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/smp_lock.h> 27 #include <linux/time.h> 28 #include <linux/uaccess.h> 29 30 #include <trace/events/block.h> 31 32 #include "trace_output.h" 33 34 #ifdef CONFIG_BLK_DEV_IO_TRACE 35 36 static unsigned int blktrace_seq __read_mostly = 1; 37 38 static struct trace_array *blk_tr; 39 static bool blk_tracer_enabled __read_mostly; 40 41 /* Select an alternative, minimalistic output than the original one */ 42 #define TRACE_BLK_OPT_CLASSIC 0x1 43 44 static struct tracer_opt blk_tracer_opts[] = { 45 /* Default disable the minimalistic output */ 46 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) }, 47 { } 48 }; 49 50 static struct tracer_flags blk_tracer_flags = { 51 .val = 0, 52 .opts = blk_tracer_opts, 53 }; 54 55 /* Global reference count of probes */ 56 static atomic_t blk_probes_ref = ATOMIC_INIT(0); 57 58 static void blk_register_tracepoints(void); 59 static void blk_unregister_tracepoints(void); 60 61 /* 62 * Send out a notify message. 63 */ 64 static void trace_note(struct blk_trace *bt, pid_t pid, int action, 65 const void *data, size_t len) 66 { 67 struct blk_io_trace *t; 68 struct ring_buffer_event *event = NULL; 69 struct ring_buffer *buffer = NULL; 70 int pc = 0; 71 int cpu = smp_processor_id(); 72 bool blk_tracer = blk_tracer_enabled; 73 74 if (blk_tracer) { 75 buffer = blk_tr->buffer; 76 pc = preempt_count(); 77 event = trace_buffer_lock_reserve(buffer, TRACE_BLK, 78 sizeof(*t) + len, 79 0, pc); 80 if (!event) 81 return; 82 t = ring_buffer_event_data(event); 83 goto record_it; 84 } 85 86 if (!bt->rchan) 87 return; 88 89 t = relay_reserve(bt->rchan, sizeof(*t) + len); 90 if (t) { 91 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; 92 t->time = ktime_to_ns(ktime_get()); 93 record_it: 94 t->device = bt->dev; 95 t->action = action; 96 t->pid = pid; 97 t->cpu = cpu; 98 t->pdu_len = len; 99 memcpy((void *) t + sizeof(*t), data, len); 100 101 if (blk_tracer) 102 trace_buffer_unlock_commit(buffer, event, 0, pc); 103 } 104 } 105 106 /* 107 * Send out a notify for this process, if we haven't done so since a trace 108 * started 109 */ 110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk) 111 { 112 tsk->btrace_seq = blktrace_seq; 113 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm)); 114 } 115 116 static void trace_note_time(struct blk_trace *bt) 117 { 118 struct timespec now; 119 unsigned long flags; 120 u32 words[2]; 121 122 getnstimeofday(&now); 123 words[0] = now.tv_sec; 124 words[1] = now.tv_nsec; 125 126 local_irq_save(flags); 127 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words)); 128 local_irq_restore(flags); 129 } 130 131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...) 132 { 133 int n; 134 va_list args; 135 unsigned long flags; 136 char *buf; 137 138 if (unlikely(bt->trace_state != Blktrace_running && 139 !blk_tracer_enabled)) 140 return; 141 142 local_irq_save(flags); 143 buf = per_cpu_ptr(bt->msg_data, smp_processor_id()); 144 va_start(args, fmt); 145 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args); 146 va_end(args); 147 148 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n); 149 local_irq_restore(flags); 150 } 151 EXPORT_SYMBOL_GPL(__trace_note_message); 152 153 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector, 154 pid_t pid) 155 { 156 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0) 157 return 1; 158 if (sector && (sector < bt->start_lba || sector > bt->end_lba)) 159 return 1; 160 if (bt->pid && pid != bt->pid) 161 return 1; 162 163 return 0; 164 } 165 166 /* 167 * Data direction bit lookup 168 */ 169 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ), 170 BLK_TC_ACT(BLK_TC_WRITE) }; 171 172 #define BLK_TC_HARDBARRIER BLK_TC_BARRIER 173 #define BLK_TC_RAHEAD BLK_TC_AHEAD 174 175 /* The ilog2() calls fall out because they're constant */ 176 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \ 177 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name)) 178 179 /* 180 * The worker for the various blk_add_trace*() types. Fills out a 181 * blk_io_trace structure and places it in a per-cpu subbuffer. 