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
v3.15
 
   1/*
   2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License version 2 as
   6 * published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11 * GNU General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public License
  14 * along with this program; if not, write to the Free Software
  15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16 *
  17 */
 
 
 
  18#include <linux/kernel.h>
  19#include <linux/blkdev.h>
  20#include <linux/blktrace_api.h>
  21#include <linux/percpu.h>
  22#include <linux/init.h>
  23#include <linux/mutex.h>
  24#include <linux/slab.h>
  25#include <linux/debugfs.h>
  26#include <linux/export.h>
  27#include <linux/time.h>
  28#include <linux/uaccess.h>
  29#include <linux/list.h>
 
 
 
  30
  31#include <trace/events/block.h>
  32
  33#include "trace_output.h"
  34
  35#ifdef CONFIG_BLK_DEV_IO_TRACE
  36
  37static unsigned int blktrace_seq __read_mostly = 1;
  38
  39static struct trace_array *blk_tr;
  40static bool blk_tracer_enabled __read_mostly;
  41
  42static LIST_HEAD(running_trace_list);
  43static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  44
  45/* Select an alternative, minimalistic output than the original one */
  46#define TRACE_BLK_OPT_CLASSIC	0x1
 
 
  47
  48static struct tracer_opt blk_tracer_opts[] = {
  49	/* Default disable the minimalistic output */
  50	{ TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
 
 
 
 
  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 atomic_t blk_probes_ref = ATOMIC_INIT(0);
 
  61
  62static void blk_register_tracepoints(void);
  63static void blk_unregister_tracepoints(void);
  64
  65/*
  66 * Send out a notify message.
  67 */
  68static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  69		       const void *data, size_t len)
  70{
  71	struct blk_io_trace *t;
  72	struct ring_buffer_event *event = NULL;
  73	struct ring_buffer *buffer = NULL;
  74	int pc = 0;
  75	int cpu = smp_processor_id();
  76	bool blk_tracer = blk_tracer_enabled;
 
  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,
  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);
  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;
 100		t->pid = pid;
 101		t->cpu = cpu;
 102		t->pdu_len = len;
 103		memcpy((void *) t + sizeof(*t), data, len);
 
 
 104
 105		if (blk_tracer)
 106			trace_buffer_unlock_commit(buffer, event, 0, pc);
 107	}
 108}
 109
 110/*
 111 * Send out a notify for this process, if we haven't done so since a trace
 112 * started
 113 */
 114static void trace_note_tsk(struct task_struct *tsk)
 115{
 116	unsigned long flags;
 117	struct blk_trace *bt;
 118
 119	tsk->btrace_seq = blktrace_seq;
 120	spin_lock_irqsave(&running_trace_lock, flags);
 121	list_for_each_entry(bt, &running_trace_list, running_list) {
 122		trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 123			   sizeof(tsk->comm));
 124	}
 125	spin_unlock_irqrestore(&running_trace_lock, flags);
 126}
 127
 128static void trace_note_time(struct blk_trace *bt)
 129{
 130	struct timespec now;
 131	unsigned long flags;
 132	u32 words[2];
 133
 134	getnstimeofday(&now);
 135	words[0] = now.tv_sec;
 
 136	words[1] = now.tv_nsec;
 137
 138	local_irq_save(flags);
 139	trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 140	local_irq_restore(flags);
 141}
 142
 143void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 
 144{
 145	int n;
 146	va_list args;
 147	unsigned long flags;
 148	char *buf;
 
 149
 150	if (unlikely(bt->trace_state != Blktrace_running &&
 151		     !blk_tracer_enabled))
 152		return;
 153
 154	/*
 155	 * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 156	 * message to the trace.
 157	 */
 158	if (!(bt->act_mask & BLK_TC_NOTIFY))
 159		return;
 160
 161	local_irq_save(flags);
 162	buf = this_cpu_ptr(bt->msg_data);
 163	va_start(args, fmt);
 164	n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 165	va_end(args);
 166
 167	trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 
 
 
 
 
 
 168	local_irq_restore(flags);
 169}
 170EXPORT_SYMBOL_GPL(__trace_note_message);
 171
 172static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 173			 pid_t pid)
 174{
 175	if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 176		return 1;
 177	if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 178		return 1;
 179	if (bt->pid && pid != bt->pid)
 180		return 1;
 181
 182	return 0;
 183}
 184
 185/*
 186 * Data direction bit lookup
 187 */
 188static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 189				 BLK_TC_ACT(BLK_TC_WRITE) };
 190
 191#define BLK_TC_RAHEAD		BLK_TC_AHEAD
 
 192
 193/* The ilog2() calls fall out because they're constant */
 194#define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 195	  (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 196
 197/*
 198 * The worker for the various blk_add_trace*() types. Fills out a
 199 * blk_io_trace structure and places it in a per-cpu subbuffer.
 200 */
 201static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 202		     int rw, u32 what, int error, int pdu_len, void *pdu_data)
 
 203{
 204	struct task_struct *tsk = current;
 205	struct ring_buffer_event *event = NULL;
 206	struct ring_buffer *buffer = NULL;
 207	struct blk_io_trace *t;
 208	unsigned long flags = 0;
 209	unsigned long *sequence;
 
 210	pid_t pid;
 211	int cpu, pc = 0;
 212	bool blk_tracer = blk_tracer_enabled;
 
 
 213
 214	if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 215		return;
 216
 217	what |= ddir_act[rw & WRITE];
 218	what |= MASK_TC_BIT(rw, SYNC);
 219	what |= MASK_TC_BIT(rw, RAHEAD);
 220	what |= MASK_TC_BIT(rw, META);
 221	what |= MASK_TC_BIT(rw, DISCARD);
 222	what |= MASK_TC_BIT(rw, FLUSH);
 223	what |= MASK_TC_BIT(rw, FUA);
 
 
 
 
 
 224
 225	pid = tsk->pid;
 226	if (act_log_check(bt, what, sector, pid))
 227		return;
 228	cpu = raw_smp_processor_id();
 229
 230	if (blk_tracer) {
 231		tracing_record_cmdline(current);
 232
 233		buffer = blk_tr->trace_buffer.buffer;
 234		pc = preempt_count();
 235		event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 236						  sizeof(*t) + pdu_len,
 237						  0, pc);
 238		if (!event)
 239			return;
 240		t = ring_buffer_event_data(event);
 241		goto record_it;
 242	}
 243
 244	if (unlikely(tsk->btrace_seq != blktrace_seq))
 245		trace_note_tsk(tsk);
 246
 247	/*
 248	 * A word about the locking here - we disable interrupts to reserve
 249	 * some space in the relay per-cpu buffer, to prevent an irq
 250	 * from coming in and stepping on our toes.
 251	 */
 252	local_irq_save(flags);
 253	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 254	if (t) {
 255		sequence = per_cpu_ptr(bt->sequence, cpu);
 256
 257		t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 258		t->sequence = ++(*sequence);
 259		t->time = ktime_to_ns(ktime_get());
 260record_it:
 261		/*
 262		 * These two are not needed in ftrace as they are in the
 263		 * generic trace_entry, filled by tracing_generic_entry_update,
 264		 * but for the trace_event->bin() synthesizer benefit we do it
 265		 * here too.
 266		 */
 267		t->cpu = cpu;
 268		t->pid = pid;
 269
 270		t->sector = sector;
 271		t->bytes = bytes;
 272		t->action = what;
 273		t->device = bt->dev;
 274		t->error = error;
 275		t->pdu_len = pdu_len;
 276
 
