Linux Audio

Check our new training course

Loading...
v4.10.11
   1/*
   2 *  linux/kernel/printk.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 * Modified to make sys_syslog() more flexible: added commands to
   7 * return the last 4k of kernel messages, regardless of whether
   8 * they've been read or not.  Added option to suppress kernel printk's
   9 * to the console.  Added hook for sending the console messages
  10 * elsewhere, in preparation for a serial line console (someday).
  11 * Ted Ts'o, 2/11/93.
  12 * Modified for sysctl support, 1/8/97, Chris Horn.
  13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  14 *     manfred@colorfullife.com
  15 * Rewrote bits to get rid of console_lock
  16 *	01Mar01 Andrew Morton
  17 */
  18
  19#include <linux/kernel.h>
  20#include <linux/mm.h>
  21#include <linux/tty.h>
  22#include <linux/tty_driver.h>
  23#include <linux/console.h>
  24#include <linux/init.h>
  25#include <linux/jiffies.h>
  26#include <linux/nmi.h>
  27#include <linux/module.h>
  28#include <linux/moduleparam.h>
  29#include <linux/delay.h>
  30#include <linux/smp.h>
  31#include <linux/security.h>
  32#include <linux/bootmem.h>
  33#include <linux/memblock.h>
  34#include <linux/syscalls.h>
  35#include <linux/kexec.h>
  36#include <linux/kdb.h>
  37#include <linux/ratelimit.h>
  38#include <linux/kmsg_dump.h>
  39#include <linux/syslog.h>
  40#include <linux/cpu.h>
  41#include <linux/notifier.h>
  42#include <linux/rculist.h>
  43#include <linux/poll.h>
  44#include <linux/irq_work.h>
  45#include <linux/utsname.h>
  46#include <linux/ctype.h>
  47#include <linux/uio.h>
 
 
 
  48
  49#include <linux/uaccess.h>
  50#include <asm/sections.h>
  51
 
  52#define CREATE_TRACE_POINTS
  53#include <trace/events/printk.h>
  54
  55#include "console_cmdline.h"
  56#include "braille.h"
  57#include "internal.h"
  58
  59int console_printk[4] = {
  60	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  61	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  62	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  63	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  64};
  65
  66/*
  67 * Low level drivers may need that to know if they can schedule in
  68 * their unblank() callback or not. So let's export it.
  69 */
  70int oops_in_progress;
  71EXPORT_SYMBOL(oops_in_progress);
  72
  73/*
  74 * console_sem protects the console_drivers list, and also
  75 * provides serialisation for access to the entire console
  76 * driver system.
  77 */
  78static DEFINE_SEMAPHORE(console_sem);
  79struct console *console_drivers;
  80EXPORT_SYMBOL_GPL(console_drivers);
  81
  82#ifdef CONFIG_LOCKDEP
  83static struct lockdep_map console_lock_dep_map = {
  84	.name = "console_lock"
  85};
  86#endif
  87
  88enum devkmsg_log_bits {
  89	__DEVKMSG_LOG_BIT_ON = 0,
  90	__DEVKMSG_LOG_BIT_OFF,
  91	__DEVKMSG_LOG_BIT_LOCK,
  92};
  93
  94enum devkmsg_log_masks {
  95	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
  96	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
  97	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
  98};
  99
 100/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 101#define DEVKMSG_LOG_MASK_DEFAULT	0
 102
 103static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 104
 105static int __control_devkmsg(char *str)
 106{
 107	if (!str)
 108		return -EINVAL;
 109
 110	if (!strncmp(str, "on", 2)) {
 111		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 112		return 2;
 113	} else if (!strncmp(str, "off", 3)) {
 114		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 115		return 3;
 116	} else if (!strncmp(str, "ratelimit", 9)) {
 117		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 118		return 9;
 119	}
 120	return -EINVAL;
 121}
 122
 123static int __init control_devkmsg(char *str)
 124{
 125	if (__control_devkmsg(str) < 0)
 126		return 1;
 127
 128	/*
 129	 * Set sysctl string accordingly:
 130	 */
 131	if (devkmsg_log == DEVKMSG_LOG_MASK_ON) {
 132		memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
 133		strncpy(devkmsg_log_str, "on", 2);
 134	} else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF) {
 135		memset(devkmsg_log_str, 0, DEVKMSG_STR_MAX_SIZE);
 136		strncpy(devkmsg_log_str, "off", 3);
 137	}
 138	/* else "ratelimit" which is set by default. */
 139
 140	/*
 141	 * Sysctl cannot change it anymore. The kernel command line setting of
 142	 * this parameter is to force the setting to be permanent throughout the
 143	 * runtime of the system. This is a precation measure against userspace
 144	 * trying to be a smarta** and attempting to change it up on us.
 145	 */
 146	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 147
 148	return 0;
 149}
 150__setup("printk.devkmsg=", control_devkmsg);
 151
 152char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 153
 154int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 155			      void __user *buffer, size_t *lenp, loff_t *ppos)
 156{
 157	char old_str[DEVKMSG_STR_MAX_SIZE];
 158	unsigned int old;
 159	int err;
 160
 161	if (write) {
 162		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 163			return -EINVAL;
 164
 165		old = devkmsg_log;
 166		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 167	}
 168
 169	err = proc_dostring(table, write, buffer, lenp, ppos);
 170	if (err)
 171		return err;
 172
 173	if (write) {
 174		err = __control_devkmsg(devkmsg_log_str);
 175
 176		/*
 177		 * Do not accept an unknown string OR a known string with
 178		 * trailing crap...
 179		 */
 180		if (err < 0 || (err + 1 != *lenp)) {
 181
 182			/* ... and restore old setting. */
 183			devkmsg_log = old;
 184			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 185
 186			return -EINVAL;
 187		}
 188	}
 189
 190	return 0;
 191}
 192
 193/*
 194 * Number of registered extended console drivers.
 195 *
 196 * If extended consoles are present, in-kernel cont reassembly is disabled
 197 * and each fragment is stored as a separate log entry with proper
 198 * continuation flag so that every emitted message has full metadata.  This
 199 * doesn't change the result for regular consoles or /proc/kmsg.  For
 200 * /dev/kmsg, as long as the reader concatenates messages according to
 201 * consecutive continuation flags, the end result should be the same too.
 202 */
 203static int nr_ext_console_drivers;
 204
 205/*
 206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 207 * macros instead of functions so that _RET_IP_ contains useful information.
 208 */
 209#define down_console_sem() do { \
 210	down(&console_sem);\
 211	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 212} while (0)
 213
 214static int __down_trylock_console_sem(unsigned long ip)
 215{
 216	if (down_trylock(&console_sem))
 
 
 
 
 
 
 
 
 
 
 
 
 217		return 1;
 218	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 219	return 0;
 220}
 221#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 222
 223#define up_console_sem() do { \
 224	mutex_release(&console_lock_dep_map, 1, _RET_IP_);\
 225	up(&console_sem);\
 226} while (0)
 
 
 
 
 
 
 
 227
 228/*
 229 * This is used for debugging the mess that is the VT code by
 230 * keeping track if we have the console semaphore held. It's
 231 * definitely not the perfect debug tool (we don't know if _WE_
 232 * hold it and are racing, but it helps tracking those weird code
 233 * paths in the console code where we end up in places I want
 234 * locked without the console sempahore held).
 235 */
 236static int console_locked, console_suspended;
 237
 238/*
 239 * If exclusive_console is non-NULL then only this console is to be printed to.
 240 */
 241static struct console *exclusive_console;
 242
 243/*
 244 *	Array of consoles built from command line options (console=)
 245 */
 246
 247#define MAX_CMDLINECONSOLES 8
 248
 249static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 250
 251static int selected_console = -1;
 252static int preferred_console = -1;
 253int console_set_on_cmdline;
 254EXPORT_SYMBOL(console_set_on_cmdline);
 255
 256/* Flag: console code may call schedule() */
 257static int console_may_schedule;
 258
 
 
 
 
 
 
 
 259/*
 260 * The printk log buffer consists of a chain of concatenated variable
 261 * length records. Every record starts with a record header, containing
 262 * the overall length of the record.
 263 *
 264 * The heads to the first and last entry in the buffer, as well as the
 265 * sequence numbers of these entries are maintained when messages are
 266 * stored.
 267 *
 268 * If the heads indicate available messages, the length in the header
 269 * tells the start next message. A length == 0 for the next message
 270 * indicates a wrap-around to the beginning of the buffer.
 271 *
 272 * Every record carries the monotonic timestamp in microseconds, as well as
 273 * the standard userspace syslog level and syslog facility. The usual
 274 * kernel messages use LOG_KERN; userspace-injected messages always carry
 275 * a matching syslog facility, by default LOG_USER. The origin of every
 276 * message can be reliably determined that way.
 277 *
 278 * The human readable log message directly follows the message header. The
 279 * length of the message text is stored in the header, the stored message
 280 * is not terminated.
 281 *
 282 * Optionally, a message can carry a dictionary of properties (key/value pairs),
 283 * to provide userspace with a machine-readable message context.
 284 *
 285 * Examples for well-defined, commonly used property names are:
 286 *   DEVICE=b12:8               device identifier
 287 *                                b12:8         block dev_t
 288 *                                c127:3        char dev_t
 289 *                                n8            netdev ifindex
 290 *                                +sound:card0  subsystem:devname
 291 *   SUBSYSTEM=pci              driver-core subsystem name
 292 *
 293 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
 294 * follows directly after a '=' character. Every property is terminated by
 295 * a '\0' character. The last property is not terminated.
 296 *
 297 * Example of a message structure:
 298 *   0000  ff 8f 00 00 00 00 00 00      monotonic time in nsec
 299 *   0008  34 00                        record is 52 bytes long
 300 *   000a        0b 00                  text is 11 bytes long
 301 *   000c              1f 00            dictionary is 23 bytes long
 302 *   000e                    03 00      LOG_KERN (facility) LOG_ERR (level)
 303 *   0010  69 74 27 73 20 61 20 6c      "it's a l"
 304 *         69 6e 65                     "ine"
 305 *   001b           44 45 56 49 43      "DEVIC"
 306 *         45 3d 62 38 3a 32 00 44      "E=b8:2\0D"
 307 *         52 49 56 45 52 3d 62 75      "RIVER=bu"
 308 *         67                           "g"
 309 *   0032     00 00 00                  padding to next message header
 310 *
 311 * The 'struct printk_log' buffer header must never be directly exported to
 312 * userspace, it is a kernel-private implementation detail that might
 313 * need to be changed in the future, when the requirements change.
 314 *
 315 * /dev/kmsg exports the structured data in the following line format:
 316 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 317 *
 318 * Users of the export format should ignore possible additional values
 319 * separated by ',', and find the message after the ';' character.
 320 *
 321 * The optional key/value pairs are attached as continuation lines starting
 322 * with a space character and terminated by a newline. All possible
 323 * non-prinatable characters are escaped in the "\xff" notation.
 324 */
 325
 326enum log_flags {
 327	LOG_NOCONS	= 1,	/* already flushed, do not print to console */
 328	LOG_NEWLINE	= 2,	/* text ended with a newline */
 329	LOG_PREFIX	= 4,	/* text started with a prefix */
 330	LOG_CONT	= 8,	/* text is a fragment of a continuation line */
 331};
 332
 333struct printk_log {
 334	u64 ts_nsec;		/* timestamp in nanoseconds */
 335	u16 len;		/* length of entire record */
 336	u16 text_len;		/* length of text buffer */
 337	u16 dict_len;		/* length of dictionary buffer */
 338	u8 facility;		/* syslog facility */
 339	u8 flags:5;		/* internal record flags */
 340	u8 level:3;		/* syslog level */
 341}
 342#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
 343__packed __aligned(4)
 344#endif
 345;
 346
 347/*
 348 * The logbuf_lock protects kmsg buffer, indices, counters.  This can be taken
 349 * within the scheduler's rq lock. It must be released before calling
 350 * console_unlock() or anything else that might wake up a process.
 351 */
 352DEFINE_RAW_SPINLOCK(logbuf_lock);
 353
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 354#ifdef CONFIG_PRINTK
 355DECLARE_WAIT_QUEUE_HEAD(log_wait);
 356/* the next printk record to read by syslog(READ) or /proc/kmsg */
 357static u64 syslog_seq;
 358static u32 syslog_idx;
 359static size_t syslog_partial;
 360
 361/* index and sequence number of the first record stored in the buffer */
 362static u64 log_first_seq;
 363static u32 log_first_idx;
 364
 365/* index and sequence number of the next record to store in the buffer */
 366static u64 log_next_seq;
 367static u32 log_next_idx;
 368
 369/* the next printk record to write to the console */
 370static u64 console_seq;
 371static u32 console_idx;
 372
 373/* the next printk record to read after the last 'clear' command */
 374static u64 clear_seq;
 375static u32 clear_idx;
 376
 377#define PREFIX_MAX		32
 378#define LOG_LINE_MAX		(1024 - PREFIX_MAX)
 379
 380#define LOG_LEVEL(v)		((v) & 0x07)
 381#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 382
 383/* record buffer */
 384#define LOG_ALIGN __alignof__(struct printk_log)
 385#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 386static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 387static char *log_buf = __log_buf;
 388static u32 log_buf_len = __LOG_BUF_LEN;
 389
 390/* Return log buffer address */
 391char *log_buf_addr_get(void)
 392{
 393	return log_buf;
 394}
 395
 396/* Return log buffer size */
 397u32 log_buf_len_get(void)
 398{
 399	return log_buf_len;
 400}
 401
 402/* human readable text of the record */
 403static char *log_text(const struct printk_log *msg)
 404{
 405	return (char *)msg + sizeof(struct printk_log);
 406}
 407
 408/* optional key/value pair dictionary attached to the record */
 409static char *log_dict(const struct printk_log *msg)
 410{
 411	return (char *)msg + sizeof(struct printk_log) + msg->text_len;
 412}
 413
 414/* get record by index; idx must point to valid msg */
 415static struct printk_log *log_from_idx(u32 idx)
 416{
 417	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
 418
 419	/*
 420	 * A length == 0 record is the end of buffer marker. Wrap around and
 421	 * read the message at the start of the buffer.
 422	 */
 423	if (!msg->len)
 424		return (struct printk_log *)log_buf;
 425	return msg;
 426}
 427
 428/* get next record; idx must point to valid msg */
 429static u32 log_next(u32 idx)
 430{
 431	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
 432
 433	/* length == 0 indicates the end of the buffer; wrap */
 434	/*
 435	 * A length == 0 record is the end of buffer marker. Wrap around and
 436	 * read the message at the start of the buffer as *this* one, and
 437	 * return the one after that.
 438	 */
 439	if (!msg->len) {
 440		msg = (struct printk_log *)log_buf;
 441		return msg->len;
 442	}
 443	return idx + msg->len;
 444}
 445
 446/*
 447 * Check whether there is enough free space for the given message.
 448 *
 449 * The same values of first_idx and next_idx mean that the buffer
 450 * is either empty or full.
 451 *
 452 * If the buffer is empty, we must respect the position of the indexes.
 453 * They cannot be reset to the beginning of the buffer.
 454 */
 455static int logbuf_has_space(u32 msg_size, bool empty)
 456{
 457	u32 free;
 458
 459	if (log_next_idx > log_first_idx || empty)
 460		free = max(log_buf_len - log_next_idx, log_first_idx);
 461	else
 462		free = log_first_idx - log_next_idx;
 463
 464	/*
 465	 * We need space also for an empty header that signalizes wrapping
 466	 * of the buffer.
 467	 */
 468	return free >= msg_size + sizeof(struct printk_log);
 469}
 470
 471static int log_make_free_space(u32 msg_size)
 472{
 473	while (log_first_seq < log_next_seq &&
 474	       !logbuf_has_space(msg_size, false)) {
 475		/* drop old messages until we have enough contiguous space */
 476		log_first_idx = log_next(log_first_idx);
 477		log_first_seq++;
 478	}
 479
 480	if (clear_seq < log_first_seq) {
 481		clear_seq = log_first_seq;
 482		clear_idx = log_first_idx;
 483	}
 484
 485	/* sequence numbers are equal, so the log buffer is empty */
 486	if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
 487		return 0;
 488
 489	return -ENOMEM;
 490}
 491
 492/* compute the message size including the padding bytes */
 493static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
 494{
 495	u32 size;
 496
 497	size = sizeof(struct printk_log) + text_len + dict_len;
 498	*pad_len = (-size) & (LOG_ALIGN - 1);
 499	size += *pad_len;
 500
 501	return size;
 502}
 503
 504/*
 505 * Define how much of the log buffer we could take at maximum. The value
 506 * must be greater than two. Note that only half of the buffer is available
 507 * when the index points to the middle.
 508 */
 509#define MAX_LOG_TAKE_PART 4
 510static const char trunc_msg[] = "<truncated>";
 511
 512static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
 513			u16 *dict_len, u32 *pad_len)
 514{
 515	/*
 516	 * The message should not take the whole buffer. Otherwise, it might
 517	 * get removed too soon.
 518	 */
 519	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 520	if (*text_len > max_text_len)
 521		*text_len = max_text_len;
 522	/* enable the warning message */
 523	*trunc_msg_len = strlen(trunc_msg);
 524	/* disable the "dict" completely */
 525	*dict_len = 0;
 526	/* compute the size again, count also the warning message */
 527	return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
 528}
 529
 530/* insert record into the buffer, discard old ones, update heads */
 531static int log_store(int facility, int level,
 532		     enum log_flags flags, u64 ts_nsec,
 533		     const char *dict, u16 dict_len,
 534		     const char *text, u16 text_len)
 535{
 536	struct printk_log *msg;
 537	u32 size, pad_len;
 538	u16 trunc_msg_len = 0;
 539
 540	/* number of '\0' padding bytes to next message */
 541	size = msg_used_size(text_len, dict_len, &pad_len);
 542
 543	if (log_make_free_space(size)) {
 544		/* truncate the message if it is too long for empty buffer */
 545		size = truncate_msg(&text_len, &trunc_msg_len,
 546				    &dict_len, &pad_len);
 547		/* survive when the log buffer is too small for trunc_msg */
 548		if (log_make_free_space(size))
 549			return 0;
 550	}
 551
 552	if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
 553		/*
 554		 * This message + an additional empty header does not fit
 555		 * at the end of the buffer. Add an empty header with len == 0
 556		 * to signify a wrap around.
 557		 */
 558		memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
 559		log_next_idx = 0;
 560	}
 561
 562	/* fill message */
 563	msg = (struct printk_log *)(log_buf + log_next_idx);
 564	memcpy(log_text(msg), text, text_len);
 565	msg->text_len = text_len;
 566	if (trunc_msg_len) {
 567		memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
 568		msg->text_len += trunc_msg_len;
 569	}
 570	memcpy(log_dict(msg), dict, dict_len);
 571	msg->dict_len = dict_len;
 572	msg->facility = facility;
 573	msg->level = level & 7;
 574	msg->flags = flags & 0x1f;
 575	if (ts_nsec > 0)
 576		msg->ts_nsec = ts_nsec;
 577	else
 578		msg->ts_nsec = local_clock();
 579	memset(log_dict(msg) + dict_len, 0, pad_len);
 580	msg->len = size;
 581
 582	/* insert message */
 583	log_next_idx += msg->len;
 584	log_next_seq++;
 585
 586	return msg->text_len;
 587}
 588
 589int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 590
 591static int syslog_action_restricted(int type)
 592{
 593	if (dmesg_restrict)
 594		return 1;
 595	/*
 596	 * Unless restricted, we allow "read all" and "get buffer size"
 597	 * for everybody.
 598	 */
 599	return type != SYSLOG_ACTION_READ_ALL &&
 600	       type != SYSLOG_ACTION_SIZE_BUFFER;
 601}
 602
 603int check_syslog_permissions(int type, int source)
 604{
 605	/*
 606	 * If this is from /proc/kmsg and we've already opened it, then we've
 607	 * already done the capabilities checks at open time.
 608	 */
 609	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 610		goto ok;
 611
 612	if (syslog_action_restricted(type)) {
 613		if (capable(CAP_SYSLOG))
 614			goto ok;
 615		/*
 616		 * For historical reasons, accept CAP_SYS_ADMIN too, with
 617		 * a warning.
 618		 */
 619		if (capable(CAP_SYS_ADMIN)) {
 620			pr_warn_once("%s (%d): Attempt to access syslog with "
 621				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
 622				     "(deprecated).\n",
 623				 current->comm, task_pid_nr(current));
 624			goto ok;
 625		}
 626		return -EPERM;
 627	}
 628ok:
 629	return security_syslog(type);
 630}
 631EXPORT_SYMBOL_GPL(check_syslog_permissions);
 632
 633static void append_char(char **pp, char *e, char c)
 634{
 635	if (*pp < e)
 636		*(*pp)++ = c;
 637}
 638
 639static ssize_t msg_print_ext_header(char *buf, size_t size,
 640				    struct printk_log *msg, u64 seq)
 641{
 642	u64 ts_usec = msg->ts_nsec;
 643
 644	do_div(ts_usec, 1000);
 645
 646	return scnprintf(buf, size, "%u,%llu,%llu,%c;",
 647		       (msg->facility << 3) | msg->level, seq, ts_usec,
 648		       msg->flags & LOG_CONT ? 'c' : '-');
 649}
 650
 651static ssize_t msg_print_ext_body(char *buf, size_t size,
 652				  char *dict, size_t dict_len,
 653				  char *text, size_t text_len)
 654{
 655	char *p = buf, *e = buf + size;
 656	size_t i;
 657
 658	/* escape non-printable characters */
 659	for (i = 0; i < text_len; i++) {
 660		unsigned char c = text[i];
 661
 662		if (c < ' ' || c >= 127 || c == '\\')
 663			p += scnprintf(p, e - p, "\\x%02x", c);
 664		else
 665			append_char(&p, e, c);
 666	}
 667	append_char(&p, e, '\n');
 668
 669	if (dict_len) {
 670		bool line = true;
 671
 672		for (i = 0; i < dict_len; i++) {
 673			unsigned char c = dict[i];
 674
 675			if (line) {
 676				append_char(&p, e, ' ');
 677				line = false;
 678			}
 679
 680			if (c == '\0') {
 681				append_char(&p, e, '\n');
 682				line = true;
 683				continue;
 684			}
 685
 686			if (c < ' ' || c >= 127 || c == '\\') {
 687				p += scnprintf(p, e - p, "\\x%02x", c);
 688				continue;
 689			}
 690
 691			append_char(&p, e, c);
 692		}
 693		append_char(&p, e, '\n');
 694	}
 695
 696	return p - buf;
 697}
 698
 699/* /dev/kmsg - userspace message inject/listen interface */
 700struct devkmsg_user {
 701	u64 seq;
 702	u32 idx;
 703	struct ratelimit_state rs;
 704	struct mutex lock;
 705	char buf[CONSOLE_EXT_LOG_MAX];
 706};
 707
 708static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 709{
 710	char *buf, *line;
 711	int level = default_message_loglevel;
 712	int facility = 1;	/* LOG_USER */
 713	struct file *file = iocb->ki_filp;
 714	struct devkmsg_user *user = file->private_data;
 715	size_t len = iov_iter_count(from);
 716	ssize_t ret = len;
 717
 718	if (!user || len > LOG_LINE_MAX)
 719		return -EINVAL;
 720
 721	/* Ignore when user logging is disabled. */
 722	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 723		return len;
 724
 725	/* Ratelimit when not explicitly enabled. */
 726	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 727		if (!___ratelimit(&user->rs, current->comm))
 728			return ret;
 729	}
 730
 731	buf = kmalloc(len+1, GFP_KERNEL);
 732	if (buf == NULL)
 733		return -ENOMEM;
 734
 735	buf[len] = '\0';
 736	if (!copy_from_iter_full(buf, len, from)) {
 737		kfree(buf);
 738		return -EFAULT;
 739	}
 740
 741	/*
 742	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 743	 * the decimal value represents 32bit, the lower 3 bit are the log
 744	 * level, the rest are the log facility.
 745	 *
 746	 * If no prefix or no userspace facility is specified, we
 747	 * enforce LOG_USER, to be able to reliably distinguish
 748	 * kernel-generated messages from userspace-injected ones.
 749	 */
 750	line = buf;
 751	if (line[0] == '<') {
 752		char *endp = NULL;
 753		unsigned int u;
 754
 755		u = simple_strtoul(line + 1, &endp, 10);
 756		if (endp && endp[0] == '>') {
 757			level = LOG_LEVEL(u);
 758			if (LOG_FACILITY(u) != 0)
 759				facility = LOG_FACILITY(u);
 760			endp++;
 761			len -= endp - line;
 762			line = endp;
 763		}
 764	}
 765
 766	printk_emit(facility, level, NULL, 0, "%s", line);
 767	kfree(buf);
 768	return ret;
 769}
 770
 771static ssize_t devkmsg_read(struct file *file, char __user *buf,
 772			    size_t count, loff_t *ppos)
 773{
 774	struct devkmsg_user *user = file->private_data;
 775	struct printk_log *msg;
 776	size_t len;
 777	ssize_t ret;
 778
 779	if (!user)
 780		return -EBADF;
 781
 782	ret = mutex_lock_interruptible(&user->lock);
 783	if (ret)
 784		return ret;
 785	raw_spin_lock_irq(&logbuf_lock);
 
