb_reserve(&e, rb, &dest_r))
1109		return 0;
1110
1111	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1112	dest_r.info->text_len = r->info->text_len;
1113	dest_r.info->facility = r->info->facility;
1114	dest_r.info->level = r->info->level;
1115	dest_r.info->flags = r->info->flags;
1116	dest_r.info->ts_nsec = r->info->ts_nsec;
1117	dest_r.info->caller_id = r->info->caller_id;
1118	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1119
1120	prb_final_commit(&e);
1121
1122	return prb_record_text_space(&e);
1123}
1124
1125static char setup_text_buf[PRINTKRB_RECORD_MAX] __initdata;
1126
1127void __init setup_log_buf(int early)
1128{
1129	struct printk_info *new_infos;
1130	unsigned int new_descs_count;
1131	struct prb_desc *new_descs;
1132	struct printk_info info;
1133	struct printk_record r;
1134	unsigned int text_size;
1135	size_t new_descs_size;
1136	size_t new_infos_size;
1137	unsigned long flags;
1138	char *new_log_buf;
1139	unsigned int free;
1140	u64 seq;
1141
1142	/*
1143	 * Some archs call setup_log_buf() multiple times - first is very
1144	 * early, e.g. from setup_arch(), and second - when percpu_areas
1145	 * are initialised.
1146	 */
1147	if (!early)
1148		set_percpu_data_ready();
1149
1150	if (log_buf != __log_buf)
1151		return;
1152
1153	if (!early && !new_log_buf_len)
1154		log_buf_add_cpu();
1155
1156	if (!new_log_buf_len)
1157		return;
1158
1159	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1160	if (new_descs_count == 0) {
1161		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1162		return;
1163	}
1164
1165	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1166	if (unlikely(!new_log_buf)) {
1167		pr_err("log_buf_len: %lu text bytes not available\n",
1168		       new_log_buf_len);
1169		return;
1170	}
1171
1172	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1173	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1174	if (unlikely(!new_descs)) {
1175		pr_err("log_buf_len: %zu desc bytes not available\n",
1176		       new_descs_size);
1177		goto err_free_log_buf;
1178	}
1179
1180	new_infos_size = new_descs_count * sizeof(struct printk_info);
1181	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1182	if (unlikely(!new_infos)) {
1183		pr_err("log_buf_len: %zu info bytes not available\n",
1184		       new_infos_size);
1185		goto err_free_descs;
1186	}
1187
1188	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1189
1190	prb_init(&printk_rb_dynamic,
1191		 new_log_buf, ilog2(new_log_buf_len),
1192		 new_descs, ilog2(new_descs_count),
1193		 new_infos);
1194
1195	local_irq_save(flags);
1196
1197	log_buf_len = new_log_buf_len;
1198	log_buf = new_log_buf;
1199	new_log_buf_len = 0;
1200
1201	free = __LOG_BUF_LEN;
1202	prb_for_each_record(0, &printk_rb_static, seq, &r) {
1203		text_size = add_to_rb(&printk_rb_dynamic, &r);
1204		if (text_size > free)
1205			free = 0;
1206		else
1207			free -= text_size;
1208	}
1209
1210	prb = &printk_rb_dynamic;
1211
1212	local_irq_restore(flags);
1213
1214	/*
1215	 * Copy any remaining messages that might have appeared from
1216	 * NMI context after copying but before switching to the
1217	 * dynamic buffer.
1218	 */
1219	prb_for_each_record(seq, &printk_rb_static, seq, &r) {
1220		text_size = add_to_rb(&printk_rb_dynamic, &r);
1221		if (text_size > free)
1222			free = 0;
1223		else
1224			free -= text_size;
1225	}
1226
1227	if (seq != prb_next_seq(&printk_rb_static)) {
1228		pr_err("dropped %llu messages\n",
1229		       prb_next_seq(&printk_rb_static) - seq);
1230	}
1231
1232	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1233	pr_info("early log buf free: %u(%u%%)\n",
1234		free, (free * 100) / __LOG_BUF_LEN);
1235	return;
1236
1237err_free_descs:
1238	memblock_free(new_descs, new_descs_size);
1239err_free_log_buf:
1240	memblock_free(new_log_buf, new_log_buf_len);
1241}
1242
1243static bool __read_mostly ignore_loglevel;
1244
1245static int __init ignore_loglevel_setup(char *str)
1246{
1247	ignore_loglevel = true;
1248	pr_info("debug: ignoring loglevel setting.\n");
1249
1250	return 0;
1251}
1252
1253early_param("ignore_loglevel", ignore_loglevel_setup);
1254module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1255MODULE_PARM_DESC(ignore_loglevel,
1256		 "ignore loglevel setting (prints all kernel messages to the console)");
1257
1258static bool suppress_message_printing(int level)
1259{
1260	return (level >= console_loglevel && !ignore_loglevel);
1261}
1262
1263#ifdef CONFIG_BOOT_PRINTK_DELAY
1264
1265static int boot_delay; /* msecs delay after each printk during bootup */
1266static unsigned long long loops_per_msec;	/* based on boot_delay */
1267
1268static int __init boot_delay_setup(char *str)
1269{
1270	unsigned long lpj;
1271
1272	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1273	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1274
1275	get_option(&str, &boot_delay);
1276	if (boot_delay > 10 * 1000)
1277		boot_delay = 0;
1278
1279	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1280		"HZ: %d, loops_per_msec: %llu\n",
1281		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1282	return 0;
1283}
1284early_param("boot_delay", boot_delay_setup);
1285
1286static void boot_delay_msec(int level)
1287{
1288	unsigned long long k;
1289	unsigned long timeout;
1290
1291	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1292		|| suppress_message_printing(level)) {
1293		return;
1294	}
1295
1296	k = (unsigned long long)loops_per_msec * boot_delay;
1297
1298	timeout = jiffies + msecs_to_jiffies(boot_delay);
1299	while (k) {
1300		k--;
1301		cpu_relax();
1302		/*
1303		 * use (volatile) jiffies to prevent
1304		 * compiler reduction; loop termination via jiffies
1305		 * is secondary and may or may not happen.
1306		 */
1307		if (time_after(jiffies, timeout))
1308			break;
1309		touch_nmi_watchdog();
1310	}
1311}
1312#else
1313static inline void boot_delay_msec(int level)
1314{
1315}
1316#endif
1317
1318static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1319module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1320
1321static size_t print_syslog(unsigned int level, char *buf)
1322{
1323	return sprintf(buf, "<%u>", level);
1324}
1325
1326static size_t print_time(u64 ts, char *buf)
1327{
1328	unsigned long rem_nsec = do_div(ts, 1000000000);
1329
1330	return sprintf(buf, "[%5lu.%06lu]",
1331		       (unsigned long)ts, rem_nsec / 1000);
1332}
1333
1334#ifdef CONFIG_PRINTK_CALLER
1335static size_t print_caller(u32 id, char *buf)
1336{
1337	char caller[12];
1338
1339	snprintf(caller, sizeof(caller), "%c%u",
1340		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1341	return sprintf(buf, "[%6s]", caller);
1342}
1343#else
1344#define print_caller(id, buf) 0
1345#endif
1346
1347static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1348				bool time, char *buf)
1349{
1350	size_t len = 0;
1351
1352	if (syslog)
1353		len = print_syslog((info->facility << 3) | info->level, buf);
1354
1355	if (time)
1356		len += print_time(info->ts_nsec, buf + len);
1357
1358	len += print_caller(info->caller_id, buf + len);
1359
1360	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1361		buf[len++] = ' ';
1362		buf[len] = '\0';
1363	}
1364
1365	return len;
1366}
1367
1368/*
1369 * Prepare the record for printing. The text is shifted within the given
1370 * buffer to avoid a need for another one. The following operations are
1371 * done:
1372 *
1373 *   - Add prefix for each line.
1374 *   - Drop truncated lines that no longer fit into the buffer.
1375 *   - Add the trailing newline that has been removed in vprintk_store().
1376 *   - Add a string terminator.
1377 *
1378 * Since the produced string is always terminated, the maximum possible
1379 * return value is @r->text_buf_size - 1;
1380 *
1381 * Return: The length of the updated/prepared text, including the added
1382 * prefixes and the newline. The terminator is not counted. The dropped
1383 * line(s) are not counted.
1384 */
1385static size_t record_print_text(struct printk_record *r, bool syslog,
1386				bool time)
1387{
1388	size_t text_len = r->info->text_len;
1389	size_t buf_size = r->text_buf_size;
1390	char *text = r->text_buf;
1391	char prefix[PRINTK_PREFIX_MAX];
1392	bool truncated = false;
1393	size_t prefix_len;
1394	size_t line_len;
1395	size_t len = 0;
1396	char *next;
1397
1398	/*
1399	 * If the message was truncated because the buffer was not large
1400	 * enough, treat the available text as if it were the full text.
1401	 */
1402	if (text_len > buf_size)
1403		text_len = buf_size;
1404
1405	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1406
1407	/*
1408	 * @text_len: bytes of unprocessed text
1409	 * @line_len: bytes of current line _without_ newline
1410	 * @text:     pointer to beginning of current line
1411	 * @len:      number of bytes prepared in r->text_buf
1412	 */
1413	for (;;) {
1414		next = memchr(text, '\n', text_len);
1415		if (next) {
1416			line_len = next - text;
1417		} else {
1418			/* Drop truncated line(s). */
1419			if (truncated)
1420				break;
1421			line_len = text_len;
1422		}
1423
1424		/*
1425		 * Truncate the text if there is not enough space to add the
1426		 * prefix and a trailing newline and a terminator.
1427		 */
1428		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1429			/* Drop even the current line if no space. */
1430			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1431				break;
1432
1433			text_len = buf_size - len - prefix_len - 1 - 1;
1434			truncated = true;
1435		}
1436
1437		memmove(text + prefix_len, text, text_len);
1438		memcpy(text, prefix, prefix_len);
1439
1440		/*
1441		 * Increment the prepared length to include the text and
1442		 * prefix that were just moved+copied. Also increment for the
1443		 * newline at the end of this line. If this is the last line,
1444		 * there is no newline, but it will be added immediately below.
1445		 */
1446		len += prefix_len + line_len + 1;
1447		if (text_len == line_len) {
1448			/*
1449			 * This is the last line. Add the trailing newline
1450			 * removed in vprintk_store().
1451			 */
1452			text[prefix_len + line_len] = '\n';
1453			break;
1454		}
1455
1456		/*
1457		 * Advance beyond the added prefix and the related line with
1458		 * its newline.
1459		 */
1460		text += prefix_len + line_len + 1;
1461
1462		/*
1463		 * The remaining text has only decreased by the line with its
1464		 * newline.
1465		 *
1466		 * Note that @text_len can become zero. It happens when @text
1467		 * ended with a newline (either due to truncation or the
1468		 * original string ending with "\n\n"). The loop is correctly
1469		 * repeated and (if not truncated) an empty line with a prefix
1470		 * will be prepared.
1471		 */
1472		text_len -= line_len + 1;
1473	}
1474
1475	/*
1476	 * If a buffer was provided, it will be terminated. Space for the
1477	 * string terminator is guaranteed to be available. The terminator is
1478	 * not counted in the return value.
1479	 */
1480	if (buf_size > 0)
1481		r->text_buf[len] = 0;
1482
1483	return len;
1484}
1485
1486static size_t get_record_print_text_size(struct printk_info *info,
1487					 unsigned int line_count,
1488					 bool syslog, bool time)
1489{
1490	char prefix[PRINTK_PREFIX_MAX];
1491	size_t prefix_len;
1492
1493	prefix_len = info_print_prefix(info, syslog, time, prefix);
1494
1495	/*
1496	 * Each line will be preceded with a prefix. The intermediate
1497	 * newlines are already within the text, but a final trailing
1498	 * newline will be added.
1499	 */
1500	return ((prefix_len * line_count) + info->text_len + 1);
1501}
1502
1503/*
1504 * Beginning with @start_seq, find the first record where it and all following
1505 * records up to (but not including) @max_seq fit into @size.
1506 *
1507 * @max_seq is simply an upper bound and does not need to exist. If the caller
1508 * does not require an upper bound, -1 can be used for @max_seq.
1509 */
1510static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1511				  bool syslog, bool time)
1512{
1513	struct printk_info info;
1514	unsigned int line_count;
1515	size_t len = 0;
1516	u64 seq;
1517
1518	/* Determine the size of the records up to @max_seq. */
1519	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1520		if (info.seq >= max_seq)
1521			break;
1522		len += get_record_print_text_size(&info, line_count, syslog, time);
1523	}
1524
1525	/*
1526	 * Adjust the upper bound for the next loop to avoid subtracting
1527	 * lengths that were never added.
1528	 */
1529	if (seq < max_seq)
1530		max_seq = seq;
1531
1532	/*
1533	 * Move first record forward until length fits into the buffer. Ignore
1534	 * newest messages that were not counted in the above cycle. Messages
1535	 * might appear and get lost in the meantime. This is a best effort
1536	 * that prevents an infinite loop that could occur with a retry.
1537	 */
1538	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1539		if (len <= size || info.seq >= max_seq)
1540			break;
1541		len -= get_record_print_text_size(&info, line_count, syslog, time);
1542	}
1543
1544	return seq;
1545}
1546
1547/* The caller is responsible for making sure @size is greater than 0. */
1548static int syslog_print(char __user *buf, int size)
1549{
1550	struct printk_info info;
1551	struct printk_record r;
1552	char *text;
1553	int len = 0;
1554	u64 seq;
1555
1556	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1557	if (!text)
1558		return -ENOMEM;
1559
1560	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1561
1562	mutex_lock(&syslog_lock);
1563
1564	/*
1565	 * Wait for the @syslog_seq record to be available. @syslog_seq may
1566	 * change while waiting.
1567	 */
1568	do {
1569		seq = syslog_seq;
1570
1571		mutex_unlock(&syslog_lock);
1572		/*
1573		 * Guarantee this task is visible on the waitqueue before
1574		 * checking the wake condition.
1575		 *
1576		 * The full memory barrier within set_current_state() of
1577		 * prepare_to_wait_event() pairs with the full memory barrier
1578		 * within wq_has_sleeper().
1579		 *
1580		 * This pairs with __wake_up_klogd:A.
1581		 */
1582		len = wait_event_interruptible(log_wait,
1583				prb_read_valid(prb, seq, NULL)); /* LMM(syslog_print:A) */
1584		mutex_lock(&syslog_lock);
1585
1586		if (len)
1587			goto out;
1588	} while (syslog_seq != seq);
1589
1590	/*
1591	 * Copy records that fit into the buffer. The above cycle makes sure
1592	 * that the first record is always available.
1593	 */
1594	do {
1595		size_t n;
1596		size_t skip;
1597		int err;
1598
1599		if (!prb_read_valid(prb, syslog_seq, &r))
1600			break;
1601
1602		if (r.info->seq != syslog_seq) {
1603			/* message is gone, move to next valid one */
1604			syslog_seq = r.info->seq;
1605			syslog_partial = 0;
1606		}
1607
1608		/*
1609		 * To keep reading/counting partial line consistent,
1610		 * use printk_time value as of the beginning of a line.
1611		 */
1612		if (!syslog_partial)
1613			syslog_time = printk_time;
1614
1615		skip = syslog_partial;
1616		n = record_print_text(&r, true, syslog_time);
1617		if (n - syslog_partial <= size) {
1618			/* message fits into buffer, move forward */
1619			syslog_seq = r.info->seq + 1;
1620			n -= syslog_partial;
1621			syslog_partial = 0;
1622		} else if (!len){
1623			/* partial read(), remember position */
1624			n = size;
1625			syslog_partial += n;
1626		} else
1627			n = 0;
1628
1629		if (!n)
1630			break;
1631
1632		mutex_unlock(&syslog_lock);
1633		err = copy_to_user(buf, text + skip, n);
1634		mutex_lock(&syslog_lock);
1635
1636		if (err) {
1637			if (!len)
1638				len = -EFAULT;
1639			break;
1640		}
1641
1642		len += n;
1643		size -= n;
1644		buf += n;
1645	} while (size);
1646out:
1647	mutex_unlock(&syslog_lock);
1648	kfree(text);
1649	return len;
1650}
1651
1652static int syslog_print_all(char __user *buf, int size, bool clear)
1653{
1654	struct printk_info info;
1655	struct printk_record r;
1656	char *text;
1657	int len = 0;
1658	u64 seq;
1659	bool time;
1660
1661	text = kmalloc(PRINTK_MESSAGE_MAX, GFP_KERNEL);
1662	if (!text)
1663		return -ENOMEM;
1664
1665	time = printk_time;
1666	/*
1667	 * Find first record that fits, including all following records,
1668	 * into the user-provided buffer for this dump.
1669	 */
1670	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1671				     size, true, time);
1672
1673	prb_rec_init_rd(&r, &info, text, PRINTK_MESSAGE_MAX);
1674
 
