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/debugfs.h> 25 #include <linux/smp_lock.h> 26 #include <linux/time.h> 27 #include <linux/uaccess.h> 28 29 #include <trace/events/block.h> 30 31 #include "trace_output.h" 32 33 #ifdef CONFIG_BLK_DEV_IO_TRACE 34 35 static unsigned int blktrace_seq __read_mostly = 1; 36 37 static struct trace_array *blk_tr; 38 static bool blk_tracer_enabled __read_mostly; 39 40 /* Select an alternative, minimalistic output than the original one */ 41 #define TRACE_BLK_OPT_CLASSIC 0x1 42 43 static struct tracer_opt blk_tracer_opts[] = { 44 /* Default disable the minimalistic output */ 45 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) }, 46 { } 47 }; 48 49 static struct tracer_flags blk_tracer_flags = { 50 .val = 0, 51 .opts = blk_tracer_opts, 52 }; 53 54 /* Global reference count of probes */ 55 static atomic_t blk_probes_ref = ATOMIC_INIT(0); 56 57 static void blk_register_tracepoints(void); 58 static void blk_unregister_tracepoints(void); 59 60 /* 61 * Send out a notify message. 62 */ 63 static void trace_note(struct blk_trace *bt, pid_t pid, int action, 64 const void *data, size_t len) 65 { 66 struct blk_io_trace *t; 67 struct ring_buffer_event *event = NULL; 68 struct ring_buffer *buffer = NULL; 69 int pc = 0; 70 int cpu = smp_processor_id(); 71 bool blk_tracer = blk_tracer_enabled; 72 73 if (blk_tracer) { 74 buffer = blk_tr->buffer; 75 pc = preempt_count(); 76 event = trace_buffer_lock_reserve(buffer, TRACE_BLK, 77 sizeof(*t) + len, 78 0, pc); 79 if (!event) 80 return; 81 t = ring_buffer_event_data(event); 82 goto record_it; 83 } 84 85 if (!bt->rchan) 86 return; 87 88 t = relay_reserve(bt->rchan, sizeof(*t) + len); 89 if (t) { 90 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; 91 t->time = ktime_to_ns(ktime_get()); 92 record_it: 93 t->device = bt->dev; 94 t->action = action; 95 t->pid = pid; 96 t->cpu = cpu; 97 t->pdu_len = len; 98 memcpy((void *) t + sizeof(*t), data, len); 99 100 if (blk_tracer) 101 trace_buffer_unlock_commit(buffer, event, 0, pc); 102 } 103 } 104 105 /* 106 * Send out a notify for this process, if we haven't done so since a trace 107 * started 108 */ 109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk) 110 { 111 tsk->btrace_seq = blktrace_seq; 112 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm)); 113 } 114 115 static void trace_note_time(struct blk_trace *bt) 116 { 117 struct timespec now; 118 unsigned long flags; 119 u32 words[2]; 120 121 getnstimeofday(&now); 122 words[0] = now.tv_sec; 123 words[1] = now.tv_nsec; 124 125 local_irq_save(flags); 126 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words)); 127 local_irq_restore(flags); 128 } 129 130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...) 131 { 132 int n; 133 va_list args; 134 unsigned long flags; 135 char *buf; 136 137 if (unlikely(bt->trace_state != Blktrace_running && 138 !blk_tracer_enabled)) 139 return; 140 141 local_irq_save(flags); 142 buf = per_cpu_ptr(bt->msg_data, smp_processor_id()); 143 va_start(args, fmt); 144 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args); 145 va_end(args); 146 147 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n); 148 local_irq_restore(flags); 149 } 150 EXPORT_SYMBOL_GPL(__trace_note_message); 151 152 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector, 153 pid_t pid) 154 { 155 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0) 156 return 1; 157 if (sector && (sector < bt->start_lba || sector > bt->end_lba)) 158 return 1; 159 if (bt->pid && pid != bt->pid) 160 return 1; 161 162 return 0; 163 } 164 165 /* 166 * Data direction bit lookup 167 */ 168 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ), 169 BLK_TC_ACT(BLK_TC_WRITE) }; 170 171 /* The ilog2() calls fall out because they're constant */ 172 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \ 173 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name)) 174 175 /* 176 * The worker for the various blk_add_trace*() types. Fills out a 177 * blk_io_trace structure and places it in a per-cpu subbuffer. 178 */ 179 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes, 180 int rw, u32 what, int error, int pdu_len, void *pdu_data) 181 { 182 struct task_struct *tsk = current; 183 struct ring_buffer_event *event = NULL; 184 struct ring_buffer *buffer = NULL; 185 struct blk_io_trace *t; 186 unsigned long flags = 0; 187 unsigned long *sequence; 188 pid_t pid; 189 int cpu, pc = 0; 190 bool blk_tracer = blk_tracer_enabled; 191 192 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer)) 193 return; 194 195 what |= ddir_act[rw & WRITE]; 196 what |= MASK_TC_BIT(rw, BARRIER); 197 what |= MASK_TC_BIT(rw, SYNCIO); 198 what |= MASK_TC_BIT(rw, AHEAD); 199 what |= MASK_TC_BIT(rw, META); 200 what |= MASK_TC_BIT(rw, DISCARD); 201 202 pid = tsk->pid; 203 if (act_log_check(bt, what, sector, pid)) 204 return; 205 cpu = raw_smp_processor_id(); 206 207 if (blk_tracer) { 208 tracing_record_cmdline(current); 209 210 buffer = blk_tr->buffer; 211 pc = preempt_count(); 212 event = trace_buffer_lock_reserve(buffer, TRACE_BLK, 213 sizeof(*t) + pdu_len, 214 0, pc); 215 if (!event) 216 return; 217 t = ring_buffer_event_data(event); 218 goto record_it; 219 } 220 221 /* 222 * A word about the locking here - we disable interrupts to reserve 223 * some space in the relay per-cpu buffer, to prevent an irq 224 * from coming in and stepping on our toes. 225 */ 226 local_irq_save(flags); 227 228 if (unlikely(tsk->btrace_seq != blktrace_seq)) 229 trace_note_tsk(bt, tsk); 230 231 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len); 232 if (t) { 233 sequence = per_cpu_ptr(bt->sequence, cpu); 234 235 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; 