 
 277		if (pdu_len)
 278			memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 279
 280		if (blk_tracer) {
 281			trace_buffer_unlock_commit(buffer, event, 0, pc);
 282			return;
 283		}
 284	}
 285
 286	local_irq_restore(flags);
 287}
 288
 289static struct dentry *blk_tree_root;
 290static DEFINE_MUTEX(blk_tree_mutex);
 291
 292static void blk_trace_free(struct blk_trace *bt)
 293{
 294	debugfs_remove(bt->msg_file);
 295	debugfs_remove(bt->dropped_file);
 296	relay_close(bt->rchan);
 297	debugfs_remove(bt->dir);
 
 
 
 
 
 
 
 
 
 
 298	free_percpu(bt->sequence);
 299	free_percpu(bt->msg_data);
 300	kfree(bt);
 301}
 302
 303static void blk_trace_cleanup(struct blk_trace *bt)
 
 
 
 
 
 
 
 
 304{
 305	blk_trace_free(bt);
 306	if (atomic_dec_and_test(&blk_probes_ref))
 307		blk_unregister_tracepoints();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 308}
 309
 310int blk_trace_remove(struct request_queue *q)
 
 
 
 
 
 
 
 
 311{
 312	struct blk_trace *bt;
 313
 314	bt = xchg(&q->blk_trace, NULL);
 
 315	if (!bt)
 316		return -EINVAL;
 317
 318	if (bt->trace_state != Blktrace_running)
 319		blk_trace_cleanup(bt);
 320
 321	return 0;
 322}
 
 
 
 
 
 
 
 
 
 
 
 323EXPORT_SYMBOL_GPL(blk_trace_remove);
 324
 325static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 326				size_t count, loff_t *ppos)
 327{
 328	struct blk_trace *bt = filp->private_data;
 329	char buf[16];
 330
 331	snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 332
 333	return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 334}
 335
 336static const struct file_operations blk_dropped_fops = {
 337	.owner =	THIS_MODULE,
 338	.open =		simple_open,
 339	.read =		blk_dropped_read,
 340	.llseek =	default_llseek,
 341};
 342
 343static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 344				size_t count, loff_t *ppos)
 345{
 346	char *msg;
 347	struct blk_trace *bt;
 348
 349	if (count >= BLK_TN_MAX_MSG)
 350		return -EINVAL;
 351
 352	msg = kmalloc(count + 1, GFP_KERNEL);
 353	if (msg == NULL)
 354		return -ENOMEM;
 355
 356	if (copy_from_user(msg, buffer, count)) {
 357		kfree(msg);
 358		return -EFAULT;
 359	}
 360
 361	msg[count] = '\0';
 362	bt = filp->private_data;
 363	__trace_note_message(bt, "%s", msg);
 364	kfree(msg);
 365
 366	return count;
 367}
 368
 369static const struct file_operations blk_msg_fops = {
 370	.owner =	THIS_MODULE,
 371	.open =		simple_open,
 372	.write =	blk_msg_write,
 373	.llseek =	noop_llseek,
 374};
 375
 376/*
 377 * Keep track of how many times we encountered a full subbuffer, to aid
 378 * the user space app in telling how many lost events there were.
 379 */
 380static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 381				     void *prev_subbuf, size_t prev_padding)
 382{
 383	struct blk_trace *bt;
 384
 385	if (!relay_buf_full(buf))
 386		return 1;
 387
 388	bt = buf->chan->private_data;
 389	atomic_inc(&bt->dropped);
 390	return 0;
 391}
 392
 393static int blk_remove_buf_file_callback(struct dentry *dentry)
 394{
 395	debugfs_remove(dentry);
 396
 397	return 0;
 398}
 399
 400static struct dentry *blk_create_buf_file_callback(const char *filename,
 401						   struct dentry *parent,
 402						   umode_t mode,
 403						   struct rchan_buf *buf,
 404						   int *is_global)
 405{
 406	return debugfs_create_file(filename, mode, parent, buf,
 407					&relay_file_operations);
 408}
 409
 410static struct rchan_callbacks blk_relay_callbacks = {
 411	.subbuf_start		= blk_subbuf_start_callback,
 412	.create_buf_file	= blk_create_buf_file_callback,
 413	.remove_buf_file	= blk_remove_buf_file_callback,
 414};
 415
 416static void blk_trace_setup_lba(struct blk_trace *bt,
 417				struct block_device *bdev)
 418{
 419	struct hd_struct *part = NULL;
 420
 421	if (bdev)
 422		part = bdev->bd_part;
 423
 424	if (part) {
 425		bt->start_lba = part->start_sect;
 426		bt->end_lba = part->start_sect + part->nr_sects;
 427	} else {
 428		bt->start_lba = 0;
 429		bt->end_lba = -1ULL;
 430	}
 431}
 432
 433/*
 434 * Setup everything required to start tracing
 435 */
 436int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 437		       struct block_device *bdev,
 438		       struct blk_user_trace_setup *buts)
 439{
 440	struct blk_trace *old_bt, *bt = NULL;
 441	struct dentry *dir = NULL;
 442	int ret, i;
 
 
 443
 444	if (!buts->buf_size || !buts->buf_nr)
 445		return -EINVAL;
 446
 447	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 448	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 449
 450	/*
 451	 * some device names have larger paths - convert the slashes
 452	 * to underscores for this to work as expected
 453	 */
 454	for (i = 0; i < strlen(buts->name); i++)
 455		if (buts->name[i] == '/')
 456			buts->name[i] = '_';
 
 
 
 
 
 
 
 
 
 457
 458	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 459	if (!bt)
 460		return -ENOMEM;
 461
 462	ret = -ENOMEM;
 463	bt->sequence = alloc_percpu(unsigned long);
 464	if (!bt->sequence)
 465		goto err;
 466
 467	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 468	if (!bt->msg_data)
 469		goto err;
 470
 471	ret = -ENOENT;
 
 
 
 
 
 
 
 
 
 472
 473	mutex_lock(&blk_tree_mutex);
 474	if (!blk_tree_root) {
 475		blk_tree_root = debugfs_create_dir("block", NULL);
 476		if (!blk_tree_root) {
 477			mutex_unlock(&blk_tree_mutex);
 478			goto err;
 479		}
 
 
 