 786	while (user->seq == log_next_seq) {
 787		if (file->f_flags & O_NONBLOCK) {
 788			ret = -EAGAIN;
 789			raw_spin_unlock_irq(&logbuf_lock);
 790			goto out;
 791		}
 792
 793		raw_spin_unlock_irq(&logbuf_lock);
 794		ret = wait_event_interruptible(log_wait,
 795					       user->seq != log_next_seq);
 796		if (ret)
 797			goto out;
 798		raw_spin_lock_irq(&logbuf_lock);
 799	}
 800
 801	if (user->seq < log_first_seq) {
 802		/* our last seen message is gone, return error and reset */
 803		user->idx = log_first_idx;
 804		user->seq = log_first_seq;
 805		ret = -EPIPE;
 806		raw_spin_unlock_irq(&logbuf_lock);
 807		goto out;
 808	}
 809
 810	msg = log_from_idx(user->idx);
 811	len = msg_print_ext_header(user->buf, sizeof(user->buf),
 812				   msg, user->seq);
 813	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
 814				  log_dict(msg), msg->dict_len,
 815				  log_text(msg), msg->text_len);
 816
 817	user->idx = log_next(user->idx);
 818	user->seq++;
 819	raw_spin_unlock_irq(&logbuf_lock);
 820
 821	if (len > count) {
 822		ret = -EINVAL;
 823		goto out;
 824	}
 825
 826	if (copy_to_user(buf, user->buf, len)) {
 827		ret = -EFAULT;
 828		goto out;
 829	}
 830	ret = len;
 831out:
 832	mutex_unlock(&user->lock);
 833	return ret;
 834}
 835
 836static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 837{
 838	struct devkmsg_user *user = file->private_data;
 839	loff_t ret = 0;
 840
 841	if (!user)
 842		return -EBADF;
 843	if (offset)
 844		return -ESPIPE;
 845
 846	raw_spin_lock_irq(&logbuf_lock);
 847	switch (whence) {
 848	case SEEK_SET:
 849		/* the first record */
 850		user->idx = log_first_idx;
 851		user->seq = log_first_seq;
 852		break;
 853	case SEEK_DATA:
 854		/*
 855		 * The first record after the last SYSLOG_ACTION_CLEAR,
 856		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 857		 * changes no global state, and does not clear anything.
 858		 */
 859		user->idx = clear_idx;
 860		user->seq = clear_seq;
 861		break;
 862	case SEEK_END:
 863		/* after the last record */
 864		user->idx = log_next_idx;
 865		user->seq = log_next_seq;
 866		break;
 867	default:
 868		ret = -EINVAL;
 869	}
 870	raw_spin_unlock_irq(&logbuf_lock);
 871	return ret;
 872}
 873
 874static unsigned int devkmsg_poll(struct file *file, poll_table *wait)
 875{
 876	struct devkmsg_user *user = file->private_data;
 877	int ret = 0;
 878
 879	if (!user)
 880		return POLLERR|POLLNVAL;
 881
 882	poll_wait(file, &log_wait, wait);
 883
 884	raw_spin_lock_irq(&logbuf_lock);
 885	if (user->seq < log_next_seq) {
 886		/* return error when data has vanished underneath us */
 887		if (user->seq < log_first_seq)
 888			ret = POLLIN|POLLRDNORM|POLLERR|POLLPRI;
 889		else
 890			ret = POLLIN|POLLRDNORM;
 891	}
 892	raw_spin_unlock_irq(&logbuf_lock);
 893
 894	return ret;
 895}
 896
 897static int devkmsg_open(struct inode *inode, struct file *file)
 898{
 899	struct devkmsg_user *user;
 900	int err;
 901
 902	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 903		return -EPERM;
 904
 905	/* write-only does not need any file context */
 906	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 907		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 908					       SYSLOG_FROM_READER);
 909		if (err)
 910			return err;
 911	}
 912
 913	user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 914	if (!user)
 915		return -ENOMEM;
 916
 917	ratelimit_default_init(&user->rs);
 918	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 919
 920	mutex_init(&user->lock);
 921
 922	raw_spin_lock_irq(&logbuf_lock);
 923	user->idx = log_first_idx;
 924	user->seq = log_first_seq;
 925	raw_spin_unlock_irq(&logbuf_lock);
 926
 927	file->private_data = user;
 928	return 0;
 929}
 930
 931static int devkmsg_release(struct inode *inode, struct file *file)
 932{
 933	struct devkmsg_user *user = file->private_data;
 934
 935	if (!user)
 936		return 0;
 937
 938	ratelimit_state_exit(&user->rs);
 939
 940	mutex_destroy(&user->lock);
 941	kfree(user);
 942	return 0;
 943}
 944
 945const struct file_operations kmsg_fops = {
 946	.open = devkmsg_open,
 947	.read = devkmsg_read,
 948	.write_iter = devkmsg_write,
 949	.llseek = devkmsg_llseek,
 950	.poll = devkmsg_poll,
 951	.release = devkmsg_release,
 952};
 953
 954#ifdef CONFIG_KEXEC_CORE
 955/*
 956 * This appends the listed symbols to /proc/vmcore
 957 *
 958 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
 959 * obtain access to symbols that are otherwise very difficult to locate.  These
 960 * symbols are specifically used so that utilities can access and extract the
 961 * dmesg log from a vmcore file after a crash.
 962 */
 963void log_buf_kexec_setup(void)
 964{
 965	VMCOREINFO_SYMBOL(log_buf);
 966	VMCOREINFO_SYMBOL(log_buf_len);
 967	VMCOREINFO_SYMBOL(log_first_idx);
 968	VMCOREINFO_SYMBOL(clear_idx);
 969	VMCOREINFO_SYMBOL(log_next_idx);
 970	/*
 971	 * Export struct printk_log size and field offsets. User space tools can
 972	 * parse it and detect any changes to structure down the line.
 973	 */
 974	VMCOREINFO_STRUCT_SIZE(printk_log);
 975	VMCOREINFO_OFFSET(printk_log, ts_nsec);
 976	VMCOREINFO_OFFSET(printk_log, len);
 977	VMCOREINFO_OFFSET(printk_log, text_len);
 978	VMCOREINFO_OFFSET(printk_log, dict_len);
 979}
 980#endif
 981
 982/* requested log_buf_len from kernel cmdline */
 983static unsigned long __initdata new_log_buf_len;
 984
 985/* we practice scaling the ring buffer by powers of 2 */
 986static void __init log_buf_len_update(unsigned size)
 987{
 988	if (size)
 989		size = roundup_pow_of_two(size);
 990	if (size > log_buf_len)
 991		new_log_buf_len = 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	unsigned size = memparse(str, &str);
 998
 999	log_buf_len_update(size);
1000
1001	return 0;
1002}
1003early_param("log_buf_len", log_buf_len_setup);
1004
1005#ifdef CONFIG_SMP
1006#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1007
1008static void __init log_buf_add_cpu(void)
1009{
1010	unsigned int cpu_extra;
1011
1012	/*
1013	 * archs should set up cpu_possible_bits properly with
1014	 * set_cpu_possible() after setup_arch() but just in
1015	 * case lets ensure this is valid.
1016	 */
1017	if (num_possible_cpus() == 1)
1018		return;
1019
1020	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1021
1022	/* by default this will only continue through for large > 64 CPUs */
1023	if (cpu_extra <= __LOG_BUF_LEN / 2)
1024		return;
1025
1026	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1027		__LOG_CPU_MAX_BUF_LEN);
1028	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1029		cpu_extra);
1030	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1031
1032	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1033}
1034#else /* !CONFIG_SMP */
1035static inline void log_buf_add_cpu(void) {}
1036#endif /* CONFIG_SMP */
1037
1038void __init setup_log_buf(int early)
1039{
1040	unsigned long flags;
1041	char *new_log_buf;
1042	int free;
1043
1044	if (log_buf != __log_buf)
1045		return;
1046
1047	if (!early && !new_log_buf_len)
1048		log_buf_add_cpu();
1049
1050	if (!new_log_buf_len)
1051		return;
1052
1053	if (early) {
1054		new_log_buf =
1055			memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1056	} else {
1057		new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1058							  LOG_ALIGN);
1059	}
1060
1061	if (unlikely(!new_log_buf)) {
1062		pr_err("log_buf_len: %ld bytes not available\n",
1063			new_log_buf_len);
1064		return;
1065	}
1066
1067	raw_spin_lock_irqsave(&logbuf_lock, flags);
1068	log_buf_len = new_log_buf_len;
1069	log_buf = new_log_buf;
1070	new_log_buf_len = 0;
1071	free = __LOG_BUF_LEN - log_next_idx;
1072	memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1073	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
1074
1075	pr_info("log_buf_len: %d bytes\n", log_buf_len);
1076	pr_info("early log buf free: %d(%d%%)\n",
1077		free, (free * 100) / __LOG_BUF_LEN);
1078}
1079
1080static bool __read_mostly ignore_loglevel;
1081
1082static int __init ignore_loglevel_setup(char *str)
1083{
1084	ignore_loglevel = true;
1085	pr_info("debug: ignoring loglevel setting.\n");
1086
1087	return 0;
1088}
1089
1090early_param("ignore_loglevel", ignore_loglevel_setup);
1091module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1092MODULE_PARM_DESC(ignore_loglevel,
1093		 "ignore loglevel setting (prints all kernel messages to the console)");
1094
1095static bool suppress_message_printing(int level)
1096{
1097	return (level >= console_loglevel && !ignore_loglevel);
1098}
1099
1100#ifdef CONFIG_BOOT_PRINTK_DELAY
1101
1102static int boot_delay; /* msecs delay after each printk during bootup */
1103static unsigned long long loops_per_msec;	/* based on boot_delay */
1104
1105static int __init boot_delay_setup(char *str)
1106{
1107	unsigned long lpj;
1108
1109	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1110	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1111
1112	get_option(&str, &boot_delay);
1113	if (boot_delay > 10 * 1000)
1114		boot_delay = 0;
1115
1116	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1117		"HZ: %d, loops_per_msec: %llu\n",
1118		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1119	return 0;
1120}
1121early_param("boot_delay", boot_delay_setup);
1122
1123static void boot_delay_msec(int level)
1124{
1125	unsigned long long k;
1126	unsigned long timeout;
1127
1128	if ((boot_delay == 0 || system_state != SYSTEM_BOOTING)
1129		|| suppress_message_printing(level)) {
1130		return;
1131	}
1132
1133	k = (unsigned long long)loops_per_msec * boot_delay;
1134
1135	timeout = jiffies + msecs_to_jiffies(boot_delay);
1136	while (k) {
1137		k--;
1138		cpu_relax();
1139		/*
1140		 * use (volatile) jiffies to prevent
1141		 * compiler reduction; loop termination via jiffies
1142		 * is secondary and may or may not happen.
1143		 */
1144		if (time_after(jiffies, timeout))
1145			break;
1146		touch_nmi_watchdog();
1147	}
1148}
1149#else
1150static inline void boot_delay_msec(int level)
1151{
1152}
1153#endif
1154
1155static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1156module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1157
1158static size_t print_time(u64 ts, char *buf)
1159{
1160	unsigned long rem_nsec;
1161
1162	if (!printk_time)
1163		return 0;
1164
1165	rem_nsec = do_div(ts, 1000000000);
1166
1167	if (!buf)
1168		return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1169
1170	return sprintf(buf, "[%5lu.%06lu] ",
1171		       (unsigned long)ts, rem_nsec / 1000);
1172}
1173
1174static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1175{
1176	size_t len = 0;
1177	unsigned int prefix = (msg->facility << 3) | msg->level;
1178
1179	if (syslog) {
1180		if (buf) {
1181			len += sprintf(buf, "<%u>", prefix);
1182		} else {
1183			len += 3;
1184			if (prefix > 999)
1185				len += 3;
1186			else if (prefix > 99)
1187				len += 2;
1188			else if (prefix > 9)
1189				len++;
1190		}
1191	}
1192
1193	len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1194	return len;
1195}
1196
1197static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1198{
1199	const char *text = log_text(msg);
1200	size_t text_size = msg->text_len;
1201	size_t len = 0;
1202
1203	do {
1204		const char *next = memchr(text, '\n', text_size);
1205		size_t text_len;
1206
1207		if (next) {
1208			text_len = next - text;
1209			next++;
1210			text_size -= next - text;
1211		} else {
1212			text_len = text_size;
1213		}
1214
1215		if (buf) {
1216			if (print_prefix(msg, syslog, NULL) +
1217			    text_len + 1 >= size - len)
1218				break;
1219
1220			len += print_prefix(msg, syslog, buf + len);
1221			memcpy(buf + len, text, text_len);
1222			len += text_len;
1223			buf[len++] = '\n';
1224		} else {
1225			/* SYSLOG_ACTION_* buffer size only calculation */
1226			len += print_prefix(msg, syslog, NULL);
1227			len += text_len;
1228			len++;
1229		}
1230
1231		text = next;
1232	} while (text);
1233
1234	return len;
1235}
1236
1237static int syslog_print(char __user *buf, int size)
1238{
1239	char *text;
1240	struct printk_log *msg;
1241	int len = 0;
1242
1243	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1244	if (!text)
1245		return -ENOMEM;
1246
1247	while (size > 0) {
1248		size_t n;
1249		size_t skip;
1250
1251		raw_spin_lock_irq(&logbuf_lock);
1252		if (syslog_seq < log_first_seq) {
1253			/* messages are gone, move to first one */
1254			syslog_seq = log_first_seq;
1255			syslog_idx = log_first_idx;
1256			syslog_partial = 0;
1257		}
1258		if (syslog_seq == log_next_seq) {
1259			raw_spin_unlock_irq(&logbuf_lock);
1260			break;
1261		}
1262
1263		skip = syslog_partial;
1264		msg = log_from_idx(syslog_idx);
1265		n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1266		if (n - syslog_partial <= size) {
1267			/* message fits into buffer, move forward */
1268			syslog_idx = log_next(syslog_idx);
1269			syslog_seq++;
1270			n -= syslog_partial;
1271			syslog_partial = 0;
1272		} else if (!len){
1273			/* partial read(), remember position */
1274			n = size;
1275			syslog_partial += n;
1276		} else
1277			n = 0;
1278		raw_spin_unlock_irq(&logbuf_lock);
1279
1280		if (!n)
1281			break;
1282
1283		if (copy_to_user(buf, text + skip, n)) {
1284			if (!len)
1285				len = -EFAULT;
1286			break;
1287		}
1288
1289		len += n;
1290		size -= n;
1291		buf += n;
1292	}
1293
1294	kfree(text);
1295	return len;
1296}
1297
1298static int syslog_print_all(char __user *buf, int size, bool clear)
1299{
1300	char *text;
1301	int len = 0;
1302
1303	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1304	if (!text)
1305		return -ENOMEM;
1306
1307	raw_spin_lock_irq(&logbuf_lock);
1308	if (buf) {
1309		u64 next_seq;
1310		u64 seq;
1311		u32 idx;
1312
1313		/*
1314		 * Find first record that fits, including all following records,
1315		 * into the user-provided buffer for this dump.
1316		 */
1317		seq = clear_seq;
1318		idx = clear_idx;
1319		while (seq < log_next_seq) {
1320			struct printk_log *msg = log_from_idx(idx);
1321
1322			len += msg_print_text(msg, true, NULL, 0);
1323			idx = log_next(idx);
1324			seq++;
1325		}
1326
1327		/* move first record forward until length fits into the buffer */
1328		seq = clear_seq;
1329		idx = clear_idx;
1330		while (len > size && seq < log_next_seq) {
1331			struct printk_log *msg = log_from_idx(idx);
1332
1333			len -= msg_print_text(msg, true, NULL, 0);
1334			idx = log_next(idx);
1335			seq++;
1336		}
1337
1338		/* last message fitting into this dump */
1339		next_seq = log_next_seq;
1340
1341		len = 0;
1342		while (len >= 0 && seq < next_seq) {
1343			struct printk_log *msg = log_from_idx(idx);
1344			int textlen;
1345
1346			textlen = msg_print_text(msg, true, text,
1347						 LOG_LINE_MAX + PREFIX_MAX);
1348			if (textlen < 0) {
1349				len = textlen;
1350				break;
1351			}
1352			idx = log_next(idx);
1353			seq++;
1354
1355			raw_spin_unlock_irq(&logbuf_lock);
1356			if (copy_to_user(buf + len, text, textlen))
1357				len = -EFAULT;
1358			else
1359				len += textlen;
1360			raw_spin_lock_irq(&logbuf_lock);
1361
1362			if (seq < log_first_seq) {
1363				/* messages are gone, move to next one */
1364				seq = log_first_seq;
1365				idx = log_first_idx;
1366			}
1367		}
1368	}
1369
1370	if (clear) {
1371		clear_seq = log_next_seq;
1372		clear_idx = log_next_idx;
1373	}
1374	raw_spin_unlock_irq(&logbuf_lock);
1375
1376	kfree(text);
1377	return len;
1378}
1379
1380int do_syslog(int type, char __user *buf, int len, int source)
1381{
1382	bool clear = false;
1383	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1384	int error;
1385
1386	error = check_syslog_permissions(type, source);
1387	if (error)
1388		goto out;
1389
1390	switch (type) {
1391	case SYSLOG_ACTION_CLOSE:	/* Close log */
1392		break;
1393	case SYSLOG_ACTION_OPEN:	/* Open log */
1394		break;
1395	case SYSLOG_ACTION_READ:	/* Read from log */
1396		error = -EINVAL;
1397		if (!buf || len < 0)
1398			goto out;
1399		error = 0;
1400		if (!len)
1401			goto out;
1402		if (!access_ok(VERIFY_WRITE, buf, len)) {
1403			error = -EFAULT;
1404			goto out;
1405		}
1406		error = wait_event_interruptible(log_wait,
1407						 syslog_seq != log_next_seq);
1408		if (error)
1409			goto out;
1410		error = syslog_print(buf, len);
1411		break;
1412	/* Read/clear last kernel messages */
1413	case SYSLOG_ACTION_READ_CLEAR:
1414		clear = true;
1415		/* FALL THRU */
1416	/* Read last kernel messages */
1417	case SYSLOG_ACTION_READ_ALL:
1418		error = -EINVAL;
1419		if (!buf || len < 0)
1420			goto out;
1421		error = 0;
1422		if (!len)
1423			goto out;
1424		if (!access_ok(VERIFY_WRITE, buf, len)) {
1425			error = -EFAULT;
1426			goto out;
1427		}
1428		error = syslog_print_all(buf, len, clear);
1429		break;
1430	/* Clear ring buffer */
1431	case SYSLOG_ACTION_CLEAR:
1432		syslog_print_all(NULL, 0, true);
1433		break;
1434	/* Disable logging to console */
1435	case SYSLOG_ACTION_CONSOLE_OFF:
1436		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1437			saved_console_loglevel = console_loglevel;
1438		console_loglevel = minimum_console_loglevel;
1439		break;
1440	/* Enable logging to console */
1441	case SYSLOG_ACTION_CONSOLE_ON:
1442		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1443			console_loglevel = saved_console_loglevel;
1444			saved_console_loglevel = LOGLEVEL_DEFAULT;
1445		}
1446		break;
1447	/* Set level of messages printed to console */
1448	case SYSLOG_ACTION_CONSOLE_LEVEL:
1449		error = -EINVAL;
1450		if (len < 1 || len > 8)
1451			goto out;
1452		if (len < minimum_console_loglevel)
1453			len = minimum_console_loglevel;
1454		console_loglevel = len;
1455		/* Implicitly re-enable logging to console */
1456		saved_console_loglevel = LOGLEVEL_DEFAULT;
1457		error = 0;
1458		break;
1459	/* Number of chars in the log buffer */
1460	case SYSLOG_ACTION_SIZE_UNREAD:
1461		raw_spin_lock_irq(&logbuf_lock);
1462		if (syslog_seq < log_first_seq) {
1463			/* messages are gone, move to first one */
1464			syslog_seq = log_first_seq;
1465			syslog_idx = log_first_idx;
1466			syslog_partial = 0;
1467		}
1468		if (source == SYSLOG_FROM_PROC) {
1469			/*
1470			 * Short-cut for poll(/"proc/kmsg") which simply checks
1471			 * for pending data, not the size; return the count of
1472			 * records, not the length.
1473			 */
1474			error = log_next_seq - syslog_seq;
1475		} else {
1476			u64 seq = syslog_seq;
1477			u32 idx = syslog_idx;
1478
1479			error = 0;
1480			while (seq < log_next_seq) {
1481				struct printk_log *msg = log_from_idx(idx);
1482
1483				error += msg_print_text(msg, true, NULL, 0);
1484				idx = log_next(idx);
1485				seq++;
1486			}
1487			error -= syslog_partial;
1488		}
1489		raw_spin_unlock_irq(&logbuf_lock);
1490		break;
1491	/* Size of the log buffer */
1492	case SYSLOG_ACTION_SIZE_BUFFER:
1493		error = log_buf_len;
1494		break;
1495	default:
1496		error = -EINVAL;
1497		break;
1498	}
1499out:
1500	return error;
1501}
1502
1503SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1504{
1505	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1506}
1507
1508/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1509 * Call the console drivers, asking them to write out
1510 * log_buf[start] to log_buf[end - 1].
1511 * The console_lock must be held.
1512 */
1513static void call_console_drivers(int level,
1514				 const char *ext_text, size_t ext_len,
1515				 const char *text, size_t len)
1516{
1517	struct console *con;
1518
1519	trace_console_rcuidle(text, len);
1520
1521	if (!console_drivers)
1522		return;
1523
1524	for_each_console(con) {
1525		if (exclusive_console && con != exclusive_console)
1526			continue;
1527		if (!(con->flags & CON_ENABLED))
1528			continue;
1529		if (!con->write)
1530			continue;
1531		if (!cpu_online(smp_processor_id()) &&
1532		    !(con->flags & CON_ANYTIME))
1533			continue;
1534		if (con->flags & CON_EXTENDED)
1535			con->write(con, ext_text, ext_len);
1536		else
1537			con->write(con, text, len);
1538	}
1539}
1540
1541/*
1542 * Zap console related locks when oopsing.
1543 * To leave time for slow consoles to print a full oops,
1544 * only zap at most once every 30 seconds.
1545 */
1546static void zap_locks(void)
1547{
1548	static unsigned long oops_timestamp;
1549
1550	if (time_after_eq(jiffies, oops_timestamp) &&
1551	    !time_after(jiffies, oops_timestamp + 30 * HZ))
1552		return;
1553
1554	oops_timestamp = jiffies;
1555
1556	debug_locks_off();
1557	/* If a crash is occurring, make sure we can't deadlock */
1558	raw_spin_lock_init(&logbuf_lock);
1559	/* And make sure that we print immediately */
1560	sema_init(&console_sem, 1);
1561}
1562
1563int printk_delay_msec __read_mostly;
1564
1565static inline void printk_delay(void)
1566{
1567	if (unlikely(printk_delay_msec)) {
1568		int m = printk_delay_msec;
1569
1570		while (m--) {
1571			mdelay(1);
1572			touch_nmi_watchdog();
1573		}
1574	}
1575}
1576
1577/*
1578 * Continuation lines are buffered, and not committed to the record buffer
1579 * until the line is complete, or a race forces it. The line fragments
1580 * though, are printed immediately to the consoles to ensure everything has
1581 * reached the console in case of a kernel crash.
1582 */
1583static struct cont {
1584	char buf[LOG_LINE_MAX];
1585	size_t len;			/* length == 0 means unused buffer */
1586	struct task_struct *owner;	/* task of first print*/
1587	u64 ts_nsec;			/* time of first print */
1588	u8 level;			/* log level of first message */
1589	u8 facility;			/* log facility of first message */
1590	enum log_flags flags;		/* prefix, newline flags */
1591} cont;
1592
1593static void cont_flush(void)
1594{
1595	if (cont.len == 0)
1596		return;
1597
1598	log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1599		  NULL, 0, cont.buf, cont.len);
1600	cont.len = 0;
1601}
1602
1603static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1604{
1605	/*
1606	 * If ext consoles are present, flush and skip in-kernel
1607	 * continuation.  See nr_ext_console_drivers definition.  Also, if
1608	 * the line gets too long, split it up in separate records.
1609	 */
1610	if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1611		cont_flush();
1612		return false;
1613	}
1614
1615	if (!cont.len) {
1616		cont.facility = facility;
1617		cont.level = level;
1618		cont.owner = current;
1619		cont.ts_nsec = local_clock();
1620		cont.flags = flags;
1621	}
1622
1623	memcpy(cont.buf + cont.len, text, len);
1624	cont.len += len;
1625
1626	// The original flags come from the first line,
1627	// but later continuations can add a newline.
1628	if (flags & LOG_NEWLINE) {
1629		cont.flags |= LOG_NEWLINE;
1630		cont_flush();
1631	}
1632
1633	if (cont.len > (sizeof(cont.buf) * 80) / 100)
1634		cont_flush();
1635
1636	return true;
1637}
1638
1639static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1640{
1641	/*
1642	 * If an earlier line was buffered, and we're a continuation
1643	 * write from the same process, try to add it to the buffer.
1644	 */
1645	if (cont.len) {
1646		if (cont.owner == current && (lflags & LOG_CONT)) {
1647			if (cont_add(facility, level, lflags, text, text_len))
1648				return text_len;
1649		}
1650		/* Otherwise, make sure it's flushed */
1651		cont_flush();
1652	}
1653
1654	/* Skip empty continuation lines that couldn't be added - they just flush */
1655	if (!text_len && (lflags & LOG_CONT))
1656		return 0;
1657
1658	/* If it doesn't end in a newline, try to buffer the current line */
1659	if (!(lflags & LOG_NEWLINE)) {
1660		if (cont_add(facility, level, lflags, text, text_len))
1661			return text_len;
1662	}
1663
1664	/* Store it in the record log */
1665	return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1666}
1667
1668asmlinkage int vprintk_emit(int facility, int level,
1669			    const char *dict, size_t dictlen,
1670			    const char *fmt, va_list args)
1671{
1672	static bool recursion_bug;
1673	static char textbuf[LOG_LINE_MAX];
1674	char *text = textbuf;
1675	size_t text_len = 0;
1676	enum log_flags lflags = 0;
1677	unsigned long flags;
1678	int this_cpu;
1679	int printed_len = 0;
1680	int nmi_message_lost;
1681	bool in_sched = false;
1682	/* cpu currently holding logbuf_lock in this function */
1683	static unsigned int logbuf_cpu = UINT_MAX;
1684
1685	if (level == LOGLEVEL_SCHED) {
1686		level = LOGLEVEL_DEFAULT;
1687		in_sched = true;
1688	}
1689
1690	boot_delay_msec(level);
1691	printk_delay();
1692
1693	local_irq_save(flags);
1694	this_cpu = smp_processor_id();
1695
1696	/*
1697	 * Ouch, printk recursed into itself!
1698	 */
1699	if (unlikely(logbuf_cpu == this_cpu)) {
1700		/*
1701		 * If a crash is occurring during printk() on this CPU,
1702		 * then try to get the crash message out but make sure
1703		 * we can't deadlock. Otherwise just return to avoid the
1704		 * recursion and return - but flag the recursion so that
1705		 * it can be printed at the next appropriate moment:
1706		 */
1707		if (!oops_in_progress && !lockdep_recursing(current)) {
1708			recursion_bug = true;
1709			local_irq_restore(flags);
1710			return 0;
1711		}
1712		zap_locks();
1713	}
1714
1715	lockdep_off();
1716	/* This stops the holder of console_sem just where we want him */
1717	raw_spin_lock(&logbuf_lock);
1718	logbuf_cpu = this_cpu;
1719
1720	if (unlikely(recursion_bug)) {
1721		static const char recursion_msg[] =
1722			"BUG: recent printk recursion!";
1723
1724		recursion_bug = false;
1725		/* emit KERN_CRIT message */
1726		printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1727					 NULL, 0, recursion_msg,
1728					 strlen(recursion_msg));
1729	}
1730
1731	nmi_message_lost = get_nmi_message_lost();
1732	if (unlikely(nmi_message_lost)) {
1733		text_len = scnprintf(textbuf, sizeof(textbuf),
1734				     "BAD LUCK: lost %d message(s) from NMI context!",
1735				     nmi_message_lost);
1736		printed_len += log_store(0, 2, LOG_PREFIX|LOG_NEWLINE, 0,
1737					 NULL, 0, textbuf, text_len);
1738	}
1739
1740	/*
1741	 * The printf needs to come first; we need the syslog
1742	 * prefix which might be passed-in as a parameter.
1743	 */
1744	text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1745
1746	/* mark and strip a trailing newline */
1747	if (text_len && text[text_len-1] == '\n') {
1748		text_len--;
1749		lflags |= LOG_NEWLINE;
1750	}
1751
1752	/* strip kernel syslog prefix and extract log level or control flags */
1753	if (facility == 0) {
1754		int kern_level;
1755
1756		while ((kern_level = printk_get_level(text)) != 0) {
1757			switch (kern_level) {
1758			case '0' ... '7':
1759				if (level == LOGLEVEL_DEFAULT)
1760					level = kern_level - '0';
1761				/* fallthrough */
1762			case 'd':	/* KERN_DEFAULT */
1763				lflags |= LOG_PREFIX;
1764				break;
1765			case 'c':	/* KERN_CONT */
1766				lflags |= LOG_CONT;
1767			}
1768
1769			text_len -= 2;
1770			text += 2;
1771		}
1772	}
1773
1774	if (level == LOGLEVEL_DEFAULT)
1775		level = default_message_loglevel;
1776
1777	if (dict)
1778		lflags |= LOG_PREFIX|LOG_NEWLINE;
1779
1780	printed_len += log_output(facility, level, lflags, dict, dictlen, text, text_len);
1781
1782	logbuf_cpu = UINT_MAX;
1783	raw_spin_unlock(&logbuf_lock);
1784	lockdep_on();
1785	local_irq_restore(flags);
1786
1787	/* If called from the scheduler, we can not call up(). */
1788	if (!in_sched) {
1789		lockdep_off();
 