236 t->sequence = ++(*sequence); 237 t->time = ktime_to_ns(ktime_get()); 238 record_it: 239 /* 240 * These two are not needed in ftrace as they are in the 241 * generic trace_entry, filled by tracing_generic_entry_update, 242 * but for the trace_event->bin() synthesizer benefit we do it 243 * here too. 244 */ 245 t->cpu = cpu; 246 t->pid = pid; 247 248 t->sector = sector; 249 t->bytes = bytes; 250 t->action = what; 251 t->device = bt->dev; 252 t->error = error; 253 t->pdu_len = pdu_len; 254 255 if (pdu_len) 256 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len); 257 258 if (blk_tracer) { 259 trace_buffer_unlock_commit(buffer, event, 0, pc); 260 return; 261 } 262 } 263 264 local_irq_restore(flags); 265 } 266 267 static struct dentry *blk_tree_root; 268 static DEFINE_MUTEX(blk_tree_mutex); 269 270 static void blk_trace_free(struct blk_trace *bt) 271 { 272 debugfs_remove(bt->msg_file); 273 debugfs_remove(bt->dropped_file); 274 relay_close(bt->rchan); 275 debugfs_remove(bt->dir); 276 free_percpu(bt->sequence); 277 free_percpu(bt->msg_data); 278 kfree(bt); 279 } 280 281 static void blk_trace_cleanup(struct blk_trace *bt) 282 { 283 blk_trace_free(bt); 284 if (atomic_dec_and_test(&blk_probes_ref)) 285 blk_unregister_tracepoints(); 286 } 287 288 int blk_trace_remove(struct request_queue *q) 289 { 290 struct blk_trace *bt; 291 292 bt = xchg(&q->blk_trace, NULL); 293 if (!bt) 294 return -EINVAL; 295 296 if (bt->trace_state != Blktrace_running) 297 blk_trace_cleanup(bt); 298 299 return 0; 300 } 301 EXPORT_SYMBOL_GPL(blk_trace_remove); 302 303 static int blk_dropped_open(struct inode *inode, struct file *filp) 304 { 305 filp->private_data = inode->i_private; 306 307 return 0; 308 } 309 310 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer, 311 size_t count, loff_t *ppos) 312 { 313 struct blk_trace *bt = filp->private_data; 314 char buf[16]; 315 316 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped)); 317 318 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf)); 319 } 320 321 static const struct file_operations blk_dropped_fops = { 322 .owner = THIS_MODULE, 323 .open = blk_dropped_open, 324 .read = blk_dropped_read, 325 }; 326 327 static int blk_msg_open(struct inode *inode, struct file *filp) 328 { 329 filp->private_data = inode->i_private; 330 331 return 0; 332 } 333 334 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer, 335 size_t count, loff_t *ppos) 336 { 337 char *msg; 338 struct blk_trace *bt; 339 340 if (count >= BLK_TN_MAX_MSG) 341 return -EINVAL; 342 343 msg = kmalloc(count + 1, GFP_KERNEL); 344 if (msg == NULL) 345 return -ENOMEM; 346 347 if (copy_from_user(msg, buffer, count)) { 348 kfree(msg); 349 return -EFAULT; 350 } 351 352 msg[count] = '\0'; 353 bt = filp->private_data; 354 __trace_note_message(bt, "%s", msg); 355 kfree(msg); 356 357 return count; 358 } 359 360 static const struct file_operations blk_msg_fops = { 361 .owner = THIS_MODULE, 362 .open = blk_msg_open, 363 .write = blk_msg_write, 364 }; 365 366 /* 367 * Keep track of how many times we encountered a full subbuffer, to aid 368 * the user space app in telling how many lost events there were. 369 */ 370 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf, 371 void *prev_subbuf, size_t prev_padding) 372 { 373 struct blk_trace *bt; 374 375 if (!relay_buf_full(buf)) 376 return 1; 377 378 bt = buf->chan->private_data; 379 atomic_inc(&bt->dropped); 380 return 0; 381 } 382 383 static int blk_remove_buf_file_callback(struct dentry *dentry) 384 { 385 debugfs_remove(dentry); 386 387 return 0; 388 } 389 390 static struct dentry *blk_create_buf_file_callback(const char *filename, 391 struct dentry *parent, 392 int mode, 393 struct rchan_buf *buf, 394 int *is_global) 395 { 396 return debugfs_create_file(filename, mode, parent, buf, 397 &relay_file_operations); 398 } 399 400 static struct rchan_callbacks blk_relay_callbacks = { 401 .subbuf_start = blk_subbuf_start_callback, 402 .create_buf_file = blk_create_buf_file_callback, 403 .remove_buf_file = blk_remove_buf_file_callback, 404 }; 405 406 static void blk_trace_setup_lba(struct blk_trace *bt, 407 struct block_device *bdev) 408 { 409 struct hd_struct *part = NULL; 410 411 if (bdev) 412 part = bdev->bd_part; 413 414 if (part) { 415 bt->start_lba = part->start_sect; 416 bt->end_lba = part->start_sect + part->nr_sects; 417 } else { 418 bt->start_lba = 0; 419 bt->end_lba = -1ULL; 420 } 421 } 422 423 /* 424 * Setup everything required to start tracing 425 */ 426 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev, 427 struct block_device *bdev, 428 struct blk_user_trace_setup *buts) 429 { 430 struct blk_trace *old_bt, *bt = NULL; 431 struct dentry *dir = NULL; 432 int ret, i; 433 434 if (!buts->buf_size || !buts->buf_nr) 435 return -EINVAL; 436 437 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE); 438 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0'; 439 440 /* 441 * some device names have larger paths - convert the slashes 442 * to underscores for this to work as expected 443 */ 444 for (i = 0; i < strlen(buts->name); i++) 445 if (buts->name[i] == '/') 446 buts->name[i] = '_'; 447 448 bt = kzalloc(sizeof(*bt), GFP_KERNEL); 449 if (!bt) 450 return -ENOMEM; 451 452 ret = -ENOMEM; 453 bt->sequence = alloc_percpu(unsigned long); 454 if (!bt->sequence) 455 goto err; 456 457 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); 458 if (!bt->msg_data) 459 goto err; 460 461 ret = -ENOENT; 462 463 mutex_lock(&blk_tree_mutex); 464 if (!blk_tree_root) { 465 blk_tree_root = debugfs_create_dir("block", NULL); 466 if (!blk_tree_root) { 467 mutex_unlock(&blk_tree_mutex); 468 goto err; 469 } 470 } 471 mutex_unlock(&blk_tree_mutex); 472 473 dir = debugfs_create_dir(buts->name, blk_tree_root); 474 475 if (!dir) 476 goto err; 477 478 bt->dir = dir; 479 bt->dev = dev; 480 atomic_set(&bt->dropped, 0); 481 482 ret = -EIO; 483 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, 484 &blk_dropped_fops); 485 if (!bt->dropped_file) 486 goto err; 487 488 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops); 489 if (!bt->msg_file) 490 goto err; 491 492 bt->rchan = relay_open("trace", dir, buts->buf_size, 493 buts->buf_nr, &blk_relay_callbacks, bt); 494 if (!bt->rchan) 495 goto err; 496 497 bt->act_mask = buts->act_mask; 498 if (!bt->act_mask) 499 bt->act_mask = (u16) -1; 500 501 blk_trace_setup_lba(bt, bdev); 502 503 /* overwrite with user settings */ 504 if (buts->start_lba) 505 bt->start_lba = buts->start_lba; 506 if (buts->end_lba) 507 bt->end_lba = buts->end_lba; 508 509 bt->pid = buts->pid; 510 bt->trace_state = Blktrace_setup; 511 512 ret = -EBUSY; 513 old_bt = xchg(&q->blk_trace, bt); 514 if (old_bt) { 515 (void) xchg(&q->blk_trace, old_bt); 516 goto err; 517 } 518 519 if (atomic_inc_return(&blk_probes_ref) == 1) 520 blk_register_tracepoints(); 521 522 return 0; 523 err: 524 blk_trace_free(bt); 525 return ret; 526 } 527 528 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev, 529 struct block_device *bdev, 530 char __user *arg) 531 { 532 struct blk_user_trace_setup buts; 533 int ret; 534 535 ret = copy_from_user(&buts, arg, sizeof(buts)); 536 if (ret) 537 return -EFAULT; 538 539 ret = do_blk_trace_setup(q, name, dev, bdev, &buts); 540 if (ret) 541 return ret; 542 543 if (copy_to_user(arg, &buts, sizeof(buts))) { 544 blk_trace_remove(q); 545 return -EFAULT; 546 } 547 return 0; 548 } 549 EXPORT_SYMBOL_GPL(blk_trace_setup); 550 551 int blk_trace_startstop(struct request_queue *q, int start) 552 { 553 int ret; 554 struct blk_trace *bt = q->blk_trace; 555 556 if (bt == NULL) 557 return -EINVAL; 558 559 /* 560 * For starting a trace, we can transition from a setup or stopped 561 * trace. For stopping a trace, the state must be running 562 */ 563 ret = -EINVAL; 564 if (start) { 565 if (bt->trace_state == Blktrace_setup || 566 bt->trace_state == Blktrace_stopped) { 567 blktrace_seq++; 568 smp_mb(); 569 bt->trace_state = Blktrace_running; 570 571 trace_note_time(bt); 572 ret = 0; 573 } 574 } else { 575 if (bt->trace_state == Blktrace_running) { 576 bt->trace_state = Blktrace_stopped; 577 relay_flush(bt->rchan); 578 ret = 0; 579 } 580 } 581 582 return ret; 583 } 584 EXPORT_SYMBOL_GPL(blk_trace_startstop); 585 586 /** 587 * blk_trace_ioctl: - handle the ioctls associated with tracing 588 * @bdev: the block device 589 * @cmd: the ioctl cmd 590 * @arg: the argument data, if any 591 * 592 **/ 593 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg) 594 { 595 struct request_queue *q; 596 int ret, start = 0; 597 char b[BDEVNAME_SIZE]; 598 599 q = bdev_get_queue(bdev); 600 if (!q) 601 return -ENXIO; 602 603 mutex_lock(&bdev->bd_mutex); 604 605 switch (cmd) { 606 case BLKTRACESETUP: 607 bdevname(bdev, b); 608 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg); 609 break; 610 case BLKTRACESTART: 611 start = 1; 612 case BLKTRACESTOP: 613 ret = blk_trace_startstop(q, start); 614 break; 615 case BLKTRACETEARDOWN: 616 ret = blk_trace_remove(q); 617 break; 618 default: 619 ret = -ENOTTY; 620 break; 621 } 622 623 mutex_unlock(&bdev->bd_mutex); 624 return ret; 625 } 626 627 /** 628 * blk_trace_shutdown: - stop and cleanup trace structures 629 * @q: the request queue associated with the device 630 * 631 **/ 632 void blk_trace_shutdown(struct request_queue *q) 633 { 634 if (q->blk_trace) { 635 blk_trace_startstop(q, 0); 636 blk_trace_remove(q); 637 } 638 } 639 640 /* 641 * blktrace probes 642 */ 643 644 /** 645 * blk_add_trace_rq - Add a trace for a request oriented action 646 * @q: queue the io is for 647 * @rq: the source request 648 * @what: the action 649 * 650 * Description: 651 * Records an action against a request. Will log the bio offset + size. 652 * 653 **/ 654 static void blk_add_trace_rq(struct request_queue *q, struct request *rq, 655 u32 what) 656 { 657 struct blk_trace *bt = q->blk_trace; 658 int rw = rq->cmd_flags & 0x03; 659 660 if (likely(!bt)) 661 return; 662 663 if (blk_discard_rq(rq)) 664 rw |= (1 << BIO_RW_DISCARD); 665 666 if (blk_pc_request(rq)) { 667 what |= BLK_TC_ACT(BLK_TC_PC); 668 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw, 669 what, rq->errors, rq->cmd_len, rq->cmd); 670 } else { 671 what |= BLK_TC_ACT(BLK_TC_FS); 672 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw, 673 what, rq->errors, 0, NULL); 674 } 675 } 676 677 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq) 678 { 679 blk_add_trace_rq(q, rq, BLK_TA_ABORT); 680 } 681 682 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq) 683 { 684 blk_add_trace_rq(q, rq, BLK_TA_INSERT); 685 } 686 687 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq) 688 { 689 blk_add_trace_rq(q, rq, BLK_TA_ISSUE); 690 } 691 692 static void blk_add_trace_rq_requeue(struct request_queue *q, 693 struct request *rq) 694 { 695 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); 696 } 697 698 static void blk_add_trace_rq_complete(struct request_queue *q, 699 struct request *rq) 700 { 701 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE); 702 } 703 704 /** 705 * blk_add_trace_bio - Add a trace for a bio oriented action 706 * @q: queue the io is for 707 * @bio: the source bio 708 * @what: the action 709 * 710 * Description: 711 * Records an action against a bio. Will log the bio offset + size. 712 * 713 **/ 714 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio, 715 u32 what) 716 { 717 struct blk_trace *bt = q->blk_trace; 718 719 if (likely(!bt)) 720 