Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4 *
   5 */
   6
   7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9#include <linux/kernel.h>
  10#include <linux/blkdev.h>
  11#include <linux/blktrace_api.h>
  12#include <linux/percpu.h>
  13#include <linux/init.h>
  14#include <linux/mutex.h>
  15#include <linux/slab.h>
  16#include <linux/debugfs.h>
  17#include <linux/export.h>
  18#include <linux/time.h>
  19#include <linux/uaccess.h>
  20#include <linux/list.h>
  21#include <linux/blk-cgroup.h>
  22
  23#include "../../block/blk.h"
  24
  25#include <trace/events/block.h>
  26
  27#include "trace_output.h"
  28
  29#ifdef CONFIG_BLK_DEV_IO_TRACE
  30
  31static unsigned int blktrace_seq __read_mostly = 1;
  32
  33static struct trace_array *blk_tr;
  34static bool blk_tracer_enabled __read_mostly;
  35
  36static LIST_HEAD(running_trace_list);
  37static __cacheline_aligned_in_smp DEFINE_RAW_SPINLOCK(running_trace_lock);
  38
  39/* Select an alternative, minimalistic output than the original one */
  40#define TRACE_BLK_OPT_CLASSIC	0x1
  41#define TRACE_BLK_OPT_CGROUP	0x2
  42#define TRACE_BLK_OPT_CGNAME	0x4
  43
  44static struct tracer_opt blk_tracer_opts[] = {
  45	/* Default disable the minimalistic output */
  46	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47#ifdef CONFIG_BLK_CGROUP
  48	{ TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  49	{ TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  50#endif
  51	{ }
  52};
  53
  54static struct tracer_flags blk_tracer_flags = {
  55	.val  = 0,
  56	.opts = blk_tracer_opts,
  57};
  58
  59/* Global reference count of probes */
  60static DEFINE_MUTEX(blk_probe_mutex);
  61static int blk_probes_ref;
  62
  63static void blk_register_tracepoints(void);
  64static void blk_unregister_tracepoints(void);
  65
  66/*
  67 * Send out a notify message.
  68 */
  69static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  70		       const void *data, size_t len, u64 cgid)
 
  71{
  72	struct blk_io_trace *t;
  73	struct ring_buffer_event *event = NULL;
  74	struct trace_buffer *buffer = NULL;
  75	unsigned int trace_ctx = 0;
  76	int cpu = smp_processor_id();
  77	bool blk_tracer = blk_tracer_enabled;
  78	ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
  79
  80	if (blk_tracer) {
  81		buffer = blk_tr->array_buffer.buffer;
  82		trace_ctx = tracing_gen_ctx_flags(0);
  83		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84						  sizeof(*t) + len + cgid_len,
  85						  trace_ctx);
  86		if (!event)
  87			return;
  88		t = ring_buffer_event_data(event);
  89		goto record_it;
  90	}
  91
  92	if (!bt->rchan)
  93		return;
  94
  95	t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  96	if (t) {
  97		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  98		t->time = ktime_to_ns(ktime_get());
  99record_it:
 100		t->device = bt->dev;
 101		t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 102		t->pid = pid;
 103		t->cpu = cpu;
 104		t->pdu_len = len + cgid_len;
 105		if (cgid_len)
 106			memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 107		memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 108
 109		if (blk_tracer)
 110			trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx);
 111	}
 112}
 113
 114/*
 115 * Send out a notify for this process, if we haven't done so since a trace
 116 * started
 117 */
 118static void trace_note_tsk(struct task_struct *tsk)
 119{
 120	unsigned long flags;
 121	struct blk_trace *bt;
 122
 123	tsk->btrace_seq = blktrace_seq;
 124	raw_spin_lock_irqsave(&running_trace_lock, flags);
 125	list_for_each_entry(bt, &running_trace_list, running_list) {
 126		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 127			   sizeof(tsk->comm), 0);
 128	}
 129	raw_spin_unlock_irqrestore(&running_trace_lock, flags);
 130}
 131
 132static void trace_note_time(struct blk_trace *bt)
 133{
 134	struct timespec64 now;
 135	unsigned long flags;
 136	u32 words[2];
 137
 138	/* need to check user space to see if this breaks in y2038 or y2106 */
 139	ktime_get_real_ts64(&now);
 140	words[0] = (u32)now.tv_sec;
 141	words[1] = now.tv_nsec;
 142
 143	local_irq_save(flags);
 144	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
 145	local_irq_restore(flags);
 146}
 147
 148void __blk_trace_note_message(struct blk_trace *bt,
 149		struct cgroup_subsys_state *css, const char *fmt, ...)
 150{
 151	int n;
 152	va_list args;
 153	unsigned long flags;
 154	char *buf;
 155	u64 cgid = 0;
 156
 157	if (unlikely(bt->trace_state != Blktrace_running &&
 158		     !blk_tracer_enabled))
 159		return;
 160
 161	/*
 162	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 163	 * message to the trace.
 164	 */
 165	if (!(bt->act_mask & BLK_TC_NOTIFY))
 166		return;
 167
 168	local_irq_save(flags);
 169	buf = this_cpu_ptr(bt->msg_data);
 170	va_start(args, fmt);
 171	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 172	va_end(args);
 173
 
 
 174#ifdef CONFIG_BLK_CGROUP
 175	if (css && (blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 176		cgid = cgroup_id(css->cgroup);
 177	else
 178		cgid = 1;
 179#endif
 180	trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, cgid);
 181	local_irq_restore(flags);
 182}
 183EXPORT_SYMBOL_GPL(__blk_trace_note_message);
 184
 185static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 186			 pid_t pid)
 187{
 188	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 189		return 1;
 190	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 191		return 1;
 192	if (bt->pid && pid != bt->pid)
 193		return 1;
 194
 195	return 0;
 196}
 197
 198/*
 199 * Data direction bit lookup
 200 */
 201static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 202				 BLK_TC_ACT(BLK_TC_WRITE) };
 203
 204#define BLK_TC_RAHEAD		BLK_TC_AHEAD
 205#define BLK_TC_PREFLUSH		BLK_TC_FLUSH
 206
 207/* The ilog2() calls fall out because they're constant */
 208#define MASK_TC_BIT(rw, __name) ((__force u32)(rw & REQ_ ## __name) <<	\
 209	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 210
 211/*
 212 * The worker for the various blk_add_trace*() types. Fills out a
 213 * blk_io_trace structure and places it in a per-cpu subbuffer.
 214 */
 215static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 216			    const blk_opf_t opf, u32 what, int error,
 217			    int pdu_len, void *pdu_data, u64 cgid)
 218{
 219	struct task_struct *tsk = current;
 220	struct ring_buffer_event *event = NULL;
 221	struct trace_buffer *buffer = NULL;
 222	struct blk_io_trace *t;
 223	unsigned long flags = 0;
 224	unsigned long *sequence;
 225	unsigned int trace_ctx = 0;
 226	pid_t pid;
 227	int cpu;
 228	bool blk_tracer = blk_tracer_enabled;
 229	ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
 230	const enum req_op op = opf & REQ_OP_MASK;
 231
 232	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 233		return;
 234
 235	what |= ddir_act[op_is_write(op) ? WRITE : READ];
 236	what |= MASK_TC_BIT(opf, SYNC);
 237	what |= MASK_TC_BIT(opf, RAHEAD);
 238	what |= MASK_TC_BIT(opf, META);
 239	what |= MASK_TC_BIT(opf, PREFLUSH);
 240	what |= MASK_TC_BIT(opf, FUA);
 241	if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 242		what |= BLK_TC_ACT(BLK_TC_DISCARD);
 243	if (op == REQ_OP_FLUSH)
 244		what |= BLK_TC_ACT(BLK_TC_FLUSH);
 245	if (cgid)
 246		what |= __BLK_TA_CGROUP;
 247
 248	pid = tsk->pid;
 249	if (act_log_check(bt, what, sector, pid))
 250		return;
 251	cpu = raw_smp_processor_id();
 252
 253	if (blk_tracer) {
 254		tracing_record_cmdline(current);
 255
 256		buffer = blk_tr->array_buffer.buffer;
 257		trace_ctx = tracing_gen_ctx_flags(0);
 258		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 259						  sizeof(*t) + pdu_len + cgid_len,
 260						  trace_ctx);
 261		if (!event)
 262			return;
 263		t = ring_buffer_event_data(event);
 264		goto record_it;
 265	}
 266
 267	if (unlikely(tsk->btrace_seq != blktrace_seq))
 268		trace_note_tsk(tsk);
 269
 270	/*
 271	 * A word about the locking here - we disable interrupts to reserve
 272	 * some space in the relay per-cpu buffer, to prevent an irq
 273	 * from coming in and stepping on our toes.
 274	 */
 275	local_irq_save(flags);
 276	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 277	if (t) {
 278		sequence = per_cpu_ptr(bt->sequence, cpu);
 279
 280		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 281		t->sequence = ++(*sequence);
 282		t->time = ktime_to_ns(ktime_get());
 283record_it:
 284		/*
 285		 * These two are not needed in ftrace as they are in the
 286		 * generic trace_entry, filled by tracing_generic_entry_update,
 287		 * but for the trace_event->bin() synthesizer benefit we do it
 288		 * here too.
 289		 */
 290		t->cpu = cpu;
 291		t->pid = pid;
 292
 293		t->sector = sector;
 294		t->bytes = bytes;
 295		t->action = what;
 296		t->device = bt->dev;
 297		t->error = error;
 298		t->pdu_len = pdu_len + cgid_len;
 299
 300		if (cgid_len)
 301			memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 302		if (pdu_len)
 303			memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 304
 305		if (blk_tracer) {
 306			trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx);
 307			return;
 308		}
 309	}
 310
 311	local_irq_restore(flags);
 312}
 313
 314static void blk_trace_free(struct request_queue *q, struct blk_trace *bt)
 315{
 
 
 316	relay_close(bt->rchan);
 317
 318	/*
 319	 * If 'bt->dir' is not set, then both 'dropped' and 'msg' are created
 320	 * under 'q->debugfs_dir', thus lookup and remove them.
 321	 */
 322	if (!bt->dir) {
 323		debugfs_remove(debugfs_lookup("dropped", q->debugfs_dir));
 324		debugfs_remove(debugfs_lookup("msg", q->debugfs_dir));
 325	} else {
 326		debugfs_remove(bt->dir);
 327	}
 328	free_percpu(bt->sequence);
 329	free_percpu(bt->msg_data);
 330	kfree(bt);
 331}
 332
 333static void get_probe_ref(void)
 334{
 335	mutex_lock(&blk_probe_mutex);
 336	if (++blk_probes_ref == 1)
 337		blk_register_tracepoints();
 338	mutex_unlock(&blk_probe_mutex);
 339}
 340
 341static void put_probe_ref(void)
 342{
 343	mutex_lock(&blk_probe_mutex);
 344	if (!--blk_probes_ref)
 345		blk_unregister_tracepoints();
 346	mutex_unlock(&blk_probe_mutex);
 347}
 348
 349static int blk_trace_start(struct blk_trace *bt)
 350{
 351	if (bt->trace_state != Blktrace_setup &&
 352	    bt->trace_state != Blktrace_stopped)
 353		return -EINVAL;
 354
 355	blktrace_seq++;
 356	smp_mb();
 357	bt->trace_state = Blktrace_running;
 358	raw_spin_lock_irq(&running_trace_lock);
 359	list_add(&bt->running_list, &running_trace_list);
 360	raw_spin_unlock_irq(&running_trace_lock);
 361	trace_note_time(bt);
 362
 363	return 0;
 364}
 365
 366static int blk_trace_stop(struct blk_trace *bt)
 367{
 368	if (bt->trace_state != Blktrace_running)
 369		return -EINVAL;
 370
 371	bt->trace_state = Blktrace_stopped;
 372	raw_spin_lock_irq(&running_trace_lock);
 373	list_del_init(&bt->running_list);
 374	raw_spin_unlock_irq(&running_trace_lock);
 375	relay_flush(bt->rchan);
 376
 377	return 0;
 378}
 379
 380static void blk_trace_cleanup(struct request_queue *q, struct blk_trace *bt)
 381{
 382	blk_trace_stop(bt);
 383	synchronize_rcu();
 384	blk_trace_free(q, bt);
 385	put_probe_ref();
 386}
 387
 388static int __blk_trace_remove(struct request_queue *q)
 389{
 390	struct blk_trace *bt;
 391
 392	bt = rcu_replace_pointer(q->blk_trace, NULL,
 393				 lockdep_is_held(&q->debugfs_mutex));
 394	if (!bt)
 395		return -EINVAL;
 396
 397	blk_trace_cleanup(q, bt);
 