 
 
 
 
1790		/*
1791		 * Try to acquire and then immediately release the console
1792		 * semaphore.  The release will print out buffers and wake up
1793		 * /dev/kmsg and syslog() users.
1794		 */
1795		if (console_trylock())
1796			console_unlock();
1797		lockdep_on();
1798	}
1799
1800	return printed_len;
1801}
1802EXPORT_SYMBOL(vprintk_emit);
1803
1804asmlinkage int vprintk(const char *fmt, va_list args)
1805{
1806	return vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1807}
1808EXPORT_SYMBOL(vprintk);
1809
1810asmlinkage int printk_emit(int facility, int level,
1811			   const char *dict, size_t dictlen,
1812			   const char *fmt, ...)
1813{
1814	va_list args;
1815	int r;
1816
1817	va_start(args, fmt);
1818	r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1819	va_end(args);
1820
1821	return r;
1822}
1823EXPORT_SYMBOL(printk_emit);
1824
1825int vprintk_default(const char *fmt, va_list args)
1826{
1827	int r;
1828
1829#ifdef CONFIG_KGDB_KDB
1830	/* Allow to pass printk() to kdb but avoid a recursion. */
1831	if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1832		r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1833		return r;
1834	}
1835#endif
1836	r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1837
1838	return r;
1839}
1840EXPORT_SYMBOL_GPL(vprintk_default);
1841
1842/**
1843 * printk - print a kernel message
1844 * @fmt: format string
1845 *
1846 * This is printk(). It can be called from any context. We want it to work.
1847 *
1848 * We try to grab the console_lock. If we succeed, it's easy - we log the
1849 * output and call the console drivers.  If we fail to get the semaphore, we
1850 * place the output into the log buffer and return. The current holder of
1851 * the console_sem will notice the new output in console_unlock(); and will
1852 * send it to the consoles before releasing the lock.
1853 *
1854 * One effect of this deferred printing is that code which calls printk() and
1855 * then changes console_loglevel may break. This is because console_loglevel
1856 * is inspected when the actual printing occurs.
1857 *
1858 * See also:
1859 * printf(3)
1860 *
1861 * See the vsnprintf() documentation for format string extensions over C99.
1862 */
1863asmlinkage __visible int printk(const char *fmt, ...)
1864{
1865	va_list args;
1866	int r;
1867
1868	va_start(args, fmt);
1869	r = vprintk_func(fmt, args);
1870	va_end(args);
1871
1872	return r;
1873}
1874EXPORT_SYMBOL(printk);
1875
1876#else /* CONFIG_PRINTK */
1877
1878#define LOG_LINE_MAX		0
1879#define PREFIX_MAX		0
1880
1881static u64 syslog_seq;
1882static u32 syslog_idx;
1883static u64 console_seq;
1884static u32 console_idx;
1885static u64 log_first_seq;
1886static u32 log_first_idx;
1887static u64 log_next_seq;
1888static char *log_text(const struct printk_log *msg) { return NULL; }
1889static char *log_dict(const struct printk_log *msg) { return NULL; }
1890static struct printk_log *log_from_idx(u32 idx) { return NULL; }
1891static u32 log_next(u32 idx) { return 0; }
1892static ssize_t msg_print_ext_header(char *buf, size_t size,
1893				    struct printk_log *msg,
1894				    u64 seq) { return 0; }
1895static ssize_t msg_print_ext_body(char *buf, size_t size,
1896				  char *dict, size_t dict_len,
1897				  char *text, size_t text_len) { return 0; }
1898static void call_console_drivers(int level,
1899				 const char *ext_text, size_t ext_len,
 
1900				 const char *text, size_t len) {}
1901static size_t msg_print_text(const struct printk_log *msg,
1902			     bool syslog, char *buf, size_t size) { return 0; }
1903static bool suppress_message_printing(int level) { return false; }
1904
1905/* Still needs to be defined for users */
1906DEFINE_PER_CPU(printk_func_t, printk_func);
1907
1908#endif /* CONFIG_PRINTK */
1909
1910#ifdef CONFIG_EARLY_PRINTK
1911struct console *early_console;
1912
1913asmlinkage __visible void early_printk(const char *fmt, ...)
1914{
1915	va_list ap;
1916	char buf[512];
1917	int n;
1918
1919	if (!early_console)
1920		return;
1921
1922	va_start(ap, fmt);
1923	n = vscnprintf(buf, sizeof(buf), fmt, ap);
1924	va_end(ap);
1925
1926	early_console->write(early_console, buf, n);
1927}
1928#endif
1929
1930static int __add_preferred_console(char *name, int idx, char *options,
1931				   char *brl_options)
1932{
1933	struct console_cmdline *c;
1934	int i;
1935
1936	/*
1937	 *	See if this tty is not yet registered, and
1938	 *	if we have a slot free.
1939	 */
1940	for (i = 0, c = console_cmdline;
1941	     i < MAX_CMDLINECONSOLES && c->name[0];
1942	     i++, c++) {
1943		if (strcmp(c->name, name) == 0 && c->index == idx) {
1944			if (!brl_options)
1945				selected_console = i;
1946			return 0;
1947		}
1948	}
1949	if (i == MAX_CMDLINECONSOLES)
1950		return -E2BIG;
1951	if (!brl_options)
1952		selected_console = i;
1953	strlcpy(c->name, name, sizeof(c->name));
1954	c->options = options;
1955	braille_set_options(c, brl_options);
1956
1957	c->index = idx;
1958	return 0;
1959}
 
 
 
 
 
 
 
 
 
 
 
1960/*
1961 * Set up a console.  Called via do_early_param() in init/main.c
1962 * for each "console=" parameter in the boot command line.
1963 */
1964static int __init console_setup(char *str)
1965{
1966	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
1967	char *s, *options, *brl_options = NULL;
1968	int idx;
1969
1970	if (_braille_console_setup(&str, &brl_options))
1971		return 1;
1972
1973	/*
1974	 * Decode str into name, index, options.
1975	 */
1976	if (str[0] >= '0' && str[0] <= '9') {
1977		strcpy(buf, "ttyS");
1978		strncpy(buf + 4, str, sizeof(buf) - 5);
1979	} else {
1980		strncpy(buf, str, sizeof(buf) - 1);
1981	}
1982	buf[sizeof(buf) - 1] = 0;
1983	options = strchr(str, ',');
1984	if (options)
1985		*(options++) = 0;
1986#ifdef __sparc__
1987	if (!strcmp(str, "ttya"))
1988		strcpy(buf, "ttyS0");
1989	if (!strcmp(str, "ttyb"))
1990		strcpy(buf, "ttyS1");
1991#endif
1992	for (s = buf; *s; s++)
1993		if (isdigit(*s) || *s == ',')
1994			break;
1995	idx = simple_strtoul(s, NULL, 10);
1996	*s = 0;
1997
1998	__add_preferred_console(buf, idx, options, brl_options);
1999	console_set_on_cmdline = 1;
2000	return 1;
2001}
2002__setup("console=", console_setup);
2003
2004/**
2005 * add_preferred_console - add a device to the list of preferred consoles.
2006 * @name: device name
2007 * @idx: device index
2008 * @options: options for this console
2009 *
2010 * The last preferred console added will be used for kernel messages
2011 * and stdin/out/err for init.  Normally this is used by console_setup
2012 * above to handle user-supplied console arguments; however it can also
2013 * be used by arch-specific code either to override the user or more
2014 * commonly to provide a default console (ie from PROM variables) when
2015 * the user has not supplied one.
2016 */
2017int add_preferred_console(char *name, int idx, char *options)
2018{
2019	return __add_preferred_console(name, idx, options, NULL);
2020}
2021
2022bool console_suspend_enabled = true;
2023EXPORT_SYMBOL(console_suspend_enabled);
2024
2025static int __init console_suspend_disable(char *str)
2026{
2027	console_suspend_enabled = false;
2028	return 1;
2029}
2030__setup("no_console_suspend", console_suspend_disable);
2031module_param_named(console_suspend, console_suspend_enabled,
2032		bool, S_IRUGO | S_IWUSR);
2033MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2034	" and hibernate operations");
2035
2036/**
2037 * suspend_console - suspend the console subsystem
2038 *
2039 * This disables printk() while we go into suspend states
2040 */
2041void suspend_console(void)
2042{
2043	if (!console_suspend_enabled)
2044		return;
2045	printk("Suspending console(s) (use no_console_suspend to debug)\n");
2046	console_lock();
2047	console_suspended = 1;
2048	up_console_sem();
2049}
2050
2051void resume_console(void)
2052{
2053	if (!console_suspend_enabled)
2054		return;
2055	down_console_sem();
2056	console_suspended = 0;
2057	console_unlock();
2058}
2059
2060/**
2061 * console_cpu_notify - print deferred console messages after CPU hotplug
2062 * @cpu: unused
2063 *
2064 * If printk() is called from a CPU that is not online yet, the messages
2065 * will be spooled but will not show up on the console.  This function is
2066 * called when a new CPU comes online (or fails to come up), and ensures
2067 * that any such output gets printed.
2068 */
2069static int console_cpu_notify(unsigned int cpu)
2070{
2071	if (!cpuhp_tasks_frozen) {
2072		console_lock();
2073		console_unlock();
 
2074	}
2075	return 0;
2076}
2077
2078/**
2079 * console_lock - lock the console system for exclusive use.
2080 *
2081 * Acquires a lock which guarantees that the caller has
2082 * exclusive access to the console system and the console_drivers list.
2083 *
2084 * Can sleep, returns nothing.
2085 */
2086void console_lock(void)
2087{
2088	might_sleep();
2089
2090	down_console_sem();
2091	if (console_suspended)
2092		return;
2093	console_locked = 1;
2094	console_may_schedule = 1;
2095}
2096EXPORT_SYMBOL(console_lock);
2097
2098/**
2099 * console_trylock - try to lock the console system for exclusive use.
2100 *
2101 * Try to acquire a lock which guarantees that the caller has exclusive
2102 * access to the console system and the console_drivers list.
2103 *
2104 * returns 1 on success, and 0 on failure to acquire the lock.
2105 */
2106int console_trylock(void)
2107{
2108	if (down_trylock_console_sem())
2109		return 0;
2110	if (console_suspended) {
2111		up_console_sem();
2112		return 0;
2113	}
2114	console_locked = 1;
2115	/*
2116	 * When PREEMPT_COUNT disabled we can't reliably detect if it's
2117	 * safe to schedule (e.g. calling printk while holding a spin_lock),
2118	 * because preempt_disable()/preempt_enable() are just barriers there
2119	 * and preempt_count() is always 0.
2120	 *
2121	 * RCU read sections have a separate preemption counter when
2122	 * PREEMPT_RCU enabled thus we must take extra care and check
2123	 * rcu_preempt_depth(), otherwise RCU read sections modify
2124	 * preempt_count().
2125	 */
2126	console_may_schedule = !oops_in_progress &&
2127			preemptible() &&
2128			!rcu_preempt_depth();
2129	return 1;
2130}
2131EXPORT_SYMBOL(console_trylock);
2132
2133int is_console_locked(void)
2134{
2135	return console_locked;
2136}
2137
2138/*
2139 * Check if we have any console that is capable of printing while cpu is
2140 * booting or shutting down. Requires console_sem.
2141 */
2142static int have_callable_console(void)
2143{
2144	struct console *con;
2145
2146	for_each_console(con)
2147		if ((con->flags & CON_ENABLED) &&
2148				(con->flags & CON_ANYTIME))
2149			return 1;
2150
2151	return 0;
2152}
2153
2154/*
2155 * Can we actually use the console at this time on this cpu?
2156 *
2157 * Console drivers may assume that per-cpu resources have been allocated. So
2158 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2159 * call them until this CPU is officially up.
2160 */
2161static inline int can_use_console(void)
2162{
2163	return cpu_online(raw_smp_processor_id()) || have_callable_console();
2164}
2165
2166/**
2167 * console_unlock - unlock the console system
2168 *
2169 * Releases the console_lock which the caller holds on the console system
2170 * and the console driver list.
2171 *
2172 * While the console_lock was held, console output may have been buffered
2173 * by printk().  If this is the case, console_unlock(); emits
2174 * the output prior to releasing the lock.
2175 *
2176 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2177 *
2178 * console_unlock(); may be called from any context.
2179 */
2180void console_unlock(void)
2181{
2182	static char ext_text[CONSOLE_EXT_LOG_MAX];
2183	static char text[LOG_LINE_MAX + PREFIX_MAX];
2184	static u64 seen_seq;
2185	unsigned long flags;
2186	bool wake_klogd = false;
2187	bool do_cond_resched, retry;
2188
2189	if (console_suspended) {
2190		up_console_sem();
2191		return;
2192	}
2193
2194	/*
2195	 * Console drivers are called under logbuf_lock, so
2196	 * @console_may_schedule should be cleared before; however, we may
2197	 * end up dumping a lot of lines, for example, if called from
2198	 * console registration path, and should invoke cond_resched()
2199	 * between lines if allowable.  Not doing so can cause a very long
2200	 * scheduling stall on a slow console leading to RCU stall and
2201	 * softlockup warnings which exacerbate the issue with more
2202	 * messages practically incapacitating the system.
 
 
 
 
2203	 */
2204	do_cond_resched = console_may_schedule;
 
2205	console_may_schedule = 0;
2206
2207again:
2208	/*
2209	 * We released the console_sem lock, so we need to recheck if
2210	 * cpu is online and (if not) is there at least one CON_ANYTIME
2211	 * console.
2212	 */
2213	if (!can_use_console()) {
2214		console_locked = 0;
2215		up_console_sem();
2216		return;
2217	}
2218
2219	for (;;) {
2220		struct printk_log *msg;
2221		size_t ext_len = 0;
2222		size_t len;
2223		int level;
2224
2225		raw_spin_lock_irqsave(&logbuf_lock, flags);
 
2226		if (seen_seq != log_next_seq) {
2227			wake_klogd = true;
2228			seen_seq = log_next_seq;
2229		}
2230
2231		if (console_seq < log_first_seq) {
2232			len = sprintf(text, "** %u printk messages dropped ** ",
2233				      (unsigned)(log_first_seq - console_seq));
2234
2235			/* messages are gone, move to first one */
2236			console_seq = log_first_seq;
2237			console_idx = log_first_idx;
2238		} else {
2239			len = 0;
2240		}
2241skip:
2242		if (console_seq == log_next_seq)
2243			break;
2244
2245		msg = log_from_idx(console_idx);
2246		level = msg->level;
2247		if (suppress_message_printing(level)) {
2248			/*
2249			 * Skip record we have buffered and already printed
2250			 * directly to the console when we received it, and
2251			 * record that has level above the console loglevel.
2252			 */
2253			console_idx = log_next(console_idx);
2254			console_seq++;
2255			goto skip;
2256		}
2257
2258		len += msg_print_text(msg, false, text + len, sizeof(text) - len);
 
 
 
2259		if (nr_ext_console_drivers) {
2260			ext_len = msg_print_ext_header(ext_text,
2261						sizeof(ext_text),
2262						msg, console_seq);
2263			ext_len += msg_print_ext_body(ext_text + ext_len,
2264						sizeof(ext_text) - ext_len,
2265						log_dict(msg), msg->dict_len,
2266						log_text(msg), msg->text_len);
2267		}
2268		console_idx = log_next(console_idx);
2269		console_seq++;
2270		raw_spin_unlock(&logbuf_lock);
2271
 
 
 
 
 
 
 
 
2272		stop_critical_timings();	/* don't trace print latency */
2273		call_console_drivers(level, ext_text, ext_len, text, len);
2274		start_critical_timings();
2275		local_irq_restore(flags);
 
 
 
 
 
 
2276
2277		if (do_cond_resched)
2278			cond_resched();
2279	}
 
2280	console_locked = 0;
2281
2282	/* Release the exclusive_console once it is used */
2283	if (unlikely(exclusive_console))
2284		exclusive_console = NULL;
2285
2286	raw_spin_unlock(&logbuf_lock);
2287
2288	up_console_sem();
2289
2290	/*
2291	 * Someone could have filled up the buffer again, so re-check if there's
2292	 * something to flush. In case we cannot trylock the console_sem again,
2293	 * there's a new owner and the console_unlock() from them will do the
2294	 * flush, no worries.
2295	 */
2296	raw_spin_lock(&logbuf_lock);
2297	retry = console_seq != log_next_seq;
2298	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
 
2299
2300	if (retry && console_trylock())
2301		goto again;
2302
 
2303	if (wake_klogd)
2304		wake_up_klogd();
2305}
2306EXPORT_SYMBOL(console_unlock);
2307
2308/**
2309 * console_conditional_schedule - yield the CPU if required
2310 *
2311 * If the console code is currently allowed to sleep, and
2312 * if this CPU should yield the CPU to another task, do
2313 * so here.
2314 *
2315 * Must be called within console_lock();.
2316 */
2317void __sched console_conditional_schedule(void)
2318{
2319	if (console_may_schedule)
2320		cond_resched();
2321}
2322EXPORT_SYMBOL(console_conditional_schedule);
2323
2324void console_unblank(void)
2325{
2326	struct console *c;
2327
2328	/*
2329	 * console_unblank can no longer be called in interrupt context unless
2330	 * oops_in_progress is set to 1..
2331	 */
2332	if (oops_in_progress) {
2333		if (down_trylock_console_sem() != 0)
2334			return;
2335	} else
2336		console_lock();
2337
2338	console_locked = 1;
2339	console_may_schedule = 0;
2340	for_each_console(c)
2341		if ((c->flags & CON_ENABLED) && c->unblank)
2342			c->unblank();
2343	console_unlock();
2344}
2345
2346/**
2347 * console_flush_on_panic - flush console content on panic
2348 *
2349 * Immediately output all pending messages no matter what.
2350 */
2351void console_flush_on_panic(void)
2352{
2353	/*
2354	 * If someone else is holding the console lock, trylock will fail
2355	 * and may_schedule may be set.  Ignore and proceed to unlock so
2356	 * that messages are flushed out.  As this can be called from any
2357	 * context and we don't want to get preempted while flushing,
2358	 * ensure may_schedule is cleared.
2359	 */
2360	console_trylock();
2361	console_may_schedule = 0;
2362	console_unlock();
2363}
2364
2365/*
2366 * Return the console tty driver structure and its associated index
2367 */
2368struct tty_driver *console_device(int *index)
2369{
2370	struct console *c;
2371	struct tty_driver *driver = NULL;
2372
2373	console_lock();
2374	for_each_console(c) {
2375		if (!c->device)
2376			continue;
2377		driver = c->device(c, index);
2378		if (driver)
2379			break;
2380	}
2381	console_unlock();
2382	return driver;
2383}
2384
2385/*
2386 * Prevent further output on the passed console device so that (for example)
2387 * serial drivers can disable console output before suspending a port, and can
2388 * re-enable output afterwards.
2389 */
2390void console_stop(struct console *console)
2391{
2392	console_lock();
2393	console->flags &= ~CON_ENABLED;
2394	console_unlock();
2395}
2396EXPORT_SYMBOL(console_stop);
2397
2398void console_start(struct console *console)
2399{
2400	console_lock();
2401	console->flags |= CON_ENABLED;
2402	console_unlock();
2403}
2404EXPORT_SYMBOL(console_start);
2405
2406static int __read_mostly keep_bootcon;
2407
2408static int __init keep_bootcon_setup(char *str)
2409{
2410	keep_bootcon = 1;
2411	pr_info("debug: skip boot console de-registration.\n");
2412
2413	return 0;
2414}
2415
2416early_param("keep_bootcon", keep_bootcon_setup);
2417
2418/*
2419 * The console driver calls this routine during kernel initialization
2420 * to register the console printing procedure with printk() and to
2421 * print any messages that were printed by the kernel before the
2422 * console driver was initialized.
2423 *
2424 * This can happen pretty early during the boot process (because of
2425 * early_printk) - sometimes before setup_arch() completes - be careful
2426 * of what kernel features are used - they may not be initialised yet.
2427 *
2428 * There are two types of consoles - bootconsoles (early_printk) and
2429 * "real" consoles (everything which is not a bootconsole) which are
2430 * handled differently.
2431 *  - Any number of bootconsoles can be registered at any time.
2432 *  - As soon as a "real" console is registered, all bootconsoles
2433 *    will be unregistered automatically.
2434 *  - Once a "real" console is registered, any attempt to register a
2435 *    bootconsoles will be rejected
2436 */
2437void register_console(struct console *newcon)
2438{
2439	int i;
2440	unsigned long flags;
2441	struct console *bcon = NULL;
2442	struct console_cmdline *c;
 