182 */ 183 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes, 184 int rw, u32 what, int error, int pdu_len, void *pdu_data) 185 { 186 struct task_struct *tsk = current; 187 struct ring_buffer_event *event = NULL; 188 struct ring_buffer *buffer = NULL; 189 struct blk_io_trace *t; 190 unsigned long flags = 0; 191 unsigned long *sequence; 192 pid_t pid; 193 int cpu, pc = 0; 194 bool blk_tracer = blk_tracer_enabled; 195 196 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer)) 197 return; 198 199 what |= ddir_act[rw & WRITE]; 200 what |= MASK_TC_BIT(rw, HARDBARRIER); 201 what |= MASK_TC_BIT(rw, SYNC); 202 what |= MASK_TC_BIT(rw, RAHEAD); 203 what |= MASK_TC_BIT(rw, META); 204 what |= MASK_TC_BIT(rw, DISCARD); 205 206 pid = tsk->pid; 207 if (act_log_check(bt, what, sector, pid)) 208 return; 209 cpu = raw_smp_processor_id(); 210 211 if (blk_tracer) { 212 tracing_record_cmdline(current); 213 214 buffer = blk_tr->buffer; 215 pc = preempt_count(); 216 event = trace_buffer_lock_reserve(buffer, TRACE_BLK, 217 sizeof(*t) + pdu_len, 218 0, pc); 219 if (!event) 220 return; 221 t = ring_buffer_event_data(event); 222 goto record_it; 223 } 224 225 /* 226 * A word about the locking here - we disable interrupts to reserve 227 * some space in the relay per-cpu buffer, to prevent an irq 228 * from coming in and stepping on our toes. 229 */ 230 local_irq_save(flags); 231 232 if (unlikely(tsk->btrace_seq != blktrace_seq)) 233 trace_note_tsk(bt, tsk); 234 235 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len); 236 if (t) { 237 sequence = per_cpu_ptr(bt->sequence, cpu); 238 239 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; 240 t->sequence = ++(*sequence); 241 t->time = ktime_to_ns(ktime_get()); 242 record_it: 243 /* 244 * These two are not needed in ftrace as they are in the 245 * generic trace_entry, filled by tracing_generic_entry_update, 246 * but for the trace_event->bin() synthesizer benefit we do it 247 * here too. 248 */ 249 t->cpu = cpu; 250 t->pid = pid; 251 252 t->sector = sector; 253 t->bytes = bytes; 254 t->action = what; 255 t->device = bt->dev; 256 t->error = error; 257 t->pdu_len = pdu_len; 258 259 if (pdu_len) 260 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len); 261 262 if (blk_tracer) { 263 trace_buffer_unlock_commit(buffer, event, 0, pc); 264 return; 265 } 266 } 267 268 local_irq_restore(flags); 269 } 270 271 static struct dentry *blk_tree_root; 272 static DEFINE_MUTEX(blk_tree_mutex); 273 274 static void blk_trace_free(struct blk_trace *bt) 275 { 276 debugfs_remove(bt->msg_file); 277 debugfs_remove(bt->dropped_file); 278 relay_close(bt->rchan); 279 debugfs_remove(bt->dir); 280 free_percpu(bt->sequence); 281 free_percpu(bt->msg_data); 282 kfree(bt); 283 } 284 285 static void blk_trace_cleanup(struct blk_trace *bt) 286 { 287 blk_trace_free(bt); 288 if (atomic_dec_and_test(&blk_probes_ref)) 289 blk_unregister_tracepoints(); 290 } 291 292 int blk_trace_remove(struct request_queue *q) 293 { 294 struct blk_trace *bt; 295 296 bt = xchg(&q->blk_trace, NULL); 297 if (!bt) 298 return -EINVAL; 299 300 if (bt->trace_state != Blktrace_running) 301 blk_trace_cleanup(bt); 302 303 return 0; 304 } 305 EXPORT_SYMBOL_GPL(blk_trace_remove); 306 307 static int blk_dropped_open(struct inode *inode, struct file *filp) 308 { 309 filp->private_data = inode->i_private; 310 311 return 0; 312 } 313 314 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer, 315 size_t count, loff_t *ppos) 316 { 317 struct blk_trace *bt = filp->private_data; 318 char buf[16]; 319 320 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped)); 321 322 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf)); 323 } 324 325 static const struct file_operations blk_dropped_fops = { 326 .owner = THIS_MODULE, 327 .open = blk_dropped_open, 328 .read = blk_dropped_read, 329 .llseek = default_llseek, 330 }; 331 332 static int blk_msg_open(struct inode *inode, struct file *filp) 333 { 334 filp->private_data = inode->i_private; 335 336 return 0; 337 } 338 339 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer, 340 size_t count, loff_t *ppos) 341 { 342 char *msg; 343 struct blk_trace *bt; 344 345 if (count >= BLK_TN_MAX_MSG) 346 return -EINVAL; 347 348 msg = kmalloc(count + 1, GFP_KERNEL); 349 if (msg == NULL) 350 return -ENOMEM; 351 352 if (copy_from_user(msg, buffer, count)) { 353 kfree(msg); 354 return -EFAULT; 355 } 356 357 msg[count] = '\0'; 358 bt = filp->private_data; 359 __trace_note_message(bt, "%s", msg); 360 kfree(msg); 361 362 return count; 363 } 364 365 static const struct file_operations blk_msg_fops = { 366 .owner = THIS_MODULE, 367 .open = blk_msg_open, 368 .write = blk_msg_write, 369 .llseek = noop_llseek, 370 }; 371 372 /* 373 * Keep track of how many times we encountered a full subbuffer, to aid 374 * the user space app in telling how many lost events there were. 375 */ 376 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf, 377 void *prev_subbuf, size_t prev_padding) 378 { 379 struct blk_trace *bt; 380 381 if (!relay_buf_full(buf)) 382 return 1; 383 384 bt = buf->chan->private_data; 385 atomic_inc(&bt->dropped); 386 return 0; 387 } 388 389 static int blk_remove_buf_file_callback(struct dentry *dentry) 390 { 391 debugfs_remove(dentry); 392 393 return 0; 394 } 395 396 static struct dentry *blk_create_buf_file_callback(const char *filename, 397 struct dentry *parent, 398 int mode, 399 struct rchan_buf *buf, 400 int *is_global) 401 { 402 return debugfs_create_file(filename, mode, parent, buf, 403 &relay_file_operations); 404 } 405 406 static struct rchan_callbacks blk_relay_callbacks = { 407 .subbuf_start = blk_subbuf_start_callback, 408 .create_buf_file = blk_create_buf_file_callback, 409 .remove_buf_file = blk_remove_buf_file_callback, 410 }; 411 412 static void blk_trace_setup_lba(struct blk_trace *bt, 413 struct block_device *bdev) 414 { 415 struct hd_struct *part = NULL; 416 417 if (bdev) 418 part = bdev->bd_part; 419 420 if (part) { 421 bt->start_lba = part->start_sect; 422 bt->end_lba = part->start_sect + part->nr_sects; 423 } else { 424 bt->start_lba = 0; 425 bt->end_lba = -1ULL; 426 } 427 } 428 429 /* 430 * Setup everything required to start tracing 431 */ 432 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev, 433 struct block_device *bdev, 434 struct blk_user_trace_setup *buts) 435 { 436 struct blk_trace *old_bt, *bt = NULL; 437 struct dentry *dir = NULL; 438 int ret, i; 439 440 if (!buts->buf_size || !buts->buf_nr) 441 return -EINVAL; 442 443 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE); 444 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0'; 445 446 /* 447 * some device names have larger paths - convert the slashes 448 * to underscores for this to work as expected 449 */ 450 for (i = 0; i < strlen(buts->name); i++) 451 if (buts->name[i] == '/') 452 buts->name[i] = '_'; 453 454 bt = kzalloc(sizeof(*bt), GFP_KERNEL); 455 if (!bt) 456 return -ENOMEM; 457 458 ret = -ENOMEM; 459 bt->sequence = alloc_percpu(unsigned long); 460 if (!bt->sequence) 461 goto err; 462 463 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); 464 if (!bt->msg_data) 465 goto err; 466 467 ret = -ENOENT; 468 469 mutex_lock(&blk_tree_mutex); 470 if (!blk_tree_root) { 471 blk_tree_root = debugfs_create_dir("block", NULL); 472 if (!blk_tree_root) { 473 mutex_unlock(&blk_tree_mutex); 474 goto err; 475 } 476 } 477 mutex_unlock(&blk_tree_mutex); 478 479 dir = debugfs_create_dir(buts->name, blk_tree_root); 480 481 if (!dir) 482 goto err; 483 484 bt->dir = dir; 485 bt->dev = dev; 486 atomic_set(&bt->dropped, 0); 487 488 ret = -EIO; 489 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, 490 &blk_dropped_fops); 491 if (!bt->dropped_file) 