 398
 399	return 0;
 400}
 401
 402int blk_trace_remove(struct request_queue *q)
 403{
 404	int ret;
 405
 406	mutex_lock(&q->debugfs_mutex);
 407	ret = __blk_trace_remove(q);
 408	mutex_unlock(&q->debugfs_mutex);
 409
 410	return ret;
 411}
 412EXPORT_SYMBOL_GPL(blk_trace_remove);
 413
 414static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 415				size_t count, loff_t *ppos)
 416{
 417	struct blk_trace *bt = filp->private_data;
 418	char buf[16];
 419
 420	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 421
 422	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 423}
 424
 425static const struct file_operations blk_dropped_fops = {
 426	.owner =	THIS_MODULE,
 427	.open =		simple_open,
 428	.read =		blk_dropped_read,
 429	.llseek =	default_llseek,
 430};
 431
 432static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 433				size_t count, loff_t *ppos)
 434{
 435	char *msg;
 436	struct blk_trace *bt;
 437
 438	if (count >= BLK_TN_MAX_MSG)
 439		return -EINVAL;
 440
 441	msg = memdup_user_nul(buffer, count);
 442	if (IS_ERR(msg))
 443		return PTR_ERR(msg);
 444
 445	bt = filp->private_data;
 446	__blk_trace_note_message(bt, NULL, "%s", msg);
 447	kfree(msg);
 448
 449	return count;
 450}
 451
 452static const struct file_operations blk_msg_fops = {
 453	.owner =	THIS_MODULE,
 454	.open =		simple_open,
 455	.write =	blk_msg_write,
 456	.llseek =	noop_llseek,
 457};
 458
 459/*
 460 * Keep track of how many times we encountered a full subbuffer, to aid
 461 * the user space app in telling how many lost events there were.
 462 */
 463static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 464				     void *prev_subbuf, size_t prev_padding)
 465{
 466	struct blk_trace *bt;
 467
 468	if (!relay_buf_full(buf))
 469		return 1;
 470
 471	bt = buf->chan->private_data;
 472	atomic_inc(&bt->dropped);
 473	return 0;
 474}
 475
 476static int blk_remove_buf_file_callback(struct dentry *dentry)
 477{
 478	debugfs_remove(dentry);
 479
 480	return 0;
 481}
 482
 483static struct dentry *blk_create_buf_file_callback(const char *filename,
 484						   struct dentry *parent,
 485						   umode_t mode,
 486						   struct rchan_buf *buf,
 487						   int *is_global)
 488{
 489	return debugfs_create_file(filename, mode, parent, buf,
 490					&relay_file_operations);
 491}
 492
 493static const struct rchan_callbacks blk_relay_callbacks = {
 494	.subbuf_start		= blk_subbuf_start_callback,
 495	.create_buf_file	= blk_create_buf_file_callback,
 496	.remove_buf_file	= blk_remove_buf_file_callback,
 497};
 498
 499static void blk_trace_setup_lba(struct blk_trace *bt,
 500				struct block_device *bdev)
 501{
 502	if (bdev) {
 503		bt->start_lba = bdev->bd_start_sect;
 504		bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
 
 
 
 
 
 505	} else {
 506		bt->start_lba = 0;
 507		bt->end_lba = -1ULL;
 508	}
 509}
 510
 511/*
 512 * Setup everything required to start tracing
 513 */
 514static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 515			      struct block_device *bdev,
 516			      struct blk_user_trace_setup *buts)
 517{
 518	struct blk_trace *bt = NULL;
 519	struct dentry *dir = NULL;
 520	int ret;
 521
 522	lockdep_assert_held(&q->debugfs_mutex);
 523
 524	if (!buts->buf_size || !buts->buf_nr)
 525		return -EINVAL;
 526
 
 
 
 527	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 528	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 529
 530	/*
 531	 * some device names have larger paths - convert the slashes
 532	 * to underscores for this to work as expected
 533	 */
 534	strreplace(buts->name, '/', '_');
 535
 536	/*
 537	 * bdev can be NULL, as with scsi-generic, this is a helpful as
 538	 * we can be.
 539	 */
 540	if (rcu_dereference_protected(q->blk_trace,
 541				      lockdep_is_held(&q->debugfs_mutex))) {
 542		pr_warn("Concurrent blktraces are not allowed on %s\n",
 543			buts->name);
 544		return -EBUSY;
 545	}
 546
 547	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 548	if (!bt)
 549		return -ENOMEM;
 550
 551	ret = -ENOMEM;
 552	bt->sequence = alloc_percpu(unsigned long);
 553	if (!bt->sequence)
 554		goto err;
 555
 556	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 557	if (!bt->msg_data)
 558		goto err;
 559
 560	/*
 561	 * When tracing the whole disk reuse the existing debugfs directory
 562	 * created by the block layer on init. For partitions block devices,
 563	 * and scsi-generic block devices we create a temporary new debugfs
 564	 * directory that will be removed once the trace ends.
 565	 */
 566	if (bdev && !bdev_is_partition(bdev))
 567		dir = q->debugfs_dir;
 568	else
 569		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 570
 571	/*
 572	 * As blktrace relies on debugfs for its interface the debugfs directory
 573	 * is required, contrary to the usual mantra of not checking for debugfs
 574	 * files or directories.
 575	 */
 576	if (IS_ERR_OR_NULL(dir)) {
 577		pr_warn("debugfs_dir not present for %s so skipping\n",
 578			buts->name);
 579		ret = -ENOENT;
 580		goto err;
 581	}
 582
 583	bt->dev = dev;
 584	atomic_set(&bt->dropped, 0);
 585	INIT_LIST_HEAD(&bt->running_list);
 586
 587	ret = -EIO;
 588	debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 589	debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 
 
 590
 591	bt->rchan = relay_open("trace", dir, buts->buf_size,
 592				buts->buf_nr, &blk_relay_callbacks, bt);
 593	if (!bt->rchan)
 594		goto err;
 595
 596	bt->act_mask = buts->act_mask;
 597	if (!bt->act_mask)
 598		bt->act_mask = (u16) -1;
 599
 600	blk_trace_setup_lba(bt, bdev);
 601
 602	/* overwrite with user settings */
 603	if (buts->start_lba)
 604		bt->start_lba = buts->start_lba;
 605	if (buts->end_lba)
 606		bt->end_lba = buts->end_lba;
 607
 608	bt->pid = buts->pid;
 609	bt->trace_state = Blktrace_setup;
 610
 611	rcu_assign_pointer(q->blk_trace, bt);
 
 
 
 612	get_probe_ref();
 613
 614	ret = 0;
 615err:
 
 
 616	if (ret)
 617		blk_trace_free(q, bt);
 618	return ret;
 619}
 620
 621static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 622			     struct block_device *bdev, char __user *arg)
 623{
 624	struct blk_user_trace_setup buts;
 625	int ret;
 626
 627	ret = copy_from_user(&buts, arg, sizeof(buts));
 628	if (ret)
 629		return -EFAULT;
 630
 631	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 632	if (ret)
 633		return ret;
 634
 635	if (copy_to_user(arg, &buts, sizeof(buts))) {
 636		__blk_trace_remove(q);
 637		return -EFAULT;
 638	}
 639	return 0;
 640}
 641
 642int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 643		    struct block_device *bdev,
 644		    char __user *arg)
 645{
 646	int ret;
 647
 648	mutex_lock(&q->debugfs_mutex);
 649	ret = __blk_trace_setup(q, name, dev, bdev, arg);
 650	mutex_unlock(&q->debugfs_mutex);
 651
 652	return ret;
 653}
 654EXPORT_SYMBOL_GPL(blk_trace_setup);
 655
 656#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 657static int compat_blk_trace_setup(struct request_queue *q, char *name,
 658				  dev_t dev, struct block_device *bdev,
 659				  char __user *arg)
 660{
 661	struct blk_user_trace_setup buts;
 662	struct compat_blk_user_trace_setup cbuts;
 663	int ret;
 664
 665	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 666		return -EFAULT;
 667
 668	buts = (struct blk_user_trace_setup) {
 669		.act_mask = cbuts.act_mask,
 670		.buf_size = cbuts.buf_size,
 671		.buf_nr = cbuts.buf_nr,
 672		.start_lba = cbuts.start_lba,
 673		.end_lba = cbuts.end_lba,
 674		.pid = cbuts.pid,
 675	};
 676
 677	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 678	if (ret)
 679		return ret;
 680
 681	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 682		__blk_trace_remove(q);
 683		return -EFAULT;
 684	}
 685
 686	return 0;
 687}
 688#endif
 689
 690static int __blk_trace_startstop(struct request_queue *q, int start)
 691{
 692	struct blk_trace *bt;
 
 693
 694	bt = rcu_dereference_protected(q->blk_trace,
 695				       lockdep_is_held(&q->debugfs_mutex));
 696	if (bt == NULL)
 697		return -EINVAL;
 698
 699	if (start)
 700		return blk_trace_start(bt);
 701	else
 702		return blk_trace_stop(bt);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 703}
 704
 705int blk_trace_startstop(struct request_queue *q, int start)
 706{
 707	int ret;
 708
 709	mutex_lock(&q->debugfs_mutex);
 710	ret = __blk_trace_startstop(q, start);
 711	mutex_unlock(&q->debugfs_mutex);
 712
 713	return ret;
 714}
 715EXPORT_SYMBOL_GPL(blk_trace_startstop);
 716
 717/*
 718 * When reading or writing the blktrace sysfs files, the references to the
 719 * opened sysfs or device files should prevent the underlying block device
 720 * from being removed. So no further delete protection is really needed.
 721 */
 722
 723/**
 724 * blk_trace_ioctl - handle the ioctls associated with tracing
 725 * @bdev:	the block device
 726 * @cmd:	the ioctl cmd
 727 * @arg:	the argument data, if any
 728 *
 729 **/
 730int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 731{
 732	struct request_queue *q;
 733	int ret, start = 0;
 734	char b[BDEVNAME_SIZE];
 735
 736	q = bdev_get_queue(bdev);
 737	if (!q)
 738		return -ENXIO;
 739
 740	mutex_lock(&q->debugfs_mutex);
 741
 742	switch (cmd) {
 743	case BLKTRACESETUP:
 744		snprintf(b, sizeof(b), "%pg", bdev);
 745		ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 746		break;
 747#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 748	case BLKTRACESETUP32:
 749		snprintf(b, sizeof(b), "%pg", bdev);
 750		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 751		break;
 752#endif
 753	case BLKTRACESTART:
 754		start = 1;
 755		fallthrough;
 756	case BLKTRACESTOP:
 757		ret = __blk_trace_startstop(q, start);
 758		break;
 759	case BLKTRACETEARDOWN:
 760		ret = __blk_trace_remove(q);
 761		break;
 762	default:
 763		ret = -ENOTTY;
 764		break;
 765	}
 766
 767	mutex_unlock(&q->debugfs_mutex);
 768	return ret;
 769}
 770
 771/**
 772 * blk_trace_shutdown - stop and cleanup trace structures
 773 * @q:    the request queue associated with the device
 774 *
 775 **/
 776void blk_trace_shutdown(struct request_queue *q)
 777{
 778	if (rcu_dereference_protected(q->blk_trace,
 779				      lockdep_is_held(&q->debugfs_mutex)))
 
 
 780		__blk_trace_remove(q);
 
 
 
 781}
 782
 783#ifdef CONFIG_BLK_CGROUP
 784static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 
 785{
 786	struct cgroup_subsys_state *blkcg_css;
 787	struct blk_trace *bt;
 788
 789	/* We don't use the 'bt' value here except as an optimization... */
 790	bt = rcu_dereference_protected(q->blk_trace, 1);
 791	if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 792		return 0;
 793
 794	blkcg_css = bio_blkcg_css(bio);
 795	if (!blkcg_css)
 796		return 0;
 797	return cgroup_id(blkcg_css->cgroup);
 798}
 799#else
 800static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 
 801{
 802	return 0;
 803}
 804#endif
 805
 806static u64
 807blk_trace_request_get_cgid(struct request *rq)
 808{
 809	if (!rq->bio)
 810		return 0;
 811	/* Use the first bio */
 812	return blk_trace_bio_get_cgid(rq->q, rq->bio);
 813}
 814
 815/*
 816 * blktrace probes
 817 */
 818
 819/**
 820 * blk_add_trace_rq - Add a trace for a request oriented action
 821 * @rq:		the source request
 822 * @error:	return status to log
 823 * @nr_bytes:	number of completed bytes
 824 * @what:	the action
 825 * @cgid:	the cgroup info
 826 *
 827 * Description:
 828 *     Records an action against a request. Will log the bio offset + size.
 829 *
 830 **/
 831static void blk_add_trace_rq(struct request *rq, blk_status_t error,
 832			     unsigned int nr_bytes, u32 what, u64 cgid)
 
 833{
 834	struct blk_trace *bt;
 835
 836	rcu_read_lock();
 837	bt = rcu_dereference(rq->q->blk_trace);
 838	if (likely(!bt)) {
 839		rcu_read_unlock();
 840		return;
 841	}
 842
 843	if (blk_rq_is_passthrough(rq))
 844		what |= BLK_TC_ACT(BLK_TC_PC);
 845	else
 846		what |= BLK_TC_ACT(BLK_TC_FS);
 847
 848	__blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, rq->cmd_flags,
 849			what, blk_status_to_errno(error), 0, NULL, cgid);
 850	rcu_read_unlock();
 851}
 852
 853static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
 
 854{
 855	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 856			 blk_trace_request_get_cgid(rq));
 857}
 858
 859static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
 
 860{
 861	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 862			 blk_trace_request_get_cgid(rq));
 863}
 864
 865static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
 866{
 867	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
 868			 blk_trace_request_get_cgid(rq));
 869}
 870
 871static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
 872{
 873	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 874			 blk_trace_request_get_cgid(rq));
 875}
 876
 877static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 878			blk_status_t error, unsigned int nr_bytes)
 879{
 880	blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 881			 blk_trace_request_get_cgid(rq));
 882}
 883
 884/**
 885 * blk_add_trace_bio - Add a trace for a bio oriented action
 886 * @q:		queue the io is for
 887 * @bio:	the source bio
 888 * @what:	the action
 889 * @error:	error, if any
 890 *
 891 * Description:
 892 *     Records an action against a bio. Will log the bio offset + size.
 893 *
 894 **/
 895static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 896			      u32 what, int error)
 897{
 898	struct blk_trace *bt;
 899
 900	rcu_read_lock();
 901	bt = rcu_dereference(q->blk_trace);
 902	if (likely(!bt)) {
 903		rcu_read_unlock();
 904		return;
 905	}
 906
 907	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 908			bio->bi_opf, what, error, 0, NULL,
 909			blk_trace_bio_get_cgid(q, bio));
 910	rcu_read_unlock();
 911}
 912
 913static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio)
 
 914{
 915	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0);
 916}
 917
 918static void blk_add_trace_bio_complete(void *ignore,
 919				       struct request_queue *q, struct bio *bio)
 
 920{
 921	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
 922			  blk_status_to_errno(bio->bi_status));
 923}
 924
 925static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio)
 
 
 
 926{
 927	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE,
 928			0);
 929}
 930
 931static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio)
 
 
 
 932{
 933	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE,
 934			0);
 935}
 936
 937static void blk_add_trace_bio_queue(void *ignore, struct bio *bio)
 
 938{
 939	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0);
 940}
 941
 942static void blk_add_trace_getrq(void *ignore, struct bio *bio)
 
 
 943{
 944	blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 945}
 946
 947static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 948{
 949	struct blk_trace *bt;
 950
 951	rcu_read_lock();
 952	bt = rcu_dereference(q->blk_trace);
 953	if (bt)
 954		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 955	rcu_read_unlock();
 956}
 957
 958static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 959				    unsigned int depth, bool explicit)
 960{
 961	struct blk_trace *bt;
 962
 963	rcu_read_lock();
 964	bt = rcu_dereference(q->blk_trace);
 965	if (bt) {
 966		__be64 rpdu = cpu_to_be64(depth);
 967		u32 what;
 968
 969		if (explicit)
 970			what = BLK_TA_UNPLUG_IO;
 971		else
 972			what = BLK_TA_UNPLUG_TIMER;
 973
 974		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
 975	}
 976	rcu_read_unlock();
 977}
 978
 979static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu)
 
 
 980{
 981	struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 982	struct blk_trace *bt;
 983
 984	rcu_read_lock();
 985	bt = rcu_dereference(q->blk_trace);
 986	if (bt) {
 987		__be64 rpdu = cpu_to_be64(pdu);
 988
 989		__blk_add_trace(bt, bio->bi_iter.bi_sector,
 990				bio->bi_iter.bi_size, bio->bi_opf, BLK_TA_SPLIT,
 991				blk_status_to_errno(bio->bi_status),
 992				sizeof(rpdu), &rpdu,
 993				blk_trace_bio_get_cgid(q, bio));
 994	}
 995	rcu_read_unlock();
 996}
 997
 998/**
 999 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1000 * @ignore:	trace callback data parameter (not used)
 
1001 * @bio:	the source bio
1002 * @dev:	source device
1003 * @from:	source sector
1004 *
1005 * Called after a bio is remapped to a different device and/or sector.
 