return; 721 722 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, 723 !bio_flagged(bio, BIO_UPTODATE), 0, NULL); 724 } 725 726 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio) 727 { 728 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE); 729 } 730 731 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio) 732 { 733 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE); 734 } 735 736 static void blk_add_trace_bio_backmerge(struct request_queue *q, 737 struct bio *bio) 738 { 739 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); 740 } 741 742 static void blk_add_trace_bio_frontmerge(struct request_queue *q, 743 struct bio *bio) 744 { 745 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); 746 } 747 748 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio) 749 { 750 blk_add_trace_bio(q, bio, BLK_TA_QUEUE); 751 } 752 753 static void blk_add_trace_getrq(struct request_queue *q, 754 struct bio *bio, int rw) 755 { 756 if (bio) 757 blk_add_trace_bio(q, bio, BLK_TA_GETRQ); 758 else { 759 struct blk_trace *bt = q->blk_trace; 760 761 if (bt) 762 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL); 763 } 764 } 765 766 767 static void blk_add_trace_sleeprq(struct request_queue *q, 768 struct bio *bio, int rw) 769 { 770 if (bio) 771 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ); 772 else { 773 struct blk_trace *bt = q->blk_trace; 774 775 if (bt) 776 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ, 777 0, 0, NULL); 778 } 779 } 780 781 static void blk_add_trace_plug(struct request_queue *q) 782 { 783 struct blk_trace *bt = q->blk_trace; 784 785 if (bt) 786 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL); 787 } 788 789 static void blk_add_trace_unplug_io(struct request_queue *q) 790 { 791 struct blk_trace *bt = q->blk_trace; 792 793 if (bt) { 794 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE]; 795 __be64 rpdu = cpu_to_be64(pdu); 796 797 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0, 798 sizeof(rpdu), &rpdu); 799 } 800 } 801 802 static void blk_add_trace_unplug_timer(struct request_queue *q) 803 { 804 struct blk_trace *bt = q->blk_trace; 805 806 if (bt) { 807 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE]; 808 __be64 rpdu = cpu_to_be64(pdu); 809 810 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0, 811 sizeof(rpdu), &rpdu); 812 } 813 } 814 815 static void blk_add_trace_split(struct request_queue *q, struct bio *bio, 816 unsigned int pdu) 817 { 818 struct blk_trace *bt = q->blk_trace; 819 820 if (bt) { 821 __be64 rpdu = cpu_to_be64(pdu); 822 823 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, 824 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE), 825 sizeof(rpdu), &rpdu); 826 } 827 } 828 829 /** 830 * blk_add_trace_remap - Add a trace for a remap operation 831 * @q: queue the io is for 832 * @bio: the source bio 833 * @dev: target device 834 * @from: source sector 835 * 836 * Description: 837 * Device mapper or raid target sometimes need to split a bio because 838 * it spans a stripe (or similar). Add a trace for that action. 839 * 840 **/ 841 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio, 842 dev_t dev, sector_t from) 843 { 844 struct blk_trace *bt = q->blk_trace; 845 struct blk_io_trace_remap r; 846 847 if (likely(!bt)) 848 return; 849 850 r.device_from = cpu_to_be32(dev); 851 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev); 852 r.sector_from = cpu_to_be64(from); 853 854 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, 855 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), 856 sizeof(r), &r); 857 } 858 859 /** 860 * blk_add_trace_rq_remap - Add a trace for a request-remap operation 861 * @q: queue the io is for 862 * @rq: the source request 863 * @dev: target device 864 * @from: source sector 865 * 866 * Description: 867 * Device mapper remaps request to other devices. 868 * Add a trace for that action. 869 * 870 **/ 871 static void blk_add_trace_rq_remap(struct request_queue *q, 872 struct request *rq, dev_t dev, 873 sector_t from) 874 { 875 struct blk_trace *bt = q->blk_trace; 876 struct blk_io_trace_remap r; 877 878 if (likely(!bt)) 879 return; 880 881 r.device_from = cpu_to_be32(dev); 882 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk)); 883 r.sector_from = cpu_to_be64(from); 884 885 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 886 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors, 887 sizeof(r), &r); 888 } 889 890 /** 891 * blk_add_driver_data - Add binary message with driver-specific data 892 * @q: queue the io is for 893 * @rq: io request 894 * @data: driver-specific data 895 * @len: length of driver-specific data 896 * 897 * Description: 898 * Some drivers might want to write driver-specific data per request. 899 * 900 **/ 901 void blk_add_driver_data(struct request_queue *q, 902 struct request *rq, 903 void *data, size_t len) 904 { 905 struct blk_trace *bt = q->blk_trace; 906 907 if (likely(!bt)) 908 return; 909 910 if (blk_pc_request(rq)) 911 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 912 BLK_TA_DRV_DATA, rq->errors, len, data); 913 else 914 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 915 BLK_TA_DRV_DATA, rq->errors, len, data); 916 } 917 EXPORT_SYMBOL_GPL(blk_add_driver_data); 918 919 static void blk_register_tracepoints(void) 920 { 921 int ret; 922 923 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort); 924 WARN_ON(ret); 925 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert); 926 WARN_ON(ret); 927 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue); 928 WARN_ON(ret); 929 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue); 930 WARN_ON(ret); 931 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete); 