 480	}
 481	mutex_unlock(&blk_tree_mutex);
 482
 483	dir = debugfs_create_dir(buts->name, blk_tree_root);
 484
 485	if (!dir)
 486		goto err;
 487
 488	bt->dir = dir;
 489	bt->dev = dev;
 490	atomic_set(&bt->dropped, 0);
 491	INIT_LIST_HEAD(&bt->running_list);
 492
 493	ret = -EIO;
 494	bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 495					       &blk_dropped_fops);
 496	if (!bt->dropped_file)
 497		goto err;
 498
 499	bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 500	if (!bt->msg_file)
 501		goto err;
 502
 503	bt->rchan = relay_open("trace", dir, buts->buf_size,
 504				buts->buf_nr, &blk_relay_callbacks, bt);
 505	if (!bt->rchan)
 506		goto err;
 507
 508	bt->act_mask = buts->act_mask;
 509	if (!bt->act_mask)
 510		bt->act_mask = (u16) -1;
 511
 512	blk_trace_setup_lba(bt, bdev);
 513
 514	/* overwrite with user settings */
 515	if (buts->start_lba)
 516		bt->start_lba = buts->start_lba;
 517	if (buts->end_lba)
 518		bt->end_lba = buts->end_lba;
 519
 520	bt->pid = buts->pid;
 521	bt->trace_state = Blktrace_setup;
 522
 523	ret = -EBUSY;
 524	old_bt = xchg(&q->blk_trace, bt);
 525	if (old_bt) {
 526		(void) xchg(&q->blk_trace, old_bt);
 527		goto err;
 528	}
 529
 530	if (atomic_inc_return(&blk_probes_ref) == 1)
 531		blk_register_tracepoints();
 532
 533	return 0;
 534err:
 535	blk_trace_free(bt);
 
 536	return ret;
 537}
 538
 539int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 540		    struct block_device *bdev,
 541		    char __user *arg)
 542{
 543	struct blk_user_trace_setup buts;
 544	int ret;
 545
 546	ret = copy_from_user(&buts, arg, sizeof(buts));
 547	if (ret)
 548		return -EFAULT;
 549
 550	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 551	if (ret)
 552		return ret;
 553
 554	if (copy_to_user(arg, &buts, sizeof(buts))) {
 555		blk_trace_remove(q);
 556		return -EFAULT;
 557	}
 558	return 0;
 559}
 
 
 
 
 
 
 
 
 
 
 
 
 
 560EXPORT_SYMBOL_GPL(blk_trace_setup);
 561
 562#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 563static int compat_blk_trace_setup(struct request_queue *q, char *name,
 564				  dev_t dev, struct block_device *bdev,
 565				  char __user *arg)
 566{
 567	struct blk_user_trace_setup buts;
 568	struct compat_blk_user_trace_setup cbuts;
 569	int ret;
 570
 571	if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 572		return -EFAULT;
 573
 574	buts = (struct blk_user_trace_setup) {
 575		.act_mask = cbuts.act_mask,
 576		.buf_size = cbuts.buf_size,
 577		.buf_nr = cbuts.buf_nr,
 578		.start_lba = cbuts.start_lba,
 579		.end_lba = cbuts.end_lba,
 580		.pid = cbuts.pid,
 581	};
 582
 583	ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 584	if (ret)
 585		return ret;
 586
 587	if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 588		blk_trace_remove(q);
 589		return -EFAULT;
 590	}
 591
 592	return 0;
 593}
 594#endif
 595
 596int blk_trace_startstop(struct request_queue *q, int start)
 597{
 598	int ret;
 599	struct blk_trace *bt = q->blk_trace;
 600
 
 
 601	if (bt == NULL)
 602		return -EINVAL;
 603
 604	/*
 605	 * For starting a trace, we can transition from a setup or stopped
 606	 * trace. For stopping a trace, the state must be running
 607	 */
 608	ret = -EINVAL;
 609	if (start) {
 610		if (bt->trace_state == Blktrace_setup ||
 611		    bt->trace_state == Blktrace_stopped) {
 612			blktrace_seq++;
 613			smp_mb();
 614			bt->trace_state = Blktrace_running;
 615			spin_lock_irq(&running_trace_lock);
 616			list_add(&bt->running_list, &running_trace_list);
 617			spin_unlock_irq(&running_trace_lock);
 618
 619			trace_note_time(bt);
 620			ret = 0;
 621		}
 622	} else {
 623		if (bt->trace_state == Blktrace_running) {
 624			bt->trace_state = Blktrace_stopped;
 625			spin_lock_irq(&running_trace_lock);
 626			list_del_init(&bt->running_list);
 627			spin_unlock_irq(&running_trace_lock);
 628			relay_flush(bt->rchan);
 629			ret = 0;
 630		}
 631	}
 632
 633	return ret;
 634}
 635EXPORT_SYMBOL_GPL(blk_trace_startstop);
 636
 
 
 
 
 
 
 637/**
 638 * blk_trace_ioctl: - handle the ioctls associated with tracing
 639 * @bdev:	the block device
 640 * @cmd:	the ioctl cmd
 641 * @arg:	the argument data, if any
 642 *
 643 **/
 644int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 645{
 646	struct request_queue *q;
 647	int ret, start = 0;
 648	char b[BDEVNAME_SIZE];
 649
 650	q = bdev_get_queue(bdev);
 651	if (!q)
 652		return -ENXIO;
 653
 654	mutex_lock(&bdev->bd_mutex);
 655
 656	switch (cmd) {
 657	case BLKTRACESETUP:
 658		bdevname(bdev, b);
 659		ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 660		break;
 661#if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 662	case BLKTRACESETUP32:
 663		bdevname(bdev, b);
 664		ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 665		break;
 666#endif
 667	case BLKTRACESTART:
 668		start = 1;
 
 669	case BLKTRACESTOP:
 670		ret = blk_trace_startstop(q, start);
 671		break;
 672	case BLKTRACETEARDOWN:
 673		ret = blk_trace_remove(q);
 674		break;
 675	default:
 676		ret = -ENOTTY;
 677		break;
 678	}
 679
 680	mutex_unlock(&bdev->bd_mutex);
 681	return ret;
 682}
 683
 684/**
 685 * blk_trace_shutdown: - stop and cleanup trace structures
 686 * @q:    the request queue associated with the device
 687 *
 688 **/
 689void blk_trace_shutdown(struct request_queue *q)
 690{
 691	if (q->blk_trace) {
 692		blk_trace_startstop(q, 0);
 693		blk_trace_remove(q);
 694	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 695}
 696
 697/*
 698 * blktrace probes
 699 */
 700
 701/**
 702 * blk_add_trace_rq - Add a trace for a request oriented action
 703 * @q:		queue the io is for
 704 * @rq:		the source request
 