1675	prb_for_each_record(seq, prb, seq, &r) {
1676		int textlen;
1677
1678		textlen = record_print_text(&r, true, time);
1679
1680		if (len + textlen > size) {
1681			seq--;
1682			break;
1683		}
1684
1685		if (copy_to_user(buf + len, text, textlen))
1686			len = -EFAULT;
1687		else
1688			len += textlen;
1689
1690		if (len < 0)
1691			break;
1692	}
1693
1694	if (clear) {
1695		mutex_lock(&syslog_lock);
1696		latched_seq_write(&clear_seq, seq);
1697		mutex_unlock(&syslog_lock);
1698	}
1699
1700	kfree(text);
1701	return len;
1702}
1703
1704static void syslog_clear(void)
1705{
1706	mutex_lock(&syslog_lock);
1707	latched_seq_write(&clear_seq, prb_next_seq(prb));
1708	mutex_unlock(&syslog_lock);
1709}
1710
1711int do_syslog(int type, char __user *buf, int len, int source)
1712{
1713	struct printk_info info;
1714	bool clear = false;
1715	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1716	int error;
1717
1718	error = check_syslog_permissions(type, source);
1719	if (error)
1720		return error;
1721
1722	switch (type) {
1723	case SYSLOG_ACTION_CLOSE:	/* Close log */
1724		break;
1725	case SYSLOG_ACTION_OPEN:	/* Open log */
1726		break;
1727	case SYSLOG_ACTION_READ:	/* Read from log */
1728		if (!buf || len < 0)
1729			return -EINVAL;
1730		if (!len)
1731			return 0;
1732		if (!access_ok(buf, len))
1733			return -EFAULT;
1734		error = syslog_print(buf, len);
1735		break;
1736	/* Read/clear last kernel messages */
1737	case SYSLOG_ACTION_READ_CLEAR:
1738		clear = true;
1739		fallthrough;
1740	/* Read last kernel messages */
1741	case SYSLOG_ACTION_READ_ALL:
1742		if (!buf || len < 0)
1743			return -EINVAL;
1744		if (!len)
1745			return 0;
1746		if (!access_ok(buf, len))
1747			return -EFAULT;
1748		error = syslog_print_all(buf, len, clear);
1749		break;
1750	/* Clear ring buffer */
1751	case SYSLOG_ACTION_CLEAR:
1752		syslog_clear();
1753		break;
1754	/* Disable logging to console */
1755	case SYSLOG_ACTION_CONSOLE_OFF:
1756		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1757			saved_console_loglevel = console_loglevel;
1758		console_loglevel = minimum_console_loglevel;
1759		break;
1760	/* Enable logging to console */
1761	case SYSLOG_ACTION_CONSOLE_ON:
1762		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1763			console_loglevel = saved_console_loglevel;
1764			saved_console_loglevel = LOGLEVEL_DEFAULT;
1765		}
1766		break;
1767	/* Set level of messages printed to console */
1768	case SYSLOG_ACTION_CONSOLE_LEVEL:
1769		if (len < 1 || len > 8)
1770			return -EINVAL;
1771		if (len < minimum_console_loglevel)
1772			len = minimum_console_loglevel;
1773		console_loglevel = len;
1774		/* Implicitly re-enable logging to console */
1775		saved_console_loglevel = LOGLEVEL_DEFAULT;
1776		break;
1777	/* Number of chars in the log buffer */
1778	case SYSLOG_ACTION_SIZE_UNREAD:
1779		mutex_lock(&syslog_lock);
1780		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1781			/* No unread messages. */
1782			mutex_unlock(&syslog_lock);
1783			return 0;
1784		}
1785		if (info.seq != syslog_seq) {
1786			/* messages are gone, move to first one */
1787			syslog_seq = info.seq;
1788			syslog_partial = 0;
1789		}
1790		if (source == SYSLOG_FROM_PROC) {
1791			/*
1792			 * Short-cut for poll(/"proc/kmsg") which simply checks
1793			 * for pending data, not the size; return the count of
1794			 * records, not the length.
1795			 */
1796			error = prb_next_seq(prb) - syslog_seq;
1797		} else {
1798			bool time = syslog_partial ? syslog_time : printk_time;
1799			unsigned int line_count;
1800			u64 seq;
1801
1802			prb_for_each_info(syslog_seq, prb, seq, &info,
1803					  &line_count) {
1804				error += get_record_print_text_size(&info, line_count,
1805								    true, time);
1806				time = printk_time;
1807			}
1808			error -= syslog_partial;
1809		}
1810		mutex_unlock(&syslog_lock);
1811		break;
1812	/* Size of the log buffer */
1813	case SYSLOG_ACTION_SIZE_BUFFER:
1814		error = log_buf_len;
1815		break;
1816	default:
1817		error = -EINVAL;
1818		break;
1819	}
1820
1821	return error;
1822}
1823
1824SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1825{
1826	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1827}
1828
1829/*
1830 * Special console_lock variants that help to reduce the risk of soft-lockups.
1831 * They allow to pass console_lock to another printk() call using a busy wait.
1832 */
1833
1834#ifdef CONFIG_LOCKDEP
1835static struct lockdep_map console_owner_dep_map = {
1836	.name = "console_owner"
1837};
1838#endif
1839
1840static DEFINE_RAW_SPINLOCK(console_owner_lock);
1841static struct task_struct *console_owner;
1842static bool console_waiter;
1843
1844/**
1845 * console_lock_spinning_enable - mark beginning of code where another
1846 *	thread might safely busy wait
1847 *
1848 * This basically converts console_lock into a spinlock. This marks
1849 * the section where the console_lock owner can not sleep, because
1850 * there may be a waiter spinning (like a spinlock). Also it must be
1851 * ready to hand over the lock at the end of the section.
1852 */
1853static void console_lock_spinning_enable(void)
1854{
1855	/*
1856	 * Do not use spinning in panic(). The panic CPU wants to keep the lock.
1857	 * Non-panic CPUs abandon the flush anyway.
1858	 *
1859	 * Just keep the lockdep annotation. The panic-CPU should avoid
1860	 * taking console_owner_lock because it might cause a deadlock.
1861	 * This looks like the easiest way how to prevent false lockdep
1862	 * reports without handling races a lockless way.
1863	 */
1864	if (panic_in_progress())
1865		goto lockdep;
1866
1867	raw_spin_lock(&console_owner_lock);
1868	console_owner = current;
1869	raw_spin_unlock(&console_owner_lock);
1870
1871lockdep:
1872	/* The waiter may spin on us after setting console_owner */
1873	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1874}
1875
1876/**
1877 * console_lock_spinning_disable_and_check - mark end of code where another
1878 *	thread was able to busy wait and check if there is a waiter
1879 * @cookie: cookie returned from console_srcu_read_lock()
1880 *
1881 * This is called at the end of the section where spinning is allowed.
1882 * It has two functions. First, it is a signal that it is no longer
1883 * safe to start busy waiting for the lock. Second, it checks if
1884 * there is a busy waiter and passes the lock rights to her.
1885 *
1886 * Important: Callers lose both the console_lock and the SRCU read lock if
1887 *	there was a busy waiter. They must not touch items synchronized by
1888 *	console_lock or SRCU read lock in this case.
1889 *
1890 * Return: 1 if the lock rights were passed, 0 otherwise.
1891 */
1892static int console_lock_spinning_disable_and_check(int cookie)
1893{
1894	int waiter;
1895
1896	/*
1897	 * Ignore spinning waiters during panic() because they might get stopped
1898	 * or blocked at any time,
1899	 *
1900	 * It is safe because nobody is allowed to start spinning during panic
1901	 * in the first place. If there has been a waiter then non panic CPUs
1902	 * might stay spinning. They would get stopped anyway. The panic context
1903	 * will never start spinning and an interrupted spin on panic CPU will
1904	 * never continue.
1905	 */
1906	if (panic_in_progress()) {
1907		/* Keep lockdep happy. */
1908		spin_release(&console_owner_dep_map, _THIS_IP_);
1909		return 0;
1910	}
1911
1912	raw_spin_lock(&console_owner_lock);
1913	waiter = READ_ONCE(console_waiter);
1914	console_owner = NULL;
1915	raw_spin_unlock(&console_owner_lock);
1916
1917	if (!waiter) {
1918		spin_release(&console_owner_dep_map, _THIS_IP_);
1919		return 0;
1920	}
1921
1922	/* The waiter is now free to continue */
1923	WRITE_ONCE(console_waiter, false);
1924
1925	spin_release(&console_owner_dep_map, _THIS_IP_);
1926
1927	/*
1928	 * Preserve lockdep lock ordering. Release the SRCU read lock before
1929	 * releasing the console_lock.
1930	 */
1931	console_srcu_read_unlock(cookie);
1932
1933	/*
1934	 * Hand off console_lock to waiter. The waiter will perform
1935	 * the up(). After this, the waiter is the console_lock owner.
1936	 */
1937	mutex_release(&console_lock_dep_map, _THIS_IP_);
1938	return 1;
1939}
1940
1941/**
1942 * console_trylock_spinning - try to get console_lock by busy waiting
1943 *
1944 * This allows to busy wait for the console_lock when the current
1945 * owner is running in specially marked sections. It means that
1946 * the current owner is running and cannot reschedule until it
1947 * is ready to lose the lock.
1948 *
1949 * Return: 1 if we got the lock, 0 othrewise
1950 */
1951static int console_trylock_spinning(void)
1952{
1953	struct task_struct *owner = NULL;
1954	bool waiter;
1955	bool spin = false;
1956	unsigned long flags;
1957
1958	if (console_trylock())
1959		return 1;
1960
1961	/*
1962	 * It's unsafe to spin once a panic has begun. If we are the
1963	 * panic CPU, we may have already halted the owner of the
1964	 * console_sem. If we are not the panic CPU, then we should
1965	 * avoid taking console_sem, so the panic CPU has a better
1966	 * chance of cleanly acquiring it later.
1967	 */
1968	if (panic_in_progress())
1969		return 0;
1970
1971	printk_safe_enter_irqsave(flags);
1972
1973	raw_spin_lock(&console_owner_lock);
1974	owner = READ_ONCE(console_owner);
1975	waiter = READ_ONCE(console_waiter);
1976	if (!waiter && owner && owner != current) {
1977		WRITE_ONCE(console_waiter, true);
1978		spin = true;
1979	}
1980	raw_spin_unlock(&console_owner_lock);
1981
1982	/*
1983	 * If there is an active printk() writing to the
1984	 * consoles, instead of having it write our data too,
1985	 * see if we can offload that load from the active
1986	 * printer, and do some printing ourselves.
1987	 * Go into a spin only if there isn't already a waiter
1988	 * spinning, and there is an active printer, and
1989	 * that active printer isn't us (recursive printk?).
1990	 */
1991	if (!spin) {
1992		printk_safe_exit_irqrestore(flags);
1993		return 0;
1994	}
1995
1996	/* We spin waiting for the owner to release us */
1997	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1998	/* Owner will clear console_waiter on hand off */
1999	while (READ_ONCE(console_waiter))
2000		cpu_relax();
2001	spin_release(&console_owner_dep_map, _THIS_IP_);
2002
2003	printk_safe_exit_irqrestore(flags);
2004	/*
2005	 * The owner passed the console lock to us.
2006	 * Since we did not spin on console lock, annotate
2007	 * this as a trylock. Otherwise lockdep will
2008	 * complain.
2009	 */
2010	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
2011
2012	/*
2013	 * Update @console_may_schedule for trylock because the previous
2014	 * owner may have been schedulable.
2015	 */
2016	console_may_schedule = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2017