932 WARN_ON(ret); 933 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce); 934 WARN_ON(ret); 935 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete); 936 WARN_ON(ret); 937 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge); 938 WARN_ON(ret); 939 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge); 940 WARN_ON(ret); 941 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue); 942 WARN_ON(ret); 943 ret = register_trace_block_getrq(blk_add_trace_getrq); 944 WARN_ON(ret); 945 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq); 946 WARN_ON(ret); 947 ret = register_trace_block_plug(blk_add_trace_plug); 948 WARN_ON(ret); 949 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer); 950 WARN_ON(ret); 951 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io); 952 WARN_ON(ret); 953 ret = register_trace_block_split(blk_add_trace_split); 954 WARN_ON(ret); 955 ret = register_trace_block_remap(blk_add_trace_remap); 956 WARN_ON(ret); 957 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap); 958 WARN_ON(ret); 959 } 960 961 static void blk_unregister_tracepoints(void) 962 { 963 unregister_trace_block_rq_remap(blk_add_trace_rq_remap); 964 unregister_trace_block_remap(blk_add_trace_remap); 965 unregister_trace_block_split(blk_add_trace_split); 966 unregister_trace_block_unplug_io(blk_add_trace_unplug_io); 967 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer); 968 unregister_trace_block_plug(blk_add_trace_plug); 969 unregister_trace_block_sleeprq(blk_add_trace_sleeprq); 970 unregister_trace_block_getrq(blk_add_trace_getrq); 971 unregister_trace_block_bio_queue(blk_add_trace_bio_queue); 972 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge); 973 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge); 974 unregister_trace_block_bio_complete(blk_add_trace_bio_complete); 975 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce); 976 unregister_trace_block_rq_complete(blk_add_trace_rq_complete); 977 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue); 978 unregister_trace_block_rq_issue(blk_add_trace_rq_issue); 979 unregister_trace_block_rq_insert(blk_add_trace_rq_insert); 980 unregister_trace_block_rq_abort(blk_add_trace_rq_abort); 981 982 tracepoint_synchronize_unregister(); 983 } 984 985 /* 986 * struct blk_io_tracer formatting routines 987 */ 988 989 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t) 990 { 991 int i = 0; 992 int tc = t->action >> BLK_TC_SHIFT; 993 994 if (t->action == BLK_TN_MESSAGE) { 995 rwbs[i++] = 'N'; 996 goto out; 997 } 998 999 if (tc & BLK_TC_DISCARD) 1000 rwbs[i++] = 'D'; 1001 else if (tc & BLK_TC_WRITE) 1002 rwbs[i++] = 'W'; 1003 else if (t->bytes) 1004 rwbs[i++] = 'R'; 1005 else 1006 rwbs[i++] = 'N'; 1007 1008 if (tc & BLK_TC_AHEAD) 1009 rwbs[i++] = 'A'; 1010 if (tc & BLK_TC_BARRIER) 1011 rwbs[i++] = 'B'; 1012 if (tc & BLK_TC_SYNC) 1013 rwbs[i++] = 'S'; 1014 if (tc & BLK_TC_META) 1015 rwbs[i++] = 'M'; 1016 out: 1017 rwbs[i] = '\0'; 1018 } 1019 1020 static inline 1021 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent) 1022 { 1023 return (const struct blk_io_trace *)ent; 1024 } 1025 1026 static inline const void *pdu_start(const struct trace_entry *ent) 1027 { 1028 return te_blk_io_trace(ent) + 1; 1029 } 1030 1031 static inline u32 t_action(const struct trace_entry *ent) 1032 { 1033 return te_blk_io_trace(ent)->action; 1034 } 1035 1036 static inline u32 t_bytes(const struct trace_entry *ent) 1037 { 1038 return te_blk_io_trace(ent)->bytes; 1039 } 1040 1041 static inline u32 t_sec(const struct trace_entry *ent) 1042 { 1043 return te_blk_io_trace(ent)->bytes >> 9; 1044 } 1045 1046 static inline unsigned long long t_sector(const struct trace_entry *ent) 1047 { 1048 return te_blk_io_trace(ent)->sector; 1049 } 1050 1051 static inline __u16 t_error(const struct trace_entry *ent) 1052 { 1053 return te_blk_io_trace(ent)->error; 1054 } 1055 1056 static __u64 get_pdu_int(const struct trace_entry *ent) 1057 { 1058 const __u64 *val = pdu_start(ent); 1059 return be64_to_cpu(*val); 1060 } 1061 1062 static void get_pdu_remap(const struct trace_entry *ent, 1063 struct blk_io_trace_remap *r) 1064 { 1065 const struct blk_io_trace_remap *__r = pdu_start(ent); 1066 __u64 sector_from = __r->sector_from; 1067 1068 r->device_from = be32_to_cpu(__r->device_from); 1069 r->device_to = be32_to_cpu(__r->device_to); 1070 r->sector_from = be64_to_cpu(sector_from); 1071 } 1072 1073 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act); 1074 1075 static int blk_log_action_classic(struct trace_iterator *iter, const char *act) 1076 { 1077 char rwbs[6]; 1078 unsigned long long ts = iter->ts; 1079 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC); 1080 unsigned secs = (unsigned long)ts; 1081 const struct blk_io_trace *t = te_blk_io_trace(iter->ent); 1082 1083 fill_rwbs(rwbs, t); 1084 1085 return trace_seq_printf(&iter->seq, 1086 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ", 1087 MAJOR(t->device), MINOR(t->device), iter->cpu, 1088 secs, nsec_rem, iter->ent->pid, act, rwbs); 1089 } 1090 1091 static int blk_log_action(struct trace_iterator *iter, const char *act) 1092 { 1093 char rwbs[6]; 1094 const struct blk_io_trace *t = te_blk_io_trace(iter->ent); 1095 1096 fill_rwbs(rwbs, t); 1097 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ", 1098 MAJOR(t->device), MINOR(t->device), act, rwbs); 1099 } 1100 1101 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent) 1102 { 1103 const unsigned char *pdu_buf; 1104 int pdu_len; 1105 int i, end, ret; 1106 1107 pdu_buf = pdu_start(ent); 1108 pdu_len = te_blk_io_trace(ent)->pdu_len; 1109 1110 if (!pdu_len) 1111 return 1; 1112 1113 /* find the last zero that needs to be printed */ 