 
 
1006 **/
1007static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev,
1008				    sector_t from)
 
1009{
1010	struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1011	struct blk_trace *bt;
1012	struct blk_io_trace_remap r;
1013
1014	rcu_read_lock();
1015	bt = rcu_dereference(q->blk_trace);
1016	if (likely(!bt)) {
1017		rcu_read_unlock();
1018		return;
1019	}
1020
1021	r.device_from = cpu_to_be32(dev);
1022	r.device_to   = cpu_to_be32(bio_dev(bio));
1023	r.sector_from = cpu_to_be64(from);
1024
1025	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1026			bio->bi_opf, BLK_TA_REMAP,
1027			blk_status_to_errno(bio->bi_status),
1028			sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1029	rcu_read_unlock();
1030}
1031
1032/**
1033 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1034 * @ignore:	trace callback data parameter (not used)
 
1035 * @rq:		the source request
1036 * @dev:	target device
1037 * @from:	source sector
1038 *
1039 * Description:
1040 *     Device mapper remaps request to other devices.
1041 *     Add a trace for that action.
1042 *
1043 **/
1044static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
 
 
1045				   sector_t from)
1046{
1047	struct blk_trace *bt;
1048	struct blk_io_trace_remap r;
1049
1050	rcu_read_lock();
1051	bt = rcu_dereference(rq->q->blk_trace);
1052	if (likely(!bt)) {
1053		rcu_read_unlock();
1054		return;
1055	}
1056
1057	r.device_from = cpu_to_be32(dev);
1058	r.device_to   = cpu_to_be32(disk_devt(rq->q->disk));
1059	r.sector_from = cpu_to_be64(from);
1060
1061	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1062			rq->cmd_flags, BLK_TA_REMAP, 0,
1063			sizeof(r), &r, blk_trace_request_get_cgid(rq));
1064	rcu_read_unlock();
1065}
1066
1067/**
1068 * blk_add_driver_data - Add binary message with driver-specific data
 
1069 * @rq:		io request
1070 * @data:	driver-specific data
1071 * @len:	length of driver-specific data
1072 *
1073 * Description:
1074 *     Some drivers might want to write driver-specific data per request.
1075 *
1076 **/
1077void blk_add_driver_data(struct request *rq, void *data, size_t len)
 
 
1078{
1079	struct blk_trace *bt;
1080
1081	rcu_read_lock();
1082	bt = rcu_dereference(rq->q->blk_trace);
1083	if (likely(!bt)) {
1084		rcu_read_unlock();
1085		return;
1086	}
1087
1088	__blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0,
1089				BLK_TA_DRV_DATA, 0, len, data,
1090				blk_trace_request_get_cgid(rq));
1091	rcu_read_unlock();
1092}
1093EXPORT_SYMBOL_GPL(blk_add_driver_data);
1094
1095static void blk_register_tracepoints(void)
1096{
1097	int ret;
1098
1099	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1100	WARN_ON(ret);
1101	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1102	WARN_ON(ret);
1103	ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1104	WARN_ON(ret);
1105	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1106	WARN_ON(ret);
1107	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1108	WARN_ON(ret);
1109	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1110	WARN_ON(ret);
1111	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1112	WARN_ON(ret);
1113	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1114	WARN_ON(ret);
1115	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1116	WARN_ON(ret);
1117	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1118	WARN_ON(ret);
1119	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1120	WARN_ON(ret);
 
 
1121	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1122	WARN_ON(ret);
1123	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1124	WARN_ON(ret);
1125	ret = register_trace_block_split(blk_add_trace_split, NULL);
1126	WARN_ON(ret);
1127	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1128	WARN_ON(ret);
1129	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1130	WARN_ON(ret);
1131}
1132
1133static void blk_unregister_tracepoints(void)
1134{
1135	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1136	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1137	unregister_trace_block_split(blk_add_trace_split, NULL);
1138	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1139	unregister_trace_block_plug(blk_add_trace_plug, NULL);
 
1140	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1141	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1142	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1143	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1144	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1145	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1146	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1147	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1148	unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1149	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1150	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1151
1152	tracepoint_synchronize_unregister();
1153}
1154
1155/*
1156 * struct blk_io_tracer formatting routines
1157 */
1158
1159static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1160{
1161	int i = 0;
1162	int tc = t->action >> BLK_TC_SHIFT;
1163
1164	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1165		rwbs[i++] = 'N';
1166		goto out;
1167	}
1168
1169	if (tc & BLK_TC_FLUSH)
1170		rwbs[i++] = 'F';
1171
1172	if (tc & BLK_TC_DISCARD)
1173		rwbs[i++] = 'D';
1174	else if (tc & BLK_TC_WRITE)
1175		rwbs[i++] = 'W';
1176	else if (t->bytes)
1177		rwbs[i++] = 'R';
1178	else
1179		rwbs[i++] = 'N';
1180
1181	if (tc & BLK_TC_FUA)
1182		rwbs[i++] = 'F';
1183	if (tc & BLK_TC_AHEAD)
1184		rwbs[i++] = 'A';
1185	if (tc & BLK_TC_SYNC)
1186		rwbs[i++] = 'S';
1187	if (tc & BLK_TC_META)
1188		rwbs[i++] = 'M';
1189out:
1190	rwbs[i] = '\0';
1191}
1192
1193static inline
1194const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1195{
1196	return (const struct blk_io_trace *)ent;
1197}
1198
1199static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1200{
1201	return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
 
1202}
1203
1204static inline u64 t_cgid(const struct trace_entry *ent)
1205{
1206	return *(u64 *)(te_blk_io_trace(ent) + 1);
1207}
1208
1209static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1210{
1211	return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
 
1212}
1213
1214static inline u32 t_action(const struct trace_entry *ent)
1215{
1216	return te_blk_io_trace(ent)->action;
1217}
1218
1219static inline u32 t_bytes(const struct trace_entry *ent)
1220{
1221	return te_blk_io_trace(ent)->bytes;
1222}
1223
1224static inline u32 t_sec(const struct trace_entry *ent)
1225{
1226	return te_blk_io_trace(ent)->bytes >> 9;
1227}
1228
1229static inline unsigned long long t_sector(const struct trace_entry *ent)
1230{
1231	return te_blk_io_trace(ent)->sector;
1232}
1233
1234static inline __u16 t_error(const struct trace_entry *ent)
1235{
1236	return te_blk_io_trace(ent)->error;
1237}
1238
1239static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1240{
1241	const __be64 *val = pdu_start(ent, has_cg);
1242	return be64_to_cpu(*val);
1243}
1244
 
 
 
 
 
 
 
 
 
 
 
1245typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1246	bool has_cg);
1247
1248static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1249	bool has_cg)
1250{
1251	char rwbs[RWBS_LEN];
1252	unsigned long long ts  = iter->ts;
1253	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1254	unsigned secs	       = (unsigned long)ts;
1255	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1256
1257	fill_rwbs(rwbs, t);
1258
1259	trace_seq_printf(&iter->seq,
1260			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1261			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1262			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1263}
1264
1265static void blk_log_action(struct trace_iterator *iter, const char *act,
1266	bool has_cg)
1267{
1268	char rwbs[RWBS_LEN];
1269	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1270
1271	fill_rwbs(rwbs, t);
1272	if (has_cg) {
1273		u64 id = t_cgid(iter->ent);
1274
1275		if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1276			char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1277
1278			cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1279				sizeof(blkcg_name_buf));
1280			trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1281				 MAJOR(t->device), MINOR(t->device),
1282				 blkcg_name_buf, act, rwbs);
1283		} else {
1284			/*
1285			 * The cgid portion used to be "INO,GEN".  Userland
1286			 * builds a FILEID_INO32_GEN fid out of them and
1287			 * opens the cgroup using open_by_handle_at(2).
1288			 * While 32bit ino setups are still the same, 64bit
1289			 * ones now use the 64bit ino as the whole ID and
1290			 * no longer use generation.
1291			 *
1292			 * Regardless of the content, always output
1293			 * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1294			 * be mapped back to @id on both 64 and 32bit ino
1295			 * setups.  See __kernfs_fh_to_dentry().
1296			 */
1297			trace_seq_printf(&iter->seq,
1298				 "%3d,%-3d %llx,%-llx %2s %3s ",
1299				 MAJOR(t->device), MINOR(t->device),
1300				 id & U32_MAX, id >> 32, act, rwbs);
1301		}
1302	} else
1303		trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1304				 MAJOR(t->device), MINOR(t->device), act, rwbs);
1305}
1306
1307static void blk_log_dump_pdu(struct trace_seq *s,
1308	const struct trace_entry *ent, bool has_cg)
1309{
1310	const unsigned char *pdu_buf;
1311	int pdu_len;
1312	int i, end;
1313
1314	pdu_buf = pdu_start(ent, has_cg);
1315	pdu_len = pdu_real_len(ent, has_cg);
1316
1317	if (!pdu_len)
1318		return;
1319
1320	/* find the last zero that needs to be printed */
1321	for (end = pdu_len - 1; end >= 0; end--)
1322		if (pdu_buf[end])
1323			break;
1324	end++;
1325
1326	trace_seq_putc(s, '(');
1327
1328	for (i = 0; i < pdu_len; i++) {
1329
1330		trace_seq_printf(s, "%s%02x",
1331				 i == 0 ? "" : " ", pdu_buf[i]);
1332
1333		/*
1334		 * stop when the rest is just zeros and indicate so
1335		 * with a ".." appended
1336		 */
1337		if (i == end && end != pdu_len - 1) {
1338			trace_seq_puts(s, " ..) ");
1339			return;
1340		}
1341	}
1342
1343	trace_seq_puts(s, ") ");
1344}
1345
1346static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1347{
1348	char cmd[TASK_COMM_LEN];
1349
1350	trace_find_cmdline(ent->pid, cmd);
1351
1352	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1353		trace_seq_printf(s, "%u ", t_bytes(ent));
1354		blk_log_dump_pdu(s, ent, has_cg);
1355		trace_seq_printf(s, "[%s]\n", cmd);
1356	} else {
1357		if (t_sec(ent))
1358			trace_seq_printf(s, "%llu + %u [%s]\n",
1359						t_sector(ent), t_sec(ent), cmd);
1360		else
1361			trace_seq_printf(s, "[%s]\n", cmd);
1362	}
1363}
1364
1365static void blk_log_with_error(struct trace_seq *s,
1366			      const struct trace_entry *ent, bool has_cg)
1367{
1368	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1369		blk_log_dump_pdu(s, ent, has_cg);
1370		trace_seq_printf(s, "[%d]\n", t_error(ent));
1371	} else {
1372		if (t_sec(ent))
1373			trace_seq_printf(s, "%llu + %u [%d]\n",
1374					 t_sector(ent),
1375					 t_sec(ent), t_error(ent));
1376		else
1377			trace_seq_printf(s, "%llu [%d]\n",
1378					 t_sector(ent), t_error(ent));
1379	}
1380}
1381
1382static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1383{
1384	const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1385
 