2018	return 1;
2019}
2020
2021/*
2022 * Recursion is tracked separately on each CPU. If NMIs are supported, an
2023 * additional NMI context per CPU is also separately tracked. Until per-CPU
2024 * is available, a separate "early tracking" is performed.
2025 */
2026static DEFINE_PER_CPU(u8, printk_count);
2027static u8 printk_count_early;
2028#ifdef CONFIG_HAVE_NMI
2029static DEFINE_PER_CPU(u8, printk_count_nmi);
2030static u8 printk_count_nmi_early;
2031#endif
2032
2033/*
2034 * Recursion is limited to keep the output sane. printk() should not require
2035 * more than 1 level of recursion (allowing, for example, printk() to trigger
2036 * a WARN), but a higher value is used in case some printk-internal errors
2037 * exist, such as the ringbuffer validation checks failing.
2038 */
2039#define PRINTK_MAX_RECURSION 3
2040
2041/*
2042 * Return a pointer to the dedicated counter for the CPU+context of the
2043 * caller.
2044 */
2045static u8 *__printk_recursion_counter(void)
2046{
2047#ifdef CONFIG_HAVE_NMI
2048	if (in_nmi()) {
2049		if (printk_percpu_data_ready())
2050			return this_cpu_ptr(&printk_count_nmi);
2051		return &printk_count_nmi_early;
2052	}
2053#endif
2054	if (printk_percpu_data_ready())
2055		return this_cpu_ptr(&printk_count);
2056	return &printk_count_early;
2057}
2058
2059/*
2060 * Enter recursion tracking. Interrupts are disabled to simplify tracking.
2061 * The caller must check the boolean return value to see if the recursion is
2062 * allowed. On failure, interrupts are not disabled.
2063 *
2064 * @recursion_ptr must be a variable of type (u8 *) and is the same variable
2065 * that is passed to printk_exit_irqrestore().
2066 */
2067#define printk_enter_irqsave(recursion_ptr, flags)	\
2068({							\
2069	bool success = true;				\
2070							\
2071	typecheck(u8 *, recursion_ptr);			\
2072	local_irq_save(flags);				\
2073	(recursion_ptr) = __printk_recursion_counter();	\
2074	if (*(recursion_ptr) > PRINTK_MAX_RECURSION) {	\
2075		local_irq_restore(flags);		\
2076		success = false;			\
2077	} else {					\
2078		(*(recursion_ptr))++;			\
2079	}						\
2080	success;					\
2081})
2082
2083/* Exit recursion tracking, restoring interrupts. */
2084#define printk_exit_irqrestore(recursion_ptr, flags)	\
2085	do {						\
2086		typecheck(u8 *, recursion_ptr);		\
2087		(*(recursion_ptr))--;			\
2088		local_irq_restore(flags);		\
2089	} while (0)
2090
2091int printk_delay_msec __read_mostly;
2092
2093static inline void printk_delay(int level)
2094{
2095	boot_delay_msec(level);
2096
2097	if (unlikely(printk_delay_msec)) {
2098		int m = printk_delay_msec;
2099
2100		while (m--) {
2101			mdelay(1);
2102			touch_nmi_watchdog();
2103		}
2104	}
2105}
2106
2107static inline u32 printk_caller_id(void)
2108{
2109	return in_task() ? task_pid_nr(current) :
2110		0x80000000 + smp_processor_id();
2111}
2112
2113/**
2114 * printk_parse_prefix - Parse level and control flags.
2115 *
2116 * @text:     The terminated text message.
2117 * @level:    A pointer to the current level value, will be updated.
2118 * @flags:    A pointer to the current printk_info flags, will be updated.
2119 *
2120 * @level may be NULL if the caller is not interested in the parsed value.
2121 * Otherwise the variable pointed to by @level must be set to
2122 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
2123 *
2124 * @flags may be NULL if the caller is not interested in the parsed value.
2125 * Otherwise the variable pointed to by @flags will be OR'd with the parsed
2126 * value.
2127 *
2128 * Return: The length of the parsed level and control flags.
2129 */
2130u16 printk_parse_prefix(const char *text, int *level,
2131			enum printk_info_flags *flags)
2132{
2133	u16 prefix_len = 0;
2134	int kern_level;
2135
2136	while (*text) {
2137		kern_level = printk_get_level(text);
2138		if (!kern_level)
2139			break;
2140
2141		switch (kern_level) {
2142		case '0' ... '7':
2143			if (level && *level == LOGLEVEL_DEFAULT)
2144				*level = kern_level - '0';
2145			break;
2146		case 'c':	/* KERN_CONT */
2147			if (flags)
2148				*flags |= LOG_CONT;
2149		}
2150
2151		prefix_len += 2;
2152		text += 2;
2153	}
2154
2155	return prefix_len;
2156}
2157
2158__printf(5, 0)
2159static u16 printk_sprint(char *text, u16 size, int facility,
2160			 enum printk_info_flags *flags, const char *fmt,
2161			 va_list args)
2162{
2163	u16 text_len;
2164
2165	text_len = vscnprintf(text, size, fmt, args);
2166
2167	/* Mark and strip a trailing newline. */
2168	if (text_len && text[text_len - 1] == '\n') {
2169		text_len--;
2170		*flags |= LOG_NEWLINE;
2171	}
2172
2173	/* Strip log level and control flags. */
2174	if (facility == 0) {
2175		u16 prefix_len;
2176
2177		prefix_len = printk_parse_prefix(text, NULL, NULL);
2178		if (prefix_len) {
2179			text_len -= prefix_len;
2180			memmove(text, text + prefix_len, text_len);
2181		}
2182	}
2183
2184	trace_console(text, text_len);
2185
2186	return text_len;
2187}
2188
2189__printf(4, 0)
2190int vprintk_store(int facility, int level,
2191		  const struct dev_printk_info *dev_info,
2192		  const char *fmt, va_list args)
2193{
2194	struct prb_reserved_entry e;
2195	enum printk_info_flags flags = 0;
2196	struct printk_record r;
2197	unsigned long irqflags;
2198	u16 trunc_msg_len = 0;
2199	char prefix_buf[8];
2200	u8 *recursion_ptr;
2201	u16 reserve_size;
2202	va_list args2;
2203	u32 caller_id;
2204	u16 text_len;
2205	int ret = 0;
2206	u64 ts_nsec;
2207
2208	if (!printk_enter_irqsave(recursion_ptr, irqflags))
2209		return 0;
2210
2211	/*
2212	 * Since the duration of printk() can vary depending on the message
2213	 * and state of the ringbuffer, grab the timestamp now so that it is
2214	 * close to the call of printk(). This provides a more deterministic
2215	 * timestamp with respect to the caller.
2216	 */
2217	ts_nsec = local_clock();
2218
2219	caller_id = printk_caller_id();
2220
2221	/*
2222	 * The sprintf needs to come first since the syslog prefix might be
2223	 * passed in as a parameter. An extra byte must be reserved so that
2224	 * later the vscnprintf() into the reserved buffer has room for the
2225	 * terminating '\0', which is not counted by vsnprintf().
2226	 */
2227	va_copy(args2, args);
2228	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2229	va_end(args2);
2230
2231	if (reserve_size > PRINTKRB_RECORD_MAX)
2232		reserve_size = PRINTKRB_RECORD_MAX;
2233
2234	/* Extract log level or control flags. */
2235	if (facility == 0)
2236		printk_parse_prefix(&prefix_buf[0], &level, &flags);
2237
2238	if (level == LOGLEVEL_DEFAULT)
2239		level = default_message_loglevel;
2240
2241	if (dev_info)
2242		flags |= LOG_NEWLINE;
2243
2244	if (flags & LOG_CONT) {
2245		prb_rec_init_wr(&r, reserve_size);
2246		if (prb_reserve_in_last(&e, prb, &r, caller_id, PRINTKRB_RECORD_MAX)) {
2247			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2248						 facility, &flags, fmt, args);
2249			r.info->text_len += text_len;
2250
2251			if (flags & LOG_NEWLINE) {
2252				r.info->flags |= LOG_NEWLINE;
2253				prb_final_commit(&e);
2254			} else {
2255				prb_commit(&e);
2256			}
2257
2258			ret = text_len;
2259			goto out;
2260		}
2261	}
2262
2263	/*
2264	 * Explicitly initialize the record before every prb_reserve() call.
2265	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2266	 * structure when they fail.
2267	 */
2268	prb_rec_init_wr(&r, reserve_size);
2269	if (!prb_reserve(&e, prb, &r)) {
2270		/* truncate the message if it is too long for empty buffer */
2271		truncate_msg(&reserve_size, &trunc_msg_len);
2272
2273		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2274		if (!prb_reserve(&e, prb, &r))
2275			goto out;
2276	}
2277
2278	/* fill message */
2279	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &flags, fmt, args);
2280	if (trunc_msg_len)
2281		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2282	r.info->text_len = text_len + trunc_msg_len;
2283	r.info->facility = facility;
2284	r.info->level = level & 7;
2285	r.info->flags = flags & 0x1f;
2286	r.info->ts_nsec = ts_nsec;
2287	r.info->caller_id = caller_id;
2288	if (dev_info)
2289		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2290
2291	/* A message without a trailing newline can be continued. */
2292	if (!(flags & LOG_NEWLINE))
2293		prb_commit(&e);
2294	else
2295		prb_final_commit(&e);
2296
2297	ret = text_len + trunc_msg_len;
2298out:
2299	printk_exit_irqrestore(recursion_ptr, irqflags);
2300	return ret;
2301}
2302
2303asmlinkage int vprintk_emit(int facility, int level,
2304			    const struct dev_printk_info *dev_info,
2305			    const char *fmt, va_list args)
2306{
2307	int printed_len;
2308	bool in_sched = false;
2309
2310	/* Suppress unimportant messages after panic happens */
2311	if (unlikely(suppress_printk))
2312		return 0;
2313
2314	/*
2315	 * The messages on the panic CPU are the most important. If
2316	 * non-panic CPUs are generating any messages, they will be
2317	 * silently dropped.
2318	 */
2319	if (other_cpu_in_panic())
2320		return 0;
2321
2322	if (level == LOGLEVEL_SCHED) {
2323		level = LOGLEVEL_DEFAULT;
2324		in_sched = true;
2325	}
2326
2327	printk_delay(level);
2328
2329	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2330
2331	/* If called from the scheduler, we can not call up(). */
2332	if (!in_sched) {
2333		/*
2334		 * The caller may be holding system-critical or
2335		 * timing-sensitive locks. Disable preemption during
2336		 * printing of all remaining records to all consoles so that
2337		 * this context can return as soon as possible. Hopefully
2338		 * another printk() caller will take over the printing.
2339		 */
2340		preempt_disable();
2341		/*
2342		 * Try to acquire and then immediately release the console
2343		 * semaphore. The release will print out buffers. With the
2344		 * spinning variant, this context tries to take over the
2345		 * printing from another printing context.
2346		 */
2347		if (console_trylock_spinning())
2348			console_unlock();
2349		preempt_enable();
2350	}
2351
2352	if (in_sched)
2353		defer_console_output();
2354	else
2355		wake_up_klogd();
2356
2357	return printed_len;
2358}
2359EXPORT_SYMBOL(vprintk_emit);
2360
2361int vprintk_default(const char *fmt, va_list args)
2362{
2363	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2364}
2365EXPORT_SYMBOL_GPL(vprintk_default);
2366
2367asmlinkage __visible int _printk(const char *fmt, ...)
2368{
2369	va_list args;
2370	int r;
2371
2372	va_start(args, fmt);
2373	r = vprintk(fmt, args);
2374	va_end(args);
2375
2376	return r;
2377}
2378EXPORT_SYMBOL(_printk);
2379
2380static bool pr_flush(int timeout_ms, bool reset_on_progress);
2381static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress);
2382
2383#else /* CONFIG_PRINTK */
2384
 