1114 for (end = pdu_len - 1; end >= 0; end--) 1115 if (pdu_buf[end]) 1116 break; 1117 end++; 1118 1119 if (!trace_seq_putc(s, '(')) 1120 return 0; 1121 1122 for (i = 0; i < pdu_len; i++) { 1123 1124 ret = trace_seq_printf(s, "%s%02x", 1125 i == 0 ? "" : " ", pdu_buf[i]); 1126 if (!ret) 1127 return ret; 1128 1129 /* 1130 * stop when the rest is just zeroes and indicate so 1131 * with a ".." appended 1132 */ 1133 if (i == end && end != pdu_len - 1) 1134 return trace_seq_puts(s, " ..) "); 1135 } 1136 1137 return trace_seq_puts(s, ") "); 1138 } 1139 1140 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent) 1141 { 1142 char cmd[TASK_COMM_LEN]; 1143 1144 trace_find_cmdline(ent->pid, cmd); 1145 1146 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { 1147 int ret; 1148 1149 ret = trace_seq_printf(s, "%u ", t_bytes(ent)); 1150 if (!ret) 1151 return 0; 1152 ret = blk_log_dump_pdu(s, ent); 1153 if (!ret) 1154 return 0; 1155 return trace_seq_printf(s, "[%s]\n", cmd); 1156 } else { 1157 if (t_sec(ent)) 1158 return trace_seq_printf(s, "%llu + %u [%s]\n", 1159 t_sector(ent), t_sec(ent), cmd); 1160 return trace_seq_printf(s, "[%s]\n", cmd); 1161 } 1162 } 1163 1164 static int blk_log_with_error(struct trace_seq *s, 1165 const struct trace_entry *ent) 1166 { 1167 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { 1168 int ret; 1169 1170 ret = blk_log_dump_pdu(s, ent); 1171 if (ret) 1172 return trace_seq_printf(s, "[%d]\n", t_error(ent)); 1173 return 0; 1174 } else { 1175 if (t_sec(ent)) 1176 return trace_seq_printf(s, "%llu + %u [%d]\n", 1177 t_sector(ent), 1178 t_sec(ent), t_error(ent)); 1179 return trace_seq_printf(s, "%llu [%d]\n", 1180 t_sector(ent), t_error(ent)); 1181 } 1182 } 1183 1184 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent) 1185 { 1186 struct blk_io_trace_remap r = { .device_from = 0, }; 1187 1188 get_pdu_remap(ent, &r); 1189 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n", 1190 t_sector(ent), t_sec(ent), 1191 MAJOR(r.device_from), MINOR(r.device_from), 1192 (unsigned long long)r.sector_from); 1193 } 1194 1195 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent) 1196 { 1197 char cmd[TASK_COMM_LEN]; 1198 1199 trace_find_cmdline(ent->pid, cmd); 1200 1201 return trace_seq_printf(s, "[%s]\n", cmd); 1202 } 1203 1204 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent) 1205 { 1206 char cmd[TASK_COMM_LEN]; 1207 1208 trace_find_cmdline(ent->pid, cmd); 1209 1210 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent)); 1211 } 1212 1213 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent) 1214 { 1215 char cmd[TASK_COMM_LEN]; 1216 1217 trace_find_cmdline(ent->pid, cmd); 1218 1219 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent), 1220 get_pdu_int(ent), cmd); 1221 } 1222 1223 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent) 1224 { 1225 int ret; 1226 const struct blk_io_trace *t = te_blk_io_trace(ent); 1227 1228 ret = trace_seq_putmem(s, t + 1, t->pdu_len); 1229 if (ret) 1230 return trace_seq_putc(s, '\n'); 1231 return ret; 1232 } 1233 1234 /* 1235 * struct tracer operations 1236 */ 1237 1238 static void blk_tracer_print_header(struct seq_file *m) 1239 { 1240 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) 1241 return; 1242 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n" 1243 "# | | | | | |\n"); 1244 } 1245 1246 static void blk_tracer_start(struct trace_array *tr) 1247 { 1248 blk_tracer_enabled = true; 1249 } 1250 1251 static int blk_tracer_init(struct trace_array *tr) 1252 { 1253 blk_tr = tr; 1254 blk_tracer_start(tr); 1255 return 0; 1256 } 1257 1258 static void blk_tracer_stop(struct trace_array *tr) 1259 { 1260 blk_tracer_enabled = false; 1261 } 1262 1263 static void blk_tracer_reset(struct trace_array *tr) 1264 { 1265 blk_tracer_stop(tr); 1266 } 1267 1268 static const struct { 1269 const char *act[2]; 1270 int (*print)(struct trace_seq *s, const struct trace_entry *ent); 1271 } what2act[] = { 1272 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic }, 1273 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic }, 1274 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic }, 1275 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic }, 1276 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic }, 1277 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error }, 1278 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic }, 1279 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error }, 1280 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug }, 1281 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug }, 1282 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug }, 1283 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic }, 1284 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split }, 1285 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic }, 1286 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap }, 1287 }; 1288 1289 static enum print_line_t print_one_line(struct trace_iterator *iter, 1290 bool classic) 1291 { 1292 struct trace_seq *s = &iter->seq; 1293 const struct blk_io_trace *t; 1294 u16 what; 1295 int ret; 1296 bool long_act; 1297 blk_log_action_t *log_action; 1298 1299 t = te_blk_io_trace(iter->ent); 1300 what = t->action & ((1 << BLK_TC_SHIFT) - 1); 1301 long_act = !!