2443
2444	if (console_drivers)
2445		for_each_console(bcon)
2446			if (WARN(bcon == newcon,
2447					"console '%s%d' already registered\n",
2448					bcon->name, bcon->index))
2449				return;
2450
2451	/*
2452	 * before we register a new CON_BOOT console, make sure we don't
2453	 * already have a valid console
2454	 */
2455	if (console_drivers && newcon->flags & CON_BOOT) {
2456		/* find the last or real console */
2457		for_each_console(bcon) {
2458			if (!(bcon->flags & CON_BOOT)) {
2459				pr_info("Too late to register bootconsole %s%d\n",
2460					newcon->name, newcon->index);
2461				return;
2462			}
2463		}
2464	}
2465
2466	if (console_drivers && console_drivers->flags & CON_BOOT)
2467		bcon = console_drivers;
2468
2469	if (preferred_console < 0 || bcon || !console_drivers)
2470		preferred_console = selected_console;
2471
2472	/*
2473	 *	See if we want to use this console driver. If we
2474	 *	didn't select a console we take the first one
2475	 *	that registers here.
2476	 */
2477	if (preferred_console < 0) {
2478		if (newcon->index < 0)
2479			newcon->index = 0;
2480		if (newcon->setup == NULL ||
2481		    newcon->setup(newcon, NULL) == 0) {
2482			newcon->flags |= CON_ENABLED;
2483			if (newcon->device) {
2484				newcon->flags |= CON_CONSDEV;
2485				preferred_console = 0;
2486			}
2487		}
2488	}
2489
2490	/*
2491	 *	See if this console matches one we selected on
2492	 *	the command line.
2493	 */
2494	for (i = 0, c = console_cmdline;
2495	     i < MAX_CMDLINECONSOLES && c->name[0];
2496	     i++, c++) {
2497		if (!newcon->match ||
2498		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
2499			/* default matching */
2500			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2501			if (strcmp(c->name, newcon->name) != 0)
2502				continue;
2503			if (newcon->index >= 0 &&
2504			    newcon->index != c->index)
2505				continue;
2506			if (newcon->index < 0)
2507				newcon->index = c->index;
2508
2509			if (_braille_register_console(newcon, c))
2510				return;
2511
2512			if (newcon->setup &&
2513			    newcon->setup(newcon, c->options) != 0)
2514				break;
2515		}
2516
2517		newcon->flags |= CON_ENABLED;
2518		if (i == selected_console) {
2519			newcon->flags |= CON_CONSDEV;
2520			preferred_console = selected_console;
2521		}
2522		break;
2523	}
2524
2525	if (!(newcon->flags & CON_ENABLED))
2526		return;
2527
2528	/*
2529	 * If we have a bootconsole, and are switching to a real console,
2530	 * don't print everything out again, since when the boot console, and
2531	 * the real console are the same physical device, it's annoying to
2532	 * see the beginning boot messages twice
2533	 */
2534	if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2535		newcon->flags &= ~CON_PRINTBUFFER;
2536
2537	/*
2538	 *	Put this console in the list - keep the
2539	 *	preferred driver at the head of the list.
2540	 */
2541	console_lock();
2542	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2543		newcon->next = console_drivers;
2544		console_drivers = newcon;
2545		if (newcon->next)
2546			newcon->next->flags &= ~CON_CONSDEV;
2547	} else {
2548		newcon->next = console_drivers->next;
2549		console_drivers->next = newcon;
2550	}
2551
2552	if (newcon->flags & CON_EXTENDED)
2553		if (!nr_ext_console_drivers++)
2554			pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2555
2556	if (newcon->flags & CON_PRINTBUFFER) {
2557		/*
2558		 * console_unlock(); will print out the buffered messages
2559		 * for us.
2560		 */
2561		raw_spin_lock_irqsave(&logbuf_lock, flags);
2562		console_seq = syslog_seq;
2563		console_idx = syslog_idx;
2564		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2565		/*
2566		 * We're about to replay the log buffer.  Only do this to the
2567		 * just-registered console to avoid excessive message spam to
2568		 * the already-registered consoles.
2569		 */
2570		exclusive_console = newcon;
2571	}
2572	console_unlock();
2573	console_sysfs_notify();
2574
2575	/*
2576	 * By unregistering the bootconsoles after we enable the real console
2577	 * we get the "console xxx enabled" message on all the consoles -
2578	 * boot consoles, real consoles, etc - this is to ensure that end
2579	 * users know there might be something in the kernel's log buffer that
2580	 * went to the bootconsole (that they do not see on the real console)
2581	 */
2582	pr_info("%sconsole [%s%d] enabled\n",
2583		(newcon->flags & CON_BOOT) ? "boot" : "" ,
2584		newcon->name, newcon->index);
2585	if (bcon &&
2586	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2587	    !keep_bootcon) {
2588		/* We need to iterate through all boot consoles, to make
2589		 * sure we print everything out, before we unregister them.
2590		 */
2591		for_each_console(bcon)
2592			if (bcon->flags & CON_BOOT)
2593				unregister_console(bcon);
2594	}
2595}
2596EXPORT_SYMBOL(register_console);
2597
2598int unregister_console(struct console *console)
2599{
2600        struct console *a, *b;
2601	int res;
2602
2603	pr_info("%sconsole [%s%d] disabled\n",
2604		(console->flags & CON_BOOT) ? "boot" : "" ,
2605		console->name, console->index);
2606
2607	res = _braille_unregister_console(console);
2608	if (res)
2609		return res;
2610
2611	res = 1;
2612	console_lock();
2613	if (console_drivers == console) {
2614		console_drivers=console->next;
2615		res = 0;
2616	} else if (console_drivers) {
2617		for (a=console_drivers->next, b=console_drivers ;
2618		     a; b=a, a=b->next) {
2619			if (a == console) {
2620				b->next = a->next;
2621				res = 0;
2622				break;
2623			}
2624		}
2625	}
2626
2627	if (!res && (console->flags & CON_EXTENDED))
2628		nr_ext_console_drivers--;
2629
2630	/*
2631	 * If this isn't the last console and it has CON_CONSDEV set, we
2632	 * need to set it on the next preferred console.
2633	 */
2634	if (console_drivers != NULL && console->flags & CON_CONSDEV)
2635		console_drivers->flags |= CON_CONSDEV;
2636
2637	console->flags &= ~CON_ENABLED;
2638	console_unlock();
2639	console_sysfs_notify();
2640	return res;
2641}
2642EXPORT_SYMBOL(unregister_console);
2643
2644/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2645 * Some boot consoles access data that is in the init section and which will
2646 * be discarded after the initcalls have been run. To make sure that no code
2647 * will access this data, unregister the boot consoles in a late initcall.
2648 *
2649 * If for some reason, such as deferred probe or the driver being a loadable
2650 * module, the real console hasn't registered yet at this point, there will
2651 * be a brief interval in which no messages are logged to the console, which
2652 * makes it difficult to diagnose problems that occur during this time.
2653 *
2654 * To mitigate this problem somewhat, only unregister consoles whose memory
2655 * intersects with the init section. Note that code exists elsewhere to get
2656 * rid of the boot console as soon as the proper console shows up, so there
2657 * won't be side-effects from postponing the removal.
2658 */
2659static int __init printk_late_init(void)
2660{
2661	struct console *con;
2662	int ret;
2663
2664	for_each_console(con) {
2665		if (!keep_bootcon && con->flags & CON_BOOT) {
 
 
 
 
 
 
 
 
 
2666			/*
2667			 * Make sure to unregister boot consoles whose data
2668			 * resides in the init section before the init section
2669			 * is discarded. Boot consoles whose data will stick
2670			 * around will automatically be unregistered when the
2671			 * proper console replaces them.
2672			 */
2673			if (init_section_intersects(con, sizeof(*con)))
2674				unregister_console(con);
 
2675		}
2676	}
2677	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2678					console_cpu_notify);
2679	WARN_ON(ret < 0);
2680	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2681					console_cpu_notify, NULL);
2682	WARN_ON(ret < 0);
2683	return 0;
2684}
2685late_initcall(printk_late_init);
2686
2687#if defined CONFIG_PRINTK
2688/*
2689 * Delayed printk version, for scheduler-internal messages:
2690 */
2691#define PRINTK_PENDING_WAKEUP	0x01
2692#define PRINTK_PENDING_OUTPUT	0x02
2693
2694static DEFINE_PER_CPU(int, printk_pending);
2695
2696static void wake_up_klogd_work_func(struct irq_work *irq_work)
2697{
2698	int pending = __this_cpu_xchg(printk_pending, 0);
2699
2700	if (pending & PRINTK_PENDING_OUTPUT) {
2701		/* If trylock fails, someone else is doing the printing */
2702		if (console_trylock())
2703			console_unlock();
2704	}
2705
2706	if (pending & PRINTK_PENDING_WAKEUP)
2707		wake_up_interruptible(&log_wait);
2708}
2709
2710static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2711	.func = wake_up_klogd_work_func,
2712	.flags = IRQ_WORK_LAZY,
2713};
2714
2715void wake_up_klogd(void)
2716{
2717	preempt_disable();
2718	if (waitqueue_active(&log_wait)) {
2719		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2720		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2721	}
2722	preempt_enable();
2723}
2724
2725int printk_deferred(const char *fmt, ...)
2726{
2727	va_list args;
2728	int r;
2729
2730	preempt_disable();
2731	va_start(args, fmt);
2732	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
2733	va_end(args);
2734
 
2735	__this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2736	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2737	preempt_enable();
2738
2739	return r;
2740}
2741
 
 
 
 
 
 
 
 
 
 
 
 
2742/*
2743 * printk rate limiting, lifted from the networking subsystem.
2744 *
2745 * This enforces a rate limit: not more than 10 kernel messages
2746 * every 5s to make a denial-of-service attack impossible.
2747 */
2748DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2749
2750int __printk_ratelimit(const char *func)
2751{
2752	return ___ratelimit(&printk_ratelimit_state, func);
2753}
2754EXPORT_SYMBOL(__printk_ratelimit);
2755
2756/**
2757 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2758 * @caller_jiffies: pointer to caller's state
2759 * @interval_msecs: minimum interval between prints
2760 *
2761 * printk_timed_ratelimit() returns true if more than @interval_msecs
2762 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2763 * returned true.
2764 */
2765bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2766			unsigned int interval_msecs)
2767{
2768	unsigned long elapsed = jiffies - *caller_jiffies;
2769
2770	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2771		return false;
2772
2773	*caller_jiffies = jiffies;
2774	return true;
2775}
2776EXPORT_SYMBOL(printk_timed_ratelimit);
2777
2778static DEFINE_SPINLOCK(dump_list_lock);
2779static LIST_HEAD(dump_list);
2780
2781/**
2782 * kmsg_dump_register - register a kernel log dumper.
2783 * @dumper: pointer to the kmsg_dumper structure
2784 *
2785 * Adds a kernel log dumper to the system. The dump callback in the
2786 * structure will be called when the kernel oopses or panics and must be
2787 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2788 */
2789int kmsg_dump_register(struct kmsg_dumper *dumper)
2790{
2791	unsigned long flags;
2792	int err = -EBUSY;
2793
2794	/* The dump callback needs to be set */
2795	if (!dumper->dump)
2796		return -EINVAL;
2797
2798	spin_lock_irqsave(&dump_list_lock, flags);
2799	/* Don't allow registering multiple times */
2800	if (!dumper->registered) {
2801		dumper->registered = 1;
2802		list_add_tail_rcu(&dumper->list, &dump_list);
2803		err = 0;
2804	}
2805	spin_unlock_irqrestore(&dump_list_lock, flags);
2806
2807	return err;
2808}
2809EXPORT_SYMBOL_GPL(kmsg_dump_register);
2810
2811/**
2812 * kmsg_dump_unregister - unregister a kmsg dumper.
2813 * @dumper: pointer to the kmsg_dumper structure
2814 *
2815 * Removes a dump device from the system. Returns zero on success and
2816 * %-EINVAL otherwise.
2817 */
2818int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2819{
2820	unsigned long flags;
2821	int err = -EINVAL;
2822
2823	spin_lock_irqsave(&dump_list_lock, flags);
2824	if (dumper->registered) {
2825		dumper->registered = 0;
2826		list_del_rcu(&dumper->list);
2827		err = 0;
2828	}
2829	spin_unlock_irqrestore(&dump_list_lock, flags);
2830	synchronize_rcu();
2831
2832	return err;
2833}
2834EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
2835
2836static bool always_kmsg_dump;
2837module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
2838
2839/**
2840 * kmsg_dump - dump kernel log to kernel message dumpers.
2841 * @reason: the reason (oops, panic etc) for dumping
2842 *
2843 * Call each of the registered dumper's dump() callback, which can
2844 * retrieve the kmsg records with kmsg_dump_get_line() or
2845 * kmsg_dump_get_buffer().
2846 */
2847void kmsg_dump(enum kmsg_dump_reason reason)
2848{
2849	struct kmsg_dumper *dumper;
2850	unsigned long flags;
2851
2852	if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
2853		return;
2854
2855	rcu_read_lock();
2856	list_for_each_entry_rcu(dumper, &dump_list, list) {
2857		if (dumper->max_reason && reason > dumper->max_reason)
2858			continue;
2859
2860		/* initialize iterator with data about the stored records */
2861		dumper->active = true;
2862
2863		raw_spin_lock_irqsave(&logbuf_lock, flags);
2864		dumper->cur_seq = clear_seq;
2865		dumper->cur_idx = clear_idx;
2866		dumper->next_seq = log_next_seq;
2867		dumper->next_idx = log_next_idx;
2868		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2869
2870		/* invoke dumper which will iterate over records */
2871		dumper->dump(dumper, reason);
2872
2873		/* reset iterator */
2874		dumper->active = false;
2875	}
2876	rcu_read_unlock();
2877}
2878
2879/**
2880 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
2881 * @dumper: registered kmsg dumper
2882 * @syslog: include the "<4>" prefixes
2883 * @line: buffer to copy the line to
2884 * @size: maximum size of the buffer
2885 * @len: length of line placed into buffer
2886 *
2887 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2888 * record, and copy one record into the provided buffer.
2889 *
2890 * Consecutive calls will return the next available record moving
2891 * towards the end of the buffer with the youngest messages.
2892 *
2893 * A return value of FALSE indicates that there are no more records to
2894 * read.
2895 *
2896 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
2897 */
2898bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
2899			       char *line, size_t size, size_t *len)
2900{
2901	struct printk_log *msg;
2902	size_t l = 0;
2903	bool ret = false;
2904
2905	if (!dumper->active)
2906		goto out;
2907
2908	if (dumper->cur_seq < log_first_seq) {
2909		/* messages are gone, move to first available one */
2910		dumper->cur_seq = log_first_seq;
2911		dumper->cur_idx = log_first_idx;
2912	}
2913
2914	/* last entry */
2915	if (dumper->cur_seq >= log_next_seq)
2916		goto out;
2917
2918	msg = log_from_idx(dumper->cur_idx);
2919	l = msg_print_text(msg, syslog, line, size);
2920
2921	dumper->cur_idx = log_next(dumper->cur_idx);
2922	dumper->cur_seq++;
2923	ret = true;
2924out:
2925	if (len)
2926		*len = l;
2927	return ret;
2928}
2929
2930/**
2931 * kmsg_dump_get_line - retrieve one kmsg log line
2932 * @dumper: registered kmsg dumper
2933 * @syslog: include the "<4>" prefixes
2934 * @line: buffer to copy the line to
2935 * @size: maximum size of the buffer
2936 * @len: length of line placed into buffer
2937 *
2938 * Start at the beginning of the kmsg buffer, with the oldest kmsg
2939 * record, and copy one record into the provided buffer.
2940 *
2941 * Consecutive calls will return the next available record moving
2942 * towards the end of the buffer with the youngest messages.
2943 *
2944 * A return value of FALSE indicates that there are no more records to
2945 * read.
2946 */
2947bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
2948			char *line, size_t size, size_t *len)
2949{
2950	unsigned long flags;
2951	bool ret;
2952
2953	raw_spin_lock_irqsave(&logbuf_lock, flags);
2954	ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
2955	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
2956
2957	return ret;
2958}
2959EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
2960
2961/**
2962 * kmsg_dump_get_buffer - copy kmsg log lines
2963 * @dumper: registered kmsg dumper
2964 * @syslog: include the "<4>" prefixes
2965 * @buf: buffer to copy the line to
2966 * @size: maximum size of the buffer
2967 * @len: length of line placed into buffer
2968 *
2969 * Start at the end of the kmsg buffer and fill the provided buffer
2970 * with as many of the the *youngest* kmsg records that fit into it.
2971 * If the buffer is large enough, all available kmsg records will be
2972 * copied with a single call.
2973 *
2974 * Consecutive calls will fill the buffer with the next block of
2975 * available older records, not including the earlier retrieved ones.
2976 *
2977 * A return value of FALSE indicates that there are no more records to
2978 * read.
2979 */
2980bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
2981			  char *buf, size_t size, size_t *len)
2982{
2983	unsigned long flags;
2984	u64 seq;
2985	u32 idx;
2986	u64 next_seq;
2987	u32 next_idx;
2988	size_t l = 0;
2989	bool ret = false;
2990
2991	if (!dumper->active)
2992		goto out;
2993
2994	raw_spin_lock_irqsave(&logbuf_lock, flags);
2995	if (dumper->cur_seq < log_first_seq) {
2996		/* messages are gone, move to first available one */
2997		dumper->cur_seq = log_first_seq;
2998		dumper->cur_idx = log_first_idx;
2999	}
3000
3001	/* last entry */
3002	if (dumper->cur_seq >= dumper->next_seq) {
3003		raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3004		goto out;
3005	}
3006
3007	/* calculate length of entire buffer */
3008	seq = dumper->cur_seq;
3009	idx = dumper->cur_idx;
3010	while (seq < dumper->next_seq) {
3011		struct printk_log *msg = log_from_idx(idx);
3012
3013		l += msg_print_text(msg, true, NULL, 0);
3014		idx = log_next(idx);
3015		seq++;
3016	}
3017
3018	/* move first record forward until length fits into the buffer */
3019	seq = dumper->cur_seq;
3020	idx = dumper->cur_idx;
3021	while (l > size && seq < dumper->next_seq) {
3022		struct printk_log *msg = log_from_idx(idx);
3023
3024		l -= msg_print_text(msg, true, NULL, 0);
3025		idx = log_next(idx);
3026		seq++;
3027	}
3028
3029	/* last message in next interation */
3030	next_seq = seq;
3031	next_idx = idx;
3032
3033	l = 0;
3034	while (seq < dumper->next_seq) {
3035		struct printk_log *msg = log_from_idx(idx);
3036
3037		l += msg_print_text(msg, syslog, buf + l, size - l);
3038		idx = log_next(idx);
3039		seq++;
3040	}
3041
3042	dumper->next_seq = next_seq;
3043	dumper->next_idx = next_idx;
3044	ret = true;
3045	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3046out:
3047	if (len)
3048		*len = l;
3049	return ret;
3050}
3051EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3052
3053/**
3054 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3055 * @dumper: registered kmsg dumper
3056 *
3057 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3058 * kmsg_dump_get_buffer() can be called again and used multiple
3059 * times within the same dumper.dump() callback.
3060 *
3061 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3062 */
3063void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3064{
3065	dumper->cur_seq = clear_seq;
3066	dumper->cur_idx = clear_idx;
3067	dumper->next_seq = log_next_seq;
3068	dumper->next_idx = log_next_idx;
3069}
3070
3071/**
3072 * kmsg_dump_rewind - reset the interator
3073 * @dumper: registered kmsg dumper
3074 *
3075 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3076 * kmsg_dump_get_buffer() can be called again and used multiple
3077 * times within the same dumper.dump() callback.
3078 */
3079void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3080{
3081	unsigned long flags;
3082
3083	raw_spin_lock_irqsave(&logbuf_lock, flags);
3084	kmsg_dump_rewind_nolock(dumper);
3085	raw_spin_unlock_irqrestore(&logbuf_lock, flags);
3086}
3087EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
3088
3089static char dump_stack_arch_desc_str[128];
3090
3091/**
3092 * dump_stack_set_arch_desc - set arch-specific str to show with task dumps
3093 * @fmt: printf-style format string
3094 * @...: arguments for the format string
3095 *
3096 * The configured string will be printed right after utsname during task
3097 * dumps.  Usually used to add arch-specific system identifiers.  If an
3098 * arch wants to make use of such an ID string, it should initialize this
3099 * as soon as possible during boot.
3100 */
3101void __init dump_stack_set_arch_desc(const char *fmt, ...)
3102{
3103	va_list args;
3104
3105	va_start(args, fmt);
3106	vsnprintf(dump_stack_arch_desc_str, sizeof(dump_stack_arch_desc_str),
3107		  fmt, args);
3108	va_end(args);
3109}
3110
3111/**
3112 * dump_stack_print_info - print generic debug info for dump_stack()
3113 * @log_lvl: log level
3114 *
3115 * Arch-specific dump_stack() implementations can use this function to
3116 * print out the same debug information as the generic dump_stack().
3117 */
3118void dump_stack_print_info(const char *log_lvl)
3119{
3120	printk("%sCPU: %d PID: %d Comm: %.20s %s %s %.*s\n",
3121	       log_lvl, raw_smp_processor_id(), current->pid, current->comm,
3122	       print_tainted(), init_utsname()->release,
3123	       (int)strcspn(init_utsname()->version, " "),
3124	       init_utsname()->version);
3125
3126	if (dump_stack_arch_desc_str[0] != '\0')
3127		printk("%sHardware name: %s\n",
3128		       log_lvl, dump_stack_arch_desc_str);
3129
3130	print_worker_info(log_lvl, current);
3131}
3132
3133/**
3134 * show_regs_print_info - print generic debug info for show_regs()
3135 * @log_lvl: log level
3136 *
3137 * show_regs() implementations can use this function to print out generic
3138 * debug information.
3139 */
3140void show_regs_print_info(const char *log_lvl)
3141{
3142	dump_stack_print_info(log_lvl);
3143
3144	printk("%stask: %p task.stack: %p\n",
3145	       log_lvl, current, task_stack_page(current));
3146}
3147
3148#endif
v4.17
   1/*
   2 *  linux/kernel/printk.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 * Modified to make sys_syslog() more flexible: added commands to
   7 * return the last 4k of kernel messages, regardless of whether
   8 * they've been read or not.  Added option to suppress kernel printk's
   9 * to the console.  Added hook for sending the console messages
  10 * elsewhere, in preparation for a serial line console (someday).
  11 * Ted Ts'o, 2/11/93.
  12 * Modified for sysctl support, 1/8/97, Chris Horn.
  13 * Fixed SMP synchronization, 08/08/99, Manfred Spraul
  14 *     manfred@colorfullife.com
  15 * Rewrote bits to get rid of console_lock
  16 *	01Mar01 Andrew Morton
  17 */
  18
  19#include <linux/kernel.h>
  20#include <linux/mm.h>
  21#include <linux/tty.h>
  22#include <linux/tty_driver.h>
  23#include <linux/console.h>
  24#include <linux/init.h>
  25#include <linux/jiffies.h>
  26#include <linux/nmi.h>
  27#include <linux/module.h>
  28#include <linux/moduleparam.h>
  29#include <linux/delay.h>
  30#include <linux/smp.h>
  31#include <linux/security.h>
  32#include <linux/bootmem.h>
  33#include <linux/memblock.h>
  34#include <linux/syscalls.h>
  35#include <linux/crash_core.h>
  36#include <linux/kdb.h>
  37#include <linux/ratelimit.h>
  38#include <linux/kmsg_dump.h>
  39#include <linux/syslog.h>
  40#include <linux/cpu.h>
  41#include <linux/notifier.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 "console_cmdline.h"
  59#include "braille.h"
  60#include "internal.h"
  61
  62int console_printk[4] = {
  63	CONSOLE_LOGLEVEL_DEFAULT,	/* console_loglevel */
  64	MESSAGE_LOGLEVEL_DEFAULT,	/* default_message_loglevel */
  65	CONSOLE_LOGLEVEL_MIN,		/* minimum_console_loglevel */
  66	CONSOLE_LOGLEVEL_DEFAULT,	/* default_console_loglevel */
  67};
  68
  69/*
  70 * Low level drivers may need that to know if they can schedule in
  71 * their unblank() callback or not. So let's export it.
  72 */
  73int oops_in_progress;
  74EXPORT_SYMBOL(oops_in_progress);
  75
  76/*
  77 * console_sem protects the console_drivers list, and also
  78 * provides serialisation for access to the entire console
  79 * driver system.
  80 */
  81static DEFINE_SEMAPHORE(console_sem);
  82struct console *console_drivers;
  83EXPORT_SYMBOL_GPL(console_drivers);
  84
  85#ifdef CONFIG_LOCKDEP
  86static struct lockdep_map console_lock_dep_map = {
  87	.name = "console_lock"
  88};
  89#endif
  90
  91enum devkmsg_log_bits {
  92	__DEVKMSG_LOG_BIT_ON = 0,
  93	__DEVKMSG_LOG_BIT_OFF,
  94	__DEVKMSG_LOG_BIT_LOCK,
  95};
  96
  97enum devkmsg_log_masks {
  98	DEVKMSG_LOG_MASK_ON             = BIT(__DEVKMSG_LOG_BIT_ON),
  99	DEVKMSG_LOG_MASK_OFF            = BIT(__DEVKMSG_LOG_BIT_OFF),
 100	DEVKMSG_LOG_MASK_LOCK           = BIT(__DEVKMSG_LOG_BIT_LOCK),
 101};
 102
 103/* Keep both the 'on' and 'off' bits clear, i.e. ratelimit by default: */
 104#define DEVKMSG_LOG_MASK_DEFAULT	0
 105
 106static unsigned int __read_mostly devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 107
 108static int __control_devkmsg(char *str)
 109{
 110	if (!str)
 111		return -EINVAL;
 112
 113	if (!strncmp(str, "on", 2)) {
 114		devkmsg_log = DEVKMSG_LOG_MASK_ON;
 115		return 2;
 116	} else if (!strncmp(str, "off", 3)) {
 117		devkmsg_log = DEVKMSG_LOG_MASK_OFF;
 118		return 3;
 119	} else if (!strncmp(str, "ratelimit", 9)) {
 120		devkmsg_log = DEVKMSG_LOG_MASK_DEFAULT;
 121		return 9;
 122	}
 123	return -EINVAL;
 124}
 125
 126static int __init control_devkmsg(char *str)
 127{
 128	if (__control_devkmsg(str) < 0)
 129		return 1;
 130
 131	/*
 132	 * Set sysctl string accordingly:
 133	 */
 134	if (devkmsg_log == DEVKMSG_LOG_MASK_ON)
 135		strcpy(devkmsg_log_str, "on");
 136	else if (devkmsg_log == DEVKMSG_LOG_MASK_OFF)
 137		strcpy(devkmsg_log_str, "off");
 
 
 