 
2385#define printk_time		false
2386
2387#define prb_read_valid(rb, seq, r)	false
2388#define prb_first_valid_seq(rb)		0
2389#define prb_next_seq(rb)		0
2390
2391static u64 syslog_seq;
2392
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2393static bool pr_flush(int timeout_ms, bool reset_on_progress) { return true; }
2394static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress) { return true; }
2395
2396#endif /* CONFIG_PRINTK */
2397
2398#ifdef CONFIG_EARLY_PRINTK
2399struct console *early_console;
2400
2401asmlinkage __visible void early_printk(const char *fmt, ...)
2402{
2403	va_list ap;
2404	char buf[512];
2405	int n;
2406
2407	if (!early_console)
2408		return;
2409
2410	va_start(ap, fmt);
2411	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2412	va_end(ap);
2413
2414	early_console->write(early_console, buf, n);
2415}
2416#endif
2417
2418static void set_user_specified(struct console_cmdline *c, bool user_specified)
2419{
2420	if (!user_specified)
2421		return;
2422
2423	/*
2424	 * @c console was defined by the user on the command line.
2425	 * Do not clear when added twice also by SPCR or the device tree.
2426	 */
2427	c->user_specified = true;
2428	/* At least one console defined by the user on the command line. */
2429	console_set_on_cmdline = 1;
2430}
2431
2432static int __add_preferred_console(const char *name, const short idx, char *options,
2433				   char *brl_options, bool user_specified)
2434{
2435	struct console_cmdline *c;
2436	int i;
2437
2438	/*
2439	 * We use a signed short index for struct console for device drivers to
2440	 * indicate a not yet assigned index or port. However, a negative index
2441	 * value is not valid for preferred console.
2442	 */
2443	if (idx < 0)
2444		return -EINVAL;
2445
2446	/*
2447	 *	See if this tty is not yet registered, and
2448	 *	if we have a slot free.
2449	 */
2450	for (i = 0, c = console_cmdline;
2451	     i < MAX_CMDLINECONSOLES && c->name[0];
2452	     i++, c++) {
2453		if (strcmp(c->name, name) == 0 && c->index == idx) {
2454			if (!brl_options)
2455				preferred_console = i;
2456			set_user_specified(c, user_specified);
2457			return 0;
2458		}
2459	}
2460	if (i == MAX_CMDLINECONSOLES)
2461		return -E2BIG;
2462	if (!brl_options)
2463		preferred_console = i;
2464	strscpy(c->name, name, sizeof(c->name));
2465	c->options = options;
2466	set_user_specified(c, user_specified);
2467	braille_set_options(c, brl_options);
2468
2469	c->index = idx;
2470	return 0;
2471}
2472
2473static int __init console_msg_format_setup(char *str)
2474{
2475	if (!strcmp(str, "syslog"))
2476		console_msg_format = MSG_FORMAT_SYSLOG;
2477	if (!strcmp(str, "default"))
2478		console_msg_format = MSG_FORMAT_DEFAULT;
2479	return 1;
2480}
2481__setup("console_msg_format=", console_msg_format_setup);
2482
2483/*
2484 * Set up a console.  Called via do_early_param() in init/main.c
2485 * for each "console=" parameter in the boot command line.
2486 */
2487static int __init console_setup(char *str)
2488{
2489	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2490	char *s, *options, *brl_options = NULL;
2491	int idx;
2492
2493	/*
2494	 * console="" or console=null have been suggested as a way to
2495	 * disable console output. Use ttynull that has been created
2496	 * for exactly this purpose.
2497	 */
2498	if (str[0] == 0 || strcmp(str, "null") == 0) {
2499		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2500		return 1;
2501	}
2502
2503	if (_braille_console_setup(&str, &brl_options))
2504		return 1;
2505
2506	/*
2507	 * Decode str into name, index, options.
2508	 */
2509	if (str[0] >= '0' && str[0] <= '9') {
2510		strcpy(buf, "ttyS");
2511		strncpy(buf + 4, str, sizeof(buf) - 5);
2512	} else {
2513		strncpy(buf, str, sizeof(buf) - 1);
2514	}
2515	buf[sizeof(buf) - 1] = 0;
2516	options = strchr(str, ',');
2517	if (options)
2518		*(options++) = 0;
2519#ifdef __sparc__
2520	if (!strcmp(str, "ttya"))
2521		strcpy(buf, "ttyS0");
2522	if (!strcmp(str, "ttyb"))
2523		strcpy(buf, "ttyS1");
2524#endif
2525	for (s = buf; *s; s++)
2526		if (isdigit(*s) || *s == ',')
2527			break;
2528	idx = simple_strtoul(s, NULL, 10);
2529	*s = 0;
2530
2531	__add_preferred_console(buf, idx, options, brl_options, true);
2532	return 1;
2533}
2534__setup("console=", console_setup);
2535
2536/**
2537 * add_preferred_console - add a device to the list of preferred consoles.
2538 * @name: device name
2539 * @idx: device index
2540 * @options: options for this console
2541 *
2542 * The last preferred console added will be used for kernel messages
2543 * and stdin/out/err for init.  Normally this is used by console_setup
2544 * above to handle user-supplied console arguments; however it can also
2545 * be used by arch-specific code either to override the user or more
2546 * commonly to provide a default console (ie from PROM variables) when
2547 * the user has not supplied one.
2548 */
2549int add_preferred_console(const char *name, const short idx, char *options)
2550{
2551	return __add_preferred_console(name, idx, options, NULL, false);
2552}
2553
2554bool console_suspend_enabled = true;
2555EXPORT_SYMBOL(console_suspend_enabled);
2556
2557static int __init console_suspend_disable(char *str)
2558{
2559	console_suspend_enabled = false;
2560	return 1;
2561}
2562__setup("no_console_suspend", console_suspend_disable);
2563module_param_named(console_suspend, console_suspend_enabled,
2564		bool, S_IRUGO | S_IWUSR);
2565MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2566	" and hibernate operations");
2567
2568static bool printk_console_no_auto_verbose;
2569
2570void console_verbose(void)
2571{
2572	if (console_loglevel && !printk_console_no_auto_verbose)
2573		console_loglevel = CONSOLE_LOGLEVEL_MOTORMOUTH;
2574}
2575EXPORT_SYMBOL_GPL(console_verbose);
2576
2577module_param_named(console_no_auto_verbose, printk_console_no_auto_verbose, bool, 0644);
2578MODULE_PARM_DESC(console_no_auto_verbose, "Disable console loglevel raise to highest on oops/panic/etc");
2579
2580/**
2581 * suspend_console - suspend the console subsystem
2582 *
2583 * This disables printk() while we go into suspend states
2584 */
2585void suspend_console(void)
2586{
2587	struct console *con;
2588
2589	if (!console_suspend_enabled)
2590		return;
2591	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2592	pr_flush(1000, true);
2593
2594	console_list_lock();
2595	for_each_console(con)
2596		console_srcu_write_flags(con, con->flags | CON_SUSPENDED);
2597	console_list_unlock();
2598
2599	/*
2600	 * Ensure that all SRCU list walks have completed. All printing
2601	 * contexts must be able to see that they are suspended so that it
2602	 * is guaranteed that all printing has stopped when this function
2603	 * completes.
2604	 */
2605	synchronize_srcu(&console_srcu);
2606}
2607
2608void resume_console(void)
2609{
2610	struct console *con;
2611
2612	if (!console_suspend_enabled)
2613		return;
2614
2615	console_list_lock();
2616	for_each_console(con)
2617		console_srcu_write_flags(con, con->flags & ~CON_SUSPENDED);
2618	console_list_unlock();
2619
2620	/*
2621	 * Ensure that all SRCU list walks have completed. All printing
2622	 * contexts must be able to see they are no longer suspended so
2623	 * that they are guaranteed to wake up and resume printing.
2624	 */
2625	synchronize_srcu(&console_srcu);
2626
2627	pr_flush(1000, true);
2628}
2629
2630/**
2631 * console_cpu_notify - print deferred console messages after CPU hotplug
2632 * @cpu: unused
2633 *
2634 * If printk() is called from a CPU that is not online yet, the messages
2635 * will be printed on the console only if there are CON_ANYTIME consoles.
2636 * This function is called when a new CPU comes online (or fails to come
2637 * up) or goes offline.
2638 */
2639static int console_cpu_notify(unsigned int cpu)
2640{
2641	if (!cpuhp_tasks_frozen) {
2642		/* If trylock fails, someone else is doing the printing */
2643		if (console_trylock())
2644			console_unlock();
2645	}
2646	return 0;
2647}
2648
2649/**
2650 * console_lock - block the console subsystem from printing
2651 *
2652 * Acquires a lock which guarantees that no consoles will
2653 * be in or enter their write() callback.
2654 *
2655 * Can sleep, returns nothing.
2656 */
2657void console_lock(void)
2658{
2659	might_sleep();
2660
2661	/* On panic, the console_lock must be left to the panic cpu. */
2662	while (other_cpu_in_panic())
2663		msleep(1000);
2664
2665	down_console_sem();
 