 705 * @nr_bytes:	number of completed bytes
 706 * @what:	the action
 
 707 *
 708 * Description:
 709 *     Records an action against a request. Will log the bio offset + size.
 710 *
 711 **/
 712static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 713			     unsigned int nr_bytes, u32 what)
 714{
 715	struct blk_trace *bt = q->blk_trace;
 716
 717	if (likely(!bt))
 
 
 
 718		return;
 
 719
 720	if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 721		what |= BLK_TC_ACT(BLK_TC_PC);
 722		__blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
 723				what, rq->errors, rq->cmd_len, rq->cmd);
 724	} else  {
 725		what |= BLK_TC_ACT(BLK_TC_FS);
 726		__blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
 727				rq->cmd_flags, what, rq->errors, 0, NULL);
 728	}
 
 729}
 730
 731static void blk_add_trace_rq_abort(void *ignore,
 732				   struct request_queue *q, struct request *rq)
 733{
 734	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
 
 735}
 736
 737static void blk_add_trace_rq_insert(void *ignore,
 738				    struct request_queue *q, struct request *rq)
 739{
 740	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
 
 741}
 742
 743static void blk_add_trace_rq_issue(void *ignore,
 744				   struct request_queue *q, struct request *rq)
 745{
 746	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
 
 747}
 748
 749static void blk_add_trace_rq_requeue(void *ignore,
 750				     struct request_queue *q,
 751				     struct request *rq)
 752{
 753	blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 
 754}
 755
 756static void blk_add_trace_rq_complete(void *ignore,
 757				      struct request_queue *q,
 758				      struct request *rq,
 759				      unsigned int nr_bytes)
 760{
 761	blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 
 762}
 763
 764/**
 765 * blk_add_trace_bio - Add a trace for a bio oriented action
 766 * @q:		queue the io is for
 767 * @bio:	the source bio
 768 * @what:	the action
 769 * @error:	error, if any
 770 *
 771 * Description:
 772 *     Records an action against a bio. Will log the bio offset + size.
 773 *
 774 **/
 775static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 776			      u32 what, int error)
 777{
 778	struct blk_trace *bt = q->blk_trace;
 779
 780	if (likely(!bt))
 
 
 
 781		return;
 782
 783	if (!error && !bio_flagged(bio, BIO_UPTODATE))
 784		error = EIO;
 785
 786	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 787			bio->bi_rw, what, error, 0, NULL);
 
 
 788}
 789
 790static void blk_add_trace_bio_bounce(void *ignore,
 791				     struct request_queue *q, struct bio *bio)
 792{
 793	blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 794}
 795
 796static void blk_add_trace_bio_complete(void *ignore,
 797				       struct request_queue *q, struct bio *bio,
 798				       int error)
 799{
 800	blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 
 801}
 802
 803static void blk_add_trace_bio_backmerge(void *ignore,
 804					struct request_queue *q,
 805					struct request *rq,
 806					struct bio *bio)
 807{
 808	blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 
 809}
 810
 811static void blk_add_trace_bio_frontmerge(void *ignore,
 812					 struct request_queue *q,
 813					 struct request *rq,
 814					 struct bio *bio)
 815{
 816	blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 
 817}
 818
 819static void blk_add_trace_bio_queue(void *ignore,
 820				    struct request_queue *q, struct bio *bio)
 821{
 822	blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 823}
 824
 825static void blk_add_trace_getrq(void *ignore,
 826				struct request_queue *q,
 827				struct bio *bio, int rw)
 828{
 829	if (bio)
 830		blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 831	else {
 832		struct blk_trace *bt = q->blk_trace;
 833
 834		if (bt)
 835			__blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 836	}
 837}
 838
 839
 840static void blk_add_trace_sleeprq(void *ignore,
 841				  struct request_queue *q,
 842				  struct bio *bio, int rw)
 843{
 844	if (bio)
 845		blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 846	else {
 847		struct blk_trace *bt = q->blk_trace;
 848
 849		if (bt)
 850			__blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 851					0, 0, NULL);
 852	}
 853}
 854
 855static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 856{
 857	struct blk_trace *bt = q->blk_trace;
 858
 
 
 859	if (bt)
 860		__blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 
 861}
 862
 863static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 864				    unsigned int depth, bool explicit)
 865{
 866	struct blk_trace *bt = q->blk_trace;
 867
 
 
 868	if (bt) {
 869		__be64 rpdu = cpu_to_be64(depth);
 870		u32 what;
 871
 872		if (explicit)
 873			what = BLK_TA_UNPLUG_IO;
 874		else
 875			what = BLK_TA_UNPLUG_TIMER;
 876
 877		__blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 878	}
 
 879}
 880
 881static void blk_add_trace_split(void *ignore,
 882				struct request_queue *q, struct bio *bio,
 883				unsigned int pdu)
 884{
 885	struct blk_trace *bt = q->blk_trace;
 
 886
 
 
 887	if (bt) {
 888		__be64 rpdu = cpu_to_be64(pdu);
 889
 890		__blk_add_trace(bt, bio->bi_iter.bi_sector,
 891				bio->bi_iter.bi_size, bio->bi_rw, BLK_TA_SPLIT,
 892				!bio_flagged(bio, BIO_UPTODATE),
 893				sizeof(rpdu), &rpdu);
 
 894	}
 
 895}
 896
 897/**
 898 * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 899 * @ignore:	trace callback data parameter (not used)
 900 * @q:		queue the io is for
 901 * @bio:	the source bio
 902 * @dev:	target device
 903 * @from:	source sector
 904 *
 905 * Description:
 906 *     Device mapper or raid target sometimes need to split a bio because
 907 *     it spans a stripe (or similar). Add a trace for that action.
 908 *
 909 **/
 910static void blk_add_trace_bio_remap(void *ignore,
 911				    struct request_queue *q, struct bio *bio,
 912				    dev_t dev, sector_t from)
 913{
 914	struct blk_trace *bt = q->blk_trace;
 
 915	struct blk_io_trace_remap r;
 916
 917	if (likely(!bt))
 
 
 
 918		return;
 
 919
 920	r.device_from = cpu_to_be32(dev);
 921	r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 922	r.sector_from = cpu_to_be64(from);
 923
 924	__blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 925			bio->bi_rw, BLK_TA_REMAP,
 926			!bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 
 
 927}
 928
 929/**
 930 * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 931 * @ignore:	trace callback data parameter (not used)
 932 * @q:		queue the io is for
 933 * @rq:		the source request
 934 * @dev:	target device
 935 * @from:	source sector
 936 *
 937 * Description:
 938 *     Device mapper remaps request to other devices.
 939 *     Add a trace for that action.
 940 *
 941 **/
 942static void blk_add_trace_rq_remap(void *ignore,
 943				   struct request_queue *q,
 944				   struct request *rq, dev_t dev,
 945				   sector_t from)
 946{
 947	struct blk_trace *bt = q->blk_trace;
 948	struct blk_io_trace_remap r;
 949
 950	if (likely(!bt))
 
 
 
 951		return;
 
