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 */
v5.14.15
   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_sem protects the console_drivers list, and also
  83 * provides serialisation for access to the entire console
  84 * driver system.
 
 
 
 
 
 
  85 */
  86static DEFINE_SEMAPHORE(console_sem);
  87struct console *console_drivers;
  88EXPORT_SYMBOL_GPL(console_drivers);
 
  89
  90/*
  91 * System may need to suppress printk message under certain
  92 * circumstances, like after kernel panic happens.
  93 */
  94int __read_mostly suppress_printk;
  95
 
 
 
 
 
 
  96#ifdef CONFIG_LOCKDEP
  97static struct lockdep_map console_lock_dep_map = {
  98	.name = "console_lock"
  99};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 100#endif
 101
 102enum devkmsg_log_bits {
 103	__DEVKMSG_LOG_BIT_ON = 0,
 104	__DEVKMSG_LOG_BIT_OFF,
 105	__DEVKMSG_LOG_BIT_LOCK,
 106};
 107
 108enum devkmsg_log_masks {
 109	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
 110	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 111	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 112};
 113
 114/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 115#define DEVKMSG_LOG_MASK_DEFAULT	0
 116
 117static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 118
 119static int __control_devkmsg(char *str)
 120{
 121	size_t len;
 122
 123	if (!str)
 124		return -EINVAL;
 125
 126	len = str_has_prefix(str, "on");
 127	if (len) {
 128		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 129		return len;
 130	}
 131
 132	len = str_has_prefix(str, "off");
 133	if (len) {
 134		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 135		return len;
 136	}
 137
 138	len = str_has_prefix(str, "ratelimit");
 139	if (len) {
 140		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 141		return len;
 142	}
 143
 144	return -EINVAL;
 145}
 146
 147static int __init control_devkmsg(char *str)
 148{
 149	if (__control_devkmsg(str) < 0)
 
 150		return 1;
 
 151
 152	/*
 153	 * Set sysctl string accordingly:
 154	 */
 155	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 156		strcpy(devkmsg_log_str, "on");
 157	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 158		strcpy(devkmsg_log_str, "off");
 159	/* else "ratelimit" which is set by default. */
 160
 161	/*
 162	 * Sysctl cannot change it anymore. The kernel command line setting of
 163	 * this parameter is to force the setting to be permanent throughout the
 164	 * runtime of the system. This is a precation measure against userspace
 165	 * trying to be a smarta** and attempting to change it up on us.
 166	 */
 167	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 168
 169	return 0;
 170}
 171__setup("printk.devkmsg=", control_devkmsg);
 172
 173char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 174
 175int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 176			      void *buffer, size_t *lenp, loff_t *ppos)
 177{
 178	char old_str[DEVKMSG_STR_MAX_SIZE];
 179	unsigned int old;
 180	int err;
 181
 182	if (write) {
 183		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 184			return -EINVAL;
 185
 186		old = devkmsg_log;
 187		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 188	}
 189
 190	err = proc_dostring(table, write, buffer, lenp, ppos);
 191	if (err)
 192		return err;
 193
 194	if (write) {
 195		err = __control_devkmsg(devkmsg_log_str);
 196
 197		/*
 198		 * Do not accept an unknown string OR a known string with
 199		 * trailing crap...
 200		 */
 201		if (err < 0 || (err + 1 != *lenp)) {
 202
 203			/* ... and restore old setting. */
 204			devkmsg_log = old;
 205			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 206
 207			return -EINVAL;
 208		}
 209	}
 210
 211	return 0;
 212}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 213
 214/* Number of registered extended console drivers. */
 215static int nr_ext_console_drivers;
 
 
 
 
 
 
 
 
 
 
 216
 217/*
 218 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 219 * macros instead of functions so that _RET_IP_ contains useful information.
 220 */
 221#define down_console_sem() do { \
 222	down(&console_sem);\
 223	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 224} while (0)
 225
 226static int __down_trylock_console_sem(unsigned long ip)
 227{
 228	int lock_failed;
 229	unsigned long flags;
 230
 231	/*
 232	 * Here and in __up_console_sem() we need to be in safe mode,
 233	 * because spindump/WARN/etc from under console ->lock will
 234	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 235	 */
 236	printk_safe_enter_irqsave(flags);
 237	lock_failed = down_trylock(&console_sem);
 238	printk_safe_exit_irqrestore(flags);
 239
 240	if (lock_failed)
 241		return 1;
 242	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 243	return 0;
 244}
 245#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 246
 247static void __up_console_sem(unsigned long ip)
 248{
 249	unsigned long flags;
 250
 251	mutex_release(&console_lock_dep_map, ip);
 252
 253	printk_safe_enter_irqsave(flags);
 254	up(&console_sem);
 255	printk_safe_exit_irqrestore(flags);
 256}
 257#define up_console_sem() __up_console_sem(_RET_IP_)
 258
 
 
 
 
 
 259/*
 260 * This is used for debugging the mess that is the VT code by
 261 * keeping track if we have the console semaphore held. It's
 262 * definitely not the perfect debug tool (we don't know if _WE_
 263 * hold it and are racing, but it helps tracking those weird code
 264 * paths in the console code where we end up in places I want
 265 * locked without the console semaphore held).
 266 */
 267static int console_locked, console_suspended;
 268
 269/*
 270 * If exclusive_console is non-NULL then only this console is to be printed to.
 271 */
 272static struct console *exclusive_console;
 273
 274/*
 275 *	Array of consoles built from command line options (console=)
 276 */
 277
 278#define MAX_CMDLINECONSOLES 8
 279
 280static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 281
 282static int preferred_console = -1;
 283static bool has_preferred_console;
 284int console_set_on_cmdline;
 285EXPORT_SYMBOL(console_set_on_cmdline);
 286
 287/* Flag: console code may call schedule() */
 288static int console_may_schedule;
 289
 290enum con_msg_format_flags {
 291	MSG_FORMAT_DEFAULT	= 0,
 292	MSG_FORMAT_SYSLOG	= (1 << 0),
 293};
 294
 295static int console_msg_format = MSG_FORMAT_DEFAULT;
 296
 297/*
 298 * The printk log buffer consists of a sequenced collection of records, each
 299 * containing variable length message text. Every record also contains its
 300 * own meta-data (@info).
 301 *
 302 * Every record meta-data carries the timestamp in microseconds, as well as
 303 * the standard userspace syslog level and syslog facility. The usual kernel
 304 * messages use LOG_KERN; userspace-injected messages always carry a matching
 305 * syslog facility, by default LOG_USER. The origin of every message can be
 306 * reliably determined that way.
 307 *
 308 * The human readable log message of a record is available in @text, the
 309 * length of the message text in @text_len. The stored message is not
 310 * terminated.
 311 *
 312 * Optionally, a record can carry a dictionary of properties (key/value
 313 * pairs), to provide userspace with a machine-readable message context.
 314 *
 315 * Examples for well-defined, commonly used property names are:
 316 *   DEVICE=b12:8               device identifier
 317 *                                b12:8         block dev_t
 318 *                                c127:3        char dev_t
 319 *                                n8            netdev ifindex
 320 *                                +sound:card0  subsystem:devname
 321 *   SUBSYSTEM=pci              driver-core subsystem name
 322 *
 323 * Valid characters in property names are [a-zA-Z0-9.-_]. Property names
 324 * and values are terminated by a '\0' character.
 325 *
 326 * Example of record values:
 327 *   record.text_buf                = "it's a line" (unterminated)
 328 *   record.info.seq                = 56
 329 *   record.info.ts_nsec            = 36863
 330 *   record.info.text_len           = 11
 331 *   record.info.facility           = 0 (LOG_KERN)
 332 *   record.info.flags              = 0
 333 *   record.info.level              = 3 (LOG_ERR)
 334 *   record.info.caller_id          = 299 (task 299)
 335 *   record.info.dev_info.subsystem = "pci" (terminated)
 336 *   record.info.dev_info.device    = "+pci:0000:00:01.0" (terminated)
 337 *
 338 * The 'struct printk_info' buffer must never be directly exported to
 339 * userspace, it is a kernel-private implementation detail that might
 340 * need to be changed in the future, when the requirements change.
 341 *
 342 * /dev/kmsg exports the structured data in the following line format:
 343 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 344 *
 345 * Users of the export format should ignore possible additional values
 346 * separated by ',', and find the message after the ';' character.
 347 *
 348 * The optional key/value pairs are attached as continuation lines starting
 349 * with a space character and terminated by a newline. All possible
 350 * non-prinatable characters are escaped in the "\xff" notation.
 351 */
 352
 353enum log_flags {
 354	LOG_NEWLINE	= 2,	/* text ended with a newline */
 355	LOG_CONT	= 8,	/* text is a fragment of a continuation line */
 356};
 357
 358/* syslog_lock protects syslog_* variables and write access to clear_seq. */
 359static DEFINE_RAW_SPINLOCK(syslog_lock);
 360
 361#ifdef CONFIG_PRINTK
 362DECLARE_WAIT_QUEUE_HEAD(log_wait);
 363/* All 3 protected by @syslog_lock. */
 364/* the next printk record to read by syslog(READ) or /proc/kmsg */
 365static u64 syslog_seq;
 366static size_t syslog_partial;
 367static bool syslog_time;
 368
 369/* All 3 protected by @console_sem. */
 370/* the next printk record to write to the console */
 371static u64 console_seq;
 372static u64 exclusive_console_stop_seq;
 373static unsigned long console_dropped;
 374
 375struct latched_seq {
 376	seqcount_latch_t	latch;
 377	u64			val[2];
 378};
 379
 380/*
 381 * The next printk record to read after the last 'clear' command. There are
 382 * two copies (updated with seqcount_latch) so that reads can locklessly
 383 * access a valid value. Writers are synchronized by @syslog_lock.
 384 */
 385static struct latched_seq clear_seq = {
 386	.latch		= SEQCNT_LATCH_ZERO(clear_seq.latch),
 387	.val[0]		= 0,
 388	.val[1]		= 0,
 389};
 390
 391#ifdef CONFIG_PRINTK_CALLER
 392#define PREFIX_MAX		48
 393#else
 394#define PREFIX_MAX		32
 395#endif
 396
 397/* the maximum size of a formatted record (i.e. with prefix added per line) */
 398#define CONSOLE_LOG_MAX		1024
 399
 
 
 