 
2666	console_locked = 1;
2667	console_may_schedule = 1;
2668}
2669EXPORT_SYMBOL(console_lock);
2670
2671/**
2672 * console_trylock - try to block the console subsystem from printing
2673 *
2674 * Try to acquire a lock which guarantees that no consoles will
2675 * be in or enter their write() callback.
2676 *
2677 * returns 1 on success, and 0 on failure to acquire the lock.
2678 */
2679int console_trylock(void)
2680{
2681	/* On panic, the console_lock must be left to the panic cpu. */
2682	if (other_cpu_in_panic())
2683		return 0;
2684	if (down_trylock_console_sem())
 
2685		return 0;
 
2686	console_locked = 1;
2687	console_may_schedule = 0;
2688	return 1;
2689}
2690EXPORT_SYMBOL(console_trylock);
2691
2692int is_console_locked(void)
2693{
2694	return console_locked;
2695}
2696EXPORT_SYMBOL(is_console_locked);
2697
2698/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2699 * Check if the given console is currently capable and allowed to print
2700 * records.
2701 *
2702 * Requires the console_srcu_read_lock.
2703 */
2704static inline bool console_is_usable(struct console *con)
2705{
2706	short flags = console_srcu_read_flags(con);
2707
2708	if (!(flags & CON_ENABLED))
2709		return false;
2710
2711	if ((flags & CON_SUSPENDED))
2712		return false;
2713
2714	if (!con->write)
2715		return false;
2716
2717	/*
2718	 * Console drivers may assume that per-cpu resources have been
2719	 * allocated. So unless they're explicitly marked as being able to
2720	 * cope (CON_ANYTIME) don't call them until this CPU is officially up.
2721	 */
2722	if (!cpu_online(raw_smp_processor_id()) && !(flags & CON_ANYTIME))
2723		return false;
2724
2725	return true;
2726}
2727
2728static void __console_unlock(void)
2729{
2730	console_locked = 0;
2731	up_console_sem();
2732}
2733
2734#ifdef CONFIG_PRINTK
2735
2736/*
2737 * Prepend the message in @pmsg->pbufs->outbuf with a "dropped message". This
2738 * is achieved by shifting the existing message over and inserting the dropped
2739 * message.
2740 *
2741 * @pmsg is the printk message to prepend.
2742 *
2743 * @dropped is the dropped count to report in the dropped message.
2744 *
2745 * If the message text in @pmsg->pbufs->outbuf does not have enough space for
2746 * the dropped message, the message text will be sufficiently truncated.
2747 *
2748 * If @pmsg->pbufs->outbuf is modified, @pmsg->outbuf_len is updated.
2749 */
2750void console_prepend_dropped(struct printk_message *pmsg, unsigned long dropped)
2751{
2752	struct printk_buffers *pbufs = pmsg->pbufs;
2753	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2754	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2755	char *scratchbuf = &pbufs->scratchbuf[0];
2756	char *outbuf = &pbufs->outbuf[0];
2757	size_t len;
2758
2759	len = scnprintf(scratchbuf, scratchbuf_sz,
2760		       "** %lu printk messages dropped **\n", dropped);
2761
2762	/*
2763	 * Make sure outbuf is sufficiently large before prepending.
2764	 * Keep at least the prefix when the message must be truncated.
2765	 * It is a rather theoretical problem when someone tries to
2766	 * use a minimalist buffer.
2767	 */
2768	if (WARN_ON_ONCE(len + PRINTK_PREFIX_MAX >= outbuf_sz))
2769		return;
2770
2771	if (pmsg->outbuf_len + len >= outbuf_sz) {
2772		/* Truncate the message, but keep it terminated. */
2773		pmsg->outbuf_len = outbuf_sz - (len + 1);
2774		outbuf[pmsg->outbuf_len] = 0;
2775	}
2776
2777	memmove(outbuf + len, outbuf, pmsg->outbuf_len + 1);
2778	memcpy(outbuf, scratchbuf, len);
2779	pmsg->outbuf_len += len;
2780}
2781
2782/*
2783 * Read and format the specified record (or a later record if the specified
2784 * record is not available).
2785 *
2786 * @pmsg will contain the formatted result. @pmsg->pbufs must point to a
2787 * struct printk_buffers.
2788 *
2789 * @seq is the record to read and format. If it is not available, the next
2790 * valid record is read.
2791 *
2792 * @is_extended specifies if the message should be formatted for extended
2793 * console output.
2794 *
2795 * @may_supress specifies if records may be skipped based on loglevel.
2796 *
2797 * Returns false if no record is available. Otherwise true and all fields
2798 * of @pmsg are valid. (See the documentation of struct printk_message
2799 * for information about the @pmsg fields.)
2800 */
2801bool printk_get_next_message(struct printk_message *pmsg, u64 seq,
2802			     bool is_extended, bool may_suppress)
2803{
2804	struct printk_buffers *pbufs = pmsg->pbufs;
2805	const size_t scratchbuf_sz = sizeof(pbufs->scratchbuf);
2806	const size_t outbuf_sz = sizeof(pbufs->outbuf);
2807	char *scratchbuf = &pbufs->scratchbuf[0];
2808	char *outbuf = &pbufs->outbuf[0];
2809	struct printk_info info;
2810	struct printk_record r;
2811	size_t len = 0;
2812
2813	/*
2814	 * Formatting extended messages requires a separate buffer, so use the
2815	 * scratch buffer to read in the ringbuffer text.
2816	 *
2817	 * Formatting normal messages is done in-place, so read the ringbuffer
2818	 * text directly into the output buffer.
2819	 */
2820	if (is_extended)
2821		prb_rec_init_rd(&r, &info, scratchbuf, scratchbuf_sz);
2822	else
2823		prb_rec_init_rd(&r, &info, outbuf, outbuf_sz);
2824
2825	if (!prb_read_valid(prb, seq, &r))
2826		return false;
2827
2828	pmsg->seq = r.info->seq;
2829	pmsg->dropped = r.info->seq - seq;
2830
2831	/* Skip record that has level above the console loglevel. */
2832	if (may_suppress && suppress_message_printing(r.info->level))
2833		goto out;
2834
2835	if (is_extended) {
2836		len = info_print_ext_header(outbuf, outbuf_sz, r.info);
2837		len += msg_print_ext_body(outbuf + len, outbuf_sz - len,
2838					  &r.text_buf[0], r.info->text_len, &r.info->dev_info);
2839	} else {
2840		len = record_print_text(&r, console_msg_format & MSG_FORMAT_SYSLOG, printk_time);
2841	}
2842out:
2843	pmsg->outbuf_len = len;
2844	return true;
2845}
2846
2847/*
2848 * Used as the printk buffers for non-panic, serialized console printing.
2849 * This is for legacy (!CON_NBCON) as well as all boot (CON_BOOT) consoles.
2850 * Its usage requires the console_lock held.
2851 */
2852struct printk_buffers printk_shared_pbufs;
2853
2854/*
2855 * Print one record for the given console. The record printed is whatever
2856 * record is the next available record for the given console.
2857 *
2858 * @handover will be set to true if a printk waiter has taken over the
2859 * console_lock, in which case the caller is no longer holding both the
2860 * console_lock and the SRCU read lock. Otherwise it is set to false.
2861 *
2862 * @cookie is the cookie from the SRCU read lock.
2863 *
2864 * Returns false if the given console has no next record to print, otherwise
2865 * true.
2866 *
2867 * Requires the console_lock and the SRCU read lock.
2868 */
2869static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
 
2870{
2871	bool is_extended = console_srcu_read_flags(con) & CON_EXTENDED;
2872	char *outbuf = &printk_shared_pbufs.outbuf[0];
2873	struct printk_message pmsg = {
2874		.pbufs = &printk_shared_pbufs,
2875	};
2876	unsigned long flags;
 
 
 
 
2877
2878	*handover = false;
2879
2880	if (!printk_get_next_message(&pmsg, con->seq, is_extended, true))
2881		return false;
2882
2883	con->dropped += pmsg.dropped;
 
 
 
 
 
 
 
2884
2885	/* Skip messages of formatted length 0. */
2886	if (pmsg.outbuf_len == 0) {
2887		con->seq = pmsg.seq + 1;
2888		goto skip;
2889	}
2890
2891	if (con->dropped && !is_extended) {
2892		console_prepend_dropped(&pmsg, con->dropped);
2893		con->dropped = 0;
 
 
 
 
 
2894	}
2895
2896	/*
2897	 * While actively printing out messages, if another printk()
2898	 * were to occur on another CPU, it may wait for this one to
2899	 * finish. This task can not be preempted if there is a
2900	 * waiter waiting to take over.
2901	 *
2902	 * Interrupts are disabled because the hand over to a waiter
2903	 * must not be interrupted until the hand over is completed
2904	 * (@console_waiter is cleared).
2905	 */
2906	printk_safe_enter_irqsave(flags);
2907	console_lock_spinning_enable();
2908
2909	/* Do not trace print latency. */
2910	stop_critical_timings();
2911
2912	/* Write everything out to the hardware. */
2913	con->write(con, outbuf, pmsg.outbuf_len);
2914
2915	start_critical_timings();
2916
2917	con->seq = pmsg.seq + 1;
2918
2919	*handover = console_lock_spinning_disable_and_check(cookie);
2920	printk_safe_exit_irqrestore(flags);
2921skip:
2922	return true;
2923}
2924
2925#else
2926
2927static bool console_emit_next_record(struct console *con, bool *handover, int cookie)
2928{
2929	*handover = false;
2930	return false;
2931}
2932
2933#endif /* CONFIG_PRINTK */
2934
2935/*
2936 * Print out all remaining records to all consoles.
2937 *
2938 * @do_cond_resched is set by the caller. It can be true only in schedulable
2939 * context.
2940 *
2941 * @next_seq is set to the sequence number after the last available record.
2942 * The value is valid only when this function returns true. It means that all
2943 * usable consoles are completely flushed.
2944 *
2945 * @handover will be set to true if a printk waiter has taken over the
2946 * console_lock, in which case the caller is no longer holding the
2947 * console_lock. Otherwise it is set to false.
2948 *
2949 * Returns true when there was at least one usable console and all messages
2950 * were flushed to all usable consoles. A returned false informs the caller
2951 * that everything was not flushed (either there were no usable consoles or
2952 * another context has taken over printing or it is a panic situation and this
2953 * is not the panic CPU). Regardless the reason, the caller should assume it
2954 * is not useful to immediately try again.
2955 *
2956 * Requires the console_lock.
2957 */
2958static bool console_flush_all(bool do_cond_resched, u64 *next_seq, bool *handover)
2959{
 
 
 
2960	bool any_usable = false;
2961	struct console *con;
2962	bool any_progress;
2963	int cookie;
2964
2965	*next_seq = 0;
2966	*handover = false;
2967
2968	do {
2969		any_progress = false;
2970
2971		cookie = console_srcu_read_lock();
2972		for_each_console_srcu(con) {
2973			bool progress;
2974
2975			if (!console_is_usable(con))
2976				continue;
2977			any_usable = true;
2978
2979			progress = console_emit_next_record(con, handover, cookie);
 
 
 
 
 
 
 
 
 
2980
2981			/*
2982			 * If a handover has occurred, the SRCU read lock
2983			 * is already released.
2984			 */
2985			if (*handover)
2986				return false;
2987
2988			/* Track the next of the highest seq flushed. */
2989			if (con->seq > *next_seq)
2990				*next_seq = con->seq;
2991
2992			if (!progress)
2993				continue;
2994			any_progress = true;
2995
2996			/* Allow panic_cpu to take over the consoles safely. */
2997			if (other_cpu_in_panic())
2998				goto abandon;
2999
3000			if (do_cond_resched)
3001				cond_resched();
3002		}
3003		console_srcu_read_unlock(cookie);
3004	} while (any_progress);
3005
3006	return any_usable;
3007
3008abandon:
3009	console_srcu_read_unlock(cookie);
3010	return false;
3011}
3012
3013/**
3014 * console_unlock - unblock the console subsystem from printing
3015 *
3016 * Releases the console_lock which the caller holds to block printing of
3017 * the console subsystem.
3018 *
3019 * While the console_lock was held, console output may have been buffered
3020 * by printk().  If this is the case, console_unlock(); emits
3021 * the output prior to releasing the lock.
3022 *
3023 * console_unlock(); may be called from any context.
3024 */
3025void console_unlock(void)
3026{
3027	bool do_cond_resched;
3028	bool handover;
3029	bool flushed;
3030	u64 next_seq;
3031
 
 
 
 
 
3032	/*
3033	 * Console drivers are called with interrupts disabled, so
3034	 * @console_may_schedule should be cleared before; however, we may
3035	 * end up dumping a lot of lines, for example, if called from
3036	 * console registration path, and should invoke cond_resched()
3037	 * between lines if allowable.  Not doing so can cause a very long
3038	 * scheduling stall on a slow console leading to RCU stall and
3039	 * softlockup warnings which exacerbate the issue with more
3040	 * messages practically incapacitating the system. Therefore, create
3041	 * a local to use for the printing loop.
3042	 */
3043	do_cond_resched = console_may_schedule;
3044
3045	do {
3046		console_may_schedule = 0;
3047
3048		flushed = console_flush_all(do_cond_resched, &next_seq, &handover);
3049		if (!handover)
3050			__console_unlock();
3051
3052		/*
3053		 * Abort if there was a failure to flush all messages to all
3054		 * usable consoles. Either it is not possible to flush (in
3055		 * which case it would be an infinite loop of retrying) or
3056		 * another context has taken over printing.
3057		 */
3058		if (!flushed)
3059			break;
3060
3061		/*
3062		 * Some context may have added new records after
3063		 * console_flush_all() but before unlocking the console.
3064		 * Re-check if there is a new record to flush. If the trylock
3065		 * fails, another context is already handling the printing.
3066		 */
3067	} while (prb_read_valid(prb, next_seq, NULL) && console_trylock());
3068}
3069EXPORT_SYMBOL(console_unlock);
3070
3071/**
3072 * console_conditional_schedule - yield the CPU if required
3073 *
3074 * If the console code is currently allowed to sleep, and
3075 * if this CPU should yield the CPU to another task, do
3076 * so here.
3077 *
3078 * Must be called within console_lock();.
3079 */
3080void __sched console_conditional_schedule(void)
3081{
3082	if (console_may_schedule)
3083		cond_resched();
3084}
3085EXPORT_SYMBOL(console_conditional_schedule);
3086
3087void console_unblank(void)
3088{
3089	bool found_unblank = false;
3090	struct console *c;
3091	int cookie;
3092
3093	/*
3094	 * First check if there are any consoles implementing the unblank()
3095	 * callback. If not, there is no reason to continue and take the
3096	 * console lock, which in particular can be dangerous if
3097	 * @oops_in_progress is set.
3098	 */
3099	cookie = console_srcu_read_lock();
3100	for_each_console_srcu(c) {
3101		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank) {
3102			found_unblank = true;
3103			break;
3104		}
3105	}
3106	console_srcu_read_unlock(cookie);
3107	if (!found_unblank)
3108		return;
3109
3110	/*
3111	 * Stop console printing because the unblank() callback may
3112	 * assume the console is not within its write() callback.
3113	 *
3114	 * If @oops_in_progress is set, this may be an atomic context.
3115	 * In that case, attempt a trylock as best-effort.
3116	 */
3117	if (oops_in_progress) {
3118		/* Semaphores are not NMI-safe. */
3119		if (in_nmi())
3120			return;
3121
3122		/*
3123		 * Attempting to trylock the console lock can deadlock
3124		 * if another CPU was stopped while modifying the
3125		 * semaphore. "Hope and pray" that this is not the
3126		 * current situation.
3127		 */
3128		if (down_trylock_console_sem() != 0)
3129			return;
3130	} else
3131		console_lock();
3132
3133	console_locked = 1;
3134	console_may_schedule = 0;
3135
3136	cookie = console_srcu_read_lock();
3137	for_each_console_srcu(c) {
3138		if ((console_srcu_read_flags(c) & CON_ENABLED) && c->unblank)
3139			c->unblank();
3140	}
3141	console_srcu_read_unlock(cookie);
3142
3143	console_unlock();
3144
3145	if (!oops_in_progress)
3146		pr_flush(1000, true);
3147}
3148
3149/**
3150 * console_flush_on_panic - flush console content on panic
3151 * @mode: flush all messages in buffer or just the pending ones
3152 *
3153 * Immediately output all pending messages no matter what.
3154 */
3155void console_flush_on_panic(enum con_flush_mode mode)
3156{
3157	bool handover;
3158	u64 next_seq;
3159
3160	/*
3161	 * Ignore the console lock and flush out the messages. Attempting a
3162	 * trylock would not be useful because:
3163	 *
3164	 *   - if it is contended, it must be ignored anyway
3165	 *   - console_lock() and console_trylock() block and fail
3166	 *     respectively in panic for non-panic CPUs
3167	 *   - semaphores are not NMI-safe
3168	 */
3169
3170	/*
3171	 * If another context is holding the console lock,
3172	 * @console_may_schedule might be set. Clear it so that
3173	 * this context does not call cond_resched() while flushing.
3174	 */
 