 138	/* else "ratelimit" which is set by default. */
 139
 140	/*
 141	 * Sysctl cannot change it anymore. The kernel command line setting of
 142	 * this parameter is to force the setting to be permanent throughout the
 143	 * runtime of the system. This is a precation measure against userspace
 144	 * trying to be a smarta** and attempting to change it up on us.
 145	 */
 146	devkmsg_log |= DEVKMSG_LOG_MASK_LOCK;
 147
 148	return 0;
 149}
 150__setup("printk.devkmsg=", control_devkmsg);
 151
 152char devkmsg_log_str[DEVKMSG_STR_MAX_SIZE] = "ratelimit";
 153
 154int devkmsg_sysctl_set_loglvl(struct ctl_table *table, int write,
 155			      void __user *buffer, size_t *lenp, loff_t *ppos)
 156{
 157	char old_str[DEVKMSG_STR_MAX_SIZE];
 158	unsigned int old;
 159	int err;
 160
 161	if (write) {
 162		if (devkmsg_log & DEVKMSG_LOG_MASK_LOCK)
 163			return -EINVAL;
 164
 165		old = devkmsg_log;
 166		strncpy(old_str, devkmsg_log_str, DEVKMSG_STR_MAX_SIZE);
 167	}
 168
 169	err = proc_dostring(table, write, buffer, lenp, ppos);
 170	if (err)
 171		return err;
 172
 173	if (write) {
 174		err = __control_devkmsg(devkmsg_log_str);
 175
 176		/*
 177		 * Do not accept an unknown string OR a known string with
 178		 * trailing crap...
 179		 */
 180		if (err < 0 || (err + 1 != *lenp)) {
 181
 182			/* ... and restore old setting. */
 183			devkmsg_log = old;
 184			strncpy(devkmsg_log_str, old_str, DEVKMSG_STR_MAX_SIZE);
 185
 186			return -EINVAL;
 187		}
 188	}
 189
 190	return 0;
 191}
 192
 193/*
 194 * Number of registered extended console drivers.
 195 *
 196 * If extended consoles are present, in-kernel cont reassembly is disabled
 197 * and each fragment is stored as a separate log entry with proper
 198 * continuation flag so that every emitted message has full metadata.  This
 199 * doesn't change the result for regular consoles or /proc/kmsg.  For
 200 * /dev/kmsg, as long as the reader concatenates messages according to
 201 * consecutive continuation flags, the end result should be the same too.
 202 */
 203static int nr_ext_console_drivers;
 204
 205/*
 206 * Helper macros to handle lockdep when locking/unlocking console_sem. We use
 207 * macros instead of functions so that _RET_IP_ contains useful information.
 208 */
 209#define down_console_sem() do { \
 210	down(&console_sem);\
 211	mutex_acquire(&console_lock_dep_map, 0, 0, _RET_IP_);\
 212} while (0)
 213
 214static int __down_trylock_console_sem(unsigned long ip)
 215{
 216	int lock_failed;
 217	unsigned long flags;
 218
 219	/*
 220	 * Here and in __up_console_sem() we need to be in safe mode,
 221	 * because spindump/WARN/etc from under console ->lock will
 222	 * deadlock in printk()->down_trylock_console_sem() otherwise.
 223	 */
 224	printk_safe_enter_irqsave(flags);
 225	lock_failed = down_trylock(&console_sem);
 226	printk_safe_exit_irqrestore(flags);
 227
 228	if (lock_failed)
 229		return 1;
 230	mutex_acquire(&console_lock_dep_map, 0, 1, ip);
 231	return 0;
 232}
 233#define down_trylock_console_sem() __down_trylock_console_sem(_RET_IP_)
 234
 235static void __up_console_sem(unsigned long ip)
 236{
 237	unsigned long flags;
 238
 239	mutex_release(&console_lock_dep_map, 1, ip);
 240
 241	printk_safe_enter_irqsave(flags);
 242	up(&console_sem);
 243	printk_safe_exit_irqrestore(flags);
 244}
 245#define up_console_sem() __up_console_sem(_RET_IP_)
 246
 247/*
 248 * This is used for debugging the mess that is the VT code by
 249 * keeping track if we have the console semaphore held. It's
 250 * definitely not the perfect debug tool (we don't know if _WE_
 251 * hold it and are racing, but it helps tracking those weird code
 252 * paths in the console code where we end up in places I want
 253 * locked without the console sempahore held).
 254 */
 255static int console_locked, console_suspended;
 256
 257/*
 258 * If exclusive_console is non-NULL then only this console is to be printed to.
 259 */
 260static struct console *exclusive_console;
 261
 262/*
 263 *	Array of consoles built from command line options (console=)
 264 */
 265
 266#define MAX_CMDLINECONSOLES 8
 267
 268static struct console_cmdline console_cmdline[MAX_CMDLINECONSOLES];
 269
 
 270static int preferred_console = -1;
 271int console_set_on_cmdline;
 272EXPORT_SYMBOL(console_set_on_cmdline);
 273
 274/* Flag: console code may call schedule() */
 275static int console_may_schedule;
 276
 277enum con_msg_format_flags {
 278	MSG_FORMAT_DEFAULT	= 0,
 279	MSG_FORMAT_SYSLOG	= (1 << 0),
 280};
 281
 282static int console_msg_format = MSG_FORMAT_DEFAULT;
 283
 284/*
 285 * The printk log buffer consists of a chain of concatenated variable
 286 * length records. Every record starts with a record header, containing
 287 * the overall length of the record.
 288 *
 289 * The heads to the first and last entry in the buffer, as well as the
 290 * sequence numbers of these entries are maintained when messages are
 291 * stored.
 292 *
 293 * If the heads indicate available messages, the length in the header
 294 * tells the start next message. A length == 0 for the next message
 295 * indicates a wrap-around to the beginning of the buffer.
 296 *
 297 * Every record carries the monotonic timestamp in microseconds, as well as
 298 * the standard userspace syslog level and syslog facility. The usual
 299 * kernel messages use LOG_KERN; userspace-injected messages always carry
 300 * a matching syslog facility, by default LOG_USER. The origin of every
 301 * message can be reliably determined that way.
 302 *
 303 * The human readable log message directly follows the message header. The
 304 * length of the message text is stored in the header, the stored message
 305 * is not terminated.
 306 *
 307 * Optionally, a message can carry a dictionary of properties (key/value pairs),
 308 * to provide userspace with a machine-readable message context.
 309 *
 310 * Examples for well-defined, commonly used property names are:
 311 *   DEVICE=b12:8               device identifier
 312 *                                b12:8         block dev_t
 313 *                                c127:3        char dev_t
 314 *                                n8            netdev ifindex
 315 *                                +sound:card0  subsystem:devname
 316 *   SUBSYSTEM=pci              driver-core subsystem name
 317 *
 318 * Valid characters in property names are [a-zA-Z0-9.-_]. The plain text value
 319 * follows directly after a '=' character. Every property is terminated by
 320 * a '\0' character. The last property is not terminated.
 321 *
 322 * Example of a message structure:
 323 *   0000  ff 8f 00 00 00 00 00 00      monotonic time in nsec
 324 *   0008  34 00                        record is 52 bytes long
 325 *   000a        0b 00                  text is 11 bytes long
 326 *   000c              1f 00            dictionary is 23 bytes long
 327 *   000e                    03 00      LOG_KERN (facility) LOG_ERR (level)
 328 *   0010  69 74 27 73 20 61 20 6c      "it's a l"
 329 *         69 6e 65                     "ine"
 330 *   001b           44 45 56 49 43      "DEVIC"
 331 *         45 3d 62 38 3a 32 00 44      "E=b8:2\0D"
 332 *         52 49 56 45 52 3d 62 75      "RIVER=bu"
 333 *         67                           "g"
 334 *   0032     00 00 00                  padding to next message header
 335 *
 336 * The 'struct printk_log' buffer header must never be directly exported to
 337 * userspace, it is a kernel-private implementation detail that might
 338 * need to be changed in the future, when the requirements change.
 339 *
 340 * /dev/kmsg exports the structured data in the following line format:
 341 *   "<level>,<sequnum>,<timestamp>,<contflag>[,additional_values, ... ];<message text>\n"
 342 *
 343 * Users of the export format should ignore possible additional values
 344 * separated by ',', and find the message after the ';' character.
 345 *
 346 * The optional key/value pairs are attached as continuation lines starting
 347 * with a space character and terminated by a newline. All possible
 348 * non-prinatable characters are escaped in the "\xff" notation.
 349 */
 350
 351enum log_flags {
 352	LOG_NOCONS	= 1,	/* already flushed, do not print to console */
 353	LOG_NEWLINE	= 2,	/* text ended with a newline */
 354	LOG_PREFIX	= 4,	/* text started with a prefix */
 355	LOG_CONT	= 8,	/* text is a fragment of a continuation line */
 356};
 357
 358struct printk_log {
 359	u64 ts_nsec;		/* timestamp in nanoseconds */
 360	u16 len;		/* length of entire record */
 361	u16 text_len;		/* length of text buffer */
 362	u16 dict_len;		/* length of dictionary buffer */
 363	u8 facility;		/* syslog facility */
 364	u8 flags:5;		/* internal record flags */
 365	u8 level:3;		/* syslog level */
 366}
 367#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
 368__packed __aligned(4)
 369#endif
 370;
 371
 372/*
 373 * The logbuf_lock protects kmsg buffer, indices, counters.  This can be taken
 374 * within the scheduler's rq lock. It must be released before calling
 375 * console_unlock() or anything else that might wake up a process.
 376 */
 377DEFINE_RAW_SPINLOCK(logbuf_lock);
 378
 379/*
 380 * Helper macros to lock/unlock logbuf_lock and switch between
 381 * printk-safe/unsafe modes.
 382 */
 383#define logbuf_lock_irq()				\
 384	do {						\
 385		printk_safe_enter_irq();		\
 386		raw_spin_lock(&logbuf_lock);		\
 387	} while (0)
 388
 389#define logbuf_unlock_irq()				\
 390	do {						\
 391		raw_spin_unlock(&logbuf_lock);		\
 392		printk_safe_exit_irq();			\
 393	} while (0)
 394
 395#define logbuf_lock_irqsave(flags)			\
 396	do {						\
 397		printk_safe_enter_irqsave(flags);	\
 398		raw_spin_lock(&logbuf_lock);		\
 399	} while (0)
 400
 401#define logbuf_unlock_irqrestore(flags)		\
 402	do {						\
 403		raw_spin_unlock(&logbuf_lock);		\
 404		printk_safe_exit_irqrestore(flags);	\
 405	} while (0)
 406
 407#ifdef CONFIG_PRINTK
 408DECLARE_WAIT_QUEUE_HEAD(log_wait);
 409/* the next printk record to read by syslog(READ) or /proc/kmsg */
 410static u64 syslog_seq;
 411static u32 syslog_idx;
 412static size_t syslog_partial;
 413
 414/* index and sequence number of the first record stored in the buffer */
 415static u64 log_first_seq;
 416static u32 log_first_idx;
 417
 418/* index and sequence number of the next record to store in the buffer */
 419static u64 log_next_seq;
 420static u32 log_next_idx;
 421
 422/* the next printk record to write to the console */
 423static u64 console_seq;
 424static u32 console_idx;
 425
 426/* the next printk record to read after the last 'clear' command */
 427static u64 clear_seq;
 428static u32 clear_idx;
 429
 430#define PREFIX_MAX		32
 431#define LOG_LINE_MAX		(1024 - PREFIX_MAX)
 432
 433#define LOG_LEVEL(v)		((v) & 0x07)
 434#define LOG_FACILITY(v)		((v) >> 3 & 0xff)
 435
 436/* record buffer */
 437#define LOG_ALIGN __alignof__(struct printk_log)
 438#define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
 439static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
 440static char *log_buf = __log_buf;
 441static u32 log_buf_len = __LOG_BUF_LEN;
 442
 443/* Return log buffer address */
 444char *log_buf_addr_get(void)
 445{
 446	return log_buf;
 447}
 448
 449/* Return log buffer size */
 450u32 log_buf_len_get(void)
 451{
 452	return log_buf_len;
 453}
 454
 455/* human readable text of the record */
 456static char *log_text(const struct printk_log *msg)
 457{
 458	return (char *)msg + sizeof(struct printk_log);
 459}
 460
 461/* optional key/value pair dictionary attached to the record */
 462static char *log_dict(const struct printk_log *msg)
 463{
 464	return (char *)msg + sizeof(struct printk_log) + msg->text_len;
 465}
 466
 467/* get record by index; idx must point to valid msg */
 468static struct printk_log *log_from_idx(u32 idx)
 469{
 470	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
 471
 472	/*
 473	 * A length == 0 record is the end of buffer marker. Wrap around and
 474	 * read the message at the start of the buffer.
 475	 */
 476	if (!msg->len)
 477		return (struct printk_log *)log_buf;
 478	return msg;
 479}
 480
 481/* get next record; idx must point to valid msg */
 482static u32 log_next(u32 idx)
 483{
 484	struct printk_log *msg = (struct printk_log *)(log_buf + idx);
 485
 486	/* length == 0 indicates the end of the buffer; wrap */
 487	/*
 488	 * A length == 0 record is the end of buffer marker. Wrap around and
 489	 * read the message at the start of the buffer as *this* one, and
 490	 * return the one after that.
 491	 */
 492	if (!msg->len) {
 493		msg = (struct printk_log *)log_buf;
 494		return msg->len;
 495	}
 496	return idx + msg->len;
 497}
 498
 499/*
 500 * Check whether there is enough free space for the given message.
 501 *
 502 * The same values of first_idx and next_idx mean that the buffer
 503 * is either empty or full.
 504 *
 505 * If the buffer is empty, we must respect the position of the indexes.
 506 * They cannot be reset to the beginning of the buffer.
 507 */
 508static int logbuf_has_space(u32 msg_size, bool empty)
 509{
 510	u32 free;
 511
 512	if (log_next_idx > log_first_idx || empty)
 513		free = max(log_buf_len - log_next_idx, log_first_idx);
 514	else
 515		free = log_first_idx - log_next_idx;
 516
 517	/*
 518	 * We need space also for an empty header that signalizes wrapping
 519	 * of the buffer.
 520	 */
 521	return free >= msg_size + sizeof(struct printk_log);
 522}
 523
 524static int log_make_free_space(u32 msg_size)
 525{
 526	while (log_first_seq < log_next_seq &&
 527	       !logbuf_has_space(msg_size, false)) {
 528		/* drop old messages until we have enough contiguous space */
 529		log_first_idx = log_next(log_first_idx);
 530		log_first_seq++;
 531	}
 532
 533	if (clear_seq < log_first_seq) {
 534		clear_seq = log_first_seq;
 535		clear_idx = log_first_idx;
 536	}
 537
 538	/* sequence numbers are equal, so the log buffer is empty */
 539	if (logbuf_has_space(msg_size, log_first_seq == log_next_seq))
 540		return 0;
 541
 542	return -ENOMEM;
 543}
 544
 545/* compute the message size including the padding bytes */
 546static u32 msg_used_size(u16 text_len, u16 dict_len, u32 *pad_len)
 547{
 548	u32 size;
 549
 550	size = sizeof(struct printk_log) + text_len + dict_len;
 551	*pad_len = (-size) & (LOG_ALIGN - 1);
 552	size += *pad_len;
 553
 554	return size;
 555}
 556
 557/*
 558 * Define how much of the log buffer we could take at maximum. The value
 559 * must be greater than two. Note that only half of the buffer is available
 560 * when the index points to the middle.
 561 */
 562#define MAX_LOG_TAKE_PART 4
 563static const char trunc_msg[] = "<truncated>";
 564
 565static u32 truncate_msg(u16 *text_len, u16 *trunc_msg_len,
 566			u16 *dict_len, u32 *pad_len)
 567{
 568	/*
 569	 * The message should not take the whole buffer. Otherwise, it might
 570	 * get removed too soon.
 571	 */
 572	u32 max_text_len = log_buf_len / MAX_LOG_TAKE_PART;
 573	if (*text_len > max_text_len)
 574		*text_len = max_text_len;
 575	/* enable the warning message */
 576	*trunc_msg_len = strlen(trunc_msg);
 577	/* disable the "dict" completely */
 578	*dict_len = 0;
 579	/* compute the size again, count also the warning message */
 580	return msg_used_size(*text_len + *trunc_msg_len, 0, pad_len);
 581}
 582
 583/* insert record into the buffer, discard old ones, update heads */
 584static int log_store(int facility, int level,
 585		     enum log_flags flags, u64 ts_nsec,
 586		     const char *dict, u16 dict_len,
 587		     const char *text, u16 text_len)
 588{
 589	struct printk_log *msg;
 590	u32 size, pad_len;
 591	u16 trunc_msg_len = 0;
 592
 593	/* number of '\0' padding bytes to next message */
 594	size = msg_used_size(text_len, dict_len, &pad_len);
 595
 596	if (log_make_free_space(size)) {
 597		/* truncate the message if it is too long for empty buffer */
 598		size = truncate_msg(&text_len, &trunc_msg_len,
 599				    &dict_len, &pad_len);
 600		/* survive when the log buffer is too small for trunc_msg */
 601		if (log_make_free_space(size))
 602			return 0;
 603	}
 604
 605	if (log_next_idx + size + sizeof(struct printk_log) > log_buf_len) {
 606		/*
 607		 * This message + an additional empty header does not fit
 608		 * at the end of the buffer. Add an empty header with len == 0
 609		 * to signify a wrap around.
 610		 */
 611		memset(log_buf + log_next_idx, 0, sizeof(struct printk_log));
 612		log_next_idx = 0;
 613	}
 614
 615	/* fill message */
 616	msg = (struct printk_log *)(log_buf + log_next_idx);
 617	memcpy(log_text(msg), text, text_len);
 618	msg->text_len = text_len;
 619	if (trunc_msg_len) {
 620		memcpy(log_text(msg) + text_len, trunc_msg, trunc_msg_len);
 621		msg->text_len += trunc_msg_len;
 622	}
 623	memcpy(log_dict(msg), dict, dict_len);
 624	msg->dict_len = dict_len;
 625	msg->facility = facility;
 626	msg->level = level & 7;
 627	msg->flags = flags & 0x1f;
 628	if (ts_nsec > 0)
 629		msg->ts_nsec = ts_nsec;
 630	else
 631		msg->ts_nsec = local_clock();
 632	memset(log_dict(msg) + dict_len, 0, pad_len);
 633	msg->len = size;
 634
 635	/* insert message */
 636	log_next_idx += msg->len;
 637	log_next_seq++;
 638
 639	return msg->text_len;
 640}
 641
 642int dmesg_restrict = IS_ENABLED(CONFIG_SECURITY_DMESG_RESTRICT);
 643
 644static int syslog_action_restricted(int type)
 645{
 646	if (dmesg_restrict)
 647		return 1;
 648	/*
 649	 * Unless restricted, we allow "read all" and "get buffer size"
 650	 * for everybody.
 651	 */
 652	return type != SYSLOG_ACTION_READ_ALL &&
 653	       type != SYSLOG_ACTION_SIZE_BUFFER;
 654}
 655
 656static int check_syslog_permissions(int type, int source)
 657{
 658	/*
 659	 * If this is from /proc/kmsg and we've already opened it, then we've
 660	 * already done the capabilities checks at open time.
 661	 */
 662	if (source == SYSLOG_FROM_PROC && type != SYSLOG_ACTION_OPEN)
 663		goto ok;
 664
 665	if (syslog_action_restricted(type)) {
 666		if (capable(CAP_SYSLOG))
 667			goto ok;
 668		/*
 669		 * For historical reasons, accept CAP_SYS_ADMIN too, with
 670		 * a warning.
 671		 */
 672		if (capable(CAP_SYS_ADMIN)) {
 673			pr_warn_once("%s (%d): Attempt to access syslog with "
 674				     "CAP_SYS_ADMIN but no CAP_SYSLOG "
 675				     "(deprecated).\n",
 676				 current->comm, task_pid_nr(current));
 677			goto ok;
 678		}
 679		return -EPERM;
 680	}
 681ok:
 682	return security_syslog(type);
 683}
 