 400/* the maximum size allowed to be reserved for a record */
 401#define LOG_LINE_MAX		(CONSOLE_LOG_MAX - PREFIX_MAX)
 402
 403#define LOG_LEVEL(v)		((v) & 0x07)
 404#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 405
 406/* record buffer */
 407#define LOG_ALIGN __alignof__(unsigned long)
 408#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 409#define LOG_BUF_LEN_MAX (u32)(1 << 31)
 410static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 411static char *log_buf = __log_buf;
 412static u32 log_buf_len = __LOG_BUF_LEN;
 413
 414/*
 415 * Define the average message size. This only affects the number of
 416 * descriptors that will be available. Underestimating is better than
 417 * overestimating (too many available descriptors is better than not enough).
 418 */
 419#define PRB_AVGBITS 5	/* 32 character average length */
 420
 421#if CONFIG_LOG_BUF_SHIFT <= PRB_AVGBITS
 422#error CONFIG_LOG_BUF_SHIFT value too small.
 423#endif
 424_DEFINE_PRINTKRB(printk_rb_static, CONFIG_LOG_BUF_SHIFT - PRB_AVGBITS,
 425		 PRB_AVGBITS, &__log_buf[0]);
 426
 427static struct printk_ringbuffer printk_rb_dynamic;
 428
 429static struct printk_ringbuffer *prb = &printk_rb_static;
 430
 431/*
 432 * We cannot access per-CPU data (e.g. per-CPU flush irq_work) before
 433 * per_cpu_areas are initialised. This variable is set to true when
 434 * it's safe to access per-CPU data.
 435 */
 436static bool __printk_percpu_data_ready __read_mostly;
 437
 438bool printk_percpu_data_ready(void)
 439{
 440	return __printk_percpu_data_ready;
 441}
 442
 443/* Must be called under syslog_lock. */
 444static void latched_seq_write(struct latched_seq *ls, u64 val)
 445{
 446	raw_write_seqcount_latch(&ls->latch);
 447	ls->val[0] = val;
 448	raw_write_seqcount_latch(&ls->latch);
 449	ls->val[1] = val;
 450}
 451
 452/* Can be called from any context. */
 453static u64 latched_seq_read_nolock(struct latched_seq *ls)
 454{
 455	unsigned int seq;
 456	unsigned int idx;
 457	u64 val;
 458
 459	do {
 460		seq = raw_read_seqcount_latch(&ls->latch);
 461		idx = seq & 0x1;
 462		val = ls->val[idx];
 463	} while (read_seqcount_latch_retry(&ls->latch, seq));
 464
 465	return val;
 466}
 467
 468/* Return log buffer address */
 469char *log_buf_addr_get(void)
 470{
 471	return log_buf;
 472}
 473
 474/* Return log buffer size */
 475u32 log_buf_len_get(void)
 476{
 477	return log_buf_len;
 478}
 479
 480/*
 481 * Define how much of the log buffer we could take at maximum. The value
 482 * must be greater than two. Note that only half of the buffer is available
 483 * when the index points to the middle.
 484 */
 485#define MAX_LOG_TAKE_PART 4
 486static const char trunc_msg[] = "<truncated>";
 487
 488static void truncate_msg(u16 *text_len, u16 *trunc_msg_len)
 489{
 490	/*
 491	 * The message should not take the whole buffer. Otherwise, it might
 492	 * get removed too soon.
 493	 */
 494	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 495
 496	if (*text_len > max_text_len)
 497		*text_len = max_text_len;
 498
 499	/* enable the warning message (if there is room) */
 500	*trunc_msg_len = strlen(trunc_msg);
 501	if (*text_len >= *trunc_msg_len)
 502		*text_len -= *trunc_msg_len;
 503	else
 504		*trunc_msg_len = 0;
 505}
 506
 507int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 508
 509static int syslog_action_restricted(int type)
 510{
 511	if (dmesg_restrict)
 512		return 1;
 513	/*
 514	 * Unless restricted, we allow "read all" and "get buffer size"
 515	 * for everybody.
 516	 */
 517	return type != SYSLOG_ACTION_READ_ALL &&
 518	       type != SYSLOG_ACTION_SIZE_BUFFER;
 519}
 520
 521static int check_syslog_permissions(int type, int source)
 522{
 523	/*
 524	 * If this is from /proc/kmsg and we've already opened it, then we've
 525	 * already done the capabilities checks at open time.
 526	 */
 527	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 528		goto ok;
 529
 530	if (syslog_action_restricted(type)) {
 531		if (capable(CAP_SYSLOG))
 532			goto ok;
 533		/*
 534		 * For historical reasons, accept CAP_SYS_ADMIN too, with
 535		 * a warning.
 536		 */
 537		if (capable(CAP_SYS_ADMIN)) {
 538			pr_warn_once("%s (%d): Attempt to access syslog with "
 539				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
 540				     "(deprecated).\n",
 541				 current->comm, task_pid_nr(current));
 542			goto ok;
 543		}
 544		return -EPERM;
 545	}
 546ok:
 547	return security_syslog(type);
 548}
 549
 550static void append_char(char **pp, char *e, char c)
 551{
 552	if (*pp < e)
 553		*(*pp)++ = c;
 554}
 555
 556static ssize_t info_print_ext_header(char *buf, size_t size,
 557				     struct printk_info *info)
 558{
 559	u64 ts_usec = info->ts_nsec;
 560	char caller[20];
 561#ifdef CONFIG_PRINTK_CALLER
 562	u32 id = info->caller_id;
 563
 564	snprintf(caller, sizeof(caller), ",caller=%c%u",
 565		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
 566#else
 567	caller[0] = '\0';
 568#endif
 569
 570	do_div(ts_usec, 1000);
 571
 572	return scnprintf(buf, size, "%u,%llu,%llu,%c%s;",
 573			 (info->facility << 3) | info->level, info->seq,
 574			 ts_usec, info->flags & LOG_CONT ? 'c' : '-', caller);
 575}
 576
 577static ssize_t msg_add_ext_text(char *buf, size_t size,
 578				const char *text, size_t text_len,
 579				unsigned char endc)
 580{
 581	char *p = buf, *e = buf + size;
 582	size_t i;
 583
 584	/* escape non-printable characters */
 585	for (i = 0; i < text_len; i++) {
 586		unsigned char c = text[i];
 587
 588		if (c < ' ' || c >= 127 || c == '\\')
 589			p += scnprintf(p, e - p, "\\x%02x", c);
 590		else
 591			append_char(&p, e, c);
 592	}
 593	append_char(&p, e, endc);
 594
 595	return p - buf;
 596}
 597
 598static ssize_t msg_add_dict_text(char *buf, size_t size,
 599				 const char *key, const char *val)
 600{
 601	size_t val_len = strlen(val);
 602	ssize_t len;
 603
 604	if (!val_len)
 605		return 0;
 606
 607	len = msg_add_ext_text(buf, size, "", 0, ' ');	/* dict prefix */
 608	len += msg_add_ext_text(buf + len, size - len, key, strlen(key), '=');
 609	len += msg_add_ext_text(buf + len, size - len, val, val_len, '\n');
 610
 611	return len;
 612}
 613
 614static ssize_t msg_print_ext_body(char *buf, size_t size,
 615				  char *text, size_t text_len,
 616				  struct dev_printk_info *dev_info)
 617{
 618	ssize_t len;
 619
 620	len = msg_add_ext_text(buf, size, text, text_len, '\n');
 621
 622	if (!dev_info)
 623		goto out;
 624
 625	len += msg_add_dict_text(buf + len, size - len, "SUBSYSTEM",
 626				 dev_info->subsystem);
 627	len += msg_add_dict_text(buf + len, size - len, "DEVICE",
 628				 dev_info->device);
 629out:
 630	return len;
 631}
 632
 633/* /dev/kmsg - userspace message inject/listen interface */
 634struct devkmsg_user {
 635	atomic64_t seq;
 636	struct ratelimit_state rs;
 637	struct mutex lock;
 638	char buf[CONSOLE_EXT_LOG_MAX];
 639
 640	struct printk_info info;
 641	char text_buf[CONSOLE_EXT_LOG_MAX];
 642	struct printk_record record;
 643};
 644
 645static __printf(3, 4) __cold
 646int devkmsg_emit(int facility, int level, const char *fmt, ...)
 647{
 648	va_list args;
 649	int r;
 650
 651	va_start(args, fmt);
 652	r = vprintk_emit(facility, level, NULL, fmt, args);
 653	va_end(args);
 654
 655	return r;
 656}
 657
 658static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 659{
 660	char *buf, *line;
 661	int level = default_message_loglevel;
 662	int facility = 1;	/* LOG_USER */
 663	struct file *file = iocb->ki_filp;
 664	struct devkmsg_user *user = file->private_data;
 665	size_t len = iov_iter_count(from);
 666	ssize_t ret = len;
 667
 668	if (!user || len > LOG_LINE_MAX)
 669		return -EINVAL;
 670
 671	/* Ignore when user logging is disabled. */
 672	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 673		return len;
 674
 675	/* Ratelimit when not explicitly enabled. */
 676	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 677		if (!___ratelimit(&user->rs, current->comm))
 678			return ret;
 679	}
 680
 681	buf = kmalloc(len+1, GFP_KERNEL);
 682	if (buf == NULL)
 683		return -ENOMEM;
 684
 685	buf[len] = '\0';
 686	if (!copy_from_iter_full(buf, len, from)) {
 687		kfree(buf);
 688		return -EFAULT;
 689	}
 690
 691	/*
 692	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 693	 * the decimal value represents 32bit, the lower 3 bit are the log
 694	 * level, the rest are the log facility.
 695	 *
 696	 * If no prefix or no userspace facility is specified, we
 697	 * enforce LOG_USER, to be able to reliably distinguish
 698	 * kernel-generated messages from userspace-injected ones.
 699	 */
 700	line = buf;
 701	if (line[0] == '<') {
 702		char *endp = NULL;
 703		unsigned int u;
 704
 705		u = simple_strtoul(line + 1, &endp, 10);
 706		if (endp && endp[0] == '>') {
 707			level = LOG_LEVEL(u);
 708			if (LOG_FACILITY(u) != 0)
 709				facility = LOG_FACILITY(u);
 710			endp++;
 711			line = endp;
 712		}
 713	}
 714
 715	devkmsg_emit(facility, level, "%s", line);
 716	kfree(buf);
 717	return ret;
 718}
 719
 720static ssize_t devkmsg_read(struct file *file, char __user *buf,
 721			    size_t count, loff_t *ppos)
 722{
 723	struct devkmsg_user *user = file->private_data;
 724	struct printk_record *r = &user->record;
 725	size_t len;
 726	ssize_t ret;
 727
 728	if (!user)
 729		return -EBADF;
 730
 731	ret = mutex_lock_interruptible(&user->lock);
 732	if (ret)
 733		return ret;
 734
 735	printk_safe_enter_irq();
 736	if (!prb_read_valid(prb, atomic64_read(&user->seq), r)) {
 737		if (file->f_flags & O_NONBLOCK) {
 738			ret = -EAGAIN;
 739			printk_safe_exit_irq();
 740			goto out;
 741		}
 742
 743		printk_safe_exit_irq();
 
 
 
 
 
 
 
 
 
 744		ret = wait_event_interruptible(log_wait,
 745				prb_read_valid(prb, atomic64_read(&user->seq), r));
 
 746		if (ret)
 747			goto out;
 748		printk_safe_enter_irq();
 749	}
 750
 751	if (r->info->seq != atomic64_read(&user->seq)) {
 752		/* our last seen message is gone, return error and reset */
 753		atomic64_set(&user->seq, r->info->seq);
 754		ret = -EPIPE;
 755		printk_safe_exit_irq();
 756		goto out;
 757	}
 758
 759	len = info_print_ext_header(user->buf, sizeof(user->buf), r->info);
 760	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
 761				  &r->text_buf[0], r->info->text_len,
 762				  &r->info->dev_info);
 763
 764	atomic64_set(&user->seq, r->info->seq + 1);
 765	printk_safe_exit_irq();
 766
 767	if (len > count) {
 768		ret = -EINVAL;
 769		goto out;
 770	}
 771
 772	if (copy_to_user(buf, user->buf, len)) {
 773		ret = -EFAULT;
 774		goto out;
 775	}
 776	ret = len;
 777out:
 778	mutex_unlock(&user->lock);
 779	return ret;
 780}
 781
 782/*
 783 * Be careful when modifying this function!!!
 784 *
 785 * Only few operations are supported because the device works only with the
 786 * entire variable length messages (records). Non-standard values are
 787 * returned in the other cases and has been this way for quite some time.
 788 * User space applications might depend on this behavior.
 789 */
 790static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 791{
 792	struct devkmsg_user *user = file->private_data;
 793	loff_t ret = 0;
 794
 795	if (!user)
 796		return -EBADF;
 797	if (offset)
 798		return -ESPIPE;
 799
 800	printk_safe_enter_irq();
 801	switch (whence) {
 802	case SEEK_SET:
 803		/* the first record */
 804		atomic64_set(&user->seq, prb_first_valid_seq(prb));
 805		break;
 806	case SEEK_DATA:
 807		/*
 808		 * The first record after the last SYSLOG_ACTION_CLEAR,
 809		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 810		 * changes no global state, and does not clear anything.
 811		 */
 812		atomic64_set(&user->seq, latched_seq_read_nolock(&clear_seq));
 813		break;
 814	case SEEK_END:
 815		/* after the last record */
 816		atomic64_set(&user->seq, prb_next_seq(prb));
 817		break;
 818	default:
 819		ret = -EINVAL;
 820	}
 821	printk_safe_exit_irq();
 822	return ret;
 823}
 824
 825static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 826{
 827	struct devkmsg_user *user = file->private_data;
 828	struct printk_info info;
 829	__poll_t ret = 0;
 830
 831	if (!user)
 832		return EPOLLERR|EPOLLNVAL;
 833
 834	poll_wait(file, &log_wait, wait);
 835
 836	printk_safe_enter_irq();
 837	if (prb_read_valid_info(prb, atomic64_read(&user->seq), &info, NULL)) {
 838		/* return error when data has vanished underneath us */
 839		if (info.seq != atomic64_read(&user->seq))
 840			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 841		else
 842			ret = EPOLLIN|EPOLLRDNORM;
 843	}
 844	printk_safe_exit_irq();
 845
 846	return ret;
 847}
 848
 849static int devkmsg_open(struct inode *inode, struct file *file)
 850{
 851	struct devkmsg_user *user;
 852	int err;
 853
 854	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 855		return -EPERM;
 856
 857	/* write-only does not need any file context */
 858	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 859		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 860					       SYSLOG_FROM_READER);
 861		if (err)
 862			return err;
 863	}
 864
 865	user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 866	if (!user)
 867		return -ENOMEM;
 868
 869	ratelimit_default_init(&user->rs);
 870	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 871
 872	mutex_init(&user->lock);
 873
 874	prb_rec_init_rd(&user->record, &user->info,
 875			&user->text_buf[0], sizeof(user->text_buf));
 876
 877	printk_safe_enter_irq();
 878	atomic64_set(&user->seq, prb_first_valid_seq(prb));
 879	printk_safe_exit_irq();
 880
 881	file->private_data = user;
 882	return 0;
 883}
 884
 885static int devkmsg_release(struct inode *inode, struct file *file)
 886{
 887	struct devkmsg_user *user = file->private_data;
 888
 889	if (!user)
 890		return 0;
 891
 892	ratelimit_state_exit(&user->rs);
 893
 894	mutex_destroy(&user->lock);
 895	kfree(user);
 896	return 0;
 897}
 898
 899const struct file_operations kmsg_fops = {
 900	.open = devkmsg_open,
 901	.read = devkmsg_read,
 902	.write_iter = devkmsg_write,
 903	.llseek = devkmsg_llseek,
 904	.poll = devkmsg_poll,
 905	.release = devkmsg_release,
 906};
 907
 908#ifdef CONFIG_CRASH_CORE
 909/*
 910 * This appends the listed symbols to /proc/vmcore
 911 *
 912 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 913 * obtain access to symbols that are otherwise very difficult to locate.  These
 914 * symbols are specifically used so that utilities can access and extract the
 915 * dmesg log from a vmcore file after a crash.
 916 */
 917void log_buf_vmcoreinfo_setup(void)
 918{
 919	struct dev_printk_info *dev_info = NULL;
 920
 921	VMCOREINFO_SYMBOL(prb);
 922	VMCOREINFO_SYMBOL(printk_rb_static);
 923	VMCOREINFO_SYMBOL(clear_seq);
 924
 925	/*
 926	 * Export struct size and field offsets. User space tools can
 927	 * parse it and detect any changes to structure down the line.
 928	 */
 929
 930	VMCOREINFO_STRUCT_SIZE(printk_ringbuffer);
 931	VMCOREINFO_OFFSET(printk_ringbuffer, desc_ring);
 932	VMCOREINFO_OFFSET(printk_ringbuffer, text_data_ring);
 933	VMCOREINFO_OFFSET(printk_ringbuffer, fail);
 934
 935	VMCOREINFO_STRUCT_SIZE(prb_desc_ring);
 936	VMCOREINFO_OFFSET(prb_desc_ring, count_bits);
 937	VMCOREINFO_OFFSET(prb_desc_ring, descs);
 938	VMCOREINFO_OFFSET(prb_desc_ring, infos);
 939	VMCOREINFO_OFFSET(prb_desc_ring, head_id);
 940	VMCOREINFO_OFFSET(prb_desc_ring, tail_id);
 941
 942	VMCOREINFO_STRUCT_SIZE(prb_desc);
 943	VMCOREINFO_OFFSET(prb_desc, state_var);
 944	VMCOREINFO_OFFSET(prb_desc, text_blk_lpos);
 945
 946	VMCOREINFO_STRUCT_SIZE(prb_data_blk_lpos);
 947	VMCOREINFO_OFFSET(prb_data_blk_lpos, begin);
 948	VMCOREINFO_OFFSET(prb_data_blk_lpos, next);
 949
 950	VMCOREINFO_STRUCT_SIZE(printk_info);
 951	VMCOREINFO_OFFSET(printk_info, seq);
 952	VMCOREINFO_OFFSET(printk_info, ts_nsec);
 953	VMCOREINFO_OFFSET(printk_info, text_len);
 954	VMCOREINFO_OFFSET(printk_info, caller_id);
 955	VMCOREINFO_OFFSET(printk_info, dev_info);
 956
 957	VMCOREINFO_STRUCT_SIZE(dev_printk_info);
 958	VMCOREINFO_OFFSET(dev_printk_info, subsystem);
 959	VMCOREINFO_LENGTH(printk_info_subsystem, sizeof(dev_info->subsystem));
 960	VMCOREINFO_OFFSET(dev_printk_info, device);
 961	VMCOREINFO_LENGTH(printk_info_device, sizeof(dev_info->device));
 962
 963	VMCOREINFO_STRUCT_SIZE(prb_data_ring);
 964	VMCOREINFO_OFFSET(prb_data_ring, size_bits);
 965	VMCOREINFO_OFFSET(prb_data_ring, data);
 966	VMCOREINFO_OFFSET(prb_data_ring, head_lpos);
 967	VMCOREINFO_OFFSET(prb_data_ring, tail_lpos);
 968
 969	VMCOREINFO_SIZE(atomic_long_t);
 970	VMCOREINFO_TYPE_OFFSET(atomic_long_t, counter);
 971
 972	VMCOREINFO_STRUCT_SIZE(latched_seq);
 973	VMCOREINFO_OFFSET(latched_seq, val);
 974}
 975#endif
 976
 977/* requested log_buf_len from kernel cmdline */
 978static unsigned long __initdata new_log_buf_len;
 979
 980/* we practice scaling the ring buffer by powers of 2 */
 981static void __init log_buf_len_update(u64 size)
 982{
 983	if (size > (u64)LOG_BUF_LEN_MAX) {
 984		size = (u64)LOG_BUF_LEN_MAX;
 985		pr_err("log_buf over 2G is not supported.\n");
 986	}
 987
 988	if (size)
 989		size = roundup_pow_of_two(size);
 990	if (size > log_buf_len)
 991		new_log_buf_len = (unsigned long)size;
 992}
 993
 994/* save requested log_buf_len since it's too early to process it */
 995static int __init log_buf_len_setup(char *str)
 996{
 997	u64 size;
 998
 999	if (!str)
1000		return -EINVAL;
1001
1002	size = memparse(str, &str);
1003
1004	log_buf_len_update(size);
1005
1006	return 0;
1007}
1008early_param("log_buf_len", log_buf_len_setup);
1009
1010#ifdef CONFIG_SMP
1011#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1012
1013static void __init log_buf_add_cpu(void)
1014{
1015	unsigned int cpu_extra;
1016
1017	/*
1018	 * archs should set up cpu_possible_bits properly with
1019	 * set_cpu_possible() after setup_arch() but just in
1020	 * case lets ensure this is valid.
1021	 */
1022	if (num_possible_cpus() == 1)
1023		return;
1024
1025	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1026
1027	/* by default this will only continue through for large > 64 CPUs */
1028	if (cpu_extra <= __LOG_BUF_LEN / 2)
1029		return;
1030
1031	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1032		__LOG_CPU_MAX_BUF_LEN);
1033	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1034		cpu_extra);
1035	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1036
1037	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1038}
1039#else /* !CONFIG_SMP */
1040static inline void log_buf_add_cpu(void) {}
1041#endif /* CONFIG_SMP */
1042
1043static void __init set_percpu_data_ready(void)
1044{
1045	printk_safe_init();
1046	/* Make sure we set this flag only after printk_safe() init is done */
1047	barrier();
1048	__printk_percpu_data_ready = true;
1049}
1050
1051static unsigned int __init add_to_rb(struct printk_ringbuffer *rb,
1052				     struct printk_record *r)
1053{
1054	struct prb_reserved_entry e;
1055	struct printk_record dest_r;
1056
1057	prb_rec_init_wr(&dest_r, r->info->text_len);
1058
1059	if (!prb_reserve(&e, rb, &dest_r))
1060		return 0;
1061
1062	memcpy(&dest_r.text_buf[0], &r->text_buf[0], r->info->text_len);
1063	dest_r.info->text_len = r->info->text_len;
1064	dest_r.info->facility = r->info->facility;
1065	dest_r.info->level = r->info->level;
1066	dest_r.info->flags = r->info->flags;
1067	dest_r.info->ts_nsec = r->info->ts_nsec;
1068	dest_r.info->caller_id = r->info->caller_id;
1069	memcpy(&dest_r.info->dev_info, &r->info->dev_info, sizeof(dest_r.info->dev_info));
1070
1071	prb_final_commit(&e);
1072
1073	return prb_record_text_space(&e);
1074}
1075
1076static char setup_text_buf[LOG_LINE_MAX] __initdata;
1077
1078void __init setup_log_buf(int early)
1079{
1080	struct printk_info *new_infos;
1081	unsigned int new_descs_count;
1082	struct prb_desc *new_descs;
1083	struct printk_info info;
1084	struct printk_record r;
 