 952
 953	r.device_from = cpu_to_be32(dev);
 954	r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 955	r.sector_from = cpu_to_be64(from);
 956
 957	__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 958			rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 959			sizeof(r), &r);
 
 960}
 961
 962/**
 963 * blk_add_driver_data - Add binary message with driver-specific data
 964 * @q:		queue the io is for
 965 * @rq:		io request
 966 * @data:	driver-specific data
 967 * @len:	length of driver-specific data
 968 *
 969 * Description:
 970 *     Some drivers might want to write driver-specific data per request.
 971 *
 972 **/
 973void blk_add_driver_data(struct request_queue *q,
 974			 struct request *rq,
 975			 void *data, size_t len)
 976{
 977	struct blk_trace *bt = q->blk_trace;
 978
 979	if (likely(!bt))
 
 
 
 980		return;
 
 981
 982	if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 983		__blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 984				BLK_TA_DRV_DATA, rq->errors, len, data);
 985	else
 986		__blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 987				BLK_TA_DRV_DATA, rq->errors, len, data);
 988}
 989EXPORT_SYMBOL_GPL(blk_add_driver_data);
 990
 991static void blk_register_tracepoints(void)
 992{
 993	int ret;
 994
 995	ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 996	WARN_ON(ret);
 997	ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 998	WARN_ON(ret);
 999	ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1000	WARN_ON(ret);
 
 
1001	ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1002	WARN_ON(ret);
1003	ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1004	WARN_ON(ret);
1005	ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1006	WARN_ON(ret);
1007	ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1008	WARN_ON(ret);
1009	ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1010	WARN_ON(ret);
1011	ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1012	WARN_ON(ret);
1013	ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1014	WARN_ON(ret);
1015	ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1016	WARN_ON(ret);
1017	ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1018	WARN_ON(ret);
1019	ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1020	WARN_ON(ret);
1021	ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1022	WARN_ON(ret);
1023	ret = register_trace_block_split(blk_add_trace_split, NULL);
1024	WARN_ON(ret);
1025	ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1026	WARN_ON(ret);
1027	ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1028	WARN_ON(ret);
1029}
1030
1031static void blk_unregister_tracepoints(void)
1032{
1033	unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1034	unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1035	unregister_trace_block_split(blk_add_trace_split, NULL);
1036	unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1037	unregister_trace_block_plug(blk_add_trace_plug, NULL);
1038	unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1039	unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1040	unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1041	unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1042	unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1043	unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1044	unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1045	unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1046	unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 
1047	unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1048	unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1049	unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1050
1051	tracepoint_synchronize_unregister();
1052}
1053
1054/*
1055 * struct blk_io_tracer formatting routines
1056 */
1057
1058static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1059{
1060	int i = 0;
1061	int tc = t->action >> BLK_TC_SHIFT;
1062
1063	if (t->action == BLK_TN_MESSAGE) {
1064		rwbs[i++] = 'N';
1065		goto out;
1066	}
1067
1068	if (tc & BLK_TC_FLUSH)
1069		rwbs[i++] = 'F';
1070
1071	if (tc & BLK_TC_DISCARD)
1072		rwbs[i++] = 'D';
1073	else if (tc & BLK_TC_WRITE)
1074		rwbs[i++] = 'W';
1075	else if (t->bytes)
1076		rwbs[i++] = 'R';
1077	else
1078		rwbs[i++] = 'N';
1079
1080	if (tc & BLK_TC_FUA)
1081		rwbs[i++] = 'F';
1082	if (tc & BLK_TC_AHEAD)
1083		rwbs[i++] = 'A';
1084	if (tc & BLK_TC_SYNC)
1085		rwbs[i++] = 'S';
1086	if (tc & BLK_TC_META)
1087		rwbs[i++] = 'M';
1088out:
1089	rwbs[i] = '\0';
1090}
1091
1092static inline
1093const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1094{
1095	return (const struct blk_io_trace *)ent;
1096}
1097
1098static inline const void *pdu_start(const struct trace_entry *ent)
1099{
1100	return te_blk_io_trace(ent) + 1;
 
 
 
 
 
 
 
 
 
 
1101}
1102
1103static inline u32 t_action(const struct trace_entry *ent)
1104{
1105	return te_blk_io_trace(ent)->action;
1106}
1107
1108static inline u32 t_bytes(const struct trace_entry *ent)
1109{
1110	return te_blk_io_trace(ent)->bytes;
1111}
1112
1113static inline u32 t_sec(const struct trace_entry *ent)
1114{
1115	return te_blk_io_trace(ent)->bytes >> 9;
1116}
1117
1118static inline unsigned long long t_sector(const struct trace_entry *ent)
1119{
1120	return te_blk_io_trace(ent)->sector;
1121}
1122
1123static inline __u16 t_error(const struct trace_entry *ent)
1124{
1125	return te_blk_io_trace(ent)->error;
1126}
1127
1128static __u64 get_pdu_int(const struct trace_entry *ent)
1129{
1130	const __u64 *val = pdu_start(ent);
1131	return be64_to_cpu(*val);
1132}
1133
1134static void get_pdu_remap(const struct trace_entry *ent,
1135			  struct blk_io_trace_remap *r)
1136{
1137	const struct blk_io_trace_remap *__r = pdu_start(ent);
1138	__u64 sector_from = __r->sector_from;
1139
1140	r->device_from = be32_to_cpu(__r->device_from);
1141	r->device_to   = be32_to_cpu(__r->device_to);
1142	r->sector_from = be64_to_cpu(sector_from);
1143}
1144
1145typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1146
1147static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
 
1148{
1149	char rwbs[RWBS_LEN];
1150	unsigned long long ts  = iter->ts;
1151	unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1152	unsigned secs	       = (unsigned long)ts;
1153	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1154
1155	fill_rwbs(rwbs, t);
1156
1157	return trace_seq_printf(&iter->seq,
1158				"%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1159				MAJOR(t->device), MINOR(t->device), iter->cpu,
1160				secs, nsec_rem, iter->ent->pid, act, rwbs);
1161}
1162
1163static int blk_log_action(struct trace_iterator *iter, const char *act)
 
1164{
1165	char rwbs[RWBS_LEN];
1166	const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1167
1168	fill_rwbs(rwbs, t);
1169	return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1170				MAJOR(t->device), MINOR(t->device), act, rwbs);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1171}
1172
1173static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
 
1174{
1175	const unsigned char *pdu_buf;
1176	int pdu_len;
1177	int i, end, ret;
1178
1179	pdu_buf = pdu_start(ent);
1180	pdu_len = te_blk_io_trace(ent)->pdu_len;
1181
1182	if (!pdu_len)
1183		return 1;
1184
1185	/* find the last zero that needs to be printed */
1186	for (end = pdu_len - 1; end >= 0; end--)
1187		if (pdu_buf[end])
1188			break;
1189	end++;
1190
1191	if (!trace_seq_putc(s, '('))
1192		return 0;
1193
1194	for (i = 0; i < pdu_len; i++) {
1195
1196		ret = trace_seq_printf(s, "%s%02x",
1197				       i == 0 ? "" : " ", pdu_buf[i]);
1198		if (!ret)
1199			return ret;
1200
1201		/*
1202		 * stop when the rest is just zeroes and indicate so
1203		 * with a ".." appended
1204		 */
1205		if (i == end && end != pdu_len - 1)
1206			return trace_seq_puts(s, " ..) ");
 