(trace_flags & TRACE_ITER_VERBOSE); 1302 log_action = classic ? &blk_log_action_classic : &blk_log_action; 1303 1304 if (t->action == BLK_TN_MESSAGE) { 1305 ret = log_action(iter, long_act ? "message" : "m"); 1306 if (ret) 1307 ret = blk_log_msg(s, iter->ent); 1308 goto out; 1309 } 1310 1311 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act))) 1312 ret = trace_seq_printf(s, "Unknown action %x\n", what); 1313 else { 1314 ret = log_action(iter, what2act[what].act[long_act]); 1315 if (ret) 1316 ret = what2act[what].print(s, iter->ent); 1317 } 1318 out: 1319 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE; 1320 } 1321 1322 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter, 1323 int flags) 1324 { 1325 return print_one_line(iter, false); 1326 } 1327 1328 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter) 1329 { 1330 struct trace_seq *s = &iter->seq; 1331 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent; 1332 const int offset = offsetof(struct blk_io_trace, sector); 1333 struct blk_io_trace old = { 1334 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION, 1335 .time = iter->ts, 1336 }; 1337 1338 if (!trace_seq_putmem(s, &old, offset)) 1339 return 0; 1340 return trace_seq_putmem(s, &t->sector, 1341 sizeof(old) - offset + t->pdu_len); 1342 } 1343 1344 static enum print_line_t 1345 blk_trace_event_print_binary(struct trace_iterator *iter, int flags) 1346 { 1347 return blk_trace_synthesize_old_trace(iter) ? 1348 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE; 1349 } 1350 1351 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter) 1352 { 1353 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) 1354 return TRACE_TYPE_UNHANDLED; 1355 1356 return print_one_line(iter, true); 1357 } 1358 1359 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set) 1360 { 1361 /* don't output context-info for blk_classic output */ 1362 if (bit == TRACE_BLK_OPT_CLASSIC) { 1363 if (set) 1364 trace_flags &= ~TRACE_ITER_CONTEXT_INFO; 1365 else 1366 trace_flags |= TRACE_ITER_CONTEXT_INFO; 1367 } 1368 return 0; 1369 } 1370 1371 static struct tracer blk_tracer __read_mostly = { 1372 .name = "blk", 1373 .init = blk_tracer_init, 1374 .reset = blk_tracer_reset, 1375 .start = blk_tracer_start, 1376 .stop = blk_tracer_stop, 1377 .print_header = blk_tracer_print_header, 1378 .print_line = blk_tracer_print_line, 1379 .flags = &blk_tracer_flags, 1380 .set_flag = blk_tracer_set_flag, 1381 }; 1382 1383 static struct trace_event trace_blk_event = { 1384 .type = TRACE_BLK, 1385 .trace = blk_trace_event_print, 1386 .binary = blk_trace_event_print_binary, 1387 }; 1388 1389 static int __init init_blk_tracer(void) 1390 { 1391 if (!register_ftrace_event(&trace_blk_event)) { 1392 pr_warning("Warning: could not register block events\n"); 1393 return 1; 1394 } 1395 1396 if (register_tracer(&blk_tracer) != 0) { 1397 pr_warning("Warning: could not register the block tracer\n"); 1398 unregister_ftrace_event(&trace_blk_event); 1399 return 1; 1400 } 1401 1402 return 0; 1403 } 1404 1405 device_initcall(init_blk_tracer); 1406 1407 static int blk_trace_remove_queue(struct request_queue *q) 1408 { 1409 struct blk_trace *bt; 1410 1411 bt = xchg(&q->blk_trace, NULL); 1412 if (bt == NULL) 1413 return -EINVAL; 1414 1415 if (atomic_dec_and_test(&blk_probes_ref)) 1416 blk_unregister_tracepoints(); 1417 1418 blk_trace_free(bt); 1419 return 0; 1420 } 1421 1422 /* 1423 * Setup everything required to start tracing 1424 */ 1425 static int blk_trace_setup_queue(struct request_queue *q, 1426 struct block_device *bdev) 1427 { 1428 struct blk_trace *old_bt, *bt = NULL; 1429 int ret = -ENOMEM; 1430 1431 bt = kzalloc(sizeof(*bt), GFP_KERNEL); 1432 if (!bt) 1433 return -ENOMEM; 1434 1435 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); 1436 if (!bt->msg_data) 1437 goto free_bt; 1438 1439 bt->dev = bdev->bd_dev; 1440 bt->act_mask = (u16)-1; 1441 1442 blk_trace_setup_lba(bt, bdev); 1443 1444 old_bt = xchg(&q->blk_trace, bt); 1445 if (old_bt != NULL) { 1446 (void)xchg(&q->blk_trace, old_bt); 1447 ret = -EBUSY; 1448 goto free_bt; 1449 } 1450 1451 if (atomic_inc_return(&blk_probes_ref) == 1) 1452 blk_register_tracepoints(); 1453 return 0; 1454 1455 free_bt: 1456 blk_trace_free(bt); 1457 return ret; 1458 } 1459 1460 /* 1461 * sysfs interface to enable and configure tracing 1462 */ 1463 1464 static ssize_t sysfs_blk_trace_attr_show(struct device *dev, 1465 struct device_attribute *attr, 1466 char *buf); 1467 static ssize_t sysfs_blk_trace_attr_store(struct device *dev, 1468 struct device_attribute *attr, 1469 const char *buf, size_t count); 1470 #define BLK_TRACE_DEVICE_ATTR(_name) \ 1471 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \ 1472 sysfs_blk_trace_attr_show, \ 1473 sysfs_blk_trace_attr_store) 1474 1475 static BLK_TRACE_DEVICE_ATTR(enable); 1476 static BLK_TRACE_DEVICE_ATTR(act_mask); 1477 static BLK_TRACE_DEVICE_ATTR(pid); 1478 static BLK_TRACE_DEVICE_ATTR(start_lba); 1479 static BLK_TRACE_DEVICE_ATTR(end_lba); 1480 1481 static struct attribute *blk_trace_attrs[] = { 1482 &dev_attr_enable.attr, 1483 &dev_attr_act_mask.attr, 1484 &dev_attr_pid.attr, 1485 &dev_attr_start_lba.attr, 1486 &dev_attr_end_lba.attr, 1487 NULL 1488 }; 1489 1490 struct attribute_group blk_trace_attr_group = { 1491 .name = "trace", 1492 .attrs = blk_trace_attrs, 1493 }; 1494 1495 static const struct { 1496 int mask; 1497 const char *str; 1498 } mask_maps[] = { 1499 { BLK_TC_READ, "read" }, 1500 { BLK_TC_WRITE, "write" }, 1501 { BLK_TC_BARRIER, "barrier" }, 1502 { BLK_TC_SYNC, "sync" }, 1503 { BLK_TC_QUEUE, "queue" }, 1504 { BLK_TC_REQUEUE, "requeue" }, 1505 { BLK_TC_ISSUE, "issue" }, 1506 { BLK_TC_COMPLETE, "complete" }, 1507 { BLK_TC_FS, "fs" }, 1508 { BLK_TC_PC, "pc" }, 1509 { BLK_TC_AHEAD, "ahead" }, 1510 { BLK_TC_META, "meta" }, 1511 { BLK_TC_DISCARD, "discard" }, 1512 { BLK_TC_DRV_DATA, "drv_data" }, 1513 }; 1514 1515 static int blk_trace_str2mask(const char *str) 1516 { 1517 int i; 1518 int mask = 0; 1519 char *buf, *s, *token; 1520 1521 buf = kstrdup(str, GFP_KERNEL); 1522 if (buf == NULL) 1523 return -ENOMEM; 1524 s = strstrip(buf); 