1085	size_t new_descs_size;
1086	size_t new_infos_size;
1087	unsigned long flags;
1088	char *new_log_buf;
1089	unsigned int free;
1090	u64 seq;
1091
1092	/*
1093	 * Some archs call setup_log_buf() multiple times - first is very
1094	 * early, e.g. from setup_arch(), and second - when percpu_areas
1095	 * are initialised.
1096	 */
1097	if (!early)
1098		set_percpu_data_ready();
1099
1100	if (log_buf != __log_buf)
1101		return;
1102
1103	if (!early && !new_log_buf_len)
1104		log_buf_add_cpu();
1105
1106	if (!new_log_buf_len)
1107		return;
1108
1109	new_descs_count = new_log_buf_len >> PRB_AVGBITS;
1110	if (new_descs_count == 0) {
1111		pr_err("new_log_buf_len: %lu too small\n", new_log_buf_len);
1112		return;
1113	}
1114
1115	new_log_buf = memblock_alloc(new_log_buf_len, LOG_ALIGN);
1116	if (unlikely(!new_log_buf)) {
1117		pr_err("log_buf_len: %lu text bytes not available\n",
1118		       new_log_buf_len);
1119		return;
1120	}
1121
1122	new_descs_size = new_descs_count * sizeof(struct prb_desc);
1123	new_descs = memblock_alloc(new_descs_size, LOG_ALIGN);
1124	if (unlikely(!new_descs)) {
1125		pr_err("log_buf_len: %zu desc bytes not available\n",
1126		       new_descs_size);
1127		goto err_free_log_buf;
1128	}
1129
1130	new_infos_size = new_descs_count * sizeof(struct printk_info);
1131	new_infos = memblock_alloc(new_infos_size, LOG_ALIGN);
1132	if (unlikely(!new_infos)) {
1133		pr_err("log_buf_len: %zu info bytes not available\n",
1134		       new_infos_size);
1135		goto err_free_descs;
1136	}
1137
1138	prb_rec_init_rd(&r, &info, &setup_text_buf[0], sizeof(setup_text_buf));
1139
1140	prb_init(&printk_rb_dynamic,
1141		 new_log_buf, ilog2(new_log_buf_len),
1142		 new_descs, ilog2(new_descs_count),
1143		 new_infos);
1144
1145	printk_safe_enter_irqsave(flags);
1146
1147	log_buf_len = new_log_buf_len;
1148	log_buf = new_log_buf;
1149	new_log_buf_len = 0;
1150
1151	free = __LOG_BUF_LEN;
1152	prb_for_each_record(0, &printk_rb_static, seq, &r)
1153		free -= add_to_rb(&printk_rb_dynamic, &r);
 
 
 
 
 
 
 
 
 
1154
1155	/*
1156	 * This is early enough that everything is still running on the
1157	 * boot CPU and interrupts are disabled. So no new messages will
1158	 * appear during the transition to the dynamic buffer.
1159	 */
1160	prb = &printk_rb_dynamic;
1161
1162	printk_safe_exit_irqrestore(flags);
 
 
 
 
1163
1164	if (seq != prb_next_seq(&printk_rb_static)) {
1165		pr_err("dropped %llu messages\n",
1166		       prb_next_seq(&printk_rb_static) - seq);
1167	}
1168
1169	pr_info("log_buf_len: %u bytes\n", log_buf_len);
1170	pr_info("early log buf free: %u(%u%%)\n",
1171		free, (free * 100) / __LOG_BUF_LEN);
1172	return;
1173
1174err_free_descs:
1175	memblock_free(__pa(new_descs), new_descs_size);
1176err_free_log_buf:
1177	memblock_free(__pa(new_log_buf), new_log_buf_len);
1178}
1179
1180static bool __read_mostly ignore_loglevel;
1181
1182static int __init ignore_loglevel_setup(char *str)
1183{
1184	ignore_loglevel = true;
1185	pr_info("debug: ignoring loglevel setting.\n");
1186
1187	return 0;
1188}
1189
1190early_param("ignore_loglevel", ignore_loglevel_setup);
1191module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1192MODULE_PARM_DESC(ignore_loglevel,
1193		 "ignore loglevel setting (prints all kernel messages to the console)");
1194
1195static bool suppress_message_printing(int level)
1196{
1197	return (level >= console_loglevel && !ignore_loglevel);
1198}
1199
1200#ifdef CONFIG_BOOT_PRINTK_DELAY
1201
1202static int boot_delay; /* msecs delay after each printk during bootup */
1203static unsigned long long loops_per_msec;	/* based on boot_delay */
1204
1205static int __init boot_delay_setup(char *str)
1206{
1207	unsigned long lpj;
1208
1209	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1210	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1211
1212	get_option(&str, &boot_delay);
1213	if (boot_delay > 10 * 1000)
1214		boot_delay = 0;
1215
1216	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1217		"HZ: %d, loops_per_msec: %llu\n",
1218		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1219	return 0;
1220}
1221early_param("boot_delay", boot_delay_setup);
1222
1223static void boot_delay_msec(int level)
1224{
1225	unsigned long long k;
1226	unsigned long timeout;
1227
1228	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1229		|| suppress_message_printing(level)) {
1230		return;
1231	}
1232
1233	k = (unsigned long long)loops_per_msec * boot_delay;
1234
1235	timeout = jiffies + msecs_to_jiffies(boot_delay);
1236	while (k) {
1237		k--;
1238		cpu_relax();
1239		/*
1240		 * use (volatile) jiffies to prevent
1241		 * compiler reduction; loop termination via jiffies
1242		 * is secondary and may or may not happen.
1243		 */
1244		if (time_after(jiffies, timeout))
1245			break;
1246		touch_nmi_watchdog();
1247	}
1248}
1249#else
1250static inline void boot_delay_msec(int level)
1251{
1252}
1253#endif
1254
1255static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1256module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1257
1258static size_t print_syslog(unsigned int level, char *buf)
1259{
1260	return sprintf(buf, "<%u>", level);
1261}
1262
1263static size_t print_time(u64 ts, char *buf)
1264{
1265	unsigned long rem_nsec = do_div(ts, 1000000000);
1266
1267	return sprintf(buf, "[%5lu.%06lu]",
1268		       (unsigned long)ts, rem_nsec / 1000);
1269}
1270
1271#ifdef CONFIG_PRINTK_CALLER
1272static size_t print_caller(u32 id, char *buf)
1273{
1274	char caller[12];
1275
1276	snprintf(caller, sizeof(caller), "%c%u",
1277		 id & 0x80000000 ? 'C' : 'T', id & ~0x80000000);
1278	return sprintf(buf, "[%6s]", caller);
1279}
1280#else
1281#define print_caller(id, buf) 0
1282#endif
1283
1284static size_t info_print_prefix(const struct printk_info  *info, bool syslog,
1285				bool time, char *buf)
1286{
1287	size_t len = 0;
1288
1289	if (syslog)
1290		len = print_syslog((info->facility << 3) | info->level, buf);
1291
1292	if (time)
1293		len += print_time(info->ts_nsec, buf + len);
1294
1295	len += print_caller(info->caller_id, buf + len);
1296
1297	if (IS_ENABLED(CONFIG_PRINTK_CALLER) || time) {
1298		buf[len++] = ' ';
1299		buf[len] = '\0';
1300	}
1301
1302	return len;
1303}
1304
1305/*
1306 * Prepare the record for printing. The text is shifted within the given
1307 * buffer to avoid a need for another one. The following operations are
1308 * done:
1309 *
1310 *   - Add prefix for each line.
1311 *   - Drop truncated lines that no longer fit into the buffer.
1312 *   - Add the trailing newline that has been removed in vprintk_store().
1313 *   - Add a string terminator.
1314 *
1315 * Since the produced string is always terminated, the maximum possible
1316 * return value is @r->text_buf_size - 1;
1317 *
1318 * Return: The length of the updated/prepared text, including the added
1319 * prefixes and the newline. The terminator is not counted. The dropped
1320 * line(s) are not counted.
1321 */
1322static size_t record_print_text(struct printk_record *r, bool syslog,
1323				bool time)
1324{
1325	size_t text_len = r->info->text_len;
1326	size_t buf_size = r->text_buf_size;
1327	char *text = r->text_buf;
1328	char prefix[PREFIX_MAX];
1329	bool truncated = false;
1330	size_t prefix_len;
1331	size_t line_len;
1332	size_t len = 0;
1333	char *next;
1334
1335	/*
1336	 * If the message was truncated because the buffer was not large
1337	 * enough, treat the available text as if it were the full text.
1338	 */
1339	if (text_len > buf_size)
1340		text_len = buf_size;
1341
1342	prefix_len = info_print_prefix(r->info, syslog, time, prefix);
1343
1344	/*
1345	 * @text_len: bytes of unprocessed text
1346	 * @line_len: bytes of current line _without_ newline
1347	 * @text:     pointer to beginning of current line
1348	 * @len:      number of bytes prepared in r->text_buf
1349	 */
1350	for (;;) {
1351		next = memchr(text, '\n', text_len);
1352		if (next) {
1353			line_len = next - text;
1354		} else {
1355			/* Drop truncated line(s). */
1356			if (truncated)
1357				break;
1358			line_len = text_len;
1359		}
1360
1361		/*
1362		 * Truncate the text if there is not enough space to add the
1363		 * prefix and a trailing newline and a terminator.
1364		 */
1365		if (len + prefix_len + text_len + 1 + 1 > buf_size) {
1366			/* Drop even the current line if no space. */
1367			if (len + prefix_len + line_len + 1 + 1 > buf_size)
1368				break;
1369
1370			text_len = buf_size - len - prefix_len - 1 - 1;
1371			truncated = true;
1372		}
1373
1374		memmove(text + prefix_len, text, text_len);
1375		memcpy(text, prefix, prefix_len);
1376
1377		/*
1378		 * Increment the prepared length to include the text and
1379		 * prefix that were just moved+copied. Also increment for the
1380		 * newline at the end of this line. If this is the last line,
1381		 * there is no newline, but it will be added immediately below.
1382		 */
1383		len += prefix_len + line_len + 1;
1384		if (text_len == line_len) {
1385			/*
1386			 * This is the last line. Add the trailing newline
1387			 * removed in vprintk_store().
1388			 */
1389			text[prefix_len + line_len] = '\n';
1390			break;
1391		}
1392
1393		/*
1394		 * Advance beyond the added prefix and the related line with
1395		 * its newline.
1396		 */
1397		text += prefix_len + line_len + 1;
1398
1399		/*
1400		 * The remaining text has only decreased by the line with its
1401		 * newline.
1402		 *
1403		 * Note that @text_len can become zero. It happens when @text
1404		 * ended with a newline (either due to truncation or the
1405		 * original string ending with "\n\n"). The loop is correctly
1406		 * repeated and (if not truncated) an empty line with a prefix
1407		 * will be prepared.
1408		 */
1409		text_len -= line_len + 1;
1410	}
1411
1412	/*
1413	 * If a buffer was provided, it will be terminated. Space for the
1414	 * string terminator is guaranteed to be available. The terminator is
1415	 * not counted in the return value.
1416	 */
1417	if (buf_size > 0)
1418		r->text_buf[len] = 0;
1419
1420	return len;
1421}
1422
1423static size_t get_record_print_text_size(struct printk_info *info,
1424					 unsigned int line_count,
1425					 bool syslog, bool time)
1426{
1427	char prefix[PREFIX_MAX];
1428	size_t prefix_len;
1429
1430	prefix_len = info_print_prefix(info, syslog, time, prefix);
1431
1432	/*
1433	 * Each line will be preceded with a prefix. The intermediate
1434	 * newlines are already within the text, but a final trailing
1435	 * newline will be added.
1436	 */
1437	return ((prefix_len * line_count) + info->text_len + 1);
1438}
1439
1440/*
1441 * Beginning with @start_seq, find the first record where it and all following
1442 * records up to (but not including) @max_seq fit into @size.
1443 *
1444 * @max_seq is simply an upper bound and does not need to exist. If the caller
1445 * does not require an upper bound, -1 can be used for @max_seq.
1446 */
1447static u64 find_first_fitting_seq(u64 start_seq, u64 max_seq, size_t size,
1448				  bool syslog, bool time)
1449{
1450	struct printk_info info;
1451	unsigned int line_count;
1452	size_t len = 0;
1453	u64 seq;
1454
1455	/* Determine the size of the records up to @max_seq. */
1456	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1457		if (info.seq >= max_seq)
1458			break;
1459		len += get_record_print_text_size(&info, line_count, syslog, time);
1460	}
1461
1462	/*
1463	 * Adjust the upper bound for the next loop to avoid subtracting
1464	 * lengths that were never added.
1465	 */
1466	if (seq < max_seq)
1467		max_seq = seq;
1468
1469	/*
1470	 * Move first record forward until length fits into the buffer. Ignore
1471	 * newest messages that were not counted in the above cycle. Messages
1472	 * might appear and get lost in the meantime. This is a best effort
1473	 * that prevents an infinite loop that could occur with a retry.
1474	 */
1475	prb_for_each_info(start_seq, prb, seq, &info, &line_count) {
1476		if (len <= size || info.seq >= max_seq)
1477			break;
1478		len -= get_record_print_text_size(&info, line_count, syslog, time);
1479	}
1480
1481	return seq;
1482}
1483
 