 
1207	}
1208
1209	return trace_seq_puts(s, ") ");
1210}
1211
1212static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1213{
1214	char cmd[TASK_COMM_LEN];
1215
1216	trace_find_cmdline(ent->pid, cmd);
1217
1218	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1219		int ret;
1220
1221		ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1222		if (!ret)
1223			return 0;
1224		ret = blk_log_dump_pdu(s, ent);
1225		if (!ret)
1226			return 0;
1227		return trace_seq_printf(s, "[%s]\n", cmd);
1228	} else {
1229		if (t_sec(ent))
1230			return trace_seq_printf(s, "%llu + %u [%s]\n",
1231						t_sector(ent), t_sec(ent), cmd);
1232		return trace_seq_printf(s, "[%s]\n", cmd);
 
1233	}
1234}
1235
1236static int blk_log_with_error(struct trace_seq *s,
1237			      const struct trace_entry *ent)
1238{
1239	if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1240		int ret;
1241
1242		ret = blk_log_dump_pdu(s, ent);
1243		if (ret)
1244			return trace_seq_printf(s, "[%d]\n", t_error(ent));
1245		return 0;
1246	} else {
1247		if (t_sec(ent))
1248			return trace_seq_printf(s, "%llu + %u [%d]\n",
1249						t_sector(ent),
1250						t_sec(ent), t_error(ent));
1251		return trace_seq_printf(s, "%llu [%d]\n",
1252					t_sector(ent), t_error(ent));
 
1253	}
1254}
1255
1256static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1257{
1258	struct blk_io_trace_remap r = { .device_from = 0, };
1259
1260	get_pdu_remap(ent, &r);
1261	return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1262				t_sector(ent), t_sec(ent),
1263				MAJOR(r.device_from), MINOR(r.device_from),
1264				(unsigned long long)r.sector_from);
1265}
1266
1267static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1268{
1269	char cmd[TASK_COMM_LEN];
1270
1271	trace_find_cmdline(ent->pid, cmd);
1272
1273	return trace_seq_printf(s, "[%s]\n", cmd);
1274}
1275
1276static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1277{
1278	char cmd[TASK_COMM_LEN];
1279
1280	trace_find_cmdline(ent->pid, cmd);
1281
1282	return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1283}
1284
1285static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1286{
1287	char cmd[TASK_COMM_LEN];
1288
1289	trace_find_cmdline(ent->pid, cmd);
1290
1291	return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1292				get_pdu_int(ent), cmd);
1293}
1294
1295static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
 
1296{
1297	int ret;
1298	const struct blk_io_trace *t = te_blk_io_trace(ent);
1299
1300	ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1301	if (ret)
1302		return trace_seq_putc(s, '\n');
1303	return ret;
1304}
1305
1306/*
1307 * struct tracer operations
1308 */
1309
1310static void blk_tracer_print_header(struct seq_file *m)
1311{
1312	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1313		return;
1314	seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1315		    "#  |     |     |           |   |   |\n");
1316}
1317
1318static void blk_tracer_start(struct trace_array *tr)
1319{
1320	blk_tracer_enabled = true;
1321}
1322
1323static int blk_tracer_init(struct trace_array *tr)
1324{
1325	blk_tr = tr;
1326	blk_tracer_start(tr);
1327	return 0;
1328}
1329
1330static void blk_tracer_stop(struct trace_array *tr)
1331{
1332	blk_tracer_enabled = false;
1333}
1334
1335static void blk_tracer_reset(struct trace_array *tr)
1336{
1337	blk_tracer_stop(tr);
1338}
1339
1340static const struct {
1341	const char *act[2];
1342	int	   (*print)(struct trace_seq *s, const struct trace_entry *ent);
 
1343} what2act[] = {
1344	[__BLK_TA_QUEUE]	= {{  "Q", "queue" },	   blk_log_generic },
1345	[__BLK_TA_BACKMERGE]	= {{  "M", "backmerge" },  blk_log_generic },
1346	[__BLK_TA_FRONTMERGE]	= {{  "F", "frontmerge" }, blk_log_generic },
1347	[__BLK_TA_GETRQ]	= {{  "G", "getrq" },	   blk_log_generic },
1348	[__BLK_TA_SLEEPRQ]	= {{  "S", "sleeprq" },	   blk_log_generic },
1349	[__BLK_TA_REQUEUE]	= {{  "R", "requeue" },	   blk_log_with_error },
1350	[__BLK_TA_ISSUE]	= {{  "D", "issue" },	   blk_log_generic },
1351	[__BLK_TA_COMPLETE]	= {{  "C", "complete" },   blk_log_with_error },
1352	[__BLK_TA_PLUG]		= {{  "P", "plug" },	   blk_log_plug },
1353	[__BLK_TA_UNPLUG_IO]	= {{  "U", "unplug_io" },  blk_log_unplug },
1354	[__BLK_TA_UNPLUG_TIMER]	= {{ "UT", "unplug_timer" }, blk_log_unplug },
1355	[__BLK_TA_INSERT]	= {{  "I", "insert" },	   blk_log_generic },
1356	[__BLK_TA_SPLIT]	= {{  "X", "split" },	   blk_log_split },
1357	[__BLK_TA_BOUNCE]	= {{  "B", "bounce" },	   blk_log_generic },
1358	[__BLK_TA_REMAP]	= {{  "A", "remap" },	   blk_log_remap },
1359};
1360
1361static enum print_line_t print_one_line(struct trace_iterator *iter,
1362					bool classic)
1363{
 
1364	struct trace_seq *s = &iter->seq;
1365	const struct blk_io_trace *t;
1366	u16 what;
1367	int ret;
1368	bool long_act;
1369	blk_log_action_t *log_action;
 
1370
1371	t	   = te_blk_io_trace(iter->ent);
1372	what	   = t->action & ((1 << BLK_TC_SHIFT) - 1);
1373	long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1374	log_action = classic ? &blk_log_action_classic : &blk_log_action;
 