3175	console_may_schedule = 0;
3176
3177	if (mode == CONSOLE_REPLAY_ALL) {
3178		struct console *c;
3179		short flags;
3180		int cookie;
3181		u64 seq;
3182
3183		seq = prb_first_valid_seq(prb);
3184
3185		cookie = console_srcu_read_lock();
3186		for_each_console_srcu(c) {
3187			flags = console_srcu_read_flags(c);
3188
3189			if (flags & CON_NBCON) {
3190				nbcon_seq_force(c, seq);
3191			} else {
3192				/*
3193				 * This is an unsynchronized assignment. On
3194				 * panic legacy consoles are only best effort.
3195				 */
3196				c->seq = seq;
3197			}
3198		}
3199		console_srcu_read_unlock(cookie);
3200	}
3201
3202	console_flush_all(false, &next_seq, &handover);
3203}
3204
3205/*
3206 * Return the console tty driver structure and its associated index
3207 */
3208struct tty_driver *console_device(int *index)
3209{
3210	struct console *c;
3211	struct tty_driver *driver = NULL;
3212	int cookie;
3213
3214	/*
3215	 * Take console_lock to serialize device() callback with
3216	 * other console operations. For example, fg_console is
3217	 * modified under console_lock when switching vt.
3218	 */
3219	console_lock();
3220
3221	cookie = console_srcu_read_lock();
3222	for_each_console_srcu(c) {
3223		if (!c->device)
3224			continue;
3225		driver = c->device(c, index);
3226		if (driver)
3227			break;
3228	}
3229	console_srcu_read_unlock(cookie);
3230
3231	console_unlock();
3232	return driver;
3233}
3234
3235/*
3236 * Prevent further output on the passed console device so that (for example)
3237 * serial drivers can disable console output before suspending a port, and can
3238 * re-enable output afterwards.
3239 */
3240void console_stop(struct console *console)
3241{
3242	__pr_flush(console, 1000, true);
3243	console_list_lock();
3244	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3245	console_list_unlock();
3246
3247	/*
3248	 * Ensure that all SRCU list walks have completed. All contexts must
3249	 * be able to see that this console is disabled so that (for example)
3250	 * the caller can suspend the port without risk of another context
3251	 * using the port.
3252	 */
3253	synchronize_srcu(&console_srcu);
3254}
3255EXPORT_SYMBOL(console_stop);
3256
3257void console_start(struct console *console)
3258{
3259	console_list_lock();
3260	console_srcu_write_flags(console, console->flags | CON_ENABLED);
3261	console_list_unlock();
3262	__pr_flush(console, 1000, true);
3263}
3264EXPORT_SYMBOL(console_start);
3265
3266static int __read_mostly keep_bootcon;
3267
3268static int __init keep_bootcon_setup(char *str)
3269{
3270	keep_bootcon = 1;
3271	pr_info("debug: skip boot console de-registration.\n");
3272
3273	return 0;
3274}
3275
3276early_param("keep_bootcon", keep_bootcon_setup);
3277
3278static int console_call_setup(struct console *newcon, char *options)
3279{
3280	int err;
3281
3282	if (!newcon->setup)
3283		return 0;
3284
3285	/* Synchronize with possible boot console. */
3286	console_lock();
3287	err = newcon->setup(newcon, options);
3288	console_unlock();
3289
3290	return err;
3291}
3292
3293/*
3294 * This is called by register_console() to try to match
3295 * the newly registered console with any of the ones selected
3296 * by either the command line or add_preferred_console() and
3297 * setup/enable it.
3298 *
3299 * Care need to be taken with consoles that are statically
3300 * enabled such as netconsole
3301 */
3302static int try_enable_preferred_console(struct console *newcon,
3303					bool user_specified)
3304{
3305	struct console_cmdline *c;
3306	int i, err;
3307
3308	for (i = 0, c = console_cmdline;
3309	     i < MAX_CMDLINECONSOLES && c->name[0];
3310	     i++, c++) {
3311		if (c->user_specified != user_specified)
3312			continue;
3313		if (!newcon->match ||
3314		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
3315			/* default matching */
3316			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
3317			if (strcmp(c->name, newcon->name) != 0)
3318				continue;
3319			if (newcon->index >= 0 &&
3320			    newcon->index != c->index)
3321				continue;
3322			if (newcon->index < 0)
3323				newcon->index = c->index;
3324
3325			if (_braille_register_console(newcon, c))
3326				return 0;
3327
3328			err = console_call_setup(newcon, c->options);
3329			if (err)
3330				return err;
3331		}
3332		newcon->flags |= CON_ENABLED;
3333		if (i == preferred_console)
3334			newcon->flags |= CON_CONSDEV;
3335		return 0;
3336	}
3337
3338	/*
3339	 * Some consoles, such as pstore and netconsole, can be enabled even
3340	 * without matching. Accept the pre-enabled consoles only when match()
3341	 * and setup() had a chance to be called.
3342	 */
3343	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
3344		return 0;
3345
3346	return -ENOENT;
3347}
3348
3349/* Try to enable the console unconditionally */
3350static void try_enable_default_console(struct console *newcon)
3351{
3352	if (newcon->index < 0)
3353		newcon->index = 0;
3354
3355	if (console_call_setup(newcon, NULL) != 0)
3356		return;
3357
3358	newcon->flags |= CON_ENABLED;
3359
3360	if (newcon->device)
3361		newcon->flags |= CON_CONSDEV;
3362}
3363
 
 
 
 
 
3364static void console_init_seq(struct console *newcon, bool bootcon_registered)
3365{
3366	struct console *con;
3367	bool handover;
3368
3369	if (newcon->flags & (CON_PRINTBUFFER | CON_BOOT)) {
3370		/* Get a consistent copy of @syslog_seq. */
3371		mutex_lock(&syslog_lock);
3372		newcon->seq = syslog_seq;
3373		mutex_unlock(&syslog_lock);
3374	} else {
3375		/* Begin with next message added to ringbuffer. */
3376		newcon->seq = prb_next_seq(prb);
3377
3378		/*
3379		 * If any enabled boot consoles are due to be unregistered
3380		 * shortly, some may not be caught up and may be the same
3381		 * device as @newcon. Since it is not known which boot console
3382		 * is the same device, flush all consoles and, if necessary,
3383		 * start with the message of the enabled boot console that is
3384		 * the furthest behind.
3385		 */
3386		if (bootcon_registered && !keep_bootcon) {
3387			/*
3388			 * Hold the console_lock to stop console printing and
3389			 * guarantee safe access to console->seq.
3390			 */
3391			console_lock();
3392
3393			/*
3394			 * Flush all consoles and set the console to start at
3395			 * the next unprinted sequence number.
3396			 */
3397			if (!console_flush_all(true, &newcon->seq, &handover)) {
3398				/*
3399				 * Flushing failed. Just choose the lowest
3400				 * sequence of the enabled boot consoles.
3401				 */
3402
3403				/*
3404				 * If there was a handover, this context no
3405				 * longer holds the console_lock.
3406				 */
3407				if (handover)
3408					console_lock();
3409
3410				newcon->seq = prb_next_seq(prb);
3411				for_each_console(con) {
3412					if ((con->flags & CON_BOOT) &&
3413					    (con->flags & CON_ENABLED) &&
3414					    con->seq < newcon->seq) {
3415						newcon->seq = con->seq;
3416					}
3417				}
3418			}
3419
3420			console_unlock();
3421		}
3422	}
3423}
3424
3425#define console_first()				\
3426	hlist_entry(console_list.first, struct console, node)
3427
3428static int unregister_console_locked(struct console *console);
3429
3430/*
3431 * The console driver calls this routine during kernel initialization
3432 * to register the console printing procedure with printk() and to
3433 * print any messages that were printed by the kernel before the
3434 * console driver was initialized.
3435 *
3436 * This can happen pretty early during the boot process (because of
3437 * early_printk) - sometimes before setup_arch() completes - be careful
3438 * of what kernel features are used - they may not be initialised yet.
3439 *
3440 * There are two types of consoles - bootconsoles (early_printk) and
3441 * "real" consoles (everything which is not a bootconsole) which are
3442 * handled differently.
3443 *  - Any number of bootconsoles can be registered at any time.
3444 *  - As soon as a "real" console is registered, all bootconsoles
3445 *    will be unregistered automatically.
3446 *  - Once a "real" console is registered, any attempt to register a
3447 *    bootconsoles will be rejected
3448 */
3449void register_console(struct console *newcon)
3450{
3451	struct console *con;
3452	bool bootcon_registered = false;
3453	bool realcon_registered = false;
3454	int err;
3455
3456	console_list_lock();
3457
3458	for_each_console(con) {
3459		if (WARN(con == newcon, "console '%s%d' already registered\n",
3460					 con->name, con->index)) {
3461			goto unlock;
3462		}
3463
3464		if (con->flags & CON_BOOT)
3465			bootcon_registered = true;
3466		else
3467			realcon_registered = true;
3468	}
3469
3470	/* Do not register boot consoles when there already is a real one. */
3471	if ((newcon->flags & CON_BOOT) && realcon_registered) {
3472		pr_info("Too late to register bootconsole %s%d\n",
3473			newcon->name, newcon->index);
3474		goto unlock;
3475	}
3476
3477	if (newcon->flags & CON_NBCON) {
3478		/*
3479		 * Ensure the nbcon console buffers can be allocated
3480		 * before modifying any global data.
3481		 */
3482		if (!nbcon_alloc(newcon))
3483			goto unlock;
3484	}
3485
3486	/*
3487	 * See if we want to enable this console driver by default.
3488	 *
3489	 * Nope when a console is preferred by the command line, device
3490	 * tree, or SPCR.
3491	 *
3492	 * The first real console with tty binding (driver) wins. More
3493	 * consoles might get enabled before the right one is found.
3494	 *
3495	 * Note that a console with tty binding will have CON_CONSDEV
3496	 * flag set and will be first in the list.
3497	 */
3498	if (preferred_console < 0) {
3499		if (hlist_empty(&console_list) || !console_first()->device ||
3500		    console_first()->flags & CON_BOOT) {
3501			try_enable_default_console(newcon);
3502		}
3503	}
3504
3505	/* See if this console matches one we selected on the command line */
3506	err = try_enable_preferred_console(newcon, true);
3507
3508	/* If not, try to match against the platform default(s) */
3509	if (err == -ENOENT)
3510		err = try_enable_preferred_console(newcon, false);
3511
3512	/* printk() messages are not printed to the Braille console. */
3513	if (err || newcon->flags & CON_BRL) {
3514		if (newcon->flags & CON_NBCON)
3515			nbcon_free(newcon);
3516		goto unlock;
3517	}
3518
3519	/*
3520	 * If we have a bootconsole, and are switching to a real console,
3521	 * don't print everything out again, since when the boot console, and
3522	 * the real console are the same physical device, it's annoying to
3523	 * see the beginning boot messages twice
3524	 */
3525	if (bootcon_registered &&
3526	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV)) {
3527		newcon->flags &= ~CON_PRINTBUFFER;
3528	}
3529
3530	newcon->dropped = 0;
3531	console_init_seq(newcon, bootcon_registered);
3532
3533	if (newcon->flags & CON_NBCON)
3534		nbcon_init(newcon);
3535
3536	/*
3537	 * Put this console in the list - keep the
3538	 * preferred driver at the head of the list.
3539	 */
3540	if (hlist_empty(&console_list)) {
3541		/* Ensure CON_CONSDEV is always set for the head. */
3542		newcon->flags |= CON_CONSDEV;
3543		hlist_add_head_rcu(&newcon->node, &console_list);
3544
3545	} else if (newcon->flags & CON_CONSDEV) {
3546		/* Only the new head can have CON_CONSDEV set. */
3547		console_srcu_write_flags(console_first(), console_first()->flags & ~CON_CONSDEV);
3548		hlist_add_head_rcu(&newcon->node, &console_list);
3549
3550	} else {
3551		hlist_add_behind_rcu(&newcon->node, console_list.first);
3552	}
3553
3554	/*
3555	 * No need to synchronize SRCU here! The caller does not rely
3556	 * on all contexts being able to see the new console before
3557	 * register_console() completes.
3558	 */
3559
3560	console_sysfs_notify();
3561
3562	/*
3563	 * By unregistering the bootconsoles after we enable the real console
3564	 * we get the "console xxx enabled" message on all the consoles -
3565	 * boot consoles, real consoles, etc - this is to ensure that end
3566	 * users know there might be something in the kernel's log buffer that
3567	 * went to the bootconsole (that they do not see on the real console)
3568	 */
3569	con_printk(KERN_INFO, newcon, "enabled\n");
3570	if (bootcon_registered &&
3571	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
3572	    !keep_bootcon) {
3573		struct hlist_node *tmp;
3574
3575		hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3576			if (con->flags & CON_BOOT)
3577				unregister_console_locked(con);
3578		}
3579	}
3580unlock:
3581	console_list_unlock();
3582}
3583EXPORT_SYMBOL(register_console);
3584
3585/* Must be called under console_list_lock(). */
3586static int unregister_console_locked(struct console *console)
3587{
3588	int res;
3589
3590	lockdep_assert_console_list_lock_held();
3591
3592	con_printk(KERN_INFO, console, "disabled\n");
3593
3594	res = _braille_unregister_console(console);
3595	if (res < 0)
3596		return res;
3597	if (res > 0)
3598		return 0;
3599
3600	/* Disable it unconditionally */
3601	console_srcu_write_flags(console, console->flags & ~CON_ENABLED);
3602
3603	if (!console_is_registered_locked(console))
3604		return -ENODEV;
3605
3606	hlist_del_init_rcu(&console->node);
3607
3608	/*
3609	 * <HISTORICAL>
3610	 * If this isn't the last console and it has CON_CONSDEV set, we
3611	 * need to set it on the next preferred console.
3612	 * </HISTORICAL>
3613	 *
3614	 * The above makes no sense as there is no guarantee that the next
3615	 * console has any device attached. Oh well....
3616	 */
3617	if (!hlist_empty(&console_list) && console->flags & CON_CONSDEV)
3618		console_srcu_write_flags(console_first(), console_first()->flags | CON_CONSDEV);
3619
3620	/*
3621	 * Ensure that all SRCU list walks have completed. All contexts
3622	 * must not be able to see this console in the list so that any
3623	 * exit/cleanup routines can be performed safely.
3624	 */
3625	synchronize_srcu(&console_srcu);
3626
3627	if (console->flags & CON_NBCON)
3628		nbcon_free(console);
3629
3630	console_sysfs_notify();
3631
3632	if (console->exit)
3633		res = console->exit(console);
3634
3635	return res;
3636}
3637
3638int unregister_console(struct console *console)
3639{
3640	int res;
3641
3642	console_list_lock();
3643	res = unregister_console_locked(console);
3644	console_list_unlock();
3645	return res;
3646}
3647EXPORT_SYMBOL(unregister_console);
3648
3649/**
3650 * console_force_preferred_locked - force a registered console preferred
3651 * @con: The registered console to force preferred.
3652 *
3653 * Must be called under console_list_lock().
3654 */
3655void console_force_preferred_locked(struct console *con)
3656{
3657	struct console *cur_pref_con;
3658
3659	if (!console_is_registered_locked(con))
3660		return;
3661
3662	cur_pref_con = console_first();
3663
3664	/* Already preferred? */
3665	if (cur_pref_con == con)
3666		return;
3667
3668	/*
3669	 * Delete, but do not re-initialize the entry. This allows the console
3670	 * to continue to appear registered (via any hlist_unhashed_lockless()
3671	 * checks), even though it was briefly removed from the console list.
3672	 */
3673	hlist_del_rcu(&con->node);
3674
3675	/*
3676	 * Ensure that all SRCU list walks have completed so that the console
3677	 * can be added to the beginning of the console list and its forward
3678	 * list pointer can be re-initialized.
3679	 */
3680	synchronize_srcu(&console_srcu);
3681
3682	con->flags |= CON_CONSDEV;
3683	WARN_ON(!con->device);
3684
3685	/* Only the new head can have CON_CONSDEV set. */
3686	console_srcu_write_flags(cur_pref_con, cur_pref_con->flags & ~CON_CONSDEV);
3687	hlist_add_head_rcu(&con->node, &console_list);
3688}
3689EXPORT_SYMBOL(console_force_preferred_locked);
3690
3691/*
3692 * Initialize the console device. This is called *early*, so
3693 * we can't necessarily depend on lots of kernel help here.
3694 * Just do some early initializations, and do the complex setup
3695 * later.
3696 */
3697void __init console_init(void)
3698{
3699	int ret;
3700	initcall_t call;
3701	initcall_entry_t *ce;
3702
3703	/* Setup the default TTY line discipline. */
3704	n_tty_init();
3705
3706	/*
3707	 * set up the console device so that later boot sequences can
3708	 * inform about problems etc..
3709	 */
3710	ce = __con_initcall_start;
3711	trace_initcall_level("console");
3712	while (ce < __con_initcall_end) {
3713		call = initcall_from_entry(ce);
3714		trace_initcall_start(call);
3715		ret = call();
3716		trace_initcall_finish(call, ret);
3717		ce++;
3718	}
3719}
3720
3721/*
3722 * Some boot consoles access data that is in the init section and which will
3723 * be discarded after the initcalls have been run. To make sure that no code
3724 * will access this data, unregister the boot consoles in a late initcall.
3725 *
3726 * If for some reason, such as deferred probe or the driver being a loadable
3727 * module, the real console hasn't registered yet at this point, there will
3728 * be a brief interval in which no messages are logged to the console, which
3729 * makes it difficult to diagnose problems that occur during this time.
3730 *
3731 * To mitigate this problem somewhat, only unregister consoles whose memory
3732 * intersects with the init section. Note that all other boot consoles will
3733 * get unregistered when the real preferred console is registered.
3734 */
3735static int __init printk_late_init(void)
3736{
3737	struct hlist_node *tmp;
3738	struct console *con;
3739	int ret;
3740
3741	console_list_lock();
3742	hlist_for_each_entry_safe(con, tmp, &console_list, node) {
3743		if (!(con->flags & CON_BOOT))
3744			continue;
3745
3746		/* Check addresses that might be used for enabled consoles. */
3747		if (init_section_intersects(con, sizeof(*con)) ||
3748		    init_section_contains(con->write, 0) ||
3749		    init_section_contains(con->read, 0) ||
3750		    init_section_contains(con->device, 0) ||
3751		    init_section_contains(con->unblank, 0) ||
3752		    init_section_contains(con->data, 0)) {
3753			/*
3754			 * Please, consider moving the reported consoles out
3755			 * of the init section.
3756			 */
3757			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3758				con->name, con->index);
3759			unregister_console_locked(con);
3760		}
3761	}
3762	console_list_unlock();
3763
3764	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3765					console_cpu_notify);
3766	WARN_ON(ret < 0);
3767	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3768					console_cpu_notify, NULL);
3769	WARN_ON(ret < 0);
3770	printk_sysctl_init();
3771	return 0;
3772}
3773late_initcall(printk_late_init);
3774
3775#if defined CONFIG_PRINTK
3776/* If @con is specified, only wait for that console. Otherwise wait for all. */
3777static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
3778{
3779	unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
3780	unsigned long remaining_jiffies = timeout_jiffies;
3781	struct console *c;
3782	u64 last_diff = 0;
3783	u64 printk_seq;
3784	short flags;
3785	int cookie;
3786	u64 diff;
3787	u64 seq;
3788
3789	might_sleep();
3790
3791	seq = prb_next_reserve_seq(prb);
3792
3793	/* Flush the consoles so that records up to @seq are printed. */
3794	console_lock();
3795	console_unlock();
3796
3797	for (;;) {
3798		unsigned long begin_jiffies;
3799		unsigned long slept_jiffies;
3800
3801		diff = 0;
3802
3803		/*
3804		 * Hold the console_lock to guarantee safe access to
3805		 * console->seq. Releasing console_lock flushes more
3806		 * records in case @seq is still not printed on all
3807		 * usable consoles.
3808		 */
3809		console_lock();
3810
3811		cookie = console_srcu_read_lock();
3812		for_each_console_srcu(c) {
3813			if (con && con != c)
3814				continue;
3815
3816			flags = console_srcu_read_flags(c);
3817
3818			/*
3819			 * If consoles are not usable, it cannot be expected
3820			 * that they make forward progress, so only increment
3821			 * @diff for usable consoles.
3822			 */
3823			if (!console_is_usable(c))
3824				continue;
3825
3826			if (flags & CON_NBCON) {
3827				printk_seq = nbcon_seq_read(c);
3828			} else {
3829				printk_seq = c->seq;
3830			}
3831
3832			if (printk_seq < seq)
3833				diff += seq - printk_seq;
3834		}
3835		console_srcu_read_unlock(cookie);
3836
3837		if (diff != last_diff && reset_on_progress)
3838			remaining_jiffies = timeout_jiffies;
 