1484static int syslog_print(char __user *buf, int size)
1485{
1486	struct printk_info info;
1487	struct printk_record r;
1488	char *text;
1489	int len = 0;
 
1490
1491	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1492	if (!text)
1493		return -ENOMEM;
1494
1495	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1496
1497	while (size > 0) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1498		size_t n;
1499		size_t skip;
 
1500
1501		printk_safe_enter_irq();
1502		raw_spin_lock(&syslog_lock);
1503		if (!prb_read_valid(prb, syslog_seq, &r)) {
1504			raw_spin_unlock(&syslog_lock);
1505			printk_safe_exit_irq();
1506			break;
1507		}
1508		if (r.info->seq != syslog_seq) {
1509			/* message is gone, move to next valid one */
1510			syslog_seq = r.info->seq;
1511			syslog_partial = 0;
1512		}
1513
1514		/*
1515		 * To keep reading/counting partial line consistent,
1516		 * use printk_time value as of the beginning of a line.
1517		 */
1518		if (!syslog_partial)
1519			syslog_time = printk_time;
1520
1521		skip = syslog_partial;
1522		n = record_print_text(&r, true, syslog_time);
1523		if (n - syslog_partial <= size) {
1524			/* message fits into buffer, move forward */
1525			syslog_seq = r.info->seq + 1;
1526			n -= syslog_partial;
1527			syslog_partial = 0;
1528		} else if (!len){
1529			/* partial read(), remember position */
1530			n = size;
1531			syslog_partial += n;
1532		} else
1533			n = 0;
1534		raw_spin_unlock(&syslog_lock);
1535		printk_safe_exit_irq();
1536
1537		if (!n)
1538			break;
1539
1540		if (copy_to_user(buf, text + skip, n)) {
 
 
 
 
1541			if (!len)
1542				len = -EFAULT;
1543			break;
1544		}
1545
1546		len += n;
1547		size -= n;
1548		buf += n;
1549	}
1550
 
1551	kfree(text);
1552	return len;
1553}
1554
1555static int syslog_print_all(char __user *buf, int size, bool clear)
1556{
1557	struct printk_info info;
1558	struct printk_record r;
1559	char *text;
1560	int len = 0;
1561	u64 seq;
1562	bool time;
1563
1564	text = kmalloc(CONSOLE_LOG_MAX, GFP_KERNEL);
1565	if (!text)
1566		return -ENOMEM;
1567
1568	time = printk_time;
1569	printk_safe_enter_irq();
1570	/*
1571	 * Find first record that fits, including all following records,
1572	 * into the user-provided buffer for this dump.
1573	 */
1574	seq = find_first_fitting_seq(latched_seq_read_nolock(&clear_seq), -1,
1575				     size, true, time);
1576
1577	prb_rec_init_rd(&r, &info, text, CONSOLE_LOG_MAX);
1578
1579	len = 0;
1580	prb_for_each_record(seq, prb, seq, &r) {
1581		int textlen;
1582
1583		textlen = record_print_text(&r, true, time);
1584
1585		if (len + textlen > size) {
1586			seq--;
1587			break;
1588		}
1589
1590		printk_safe_exit_irq();
1591		if (copy_to_user(buf + len, text, textlen))
1592			len = -EFAULT;
1593		else
1594			len += textlen;
1595		printk_safe_enter_irq();
1596
1597		if (len < 0)
1598			break;
1599	}
1600
1601	if (clear) {
1602		raw_spin_lock(&syslog_lock);
1603		latched_seq_write(&clear_seq, seq);
1604		raw_spin_unlock(&syslog_lock);
1605	}
1606	printk_safe_exit_irq();
1607
1608	kfree(text);
1609	return len;
1610}
1611
1612static void syslog_clear(void)
1613{
1614	printk_safe_enter_irq();
1615	raw_spin_lock(&syslog_lock);
1616	latched_seq_write(&clear_seq, prb_next_seq(prb));
1617	raw_spin_unlock(&syslog_lock);
1618	printk_safe_exit_irq();
1619}
1620
1621/* Return a consistent copy of @syslog_seq. */
1622static u64 read_syslog_seq_irq(void)
1623{
1624	u64 seq;
1625
1626	raw_spin_lock_irq(&syslog_lock);
1627	seq = syslog_seq;
1628	raw_spin_unlock_irq(&syslog_lock);
1629
1630	return seq;
1631}
1632
1633int do_syslog(int type, char __user *buf, int len, int source)
1634{
1635	struct printk_info info;
1636	bool clear = false;
1637	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1638	int error;
1639
1640	error = check_syslog_permissions(type, source);
1641	if (error)
1642		return error;
1643
1644	switch (type) {
1645	case SYSLOG_ACTION_CLOSE:	/* Close log */
1646		break;
1647	case SYSLOG_ACTION_OPEN:	/* Open log */
1648		break;
1649	case SYSLOG_ACTION_READ:	/* Read from log */
1650		if (!buf || len < 0)
1651			return -EINVAL;
1652		if (!len)
1653			return 0;
1654		if (!access_ok(buf, len))
1655			return -EFAULT;
1656
1657		error = wait_event_interruptible(log_wait,
1658				prb_read_valid(prb, read_syslog_seq_irq(), NULL));
1659		if (error)
1660			return error;
1661		error = syslog_print(buf, len);
1662		break;
1663	/* Read/clear last kernel messages */
1664	case SYSLOG_ACTION_READ_CLEAR:
1665		clear = true;
1666		fallthrough;
1667	/* Read last kernel messages */
1668	case SYSLOG_ACTION_READ_ALL:
1669		if (!buf || len < 0)
1670			return -EINVAL;
1671		if (!len)
1672			return 0;
1673		if (!access_ok(buf, len))
1674			return -EFAULT;
1675		error = syslog_print_all(buf, len, clear);
1676		break;
1677	/* Clear ring buffer */
1678	case SYSLOG_ACTION_CLEAR:
1679		syslog_clear();
1680		break;
1681	/* Disable logging to console */
1682	case SYSLOG_ACTION_CONSOLE_OFF:
1683		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1684			saved_console_loglevel = console_loglevel;
1685		console_loglevel = minimum_console_loglevel;
1686		break;
1687	/* Enable logging to console */
1688	case SYSLOG_ACTION_CONSOLE_ON:
1689		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1690			console_loglevel = saved_console_loglevel;
1691			saved_console_loglevel = LOGLEVEL_DEFAULT;
1692		}
1693		break;
1694	/* Set level of messages printed to console */
1695	case SYSLOG_ACTION_CONSOLE_LEVEL:
1696		if (len < 1 || len > 8)
1697			return -EINVAL;
1698		if (len < minimum_console_loglevel)
1699			len = minimum_console_loglevel;
1700		console_loglevel = len;
1701		/* Implicitly re-enable logging to console */
1702		saved_console_loglevel = LOGLEVEL_DEFAULT;
1703		break;
1704	/* Number of chars in the log buffer */
1705	case SYSLOG_ACTION_SIZE_UNREAD:
1706		printk_safe_enter_irq();
1707		raw_spin_lock(&syslog_lock);
1708		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
1709			/* No unread messages. */
1710			raw_spin_unlock(&syslog_lock);
1711			printk_safe_exit_irq();
1712			return 0;
1713		}
1714		if (info.seq != syslog_seq) {
1715			/* messages are gone, move to first one */
1716			syslog_seq = info.seq;
1717			syslog_partial = 0;
1718		}
1719		if (source == SYSLOG_FROM_PROC) {
1720			/*
1721			 * Short-cut for poll(/"proc/kmsg") which simply checks
1722			 * for pending data, not the size; return the count of
1723			 * records, not the length.
1724			 */
1725			error = prb_next_seq(prb) - syslog_seq;
1726		} else {
1727			bool time = syslog_partial ? syslog_time : printk_time;
1728			unsigned int line_count;
1729			u64 seq;
1730
1731			prb_for_each_info(syslog_seq, prb, seq, &info,
1732					  &line_count) {
1733				error += get_record_print_text_size(&info, line_count,
1734								    true, time);
1735				time = printk_time;
1736			}
1737			error -= syslog_partial;
1738		}
1739		raw_spin_unlock(&syslog_lock);
1740		printk_safe_exit_irq();
1741		break;
1742	/* Size of the log buffer */
1743	case SYSLOG_ACTION_SIZE_BUFFER:
1744		error = log_buf_len;
1745		break;
1746	default:
1747		error = -EINVAL;
1748		break;
1749	}
1750
1751	return error;
1752}
1753
1754SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1755{
1756	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1757}
1758
1759/*
1760 * Special console_lock variants that help to reduce the risk of soft-lockups.
1761 * They allow to pass console_lock to another printk() call using a busy wait.
1762 */
1763
1764#ifdef CONFIG_LOCKDEP
1765static struct lockdep_map console_owner_dep_map = {
1766	.name = "console_owner"
1767};
1768#endif
1769
1770static DEFINE_RAW_SPINLOCK(console_owner_lock);
1771static struct task_struct *console_owner;
1772static bool console_waiter;
1773
1774/**
1775 * console_lock_spinning_enable - mark beginning of code where another
1776 *	thread might safely busy wait
1777 *
1778 * This basically converts console_lock into a spinlock. This marks
1779 * the section where the console_lock owner can not sleep, because
1780 * there may be a waiter spinning (like a spinlock). Also it must be
1781 * ready to hand over the lock at the end of the section.
1782 */
1783static void console_lock_spinning_enable(void)
1784{
1785	raw_spin_lock(&console_owner_lock);
1786	console_owner = current;
1787	raw_spin_unlock(&console_owner_lock);
1788
1789	/* The waiter may spin on us after setting console_owner */
1790	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1791}
1792
1793/**
1794 * console_lock_spinning_disable_and_check - mark end of code where another
1795 *	thread was able to busy wait and check if there is a waiter
 
1796 *
1797 * This is called at the end of the section where spinning is allowed.
1798 * It has two functions. First, it is a signal that it is no longer
1799 * safe to start busy waiting for the lock. Second, it checks if
1800 * there is a busy waiter and passes the lock rights to her.
1801 *
1802 * Important: Callers lose the lock if there was a busy waiter.
1803 *	They must not touch items synchronized by console_lock
1804 *	in this case.
1805 *
1806 * Return: 1 if the lock rights were passed, 0 otherwise.
1807 */
1808static int console_lock_spinning_disable_and_check(void)
1809{
1810	int waiter;
1811
1812	raw_spin_lock(&console_owner_lock);
1813	waiter = READ_ONCE(console_waiter);
1814	console_owner = NULL;
1815	raw_spin_unlock(&console_owner_lock);
1816
1817	if (!waiter) {
1818		spin_release(&console_owner_dep_map, _THIS_IP_);
1819		return 0;
1820	}
1821
1822	/* The waiter is now free to continue */
1823	WRITE_ONCE(console_waiter, false);
1824
1825	spin_release(&console_owner_dep_map, _THIS_IP_);
1826
1827	/*
 
 
 
 
 
 
1828	 * Hand off console_lock to waiter. The waiter will perform
1829	 * the up(). After this, the waiter is the console_lock owner.
1830	 */
1831	mutex_release(&console_lock_dep_map, _THIS_IP_);
1832	return 1;
1833}
1834
1835/**
1836 * console_trylock_spinning - try to get console_lock by busy waiting
1837 *
1838 * This allows to busy wait for the console_lock when the current
1839 * owner is running in specially marked sections. It means that
1840 * the current owner is running and cannot reschedule until it
1841 * is ready to lose the lock.
1842 *
1843 * Return: 1 if we got the lock, 0 othrewise
1844 */
1845static int console_trylock_spinning(void)
1846{
1847	struct task_struct *owner = NULL;
1848	bool waiter;
1849	bool spin = false;
1850	unsigned long flags;
1851
1852	if (console_trylock())
1853		return 1;
1854
 
 
 
 
 
 
 
 
 