1386	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1387			 t_sector(ent), t_sec(ent),
1388			 MAJOR(be32_to_cpu(__r->device_from)),
1389			 MINOR(be32_to_cpu(__r->device_from)),
1390			 be64_to_cpu(__r->sector_from));
1391}
1392
1393static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1394{
1395	char cmd[TASK_COMM_LEN];
1396
1397	trace_find_cmdline(ent->pid, cmd);
1398
1399	trace_seq_printf(s, "[%s]\n", cmd);
1400}
1401
1402static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1403{
1404	char cmd[TASK_COMM_LEN];
1405
1406	trace_find_cmdline(ent->pid, cmd);
1407
1408	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1409}
1410
1411static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1412{
1413	char cmd[TASK_COMM_LEN];
1414
1415	trace_find_cmdline(ent->pid, cmd);
1416
1417	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1418			 get_pdu_int(ent, has_cg), cmd);
1419}
1420
1421static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1422			bool has_cg)
1423{
1424
1425	trace_seq_putmem(s, pdu_start(ent, has_cg),
1426		pdu_real_len(ent, has_cg));
1427	trace_seq_putc(s, '\n');
1428}
1429
1430/*
1431 * struct tracer operations
1432 */
1433
1434static void blk_tracer_print_header(struct seq_file *m)
1435{
1436	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1437		return;
1438	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1439		    "#  |     |     |           |   |   |\n");
1440}
1441
1442static void blk_tracer_start(struct trace_array *tr)
1443{
1444	blk_tracer_enabled = true;
1445}
1446
1447static int blk_tracer_init(struct trace_array *tr)
1448{
1449	blk_tr = tr;
1450	blk_tracer_start(tr);
1451	return 0;
1452}
1453
1454static void blk_tracer_stop(struct trace_array *tr)
1455{
1456	blk_tracer_enabled = false;
1457}
1458
1459static void blk_tracer_reset(struct trace_array *tr)
1460{
1461	blk_tracer_stop(tr);
1462}
1463
1464static const struct {
1465	const char *act[2];
1466	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent,
1467			    bool has_cg);
1468} what2act[] = {
1469	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1470	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1471	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1472	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1473	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1474	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1475	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1476	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1477	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1478	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1479	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1480	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1481	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1482	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1483	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1484};
1485
1486static enum print_line_t print_one_line(struct trace_iterator *iter,
1487					bool classic)
1488{
1489	struct trace_array *tr = iter->tr;
1490	struct trace_seq *s = &iter->seq;
1491	const struct blk_io_trace *t;
1492	u16 what;
1493	bool long_act;
1494	blk_log_action_t *log_action;
1495	bool has_cg;
1496
1497	t	   = te_blk_io_trace(iter->ent);
1498	what	   = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1499	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1500	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1501	has_cg	   = t->action & __BLK_TA_CGROUP;
1502
1503	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1504		log_action(iter, long_act ? "message" : "m", has_cg);
1505		blk_log_msg(s, iter->ent, has_cg);
1506		return trace_handle_return(s);
1507	}
1508
1509	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1510		trace_seq_printf(s, "Unknown action %x\n", what);
1511	else {
1512		log_action(iter, what2act[what].act[long_act], has_cg);
1513		what2act[what].print(s, iter->ent, has_cg);
1514	}
1515
1516	return trace_handle_return(s);
1517}
1518
1519static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1520					       int flags, struct trace_event *event)
1521{
1522	return print_one_line(iter, false);
1523}
1524
1525static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1526{
1527	struct trace_seq *s = &iter->seq;
1528	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1529	const int offset = offsetof(struct blk_io_trace, sector);
1530	struct blk_io_trace old = {
1531		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1532		.time     = iter->ts,
1533	};
1534
1535	trace_seq_putmem(s, &old, offset);
1536	trace_seq_putmem(s, &t->sector,
1537			 sizeof(old) - offset + t->pdu_len);
1538}
1539
1540static enum print_line_t
1541blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1542			     struct trace_event *event)
1543{
1544	blk_trace_synthesize_old_trace(iter);
1545
1546	return trace_handle_return(&iter->seq);
1547}
1548
1549static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1550{
1551	if ((iter->ent->type != TRACE_BLK) ||
1552	    !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1553		return TRACE_TYPE_UNHANDLED;
1554
1555	return print_one_line(iter, true);
1556}
1557
1558static int
1559blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1560{
1561	/* don't output context-info for blk_classic output */
1562	if (bit == TRACE_BLK_OPT_CLASSIC) {
1563		if (set)
1564			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1565		else
1566			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1567	}
1568	return 0;
1569}
1570
1571static struct tracer blk_tracer __read_mostly = {
1572	.name		= "blk",
1573	.init		= blk_tracer_init,
1574	.reset		= blk_tracer_reset,
1575	.start		= blk_tracer_start,
1576	.stop		= blk_tracer_stop,
1577	.print_header	= blk_tracer_print_header,
1578	.print_line	= blk_tracer_print_line,
1579	.flags		= &blk_tracer_flags,
1580	.set_flag	= blk_tracer_set_flag,
1581};
1582
1583static struct trace_event_functions trace_blk_event_funcs = {
1584	.trace		= blk_trace_event_print,
1585	.binary		= blk_trace_event_print_binary,
1586};
1587
1588static struct trace_event trace_blk_event = {
1589	.type		= TRACE_BLK,
1590	.funcs		= &trace_blk_event_funcs,
1591};
1592
1593static int __init init_blk_tracer(void)
1594{
1595	if (!register_trace_event(&trace_blk_event)) {
1596		pr_warn("Warning: could not register block events\n");
1597		return 1;
1598	}
1599
1600	if (register_tracer(&blk_tracer) != 0) {
1601		pr_warn("Warning: could not register the block tracer\n");
1602		unregister_trace_event(&trace_blk_event);
1603		return 1;
1604	}
1605
1606	return 0;
1607}
1608
1609device_initcall(init_blk_tracer);
1610
1611static int blk_trace_remove_queue(struct request_queue *q)
1612{
1613	struct blk_trace *bt;
1614
1615	bt = rcu_replace_pointer(q->blk_trace, NULL,
1616				 lockdep_is_held(&q->debugfs_mutex));
1617	if (bt == NULL)
1618		return -EINVAL;
1619
1620	blk_trace_stop(bt);
1621
1622	put_probe_ref();
1623	synchronize_rcu();
1624	blk_trace_free(q, bt);
1625	return 0;
1626}
1627
1628/*
1629 * Setup everything required to start tracing
1630 */
1631static int blk_trace_setup_queue(struct request_queue *q,
1632				 struct block_device *bdev)
1633{
1634	struct blk_trace *bt = NULL;
1635	int ret = -ENOMEM;
1636
1637	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1638	if (!bt)
1639		return -ENOMEM;
1640
1641	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1642	if (!bt->msg_data)
1643		goto free_bt;
1644
1645	bt->dev = bdev->bd_dev;
1646	bt->act_mask = (u16)-1;
1647
1648	blk_trace_setup_lba(bt, bdev);
1649
1650	rcu_assign_pointer(q->blk_trace, bt);
 
 
 
1651	get_probe_ref();
1652	return 0;
1653
1654free_bt:
1655	blk_trace_free(q, bt);
1656	return ret;
1657}
1658
1659/*
1660 * sysfs interface to enable and configure tracing
1661 */
1662
1663static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1664					 struct device_attribute *attr,
1665					 char *buf);
1666static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1667					  struct device_attribute *attr,
1668					  const char *buf, size_t count);
1669#define BLK_TRACE_DEVICE_ATTR(_name) \
1670	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1671		    sysfs_blk_trace_attr_show, \
1672		    sysfs_blk_trace_attr_store)
1673
1674static BLK_TRACE_DEVICE_ATTR(enable);
1675static BLK_TRACE_DEVICE_ATTR(act_mask);
1676static BLK_TRACE_DEVICE_ATTR(pid);
1677static BLK_TRACE_DEVICE_ATTR(start_lba);
1678static BLK_TRACE_DEVICE_ATTR(end_lba);
1679
1680static struct attribute *blk_trace_attrs[] = {
1681	&dev_attr_enable.attr,
1682	&dev_attr_act_mask.attr,
1683	&dev_attr_pid.attr,
1684	&dev_attr_start_lba.attr,
1685	&dev_attr_end_lba.attr,
1686	NULL
1687};
1688
1689struct attribute_group blk_trace_attr_group = {
1690	.name  = "trace",
1691	.attrs = blk_trace_attrs,
1692};
1693
1694static const struct {
1695	int mask;
1696	const char *str;
1697} mask_maps[] = {
1698	{ BLK_TC_READ,		"read"		},
1699	{ BLK_TC_WRITE,		"write"		},
1700	{ BLK_TC_FLUSH,		"flush"		},
1701	{ BLK_TC_SYNC,		"sync"		},
1702	{ BLK_TC_QUEUE,		"queue"		},
1703	{ BLK_TC_REQUEUE,	"requeue"	},
1704	{ BLK_TC_ISSUE,		"issue"		},
1705	{ BLK_TC_COMPLETE,	"complete"	},
1706	{ BLK_TC_FS,		"fs"		},
1707	{ BLK_TC_PC,		"pc"		},
1708	{ BLK_TC_NOTIFY,	"notify"	},
1709	{ BLK_TC_AHEAD,		"ahead"		},
1710	{ BLK_TC_META,		"meta"		},
1711	{ BLK_TC_DISCARD,	"discard"	},
1712	{ BLK_TC_DRV_DATA,	"drv_data"	},
1713	{ BLK_TC_FUA,		"fua"		},
1714};
1715
1716static int blk_trace_str2mask(const char *str)
1717{
1718	int i;
1719	int mask = 0;
1720	char *buf, *s, *token;
1721
1722	buf = kstrdup(str, GFP_KERNEL);
1723	if (buf == NULL)
1724		return -ENOMEM;
1725	s = strstrip(buf);
1726
1727	while (1) {
1728		token = strsep(&s, ",");
1729		if (token == NULL)
1730			break;
1731
1732		if (*token == '\0')
1733			continue;
1734
1735		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1736			if (strcasecmp(token, mask_maps[i].str) == 0) {
1737				mask |= mask_maps[i].mask;
1738				break;
1739			}
1740		}
1741		if (i == ARRAY_SIZE(mask_maps)) {
1742			mask = -EINVAL;
1743			break;
1744		}
1745	}
1746	kfree(buf);
1747
1748	return mask;
1749}
1750
1751static ssize_t blk_trace_mask2str(char *buf, int mask)
1752{
1753	int i;
1754	char *p = buf;
1755
1756	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1757		if (mask & mask_maps[i].mask) {
1758			p += sprintf(p, "%s%s",
1759				    (p == buf) ? "" : ",", mask_maps[i].str);
1760		}
1761	}
1762	*p++ = '\n';
1763
1764	return p - buf;
1765}
1766
 
 
 
 
 
 
 
 
1767static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1768					 struct device_attribute *attr,
1769					 char *buf)
1770{
1771	struct block_device *bdev = dev_to_bdev(dev);
1772	struct request_queue *q = bdev_get_queue(bdev);
1773	struct blk_trace *bt;
1774	ssize_t ret = -ENXIO;
1775
1776	mutex_lock(&q->debugfs_mutex);
 
 
 
 
 
 
 
 
1777
1778	bt = rcu_dereference_protected(q->blk_trace,
1779				       lockdep_is_held(&q->debugfs_mutex));
1780	if (attr == &dev_attr_enable) {
1781		ret = sprintf(buf, "%u\n", !!bt);
1782		goto out_unlock_bdev;
1783	}
1784
1785	if (bt == NULL)
1786		ret = sprintf(buf, "disabled\n");
1787	else if (attr == &dev_attr_act_mask)
1788		ret = blk_trace_mask2str(buf, bt->act_mask);
1789	else if (attr == &dev_attr_pid)
1790		ret = sprintf(buf, "%u\n", bt->pid);
1791	else if (attr == &dev_attr_start_lba)
1792		ret = sprintf(buf, "%llu\n", bt->start_lba);
1793	else if (attr == &dev_attr_end_lba)
1794		ret = sprintf(buf, "%llu\n", bt->end_lba);
1795
1796out_unlock_bdev:
1797	mutex_unlock(&q->debugfs_mutex);
 
 
 
1798	return ret;
1799}
1800
1801static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1802					  struct device_attribute *attr,
1803					  const char *buf, size_t count)
1804{
1805	struct block_device *bdev = dev_to_bdev(dev);
1806	struct request_queue *q = bdev_get_queue(bdev);
1807	struct blk_trace *bt;
1808	u64 value;
1809	ssize_t ret = -EINVAL;
1810
1811	if (count == 0)
1812		goto out;
1813
1814	if (attr == &dev_attr_act_mask) {
1815		if (kstrtoull(buf, 0, &value)) {
1816			/* Assume it is a list of trace category names */
1817			ret = blk_trace_str2mask(buf);
1818			if (ret < 0)
1819				goto out;
1820			value = ret;
1821		}
1822	} else {
1823		if (kstrtoull(buf, 0, &value))
1824			goto out;
1825	}
1826
1827	mutex_lock(&q->debugfs_mutex);
 
 
 
 
 
 
 
 
 
 
 
1828
1829	bt = rcu_dereference_protected(q->blk_trace,
1830				       lockdep_is_held(&q->debugfs_mutex));
1831	if (attr == &dev_attr_enable) {
1832		if (!!value == !!bt) {
1833			ret = 0;
1834			goto out_unlock_bdev;
1835		}
1836		if (value)
1837			ret = blk_trace_setup_queue(q, bdev);
1838		else
1839			ret = blk_trace_remove_queue(q);
1840		goto out_unlock_bdev;
1841	}
1842
1843	ret = 0;
1844	if (bt == NULL) {
1845		ret = blk_trace_setup_queue(q, bdev);
1846		bt = rcu_dereference_protected(q->blk_trace,
1847				lockdep_is_held(&q->debugfs_mutex));
1848	}
1849
1850	if (ret == 0) {
1851		if (attr == &dev_attr_act_mask)
1852			bt->act_mask = value;
1853		else if (attr == &dev_attr_pid)
1854			bt->pid = value;
1855		else if (attr == &dev_attr_start_lba)
1856			bt->start_lba = value;
1857		else if (attr == &dev_attr_end_lba)
1858			bt->end_lba = value;
1859	}
1860
1861out_unlock_bdev:
1862	mutex_unlock(&q->debugfs_mutex);
 
 
1863out:
1864	return ret ? ret : count;
1865}
 
 
 
 
 
 
 
 
 
 
 
1866#endif /* CONFIG_BLK_DEV_IO_TRACE */
1867
1868#ifdef CONFIG_EVENT_TRACING
1869
1870/**
1871 * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string.
1872 * @rwbs:	buffer to be filled
1873 * @opf:	request operation type (REQ_OP_XXX) and flags for the tracepoint
1874 *
1875 * Description:
1876 *     Maps each request operation and flag to a single character and fills the
1877 *     buffer provided by the caller with resulting string.
1878 *
1879 **/
1880void blk_fill_rwbs(char *rwbs, blk_opf_t opf)
1881{
1882	int i = 0;
1883
1884	if (opf & REQ_PREFLUSH)
1885		rwbs[i++] = 'F';
1886
1887	switch (opf & REQ_OP_MASK) {
1888	case REQ_OP_WRITE:
 