 
 
 
 
 
 
3839
3840		console_unlock();
3841
3842		/* Note: @diff is 0 if there are no usable consoles. */
3843		if (diff == 0 || remaining_jiffies == 0)
3844			break;
3845
3846		/* msleep(1) might sleep much longer. Check time by jiffies. */
3847		begin_jiffies = jiffies;
3848		msleep(1);
3849		slept_jiffies = jiffies - begin_jiffies;
3850
3851		remaining_jiffies -= min(slept_jiffies, remaining_jiffies);
 
 
 
 
3852
3853		last_diff = diff;
3854	}
3855
3856	return (diff == 0);
3857}
3858
3859/**
3860 * pr_flush() - Wait for printing threads to catch up.
3861 *
3862 * @timeout_ms:        The maximum time (in ms) to wait.
3863 * @reset_on_progress: Reset the timeout if forward progress is seen.
3864 *
3865 * A value of 0 for @timeout_ms means no waiting will occur. A value of -1
3866 * represents infinite waiting.
3867 *
3868 * If @reset_on_progress is true, the timeout will be reset whenever any
3869 * printer has been seen to make some forward progress.
3870 *
3871 * Context: Process context. May sleep while acquiring console lock.
3872 * Return: true if all usable printers are caught up.
3873 */
3874static bool pr_flush(int timeout_ms, bool reset_on_progress)
3875{
3876	return __pr_flush(NULL, timeout_ms, reset_on_progress);
3877}
3878
3879/*
3880 * Delayed printk version, for scheduler-internal messages:
3881 */
3882#define PRINTK_PENDING_WAKEUP	0x01
3883#define PRINTK_PENDING_OUTPUT	0x02
3884
3885static DEFINE_PER_CPU(int, printk_pending);
3886
3887static void wake_up_klogd_work_func(struct irq_work *irq_work)
3888{
3889	int pending = this_cpu_xchg(printk_pending, 0);
3890
3891	if (pending & PRINTK_PENDING_OUTPUT) {
3892		/* If trylock fails, someone else is doing the printing */
3893		if (console_trylock())
3894			console_unlock();
3895	}
3896
3897	if (pending & PRINTK_PENDING_WAKEUP)
3898		wake_up_interruptible(&log_wait);
3899}
3900
3901static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3902	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3903
3904static void __wake_up_klogd(int val)
3905{
3906	if (!printk_percpu_data_ready())
3907		return;
3908
3909	preempt_disable();
3910	/*
3911	 * Guarantee any new records can be seen by tasks preparing to wait
3912	 * before this context checks if the wait queue is empty.
3913	 *
3914	 * The full memory barrier within wq_has_sleeper() pairs with the full
3915	 * memory barrier within set_current_state() of
3916	 * prepare_to_wait_event(), which is called after ___wait_event() adds
3917	 * the waiter but before it has checked the wait condition.
3918	 *
3919	 * This pairs with devkmsg_read:A and syslog_print:A.
3920	 */
3921	if (wq_has_sleeper(&log_wait) || /* LMM(__wake_up_klogd:A) */
3922	    (val & PRINTK_PENDING_OUTPUT)) {
3923		this_cpu_or(printk_pending, val);
3924		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3925	}
3926	preempt_enable();
3927}
3928
3929/**
3930 * wake_up_klogd - Wake kernel logging daemon
3931 *
3932 * Use this function when new records have been added to the ringbuffer
3933 * and the console printing of those records has already occurred or is
3934 * known to be handled by some other context. This function will only
3935 * wake the logging daemon.
3936 *
3937 * Context: Any context.
3938 */
3939void wake_up_klogd(void)
3940{
3941	__wake_up_klogd(PRINTK_PENDING_WAKEUP);
3942}
3943
3944/**
3945 * defer_console_output - Wake kernel logging daemon and trigger
3946 *	console printing in a deferred context
3947 *
3948 * Use this function when new records have been added to the ringbuffer,
3949 * this context is responsible for console printing those records, but
3950 * the current context is not allowed to perform the console printing.
3951 * Trigger an irq_work context to perform the console printing. This
3952 * function also wakes the logging daemon.
3953 *
3954 * Context: Any context.
3955 */
3956void defer_console_output(void)
3957{
3958	/*
3959	 * New messages may have been added directly to the ringbuffer
3960	 * using vprintk_store(), so wake any waiters as well.
3961	 */
3962	__wake_up_klogd(PRINTK_PENDING_WAKEUP | PRINTK_PENDING_OUTPUT);
3963}
3964
3965void printk_trigger_flush(void)
3966{
3967	defer_console_output();
3968}
3969
3970int vprintk_deferred(const char *fmt, va_list args)
3971{
3972	return vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
 
 
 
 
 