 
1855	printk_safe_enter_irqsave(flags);
1856
1857	raw_spin_lock(&console_owner_lock);
1858	owner = READ_ONCE(console_owner);
1859	waiter = READ_ONCE(console_waiter);
1860	if (!waiter && owner && owner != current) {
1861		WRITE_ONCE(console_waiter, true);
1862		spin = true;
1863	}
1864	raw_spin_unlock(&console_owner_lock);
1865
1866	/*
1867	 * If there is an active printk() writing to the
1868	 * consoles, instead of having it write our data too,
1869	 * see if we can offload that load from the active
1870	 * printer, and do some printing ourselves.
1871	 * Go into a spin only if there isn't already a waiter
1872	 * spinning, and there is an active printer, and
1873	 * that active printer isn't us (recursive printk?).
1874	 */
1875	if (!spin) {
1876		printk_safe_exit_irqrestore(flags);
1877		return 0;
1878	}
1879
1880	/* We spin waiting for the owner to release us */
1881	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1882	/* Owner will clear console_waiter on hand off */
1883	while (READ_ONCE(console_waiter))
1884		cpu_relax();
1885	spin_release(&console_owner_dep_map, _THIS_IP_);
1886
1887	printk_safe_exit_irqrestore(flags);
1888	/*
1889	 * The owner passed the console lock to us.
1890	 * Since we did not spin on console lock, annotate
1891	 * this as a trylock. Otherwise lockdep will
1892	 * complain.
1893	 */
1894	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1895
1896	return 1;
1897}
1898
1899/*
1900 * Call the console drivers, asking them to write out
1901 * log_buf[start] to log_buf[end - 1].
1902 * The console_lock must be held.
1903 */
1904static void call_console_drivers(const char *ext_text, size_t ext_len,
1905				 const char *text, size_t len)
1906{
1907	static char dropped_text[64];
1908	size_t dropped_len = 0;
1909	struct console *con;
 
 
 
 
 
 
1910
1911	trace_console_rcuidle(text, len);
 
1912
1913	if (!console_drivers)
1914		return;
 
 
 
 
 
 
 
 
 
1915
1916	if (console_dropped) {
1917		dropped_len = snprintf(dropped_text, sizeof(dropped_text),
1918				       "** %lu printk messages dropped **\n",
1919				       console_dropped);
1920		console_dropped = 0;
1921	}
 
1922
1923	for_each_console(con) {
1924		if (exclusive_console && con != exclusive_console)
1925			continue;
1926		if (!(con->flags & CON_ENABLED))
1927			continue;
1928		if (!con->write)
1929			continue;
1930		if (!cpu_online(smp_processor_id()) &&
1931		    !(con->flags & CON_ANYTIME))
1932			continue;
1933		if (con->flags & CON_EXTENDED)
1934			con->write(con, ext_text, ext_len);
1935		else {
1936			if (dropped_len)
1937				con->write(con, dropped_text, dropped_len);
1938			con->write(con, text, len);
1939		}
1940	}
 
 
 
 
1941}
1942
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1943int printk_delay_msec __read_mostly;
1944
1945static inline void printk_delay(void)
1946{
 
 
1947	if (unlikely(printk_delay_msec)) {
1948		int m = printk_delay_msec;
1949
1950		while (m--) {
1951			mdelay(1);
1952			touch_nmi_watchdog();
1953		}
1954	}
1955}
1956
1957static inline u32 printk_caller_id(void)
1958{
1959	return in_task() ? task_pid_nr(current) :
1960		0x80000000 + raw_smp_processor_id();
1961}
1962
1963/**
1964 * parse_prefix - Parse level and control flags.
1965 *
1966 * @text:     The terminated text message.
1967 * @level:    A pointer to the current level value, will be updated.
1968 * @lflags:   A pointer to the current log flags, will be updated.
1969 *
1970 * @level may be NULL if the caller is not interested in the parsed value.
1971 * Otherwise the variable pointed to by @level must be set to
1972 * LOGLEVEL_DEFAULT in order to be updated with the parsed value.
1973 *
1974 * @lflags may be NULL if the caller is not interested in the parsed value.
1975 * Otherwise the variable pointed to by @lflags will be OR'd with the parsed
1976 * value.
1977 *
1978 * Return: The length of the parsed level and control flags.
1979 */
1980static u16 parse_prefix(char *text, int *level, enum log_flags *lflags)
 
1981{
1982	u16 prefix_len = 0;
1983	int kern_level;
1984
1985	while (*text) {
1986		kern_level = printk_get_level(text);
1987		if (!kern_level)
1988			break;
1989
1990		switch (kern_level) {
1991		case '0' ... '7':
1992			if (level && *level == LOGLEVEL_DEFAULT)
1993				*level = kern_level - '0';
1994			break;
1995		case 'c':	/* KERN_CONT */
1996			if (lflags)
1997				*lflags |= LOG_CONT;
1998		}
1999
2000		prefix_len += 2;
2001		text += 2;
2002	}
2003
2004	return prefix_len;
2005}
2006
2007static u16 printk_sprint(char *text, u16 size, int facility, enum log_flags *lflags,
2008			 const char *fmt, va_list args)
 
 
2009{
2010	u16 text_len;
2011
2012	text_len = vscnprintf(text, size, fmt, args);
2013
2014	/* Mark and strip a trailing newline. */
2015	if (text_len && text[text_len - 1] == '\n') {
2016		text_len--;
2017		*lflags |= LOG_NEWLINE;
2018	}
2019
2020	/* Strip log level and control flags. */
2021	if (facility == 0) {
2022		u16 prefix_len;
2023
2024		prefix_len = parse_prefix(text, NULL, NULL);
2025		if (prefix_len) {
2026			text_len -= prefix_len;
2027			memmove(text, text + prefix_len, text_len);
2028		}
2029	}
2030
 
 
2031	return text_len;
2032}
2033
2034__printf(4, 0)
2035int vprintk_store(int facility, int level,
2036		  const struct dev_printk_info *dev_info,
2037		  const char *fmt, va_list args)
2038{
2039	const u32 caller_id = printk_caller_id();
2040	struct prb_reserved_entry e;
2041	enum log_flags lflags = 0;
2042	struct printk_record r;
 
2043	u16 trunc_msg_len = 0;
2044	char prefix_buf[8];
 
2045	u16 reserve_size;
2046	va_list args2;
 
2047	u16 text_len;
 
2048	u64 ts_nsec;
2049
 
 
 
2050	/*
2051	 * Since the duration of printk() can vary depending on the message
2052	 * and state of the ringbuffer, grab the timestamp now so that it is
2053	 * close to the call of printk(). This provides a more deterministic
2054	 * timestamp with respect to the caller.
2055	 */
2056	ts_nsec = local_clock();
2057
 
 
2058	/*
2059	 * The sprintf needs to come first since the syslog prefix might be
2060	 * passed in as a parameter. An extra byte must be reserved so that
2061	 * later the vscnprintf() into the reserved buffer has room for the
2062	 * terminating '\0', which is not counted by vsnprintf().
2063	 */
2064	va_copy(args2, args);
2065	reserve_size = vsnprintf(&prefix_buf[0], sizeof(prefix_buf), fmt, args2) + 1;
2066	va_end(args2);
2067
2068	if (reserve_size > LOG_LINE_MAX)
2069		reserve_size = LOG_LINE_MAX;
2070
2071	/* Extract log level or control flags. */
2072	if (facility == 0)
2073		parse_prefix(&prefix_buf[0], &level, &lflags);
2074
2075	if (level == LOGLEVEL_DEFAULT)
2076		level = default_message_loglevel;
2077
2078	if (dev_info)
2079		lflags |= LOG_NEWLINE;
2080
2081	if (lflags & LOG_CONT) {
2082		prb_rec_init_wr(&r, reserve_size);
2083		if (prb_reserve_in_last(&e, prb, &r, caller_id, LOG_LINE_MAX)) {
2084			text_len = printk_sprint(&r.text_buf[r.info->text_len], reserve_size,
2085						 facility, &lflags, fmt, args);
2086			r.info->text_len += text_len;
2087
2088			if (lflags & LOG_NEWLINE) {
2089				r.info->flags |= LOG_NEWLINE;
2090				prb_final_commit(&e);
2091			} else {
2092				prb_commit(&e);
2093			}
2094
2095			return text_len;
 
2096		}
2097	}
2098
2099	/*
2100	 * Explicitly initialize the record before every prb_reserve() call.
2101	 * prb_reserve_in_last() and prb_reserve() purposely invalidate the
2102	 * structure when they fail.
2103	 */
2104	prb_rec_init_wr(&r, reserve_size);
2105	if (!prb_reserve(&e, prb, &r)) {
2106		/* truncate the message if it is too long for empty buffer */
2107		truncate_msg(&reserve_size, &trunc_msg_len);
2108
2109		prb_rec_init_wr(&r, reserve_size + trunc_msg_len);
2110		if (!prb_reserve(&e, prb, &r))
2111			return 0;
2112	}
2113
2114	/* fill message */
2115	text_len = printk_sprint(&r.text_buf[0], reserve_size, facility, &lflags, fmt, args);
2116	if (trunc_msg_len)
2117		memcpy(&r.text_buf[text_len], trunc_msg, trunc_msg_len);
2118	r.info->text_len = text_len + trunc_msg_len;
2119	r.info->facility = facility;
2120	r.info->level = level & 7;
2121	r.info->flags = lflags & 0x1f;
2122	r.info->ts_nsec = ts_nsec;
2123	r.info->caller_id = caller_id;
2124	if (dev_info)
2125		memcpy(&r.info->dev_info, dev_info, sizeof(r.info->dev_info));
2126
2127	/* A message without a trailing newline can be continued. */
2128	if (!(lflags & LOG_NEWLINE))
2129		prb_commit(&e);
2130	else
2131		prb_final_commit(&e);
2132
2133	return (text_len + trunc_msg_len);
 
 
 
2134}
2135
2136asmlinkage int vprintk_emit(int facility, int level,
2137			    const struct dev_printk_info *dev_info,
2138			    const char *fmt, va_list args)
2139{
2140	int printed_len;
2141	bool in_sched = false;
2142	unsigned long flags;
2143
2144	/* Suppress unimportant messages after panic happens */
2145	if (unlikely(suppress_printk))
2146		return 0;
2147
 
 
 
 
2148	if (level == LOGLEVEL_SCHED) {
2149		level = LOGLEVEL_DEFAULT;
2150		in_sched = true;
2151	}
2152
2153	boot_delay_msec(level);
2154	printk_delay();
2155
2156	printk_safe_enter_irqsave(flags);
2157	printed_len = vprintk_store(facility, level, dev_info, fmt, args);
2158	printk_safe_exit_irqrestore(flags);
2159
2160	/* If called from the scheduler, we can not call up(). */
2161	if (!in_sched) {
2162		/*
2163		 * Disable preemption to avoid being preempted while holding
2164		 * console_sem which would prevent anyone from printing to
2165		 * console
 
 
2166		 */
2167		preempt_disable();
2168		/*
2169		 * Try to acquire and then immediately release the console
2170		 * semaphore.  The release will print out buffers and wake up
2171		 * /dev/kmsg and syslog() users.
 
2172		 */
2173		if (console_trylock_spinning())
2174			console_unlock();
2175		preempt_enable();
2176	}
2177
2178	wake_up_klogd();
2179	return printed_len;
2180}
2181EXPORT_SYMBOL(vprintk_emit);
2182
2183int vprintk_default(const char *fmt, va_list args)
2184{
2185	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, fmt, args);
2186}
2187EXPORT_SYMBOL_GPL(vprintk_default);
2188
2189/**
2190 * printk - print a kernel message
2191 * @fmt: format string
2192 *
2193 * This is printk(). It can be called from any context. We want it to work.
2194 *
2195 * We try to grab the console_lock. If we succeed, it's easy - we log the
2196 * output and call the console drivers.  If we fail to get the semaphore, we
2197 * place the output into the log buffer and return. The current holder of
2198 * the console_sem will notice the new output in console_unlock(); and will
2199 * send it to the consoles before releasing the lock.
2200 *
2201 * One effect of this deferred printing is that code which calls printk() and
2202 * then changes console_loglevel may break. This is because console_loglevel
2203 * is inspected when the actual printing occurs.
2204 *
2205 * See also:
2206 * printf(3)
2207 *
2208 * See the vsnprintf() documentation for format string extensions over C99.
2209 */
2210asmlinkage __visible int printk(const char *fmt, ...)
2211{
2212	va_list args;
2213	int r;
2214
2215	va_start(args, fmt);
2216	r = vprintk(fmt, args);
2217	va_end(args);
2218
2219	return r;
2220}
2221EXPORT_SYMBOL(printk);
 
 
 
2222
2223#else /* CONFIG_PRINTK */
2224
2225#define CONSOLE_LOG_MAX		0
 
2226#define printk_time		false
2227
2228#define prb_read_valid(rb, seq, r)	false
2229#define prb_first_valid_seq(rb)		0
 
2230
2231static u64 syslog_seq;
2232static u64 console_seq;
2233static u64 exclusive_console_stop_seq;
2234static unsigned long console_dropped;
2235
2236static size_t record_print_text(const struct printk_record *r,
2237				bool syslog, bool time)
2238{
2239	return 0;
2240}
2241static ssize_t info_print_ext_header(char *buf, size_t size,
2242				     struct printk_info *info)
2243{
2244	return 0;
2245}
2246static ssize_t msg_print_ext_body(char *buf, size_t size,
2247				  char *text, size_t text_len,
2248				  struct dev_printk_info *dev_info) { return 0; }
2249static void console_lock_spinning_enable(void) { }
2250static int console_lock_spinning_disable_and_check(void) { return 0; }
2251static void call_console_drivers(const char *ext_text, size_t ext_len,
2252				 const char *text, size_t len) {}
 