 684
 685static void append_char(char **pp, char *e, char c)
 686{
 687	if (*pp < e)
 688		*(*pp)++ = c;
 689}
 690
 691static ssize_t msg_print_ext_header(char *buf, size_t size,
 692				    struct printk_log *msg, u64 seq)
 693{
 694	u64 ts_usec = msg->ts_nsec;
 695
 696	do_div(ts_usec, 1000);
 697
 698	return scnprintf(buf, size, "%u,%llu,%llu,%c;",
 699		       (msg->facility << 3) | msg->level, seq, ts_usec,
 700		       msg->flags & LOG_CONT ? 'c' : '-');
 701}
 702
 703static ssize_t msg_print_ext_body(char *buf, size_t size,
 704				  char *dict, size_t dict_len,
 705				  char *text, size_t text_len)
 706{
 707	char *p = buf, *e = buf + size;
 708	size_t i;
 709
 710	/* escape non-printable characters */
 711	for (i = 0; i < text_len; i++) {
 712		unsigned char c = text[i];
 713
 714		if (c < ' ' || c >= 127 || c == '\\')
 715			p += scnprintf(p, e - p, "\\x%02x", c);
 716		else
 717			append_char(&p, e, c);
 718	}
 719	append_char(&p, e, '\n');
 720
 721	if (dict_len) {
 722		bool line = true;
 723
 724		for (i = 0; i < dict_len; i++) {
 725			unsigned char c = dict[i];
 726
 727			if (line) {
 728				append_char(&p, e, ' ');
 729				line = false;
 730			}
 731
 732			if (c == '\0') {
 733				append_char(&p, e, '\n');
 734				line = true;
 735				continue;
 736			}
 737
 738			if (c < ' ' || c >= 127 || c == '\\') {
 739				p += scnprintf(p, e - p, "\\x%02x", c);
 740				continue;
 741			}
 742
 743			append_char(&p, e, c);
 744		}
 745		append_char(&p, e, '\n');
 746	}
 747
 748	return p - buf;
 749}
 750
 751/* /dev/kmsg - userspace message inject/listen interface */
 752struct devkmsg_user {
 753	u64 seq;
 754	u32 idx;
 755	struct ratelimit_state rs;
 756	struct mutex lock;
 757	char buf[CONSOLE_EXT_LOG_MAX];
 758};
 759
 760static ssize_t devkmsg_write(struct kiocb *iocb, struct iov_iter *from)
 761{
 762	char *buf, *line;
 763	int level = default_message_loglevel;
 764	int facility = 1;	/* LOG_USER */
 765	struct file *file = iocb->ki_filp;
 766	struct devkmsg_user *user = file->private_data;
 767	size_t len = iov_iter_count(from);
 768	ssize_t ret = len;
 769
 770	if (!user || len > LOG_LINE_MAX)
 771		return -EINVAL;
 772
 773	/* Ignore when user logging is disabled. */
 774	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 775		return len;
 776
 777	/* Ratelimit when not explicitly enabled. */
 778	if (!(devkmsg_log & DEVKMSG_LOG_MASK_ON)) {
 779		if (!___ratelimit(&user->rs, current->comm))
 780			return ret;
 781	}
 782
 783	buf = kmalloc(len+1, GFP_KERNEL);
 784	if (buf == NULL)
 785		return -ENOMEM;
 786
 787	buf[len] = '\0';
 788	if (!copy_from_iter_full(buf, len, from)) {
 789		kfree(buf);
 790		return -EFAULT;
 791	}
 792
 793	/*
 794	 * Extract and skip the syslog prefix <[0-9]*>. Coming from userspace
 795	 * the decimal value represents 32bit, the lower 3 bit are the log
 796	 * level, the rest are the log facility.
 797	 *
 798	 * If no prefix or no userspace facility is specified, we
 799	 * enforce LOG_USER, to be able to reliably distinguish
 800	 * kernel-generated messages from userspace-injected ones.
 801	 */
 802	line = buf;
 803	if (line[0] == '<') {
 804		char *endp = NULL;
 805		unsigned int u;
 806
 807		u = simple_strtoul(line + 1, &endp, 10);
 808		if (endp && endp[0] == '>') {
 809			level = LOG_LEVEL(u);
 810			if (LOG_FACILITY(u) != 0)
 811				facility = LOG_FACILITY(u);
 812			endp++;
 813			len -= endp - line;
 814			line = endp;
 815		}
 816	}
 817
 818	printk_emit(facility, level, NULL, 0, "%s", line);
 819	kfree(buf);
 820	return ret;
 821}
 822
 823static ssize_t devkmsg_read(struct file *file, char __user *buf,
 824			    size_t count, loff_t *ppos)
 825{
 826	struct devkmsg_user *user = file->private_data;
 827	struct printk_log *msg;
 828	size_t len;
 829	ssize_t ret;
 830
 831	if (!user)
 832		return -EBADF;
 833
 834	ret = mutex_lock_interruptible(&user->lock);
 835	if (ret)
 836		return ret;
 837
 838	logbuf_lock_irq();
 839	while (user->seq == log_next_seq) {
 840		if (file->f_flags & O_NONBLOCK) {
 841			ret = -EAGAIN;
 842			logbuf_unlock_irq();
 843			goto out;
 844		}
 845
 846		logbuf_unlock_irq();
 847		ret = wait_event_interruptible(log_wait,
 848					       user->seq != log_next_seq);
 849		if (ret)
 850			goto out;
 851		logbuf_lock_irq();
 852	}
 853
 854	if (user->seq < log_first_seq) {
 855		/* our last seen message is gone, return error and reset */
 856		user->idx = log_first_idx;
 857		user->seq = log_first_seq;
 858		ret = -EPIPE;
 859		logbuf_unlock_irq();
 860		goto out;
 861	}
 862
 863	msg = log_from_idx(user->idx);
 864	len = msg_print_ext_header(user->buf, sizeof(user->buf),
 865				   msg, user->seq);
 866	len += msg_print_ext_body(user->buf + len, sizeof(user->buf) - len,
 867				  log_dict(msg), msg->dict_len,
 868				  log_text(msg), msg->text_len);
 869
 870	user->idx = log_next(user->idx);
 871	user->seq++;
 872	logbuf_unlock_irq();
 873
 874	if (len > count) {
 875		ret = -EINVAL;
 876		goto out;
 877	}
 878
 879	if (copy_to_user(buf, user->buf, len)) {
 880		ret = -EFAULT;
 881		goto out;
 882	}
 883	ret = len;
 884out:
 885	mutex_unlock(&user->lock);
 886	return ret;
 887}
 888
 889static loff_t devkmsg_llseek(struct file *file, loff_t offset, int whence)
 890{
 891	struct devkmsg_user *user = file->private_data;
 892	loff_t ret = 0;
 893
 894	if (!user)
 895		return -EBADF;
 896	if (offset)
 897		return -ESPIPE;
 898
 899	logbuf_lock_irq();
 900	switch (whence) {
 901	case SEEK_SET:
 902		/* the first record */
 903		user->idx = log_first_idx;
 904		user->seq = log_first_seq;
 905		break;
 906	case SEEK_DATA:
 907		/*
 908		 * The first record after the last SYSLOG_ACTION_CLEAR,
 909		 * like issued by 'dmesg -c'. Reading /dev/kmsg itself
 910		 * changes no global state, and does not clear anything.
 911		 */
 912		user->idx = clear_idx;
 913		user->seq = clear_seq;
 914		break;
 915	case SEEK_END:
 916		/* after the last record */
 917		user->idx = log_next_idx;
 918		user->seq = log_next_seq;
 919		break;
 920	default:
 921		ret = -EINVAL;
 922	}
 923	logbuf_unlock_irq();
 924	return ret;
 925}
 926
 927static __poll_t devkmsg_poll(struct file *file, poll_table *wait)
 928{
 929	struct devkmsg_user *user = file->private_data;
 930	__poll_t ret = 0;
 931
 932	if (!user)
 933		return EPOLLERR|EPOLLNVAL;
 934
 935	poll_wait(file, &log_wait, wait);
 936
 937	logbuf_lock_irq();
 938	if (user->seq < log_next_seq) {
 939		/* return error when data has vanished underneath us */
 940		if (user->seq < log_first_seq)
 941			ret = EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
 942		else
 943			ret = EPOLLIN|EPOLLRDNORM;
 944	}
 945	logbuf_unlock_irq();
 946
 947	return ret;
 948}
 949
 950static int devkmsg_open(struct inode *inode, struct file *file)
 951{
 952	struct devkmsg_user *user;
 953	int err;
 954
 955	if (devkmsg_log & DEVKMSG_LOG_MASK_OFF)
 956		return -EPERM;
 957
 958	/* write-only does not need any file context */
 959	if ((file->f_flags & O_ACCMODE) != O_WRONLY) {
 960		err = check_syslog_permissions(SYSLOG_ACTION_READ_ALL,
 961					       SYSLOG_FROM_READER);
 962		if (err)
 963			return err;
 964	}
 965
 966	user = kmalloc(sizeof(struct devkmsg_user), GFP_KERNEL);
 967	if (!user)
 968		return -ENOMEM;
 969
 970	ratelimit_default_init(&user->rs);
 971	ratelimit_set_flags(&user->rs, RATELIMIT_MSG_ON_RELEASE);
 972
 973	mutex_init(&user->lock);
 974
 975	logbuf_lock_irq();
 976	user->idx = log_first_idx;
 977	user->seq = log_first_seq;
 978	logbuf_unlock_irq();
 979
 980	file->private_data = user;
 981	return 0;
 982}
 983
 984static int devkmsg_release(struct inode *inode, struct file *file)
 985{
 986	struct devkmsg_user *user = file->private_data;
 987
 988	if (!user)
 989		return 0;
 990
 991	ratelimit_state_exit(&user->rs);
 992
 993	mutex_destroy(&user->lock);
 994	kfree(user);
 995	return 0;
 996}
 997
 998const struct file_operations kmsg_fops = {
 999	.open = devkmsg_open,
1000	.read = devkmsg_read,
1001	.write_iter = devkmsg_write,
1002	.llseek = devkmsg_llseek,
1003	.poll = devkmsg_poll,
1004	.release = devkmsg_release,
1005};
1006
1007#ifdef CONFIG_CRASH_CORE
1008/*
1009 * This appends the listed symbols to /proc/vmcore
1010 *
1011 * /proc/vmcore is used by various utilities, like crash and makedumpfile to
1012 * obtain access to symbols that are otherwise very difficult to locate.  These
1013 * symbols are specifically used so that utilities can access and extract the
1014 * dmesg log from a vmcore file after a crash.
1015 */
1016void log_buf_vmcoreinfo_setup(void)
1017{
1018	VMCOREINFO_SYMBOL(log_buf);
1019	VMCOREINFO_SYMBOL(log_buf_len);
1020	VMCOREINFO_SYMBOL(log_first_idx);
1021	VMCOREINFO_SYMBOL(clear_idx);
1022	VMCOREINFO_SYMBOL(log_next_idx);
1023	/*
1024	 * Export struct printk_log size and field offsets. User space tools can
1025	 * parse it and detect any changes to structure down the line.
1026	 */
1027	VMCOREINFO_STRUCT_SIZE(printk_log);
1028	VMCOREINFO_OFFSET(printk_log, ts_nsec);
1029	VMCOREINFO_OFFSET(printk_log, len);
1030	VMCOREINFO_OFFSET(printk_log, text_len);
1031	VMCOREINFO_OFFSET(printk_log, dict_len);
1032}
1033#endif
1034
1035/* requested log_buf_len from kernel cmdline */
1036static unsigned long __initdata new_log_buf_len;
1037
1038/* we practice scaling the ring buffer by powers of 2 */
1039static void __init log_buf_len_update(unsigned size)
1040{
1041	if (size)
1042		size = roundup_pow_of_two(size);
1043	if (size > log_buf_len)
1044		new_log_buf_len = size;
1045}
1046
1047/* save requested log_buf_len since it's too early to process it */
1048static int __init log_buf_len_setup(char *str)
1049{
1050	unsigned size = memparse(str, &str);
1051
1052	log_buf_len_update(size);
1053
1054	return 0;
1055}
1056early_param("log_buf_len", log_buf_len_setup);
1057
1058#ifdef CONFIG_SMP
1059#define __LOG_CPU_MAX_BUF_LEN (1 << CONFIG_LOG_CPU_MAX_BUF_SHIFT)
1060
1061static void __init log_buf_add_cpu(void)
1062{
1063	unsigned int cpu_extra;
1064
1065	/*
1066	 * archs should set up cpu_possible_bits properly with
1067	 * set_cpu_possible() after setup_arch() but just in
1068	 * case lets ensure this is valid.
1069	 */
1070	if (num_possible_cpus() == 1)
1071		return;
1072
1073	cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MAX_BUF_LEN;
1074
1075	/* by default this will only continue through for large > 64 CPUs */
1076	if (cpu_extra <= __LOG_BUF_LEN / 2)
1077		return;
1078
1079	pr_info("log_buf_len individual max cpu contribution: %d bytes\n",
1080		__LOG_CPU_MAX_BUF_LEN);
1081	pr_info("log_buf_len total cpu_extra contributions: %d bytes\n",
1082		cpu_extra);
1083	pr_info("log_buf_len min size: %d bytes\n", __LOG_BUF_LEN);
1084
1085	log_buf_len_update(cpu_extra + __LOG_BUF_LEN);
1086}
1087#else /* !CONFIG_SMP */
1088static inline void log_buf_add_cpu(void) {}
1089#endif /* CONFIG_SMP */
1090
1091void __init setup_log_buf(int early)
1092{
1093	unsigned long flags;
1094	char *new_log_buf;
1095	int free;
1096
1097	if (log_buf != __log_buf)
1098		return;
1099
1100	if (!early && !new_log_buf_len)
1101		log_buf_add_cpu();
1102
1103	if (!new_log_buf_len)
1104		return;
1105
1106	if (early) {
1107		new_log_buf =
1108			memblock_virt_alloc(new_log_buf_len, LOG_ALIGN);
1109	} else {
1110		new_log_buf = memblock_virt_alloc_nopanic(new_log_buf_len,
1111							  LOG_ALIGN);
1112	}
1113
1114	if (unlikely(!new_log_buf)) {
1115		pr_err("log_buf_len: %ld bytes not available\n",
1116			new_log_buf_len);
1117		return;
1118	}
1119
1120	logbuf_lock_irqsave(flags);
1121	log_buf_len = new_log_buf_len;
1122	log_buf = new_log_buf;
1123	new_log_buf_len = 0;
1124	free = __LOG_BUF_LEN - log_next_idx;
1125	memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
1126	logbuf_unlock_irqrestore(flags);
1127
1128	pr_info("log_buf_len: %d bytes\n", log_buf_len);
1129	pr_info("early log buf free: %d(%d%%)\n",
1130		free, (free * 100) / __LOG_BUF_LEN);
1131}
1132
1133static bool __read_mostly ignore_loglevel;
1134
1135static int __init ignore_loglevel_setup(char *str)
1136{
1137	ignore_loglevel = true;
1138	pr_info("debug: ignoring loglevel setting.\n");
1139
1140	return 0;
1141}
1142
1143early_param("ignore_loglevel", ignore_loglevel_setup);
1144module_param(ignore_loglevel, bool, S_IRUGO | S_IWUSR);
1145MODULE_PARM_DESC(ignore_loglevel,
1146		 "ignore loglevel setting (prints all kernel messages to the console)");
1147
1148static bool suppress_message_printing(int level)
1149{
1150	return (level >= console_loglevel && !ignore_loglevel);
1151}
1152
1153#ifdef CONFIG_BOOT_PRINTK_DELAY
1154
1155static int boot_delay; /* msecs delay after each printk during bootup */
1156static unsigned long long loops_per_msec;	/* based on boot_delay */
1157
1158static int __init boot_delay_setup(char *str)
1159{
1160	unsigned long lpj;
1161
1162	lpj = preset_lpj ? preset_lpj : 1000000;	/* some guess */
1163	loops_per_msec = (unsigned long long)lpj / 1000 * HZ;
1164
1165	get_option(&str, &boot_delay);
1166	if (boot_delay > 10 * 1000)
1167		boot_delay = 0;
1168
1169	pr_debug("boot_delay: %u, preset_lpj: %ld, lpj: %lu, "
1170		"HZ: %d, loops_per_msec: %llu\n",
1171		boot_delay, preset_lpj, lpj, HZ, loops_per_msec);
1172	return 0;
1173}
1174early_param("boot_delay", boot_delay_setup);
1175
1176static void boot_delay_msec(int level)
1177{
1178	unsigned long long k;
1179	unsigned long timeout;
1180
1181	if ((boot_delay == 0 || system_state >= SYSTEM_RUNNING)
1182		|| suppress_message_printing(level)) {
1183		return;
1184	}
1185
1186	k = (unsigned long long)loops_per_msec * boot_delay;
1187
1188	timeout = jiffies + msecs_to_jiffies(boot_delay);
1189	while (k) {
1190		k--;
1191		cpu_relax();
1192		/*
1193		 * use (volatile) jiffies to prevent
1194		 * compiler reduction; loop termination via jiffies
1195		 * is secondary and may or may not happen.
1196		 */
1197		if (time_after(jiffies, timeout))
1198			break;
1199		touch_nmi_watchdog();
1200	}
1201}
1202#else
1203static inline void boot_delay_msec(int level)
1204{
1205}
1206#endif
1207
1208static bool printk_time = IS_ENABLED(CONFIG_PRINTK_TIME);
1209module_param_named(time, printk_time, bool, S_IRUGO | S_IWUSR);
1210
1211static size_t print_time(u64 ts, char *buf)
1212{
1213	unsigned long rem_nsec;
1214
1215	if (!printk_time)
1216		return 0;
1217
1218	rem_nsec = do_div(ts, 1000000000);
1219
1220	if (!buf)
1221		return snprintf(NULL, 0, "[%5lu.000000] ", (unsigned long)ts);
1222
1223	return sprintf(buf, "[%5lu.%06lu] ",
1224		       (unsigned long)ts, rem_nsec / 1000);
1225}
1226
1227static size_t print_prefix(const struct printk_log *msg, bool syslog, char *buf)
1228{
1229	size_t len = 0;
1230	unsigned int prefix = (msg->facility << 3) | msg->level;
1231
1232	if (syslog) {
1233		if (buf) {
1234			len += sprintf(buf, "<%u>", prefix);
1235		} else {
1236			len += 3;
1237			if (prefix > 999)
1238				len += 3;
1239			else if (prefix > 99)
1240				len += 2;
1241			else if (prefix > 9)
1242				len++;
1243		}
1244	}
1245
1246	len += print_time(msg->ts_nsec, buf ? buf + len : NULL);
1247	return len;
1248}
1249
1250static size_t msg_print_text(const struct printk_log *msg, bool syslog, char *buf, size_t size)
1251{
1252	const char *text = log_text(msg);
1253	size_t text_size = msg->text_len;
1254	size_t len = 0;
1255
1256	do {
1257		const char *next = memchr(text, '\n', text_size);
1258		size_t text_len;
1259
1260		if (next) {
1261			text_len = next - text;
1262			next++;
1263			text_size -= next - text;
1264		} else {
1265			text_len = text_size;
1266		}
1267
1268		if (buf) {
1269			if (print_prefix(msg, syslog, NULL) +
1270			    text_len + 1 >= size - len)
1271				break;
1272
1273			len += print_prefix(msg, syslog, buf + len);
1274			memcpy(buf + len, text, text_len);
1275			len += text_len;
1276			buf[len++] = '\n';
1277		} else {
1278			/* SYSLOG_ACTION_* buffer size only calculation */
1279			len += print_prefix(msg, syslog, NULL);
1280			len += text_len;
1281			len++;
1282		}
1283
1284		text = next;
1285	} while (text);
1286
1287	return len;
1288}
1289
1290static int syslog_print(char __user *buf, int size)
1291{
1292	char *text;
1293	struct printk_log *msg;
1294	int len = 0;
1295
1296	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1297	if (!text)
1298		return -ENOMEM;
1299
1300	while (size > 0) {
1301		size_t n;
1302		size_t skip;
1303
1304		logbuf_lock_irq();
1305		if (syslog_seq < log_first_seq) {
1306			/* messages are gone, move to first one */
1307			syslog_seq = log_first_seq;
1308			syslog_idx = log_first_idx;
1309			syslog_partial = 0;
1310		}
1311		if (syslog_seq == log_next_seq) {
1312			logbuf_unlock_irq();
1313			break;
1314		}
1315
1316		skip = syslog_partial;
1317		msg = log_from_idx(syslog_idx);
1318		n = msg_print_text(msg, true, text, LOG_LINE_MAX + PREFIX_MAX);
1319		if (n - syslog_partial <= size) {
1320			/* message fits into buffer, move forward */
1321			syslog_idx = log_next(syslog_idx);
1322			syslog_seq++;
1323			n -= syslog_partial;
1324			syslog_partial = 0;
1325		} else if (!len){
1326			/* partial read(), remember position */
1327			n = size;
1328			syslog_partial += n;
1329		} else
1330			n = 0;
1331		logbuf_unlock_irq();
1332
1333		if (!n)
1334			break;
1335
1336		if (copy_to_user(buf, text + skip, n)) {
1337			if (!len)
1338				len = -EFAULT;
1339			break;
1340		}
1341
1342		len += n;
1343		size -= n;
1344		buf += n;
1345	}
1346
1347	kfree(text);
1348	return len;
1349}
1350
1351static int syslog_print_all(char __user *buf, int size, bool clear)
1352{
1353	char *text;
1354	int len = 0;
1355
1356	text = kmalloc(LOG_LINE_MAX + PREFIX_MAX, GFP_KERNEL);
1357	if (!text)
1358		return -ENOMEM;
1359
1360	logbuf_lock_irq();
1361	if (buf) {
1362		u64 next_seq;
1363		u64 seq;
1364		u32 idx;
1365
1366		/*
1367		 * Find first record that fits, including all following records,
1368		 * into the user-provided buffer for this dump.
1369		 */
1370		seq = clear_seq;
1371		idx = clear_idx;
1372		while (seq < log_next_seq) {
1373			struct printk_log *msg = log_from_idx(idx);
1374
1375			len += msg_print_text(msg, true, NULL, 0);
1376			idx = log_next(idx);
1377			seq++;
1378		}
1379
1380		/* move first record forward until length fits into the buffer */
1381		seq = clear_seq;
1382		idx = clear_idx;
1383		while (len > size && seq < log_next_seq) {
1384			struct printk_log *msg = log_from_idx(idx);
1385
1386			len -= msg_print_text(msg, true, NULL, 0);
1387			idx = log_next(idx);
1388			seq++;
1389		}
1390
1391		/* last message fitting into this dump */
1392		next_seq = log_next_seq;
1393
1394		len = 0;
1395		while (len >= 0 && seq < next_seq) {
1396			struct printk_log *msg = log_from_idx(idx);
1397			int textlen;
1398
1399			textlen = msg_print_text(msg, true, text,
1400						 LOG_LINE_MAX + PREFIX_MAX);
1401			if (textlen < 0) {
1402				len = textlen;
1403				break;
1404			}
1405			idx = log_next(idx);
1406			seq++;
1407
1408			logbuf_unlock_irq();
1409			if (copy_to_user(buf + len, text, textlen))
1410				len = -EFAULT;
1411			else
1412				len += textlen;
1413			logbuf_lock_irq();
1414
1415			if (seq < log_first_seq) {
1416				/* messages are gone, move to next one */
1417				seq = log_first_seq;
1418				idx = log_first_idx;
1419			}
1420		}
1421	}
1422
1423	if (clear) {
1424		clear_seq = log_next_seq;
1425		clear_idx = log_next_idx;
1426	}
1427	logbuf_unlock_irq();
1428
1429	kfree(text);
1430	return len;
1431}
1432
1433int do_syslog(int type, char __user *buf, int len, int source)
1434{
1435	bool clear = false;
1436	static int saved_console_loglevel = LOGLEVEL_DEFAULT;
1437	int error;
1438
1439	error = check_syslog_permissions(type, source);
1440	if (error)
1441		return error;
1442
1443	switch (type) {
1444	case SYSLOG_ACTION_CLOSE:	/* Close log */
1445		break;
1446	case SYSLOG_ACTION_OPEN:	/* Open log */
1447		break;
1448	case SYSLOG_ACTION_READ:	/* Read from log */
 
1449		if (!buf || len < 0)
1450			return -EINVAL;
 
1451		if (!len)
1452			return 0;
1453		if (!access_ok(VERIFY_WRITE, buf, len))
1454			return -EFAULT;
 
 
1455		error = wait_event_interruptible(log_wait,
1456						 syslog_seq != log_next_seq);
1457		if (error)
1458			return error;
1459		error = syslog_print(buf, len);
1460		break;
1461	/* Read/clear last kernel messages */
1462	case SYSLOG_ACTION_READ_CLEAR:
1463		clear = true;
1464		/* FALL THRU */
1465	/* Read last kernel messages */
1466	case SYSLOG_ACTION_READ_ALL:
 
1467		if (!buf || len < 0)
1468			return -EINVAL;
 
1469		if (!len)
1470			return 0;
1471		if (!access_ok(VERIFY_WRITE, buf, len))
1472			return -EFAULT;
 
 
1473		error = syslog_print_all(buf, len, clear);
1474		break;
1475	/* Clear ring buffer */
1476	case SYSLOG_ACTION_CLEAR:
1477		syslog_print_all(NULL, 0, true);
1478		break;
1479	/* Disable logging to console */
1480	case SYSLOG_ACTION_CONSOLE_OFF:
1481		if (saved_console_loglevel == LOGLEVEL_DEFAULT)
1482			saved_console_loglevel = console_loglevel;
1483		console_loglevel = minimum_console_loglevel;
1484		break;
1485	/* Enable logging to console */
1486	case SYSLOG_ACTION_CONSOLE_ON:
1487		if (saved_console_loglevel != LOGLEVEL_DEFAULT) {
1488			console_loglevel = saved_console_loglevel;
1489			saved_console_loglevel = LOGLEVEL_DEFAULT;
1490		}
1491		break;
1492	/* Set level of messages printed to console */
1493	case SYSLOG_ACTION_CONSOLE_LEVEL:
 
1494		if (len < 1 || len > 8)
1495			return -EINVAL;
1496		if (len < minimum_console_loglevel)
1497			len = minimum_console_loglevel;
1498		console_loglevel = len;
1499		/* Implicitly re-enable logging to console */
1500		saved_console_loglevel = LOGLEVEL_DEFAULT;
 
1501		break;
1502	/* Number of chars in the log buffer */
1503	case SYSLOG_ACTION_SIZE_UNREAD:
1504		logbuf_lock_irq();
1505		if (syslog_seq < log_first_seq) {
1506			/* messages are gone, move to first one */
1507			syslog_seq = log_first_seq;
1508			syslog_idx = log_first_idx;
1509			syslog_partial = 0;
1510		}
1511		if (source == SYSLOG_FROM_PROC) {
1512			/*
1513			 * Short-cut for poll(/"proc/kmsg") which simply checks
1514			 * for pending data, not the size; return the count of
1515			 * records, not the length.
1516			 */
1517			error = log_next_seq - syslog_seq;
1518		} else {
1519			u64 seq = syslog_seq;
1520			u32 idx = syslog_idx;
1521
 
1522			while (seq < log_next_seq) {
1523				struct printk_log *msg = log_from_idx(idx);
1524
1525				error += msg_print_text(msg, true, NULL, 0);
1526				idx = log_next(idx);
1527				seq++;
1528			}
1529			error -= syslog_partial;
1530		}
1531		logbuf_unlock_irq();
1532		break;
1533	/* Size of the log buffer */
1534	case SYSLOG_ACTION_SIZE_BUFFER:
1535		error = log_buf_len;
1536		break;
1537	default:
1538		error = -EINVAL;
1539		break;
1540	}
1541
1542	return error;
1543}
1544
1545SYSCALL_DEFINE3(syslog, int, type, char __user *, buf, int, len)
1546{
1547	return do_syslog(type, buf, len, SYSLOG_FROM_READER);
1548}
1549
1550/*
1551 * Special console_lock variants that help to reduce the risk of soft-lockups.
1552 * They allow to pass console_lock to another printk() call using a busy wait.
1553 */
1554
1555#ifdef CONFIG_LOCKDEP
1556static struct lockdep_map console_owner_dep_map = {
1557	.name = "console_owner"
1558};
1559#endif
1560
1561static DEFINE_RAW_SPINLOCK(console_owner_lock);
1562static struct task_struct *console_owner;
1563static bool console_waiter;
1564
1565/**
1566 * console_lock_spinning_enable - mark beginning of code where another
1567 *	thread might safely busy wait
1568 *
1569 * This basically converts console_lock into a spinlock. This marks
1570 * the section where the console_lock owner can not sleep, because
1571 * there may be a waiter spinning (like a spinlock). Also it must be
1572 * ready to hand over the lock at the end of the section.
1573 */
1574static void console_lock_spinning_enable(void)
1575{
1576	raw_spin_lock(&console_owner_lock);
1577	console_owner = current;
1578	raw_spin_unlock(&console_owner_lock);
1579
1580	/* The waiter may spin on us after setting console_owner */
1581	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1582}
1583
1584/**
1585 * console_lock_spinning_disable_and_check - mark end of code where another
1586 *	thread was able to busy wait and check if there is a waiter
1587 *
1588 * This is called at the end of the section where spinning is allowed.
1589 * It has two functions. First, it is a signal that it is no longer
1590 * safe to start busy waiting for the lock. Second, it checks if
1591 * there is a busy waiter and passes the lock rights to her.
1592 *
1593 * Important: Callers lose the lock if there was a busy waiter.
1594 *	They must not touch items synchronized by console_lock
1595 *	in this case.
1596 *
1597 * Return: 1 if the lock rights were passed, 0 otherwise.
1598 */
1599static int console_lock_spinning_disable_and_check(void)
1600{
1601	int waiter;
1602
1603	raw_spin_lock(&console_owner_lock);
1604	waiter = READ_ONCE(console_waiter);
1605	console_owner = NULL;
1606	raw_spin_unlock(&console_owner_lock);
1607
1608	if (!waiter) {
1609		spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1610		return 0;
1611	}
1612
1613	/* The waiter is now free to continue */
1614	WRITE_ONCE(console_waiter, false);
1615
1616	spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1617
1618	/*
1619	 * Hand off console_lock to waiter. The waiter will perform
1620	 * the up(). After this, the waiter is the console_lock owner.
1621	 */
1622	mutex_release(&console_lock_dep_map, 1, _THIS_IP_);
1623	return 1;
1624}
1625
1626/**
1627 * console_trylock_spinning - try to get console_lock by busy waiting
1628 *
1629 * This allows to busy wait for the console_lock when the current
1630 * owner is running in specially marked sections. It means that
1631 * the current owner is running and cannot reschedule until it
1632 * is ready to lose the lock.
1633 *
1634 * Return: 1 if we got the lock, 0 othrewise
1635 */
1636static int console_trylock_spinning(void)
1637{
1638	struct task_struct *owner = NULL;
1639	bool waiter;
1640	bool spin = false;
1641	unsigned long flags;
1642
1643	if (console_trylock())
1644		return 1;
1645
1646	printk_safe_enter_irqsave(flags);
1647
1648	raw_spin_lock(&console_owner_lock);
1649	owner = READ_ONCE(console_owner);
1650	waiter = READ_ONCE(console_waiter);
1651	if (!waiter && owner && owner != current) {
1652		WRITE_ONCE(console_waiter, true);
1653		spin = true;
1654	}
1655	raw_spin_unlock(&console_owner_lock);
1656
1657	/*
1658	 * If there is an active printk() writing to the
1659	 * consoles, instead of having it write our data too,
1660	 * see if we can offload that load from the active
1661	 * printer, and do some printing ourselves.
1662	 * Go into a spin only if there isn't already a waiter
1663	 * spinning, and there is an active printer, and
1664	 * that active printer isn't us (recursive printk?).
1665	 */
1666	if (!spin) {
1667		printk_safe_exit_irqrestore(flags);
1668		return 0;
1669	}
1670
1671	/* We spin waiting for the owner to release us */
1672	spin_acquire(&console_owner_dep_map, 0, 0, _THIS_IP_);
1673	/* Owner will clear console_waiter on hand off */
1674	while (READ_ONCE(console_waiter))
1675		cpu_relax();
1676	spin_release(&console_owner_dep_map, 1, _THIS_IP_);
1677
1678	printk_safe_exit_irqrestore(flags);
1679	/*
1680	 * The owner passed the console lock to us.
1681	 * Since we did not spin on console lock, annotate
1682	 * this as a trylock. Otherwise lockdep will
1683	 * complain.
1684	 */
1685	mutex_acquire(&console_lock_dep_map, 0, 1, _THIS_IP_);
1686
1687	return 1;
1688}
1689
1690/*
1691 * Call the console drivers, asking them to write out
1692 * log_buf[start] to log_buf[end - 1].
1693 * The console_lock must be held.
1694 */
1695static void call_console_drivers(const char *ext_text, size_t ext_len,
 