1889		rwbs[i++] = 'W';
1890		break;
1891	case REQ_OP_DISCARD:
1892		rwbs[i++] = 'D';
1893		break;
1894	case REQ_OP_SECURE_ERASE:
1895		rwbs[i++] = 'D';
1896		rwbs[i++] = 'E';
1897		break;
1898	case REQ_OP_FLUSH:
1899		rwbs[i++] = 'F';
1900		break;
1901	case REQ_OP_READ:
1902		rwbs[i++] = 'R';
1903		break;
1904	default:
1905		rwbs[i++] = 'N';
1906	}
1907
1908	if (opf & REQ_FUA)
1909		rwbs[i++] = 'F';
1910	if (opf & REQ_RAHEAD)
1911		rwbs[i++] = 'A';
1912	if (opf & REQ_SYNC)
1913		rwbs[i++] = 'S';
1914	if (opf & REQ_META)
1915		rwbs[i++] = 'M';
1916
1917	rwbs[i] = '\0';
1918}
1919EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1920
1921#endif /* CONFIG_EVENT_TRACING */
1922
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4 *
   5 */
 
 
 
   6#include <linux/kernel.h>
   7#include <linux/blkdev.h>
   8#include <linux/blktrace_api.h>
   9#include <linux/percpu.h>
  10#include <linux/init.h>
  11#include <linux/mutex.h>
  12#include <linux/slab.h>
  13#include <linux/debugfs.h>
  14#include <linux/export.h>
  15#include <linux/time.h>
  16#include <linux/uaccess.h>
  17#include <linux/list.h>
  18#include <linux/blk-cgroup.h>
  19
  20#include "../../block/blk.h"
  21
  22#include <trace/events/block.h>
  23
  24#include "trace_output.h"
  25
  26#ifdef CONFIG_BLK_DEV_IO_TRACE
  27
  28static unsigned int blktrace_seq __read_mostly = 1;
  29
  30static struct trace_array *blk_tr;
  31static bool blk_tracer_enabled __read_mostly;
  32
  33static LIST_HEAD(running_trace_list);
  34static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  35
  36/* Select an alternative, minimalistic output than the original one */
  37#define TRACE_BLK_OPT_CLASSIC	0x1
  38#define TRACE_BLK_OPT_CGROUP	0x2
  39#define TRACE_BLK_OPT_CGNAME	0x4
  40
  41static struct tracer_opt blk_tracer_opts[] = {
  42	/* Default disable the minimalistic output */
  43	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  44#ifdef CONFIG_BLK_CGROUP
  45	{ TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  46	{ TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  47#endif
  48	{ }
  49};
  50
  51static struct tracer_flags blk_tracer_flags = {
  52	.val  = 0,
  53	.opts = blk_tracer_opts,
  54};
  55
  56/* Global reference count of probes */
  57static DEFINE_MUTEX(blk_probe_mutex);
  58static int blk_probes_ref;
  59
  60static void blk_register_tracepoints(void);
  61static void blk_unregister_tracepoints(void);
  62
  63/*
  64 * Send out a notify message.
  65 */
  66static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  67		       const void *data, size_t len,
  68		       union kernfs_node_id *cgid)
  69{
  70	struct blk_io_trace *t;
  71	struct ring_buffer_event *event = NULL;
  72	struct ring_buffer *buffer = NULL;
  73	int pc = 0;
  74	int cpu = smp_processor_id();
  75	bool blk_tracer = blk_tracer_enabled;
  76	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  77
  78	if (blk_tracer) {
  79		buffer = blk_tr->trace_buffer.buffer;
  80		pc = preempt_count();
  81		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  82						  sizeof(*t) + len + cgid_len,
  83						  0, pc);
  84		if (!event)
  85			return;
  86		t = ring_buffer_event_data(event);
  87		goto record_it;
  88	}
  89
  90	if (!bt->rchan)
  91		return;
  92
  93	t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  94	if (t) {
  95		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  96		t->time = ktime_to_ns(ktime_get());
  97record_it:
  98		t->device = bt->dev;
  99		t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 100		t->pid = pid;
 101		t->cpu = cpu;
 102		t->pdu_len = len + cgid_len;
 103		if (cgid)
 104			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 105		memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 106
 107		if (blk_tracer)
 108			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 109	}
 110}
 111
 112/*
 113 * Send out a notify for this process, if we haven't done so since a trace
 114 * started
 115 */
 116static void trace_note_tsk(struct task_struct *tsk)
 117{
 118	unsigned long flags;
 119	struct blk_trace *bt;
 120
 121	tsk->btrace_seq = blktrace_seq;
 122	spin_lock_irqsave(&running_trace_lock, flags);
 123	list_for_each_entry(bt, &running_trace_list, running_list) {
 124		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 125			   sizeof(tsk->comm), NULL);
 126	}
 127	spin_unlock_irqrestore(&running_trace_lock, flags);
 128}
 129
 130static void trace_note_time(struct blk_trace *bt)
 131{
 132	struct timespec64 now;
 133	unsigned long flags;
 134	u32 words[2];
 135
 136	/* need to check user space to see if this breaks in y2038 or y2106 */
 137	ktime_get_real_ts64(&now);
 138	words[0] = (u32)now.tv_sec;
 139	words[1] = now.tv_nsec;
 140
 141	local_irq_save(flags);
 142	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 143	local_irq_restore(flags);
 144}
 145
 146void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 147	const char *fmt, ...)
 148{
 149	int n;
 150	va_list args;
 151	unsigned long flags;
 152	char *buf;
 
 153
 154	if (unlikely(bt->trace_state != Blktrace_running &&
 155		     !blk_tracer_enabled))
 156		return;
 157
 158	/*
 159	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 160	 * message to the trace.
 161	 */
 162	if (!(bt->act_mask & BLK_TC_NOTIFY))
 163		return;
 164
 165	local_irq_save(flags);
 166	buf = this_cpu_ptr(bt->msg_data);
 167	va_start(args, fmt);
 168	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 169	va_end(args);
 170
 171	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 172		blkcg = NULL;
 173#ifdef CONFIG_BLK_CGROUP
 174	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 175		blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 176#else
 177	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 178#endif
 
 179	local_irq_restore(flags);
 180}
 181EXPORT_SYMBOL_GPL(__trace_note_message);
 182
 183static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 184			 pid_t pid)
 185{
 186	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 187		return 1;
 188	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 189		return 1;
 190	if (bt->pid && pid != bt->pid)
 191		return 1;
 192
 193	return 0;
 194}
 195
 196/*
 197 * Data direction bit lookup
 198 */
 199static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 200				 BLK_TC_ACT(BLK_TC_WRITE) };
 201
 202#define BLK_TC_RAHEAD		BLK_TC_AHEAD
 203#define BLK_TC_PREFLUSH		BLK_TC_FLUSH
 204
 205/* The ilog2() calls fall out because they're constant */
 206#define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 207	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 208
 209/*
 210 * The worker for the various blk_add_trace*() types. Fills out a
 211 * blk_io_trace structure and places it in a per-cpu subbuffer.
 212 */
 213static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 214		     int op, int op_flags, u32 what, int error, int pdu_len,
 215		     void *pdu_data, union kernfs_node_id *cgid)
 216{
 217	struct task_struct *tsk = current;
 218	struct ring_buffer_event *event = NULL;
 219	struct ring_buffer *buffer = NULL;
 220	struct blk_io_trace *t;
 221	unsigned long flags = 0;
 222	unsigned long *sequence;
 
 223	pid_t pid;
 224	int cpu, pc = 0;
 225	bool blk_tracer = blk_tracer_enabled;
 226	ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 
 227
 228	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 229		return;
 230
 231	what |= ddir_act[op_is_write(op) ? WRITE : READ];
 232	what |= MASK_TC_BIT(op_flags, SYNC);
 233	what |= MASK_TC_BIT(op_flags, RAHEAD);
 234	what |= MASK_TC_BIT(op_flags, META);
 235	what |= MASK_TC_BIT(op_flags, PREFLUSH);
 236	what |= MASK_TC_BIT(op_flags, FUA);
 237	if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 238		what |= BLK_TC_ACT(BLK_TC_DISCARD);
 239	if (op == REQ_OP_FLUSH)
 240		what |= BLK_TC_ACT(BLK_TC_FLUSH);
 241	if (cgid)
 242		what |= __BLK_TA_CGROUP;
 243
 244	pid = tsk->pid;
 245	if (act_log_check(bt, what, sector, pid))
 246		return;
 247	cpu = raw_smp_processor_id();
 248
 249	if (blk_tracer) {
 250		tracing_record_cmdline(current);
 251
 252		buffer = blk_tr->trace_buffer.buffer;
 253		pc = preempt_count();
 254		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 255						  sizeof(*t) + pdu_len + cgid_len,
 256						  0, pc);
 257		if (!event)
 258			return;
 259		t = ring_buffer_event_data(event);
 260		goto record_it;
 261	}
 262
 263	if (unlikely(tsk->btrace_seq != blktrace_seq))
 264		trace_note_tsk(tsk);
 265
 266	/*
 267	 * A word about the locking here - we disable interrupts to reserve
 268	 * some space in the relay per-cpu buffer, to prevent an irq
 269	 * from coming in and stepping on our toes.
 270	 */
 271	local_irq_save(flags);
 272	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 273	if (t) {
 274		sequence = per_cpu_ptr(bt->sequence, cpu);
 275
 276		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 277		t->sequence = ++(*sequence);
 278		t->time = ktime_to_ns(ktime_get());
 279record_it:
 280		/*
 281		 * These two are not needed in ftrace as they are in the
 282		 * generic trace_entry, filled by tracing_generic_entry_update,
 283		 * but for the trace_event->bin() synthesizer benefit we do it
 284		 * here too.
 285		 */
 286		t->cpu = cpu;
 287		t->pid = pid;
 288
 289		t->sector = sector;
 290		t->bytes = bytes;
 291		t->action = what;
 292		t->device = bt->dev;
 293		t->error = error;
 294		t->pdu_len = pdu_len + cgid_len;
 295
 296		if (cgid_len)
 297			memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 298		if (pdu_len)
 299			memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 300
 301		if (blk_tracer) {
 302			trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 303			return;
 304		}
 305	}
 306
 307	local_irq_restore(flags);
 308}
 309
 310static void blk_trace_free(struct blk_trace *bt)
 311{
 312	debugfs_remove(bt->msg_file);
 313	debugfs_remove(bt->dropped_file);
 314	relay_close(bt->rchan);
 315	debugfs_remove(bt->dir);
 
 
 
 
 
 
 
 
 
 
 316	free_percpu(bt->sequence);
 317	free_percpu(bt->msg_data);
 318	kfree(bt);
 319}
 320
 321static void get_probe_ref(void)
 322{
 323	mutex_lock(&blk_probe_mutex);
 324	if (++blk_probes_ref == 1)
 325		blk_register_tracepoints();
 326	mutex_unlock(&blk_probe_mutex);
 327}
 328
 329static void put_probe_ref(void)
 330{
 331	mutex_lock(&blk_probe_mutex);
 332	if (!--blk_probes_ref)
 333		blk_unregister_tracepoints();
 334	mutex_unlock(&blk_probe_mutex);
 335}
 336
 337static void blk_trace_cleanup(struct blk_trace *bt)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 338{
 339	blk_trace_free(bt);
 
 
 340	put_probe_ref();
 341}
 342
 343static int __blk_trace_remove(struct request_queue *q)
 344{
 345	struct blk_trace *bt;
 346
 347	bt = xchg(&q->blk_trace, NULL);
 
 348	if (!bt)
 349		return -EINVAL;
 350
 351	if (bt->trace_state != Blktrace_running)
 352		blk_trace_cleanup(bt);
 353
 354	return 0;
 355}
 356
 357int blk_trace_remove(struct request_queue *q)
 358{
 359	int ret;
 360
 361	mutex_lock(&q->blk_trace_mutex);
 362	ret = __blk_trace_remove(q);
 363	mutex_unlock(&q->blk_trace_mutex);
 364
 365	return ret;
 366}
 367EXPORT_SYMBOL_GPL(blk_trace_remove);
 368
 369static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 370				size_t count, loff_t *ppos)
 371{
 372	struct blk_trace *bt = filp->private_data;
 373	char buf[16];
 374
 375	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 376
 377	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 378}
 379
 380static const struct file_operations blk_dropped_fops = {
 381	.owner =	THIS_MODULE,
 382	.open =		simple_open,
 383	.read =		blk_dropped_read,
 384	.llseek =	default_llseek,
 385};
 386
 387static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 388				size_t count, loff_t *ppos)
 389{
 390	char *msg;
 391	struct blk_trace *bt;
 392
 393	if (count >= BLK_TN_MAX_MSG)
 394		return -EINVAL;
 395
 396	msg = memdup_user_nul(buffer, count);
 397	if (IS_ERR(msg))
 398		return PTR_ERR(msg);
 399
 400	bt = filp->private_data;
 401	__trace_note_message(bt, NULL, "%s", msg);
 402	kfree(msg);
 403
 404	return count;
 405}
 406
 407static const struct file_operations blk_msg_fops = {
 408	.owner =	THIS_MODULE,
 409	.open =		simple_open,
 410	.write =	blk_msg_write,
 411	.llseek =	noop_llseek,
 412};
 413
 414/*
 415 * Keep track of how many times we encountered a full subbuffer, to aid
 416 * the user space app in telling how many lost events there were.
 417 */
 418static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 419				     void *prev_subbuf, size_t prev_padding)
 420{
 421	struct blk_trace *bt;
 422
 423	if (!relay_buf_full(buf))
 424		return 1;
 425
 426	bt = buf->chan->private_data;
 427	atomic_inc(&bt->dropped);
 428	return 0;
 429}
 430
 431static int blk_remove_buf_file_callback(struct dentry *dentry)
 432{
 433	debugfs_remove(dentry);
 434
 435	return 0;
 436}
 437
 438static struct dentry *blk_create_buf_file_callback(const char *filename,
 439						   struct dentry *parent,
 440						   umode_t mode,
 441						   struct rchan_buf *buf,
 442						   int *is_global)
 443{
 444	return debugfs_create_file(filename, mode, parent, buf,
 445					&relay_file_operations);
 446}
 447
 448static struct rchan_callbacks blk_relay_callbacks = {
 449	.subbuf_start		= blk_subbuf_start_callback,
 450	.create_buf_file	= blk_create_buf_file_callback,
 451	.remove_buf_file	= blk_remove_buf_file_callback,
 452};
 453
 454static void blk_trace_setup_lba(struct blk_trace *bt,
 455				struct block_device *bdev)
 456{
 457	struct hd_struct *part = NULL;
 458
 459	if (bdev)
 460		part = bdev->bd_part;
 461
 462	if (part) {
 463		bt->start_lba = part->start_sect;
 464		bt->end_lba = part->start_sect + part->nr_sects;
 465	} else {
 466		bt->start_lba = 0;
 467		bt->end_lba = -1ULL;
 468	}
 469}
 470
 471/*
 472 * Setup everything required to start tracing
 473 */
 474static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 475			      struct block_device *bdev,
 476			      struct blk_user_trace_setup *buts)
 477{
 478	struct blk_trace *bt = NULL;
 479	struct dentry *dir = NULL;
 480	int ret;
 481
 