 
2253static bool suppress_message_printing(int level) { return false; }
 
 
2254
2255#endif /* CONFIG_PRINTK */
2256
2257#ifdef CONFIG_EARLY_PRINTK
2258struct console *early_console;
2259
2260asmlinkage __visible void early_printk(const char *fmt, ...)
2261{
2262	va_list ap;
2263	char buf[512];
2264	int n;
2265
2266	if (!early_console)
2267		return;
2268
2269	va_start(ap, fmt);
2270	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2271	va_end(ap);
2272
2273	early_console->write(early_console, buf, n);
2274}
2275#endif
2276
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2277static int __add_preferred_console(char *name, int idx, char *options,
2278				   char *brl_options, bool user_specified)
2279{
2280	struct console_cmdline *c;
2281	int i;
2282
2283	/*
2284	 *	See if this tty is not yet registered, and
2285	 *	if we have a slot free.
2286	 */
2287	for (i = 0, c = console_cmdline;
2288	     i < MAX_CMDLINECONSOLES && c->name[0];
2289	     i++, c++) {
2290		if (strcmp(c->name, name) == 0 && c->index == idx) {
2291			if (!brl_options)
2292				preferred_console = i;
2293			if (user_specified)
2294				c->user_specified = true;
2295			return 0;
2296		}
2297	}
2298	if (i == MAX_CMDLINECONSOLES)
2299		return -E2BIG;
2300	if (!brl_options)
2301		preferred_console = i;
2302	strlcpy(c->name, name, sizeof(c->name));
2303	c->options = options;
2304	c->user_specified = user_specified;
2305	braille_set_options(c, brl_options);
2306
2307	c->index = idx;
2308	return 0;
2309}
2310
2311static int __init console_msg_format_setup(char *str)
2312{
2313	if (!strcmp(str, "syslog"))
2314		console_msg_format = MSG_FORMAT_SYSLOG;
2315	if (!strcmp(str, "default"))
2316		console_msg_format = MSG_FORMAT_DEFAULT;
2317	return 1;
2318}
2319__setup("console_msg_format=", console_msg_format_setup);
2320
2321/*
2322 * Set up a console.  Called via do_early_param() in init/main.c
2323 * for each "console=" parameter in the boot command line.
2324 */
2325static int __init console_setup(char *str)
2326{
2327	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2328	char *s, *options, *brl_options = NULL;
2329	int idx;
2330
2331	/*
2332	 * console="" or console=null have been suggested as a way to
2333	 * disable console output. Use ttynull that has been created
2334	 * for exactly this purpose.
2335	 */
2336	if (str[0] == 0 || strcmp(str, "null") == 0) {
2337		__add_preferred_console("ttynull", 0, NULL, NULL, true);
2338		return 1;
2339	}
2340
2341	if (_braille_console_setup(&str, &brl_options))
2342		return 1;
2343
2344	/*
2345	 * Decode str into name, index, options.
2346	 */
2347	if (str[0] >= '0' && str[0] <= '9') {
2348		strcpy(buf, "ttyS");
2349		strncpy(buf + 4, str, sizeof(buf) - 5);
2350	} else {
2351		strncpy(buf, str, sizeof(buf) - 1);
2352	}
2353	buf[sizeof(buf) - 1] = 0;
2354	options = strchr(str, ',');
2355	if (options)
2356		*(options++) = 0;
2357#ifdef __sparc__
2358	if (!strcmp(str, "ttya"))
2359		strcpy(buf, "ttyS0");
2360	if (!strcmp(str, "ttyb"))
2361		strcpy(buf, "ttyS1");
2362#endif
2363	for (s = buf; *s; s++)
2364		if (isdigit(*s) || *s == ',')
2365			break;
2366	idx = simple_strtoul(s, NULL, 10);
2367	*s = 0;
2368
2369	__add_preferred_console(buf, idx, options, brl_options, true);
2370	console_set_on_cmdline = 1;
2371	return 1;
2372}
2373__setup("console=", console_setup);
2374
2375/**
2376 * add_preferred_console - add a device to the list of preferred consoles.
2377 * @name: device name
2378 * @idx: device index
2379 * @options: options for this console
2380 *
2381 * The last preferred console added will be used for kernel messages
2382 * and stdin/out/err for init.  Normally this is used by console_setup
2383 * above to handle user-supplied console arguments; however it can also
2384 * be used by arch-specific code either to override the user or more
2385 * commonly to provide a default console (ie from PROM variables) when
2386 * the user has not supplied one.
2387 */
2388int add_preferred_console(char *name, int idx, char *options)
2389{
2390	return __add_preferred_console(name, idx, options, NULL, false);
2391}
2392
2393bool console_suspend_enabled = true;
2394EXPORT_SYMBOL(console_suspend_enabled);
2395
2396static int __init console_suspend_disable(char *str)
2397{
2398	console_suspend_enabled = false;
2399	return 1;
2400}
2401__setup("no_console_suspend", console_suspend_disable);
2402module_param_named(console_suspend, console_suspend_enabled,
2403		bool, S_IRUGO | S_IWUSR);
2404MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2405	" and hibernate operations");
2406
 
 
 
 
 
 
 
 
 
 
 
 
2407/**
2408 * suspend_console - suspend the console subsystem
2409 *
2410 * This disables printk() while we go into suspend states
2411 */
2412void suspend_console(void)
2413{
2414	if (!console_suspend_enabled)
2415		return;
2416	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
 
2417	console_lock();
2418	console_suspended = 1;
2419	up_console_sem();
2420}
2421
2422void resume_console(void)
2423{
2424	if (!console_suspend_enabled)
2425		return;
2426	down_console_sem();
2427	console_suspended = 0;
2428	console_unlock();
 
2429}
2430
2431/**
2432 * console_cpu_notify - print deferred console messages after CPU hotplug
2433 * @cpu: unused
2434 *
2435 * If printk() is called from a CPU that is not online yet, the messages
2436 * will be printed on the console only if there are CON_ANYTIME consoles.
2437 * This function is called when a new CPU comes online (or fails to come
2438 * up) or goes offline.
2439 */
2440static int console_cpu_notify(unsigned int cpu)
2441{
2442	if (!cpuhp_tasks_frozen) {
2443		/* If trylock fails, someone else is doing the printing */
2444		if (console_trylock())
2445			console_unlock();
2446	}
2447	return 0;
2448}
2449
2450/**
2451 * console_lock - lock the console system for exclusive use.
2452 *
2453 * Acquires a lock which guarantees that the caller has
2454 * exclusive access to the console system and the console_drivers list.
2455 *
2456 * Can sleep, returns nothing.
2457 */
2458void console_lock(void)
2459{
2460	might_sleep();
2461
2462	down_console_sem();
2463	if (console_suspended)
2464		return;
2465	console_locked = 1;
2466	console_may_schedule = 1;
2467}
2468EXPORT_SYMBOL(console_lock);
2469
2470/**
2471 * console_trylock - try to lock the console system for exclusive use.
2472 *
2473 * Try to acquire a lock which guarantees that the caller has exclusive
2474 * access to the console system and the console_drivers list.
2475 *
2476 * returns 1 on success, and 0 on failure to acquire the lock.
2477 */
2478int console_trylock(void)
2479{
2480	if (down_trylock_console_sem())
2481		return 0;
2482	if (console_suspended) {
2483		up_console_sem();
2484		return 0;
2485	}
2486	console_locked = 1;
2487	console_may_schedule = 0;
2488	return 1;
2489}
2490EXPORT_SYMBOL(console_trylock);
2491
2492int is_console_locked(void)
2493{
2494	return console_locked;
2495}
2496EXPORT_SYMBOL(is_console_locked);
2497
2498/*
2499 * Check if we have any console that is capable of printing while cpu is
2500 * booting or shutting down. Requires console_sem.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2501 */
2502static int have_callable_console(void)
2503{
2504	struct console *con;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2505
2506	for_each_console(con)
2507		if ((con->flags & CON_ENABLED) &&
2508				(con->flags & CON_ANYTIME))
2509			return 1;
2510
2511	return 0;
 
 
 
2512}
2513
2514/*
2515 * Can we actually use the console at this time on this cpu?
 
 
 
 
 
 
 
2516 *
2517 * Console drivers may assume that per-cpu resources have been allocated. So
2518 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2519 * call them until this CPU is officially up.
 
 
 
 
 
 
 
 
 
2520 */
2521static inline int can_use_console(void)
2522{
2523	return cpu_online(raw_smp_processor_id()) || have_callable_console();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2524}
2525
2526/**
2527 * console_unlock - unlock the console system
2528 *
2529 * Releases the console_lock which the caller holds on the console system
2530 * and the console driver list.
2531 *
2532 * While the console_lock was held, console output may have been buffered
2533 * by printk().  If this is the case, console_unlock(); emits
2534 * the output prior to releasing the lock.
2535 *
2536 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2537 *
2538 * console_unlock(); may be called from any context.
2539 */
2540void console_unlock(void)
2541{
2542	static char ext_text[CONSOLE_EXT_LOG_MAX];
2543	static char text[CONSOLE_LOG_MAX];
2544	unsigned long flags;
2545	bool do_cond_resched, retry;
2546	struct printk_info info;
2547	struct printk_record r;
2548	u64 __maybe_unused next_seq;
2549
2550	if (console_suspended) {
2551		up_console_sem();
2552		return;
2553	}
2554
2555	prb_rec_init_rd(&r, &info, text, sizeof(text));
2556
2557	/*
2558	 * Console drivers are called with interrupts disabled, so
2559	 * @console_may_schedule should be cleared before; however, we may
2560	 * end up dumping a lot of lines, for example, if called from
2561	 * console registration path, and should invoke cond_resched()
2562	 * between lines if allowable.  Not doing so can cause a very long
2563	 * scheduling stall on a slow console leading to RCU stall and
2564	 * softlockup warnings which exacerbate the issue with more
2565	 * messages practically incapacitating the system.
2566	 *
2567	 * console_trylock() is not able to detect the preemptive
2568	 * context reliably. Therefore the value must be stored before
2569	 * and cleared after the "again" goto label.
2570	 */
2571	do_cond_resched = console_may_schedule;
2572again:
2573	console_may_schedule = 0;
2574
2575	/*
2576	 * We released the console_sem lock, so we need to recheck if
2577	 * cpu is online and (if not) is there at least one CON_ANYTIME
2578	 * console.
2579	 */
2580	if (!can_use_console()) {
2581		console_locked = 0;
2582		up_console_sem();
2583		return;
2584	}
2585
2586	for (;;) {
2587		size_t ext_len = 0;
2588		size_t len;
2589
2590		printk_safe_enter_irqsave(flags);
2591skip:
2592		if (!prb_read_valid(prb, console_seq, &r))
2593			break;
2594
2595		if (console_seq != r.info->seq) {
2596			console_dropped += r.info->seq - console_seq;
2597			console_seq = r.info->seq;
2598		}
2599
2600		if (suppress_message_printing(r.info->level)) {
2601			/*
2602			 * Skip record we have buffered and already printed
2603			 * directly to the console when we received it, and
2604			 * record that has level above the console loglevel.
2605			 */
2606			console_seq++;
2607			goto skip;
2608		}
2609
2610		/* Output to all consoles once old messages replayed. */
2611		if (unlikely(exclusive_console &&
2612			     console_seq >= exclusive_console_stop_seq)) {
2613			exclusive_console = NULL;
2614		}
2615
2616		/*
2617		 * Handle extended console text first because later
2618		 * record_print_text() will modify the record buffer in-place.
 
 
2619		 */
2620		if (nr_ext_console_drivers) {
2621			ext_len = info_print_ext_header(ext_text,
2622						sizeof(ext_text),
2623						r.info);
2624			ext_len += msg_print_ext_body(ext_text + ext_len,
2625						sizeof(ext_text) - ext_len,
2626						&r.text_buf[0],
2627						r.info->text_len,
2628						&r.info->dev_info);
2629		}
2630		len = record_print_text(&r,
2631				console_msg_format & MSG_FORMAT_SYSLOG,
2632				printk_time);
2633		console_seq++;
2634
2635		/*
2636		 * While actively printing out messages, if another printk()
2637		 * were to occur on another CPU, it may wait for this one to
2638		 * finish. This task can not be preempted if there is a
2639		 * waiter waiting to take over.
2640		 */
2641		console_lock_spinning_enable();
2642
2643		stop_critical_timings();	/* don't trace print latency */
2644		call_console_drivers(ext_text, ext_len, text, len);
2645		start_critical_timings();
2646
2647		if (console_lock_spinning_disable_and_check()) {
2648			printk_safe_exit_irqrestore(flags);
2649			return;
2650		}
2651
2652		printk_safe_exit_irqrestore(flags);
2653
2654		if (do_cond_resched)
2655			cond_resched();
2656	}
2657
2658	/* Get consistent value of the next-to-be-used sequence number. */
2659	next_seq = console_seq;
2660
2661	console_locked = 0;
2662	up_console_sem();
2663
2664	/*
2665	 * Someone could have filled up the buffer again, so re-check if there's
2666	 * something to flush. In case we cannot trylock the console_sem again,
2667	 * there's a new owner and the console_unlock() from them will do the
2668	 * flush, no worries.
2669	 */
2670	retry = prb_read_valid(prb, next_seq, NULL);
2671	printk_safe_exit_irqrestore(flags);
2672
2673	if (retry && console_trylock())
2674		goto again;
2675}
2676EXPORT_SYMBOL(console_unlock);
2677
2678/**
2679 * console_conditional_schedule - yield the CPU if required
2680 *
2681 * If the console code is currently allowed to sleep, and
2682 * if this CPU should yield the CPU to another task, do
2683 * so here.
2684 *
2685 * Must be called within console_lock();.
2686 */
2687void __sched console_conditional_schedule(void)
2688{
2689	if (console_may_schedule)
2690		cond_resched();
2691}
2692EXPORT_SYMBOL(console_conditional_schedule);
2693
2694void console_unblank(void)
2695{
2696	struct console *c;
 
2697
2698	/*
2699	 * console_unblank can no longer be called in interrupt context unless
2700	 * oops_in_progress is set to 1..
 