1696				 const char *text, size_t len)
1697{
1698	struct console *con;
1699
1700	trace_console_rcuidle(text, len);
1701
1702	if (!console_drivers)
1703		return;
1704
1705	for_each_console(con) {
1706		if (exclusive_console && con != exclusive_console)
1707			continue;
1708		if (!(con->flags & CON_ENABLED))
1709			continue;
1710		if (!con->write)
1711			continue;
1712		if (!cpu_online(smp_processor_id()) &&
1713		    !(con->flags & CON_ANYTIME))
1714			continue;
1715		if (con->flags & CON_EXTENDED)
1716			con->write(con, ext_text, ext_len);
1717		else
1718			con->write(con, text, len);
1719	}
1720}
1721
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1722int printk_delay_msec __read_mostly;
1723
1724static inline void printk_delay(void)
1725{
1726	if (unlikely(printk_delay_msec)) {
1727		int m = printk_delay_msec;
1728
1729		while (m--) {
1730			mdelay(1);
1731			touch_nmi_watchdog();
1732		}
1733	}
1734}
1735
1736/*
1737 * Continuation lines are buffered, and not committed to the record buffer
1738 * until the line is complete, or a race forces it. The line fragments
1739 * though, are printed immediately to the consoles to ensure everything has
1740 * reached the console in case of a kernel crash.
1741 */
1742static struct cont {
1743	char buf[LOG_LINE_MAX];
1744	size_t len;			/* length == 0 means unused buffer */
1745	struct task_struct *owner;	/* task of first print*/
1746	u64 ts_nsec;			/* time of first print */
1747	u8 level;			/* log level of first message */
1748	u8 facility;			/* log facility of first message */
1749	enum log_flags flags;		/* prefix, newline flags */
1750} cont;
1751
1752static void cont_flush(void)
1753{
1754	if (cont.len == 0)
1755		return;
1756
1757	log_store(cont.facility, cont.level, cont.flags, cont.ts_nsec,
1758		  NULL, 0, cont.buf, cont.len);
1759	cont.len = 0;
1760}
1761
1762static bool cont_add(int facility, int level, enum log_flags flags, const char *text, size_t len)
1763{
1764	/*
1765	 * If ext consoles are present, flush and skip in-kernel
1766	 * continuation.  See nr_ext_console_drivers definition.  Also, if
1767	 * the line gets too long, split it up in separate records.
1768	 */
1769	if (nr_ext_console_drivers || cont.len + len > sizeof(cont.buf)) {
1770		cont_flush();
1771		return false;
1772	}
1773
1774	if (!cont.len) {
1775		cont.facility = facility;
1776		cont.level = level;
1777		cont.owner = current;
1778		cont.ts_nsec = local_clock();
1779		cont.flags = flags;
1780	}
1781
1782	memcpy(cont.buf + cont.len, text, len);
1783	cont.len += len;
1784
1785	// The original flags come from the first line,
1786	// but later continuations can add a newline.
1787	if (flags & LOG_NEWLINE) {
1788		cont.flags |= LOG_NEWLINE;
1789		cont_flush();
1790	}
1791
1792	if (cont.len > (sizeof(cont.buf) * 80) / 100)
1793		cont_flush();
1794
1795	return true;
1796}
1797
1798static size_t log_output(int facility, int level, enum log_flags lflags, const char *dict, size_t dictlen, char *text, size_t text_len)
1799{
1800	/*
1801	 * If an earlier line was buffered, and we're a continuation
1802	 * write from the same process, try to add it to the buffer.
1803	 */
1804	if (cont.len) {
1805		if (cont.owner == current && (lflags & LOG_CONT)) {
1806			if (cont_add(facility, level, lflags, text, text_len))
1807				return text_len;
1808		}
1809		/* Otherwise, make sure it's flushed */
1810		cont_flush();
1811	}
1812
1813	/* Skip empty continuation lines that couldn't be added - they just flush */
1814	if (!text_len && (lflags & LOG_CONT))
1815		return 0;
1816
1817	/* If it doesn't end in a newline, try to buffer the current line */
1818	if (!(lflags & LOG_NEWLINE)) {
1819		if (cont_add(facility, level, lflags, text, text_len))
1820			return text_len;
1821	}
1822
1823	/* Store it in the record log */
1824	return log_store(facility, level, lflags, 0, dict, dictlen, text, text_len);
1825}
1826
1827asmlinkage int vprintk_emit(int facility, int level,
1828			    const char *dict, size_t dictlen,
1829			    const char *fmt, va_list args)
1830{
 
1831	static char textbuf[LOG_LINE_MAX];
1832	char *text = textbuf;
1833	size_t text_len;
1834	enum log_flags lflags = 0;
1835	unsigned long flags;
1836	int printed_len;
 
 
1837	bool in_sched = false;
 
 
1838
1839	if (level == LOGLEVEL_SCHED) {
1840		level = LOGLEVEL_DEFAULT;
1841		in_sched = true;
1842	}
1843
1844	boot_delay_msec(level);
1845	printk_delay();
1846
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1847	/* This stops the holder of console_sem just where we want him */
1848	logbuf_lock_irqsave(flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1849	/*
1850	 * The printf needs to come first; we need the syslog
1851	 * prefix which might be passed-in as a parameter.
1852	 */
1853	text_len = vscnprintf(text, sizeof(textbuf), fmt, args);
1854
1855	/* mark and strip a trailing newline */
1856	if (text_len && text[text_len-1] == '\n') {
1857		text_len--;
1858		lflags |= LOG_NEWLINE;
1859	}
1860
1861	/* strip kernel syslog prefix and extract log level or control flags */
1862	if (facility == 0) {
1863		int kern_level;
1864
1865		while ((kern_level = printk_get_level(text)) != 0) {
1866			switch (kern_level) {
1867			case '0' ... '7':
1868				if (level == LOGLEVEL_DEFAULT)
1869					level = kern_level - '0';
1870				/* fallthrough */
1871			case 'd':	/* KERN_DEFAULT */
1872				lflags |= LOG_PREFIX;
1873				break;
1874			case 'c':	/* KERN_CONT */
1875				lflags |= LOG_CONT;
1876			}
1877
1878			text_len -= 2;
1879			text += 2;
1880		}
1881	}
1882
1883	if (level == LOGLEVEL_DEFAULT)
1884		level = default_message_loglevel;
1885
1886	if (dict)
1887		lflags |= LOG_PREFIX|LOG_NEWLINE;
1888
1889	printed_len = log_output(facility, level, lflags, dict, dictlen, text, text_len);
1890
1891	logbuf_unlock_irqrestore(flags);
 
 
 
1892
1893	/* If called from the scheduler, we can not call up(). */
1894	if (!in_sched) {
1895		/*
1896		 * Disable preemption to avoid being preempted while holding
1897		 * console_sem which would prevent anyone from printing to
1898		 * console
1899		 */
1900		preempt_disable();
1901		/*
1902		 * Try to acquire and then immediately release the console
1903		 * semaphore.  The release will print out buffers and wake up
1904		 * /dev/kmsg and syslog() users.
1905		 */
1906		if (console_trylock_spinning())
1907			console_unlock();
1908		preempt_enable();
1909	}
1910
1911	return printed_len;
1912}
1913EXPORT_SYMBOL(vprintk_emit);
1914
1915asmlinkage int vprintk(const char *fmt, va_list args)
1916{
1917	return vprintk_func(fmt, args);
1918}
1919EXPORT_SYMBOL(vprintk);
1920
1921asmlinkage int printk_emit(int facility, int level,
1922			   const char *dict, size_t dictlen,
1923			   const char *fmt, ...)
1924{
1925	va_list args;
1926	int r;
1927
1928	va_start(args, fmt);
1929	r = vprintk_emit(facility, level, dict, dictlen, fmt, args);
1930	va_end(args);
1931
1932	return r;
1933}
1934EXPORT_SYMBOL(printk_emit);
1935
1936int vprintk_default(const char *fmt, va_list args)
1937{
1938	int r;
1939
1940#ifdef CONFIG_KGDB_KDB
1941	/* Allow to pass printk() to kdb but avoid a recursion. */
1942	if (unlikely(kdb_trap_printk && kdb_printf_cpu < 0)) {
1943		r = vkdb_printf(KDB_MSGSRC_PRINTK, fmt, args);
1944		return r;
1945	}
1946#endif
1947	r = vprintk_emit(0, LOGLEVEL_DEFAULT, NULL, 0, fmt, args);
1948
1949	return r;
1950}
1951EXPORT_SYMBOL_GPL(vprintk_default);
1952
1953/**
1954 * printk - print a kernel message
1955 * @fmt: format string
1956 *
1957 * This is printk(). It can be called from any context. We want it to work.
1958 *
1959 * We try to grab the console_lock. If we succeed, it's easy - we log the
1960 * output and call the console drivers.  If we fail to get the semaphore, we
1961 * place the output into the log buffer and return. The current holder of
1962 * the console_sem will notice the new output in console_unlock(); and will
1963 * send it to the consoles before releasing the lock.
1964 *
1965 * One effect of this deferred printing is that code which calls printk() and
1966 * then changes console_loglevel may break. This is because console_loglevel
1967 * is inspected when the actual printing occurs.
1968 *
1969 * See also:
1970 * printf(3)
1971 *
1972 * See the vsnprintf() documentation for format string extensions over C99.
1973 */
1974asmlinkage __visible int printk(const char *fmt, ...)
1975{
1976	va_list args;
1977	int r;
1978
1979	va_start(args, fmt);
1980	r = vprintk_func(fmt, args);
1981	va_end(args);
1982
1983	return r;
1984}
1985EXPORT_SYMBOL(printk);
1986
1987#else /* CONFIG_PRINTK */
1988
1989#define LOG_LINE_MAX		0
1990#define PREFIX_MAX		0
1991
1992static u64 syslog_seq;
1993static u32 syslog_idx;
1994static u64 console_seq;
1995static u32 console_idx;
1996static u64 log_first_seq;
1997static u32 log_first_idx;
1998static u64 log_next_seq;
1999static char *log_text(const struct printk_log *msg) { return NULL; }
2000static char *log_dict(const struct printk_log *msg) { return NULL; }
2001static struct printk_log *log_from_idx(u32 idx) { return NULL; }
2002static u32 log_next(u32 idx) { return 0; }
2003static ssize_t msg_print_ext_header(char *buf, size_t size,
2004				    struct printk_log *msg,
2005				    u64 seq) { return 0; }
2006static ssize_t msg_print_ext_body(char *buf, size_t size,
2007				  char *dict, size_t dict_len,
2008				  char *text, size_t text_len) { return 0; }
2009static void console_lock_spinning_enable(void) { }
2010static int console_lock_spinning_disable_and_check(void) { return 0; }
2011static void call_console_drivers(const char *ext_text, size_t ext_len,
2012				 const char *text, size_t len) {}
2013static size_t msg_print_text(const struct printk_log *msg,
2014			     bool syslog, char *buf, size_t size) { return 0; }
2015static bool suppress_message_printing(int level) { return false; }
2016
 
 
 
2017#endif /* CONFIG_PRINTK */
2018
2019#ifdef CONFIG_EARLY_PRINTK
2020struct console *early_console;
2021
2022asmlinkage __visible void early_printk(const char *fmt, ...)
2023{
2024	va_list ap;
2025	char buf[512];
2026	int n;
2027
2028	if (!early_console)
2029		return;
2030
2031	va_start(ap, fmt);
2032	n = vscnprintf(buf, sizeof(buf), fmt, ap);
2033	va_end(ap);
2034
2035	early_console->write(early_console, buf, n);
2036}
2037#endif
2038
2039static int __add_preferred_console(char *name, int idx, char *options,
2040				   char *brl_options)
2041{
2042	struct console_cmdline *c;
2043	int i;
2044
2045	/*
2046	 *	See if this tty is not yet registered, and
2047	 *	if we have a slot free.
2048	 */
2049	for (i = 0, c = console_cmdline;
2050	     i < MAX_CMDLINECONSOLES && c->name[0];
2051	     i++, c++) {
2052		if (strcmp(c->name, name) == 0 && c->index == idx) {
2053			if (!brl_options)
2054				preferred_console = i;
2055			return 0;
2056		}
2057	}
2058	if (i == MAX_CMDLINECONSOLES)
2059		return -E2BIG;
2060	if (!brl_options)
2061		preferred_console = i;
2062	strlcpy(c->name, name, sizeof(c->name));
2063	c->options = options;
2064	braille_set_options(c, brl_options);
2065
2066	c->index = idx;
2067	return 0;
2068}
2069
2070static int __init console_msg_format_setup(char *str)
2071{
2072	if (!strcmp(str, "syslog"))
2073		console_msg_format = MSG_FORMAT_SYSLOG;
2074	if (!strcmp(str, "default"))
2075		console_msg_format = MSG_FORMAT_DEFAULT;
2076	return 1;
2077}
2078__setup("console_msg_format=", console_msg_format_setup);
2079
2080/*
2081 * Set up a console.  Called via do_early_param() in init/main.c
2082 * for each "console=" parameter in the boot command line.
2083 */
2084static int __init console_setup(char *str)
2085{
2086	char buf[sizeof(console_cmdline[0].name) + 4]; /* 4 for "ttyS" */
2087	char *s, *options, *brl_options = NULL;
2088	int idx;
2089
2090	if (_braille_console_setup(&str, &brl_options))
2091		return 1;
2092
2093	/*
2094	 * Decode str into name, index, options.
2095	 */
2096	if (str[0] >= '0' && str[0] <= '9') {
2097		strcpy(buf, "ttyS");
2098		strncpy(buf + 4, str, sizeof(buf) - 5);
2099	} else {
2100		strncpy(buf, str, sizeof(buf) - 1);
2101	}
2102	buf[sizeof(buf) - 1] = 0;
2103	options = strchr(str, ',');
2104	if (options)
2105		*(options++) = 0;
2106#ifdef __sparc__
2107	if (!strcmp(str, "ttya"))
2108		strcpy(buf, "ttyS0");
2109	if (!strcmp(str, "ttyb"))
2110		strcpy(buf, "ttyS1");
2111#endif
2112	for (s = buf; *s; s++)
2113		if (isdigit(*s) || *s == ',')
2114			break;
2115	idx = simple_strtoul(s, NULL, 10);
2116	*s = 0;
2117
2118	__add_preferred_console(buf, idx, options, brl_options);
2119	console_set_on_cmdline = 1;
2120	return 1;
2121}
2122__setup("console=", console_setup);
2123
2124/**
2125 * add_preferred_console - add a device to the list of preferred consoles.
2126 * @name: device name
2127 * @idx: device index
2128 * @options: options for this console
2129 *
2130 * The last preferred console added will be used for kernel messages
2131 * and stdin/out/err for init.  Normally this is used by console_setup
2132 * above to handle user-supplied console arguments; however it can also
2133 * be used by arch-specific code either to override the user or more
2134 * commonly to provide a default console (ie from PROM variables) when
2135 * the user has not supplied one.
2136 */
2137int add_preferred_console(char *name, int idx, char *options)
2138{
2139	return __add_preferred_console(name, idx, options, NULL);
2140}
2141
2142bool console_suspend_enabled = true;
2143EXPORT_SYMBOL(console_suspend_enabled);
2144
2145static int __init console_suspend_disable(char *str)
2146{
2147	console_suspend_enabled = false;
2148	return 1;
2149}
2150__setup("no_console_suspend", console_suspend_disable);
2151module_param_named(console_suspend, console_suspend_enabled,
2152		bool, S_IRUGO | S_IWUSR);
2153MODULE_PARM_DESC(console_suspend, "suspend console during suspend"
2154	" and hibernate operations");
2155
2156/**
2157 * suspend_console - suspend the console subsystem
2158 *
2159 * This disables printk() while we go into suspend states
2160 */
2161void suspend_console(void)
2162{
2163	if (!console_suspend_enabled)
2164		return;
2165	pr_info("Suspending console(s) (use no_console_suspend to debug)\n");
2166	console_lock();
2167	console_suspended = 1;
2168	up_console_sem();
2169}
2170
2171void resume_console(void)
2172{
2173	if (!console_suspend_enabled)
2174		return;
2175	down_console_sem();
2176	console_suspended = 0;
2177	console_unlock();
2178}
2179
2180/**
2181 * console_cpu_notify - print deferred console messages after CPU hotplug
2182 * @cpu: unused
2183 *
2184 * If printk() is called from a CPU that is not online yet, the messages
2185 * will be printed on the console only if there are CON_ANYTIME consoles.
2186 * This function is called when a new CPU comes online (or fails to come
2187 * up) or goes offline.
2188 */
2189static int console_cpu_notify(unsigned int cpu)
2190{
2191	if (!cpuhp_tasks_frozen) {
2192		/* If trylock fails, someone else is doing the printing */
2193		if (console_trylock())
2194			console_unlock();
2195	}
2196	return 0;
2197}
2198
2199/**
2200 * console_lock - lock the console system for exclusive use.
2201 *
2202 * Acquires a lock which guarantees that the caller has
2203 * exclusive access to the console system and the console_drivers list.
2204 *
2205 * Can sleep, returns nothing.
2206 */
2207void console_lock(void)
2208{
2209	might_sleep();
2210
2211	down_console_sem();
2212	if (console_suspended)
2213		return;
2214	console_locked = 1;
2215	console_may_schedule = 1;
2216}
2217EXPORT_SYMBOL(console_lock);
2218
2219/**
2220 * console_trylock - try to lock the console system for exclusive use.
2221 *
2222 * Try to acquire a lock which guarantees that the caller has exclusive
2223 * access to the console system and the console_drivers list.
2224 *
2225 * returns 1 on success, and 0 on failure to acquire the lock.
2226 */
2227int console_trylock(void)
2228{
2229	if (down_trylock_console_sem())
2230		return 0;
2231	if (console_suspended) {
2232		up_console_sem();
2233		return 0;
2234	}
2235	console_locked = 1;
2236	console_may_schedule = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
2237	return 1;
2238}
2239EXPORT_SYMBOL(console_trylock);
2240
2241int is_console_locked(void)
2242{
2243	return console_locked;
2244}
2245
2246/*
2247 * Check if we have any console that is capable of printing while cpu is
2248 * booting or shutting down. Requires console_sem.
2249 */
2250static int have_callable_console(void)
2251{
2252	struct console *con;
2253
2254	for_each_console(con)
2255		if ((con->flags & CON_ENABLED) &&
2256				(con->flags & CON_ANYTIME))
2257			return 1;
2258
2259	return 0;
2260}
2261
2262/*
2263 * Can we actually use the console at this time on this cpu?
2264 *
2265 * Console drivers may assume that per-cpu resources have been allocated. So
2266 * unless they're explicitly marked as being able to cope (CON_ANYTIME) don't
2267 * call them until this CPU is officially up.
2268 */
2269static inline int can_use_console(void)
2270{
2271	return cpu_online(raw_smp_processor_id()) || have_callable_console();
2272}
2273
2274/**
2275 * console_unlock - unlock the console system
2276 *
2277 * Releases the console_lock which the caller holds on the console system
2278 * and the console driver list.
2279 *
2280 * While the console_lock was held, console output may have been buffered
2281 * by printk().  If this is the case, console_unlock(); emits
2282 * the output prior to releasing the lock.
2283 *
2284 * If there is output waiting, we wake /dev/kmsg and syslog() users.
2285 *
2286 * console_unlock(); may be called from any context.
2287 */
2288void console_unlock(void)
2289{
2290	static char ext_text[CONSOLE_EXT_LOG_MAX];
2291	static char text[LOG_LINE_MAX + PREFIX_MAX];
2292	static u64 seen_seq;
2293	unsigned long flags;
2294	bool wake_klogd = false;
2295	bool do_cond_resched, retry;
2296
2297	if (console_suspended) {
2298		up_console_sem();
2299		return;
2300	}
2301
2302	/*
2303	 * Console drivers are called with interrupts disabled, so
2304	 * @console_may_schedule should be cleared before; however, we may
2305	 * end up dumping a lot of lines, for example, if called from
2306	 * console registration path, and should invoke cond_resched()
2307	 * between lines if allowable.  Not doing so can cause a very long
2308	 * scheduling stall on a slow console leading to RCU stall and
2309	 * softlockup warnings which exacerbate the issue with more
2310	 * messages practically incapacitating the system.
2311	 *
2312	 * console_trylock() is not able to detect the preemptive
2313	 * context reliably. Therefore the value must be stored before
2314	 * and cleared after the the "again" goto label.
2315	 */
2316	do_cond_resched = console_may_schedule;
2317again:
2318	console_may_schedule = 0;
2319
 
2320	/*
2321	 * We released the console_sem lock, so we need to recheck if
2322	 * cpu is online and (if not) is there at least one CON_ANYTIME
2323	 * console.
2324	 */
2325	if (!can_use_console()) {
2326		console_locked = 0;
2327		up_console_sem();
2328		return;
2329	}
2330
2331	for (;;) {
2332		struct printk_log *msg;
2333		size_t ext_len = 0;
2334		size_t len;
 
2335
2336		printk_safe_enter_irqsave(flags);
2337		raw_spin_lock(&logbuf_lock);
2338		if (seen_seq != log_next_seq) {
2339			wake_klogd = true;
2340			seen_seq = log_next_seq;
2341		}
2342
2343		if (console_seq < log_first_seq) {
2344			len = sprintf(text, "** %u printk messages dropped **\n",
2345				      (unsigned)(log_first_seq - console_seq));
2346
2347			/* messages are gone, move to first one */
2348			console_seq = log_first_seq;
2349			console_idx = log_first_idx;
2350		} else {
2351			len = 0;
2352		}
2353skip:
2354		if (console_seq == log_next_seq)
2355			break;
2356
2357		msg = log_from_idx(console_idx);
2358		if (suppress_message_printing(msg->level)) {
 