1525 1526 while (1) { 1527 token = strsep(&s, ","); 1528 if (token == NULL) 1529 break; 1530 1531 if (*token == '\0') 1532 continue; 1533 1534 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { 1535 if (strcasecmp(token, mask_maps[i].str) == 0) { 1536 mask |= mask_maps[i].mask; 1537 break; 1538 } 1539 } 1540 if (i == ARRAY_SIZE(mask_maps)) { 1541 mask = -EINVAL; 1542 break; 1543 } 1544 } 1545 kfree(buf); 1546 1547 return mask; 1548 } 1549 1550 static ssize_t blk_trace_mask2str(char *buf, int mask) 1551 { 1552 int i; 1553 char *p = buf; 1554 1555 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { 1556 if (mask & mask_maps[i].mask) { 1557 p += sprintf(p, "%s%s", 1558 (p == buf) ? "" : ",", mask_maps[i].str); 1559 } 1560 } 1561 *p++ = '\n'; 1562 1563 return p - buf; 1564 } 1565 1566 static struct request_queue *blk_trace_get_queue(struct block_device *bdev) 1567 { 1568 if (bdev->bd_disk == NULL) 1569 return NULL; 1570 1571 return bdev_get_queue(bdev); 1572 } 1573 1574 static ssize_t sysfs_blk_trace_attr_show(struct device *dev, 1575 struct device_attribute *attr, 1576 char *buf) 1577 { 1578 struct hd_struct *p = dev_to_part(dev); 1579 struct request_queue *q; 1580 struct block_device *bdev; 1581 ssize_t ret = -ENXIO; 1582 1583 lock_kernel(); 1584 bdev = bdget(part_devt(p)); 1585 if (bdev == NULL) 1586 goto out_unlock_kernel; 1587 1588 q = blk_trace_get_queue(bdev); 1589 if (q == NULL) 1590 goto out_bdput; 1591 1592 mutex_lock(&bdev->bd_mutex); 1593 1594 if (attr == &dev_attr_enable) { 1595 ret = sprintf(buf, "%u\n", !!q->blk_trace); 1596 goto out_unlock_bdev; 1597 } 1598 1599 if (q->blk_trace == NULL) 1600 ret = sprintf(buf, "disabled\n"); 1601 else if (attr == &dev_attr_act_mask) 1602 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask); 1603 else if (attr == &dev_attr_pid) 1604 ret = sprintf(buf, "%u\n", q->blk_trace->pid); 1605 else if (attr == &dev_attr_start_lba) 1606 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba); 1607 else if (attr == &dev_attr_end_lba) 1608 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba); 1609 1610 out_unlock_bdev: 1611 mutex_unlock(&bdev->bd_mutex); 1612 out_bdput: 1613 bdput(bdev); 1614 out_unlock_kernel: 1615 unlock_kernel(); 1616 return ret; 1617 } 1618 1619 static ssize_t sysfs_blk_trace_attr_store(struct device *dev, 1620 struct device_attribute *attr, 1621 const char *buf, size_t count) 1622 { 1623 struct block_device *bdev; 1624 struct request_queue *q; 1625 struct hd_struct *p; 1626 u64 value; 1627 ssize_t ret = -EINVAL; 1628 1629 if (count == 0) 1630 goto out; 1631 1632 if (attr == &dev_attr_act_mask) { 1633 if (sscanf(buf, "%llx", &value) != 1) { 1634 /* Assume it is a list of trace category names */ 1635 ret = blk_trace_str2mask(buf); 1636 if (ret < 0) 1637 goto out; 1638 value = ret; 1639 } 1640 } else if (sscanf(buf, "%llu", &value) != 1) 1641 goto out; 1642 1643 ret = -ENXIO; 1644 1645 lock_kernel(); 1646 p = dev_to_part(dev); 1647 bdev = bdget(part_devt(p)); 1648 if (bdev == NULL) 1649 goto out_unlock_kernel; 1650 1651 q = blk_trace_get_queue(bdev); 1652 if (q == NULL) 1653 goto out_bdput; 1654 1655 mutex_lock(&bdev->bd_mutex); 1656 1657 if (attr == &dev_attr_enable) { 1658 if (value) 1659 ret = blk_trace_setup_queue(q, bdev); 1660 else 1661 ret = blk_trace_remove_queue(q); 1662 goto out_unlock_bdev; 1663 } 1664 1665 ret = 0; 1666 if (q->blk_trace == NULL) 1667 ret = blk_trace_setup_queue(q, bdev); 1668 1669 if (ret == 0) { 1670 if (attr == &dev_attr_act_mask) 1671 q->blk_trace->act_mask = value; 1672 else if (attr == &dev_attr_pid) 1673 q->blk_trace->pid = value; 1674 else if (attr == &dev_attr_start_lba) 1675 q->blk_trace->start_lba = value; 1676 else if (attr == &dev_attr_end_lba) 1677 q->blk_trace->end_lba = value; 1678 } 1679 1680 out_unlock_bdev: 1681 mutex_unlock(&bdev->bd_mutex); 1682 out_bdput: 1683 bdput(bdev); 1684 out_unlock_kernel: 1685 unlock_kernel(); 1686 out: 1687 return ret ? ret : count; 1688 } 1689 1690 int blk_trace_init_sysfs(struct device *dev) 1691 { 1692 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group); 1693 } 1694 1695 void blk_trace_remove_sysfs(struct device *dev) 1696 { 1697 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group); 1698 } 1699 1700 #endif /* CONFIG_BLK_DEV_IO_TRACE */ 1701 1702 #ifdef CONFIG_EVENT_TRACING 1703 1704 void blk_dump_cmd(char *buf, struct request *rq) 1705 { 1706 int i, end; 1707 int len = rq->cmd_len; 1708 unsigned char *cmd = rq->cmd; 1709 1710 if (!blk_pc_request(rq)) { 1711 buf[0] = '\0'; 1712 return; 1713 } 1714 1715 for (end = len - 1; end >= 0; end--) 1716 if (cmd[end]) 1717 break; 1718 end++; 1719 1720 for (i = 0; i < len; i++) { 1721 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]); 1722 if (i == end && end != len - 1) { 1723 sprintf(buf, " .."); 1724 break; 1725 } 1726 } 1727 } 1728 1729 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes) 1730 { 1731 int i = 0; 1732 1733 if (rw & WRITE) 1734 rwbs[i++] = 'W'; 1735 else if (rw & 1 << BIO_RW_DISCARD) 1736 rwbs[i++] = 'D'; 1737 else if (bytes) 1738 rwbs[i++] = 'R'; 1739 else 1740 rwbs[i++] = 'N'; 1741 1742 if (rw & 1 << BIO_RW_AHEAD) 1743 rwbs[i++] = 'A'; 1744 if (rw & 1 << BIO_RW_BARRIER) 1745 rwbs[i++] = 'B'; 1746 if (rw & 1 << BIO_RW_SYNCIO) 1747 rwbs[i++] = 'S'; 1748 if (rw & 1 << BIO_RW_META) 1749 rwbs[i++] = 'M'; 1750 1751 rwbs[i] = '\0'; 1752 } 1753 1754 void blk_fill_rwbs_rq(char *rwbs, struct request *rq) 1755 { 1756 int rw = rq->cmd_flags & 0x03; 1757 int bytes; 1758 1759 if (blk_discard_rq(rq)) 1760 rw |= (1 << BIO_RW_DISCARD); 1761 1762 bytes = blk_rq_bytes(rq); 1763 1764 blk_fill_rwbs(rwbs, rw, bytes); 1765 } 1766 1767 #endif /* CONFIG_EVENT_TRACING */ 1768 1769