 
 
2701	 */
2702	if (oops_in_progress) {
2703		if (down_trylock_console_sem() != 0)
2704			return;
2705	} else
2706		console_lock();
2707
2708	console_locked = 1;
2709	console_may_schedule = 0;
2710	for_each_console(c)
2711		if ((c->flags & CON_ENABLED) && c->unblank)
 
 
2712			c->unblank();
 
 
 
2713	console_unlock();
 
 
 
2714}
2715
2716/**
2717 * console_flush_on_panic - flush console content on panic
2718 * @mode: flush all messages in buffer or just the pending ones
2719 *
2720 * Immediately output all pending messages no matter what.
2721 */
2722void console_flush_on_panic(enum con_flush_mode mode)
2723{
2724	/*
2725	 * If someone else is holding the console lock, trylock will fail
2726	 * and may_schedule may be set.  Ignore and proceed to unlock so
2727	 * that messages are flushed out.  As this can be called from any
2728	 * context and we don't want to get preempted while flushing,
2729	 * ensure may_schedule is cleared.
2730	 */
2731	console_trylock();
2732	console_may_schedule = 0;
2733
2734	if (mode == CONSOLE_REPLAY_ALL) {
2735		unsigned long flags;
 
 
 
 
2736
2737		printk_safe_enter_irqsave(flags);
2738		console_seq = prb_first_valid_seq(prb);
2739		printk_safe_exit_irqrestore(flags);
 
 
 
 
 
 
 
2740	}
2741	console_unlock();
2742}
2743
2744/*
2745 * Return the console tty driver structure and its associated index
2746 */
2747struct tty_driver *console_device(int *index)
2748{
2749	struct console *c;
2750	struct tty_driver *driver = NULL;
 
2751
 
 
 
 
 
2752	console_lock();
2753	for_each_console(c) {
 
 
2754		if (!c->device)
2755			continue;
2756		driver = c->device(c, index);
2757		if (driver)
2758			break;
2759	}
 
 
2760	console_unlock();
2761	return driver;
2762}
2763
2764/*
2765 * Prevent further output on the passed console device so that (for example)
2766 * serial drivers can disable console output before suspending a port, and can
2767 * re-enable output afterwards.
2768 */
2769void console_stop(struct console *console)
2770{
2771	console_lock();
2772	console->flags &= ~CON_ENABLED;
2773	console_unlock();
 
 
 
 
 
 
 
 
 
2774}
2775EXPORT_SYMBOL(console_stop);
2776
2777void console_start(struct console *console)
2778{
2779	console_lock();
2780	console->flags |= CON_ENABLED;
2781	console_unlock();
 
2782}
2783EXPORT_SYMBOL(console_start);
2784
2785static int __read_mostly keep_bootcon;
2786
2787static int __init keep_bootcon_setup(char *str)
2788{
2789	keep_bootcon = 1;
2790	pr_info("debug: skip boot console de-registration.\n");
2791
2792	return 0;
2793}
2794
2795early_param("keep_bootcon", keep_bootcon_setup);
2796
2797/*
2798 * This is called by register_console() to try to match
2799 * the newly registered console with any of the ones selected
2800 * by either the command line or add_preferred_console() and
2801 * setup/enable it.
2802 *
2803 * Care need to be taken with consoles that are statically
2804 * enabled such as netconsole
2805 */
2806static int try_enable_new_console(struct console *newcon, bool user_specified)
 
2807{
2808	struct console_cmdline *c;
2809	int i, err;
2810
2811	for (i = 0, c = console_cmdline;
2812	     i < MAX_CMDLINECONSOLES && c->name[0];
2813	     i++, c++) {
2814		if (c->user_specified != user_specified)
2815			continue;
2816		if (!newcon->match ||
2817		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
2818			/* default matching */
2819			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2820			if (strcmp(c->name, newcon->name) != 0)
2821				continue;
2822			if (newcon->index >= 0 &&
2823			    newcon->index != c->index)
2824				continue;
2825			if (newcon->index < 0)
2826				newcon->index = c->index;
2827
2828			if (_braille_register_console(newcon, c))
2829				return 0;
2830
2831			if (newcon->setup &&
2832			    (err = newcon->setup(newcon, c->options)) != 0)
2833				return err;
2834		}
2835		newcon->flags |= CON_ENABLED;
2836		if (i == preferred_console) {
2837			newcon->flags |= CON_CONSDEV;
2838			has_preferred_console = true;
2839		}
2840		return 0;
2841	}
2842
2843	/*
2844	 * Some consoles, such as pstore and netconsole, can be enabled even
2845	 * without matching. Accept the pre-enabled consoles only when match()
2846	 * and setup() had a chance to be called.
2847	 */
2848	if (newcon->flags & CON_ENABLED && c->user_specified ==	user_specified)
2849		return 0;
2850
2851	return -ENOENT;
2852}
2853
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2854/*
2855 * The console driver calls this routine during kernel initialization
2856 * to register the console printing procedure with printk() and to
2857 * print any messages that were printed by the kernel before the
2858 * console driver was initialized.
2859 *
2860 * This can happen pretty early during the boot process (because of
2861 * early_printk) - sometimes before setup_arch() completes - be careful
2862 * of what kernel features are used - they may not be initialised yet.
2863 *
2864 * There are two types of consoles - bootconsoles (early_printk) and
2865 * "real" consoles (everything which is not a bootconsole) which are
2866 * handled differently.
2867 *  - Any number of bootconsoles can be registered at any time.
2868 *  - As soon as a "real" console is registered, all bootconsoles
2869 *    will be unregistered automatically.
2870 *  - Once a "real" console is registered, any attempt to register a
2871 *    bootconsoles will be rejected
2872 */
2873void register_console(struct console *newcon)
2874{
2875	unsigned long flags;
2876	struct console *bcon = NULL;
 
2877	int err;
2878
2879	for_each_console(bcon) {
2880		if (WARN(bcon == newcon, "console '%s%d' already registered\n",
2881					 bcon->name, bcon->index))
2882			return;
2883	}
2884
2885	/*
2886	 * before we register a new CON_BOOT console, make sure we don't
2887	 * already have a valid console
2888	 */
2889	if (newcon->flags & CON_BOOT) {
2890		for_each_console(bcon) {
2891			if (!(bcon->flags & CON_BOOT)) {
2892				pr_info("Too late to register bootconsole %s%d\n",
2893					newcon->name, newcon->index);
2894				return;
2895			}
2896		}
 
 
 
 
 
2897	}
2898
2899	if (console_drivers && console_drivers->flags & CON_BOOT)
2900		bcon = console_drivers;
2901
2902	if (!has_preferred_console || bcon || !console_drivers)
2903		has_preferred_console = preferred_console >= 0;
 
2904
2905	/*
2906	 *	See if we want to use this console driver. If we
2907	 *	didn't select a console we take the first one
2908	 *	that registers here.
2909	 */
2910	if (!has_preferred_console) {
2911		if (newcon->index < 0)
2912			newcon->index = 0;
2913		if (newcon->setup == NULL ||
2914		    newcon->setup(newcon, NULL) == 0) {
2915			newcon->flags |= CON_ENABLED;
2916			if (newcon->device) {
2917				newcon->flags |= CON_CONSDEV;
2918				has_preferred_console = true;
2919			}
 
2920		}
2921	}
2922
2923	/* See if this console matches one we selected on the command line */
2924	err = try_enable_new_console(newcon, true);
2925
2926	/* If not, try to match against the platform default(s) */
2927	if (err == -ENOENT)
2928		err = try_enable_new_console(newcon, false);
2929
2930	/* printk() messages are not printed to the Braille console. */
2931	if (err || newcon->flags & CON_BRL)
2932		return;
2933
2934	/*
2935	 * If we have a bootconsole, and are switching to a real console,
2936	 * don't print everything out again, since when the boot console, and
2937	 * the real console are the same physical device, it's annoying to
2938	 * see the beginning boot messages twice
2939	 */
2940	if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
 
2941		newcon->flags &= ~CON_PRINTBUFFER;
 
 
 
 
2942
2943	/*
2944	 *	Put this console in the list - keep the
2945	 *	preferred driver at the head of the list.
2946	 */
2947	console_lock();
2948	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2949		newcon->next = console_drivers;
2950		console_drivers = newcon;
2951		if (newcon->next)
2952			newcon->next->flags &= ~CON_CONSDEV;
2953		/* Ensure this flag is always set for the head of the list */
2954		newcon->flags |= CON_CONSDEV;
 
 
 
 
 
 
 
2955	} else {
2956		newcon->next = console_drivers->next;
2957		console_drivers->next = newcon;
2958	}
2959
2960	if (newcon->flags & CON_EXTENDED)
2961		nr_ext_console_drivers++;
2962
2963	if (newcon->flags & CON_PRINTBUFFER) {
2964		/*
2965		 * console_unlock(); will print out the buffered messages
2966		 * for us.
2967		 *
2968		 * We're about to replay the log buffer.  Only do this to the
2969		 * just-registered console to avoid excessive message spam to
2970		 * the already-registered consoles.
2971		 *
2972		 * Set exclusive_console with disabled interrupts to reduce
2973		 * race window with eventual console_flush_on_panic() that
2974		 * ignores console_lock.
2975		 */
2976		exclusive_console = newcon;
2977		exclusive_console_stop_seq = console_seq;
2978
2979		/* Get a consistent copy of @syslog_seq. */
2980		raw_spin_lock_irqsave(&syslog_lock, flags);
2981		console_seq = syslog_seq;
2982		raw_spin_unlock_irqrestore(&syslog_lock, flags);
2983	}
2984	console_unlock();
2985	console_sysfs_notify();
2986
2987	/*
2988	 * By unregistering the bootconsoles after we enable the real console
2989	 * we get the "console xxx enabled" message on all the consoles -
2990	 * boot consoles, real consoles, etc - this is to ensure that end
2991	 * users know there might be something in the kernel's log buffer that
2992	 * went to the bootconsole (that they do not see on the real console)
2993	 */
2994	pr_info("%sconsole [%s%d] enabled\n",
2995		(newcon->flags & CON_BOOT) ? "boot" : "" ,
2996		newcon->name, newcon->index);
2997	if (bcon &&
2998	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2999	    !keep_bootcon) {
3000		/* We need to iterate through all boot consoles, to make
3001		 * sure we print everything out, before we unregister them.
3002		 */
3003		for_each_console(bcon)
3004			if (bcon->flags & CON_BOOT)
3005				unregister_console(bcon);
3006	}
 
 
3007}
3008EXPORT_SYMBOL(register_console);
3009
3010int unregister_console(struct console *console)
 
3011{
3012	struct console *con;
3013	int res;
3014
3015	pr_info("%sconsole [%s%d] disabled\n",
3016		(console->flags & CON_BOOT) ? "boot" : "" ,
3017		console->name, console->index);
3018
3019	res = _braille_unregister_console(console);
3020	if (res < 0)
3021		return res;
3022	if (res > 0)
3023		return 0;
3024
3025	res = -ENODEV;
3026	console_lock();
3027	if (console_drivers == console) {
3028		console_drivers=console->next;
3029		res = 0;
3030	} else {
3031		for_each_console(con) {
3032			if (con->next == console) {
3033				con->next = console->next;
3034				res = 0;
3035				break;
3036			}
3037		}
3038	}
3039
3040	if (res)
3041		goto out_disable_unlock;
3042
3043	if (console->flags & CON_EXTENDED)
3044		nr_ext_console_drivers--;
3045
3046	/*
 
3047	 * If this isn't the last console and it has CON_CONSDEV set, we
3048	 * need to set it on the next preferred console.
 
 
 
 
 
 
 
 
 
 
 
 
3049	 */
3050	if (console_drivers != NULL && console->flags & CON_CONSDEV)
3051		console_drivers->flags |= CON_CONSDEV;
3052
3053	console->flags &= ~CON_ENABLED;
3054	console_unlock();
3055	console_sysfs_notify();
3056
3057	if (console->exit)
3058		res = console->exit(console);
3059
3060	return res;
 
3061
3062out_disable_unlock:
3063	console->flags &= ~CON_ENABLED;
3064	console_unlock();
3065
 
 
 
3066	return res;
3067}
3068EXPORT_SYMBOL(unregister_console);
3069
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3070/*
3071 * Initialize the console device. This is called *early*, so
3072 * we can't necessarily depend on lots of kernel help here.
3073 * Just do some early initializations, and do the complex setup
3074 * later.
3075 */
3076void __init console_init(void)
3077{
3078	int ret;
3079	initcall_t call;
3080	initcall_entry_t *ce;
3081
3082	/* Setup the default TTY line discipline. */
3083	n_tty_init();
3084
3085	/*
3086	 * set up the console device so that later boot sequences can
3087	 * inform about problems etc..
3088	 */
3089	ce = __con_initcall_start;
3090	trace_initcall_level("console");
3091	while (ce < __con_initcall_end) {
3092		call = initcall_from_entry(ce);
3093		trace_initcall_start(call);
3094		ret = call();
3095		trace_initcall_finish(call, ret);
3096		ce++;
3097	}
3098}
3099
3100/*
3101 * Some boot consoles access data that is in the init section and which will
3102 * be discarded after the initcalls have been run. To make sure that no code
3103 * will access this data, unregister the boot consoles in a late initcall.
3104 *
3105 * If for some reason, such as deferred probe or the driver being a loadable
3106 * module, the real console hasn't registered yet at this point, there will
3107 * be a brief interval in which no messages are logged to the console, which
3108 * makes it difficult to diagnose problems that occur during this time.
3109 *
3110 * To mitigate this problem somewhat, only unregister consoles whose memory
3111 * intersects with the init section. Note that all other boot consoles will
3112 * get unregistered when the real preferred console is registered.
3113 */
3114static int __init printk_late_init(void)
3115{
 
3116	struct console *con;
3117	int ret;
3118
3119	for_each_console(con) {
 
3120		if (!(con->flags & CON_BOOT))
3121			continue;
3122
3123		/* Check addresses that might be used for enabled consoles. */
3124		if (init_section_intersects(con, sizeof(*con)) ||
3125		    init_section_contains(con->write, 0) ||
3126		    init_section_contains(con->read, 0) ||
3127		    init_section_contains(con->device, 0) ||
3128		    init_section_contains(con->unblank, 0) ||
3129		    init_section_contains(con->data, 0)) {
3130			/*
3131			 * Please, consider moving the reported consoles out
3132			 * of the init section.
3133			 */
3134			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
3135				con->name, con->index);
3136			unregister_console(con);
3137		}
3138	}
 
 
3139	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
3140					console_cpu_notify);
3141	WARN_ON(ret < 0);
3142	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
3143					console_cpu_notify, NULL);
3144	WARN_ON(ret < 0);
 
3145	return 0;
3146}
3147late_initcall(printk_late_init);
3148
3149#if defined CONFIG_PRINTK
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3150/*
3151 * Delayed printk version, for scheduler-internal messages:
3152 */
3153#define PRINTK_PENDING_WAKEUP	0x01
3154#define PRINTK_PENDING_OUTPUT	0x02
3155
3156static DEFINE_PER_CPU(int, printk_pending);
3157
3158static void wake_up_klogd_work_func(struct irq_work *irq_work)
3159{
3160	int pending = __this_cpu_xchg(printk_pending, 0);
3161
3162	if (pending & PRINTK_PENDING_OUTPUT) {
3163		/* If trylock fails, someone else is doing the printing */
3164		if (console_trylock())
3165			console_unlock();
3166	}
3167
3168	if (pending & PRINTK_PENDING_WAKEUP)
3169		wake_up_interruptible(&log_wait);
3170}
3171
3172static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) =
3173	IRQ_WORK_INIT_LAZY(wake_up_klogd_work_func);
3174
3175void wake_up_klogd(void)
3176{
3177	if (!printk_percpu_data_ready())
3178		return;
3179
3180	preempt_disable();
3181	if (waitqueue_active(&log_wait)) {
3182		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
 
 
 
 
 
 
 
 
 
 
 
 
3183		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3184	}
3185	preempt_enable();
3186}
3187
 
 
 
 
 
3188void defer_console_output(void)
3189{
3190	if (!printk_percpu_data_ready())
3191		return;
 
 
 