1375
1376	if (t->action == BLK_TN_MESSAGE) {
1377		ret = log_action(iter, long_act ? "message" : "m");
1378		if (ret)
1379			ret = blk_log_msg(s, iter->ent);
1380		goto out;
1381	}
1382
1383	if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1384		ret = trace_seq_printf(s, "Unknown action %x\n", what);
1385	else {
1386		ret = log_action(iter, what2act[what].act[long_act]);
1387		if (ret)
1388			ret = what2act[what].print(s, iter->ent);
1389	}
1390out:
1391	return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1392}
1393
1394static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1395					       int flags, struct trace_event *event)
1396{
1397	return print_one_line(iter, false);
1398}
1399
1400static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1401{
1402	struct trace_seq *s = &iter->seq;
1403	struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1404	const int offset = offsetof(struct blk_io_trace, sector);
1405	struct blk_io_trace old = {
1406		.magic	  = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1407		.time     = iter->ts,
1408	};
1409
1410	if (!trace_seq_putmem(s, &old, offset))
1411		return 0;
1412	return trace_seq_putmem(s, &t->sector,
1413				sizeof(old) - offset + t->pdu_len);
1414}
1415
1416static enum print_line_t
1417blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1418			     struct trace_event *event)
1419{
1420	return blk_trace_synthesize_old_trace(iter) ?
1421			TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 
1422}
1423
1424static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1425{
1426	if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
 
1427		return TRACE_TYPE_UNHANDLED;
1428
1429	return print_one_line(iter, true);
1430}
1431
1432static int
1433blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1434{
1435	/* don't output context-info for blk_classic output */
1436	if (bit == TRACE_BLK_OPT_CLASSIC) {
1437		if (set)
1438			trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1439		else
1440			trace_flags |= TRACE_ITER_CONTEXT_INFO;
1441	}
1442	return 0;
1443}
1444
1445static struct tracer blk_tracer __read_mostly = {
1446	.name		= "blk",
1447	.init		= blk_tracer_init,
1448	.reset		= blk_tracer_reset,
1449	.start		= blk_tracer_start,
1450	.stop		= blk_tracer_stop,
1451	.print_header	= blk_tracer_print_header,
1452	.print_line	= blk_tracer_print_line,
1453	.flags		= &blk_tracer_flags,
1454	.set_flag	= blk_tracer_set_flag,
1455};
1456
1457static struct trace_event_functions trace_blk_event_funcs = {
1458	.trace		= blk_trace_event_print,
1459	.binary		= blk_trace_event_print_binary,
1460};
1461
1462static struct trace_event trace_blk_event = {
1463	.type		= TRACE_BLK,
1464	.funcs		= &trace_blk_event_funcs,
1465};
1466
1467static int __init init_blk_tracer(void)
1468{
1469	if (!register_ftrace_event(&trace_blk_event)) {
1470		pr_warning("Warning: could not register block events\n");
1471		return 1;
1472	}
1473
1474	if (register_tracer(&blk_tracer) != 0) {
1475		pr_warning("Warning: could not register the block tracer\n");
1476		unregister_ftrace_event(&trace_blk_event);
1477		return 1;
1478	}
1479
1480	return 0;
1481}
1482
1483device_initcall(init_blk_tracer);
1484
1485static int blk_trace_remove_queue(struct request_queue *q)
1486{
1487	struct blk_trace *bt;
1488
1489	bt = xchg(&q->blk_trace, NULL);
 
1490	if (bt == NULL)
1491		return -EINVAL;
1492
1493	if (atomic_dec_and_test(&blk_probes_ref))
1494		blk_unregister_tracepoints();
1495
1496	spin_lock_irq(&running_trace_lock);
1497	list_del(&bt->running_list);
1498	spin_unlock_irq(&running_trace_lock);
1499	blk_trace_free(bt);
1500	return 0;
1501}
1502
1503/*
1504 * Setup everything required to start tracing
1505 */
1506static int blk_trace_setup_queue(struct request_queue *q,
1507				 struct block_device *bdev)
1508{
1509	struct blk_trace *old_bt, *bt = NULL;
1510	int ret = -ENOMEM;
1511
1512	bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1513	if (!bt)
1514		return -ENOMEM;
1515
1516	bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1517	if (!bt->msg_data)
1518		goto free_bt;
1519
1520	bt->dev = bdev->bd_dev;
1521	bt->act_mask = (u16)-1;
1522
1523	blk_trace_setup_lba(bt, bdev);
1524
1525	old_bt = xchg(&q->blk_trace, bt);
1526	if (old_bt != NULL) {
1527		(void)xchg(&q->blk_trace, old_bt);
1528		ret = -EBUSY;
1529		goto free_bt;
1530	}
1531
1532	if (atomic_inc_return(&blk_probes_ref) == 1)
1533		blk_register_tracepoints();
1534	return 0;
1535
1536free_bt:
1537	blk_trace_free(bt);
1538	return ret;
1539}
1540
1541/*
1542 * sysfs interface to enable and configure tracing
1543 */
1544
1545static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1546					 struct device_attribute *attr,
1547					 char *buf);
1548static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1549					  struct device_attribute *attr,
1550					  const char *buf, size_t count);
1551#define BLK_TRACE_DEVICE_ATTR(_name) \
1552	DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1553		    sysfs_blk_trace_attr_show, \
1554		    sysfs_blk_trace_attr_store)
1555
1556static BLK_TRACE_DEVICE_ATTR(enable);
1557static BLK_TRACE_DEVICE_ATTR(act_mask);
1558static BLK_TRACE_DEVICE_ATTR(pid);
1559static BLK_TRACE_DEVICE_ATTR(start_lba);
1560static BLK_TRACE_DEVICE_ATTR(end_lba);
1561
1562static struct attribute *blk_trace_attrs[] = {
1563	&dev_attr_enable.attr,
1564	&dev_attr_act_mask.attr,
1565	&dev_attr_pid.attr,
1566	&dev_attr_start_lba.attr,
1567	&dev_attr_end_lba.attr,
1568	NULL
1569};
1570
1571struct attribute_group blk_trace_attr_group = {
1572	.name  = "trace",
1573	.attrs = blk_trace_attrs,
1574};
1575
1576static const struct {
1577	int mask;
1578	const char *str;
1579} mask_maps[] = {
1580	{ BLK_TC_READ,		"read"		},
1581	{ BLK_TC_WRITE,		"write"		},
1582	{ BLK_TC_FLUSH,		"flush"		},
1583	{ BLK_TC_SYNC,		"sync"		},
1584	{ BLK_TC_QUEUE,		"queue"		},
1585	{ BLK_TC_REQUEUE,	"requeue"	},
1586	{ BLK_TC_ISSUE,		"issue"		},
1587	{ BLK_TC_COMPLETE,	"complete"	},
1588	{ BLK_TC_FS,		"fs"		},
1589	{ BLK_TC_PC,		"pc"		},
 