 
 482	if (!buts->buf_size || !buts->buf_nr)
 483		return -EINVAL;
 484
 485	if (!blk_debugfs_root)
 486		return -ENOENT;
 487
 488	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 489	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 490
 491	/*
 492	 * some device names have larger paths - convert the slashes
 493	 * to underscores for this to work as expected
 494	 */
 495	strreplace(buts->name, '/', '_');
 496
 
 
 
 
 
 
 
 
 
 
 
 497	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 498	if (!bt)
 499		return -ENOMEM;
 500
 501	ret = -ENOMEM;
 502	bt->sequence = alloc_percpu(unsigned long);
 503	if (!bt->sequence)
 504		goto err;
 505
 506	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 507	if (!bt->msg_data)
 508		goto err;
 509
 510	ret = -ENOENT;
 
 
 
 
 
 
 
 
 
 511
 512	dir = debugfs_lookup(buts->name, blk_debugfs_root);
 513	if (!dir)
 514		bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 
 
 
 
 
 
 
 
 515
 516	bt->dev = dev;
 517	atomic_set(&bt->dropped, 0);
 518	INIT_LIST_HEAD(&bt->running_list);
 519
 520	ret = -EIO;
 521	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 522					       &blk_dropped_fops);
 523
 524	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 525
 526	bt->rchan = relay_open("trace", dir, buts->buf_size,
 527				buts->buf_nr, &blk_relay_callbacks, bt);
 528	if (!bt->rchan)
 529		goto err;
 530
 531	bt->act_mask = buts->act_mask;
 532	if (!bt->act_mask)
 533		bt->act_mask = (u16) -1;
 534
 535	blk_trace_setup_lba(bt, bdev);
 536
 537	/* overwrite with user settings */
 538	if (buts->start_lba)
 539		bt->start_lba = buts->start_lba;
 540	if (buts->end_lba)
 541		bt->end_lba = buts->end_lba;
 542
 543	bt->pid = buts->pid;
 544	bt->trace_state = Blktrace_setup;
 545
 546	ret = -EBUSY;
 547	if (cmpxchg(&q->blk_trace, NULL, bt))
 548		goto err;
 549
 550	get_probe_ref();
 551
 552	ret = 0;
 553err:
 554	if (dir && !bt->dir)
 555		dput(dir);
 556	if (ret)
 557		blk_trace_free(bt);
 558	return ret;
 559}
 560
 561static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 562			     struct block_device *bdev, char __user *arg)
 563{
 564	struct blk_user_trace_setup buts;
 565	int ret;
 566
 567	ret = copy_from_user(&buts, arg, sizeof(buts));
 568	if (ret)
 569		return -EFAULT;
 570
 571	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 572	if (ret)
 573		return ret;
 574
 575	if (copy_to_user(arg, &buts, sizeof(buts))) {
 576		__blk_trace_remove(q);
 577		return -EFAULT;
 578	}
 579	return 0;
 580}
 581
 582int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 583		    struct block_device *bdev,
 584		    char __user *arg)
 585{
 586	int ret;
 587
 588	mutex_lock(&q->blk_trace_mutex);
 589	ret = __blk_trace_setup(q, name, dev, bdev, arg);
 590	mutex_unlock(&q->blk_trace_mutex);
 591
 592	return ret;
 593}
 594EXPORT_SYMBOL_GPL(blk_trace_setup);
 595
 596#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 597static int compat_blk_trace_setup(struct request_queue *q, char *name,
 598				  dev_t dev, struct block_device *bdev,
 599				  char __user *arg)
 600{
 601	struct blk_user_trace_setup buts;
 602	struct compat_blk_user_trace_setup cbuts;
 603	int ret;
 604
 605	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 606		return -EFAULT;
 607
 608	buts = (struct blk_user_trace_setup) {
 609		.act_mask = cbuts.act_mask,
 610		.buf_size = cbuts.buf_size,
 611		.buf_nr = cbuts.buf_nr,
 612		.start_lba = cbuts.start_lba,
 613		.end_lba = cbuts.end_lba,
 614		.pid = cbuts.pid,
 615	};
 616
 617	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 618	if (ret)
 619		return ret;
 620
 621	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 622		__blk_trace_remove(q);
 623		return -EFAULT;
 624	}
 625
 626	return 0;
 627}
 628#endif
 629
 630static int __blk_trace_startstop(struct request_queue *q, int start)
 631{
 632	int ret;
 633	struct blk_trace *bt = q->blk_trace;
 634
 
 
 635	if (bt == NULL)
 636		return -EINVAL;
 637
 638	/*
 639	 * For starting a trace, we can transition from a setup or stopped
 640	 * trace. For stopping a trace, the state must be running
 641	 */
 642	ret = -EINVAL;
 643	if (start) {
 644		if (bt->trace_state == Blktrace_setup ||
 645		    bt->trace_state == Blktrace_stopped) {
 646			blktrace_seq++;
 647			smp_mb();
 648			bt->trace_state = Blktrace_running;
 649			spin_lock_irq(&running_trace_lock);
 650			list_add(&bt->running_list, &running_trace_list);
 651			spin_unlock_irq(&running_trace_lock);
 652
 653			trace_note_time(bt);
 654			ret = 0;
 655		}
 656	} else {
 657		if (bt->trace_state == Blktrace_running) {
 658			bt->trace_state = Blktrace_stopped;
 659			spin_lock_irq(&running_trace_lock);
 660			list_del_init(&bt->running_list);
 661			spin_unlock_irq(&running_trace_lock);
 662			relay_flush(bt->rchan);
 663			ret = 0;
 664		}
 665	}
 666
 667	return ret;
 668}
 669
 670int blk_trace_startstop(struct request_queue *q, int start)
 671{
 672	int ret;
 673
 674	mutex_lock(&q->blk_trace_mutex);
 675	ret = __blk_trace_startstop(q, start);
 676	mutex_unlock(&q->blk_trace_mutex);
 677
 678	return ret;
 679}
 680EXPORT_SYMBOL_GPL(blk_trace_startstop);
 681
 682/*
 683 * When reading or writing the blktrace sysfs files, the references to the
 684 * opened sysfs or device files should prevent the underlying block device
 685 * from being removed. So no further delete protection is really needed.
 686 */
 687
 688/**
 689 * blk_trace_ioctl: - handle the ioctls associated with tracing
 690 * @bdev:	the block device
 691 * @cmd:	the ioctl cmd
 692 * @arg:	the argument data, if any
 693 *
 694 **/
 695int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 696{
 697	struct request_queue *q;
 698	int ret, start = 0;
 699	char b[BDEVNAME_SIZE];
 700
 701	q = bdev_get_queue(bdev);
 702	if (!q)
 703		return -ENXIO;
 704
 705	mutex_lock(&q->blk_trace_mutex);
 706
 707	switch (cmd) {
 708	case BLKTRACESETUP:
 709		bdevname(bdev, b);
 710		ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 711		break;
 712#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 713	case BLKTRACESETUP32:
 714		bdevname(bdev, b);
 715		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 716		break;
 717#endif
 718	case BLKTRACESTART:
 719		start = 1;
 720		/* fall through */
 721	case BLKTRACESTOP:
 722		ret = __blk_trace_startstop(q, start);
 723		break;
 724	case BLKTRACETEARDOWN:
 725		ret = __blk_trace_remove(q);
 726		break;
 727	default:
 728		ret = -ENOTTY;
 729		break;
 730	}
 731
 732	mutex_unlock(&q->blk_trace_mutex);
 733	return ret;
 734}
 735
 736/**
 737 * blk_trace_shutdown: - stop and cleanup trace structures
 738 * @q:    the request queue associated with the device
 739 *
 740 **/
 741void blk_trace_shutdown(struct request_queue *q)
 742{
 743	mutex_lock(&q->blk_trace_mutex);
 744
 745	if (q->blk_trace) {
 746		__blk_trace_startstop(q, 0);
 747		__blk_trace_remove(q);
 748	}
 749
 750	mutex_unlock(&q->blk_trace_mutex);
 751}
 752
 753#ifdef CONFIG_BLK_CGROUP
 754static union kernfs_node_id *
 755blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 756{
 757	struct blk_trace *bt = q->blk_trace;
 
 758
 
 
 759	if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 760		return NULL;
 761
 762	if (!bio->bi_blkg)
 763		return NULL;
 764	return cgroup_get_kernfs_id(bio_blkcg(bio)->css.cgroup);
 
 765}
 766#else
 767static union kernfs_node_id *
 768blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 769{
 770	return NULL;
 771}
 772#endif
 773
 774static union kernfs_node_id *
 775blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 776{
 777	if (!rq->bio)
 778		return NULL;
 779	/* Use the first bio */
 780	return blk_trace_bio_get_cgid(q, rq->bio);
 781}
 782
 783/*
 784 * blktrace probes
 785 */
 786
 787/**
 788 * blk_add_trace_rq - Add a trace for a request oriented action
 789 * @rq:		the source request
 790 * @error:	return status to log
 791 * @nr_bytes:	number of completed bytes
 792 * @what:	the action
 793 * @cgid:	the cgroup info
 794 *
 795 * Description:
 796 *     Records an action against a request. Will log the bio offset + size.
 797 *
 798 **/
 799static void blk_add_trace_rq(struct request *rq, int error,
 800			     unsigned int nr_bytes, u32 what,
 801			     union kernfs_node_id *cgid)
 802{
 803	struct blk_trace *bt = rq->q->blk_trace;
 804
 805	if (likely(!bt))
 
 
 
 806		return;
 
 807
 808	if (blk_rq_is_passthrough(rq))
 809		what |= BLK_TC_ACT(BLK_TC_PC);
 810	else
 811		what |= BLK_TC_ACT(BLK_TC_FS);
 812
 813	__blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 814			rq->cmd_flags, what, error, 0, NULL, cgid);
 
 815}
 816
 817static void blk_add_trace_rq_insert(void *ignore,
 818				    struct request_queue *q, struct request *rq)
 819{
 820	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 821			 blk_trace_request_get_cgid(q, rq));
 822}
 823
 824static void blk_add_trace_rq_issue(void *ignore,
 825				   struct request_queue *q, struct request *rq)
 826{
 827	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 828			 blk_trace_request_get_cgid(q, rq));
 829}
 830
 831static void blk_add_trace_rq_requeue(void *ignore,
 832				     struct request_queue *q,
 833				     struct request *rq)
 
 
 
 
 834{
 835	blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 836			 blk_trace_request_get_cgid(q, rq));
 837}
 838
 839static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 840			int error, unsigned int nr_bytes)
 841{
 842	blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 843			 blk_trace_request_get_cgid(rq->q, rq));
 844}
 845
 846/**
 847 * blk_add_trace_bio - Add a trace for a bio oriented action
 848 * @q:		queue the io is for
 849 * @bio:	the source bio
 850 * @what:	the action
 851 * @error:	error, if any
 852 *
 853 * Description:
 854 *     Records an action against a bio. Will log the bio offset + size.
 855 *
 856 **/
 857static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 858			      u32 what, int error)
 859{
 860	struct blk_trace *bt = q->blk_trace;
 861
 862	if (likely(!bt))
 
 
 
 863		return;
 
 864
 865	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 866			bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 867			blk_trace_bio_get_cgid(q, bio));
 
 868}
 869
 870static void blk_add_trace_bio_bounce(void *ignore,
 871				     struct request_queue *q, struct bio *bio)
 872{
 873	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 874}
 875
 876static void blk_add_trace_bio_complete(void *ignore,
 877				       struct request_queue *q, struct bio *bio,
 878				       int error)
 879{
 880	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 
 881}
 882
 883static void blk_add_trace_bio_backmerge(void *ignore,
 884					struct request_queue *q,
 885					struct request *rq,
 886					struct bio *bio)
 887{
 888	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 
 889}
 890
 891static void blk_add_trace_bio_frontmerge(void *ignore,
 892					 struct request_queue *q,
 893					 struct request *rq,
 894					 struct bio *bio)
 895{
 896	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 
 897}
 898
 899static void blk_add_trace_bio_queue(void *ignore,
 900				    struct request_queue *q, struct bio *bio)
 901{
 902	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 903}
 904
 905static void blk_add_trace_getrq(void *ignore,
 906				struct request_queue *q,
 907				struct bio *bio, int rw)
 908{
 909	if (bio)
 910		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 911	else {
 912		struct blk_trace *bt = q->blk_trace;
 913
 914		if (bt)
 915			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 916					NULL, NULL);
 917	}
 918}
 919
 920
 921static void blk_add_trace_sleeprq(void *ignore,
 922				  struct request_queue *q,
 923				  struct bio *bio, int rw)
 924{
 925	if (bio)
 926		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 927	else {
 928		struct blk_trace *bt = q->blk_trace;
 929
 930		if (bt)
 931			__blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 932					0, 0, NULL, NULL);
 933	}
 934}
 935
 936static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 937{
 938	struct blk_trace *bt = q->blk_trace;
 939
 
 
 940	if (bt)
 941		__blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
 
 942}
 943
 944static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 945				    unsigned int depth, bool explicit)
 946{
 947	struct blk_trace *bt = q->blk_trace;
 948
 