3973}
3974
3975int _printk_deferred(const char *fmt, ...)
3976{
3977	va_list args;
3978	int r;
3979
3980	va_start(args, fmt);
3981	r = vprintk_deferred(fmt, args);
3982	va_end(args);
3983
3984	return r;
3985}
3986
3987/*
3988 * printk rate limiting, lifted from the networking subsystem.
3989 *
3990 * This enforces a rate limit: not more than 10 kernel messages
3991 * every 5s to make a denial-of-service attack impossible.
3992 */
3993DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3994
3995int __printk_ratelimit(const char *func)
3996{
3997	return ___ratelimit(&printk_ratelimit_state, func);
3998}
3999EXPORT_SYMBOL(__printk_ratelimit);
4000
4001/**
4002 * printk_timed_ratelimit - caller-controlled printk ratelimiting
4003 * @caller_jiffies: pointer to caller's state
4004 * @interval_msecs: minimum interval between prints
4005 *
4006 * printk_timed_ratelimit() returns true if more than @interval_msecs
4007 * milliseconds have elapsed since the last time printk_timed_ratelimit()
4008 * returned true.
4009 */
4010bool printk_timed_ratelimit(unsigned long *caller_jiffies,
4011			unsigned int interval_msecs)
4012{
4013	unsigned long elapsed = jiffies - *caller_jiffies;
4014
4015	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
4016		return false;
4017
4018	*caller_jiffies = jiffies;
4019	return true;
4020}
4021EXPORT_SYMBOL(printk_timed_ratelimit);
4022
4023static DEFINE_SPINLOCK(dump_list_lock);
4024static LIST_HEAD(dump_list);
4025
4026/**
4027 * kmsg_dump_register - register a kernel log dumper.
4028 * @dumper: pointer to the kmsg_dumper structure
4029 *
4030 * Adds a kernel log dumper to the system. The dump callback in the
4031 * structure will be called when the kernel oopses or panics and must be
4032 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
4033 */
4034int kmsg_dump_register(struct kmsg_dumper *dumper)
4035{
4036	unsigned long flags;
4037	int err = -EBUSY;
4038
4039	/* The dump callback needs to be set */
4040	if (!dumper->dump)
4041		return -EINVAL;
4042
4043	spin_lock_irqsave(&dump_list_lock, flags);
4044	/* Don't allow registering multiple times */
4045	if (!dumper->registered) {
4046		dumper->registered = 1;
4047		list_add_tail_rcu(&dumper->list, &dump_list);
4048		err = 0;
4049	}
4050	spin_unlock_irqrestore(&dump_list_lock, flags);
4051
4052	return err;
4053}
4054EXPORT_SYMBOL_GPL(kmsg_dump_register);
4055
4056/**
4057 * kmsg_dump_unregister - unregister a kmsg dumper.
4058 * @dumper: pointer to the kmsg_dumper structure
4059 *
4060 * Removes a dump device from the system. Returns zero on success and
4061 * %-EINVAL otherwise.
4062 */
4063int kmsg_dump_unregister(struct kmsg_dumper *dumper)
4064{
4065	unsigned long flags;
4066	int err = -EINVAL;
4067
4068	spin_lock_irqsave(&dump_list_lock, flags);
4069	if (dumper->registered) {
4070		dumper->registered = 0;
4071		list_del_rcu(&dumper->list);
4072		err = 0;
4073	}
4074	spin_unlock_irqrestore(&dump_list_lock, flags);
4075	synchronize_rcu();
4076
4077	return err;
4078}
4079EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
4080
4081static bool always_kmsg_dump;
4082module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
4083
4084const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
4085{
4086	switch (reason) {
4087	case KMSG_DUMP_PANIC:
4088		return "Panic";
4089	case KMSG_DUMP_OOPS:
4090		return "Oops";
4091	case KMSG_DUMP_EMERG:
4092		return "Emergency";
4093	case KMSG_DUMP_SHUTDOWN:
4094		return "Shutdown";
4095	default:
4096		return "Unknown";
4097	}
4098}
4099EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
4100
4101/**
4102 * kmsg_dump - dump kernel log to kernel message dumpers.
4103 * @reason: the reason (oops, panic etc) for dumping
4104 *
4105 * Call each of the registered dumper's dump() callback, which can
4106 * retrieve the kmsg records with kmsg_dump_get_line() or
4107 * kmsg_dump_get_buffer().
4108 */
4109void kmsg_dump(enum kmsg_dump_reason reason)
4110{
4111	struct kmsg_dumper *dumper;
4112
4113	rcu_read_lock();
4114	list_for_each_entry_rcu(dumper, &dump_list, list) {
4115		enum kmsg_dump_reason max_reason = dumper->max_reason;
4116
4117		/*
4118		 * If client has not provided a specific max_reason, default
4119		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
4120		 */
4121		if (max_reason == KMSG_DUMP_UNDEF) {
4122			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
4123							KMSG_DUMP_OOPS;
4124		}
4125		if (reason > max_reason)
4126			continue;
4127
4128		/* invoke dumper which will iterate over records */
4129		dumper->dump(dumper, reason);
4130	}
4131	rcu_read_unlock();
4132}
4133
4134/**
4135 * kmsg_dump_get_line - retrieve one kmsg log line
4136 * @iter: kmsg dump iterator
4137 * @syslog: include the "<4>" prefixes
4138 * @line: buffer to copy the line to
4139 * @size: maximum size of the buffer
4140 * @len: length of line placed into buffer
4141 *
4142 * Start at the beginning of the kmsg buffer, with the oldest kmsg
4143 * record, and copy one record into the provided buffer.
4144 *
4145 * Consecutive calls will return the next available record moving
4146 * towards the end of the buffer with the youngest messages.
4147 *
4148 * A return value of FALSE indicates that there are no more records to
4149 * read.
4150 */
4151bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
4152			char *line, size_t size, size_t *len)
4153{
4154	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4155	struct printk_info info;
4156	unsigned int line_count;
4157	struct printk_record r;
4158	size_t l = 0;
4159	bool ret = false;
4160
4161	if (iter->cur_seq < min_seq)
4162		iter->cur_seq = min_seq;
4163
4164	prb_rec_init_rd(&r, &info, line, size);
4165
4166	/* Read text or count text lines? */
4167	if (line) {
4168		if (!prb_read_valid(prb, iter->cur_seq, &r))
4169			goto out;
4170		l = record_print_text(&r, syslog, printk_time);
4171	} else {
4172		if (!prb_read_valid_info(prb, iter->cur_seq,
4173					 &info, &line_count)) {
4174			goto out;
4175		}
4176		l = get_record_print_text_size(&info, line_count, syslog,
4177					       printk_time);
4178
4179	}
4180
4181	iter->cur_seq = r.info->seq + 1;
4182	ret = true;
4183out:
4184	if (len)
4185		*len = l;
4186	return ret;
4187}
4188EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
4189
4190/**
4191 * kmsg_dump_get_buffer - copy kmsg log lines
4192 * @iter: kmsg dump iterator
4193 * @syslog: include the "<4>" prefixes
4194 * @buf: buffer to copy the line to
4195 * @size: maximum size of the buffer
4196 * @len_out: length of line placed into buffer
4197 *
4198 * Start at the end of the kmsg buffer and fill the provided buffer
4199 * with as many of the *youngest* kmsg records that fit into it.
4200 * If the buffer is large enough, all available kmsg records will be
4201 * copied with a single call.
4202 *
4203 * Consecutive calls will fill the buffer with the next block of
4204 * available older records, not including the earlier retrieved ones.
4205 *
4206 * A return value of FALSE indicates that there are no more records to
4207 * read.
4208 */
4209bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
4210			  char *buf, size_t size, size_t *len_out)
4211{
4212	u64 min_seq = latched_seq_read_nolock(&clear_seq);
4213	struct printk_info info;
4214	struct printk_record r;
4215	u64 seq;
4216	u64 next_seq;
4217	size_t len = 0;
4218	bool ret = false;
4219	bool time = printk_time;
4220
4221	if (!buf || !size)
4222		goto out;
4223
4224	if (iter->cur_seq < min_seq)
4225		iter->cur_seq = min_seq;
4226
4227	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
4228		if (info.seq != iter->cur_seq) {
4229			/* messages are gone, move to first available one */
4230			iter->cur_seq = info.seq;
4231		}
4232	}
4233
4234	/* last entry */
4235	if (iter->cur_seq >= iter->next_seq)
4236		goto out;
4237
4238	/*
4239	 * Find first record that fits, including all following records,
4240	 * into the user-provided buffer for this dump. Pass in size-1
4241	 * because this function (by way of record_print_text()) will
4242	 * not write more than size-1 bytes of text into @buf.
4243	 */
4244	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
4245				     size - 1, syslog, time);
4246
4247	/*
4248	 * Next kmsg_dump_get_buffer() invocation will dump block of
4249	 * older records stored right before this one.
4250	 */
4251	next_seq = seq;
4252
4253	prb_rec_init_rd(&r, &info, buf, size);
4254
 
4255	prb_for_each_record(seq, prb, seq, &r) {
4256		if (r.info->seq >= iter->next_seq)
4257			break;
4258
4259		len += record_print_text(&r, syslog, time);
4260
4261		/* Adjust record to store to remaining buffer space. */
4262		prb_rec_init_rd(&r, &info, buf + len, size - len);
4263	}
4264
4265	iter->next_seq = next_seq;
4266	ret = true;
4267out:
4268	if (len_out)
4269		*len_out = len;
4270	return ret;
4271}
4272EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
4273
4274/**
4275 * kmsg_dump_rewind - reset the iterator
4276 * @iter: kmsg dump iterator
4277 *
4278 * Reset the dumper's iterator so that kmsg_dump_get_line() and
4279 * kmsg_dump_get_buffer() can be called again and used multiple
4280 * times within the same dumper.dump() callback.
4281 */
4282void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
4283{
4284	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
4285	iter->next_seq = prb_next_seq(prb);
4286}
4287EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
4288
4289#endif
4290
4291#ifdef CONFIG_SMP
4292static atomic_t printk_cpu_sync_owner = ATOMIC_INIT(-1);
4293static atomic_t printk_cpu_sync_nested = ATOMIC_INIT(0);
4294
4295/**
4296 * __printk_cpu_sync_wait() - Busy wait until the printk cpu-reentrant
4297 *                            spinning lock is not owned by any CPU.
4298 *
4299 * Context: Any context.
4300 */
4301void __printk_cpu_sync_wait(void)
4302{
4303	do {
4304		cpu_relax();
4305	} while (atomic_read(&printk_cpu_sync_owner) != -1);
4306}
4307EXPORT_SYMBOL(__printk_cpu_sync_wait);
4308
4309/**
4310 * __printk_cpu_sync_try_get() - Try to acquire the printk cpu-reentrant
4311 *                               spinning lock.
4312 *
4313 * If no processor has the lock, the calling processor takes the lock and
4314 * becomes the owner. If the calling processor is already the owner of the
4315 * lock, this function succeeds immediately.
4316 *
4317 * Context: Any context. Expects interrupts to be disabled.
4318 * Return: 1 on success, otherwise 0.
4319 */
4320int __printk_cpu_sync_try_get(void)
4321{
4322	int cpu;
4323	int old;
4324
4325	cpu = smp_processor_id();
4326
4327	/*
4328	 * Guarantee loads and stores from this CPU when it is the lock owner
4329	 * are _not_ visible to the previous lock owner. This pairs with
4330	 * __printk_cpu_sync_put:B.
4331	 *
4332	 * Memory barrier involvement:
4333	 *
4334	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4335	 * then __printk_cpu_sync_put:A can never read from
4336	 * __printk_cpu_sync_try_get:B.
4337	 *
4338	 * Relies on:
4339	 *
4340	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4341	 * of the previous CPU
4342	 *    matching
4343	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4344	 * __printk_cpu_sync_try_get:B of this CPU
4345	 */
4346	old = atomic_cmpxchg_acquire(&printk_cpu_sync_owner, -1,
4347				     cpu); /* LMM(__printk_cpu_sync_try_get:A) */
4348	if (old == -1) {
4349		/*
4350		 * This CPU is now the owner and begins loading/storing
4351		 * data: LMM(__printk_cpu_sync_try_get:B)
4352		 */
4353		return 1;
4354
4355	} else if (old == cpu) {
4356		/* This CPU is already the owner. */
4357		atomic_inc(&printk_cpu_sync_nested);
4358		return 1;
4359	}
4360
4361	return 0;
4362}
4363EXPORT_SYMBOL(__printk_cpu_sync_try_get);
4364
4365/**
4366 * __printk_cpu_sync_put() - Release the printk cpu-reentrant spinning lock.
4367 *
4368 * The calling processor must be the owner of the lock.
4369 *
4370 * Context: Any context. Expects interrupts to be disabled.
4371 */
4372void __printk_cpu_sync_put(void)
4373{
4374	if (atomic_read(&printk_cpu_sync_nested)) {
4375		atomic_dec(&printk_cpu_sync_nested);
4376		return;
4377	}
4378
4379	/*
4380	 * This CPU is finished loading/storing data:
4381	 * LMM(__printk_cpu_sync_put:A)
4382	 */
4383
4384	/*
4385	 * Guarantee loads and stores from this CPU when it was the
4386	 * lock owner are visible to the next lock owner. This pairs
4387	 * with __printk_cpu_sync_try_get:A.
4388	 *
4389	 * Memory barrier involvement:
4390	 *
4391	 * If __printk_cpu_sync_try_get:A reads from __printk_cpu_sync_put:B,
4392	 * then __printk_cpu_sync_try_get:B reads from __printk_cpu_sync_put:A.
4393	 *
4394	 * Relies on:
4395	 *
4396	 * RELEASE from __printk_cpu_sync_put:A to __printk_cpu_sync_put:B
4397	 * of this CPU
4398	 *    matching
4399	 * ACQUIRE from __printk_cpu_sync_try_get:A to
4400	 * __printk_cpu_sync_try_get:B of the next CPU
4401	 */
4402	atomic_set_release(&printk_cpu_sync_owner,
4403			   -1); /* LMM(__printk_cpu_sync_put:B) */
4404}
4405EXPORT_SYMBOL(__printk_cpu_sync_put);
4406#endif /* CONFIG_SMP */