1590	{ BLK_TC_AHEAD,		"ahead"		},
1591	{ BLK_TC_META,		"meta"		},
1592	{ BLK_TC_DISCARD,	"discard"	},
1593	{ BLK_TC_DRV_DATA,	"drv_data"	},
1594	{ BLK_TC_FUA,		"fua"		},
1595};
1596
1597static int blk_trace_str2mask(const char *str)
1598{
1599	int i;
1600	int mask = 0;
1601	char *buf, *s, *token;
1602
1603	buf = kstrdup(str, GFP_KERNEL);
1604	if (buf == NULL)
1605		return -ENOMEM;
1606	s = strstrip(buf);
1607
1608	while (1) {
1609		token = strsep(&s, ",");
1610		if (token == NULL)
1611			break;
1612
1613		if (*token == '\0')
1614			continue;
1615
1616		for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1617			if (strcasecmp(token, mask_maps[i].str) == 0) {
1618				mask |= mask_maps[i].mask;
1619				break;
1620			}
1621		}
1622		if (i == ARRAY_SIZE(mask_maps)) {
1623			mask = -EINVAL;
1624			break;
1625		}
1626	}
1627	kfree(buf);
1628
1629	return mask;
1630}
1631
1632static ssize_t blk_trace_mask2str(char *buf, int mask)
1633{
1634	int i;
1635	char *p = buf;
1636
1637	for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1638		if (mask & mask_maps[i].mask) {
1639			p += sprintf(p, "%s%s",
1640				    (p == buf) ? "" : ",", mask_maps[i].str);
1641		}
1642	}
1643	*p++ = '\n';
1644
1645	return p - buf;
1646}
1647
1648static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1649{
1650	if (bdev->bd_disk == NULL)
1651		return NULL;
1652
1653	return bdev_get_queue(bdev);
1654}
1655
1656static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1657					 struct device_attribute *attr,
1658					 char *buf)
1659{
1660	struct hd_struct *p = dev_to_part(dev);
1661	struct request_queue *q;
1662	struct block_device *bdev;
1663	ssize_t ret = -ENXIO;
1664
1665	bdev = bdget(part_devt(p));
1666	if (bdev == NULL)
1667		goto out;
1668
1669	q = blk_trace_get_queue(bdev);
1670	if (q == NULL)
1671		goto out_bdput;
1672
1673	mutex_lock(&bdev->bd_mutex);
1674
 
 
1675	if (attr == &dev_attr_enable) {
1676		ret = sprintf(buf, "%u\n", !!q->blk_trace);
1677		goto out_unlock_bdev;
1678	}
1679
1680	if (q->blk_trace == NULL)
1681		ret = sprintf(buf, "disabled\n");
1682	else if (attr == &dev_attr_act_mask)
1683		ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1684	else if (attr == &dev_attr_pid)
1685		ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1686	else if (attr == &dev_attr_start_lba)
1687		ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1688	else if (attr == &dev_attr_end_lba)
1689		ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1690
1691out_unlock_bdev:
1692	mutex_unlock(&bdev->bd_mutex);
1693out_bdput:
1694	bdput(bdev);
1695out:
1696	return ret;
1697}
1698
1699static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1700					  struct device_attribute *attr,
1701					  const char *buf, size_t count)
1702{
1703	struct block_device *bdev;
1704	struct request_queue *q;
1705	struct hd_struct *p;
1706	u64 value;
1707	ssize_t ret = -EINVAL;
1708
1709	if (count == 0)
1710		goto out;
1711
1712	if (attr == &dev_attr_act_mask) {
1713		if (sscanf(buf, "%llx", &value) != 1) {
1714			/* Assume it is a list of trace category names */
1715			ret = blk_trace_str2mask(buf);
1716			if (ret < 0)
1717				goto out;
1718			value = ret;
1719		}
1720	} else if (sscanf(buf, "%llu", &value) != 1)
1721		goto out;
1722
1723	ret = -ENXIO;
1724
1725	p = dev_to_part(dev);
1726	bdev = bdget(part_devt(p));
1727	if (bdev == NULL)
1728		goto out;
1729
1730	q = blk_trace_get_queue(bdev);
1731	if (q == NULL)
1732		goto out_bdput;
1733
1734	mutex_lock(&bdev->bd_mutex);
1735
 
 
1736	if (attr == &dev_attr_enable) {
 
 
 
 
1737		if (value)
1738			ret = blk_trace_setup_queue(q, bdev);
1739		else
1740			ret = blk_trace_remove_queue(q);
1741		goto out_unlock_bdev;
1742	}
1743
1744	ret = 0;
1745	if (q->blk_trace == NULL)
1746		ret = blk_trace_setup_queue(q, bdev);
 
 
 
1747
1748	if (ret == 0) {
1749		if (attr == &dev_attr_act_mask)
1750			q->blk_trace->act_mask = value;
1751		else if (attr == &dev_attr_pid)
1752			q->blk_trace->pid = value;
1753		else if (attr == &dev_attr_start_lba)
1754			q->blk_trace->start_lba = value;
1755		else if (attr == &dev_attr_end_lba)
1756			q->blk_trace->end_lba = value;
1757	}
1758
1759out_unlock_bdev:
1760	mutex_unlock(&bdev->bd_mutex);
1761out_bdput:
1762	bdput(bdev);
1763out:
1764	return ret ? ret : count;
1765}
1766
1767int blk_trace_init_sysfs(struct device *dev)
1768{
1769	return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1770}
1771
1772void blk_trace_remove_sysfs(struct device *dev)
1773{
1774	sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1775}
1776
1777#endif /* CONFIG_BLK_DEV_IO_TRACE */
1778
1779#ifdef CONFIG_EVENT_TRACING
1780
1781void blk_dump_cmd(char *buf, struct request *rq)
1782{
1783	int i, end;
1784	int len = rq->cmd_len;
1785	unsigned char *cmd = rq->cmd;
1786
1787	if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1788		buf[0] = '\0';
1789		return;
1790	}
1791
1792	for (end = len - 1; end >= 0; end--)
1793		if (cmd[end])
1794			break;
1795	end++;
1796
1797	for (i = 0; i < len; i++) {
1798		buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1799		if (i == end && end != len - 1) {
1800			sprintf(buf, " ..");
1801			break;
1802		}
1803	}
1804}
1805
1806void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1807{
1808	int i = 0;
1809
1810	if (rw & REQ_FLUSH)
1811		rwbs[i++] = 'F';
1812
1813	if (rw & WRITE)
 
1814		rwbs[i++] = 'W';
1815	else if (rw & REQ_DISCARD)
 
 
 
 
1816		rwbs[i++] = 'D';
1817	else if (bytes)
 
 
 
 
 
1818		rwbs[i++] = 'R';
1819	else
 
1820		rwbs[i++] = 'N';
 
1821
1822	if (rw & REQ_FUA)
1823		rwbs[i++] = 'F';
1824	if (rw & REQ_RAHEAD)
1825		rwbs[i++] = 'A';
1826	if (rw & REQ_SYNC)
1827		rwbs[i++] = 'S';
1828	if (rw & REQ_META)
1829		rwbs[i++] = 'M';
1830	if (rw & REQ_SECURE)
1831		rwbs[i++] = 'E';
1832
1833	rwbs[i] = '\0';
1834}
1835EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1836
1837#endif /* CONFIG_EVENT_TRACING */
1838