 
3192
3193	preempt_disable();
3194	__this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
3195	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
3196	preempt_enable();
3197}
3198
3199int vprintk_deferred(const char *fmt, va_list args)
3200{
3201	int r;
3202
3203	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, fmt, args);
3204	defer_console_output();
3205
3206	return r;
3207}
3208
3209int printk_deferred(const char *fmt, ...)
3210{
3211	va_list args;
3212	int r;
3213
3214	va_start(args, fmt);
3215	r = vprintk_deferred(fmt, args);
3216	va_end(args);
3217
3218	return r;
3219}
3220
3221/*
3222 * printk rate limiting, lifted from the networking subsystem.
3223 *
3224 * This enforces a rate limit: not more than 10 kernel messages
3225 * every 5s to make a denial-of-service attack impossible.
3226 */
3227DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
3228
3229int __printk_ratelimit(const char *func)
3230{
3231	return ___ratelimit(&printk_ratelimit_state, func);
3232}
3233EXPORT_SYMBOL(__printk_ratelimit);
3234
3235/**
3236 * printk_timed_ratelimit - caller-controlled printk ratelimiting
3237 * @caller_jiffies: pointer to caller's state
3238 * @interval_msecs: minimum interval between prints
3239 *
3240 * printk_timed_ratelimit() returns true if more than @interval_msecs
3241 * milliseconds have elapsed since the last time printk_timed_ratelimit()
3242 * returned true.
3243 */
3244bool printk_timed_ratelimit(unsigned long *caller_jiffies,
3245			unsigned int interval_msecs)
3246{
3247	unsigned long elapsed = jiffies - *caller_jiffies;
3248
3249	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
3250		return false;
3251
3252	*caller_jiffies = jiffies;
3253	return true;
3254}
3255EXPORT_SYMBOL(printk_timed_ratelimit);
3256
3257static DEFINE_SPINLOCK(dump_list_lock);
3258static LIST_HEAD(dump_list);
3259
3260/**
3261 * kmsg_dump_register - register a kernel log dumper.
3262 * @dumper: pointer to the kmsg_dumper structure
3263 *
3264 * Adds a kernel log dumper to the system. The dump callback in the
3265 * structure will be called when the kernel oopses or panics and must be
3266 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
3267 */
3268int kmsg_dump_register(struct kmsg_dumper *dumper)
3269{
3270	unsigned long flags;
3271	int err = -EBUSY;
3272
3273	/* The dump callback needs to be set */
3274	if (!dumper->dump)
3275		return -EINVAL;
3276
3277	spin_lock_irqsave(&dump_list_lock, flags);
3278	/* Don't allow registering multiple times */
3279	if (!dumper->registered) {
3280		dumper->registered = 1;
3281		list_add_tail_rcu(&dumper->list, &dump_list);
3282		err = 0;
3283	}
3284	spin_unlock_irqrestore(&dump_list_lock, flags);
3285
3286	return err;
3287}
3288EXPORT_SYMBOL_GPL(kmsg_dump_register);
3289
3290/**
3291 * kmsg_dump_unregister - unregister a kmsg dumper.
3292 * @dumper: pointer to the kmsg_dumper structure
3293 *
3294 * Removes a dump device from the system. Returns zero on success and
3295 * %-EINVAL otherwise.
3296 */
3297int kmsg_dump_unregister(struct kmsg_dumper *dumper)
3298{
3299	unsigned long flags;
3300	int err = -EINVAL;
3301
3302	spin_lock_irqsave(&dump_list_lock, flags);
3303	if (dumper->registered) {
3304		dumper->registered = 0;
3305		list_del_rcu(&dumper->list);
3306		err = 0;
3307	}
3308	spin_unlock_irqrestore(&dump_list_lock, flags);
3309	synchronize_rcu();
3310
3311	return err;
3312}
3313EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3314
3315static bool always_kmsg_dump;
3316module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3317
3318const char *kmsg_dump_reason_str(enum kmsg_dump_reason reason)
3319{
3320	switch (reason) {
3321	case KMSG_DUMP_PANIC:
3322		return "Panic";
3323	case KMSG_DUMP_OOPS:
3324		return "Oops";
3325	case KMSG_DUMP_EMERG:
3326		return "Emergency";
3327	case KMSG_DUMP_SHUTDOWN:
3328		return "Shutdown";
3329	default:
3330		return "Unknown";
3331	}
3332}
3333EXPORT_SYMBOL_GPL(kmsg_dump_reason_str);
3334
3335/**
3336 * kmsg_dump - dump kernel log to kernel message dumpers.
3337 * @reason: the reason (oops, panic etc) for dumping
3338 *
3339 * Call each of the registered dumper's dump() callback, which can
3340 * retrieve the kmsg records with kmsg_dump_get_line() or
3341 * kmsg_dump_get_buffer().
3342 */
3343void kmsg_dump(enum kmsg_dump_reason reason)
3344{
3345	struct kmsg_dumper *dumper;
3346
3347	rcu_read_lock();
3348	list_for_each_entry_rcu(dumper, &dump_list, list) {
3349		enum kmsg_dump_reason max_reason = dumper->max_reason;
3350
3351		/*
3352		 * If client has not provided a specific max_reason, default
3353		 * to KMSG_DUMP_OOPS, unless always_kmsg_dump was set.
3354		 */
3355		if (max_reason == KMSG_DUMP_UNDEF) {
3356			max_reason = always_kmsg_dump ? KMSG_DUMP_MAX :
3357							KMSG_DUMP_OOPS;
3358		}
3359		if (reason > max_reason)
3360			continue;
3361
3362		/* invoke dumper which will iterate over records */
3363		dumper->dump(dumper, reason);
3364	}
3365	rcu_read_unlock();
3366}
3367
3368/**
3369 * kmsg_dump_get_line - retrieve one kmsg log line
3370 * @iter: kmsg dump iterator
3371 * @syslog: include the "<4>" prefixes
3372 * @line: buffer to copy the line to
3373 * @size: maximum size of the buffer
3374 * @len: length of line placed into buffer
3375 *
3376 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3377 * record, and copy one record into the provided buffer.
3378 *
3379 * Consecutive calls will return the next available record moving
3380 * towards the end of the buffer with the youngest messages.
3381 *
3382 * A return value of FALSE indicates that there are no more records to
3383 * read.
3384 */
3385bool kmsg_dump_get_line(struct kmsg_dump_iter *iter, bool syslog,
3386			char *line, size_t size, size_t *len)
3387{
3388	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3389	struct printk_info info;
3390	unsigned int line_count;
3391	struct printk_record r;
3392	unsigned long flags;
3393	size_t l = 0;
3394	bool ret = false;
3395
3396	if (iter->cur_seq < min_seq)
3397		iter->cur_seq = min_seq;
3398
3399	printk_safe_enter_irqsave(flags);
3400	prb_rec_init_rd(&r, &info, line, size);
3401
3402	/* Read text or count text lines? */
3403	if (line) {
3404		if (!prb_read_valid(prb, iter->cur_seq, &r))
3405			goto out;
3406		l = record_print_text(&r, syslog, printk_time);
3407	} else {
3408		if (!prb_read_valid_info(prb, iter->cur_seq,
3409					 &info, &line_count)) {
3410			goto out;
3411		}
3412		l = get_record_print_text_size(&info, line_count, syslog,
3413					       printk_time);
3414
3415	}
3416
3417	iter->cur_seq = r.info->seq + 1;
3418	ret = true;
3419out:
3420	printk_safe_exit_irqrestore(flags);
3421	if (len)
3422		*len = l;
3423	return ret;
3424}
3425EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3426
3427/**
3428 * kmsg_dump_get_buffer - copy kmsg log lines
3429 * @iter: kmsg dump iterator
3430 * @syslog: include the "<4>" prefixes
3431 * @buf: buffer to copy the line to
3432 * @size: maximum size of the buffer
3433 * @len_out: length of line placed into buffer
3434 *
3435 * Start at the end of the kmsg buffer and fill the provided buffer
3436 * with as many of the *youngest* kmsg records that fit into it.
3437 * If the buffer is large enough, all available kmsg records will be
3438 * copied with a single call.
3439 *
3440 * Consecutive calls will fill the buffer with the next block of
3441 * available older records, not including the earlier retrieved ones.
3442 *
3443 * A return value of FALSE indicates that there are no more records to
3444 * read.
3445 */
3446bool kmsg_dump_get_buffer(struct kmsg_dump_iter *iter, bool syslog,
3447			  char *buf, size_t size, size_t *len_out)
3448{
3449	u64 min_seq = latched_seq_read_nolock(&clear_seq);
3450	struct printk_info info;
3451	struct printk_record r;
3452	unsigned long flags;
3453	u64 seq;
3454	u64 next_seq;
3455	size_t len = 0;
3456	bool ret = false;
3457	bool time = printk_time;
3458
3459	if (!buf || !size)
3460		goto out;
3461
3462	if (iter->cur_seq < min_seq)
3463		iter->cur_seq = min_seq;
3464
3465	printk_safe_enter_irqsave(flags);
3466	if (prb_read_valid_info(prb, iter->cur_seq, &info, NULL)) {
3467		if (info.seq != iter->cur_seq) {
3468			/* messages are gone, move to first available one */
3469			iter->cur_seq = info.seq;
3470		}
3471	}
3472
3473	/* last entry */
3474	if (iter->cur_seq >= iter->next_seq) {
3475		printk_safe_exit_irqrestore(flags);
3476		goto out;
3477	}
3478
3479	/*
3480	 * Find first record that fits, including all following records,
3481	 * into the user-provided buffer for this dump. Pass in size-1
3482	 * because this function (by way of record_print_text()) will
3483	 * not write more than size-1 bytes of text into @buf.
3484	 */
3485	seq = find_first_fitting_seq(iter->cur_seq, iter->next_seq,
3486				     size - 1, syslog, time);
3487
3488	/*
3489	 * Next kmsg_dump_get_buffer() invocation will dump block of
3490	 * older records stored right before this one.
3491	 */
3492	next_seq = seq;
3493
3494	prb_rec_init_rd(&r, &info, buf, size);
3495
3496	len = 0;
3497	prb_for_each_record(seq, prb, seq, &r) {
3498		if (r.info->seq >= iter->next_seq)
3499			break;
3500
3501		len += record_print_text(&r, syslog, time);
3502
3503		/* Adjust record to store to remaining buffer space. */
3504		prb_rec_init_rd(&r, &info, buf + len, size - len);
3505	}
3506
3507	iter->next_seq = next_seq;
3508	ret = true;
3509	printk_safe_exit_irqrestore(flags);
3510out:
3511	if (len_out)
3512		*len_out = len;
3513	return ret;
3514}
3515EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3516
3517/**
3518 * kmsg_dump_rewind - reset the iterator
3519 * @iter: kmsg dump iterator
3520 *
3521 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3522 * kmsg_dump_get_buffer() can be called again and used multiple
3523 * times within the same dumper.dump() callback.
3524 */
3525void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
3526{
3527	unsigned long flags;
3528
3529	printk_safe_enter_irqsave(flags);
3530	iter->cur_seq = latched_seq_read_nolock(&clear_seq);
3531	iter->next_seq = prb_next_seq(prb);
3532	printk_safe_exit_irqrestore(flags);
3533}
3534EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3535
3536#endif
3537
3538#ifdef CONFIG_SMP
3539static atomic_t printk_cpulock_owner = ATOMIC_INIT(-1);
3540static atomic_t printk_cpulock_nested = ATOMIC_INIT(0);
3541
3542/**
3543 * __printk_wait_on_cpu_lock() - Busy wait until the printk cpu-reentrant
3544 *                               spinning lock is not owned by any CPU.
3545 *
3546 * Context: Any context.
3547 */
3548void __printk_wait_on_cpu_lock(void)
3549{
3550	do {
3551		cpu_relax();
3552	} while (atomic_read(&printk_cpulock_owner) != -1);
3553}
3554EXPORT_SYMBOL(__printk_wait_on_cpu_lock);
3555
3556/**
3557 * __printk_cpu_trylock() - Try to acquire the printk cpu-reentrant
3558 *                          spinning lock.
3559 *
3560 * If no processor has the lock, the calling processor takes the lock and
3561 * becomes the owner. If the calling processor is already the owner of the
3562 * lock, this function succeeds immediately.
3563 *
3564 * Context: Any context. Expects interrupts to be disabled.
3565 * Return: 1 on success, otherwise 0.
3566 */
3567int __printk_cpu_trylock(void)
3568{
3569	int cpu;
3570	int old;
3571
3572	cpu = smp_processor_id();
3573
3574	/*
3575	 * Guarantee loads and stores from this CPU when it is the lock owner
3576	 * are _not_ visible to the previous lock owner. This pairs with
3577	 * __printk_cpu_unlock:B.
3578	 *
3579	 * Memory barrier involvement:
3580	 *
3581	 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B, then
3582	 * __printk_cpu_unlock:A can never read from __printk_cpu_trylock:B.
 
3583	 *
3584	 * Relies on:
3585	 *
3586	 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3587	 * of the previous CPU
3588	 *    matching
3589	 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3590	 * of this CPU
3591	 */
3592	old = atomic_cmpxchg_acquire(&printk_cpulock_owner, -1,
3593				     cpu); /* LMM(__printk_cpu_trylock:A) */
3594	if (old == -1) {
3595		/*
3596		 * This CPU is now the owner and begins loading/storing
3597		 * data: LMM(__printk_cpu_trylock:B)
3598		 */
3599		return 1;
3600
3601	} else if (old == cpu) {
3602		/* This CPU is already the owner. */
3603		atomic_inc(&printk_cpulock_nested);
3604		return 1;
3605	}
3606
3607	return 0;
3608}
3609EXPORT_SYMBOL(__printk_cpu_trylock);
3610
3611/**
3612 * __printk_cpu_unlock() - Release the printk cpu-reentrant spinning lock.
3613 *
3614 * The calling processor must be the owner of the lock.
3615 *
3616 * Context: Any context. Expects interrupts to be disabled.
3617 */
3618void __printk_cpu_unlock(void)
3619{
3620	if (atomic_read(&printk_cpulock_nested)) {
3621		atomic_dec(&printk_cpulock_nested);
3622		return;
3623	}
3624
3625	/*
3626	 * This CPU is finished loading/storing data:
3627	 * LMM(__printk_cpu_unlock:A)
3628	 */
3629
3630	/*
3631	 * Guarantee loads and stores from this CPU when it was the
3632	 * lock owner are visible to the next lock owner. This pairs
3633	 * with __printk_cpu_trylock:A.
3634	 *
3635	 * Memory barrier involvement:
3636	 *
3637	 * If __printk_cpu_trylock:A reads from __printk_cpu_unlock:B,
3638	 * then __printk_cpu_trylock:B reads from __printk_cpu_unlock:A.
3639	 *
3640	 * Relies on:
3641	 *
3642	 * RELEASE from __printk_cpu_unlock:A to __printk_cpu_unlock:B
3643	 * of this CPU
3644	 *    matching
3645	 * ACQUIRE from __printk_cpu_trylock:A to __printk_cpu_trylock:B
3646	 * of the next CPU
3647	 */
3648	atomic_set_release(&printk_cpulock_owner,
3649			   -1); /* LMM(__printk_cpu_unlock:B) */
3650}
3651EXPORT_SYMBOL(__printk_cpu_unlock);
3652#endif /* CONFIG_SMP */