 
 949	if (bt) {
 950		__be64 rpdu = cpu_to_be64(depth);
 951		u32 what;
 952
 953		if (explicit)
 954			what = BLK_TA_UNPLUG_IO;
 955		else
 956			what = BLK_TA_UNPLUG_TIMER;
 957
 958		__blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
 959	}
 
 960}
 961
 962static void blk_add_trace_split(void *ignore,
 963				struct request_queue *q, struct bio *bio,
 964				unsigned int pdu)
 965{
 966	struct blk_trace *bt = q->blk_trace;
 
 967
 
 
 968	if (bt) {
 969		__be64 rpdu = cpu_to_be64(pdu);
 970
 971		__blk_add_trace(bt, bio->bi_iter.bi_sector,
 972				bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 973				BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
 974				&rpdu, blk_trace_bio_get_cgid(q, bio));
 
 975	}
 
 976}
 977
 978/**
 979 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 980 * @ignore:	trace callback data parameter (not used)
 981 * @q:		queue the io is for
 982 * @bio:	the source bio
 983 * @dev:	target device
 984 * @from:	source sector
 985 *
 986 * Description:
 987 *     Device mapper or raid target sometimes need to split a bio because
 988 *     it spans a stripe (or similar). Add a trace for that action.
 989 *
 990 **/
 991static void blk_add_trace_bio_remap(void *ignore,
 992				    struct request_queue *q, struct bio *bio,
 993				    dev_t dev, sector_t from)
 994{
 995	struct blk_trace *bt = q->blk_trace;
 
 996	struct blk_io_trace_remap r;
 997
 998	if (likely(!bt))
 
 
 
 999		return;
 
1000
1001	r.device_from = cpu_to_be32(dev);
1002	r.device_to   = cpu_to_be32(bio_dev(bio));
1003	r.sector_from = cpu_to_be64(from);
1004
1005	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1006			bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
 
1007			sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
 
1008}
1009
1010/**
1011 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1012 * @ignore:	trace callback data parameter (not used)
1013 * @q:		queue the io is for
1014 * @rq:		the source request
1015 * @dev:	target device
1016 * @from:	source sector
1017 *
1018 * Description:
1019 *     Device mapper remaps request to other devices.
1020 *     Add a trace for that action.
1021 *
1022 **/
1023static void blk_add_trace_rq_remap(void *ignore,
1024				   struct request_queue *q,
1025				   struct request *rq, dev_t dev,
1026				   sector_t from)
1027{
1028	struct blk_trace *bt = q->blk_trace;
1029	struct blk_io_trace_remap r;
1030
1031	if (likely(!bt))
 
 
 
1032		return;
 
1033
1034	r.device_from = cpu_to_be32(dev);
1035	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1036	r.sector_from = cpu_to_be64(from);
1037
1038	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1039			rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1040			sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
 
1041}
1042
1043/**
1044 * blk_add_driver_data - Add binary message with driver-specific data
1045 * @q:		queue the io is for
1046 * @rq:		io request
1047 * @data:	driver-specific data
1048 * @len:	length of driver-specific data
1049 *
1050 * Description:
1051 *     Some drivers might want to write driver-specific data per request.
1052 *
1053 **/
1054void blk_add_driver_data(struct request_queue *q,
1055			 struct request *rq,
1056			 void *data, size_t len)
1057{
1058	struct blk_trace *bt = q->blk_trace;
1059
1060	if (likely(!bt))
 
 
 
1061		return;
 
1062
1063	__blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1064				BLK_TA_DRV_DATA, 0, len, data,
1065				blk_trace_request_get_cgid(q, rq));
 
1066}
1067EXPORT_SYMBOL_GPL(blk_add_driver_data);
1068
1069static void blk_register_tracepoints(void)
1070{
1071	int ret;
1072
1073	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1074	WARN_ON(ret);
1075	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1076	WARN_ON(ret);
 
 
1077	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1078	WARN_ON(ret);
1079	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1080	WARN_ON(ret);
1081	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1082	WARN_ON(ret);
1083	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1084	WARN_ON(ret);
1085	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1086	WARN_ON(ret);
1087	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1088	WARN_ON(ret);
1089	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1090	WARN_ON(ret);
1091	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1092	WARN_ON(ret);
1093	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1094	WARN_ON(ret);
1095	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1096	WARN_ON(ret);
1097	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1098	WARN_ON(ret);
1099	ret = register_trace_block_split(blk_add_trace_split, NULL);
1100	WARN_ON(ret);
1101	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1102	WARN_ON(ret);
1103	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1104	WARN_ON(ret);
1105}
1106
1107static void blk_unregister_tracepoints(void)
1108{
1109	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1110	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1111	unregister_trace_block_split(blk_add_trace_split, NULL);
1112	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1113	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1114	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1115	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1116	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1117	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1118	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1119	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1120	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1121	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1122	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 
1123	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1124	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1125
1126	tracepoint_synchronize_unregister();
1127}
1128
1129/*
1130 * struct blk_io_tracer formatting routines
1131 */
1132
1133static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1134{
1135	int i = 0;
1136	int tc = t->action >> BLK_TC_SHIFT;
1137
1138	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1139		rwbs[i++] = 'N';
1140		goto out;
1141	}
1142
1143	if (tc & BLK_TC_FLUSH)
1144		rwbs[i++] = 'F';
1145
1146	if (tc & BLK_TC_DISCARD)
1147		rwbs[i++] = 'D';
1148	else if (tc & BLK_TC_WRITE)
1149		rwbs[i++] = 'W';
1150	else if (t->bytes)
1151		rwbs[i++] = 'R';
1152	else
1153		rwbs[i++] = 'N';
1154
1155	if (tc & BLK_TC_FUA)
1156		rwbs[i++] = 'F';
1157	if (tc & BLK_TC_AHEAD)
1158		rwbs[i++] = 'A';
1159	if (tc & BLK_TC_SYNC)
1160		rwbs[i++] = 'S';
1161	if (tc & BLK_TC_META)
1162		rwbs[i++] = 'M';
1163out:
1164	rwbs[i] = '\0';
1165}
1166
1167static inline
1168const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1169{
1170	return (const struct blk_io_trace *)ent;
1171}
1172
1173static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1174{
1175	return (void *)(te_blk_io_trace(ent) + 1) +
1176		(has_cg ? sizeof(union kernfs_node_id) : 0);
1177}
1178
1179static inline const void *cgid_start(const struct trace_entry *ent)
1180{
1181	return (void *)(te_blk_io_trace(ent) + 1);
1182}
1183
1184static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1185{
1186	return te_blk_io_trace(ent)->pdu_len -
1187			(has_cg ? sizeof(union kernfs_node_id) : 0);
1188}
1189
1190static inline u32 t_action(const struct trace_entry *ent)
1191{
1192	return te_blk_io_trace(ent)->action;
1193}
1194
1195static inline u32 t_bytes(const struct trace_entry *ent)
1196{
1197	return te_blk_io_trace(ent)->bytes;
1198}
1199
1200static inline u32 t_sec(const struct trace_entry *ent)
1201{
1202	return te_blk_io_trace(ent)->bytes >> 9;
1203}
1204
1205static inline unsigned long long t_sector(const struct trace_entry *ent)
1206{
1207	return te_blk_io_trace(ent)->sector;
1208}
1209
1210static inline __u16 t_error(const struct trace_entry *ent)
1211{
1212	return te_blk_io_trace(ent)->error;
1213}
1214
1215static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1216{
1217	const __u64 *val = pdu_start(ent, has_cg);
1218	return be64_to_cpu(*val);
1219}
1220
1221static void get_pdu_remap(const struct trace_entry *ent,
1222			  struct blk_io_trace_remap *r, bool has_cg)
1223{
1224	const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1225	__u64 sector_from = __r->sector_from;
1226
1227	r->device_from = be32_to_cpu(__r->device_from);
1228	r->device_to   = be32_to_cpu(__r->device_to);
1229	r->sector_from = be64_to_cpu(sector_from);
1230}
1231
1232typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1233	bool has_cg);
1234
1235static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1236	bool has_cg)
1237{
1238	char rwbs[RWBS_LEN];
1239	unsigned long long ts  = iter->ts;
1240	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1241	unsigned secs	       = (unsigned long)ts;
1242	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1243
1244	fill_rwbs(rwbs, t);
1245
1246	trace_seq_printf(&iter->seq,
1247			 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1248			 MAJOR(t->device), MINOR(t->device), iter->cpu,
1249			 secs, nsec_rem, iter->ent->pid, act, rwbs);
1250}
1251
1252static void blk_log_action(struct trace_iterator *iter, const char *act,
1253	bool has_cg)
1254{
1255	char rwbs[RWBS_LEN];
1256	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1257
1258	fill_rwbs(rwbs, t);
1259	if (has_cg) {
1260		const union kernfs_node_id *id = cgid_start(iter->ent);
1261
1262		if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1263			char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1264
1265			cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1266				sizeof(blkcg_name_buf));
1267			trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1268				 MAJOR(t->device), MINOR(t->device),
1269				 blkcg_name_buf, act, rwbs);
1270		} else
 
 
 
 
 
 
 
 
 
 
 
 
 
1271			trace_seq_printf(&iter->seq,
1272				 "%3d,%-3d %x,%-x %2s %3s ",
1273				 MAJOR(t->device), MINOR(t->device),
1274				 id->ino, id->generation, act, rwbs);
 
1275	} else
1276		trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1277				 MAJOR(t->device), MINOR(t->device), act, rwbs);
1278}
1279
1280static void blk_log_dump_pdu(struct trace_seq *s,
1281	const struct trace_entry *ent, bool has_cg)
1282{
1283	const unsigned char *pdu_buf;
1284	int pdu_len;
1285	int i, end;
1286
1287	pdu_buf = pdu_start(ent, has_cg);
1288	pdu_len = pdu_real_len(ent, has_cg);
1289
1290	if (!pdu_len)
1291		return;
1292
1293	/* find the last zero that needs to be printed */
1294	for (end = pdu_len - 1; end >= 0; end--)
1295		if (pdu_buf[end])
1296			break;
1297	end++;
1298
1299	trace_seq_putc(s, '(');
1300
1301	for (i = 0; i < pdu_len; i++) {
1302
1303		trace_seq_printf(s, "%s%02x",
1304				 i == 0 ? "" : " ", pdu_buf[i]);
1305
1306		/*
1307		 * stop when the rest is just zeroes and indicate so
1308		 * with a ".." appended
1309		 */
1310		if (i == end && end != pdu_len - 1) {
1311			trace_seq_puts(s, " ..) ");
1312			return;
1313		}
1314	}
1315
1316	trace_seq_puts(s, ") ");
1317}
1318
1319static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1320{
1321	char cmd[TASK_COMM_LEN];
1322
1323	trace_find_cmdline(ent->pid, cmd);
1324
1325	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1326		trace_seq_printf(s, "%u ", t_bytes(ent));
1327		blk_log_dump_pdu(s, ent, has_cg);
1328		trace_seq_printf(s, "[%s]\n", cmd);
1329	} else {
1330		if (t_sec(ent))
1331			trace_seq_printf(s, "%llu + %u [%s]\n",
1332						t_sector(ent), t_sec(ent), cmd);
1333		else
1334			trace_seq_printf(s, "[%s]\n", cmd);
1335	}
1336}
1337
1338static void blk_log_with_error(struct trace_seq *s,
1339			      const struct trace_entry *ent, bool has_cg)
1340{
1341	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1342		blk_log_dump_pdu(s, ent, has_cg);
1343		trace_seq_printf(s, "[%d]\n", t_error(ent));
1344	} else {
1345		if (t_sec(ent))
1346			trace_seq_printf(s, "%llu + %u [%d]\n",
1347					 t_sector(ent),
1348					 t_sec(ent), t_error(ent));
1349		else
1350			trace_seq_printf(s, "%llu [%d]\n",
1351					 t_sector(ent), t_error(ent));
1352	}
1353}
1354
1355static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1356{
1357	struct blk_io_trace_remap r = { .device_from = 0, };
1358
1359	get_pdu_remap(ent, &r, has_cg);
1360	trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1361			 t_sector(ent), t_sec(ent),
1362			 MAJOR(r.device_from), MINOR(r.device_from),
1363			 (unsigned long long)r.sector_from);
 