2359			/*
2360			 * Skip record we have buffered and already printed
2361			 * directly to the console when we received it, and
2362			 * record that has level above the console loglevel.
2363			 */
2364			console_idx = log_next(console_idx);
2365			console_seq++;
2366			goto skip;
2367		}
2368
2369		len += msg_print_text(msg,
2370				console_msg_format & MSG_FORMAT_SYSLOG,
2371				text + len,
2372				sizeof(text) - len);
2373		if (nr_ext_console_drivers) {
2374			ext_len = msg_print_ext_header(ext_text,
2375						sizeof(ext_text),
2376						msg, console_seq);
2377			ext_len += msg_print_ext_body(ext_text + ext_len,
2378						sizeof(ext_text) - ext_len,
2379						log_dict(msg), msg->dict_len,
2380						log_text(msg), msg->text_len);
2381		}
2382		console_idx = log_next(console_idx);
2383		console_seq++;
2384		raw_spin_unlock(&logbuf_lock);
2385
2386		/*
2387		 * While actively printing out messages, if another printk()
2388		 * were to occur on another CPU, it may wait for this one to
2389		 * finish. This task can not be preempted if there is a
2390		 * waiter waiting to take over.
2391		 */
2392		console_lock_spinning_enable();
2393
2394		stop_critical_timings();	/* don't trace print latency */
2395		call_console_drivers(ext_text, ext_len, text, len);
2396		start_critical_timings();
2397
2398		if (console_lock_spinning_disable_and_check()) {
2399			printk_safe_exit_irqrestore(flags);
2400			goto out;
2401		}
2402
2403		printk_safe_exit_irqrestore(flags);
2404
2405		if (do_cond_resched)
2406			cond_resched();
2407	}
2408
2409	console_locked = 0;
2410
2411	/* Release the exclusive_console once it is used */
2412	if (unlikely(exclusive_console))
2413		exclusive_console = NULL;
2414
2415	raw_spin_unlock(&logbuf_lock);
2416
2417	up_console_sem();
2418
2419	/*
2420	 * Someone could have filled up the buffer again, so re-check if there's
2421	 * something to flush. In case we cannot trylock the console_sem again,
2422	 * there's a new owner and the console_unlock() from them will do the
2423	 * flush, no worries.
2424	 */
2425	raw_spin_lock(&logbuf_lock);
2426	retry = console_seq != log_next_seq;
2427	raw_spin_unlock(&logbuf_lock);
2428	printk_safe_exit_irqrestore(flags);
2429
2430	if (retry && console_trylock())
2431		goto again;
2432
2433out:
2434	if (wake_klogd)
2435		wake_up_klogd();
2436}
2437EXPORT_SYMBOL(console_unlock);
2438
2439/**
2440 * console_conditional_schedule - yield the CPU if required
2441 *
2442 * If the console code is currently allowed to sleep, and
2443 * if this CPU should yield the CPU to another task, do
2444 * so here.
2445 *
2446 * Must be called within console_lock();.
2447 */
2448void __sched console_conditional_schedule(void)
2449{
2450	if (console_may_schedule)
2451		cond_resched();
2452}
2453EXPORT_SYMBOL(console_conditional_schedule);
2454
2455void console_unblank(void)
2456{
2457	struct console *c;
2458
2459	/*
2460	 * console_unblank can no longer be called in interrupt context unless
2461	 * oops_in_progress is set to 1..
2462	 */
2463	if (oops_in_progress) {
2464		if (down_trylock_console_sem() != 0)
2465			return;
2466	} else
2467		console_lock();
2468
2469	console_locked = 1;
2470	console_may_schedule = 0;
2471	for_each_console(c)
2472		if ((c->flags & CON_ENABLED) && c->unblank)
2473			c->unblank();
2474	console_unlock();
2475}
2476
2477/**
2478 * console_flush_on_panic - flush console content on panic
2479 *
2480 * Immediately output all pending messages no matter what.
2481 */
2482void console_flush_on_panic(void)
2483{
2484	/*
2485	 * If someone else is holding the console lock, trylock will fail
2486	 * and may_schedule may be set.  Ignore and proceed to unlock so
2487	 * that messages are flushed out.  As this can be called from any
2488	 * context and we don't want to get preempted while flushing,
2489	 * ensure may_schedule is cleared.
2490	 */
2491	console_trylock();
2492	console_may_schedule = 0;
2493	console_unlock();
2494}
2495
2496/*
2497 * Return the console tty driver structure and its associated index
2498 */
2499struct tty_driver *console_device(int *index)
2500{
2501	struct console *c;
2502	struct tty_driver *driver = NULL;
2503
2504	console_lock();
2505	for_each_console(c) {
2506		if (!c->device)
2507			continue;
2508		driver = c->device(c, index);
2509		if (driver)
2510			break;
2511	}
2512	console_unlock();
2513	return driver;
2514}
2515
2516/*
2517 * Prevent further output on the passed console device so that (for example)
2518 * serial drivers can disable console output before suspending a port, and can
2519 * re-enable output afterwards.
2520 */
2521void console_stop(struct console *console)
2522{
2523	console_lock();
2524	console->flags &= ~CON_ENABLED;
2525	console_unlock();
2526}
2527EXPORT_SYMBOL(console_stop);
2528
2529void console_start(struct console *console)
2530{
2531	console_lock();
2532	console->flags |= CON_ENABLED;
2533	console_unlock();
2534}
2535EXPORT_SYMBOL(console_start);
2536
2537static int __read_mostly keep_bootcon;
2538
2539static int __init keep_bootcon_setup(char *str)
2540{
2541	keep_bootcon = 1;
2542	pr_info("debug: skip boot console de-registration.\n");
2543
2544	return 0;
2545}
2546
2547early_param("keep_bootcon", keep_bootcon_setup);
2548
2549/*
2550 * The console driver calls this routine during kernel initialization
2551 * to register the console printing procedure with printk() and to
2552 * print any messages that were printed by the kernel before the
2553 * console driver was initialized.
2554 *
2555 * This can happen pretty early during the boot process (because of
2556 * early_printk) - sometimes before setup_arch() completes - be careful
2557 * of what kernel features are used - they may not be initialised yet.
2558 *
2559 * There are two types of consoles - bootconsoles (early_printk) and
2560 * "real" consoles (everything which is not a bootconsole) which are
2561 * handled differently.
2562 *  - Any number of bootconsoles can be registered at any time.
2563 *  - As soon as a "real" console is registered, all bootconsoles
2564 *    will be unregistered automatically.
2565 *  - Once a "real" console is registered, any attempt to register a
2566 *    bootconsoles will be rejected
2567 */
2568void register_console(struct console *newcon)
2569{
2570	int i;
2571	unsigned long flags;
2572	struct console *bcon = NULL;
2573	struct console_cmdline *c;
2574	static bool has_preferred;
2575
2576	if (console_drivers)
2577		for_each_console(bcon)
2578			if (WARN(bcon == newcon,
2579					"console '%s%d' already registered\n",
2580					bcon->name, bcon->index))
2581				return;
2582
2583	/*
2584	 * before we register a new CON_BOOT console, make sure we don't
2585	 * already have a valid console
2586	 */
2587	if (console_drivers && newcon->flags & CON_BOOT) {
2588		/* find the last or real console */
2589		for_each_console(bcon) {
2590			if (!(bcon->flags & CON_BOOT)) {
2591				pr_info("Too late to register bootconsole %s%d\n",
2592					newcon->name, newcon->index);
2593				return;
2594			}
2595		}
2596	}
2597
2598	if (console_drivers && console_drivers->flags & CON_BOOT)
2599		bcon = console_drivers;
2600
2601	if (!has_preferred || bcon || !console_drivers)
2602		has_preferred = preferred_console >= 0;
2603
2604	/*
2605	 *	See if we want to use this console driver. If we
2606	 *	didn't select a console we take the first one
2607	 *	that registers here.
2608	 */
2609	if (!has_preferred) {
2610		if (newcon->index < 0)
2611			newcon->index = 0;
2612		if (newcon->setup == NULL ||
2613		    newcon->setup(newcon, NULL) == 0) {
2614			newcon->flags |= CON_ENABLED;
2615			if (newcon->device) {
2616				newcon->flags |= CON_CONSDEV;
2617				has_preferred = true;
2618			}
2619		}
2620	}
2621
2622	/*
2623	 *	See if this console matches one we selected on
2624	 *	the command line.
2625	 */
2626	for (i = 0, c = console_cmdline;
2627	     i < MAX_CMDLINECONSOLES && c->name[0];
2628	     i++, c++) {
2629		if (!newcon->match ||
2630		    newcon->match(newcon, c->name, c->index, c->options) != 0) {
2631			/* default matching */
2632			BUILD_BUG_ON(sizeof(c->name) != sizeof(newcon->name));
2633			if (strcmp(c->name, newcon->name) != 0)
2634				continue;
2635			if (newcon->index >= 0 &&
2636			    newcon->index != c->index)
2637				continue;
2638			if (newcon->index < 0)
2639				newcon->index = c->index;
2640
2641			if (_braille_register_console(newcon, c))
2642				return;
2643
2644			if (newcon->setup &&
2645			    newcon->setup(newcon, c->options) != 0)
2646				break;
2647		}
2648
2649		newcon->flags |= CON_ENABLED;
2650		if (i == preferred_console) {
2651			newcon->flags |= CON_CONSDEV;
2652			has_preferred = true;
2653		}
2654		break;
2655	}
2656
2657	if (!(newcon->flags & CON_ENABLED))
2658		return;
2659
2660	/*
2661	 * If we have a bootconsole, and are switching to a real console,
2662	 * don't print everything out again, since when the boot console, and
2663	 * the real console are the same physical device, it's annoying to
2664	 * see the beginning boot messages twice
2665	 */
2666	if (bcon && ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV))
2667		newcon->flags &= ~CON_PRINTBUFFER;
2668
2669	/*
2670	 *	Put this console in the list - keep the
2671	 *	preferred driver at the head of the list.
2672	 */
2673	console_lock();
2674	if ((newcon->flags & CON_CONSDEV) || console_drivers == NULL) {
2675		newcon->next = console_drivers;
2676		console_drivers = newcon;
2677		if (newcon->next)
2678			newcon->next->flags &= ~CON_CONSDEV;
2679	} else {
2680		newcon->next = console_drivers->next;
2681		console_drivers->next = newcon;
2682	}
2683
2684	if (newcon->flags & CON_EXTENDED)
2685		if (!nr_ext_console_drivers++)
2686			pr_info("printk: continuation disabled due to ext consoles, expect more fragments in /dev/kmsg\n");
2687
2688	if (newcon->flags & CON_PRINTBUFFER) {
2689		/*
2690		 * console_unlock(); will print out the buffered messages
2691		 * for us.
2692		 */
2693		logbuf_lock_irqsave(flags);
2694		console_seq = syslog_seq;
2695		console_idx = syslog_idx;
2696		logbuf_unlock_irqrestore(flags);
2697		/*
2698		 * We're about to replay the log buffer.  Only do this to the
2699		 * just-registered console to avoid excessive message spam to
2700		 * the already-registered consoles.
2701		 */
2702		exclusive_console = newcon;
2703	}
2704	console_unlock();
2705	console_sysfs_notify();
2706
2707	/*
2708	 * By unregistering the bootconsoles after we enable the real console
2709	 * we get the "console xxx enabled" message on all the consoles -
2710	 * boot consoles, real consoles, etc - this is to ensure that end
2711	 * users know there might be something in the kernel's log buffer that
2712	 * went to the bootconsole (that they do not see on the real console)
2713	 */
2714	pr_info("%sconsole [%s%d] enabled\n",
2715		(newcon->flags & CON_BOOT) ? "boot" : "" ,
2716		newcon->name, newcon->index);
2717	if (bcon &&
2718	    ((newcon->flags & (CON_CONSDEV | CON_BOOT)) == CON_CONSDEV) &&
2719	    !keep_bootcon) {
2720		/* We need to iterate through all boot consoles, to make
2721		 * sure we print everything out, before we unregister them.
2722		 */
2723		for_each_console(bcon)
2724			if (bcon->flags & CON_BOOT)
2725				unregister_console(bcon);
2726	}
2727}
2728EXPORT_SYMBOL(register_console);
2729
2730int unregister_console(struct console *console)
2731{
2732        struct console *a, *b;
2733	int res;
2734
2735	pr_info("%sconsole [%s%d] disabled\n",
2736		(console->flags & CON_BOOT) ? "boot" : "" ,
2737		console->name, console->index);
2738
2739	res = _braille_unregister_console(console);
2740	if (res)
2741		return res;
2742
2743	res = 1;
2744	console_lock();
2745	if (console_drivers == console) {
2746		console_drivers=console->next;
2747		res = 0;
2748	} else if (console_drivers) {
2749		for (a=console_drivers->next, b=console_drivers ;
2750		     a; b=a, a=b->next) {
2751			if (a == console) {
2752				b->next = a->next;
2753				res = 0;
2754				break;
2755			}
2756		}
2757	}
2758
2759	if (!res && (console->flags & CON_EXTENDED))
2760		nr_ext_console_drivers--;
2761
2762	/*
2763	 * If this isn't the last console and it has CON_CONSDEV set, we
2764	 * need to set it on the next preferred console.
2765	 */
2766	if (console_drivers != NULL && console->flags & CON_CONSDEV)
2767		console_drivers->flags |= CON_CONSDEV;
2768
2769	console->flags &= ~CON_ENABLED;
2770	console_unlock();
2771	console_sysfs_notify();
2772	return res;
2773}
2774EXPORT_SYMBOL(unregister_console);
2775
2776/*
2777 * Initialize the console device. This is called *early*, so
2778 * we can't necessarily depend on lots of kernel help here.
2779 * Just do some early initializations, and do the complex setup
2780 * later.
2781 */
2782void __init console_init(void)
2783{
2784	int ret;
2785	initcall_t *call;
2786
2787	/* Setup the default TTY line discipline. */
2788	n_tty_init();
2789
2790	/*
2791	 * set up the console device so that later boot sequences can
2792	 * inform about problems etc..
2793	 */
2794	call = __con_initcall_start;
2795	trace_initcall_level("console");
2796	while (call < __con_initcall_end) {
2797		trace_initcall_start((*call));
2798		ret = (*call)();
2799		trace_initcall_finish((*call), ret);
2800		call++;
2801	}
2802}
2803
2804/*
2805 * Some boot consoles access data that is in the init section and which will
2806 * be discarded after the initcalls have been run. To make sure that no code
2807 * will access this data, unregister the boot consoles in a late initcall.
2808 *
2809 * If for some reason, such as deferred probe or the driver being a loadable
2810 * module, the real console hasn't registered yet at this point, there will
2811 * be a brief interval in which no messages are logged to the console, which
2812 * makes it difficult to diagnose problems that occur during this time.
2813 *
2814 * To mitigate this problem somewhat, only unregister consoles whose memory
2815 * intersects with the init section. Note that all other boot consoles will
2816 * get unregistred when the real preferred console is registered.
 
2817 */
2818static int __init printk_late_init(void)
2819{
2820	struct console *con;
2821	int ret;
2822
2823	for_each_console(con) {
2824		if (!(con->flags & CON_BOOT))
2825			continue;
2826
2827		/* Check addresses that might be used for enabled consoles. */
2828		if (init_section_intersects(con, sizeof(*con)) ||
2829		    init_section_contains(con->write, 0) ||
2830		    init_section_contains(con->read, 0) ||
2831		    init_section_contains(con->device, 0) ||
2832		    init_section_contains(con->unblank, 0) ||
2833		    init_section_contains(con->data, 0)) {
2834			/*
2835			 * Please, consider moving the reported consoles out
2836			 * of the init section.
 
 
 
2837			 */
2838			pr_warn("bootconsole [%s%d] uses init memory and must be disabled even before the real one is ready\n",
2839				con->name, con->index);
2840			unregister_console(con);
2841		}
2842	}
2843	ret = cpuhp_setup_state_nocalls(CPUHP_PRINTK_DEAD, "printk:dead", NULL,
2844					console_cpu_notify);
2845	WARN_ON(ret < 0);
2846	ret = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "printk:online",
2847					console_cpu_notify, NULL);
2848	WARN_ON(ret < 0);
2849	return 0;
2850}
2851late_initcall(printk_late_init);
2852
2853#if defined CONFIG_PRINTK
2854/*
2855 * Delayed printk version, for scheduler-internal messages:
2856 */
2857#define PRINTK_PENDING_WAKEUP	0x01
2858#define PRINTK_PENDING_OUTPUT	0x02
2859
2860static DEFINE_PER_CPU(int, printk_pending);
2861
2862static void wake_up_klogd_work_func(struct irq_work *irq_work)
2863{
2864	int pending = __this_cpu_xchg(printk_pending, 0);
2865
2866	if (pending & PRINTK_PENDING_OUTPUT) {
2867		/* If trylock fails, someone else is doing the printing */
2868		if (console_trylock())
2869			console_unlock();
2870	}
2871
2872	if (pending & PRINTK_PENDING_WAKEUP)
2873		wake_up_interruptible(&log_wait);
2874}
2875
2876static DEFINE_PER_CPU(struct irq_work, wake_up_klogd_work) = {
2877	.func = wake_up_klogd_work_func,
2878	.flags = IRQ_WORK_LAZY,
2879};
2880
2881void wake_up_klogd(void)
2882{
2883	preempt_disable();
2884	if (waitqueue_active(&log_wait)) {
2885		this_cpu_or(printk_pending, PRINTK_PENDING_WAKEUP);
2886		irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2887	}
2888	preempt_enable();
2889}
2890
2891int vprintk_deferred(const char *fmt, va_list args)
2892{
 
2893	int r;
2894
 
 
2895	r = vprintk_emit(0, LOGLEVEL_SCHED, NULL, 0, fmt, args);
 
2896
2897	preempt_disable();
2898	__this_cpu_or(printk_pending, PRINTK_PENDING_OUTPUT);
2899	irq_work_queue(this_cpu_ptr(&wake_up_klogd_work));
2900	preempt_enable();
2901
2902	return r;
2903}
2904
2905int printk_deferred(const char *fmt, ...)
2906{
2907	va_list args;
2908	int r;
2909
2910	va_start(args, fmt);
2911	r = vprintk_deferred(fmt, args);
2912	va_end(args);
2913
2914	return r;
2915}
2916
2917/*
2918 * printk rate limiting, lifted from the networking subsystem.
2919 *
2920 * This enforces a rate limit: not more than 10 kernel messages
2921 * every 5s to make a denial-of-service attack impossible.
2922 */
2923DEFINE_RATELIMIT_STATE(printk_ratelimit_state, 5 * HZ, 10);
2924
2925int __printk_ratelimit(const char *func)
2926{
2927	return ___ratelimit(&printk_ratelimit_state, func);
2928}
2929EXPORT_SYMBOL(__printk_ratelimit);
2930
2931/**
2932 * printk_timed_ratelimit - caller-controlled printk ratelimiting
2933 * @caller_jiffies: pointer to caller's state
2934 * @interval_msecs: minimum interval between prints
2935 *
2936 * printk_timed_ratelimit() returns true if more than @interval_msecs
2937 * milliseconds have elapsed since the last time printk_timed_ratelimit()
2938 * returned true.
2939 */
2940bool printk_timed_ratelimit(unsigned long *caller_jiffies,
2941			unsigned int interval_msecs)
2942{
2943	unsigned long elapsed = jiffies - *caller_jiffies;
2944
2945	if (*caller_jiffies && elapsed <= msecs_to_jiffies(interval_msecs))
2946		return false;
2947
2948	*caller_jiffies = jiffies;
2949	return true;
2950}
2951EXPORT_SYMBOL(printk_timed_ratelimit);
2952
2953static DEFINE_SPINLOCK(dump_list_lock);
2954static LIST_HEAD(dump_list);
2955
2956/**
2957 * kmsg_dump_register - register a kernel log dumper.
2958 * @dumper: pointer to the kmsg_dumper structure
2959 *
2960 * Adds a kernel log dumper to the system. The dump callback in the
2961 * structure will be called when the kernel oopses or panics and must be
2962 * set. Returns zero on success and %-EINVAL or %-EBUSY otherwise.
2963 */
2964int kmsg_dump_register(struct kmsg_dumper *dumper)
2965{
2966	unsigned long flags;
2967	int err = -EBUSY;
2968
2969	/* The dump callback needs to be set */
2970	if (!dumper->dump)
2971		return -EINVAL;
2972
2973	spin_lock_irqsave(&dump_list_lock, flags);
2974	/* Don't allow registering multiple times */
2975	if (!dumper->registered) {
2976		dumper->registered = 1;
2977		list_add_tail_rcu(&dumper->list, &dump_list);
2978		err = 0;
2979	}
2980	spin_unlock_irqrestore(&dump_list_lock, flags);
2981
2982	return err;
2983}
2984EXPORT_SYMBOL_GPL(kmsg_dump_register);
2985
2986/**
2987 * kmsg_dump_unregister - unregister a kmsg dumper.
2988 * @dumper: pointer to the kmsg_dumper structure
2989 *
2990 * Removes a dump device from the system. Returns zero on success and
2991 * %-EINVAL otherwise.
2992 */
2993int kmsg_dump_unregister(struct kmsg_dumper *dumper)
2994{
2995	unsigned long flags;
2996	int err = -EINVAL;
2997
2998	spin_lock_irqsave(&dump_list_lock, flags);
2999	if (dumper->registered) {
3000		dumper->registered = 0;
3001		list_del_rcu(&dumper->list);
3002		err = 0;
3003	}
3004	spin_unlock_irqrestore(&dump_list_lock, flags);
3005	synchronize_rcu();
3006
3007	return err;
3008}
3009EXPORT_SYMBOL_GPL(kmsg_dump_unregister);
3010
3011static bool always_kmsg_dump;
3012module_param_named(always_kmsg_dump, always_kmsg_dump, bool, S_IRUGO | S_IWUSR);
3013
3014/**
3015 * kmsg_dump - dump kernel log to kernel message dumpers.
3016 * @reason: the reason (oops, panic etc) for dumping
3017 *
3018 * Call each of the registered dumper's dump() callback, which can
3019 * retrieve the kmsg records with kmsg_dump_get_line() or
3020 * kmsg_dump_get_buffer().
3021 */
3022void kmsg_dump(enum kmsg_dump_reason reason)
3023{
3024	struct kmsg_dumper *dumper;
3025	unsigned long flags;
3026
3027	if ((reason > KMSG_DUMP_OOPS) && !always_kmsg_dump)
3028		return;
3029
3030	rcu_read_lock();
3031	list_for_each_entry_rcu(dumper, &dump_list, list) {
3032		if (dumper->max_reason && reason > dumper->max_reason)
3033			continue;
3034
3035		/* initialize iterator with data about the stored records */
3036		dumper->active = true;
3037
3038		logbuf_lock_irqsave(flags);
3039		dumper->cur_seq = clear_seq;
3040		dumper->cur_idx = clear_idx;
3041		dumper->next_seq = log_next_seq;
3042		dumper->next_idx = log_next_idx;
3043		logbuf_unlock_irqrestore(flags);
3044
3045		/* invoke dumper which will iterate over records */
3046		dumper->dump(dumper, reason);
3047
3048		/* reset iterator */
3049		dumper->active = false;
3050	}
3051	rcu_read_unlock();
3052}
3053
3054/**
3055 * kmsg_dump_get_line_nolock - retrieve one kmsg log line (unlocked version)
3056 * @dumper: registered kmsg dumper
3057 * @syslog: include the "<4>" prefixes
3058 * @line: buffer to copy the line to
3059 * @size: maximum size of the buffer
3060 * @len: length of line placed into buffer
3061 *
3062 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3063 * record, and copy one record into the provided buffer.
3064 *
3065 * Consecutive calls will return the next available record moving
3066 * towards the end of the buffer with the youngest messages.
3067 *
3068 * A return value of FALSE indicates that there are no more records to
3069 * read.
3070 *
3071 * The function is similar to kmsg_dump_get_line(), but grabs no locks.
3072 */
3073bool kmsg_dump_get_line_nolock(struct kmsg_dumper *dumper, bool syslog,
3074			       char *line, size_t size, size_t *len)
3075{
3076	struct printk_log *msg;
3077	size_t l = 0;
3078	bool ret = false;
3079
3080	if (!dumper->active)
3081		goto out;
3082
3083	if (dumper->cur_seq < log_first_seq) {
3084		/* messages are gone, move to first available one */
3085		dumper->cur_seq = log_first_seq;
3086		dumper->cur_idx = log_first_idx;
3087	}
3088
3089	/* last entry */
3090	if (dumper->cur_seq >= log_next_seq)
3091		goto out;
3092
3093	msg = log_from_idx(dumper->cur_idx);
3094	l = msg_print_text(msg, syslog, line, size);
3095
3096	dumper->cur_idx = log_next(dumper->cur_idx);
3097	dumper->cur_seq++;
3098	ret = true;
3099out:
3100	if (len)
3101		*len = l;
3102	return ret;
3103}
3104
3105/**
3106 * kmsg_dump_get_line - retrieve one kmsg log line
3107 * @dumper: registered kmsg dumper
3108 * @syslog: include the "<4>" prefixes
3109 * @line: buffer to copy the line to
3110 * @size: maximum size of the buffer
3111 * @len: length of line placed into buffer
3112 *
3113 * Start at the beginning of the kmsg buffer, with the oldest kmsg
3114 * record, and copy one record into the provided buffer.
3115 *
3116 * Consecutive calls will return the next available record moving
3117 * towards the end of the buffer with the youngest messages.
3118 *
3119 * A return value of FALSE indicates that there are no more records to
3120 * read.
3121 */
3122bool kmsg_dump_get_line(struct kmsg_dumper *dumper, bool syslog,
3123			char *line, size_t size, size_t *len)
3124{
3125	unsigned long flags;
3126	bool ret;
3127
3128	logbuf_lock_irqsave(flags);
3129	ret = kmsg_dump_get_line_nolock(dumper, syslog, line, size, len);
3130	logbuf_unlock_irqrestore(flags);
3131
3132	return ret;
3133}
3134EXPORT_SYMBOL_GPL(kmsg_dump_get_line);
3135
3136/**
3137 * kmsg_dump_get_buffer - copy kmsg log lines
3138 * @dumper: registered kmsg dumper
3139 * @syslog: include the "<4>" prefixes
3140 * @buf: buffer to copy the line to
3141 * @size: maximum size of the buffer
3142 * @len: length of line placed into buffer
3143 *
3144 * Start at the end of the kmsg buffer and fill the provided buffer
3145 * with as many of the the *youngest* kmsg records that fit into it.
3146 * If the buffer is large enough, all available kmsg records will be
3147 * copied with a single call.
3148 *
3149 * Consecutive calls will fill the buffer with the next block of
3150 * available older records, not including the earlier retrieved ones.
3151 *
3152 * A return value of FALSE indicates that there are no more records to
3153 * read.
3154 */
3155bool kmsg_dump_get_buffer(struct kmsg_dumper *dumper, bool syslog,
3156			  char *buf, size_t size, size_t *len)
3157{
3158	unsigned long flags;
3159	u64 seq;
3160	u32 idx;
3161	u64 next_seq;
3162	u32 next_idx;
3163	size_t l = 0;
3164	bool ret = false;
3165
3166	if (!dumper->active)
3167		goto out;
3168
3169	logbuf_lock_irqsave(flags);
3170	if (dumper->cur_seq < log_first_seq) {
3171		/* messages are gone, move to first available one */
3172		dumper->cur_seq = log_first_seq;
3173		dumper->cur_idx = log_first_idx;
3174	}
3175
3176	/* last entry */
3177	if (dumper->cur_seq >= dumper->next_seq) {
3178		logbuf_unlock_irqrestore(flags);
3179		goto out;
3180	}
3181
3182	/* calculate length of entire buffer */
3183	seq = dumper->cur_seq;
3184	idx = dumper->cur_idx;
3185	while (seq < dumper->next_seq) {
3186		struct printk_log *msg = log_from_idx(idx);
3187
3188		l += msg_print_text(msg, true, NULL, 0);
3189		idx = log_next(idx);
3190		seq++;
3191	}
3192
3193	/* move first record forward until length fits into the buffer */
3194	seq = dumper->cur_seq;
3195	idx = dumper->cur_idx;
3196	while (l > size && seq < dumper->next_seq) {
3197		struct printk_log *msg = log_from_idx(idx);
3198
3199		l -= msg_print_text(msg, true, NULL, 0);
3200		idx = log_next(idx);
3201		seq++;
3202	}
3203
3204	/* last message in next interation */
3205	next_seq = seq;
3206	next_idx = idx;
3207
3208	l = 0;
3209	while (seq < dumper->next_seq) {
3210		struct printk_log *msg = log_from_idx(idx);
3211
3212		l += msg_print_text(msg, syslog, buf + l, size - l);
3213		idx = log_next(idx);
3214		seq++;
3215	}
3216
3217	dumper->next_seq = next_seq;
3218	dumper->next_idx = next_idx;
3219	ret = true;
3220	logbuf_unlock_irqrestore(flags);
3221out:
3222	if (len)
3223		*len = l;
3224	return ret;
3225}
3226EXPORT_SYMBOL_GPL(kmsg_dump_get_buffer);
3227
3228/**
3229 * kmsg_dump_rewind_nolock - reset the interator (unlocked version)
3230 * @dumper: registered kmsg dumper
3231 *
3232 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3233 * kmsg_dump_get_buffer() can be called again and used multiple
3234 * times within the same dumper.dump() callback.
3235 *
3236 * The function is similar to kmsg_dump_rewind(), but grabs no locks.
3237 */
3238void kmsg_dump_rewind_nolock(struct kmsg_dumper *dumper)
3239{
3240	dumper->cur_seq = clear_seq;
3241	dumper->cur_idx = clear_idx;
3242	dumper->next_seq = log_next_seq;
3243	dumper->next_idx = log_next_idx;
3244}
3245
3246/**
3247 * kmsg_dump_rewind - reset the interator
3248 * @dumper: registered kmsg dumper
3249 *
3250 * Reset the dumper's iterator so that kmsg_dump_get_line() and
3251 * kmsg_dump_get_buffer() can be called again and used multiple
3252 * times within the same dumper.dump() callback.
3253 */
3254void kmsg_dump_rewind(struct kmsg_dumper *dumper)
3255{
3256	unsigned long flags;
3257
3258	logbuf_lock_irqsave(flags);
3259	kmsg_dump_rewind_nolock(dumper);
3260	logbuf_unlock_irqrestore(flags);
3261}
3262EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3263
3264#endif