1364}
1365
1366static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1367{
1368	char cmd[TASK_COMM_LEN];
1369
1370	trace_find_cmdline(ent->pid, cmd);
1371
1372	trace_seq_printf(s, "[%s]\n", cmd);
1373}
1374
1375static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1376{
1377	char cmd[TASK_COMM_LEN];
1378
1379	trace_find_cmdline(ent->pid, cmd);
1380
1381	trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1382}
1383
1384static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1385{
1386	char cmd[TASK_COMM_LEN];
1387
1388	trace_find_cmdline(ent->pid, cmd);
1389
1390	trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1391			 get_pdu_int(ent, has_cg), cmd);
1392}
1393
1394static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1395			bool has_cg)
1396{
1397
1398	trace_seq_putmem(s, pdu_start(ent, has_cg),
1399		pdu_real_len(ent, has_cg));
1400	trace_seq_putc(s, '\n');
1401}
1402
1403/*
1404 * struct tracer operations
1405 */
1406
1407static void blk_tracer_print_header(struct seq_file *m)
1408{
1409	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1410		return;
1411	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1412		    "#  |     |     |           |   |   |\n");
1413}
1414
1415static void blk_tracer_start(struct trace_array *tr)
1416{
1417	blk_tracer_enabled = true;
1418}
1419
1420static int blk_tracer_init(struct trace_array *tr)
1421{
1422	blk_tr = tr;
1423	blk_tracer_start(tr);
1424	return 0;
1425}
1426
1427static void blk_tracer_stop(struct trace_array *tr)
1428{
1429	blk_tracer_enabled = false;
1430}
1431
1432static void blk_tracer_reset(struct trace_array *tr)
1433{
1434	blk_tracer_stop(tr);
1435}
1436
1437static const struct {
1438	const char *act[2];
1439	void	   (*print)(struct trace_seq *s, const struct trace_entry *ent,
1440			    bool has_cg);
1441} what2act[] = {
1442	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1443	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1444	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1445	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1446	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1447	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1448	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1449	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1450	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1451	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1452	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1453	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1454	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1455	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1456	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1457};
1458
1459static enum print_line_t print_one_line(struct trace_iterator *iter,
1460					bool classic)
1461{
1462	struct trace_array *tr = iter->tr;
1463	struct trace_seq *s = &iter->seq;
1464	const struct blk_io_trace *t;
1465	u16 what;
1466	bool long_act;
1467	blk_log_action_t *log_action;
1468	bool has_cg;
1469
1470	t	   = te_blk_io_trace(iter->ent);
1471	what	   = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1472	long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1473	log_action = classic ? &blk_log_action_classic : &blk_log_action;
1474	has_cg	   = t->action & __BLK_TA_CGROUP;
1475
1476	if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1477		log_action(iter, long_act ? "message" : "m", has_cg);
1478		blk_log_msg(s, iter->ent, has_cg);
1479		return trace_handle_return(s);
1480	}
1481
1482	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1483		trace_seq_printf(s, "Unknown action %x\n", what);
1484	else {
1485		log_action(iter, what2act[what].act[long_act], has_cg);
1486		what2act[what].print(s, iter->ent, has_cg);
1487	}
1488
1489	return trace_handle_return(s);
1490}
1491
1492static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1493					       int flags, struct trace_event *event)
1494{
1495	return print_one_line(iter, false);
1496}
1497
1498static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1499{
1500	struct trace_seq *s = &iter->seq;
1501	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1502	const int offset = offsetof(struct blk_io_trace, sector);
1503	struct blk_io_trace old = {
1504		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1505		.time     = iter->ts,
1506	};
1507
1508	trace_seq_putmem(s, &old, offset);
1509	trace_seq_putmem(s, &t->sector,
1510			 sizeof(old) - offset + t->pdu_len);
1511}
1512
1513static enum print_line_t
1514blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1515			     struct trace_event *event)
1516{
1517	blk_trace_synthesize_old_trace(iter);
1518
1519	return trace_handle_return(&iter->seq);
1520}
1521
1522static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1523{
1524	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
 
1525		return TRACE_TYPE_UNHANDLED;
1526
1527	return print_one_line(iter, true);
1528}
1529
1530static int
1531blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1532{
1533	/* don't output context-info for blk_classic output */
1534	if (bit == TRACE_BLK_OPT_CLASSIC) {
1535		if (set)
1536			tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1537		else
1538			tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1539	}
1540	return 0;
1541}
1542
1543static struct tracer blk_tracer __read_mostly = {
1544	.name		= "blk",
1545	.init		= blk_tracer_init,
1546	.reset		= blk_tracer_reset,
1547	.start		= blk_tracer_start,
1548	.stop		= blk_tracer_stop,
1549	.print_header	= blk_tracer_print_header,
1550	.print_line	= blk_tracer_print_line,
1551	.flags		= &blk_tracer_flags,
1552	.set_flag	= blk_tracer_set_flag,
1553};
1554
1555static struct trace_event_functions trace_blk_event_funcs = {
1556	.trace		= blk_trace_event_print,
1557	.binary		= blk_trace_event_print_binary,
1558};
1559
1560static struct trace_event trace_blk_event = {
1561	.type		= TRACE_BLK,
1562	.funcs		= &trace_blk_event_funcs,
1563};
1564
1565static int __init init_blk_tracer(void)
1566{
1567	if (!register_trace_event(&trace_blk_event)) {
1568		pr_warn("Warning: could not register block events\n");
1569		return 1;
1570	}
1571
1572	if (register_tracer(&blk_tracer) != 0) {
1573		pr_warn("Warning: could not register the block tracer\n");
1574		unregister_trace_event(&trace_blk_event);
1575		return 1;
1576	}
1577
1578	return 0;
1579}
1580
1581device_initcall(init_blk_tracer);
1582
1583static int blk_trace_remove_queue(struct request_queue *q)
1584{
1585	struct blk_trace *bt;
1586
1587	bt = xchg(&q->blk_trace, NULL);
 
1588	if (bt == NULL)
1589		return -EINVAL;
1590
 
 
1591	put_probe_ref();
1592	blk_trace_free(bt);
 
1593	return 0;
1594}
1595
1596/*
1597 * Setup everything required to start tracing
1598 */
1599static int blk_trace_setup_queue(struct request_queue *q,
1600				 struct block_device *bdev)
1601{
1602	struct blk_trace *bt = NULL;
1603	int ret = -ENOMEM;
1604
1605	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1606	if (!bt)
1607		return -ENOMEM;
1608
1609	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1610	if (!bt->msg_data)
1611		goto free_bt;
1612
1613	bt->dev = bdev->bd_dev;
1614	bt->act_mask = (u16)-1;
1615
1616	blk_trace_setup_lba(bt, bdev);
1617
1618	ret = -EBUSY;
1619	if (cmpxchg(&q->blk_trace, NULL, bt))
1620		goto free_bt;
1621
1622	get_probe_ref();
1623	return 0;
1624
1625free_bt:
1626	blk_trace_free(bt);
1627	return ret;
1628}
1629
1630/*
1631 * sysfs interface to enable and configure tracing
1632 */
1633
1634static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1635					 struct device_attribute *attr,
1636					 char *buf);
1637static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1638					  struct device_attribute *attr,
1639					  const char *buf, size_t count);
1640#define BLK_TRACE_DEVICE_ATTR(_name) \
1641	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1642		    sysfs_blk_trace_attr_show, \
1643		    sysfs_blk_trace_attr_store)
1644
1645static BLK_TRACE_DEVICE_ATTR(enable);
1646static BLK_TRACE_DEVICE_ATTR(act_mask);
1647static BLK_TRACE_DEVICE_ATTR(pid);
1648static BLK_TRACE_DEVICE_ATTR(start_lba);
1649static BLK_TRACE_DEVICE_ATTR(end_lba);
1650
1651static struct attribute *blk_trace_attrs[] = {
1652	&dev_attr_enable.attr,
1653	&dev_attr_act_mask.attr,
1654	&dev_attr_pid.attr,
1655	&dev_attr_start_lba.attr,
1656	&dev_attr_end_lba.attr,
1657	NULL
1658};
1659
1660struct attribute_group blk_trace_attr_group = {
1661	.name  = "trace",
1662	.attrs = blk_trace_attrs,
1663};
1664
1665static const struct {
1666	int mask;
1667	const char *str;
1668} mask_maps[] = {
1669	{ BLK_TC_READ,		"read"		},
1670	{ BLK_TC_WRITE,		"write"		},
1671	{ BLK_TC_FLUSH,		"flush"		},
1672	{ BLK_TC_SYNC,		"sync"		},
1673	{ BLK_TC_QUEUE,		"queue"		},
1674	{ BLK_TC_REQUEUE,	"requeue"	},
1675	{ BLK_TC_ISSUE,		"issue"		},
1676	{ BLK_TC_COMPLETE,	"complete"	},
1677	{ BLK_TC_FS,		"fs"		},
1678	{ BLK_TC_PC,		"pc"		},
1679	{ BLK_TC_NOTIFY,	"notify"	},
1680	{ BLK_TC_AHEAD,		"ahead"		},
1681	{ BLK_TC_META,		"meta"		},
1682	{ BLK_TC_DISCARD,	"discard"	},
1683	{ BLK_TC_DRV_DATA,	"drv_data"	},
1684	{ BLK_TC_FUA,		"fua"		},
1685};
1686
1687static int blk_trace_str2mask(const char *str)
1688{
1689	int i;
1690	int mask = 0;
1691	char *buf, *s, *token;
1692
1693	buf = kstrdup(str, GFP_KERNEL);
1694	if (buf == NULL)
1695		return -ENOMEM;
1696	s = strstrip(buf);
1697
1698	while (1) {
1699		token = strsep(&s, ",");
1700		if (token == NULL)
1701			break;
1702
1703		if (*token == '\0')
1704			continue;
1705
1706		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1707			if (strcasecmp(token, mask_maps[i].str) == 0) {
1708				mask |= mask_maps[i].mask;
1709				break;
1710			}
1711		}
1712		if (i == ARRAY_SIZE(mask_maps)) {
1713			mask = -EINVAL;
1714			break;
1715		}
1716	}
1717	kfree(buf);
1718
1719	return mask;
1720}
1721
1722static ssize_t blk_trace_mask2str(char *buf, int mask)
1723{
1724	int i;
1725	char *p = buf;
1726
1727	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1728		if (mask & mask_maps[i].mask) {
1729			p += sprintf(p, "%s%s",
1730				    (p == buf) ? "" : ",", mask_maps[i].str);
1731		}
1732	}
1733	*p++ = '\n';
1734
1735	return p - buf;
1736}
1737
1738static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1739{
1740	if (bdev->bd_disk == NULL)
1741		return NULL;
1742
1743	return bdev_get_queue(bdev);
1744}
1745
1746static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1747					 struct device_attribute *attr,
1748					 char *buf)
1749{
1750	struct hd_struct *p = dev_to_part(dev);
1751	struct request_queue *q;
1752	struct block_device *bdev;
1753	ssize_t ret = -ENXIO;
1754
1755	bdev = bdget(part_devt(p));
1756	if (bdev == NULL)
1757		goto out;
1758
1759	q = blk_trace_get_queue(bdev);
1760	if (q == NULL)
1761		goto out_bdput;
1762
1763	mutex_lock(&q->blk_trace_mutex);
1764
 
 
1765	if (attr == &dev_attr_enable) {
1766		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1767		goto out_unlock_bdev;
1768	}
1769
1770	if (q->blk_trace == NULL)
1771		ret = sprintf(buf, "disabled\n");
1772	else if (attr == &dev_attr_act_mask)
1773		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1774	else if (attr == &dev_attr_pid)
1775		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1776	else if (attr == &dev_attr_start_lba)
1777		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1778	else if (attr == &dev_attr_end_lba)
1779		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1780
1781out_unlock_bdev:
1782	mutex_unlock(&q->blk_trace_mutex);
1783out_bdput:
1784	bdput(bdev);
1785out:
1786	return ret;
1787}
1788
1789static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1790					  struct device_attribute *attr,
1791					  const char *buf, size_t count)
1792{
1793	struct block_device *bdev;
1794	struct request_queue *q;
1795	struct hd_struct *p;
1796	u64 value;
1797	ssize_t ret = -EINVAL;
1798
1799	if (count == 0)
1800		goto out;
1801
1802	if (attr == &dev_attr_act_mask) {
1803		if (kstrtoull(buf, 0, &value)) {
1804			/* Assume it is a list of trace category names */
1805			ret = blk_trace_str2mask(buf);
1806			if (ret < 0)
1807				goto out;
1808			value = ret;
1809		}
1810	} else if (kstrtoull(buf, 0, &value))
1811		goto out;
 
 
1812
1813	ret = -ENXIO;
1814
1815	p = dev_to_part(dev);
1816	bdev = bdget(part_devt(p));
1817	if (bdev == NULL)
1818		goto out;
1819
1820	q = blk_trace_get_queue(bdev);
1821	if (q == NULL)
1822		goto out_bdput;
1823
1824	mutex_lock(&q->blk_trace_mutex);
1825
 
 
1826	if (attr == &dev_attr_enable) {
1827		if (!!value == !!q->blk_trace) {
1828			ret = 0;
1829			goto out_unlock_bdev;
1830		}
1831		if (value)
1832			ret = blk_trace_setup_queue(q, bdev);
1833		else
1834			ret = blk_trace_remove_queue(q);
1835		goto out_unlock_bdev;
1836	}
1837
1838	ret = 0;
1839	if (q->blk_trace == NULL)
1840		ret = blk_trace_setup_queue(q, bdev);
 
 
 
1841
1842	if (ret == 0) {
1843		if (attr == &dev_attr_act_mask)
1844			q->blk_trace->act_mask = value;
1845		else if (attr == &dev_attr_pid)
1846			q->blk_trace->pid = value;
1847		else if (attr == &dev_attr_start_lba)
1848			q->blk_trace->start_lba = value;
1849		else if (attr == &dev_attr_end_lba)
1850			q->blk_trace->end_lba = value;
1851	}
1852
1853out_unlock_bdev:
1854	mutex_unlock(&q->blk_trace_mutex);
1855out_bdput:
1856	bdput(bdev);
1857out:
1858	return ret ? ret : count;
1859}
1860
1861int blk_trace_init_sysfs(struct device *dev)
1862{
1863	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1864}
1865
1866void blk_trace_remove_sysfs(struct device *dev)
1867{
1868	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1869}
1870
1871#endif /* CONFIG_BLK_DEV_IO_TRACE */
1872
1873#ifdef CONFIG_EVENT_TRACING
1874
1875void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
 
 
 
 
 
 
 
 
 
 
1876{
1877	int i = 0;
1878
1879	if (op & REQ_PREFLUSH)
1880		rwbs[i++] = 'F';
1881
1882	switch (op & REQ_OP_MASK) {
1883	case REQ_OP_WRITE:
1884	case REQ_OP_WRITE_SAME:
1885		rwbs[i++] = 'W';
1886		break;
1887	case REQ_OP_DISCARD:
1888		rwbs[i++] = 'D';
1889		break;
1890	case REQ_OP_SECURE_ERASE:
1891		rwbs[i++] = 'D';
1892		rwbs[i++] = 'E';
1893		break;
1894	case REQ_OP_FLUSH:
1895		rwbs[i++] = 'F';
1896		break;
1897	case REQ_OP_READ:
1898		rwbs[i++] = 'R';
1899		break;
1900	default:
1901		rwbs[i++] = 'N';
1902	}
1903
1904	if (op & REQ_FUA)
1905		rwbs[i++] = 'F';
1906	if (op & REQ_RAHEAD)
1907		rwbs[i++] = 'A';
1908	if (op & REQ_SYNC)
1909		rwbs[i++] = 'S';
1910	if (op & REQ_META)
1911		rwbs[i++] = 'M';
1912
1913	rwbs[i] = '\0';
1914}
1915EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1916
1917#endif /* CONFIG_EVENT_TRACING */
1918