492 goto err; 493 494 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops); 495 if (!bt->msg_file) 496 goto err; 497 498 bt->rchan = relay_open("trace", dir, buts->buf_size, 499 buts->buf_nr, &blk_relay_callbacks, bt); 500 if (!bt->rchan) 501 goto err; 502 503 bt->act_mask = buts->act_mask; 504 if (!bt->act_mask) 505 bt->act_mask = (u16) -1; 506 507 blk_trace_setup_lba(bt, bdev); 508 509 /* overwrite with user settings */ 510 if (buts->start_lba) 511 bt->start_lba = buts->start_lba; 512 if (buts->end_lba) 513 bt->end_lba = buts->end_lba; 514 515 bt->pid = buts->pid; 516 bt->trace_state = Blktrace_setup; 517 518 ret = -EBUSY; 519 old_bt = xchg(&q->blk_trace, bt); 520 if (old_bt) { 521 (void) xchg(&q->blk_trace, old_bt); 522 goto err; 523 } 524 525 if (atomic_inc_return(&blk_probes_ref) == 1) 526 blk_register_tracepoints(); 527 528 return 0; 529 err: 530 blk_trace_free(bt); 531 return ret; 532 } 533 534 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev, 535 struct block_device *bdev, 536 char __user *arg) 537 { 538 struct blk_user_trace_setup buts; 539 int ret; 540 541 ret = copy_from_user(&buts, arg, sizeof(buts)); 542 if (ret) 543 return -EFAULT; 544 545 ret = do_blk_trace_setup(q, name, dev, bdev, &buts); 546 if (ret) 547 return ret; 548 549 if (copy_to_user(arg, &buts, sizeof(buts))) { 550 blk_trace_remove(q); 551 return -EFAULT; 552 } 553 return 0; 554 } 555 EXPORT_SYMBOL_GPL(blk_trace_setup); 556 557 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64) 558 static int compat_blk_trace_setup(struct request_queue *q, char *name, 559 dev_t dev, struct block_device *bdev, 560 char __user *arg) 561 { 562 struct blk_user_trace_setup buts; 563 struct compat_blk_user_trace_setup cbuts; 564 int ret; 565 566 if (copy_from_user(&cbuts, arg, sizeof(cbuts))) 567 return -EFAULT; 568 569 buts = (struct blk_user_trace_setup) { 570 .act_mask = cbuts.act_mask, 571 .buf_size = cbuts.buf_size, 572 .buf_nr = cbuts.buf_nr, 573 .start_lba = cbuts.start_lba, 574 .end_lba = cbuts.end_lba, 575 .pid = cbuts.pid, 576 }; 577 memcpy(&buts.name, &cbuts.name, 32); 578 579 ret = do_blk_trace_setup(q, name, dev, bdev, &buts); 580 if (ret) 581 return ret; 582 583 if (copy_to_user(arg, &buts.name, 32)) { 584 blk_trace_remove(q); 585 return -EFAULT; 586 } 587 588 return 0; 589 } 590 #endif 591 592 int blk_trace_startstop(struct request_queue *q, int start) 593 { 594 int ret; 595 struct blk_trace *bt = q->blk_trace; 596 597 if (bt == NULL) 598 return -EINVAL; 599 600 /* 601 * For starting a trace, we can transition from a setup or stopped 602 * trace. For stopping a trace, the state must be running 603 */ 604 ret = -EINVAL; 605 if (start) { 606 if (bt->trace_state == Blktrace_setup || 607 bt->trace_state == Blktrace_stopped) { 608 blktrace_seq++; 609 smp_mb(); 610 bt->trace_state = Blktrace_running; 611 612 trace_note_time(bt); 613 ret = 0; 614 } 615 } else { 616 if (bt->trace_state == Blktrace_running) { 617 bt->trace_state = Blktrace_stopped; 618 relay_flush(bt->rchan); 619 ret = 0; 620 } 621 } 622 623 return ret; 624 } 625 EXPORT_SYMBOL_GPL(blk_trace_startstop); 626 627 /** 628 * blk_trace_ioctl: - handle the ioctls associated with tracing 629 * @bdev: the block device 630 * @cmd: the ioctl cmd 631 * @arg: the argument data, if any 632 * 633 **/ 634 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg) 635 { 636 struct request_queue *q; 637 int ret, start = 0; 638 char b[BDEVNAME_SIZE]; 639 640 q = bdev_get_queue(bdev); 641 if (!q) 642 return -ENXIO; 643 644 lock_kernel(); 645 mutex_lock(&bdev->bd_mutex); 646 647 switch (cmd) { 648 case BLKTRACESETUP: 649 bdevname(bdev, b); 650 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg); 651 break; 652 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64) 653 case BLKTRACESETUP32: 654 bdevname(bdev, b); 655 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg); 656 break; 657 #endif 658 case BLKTRACESTART: 659 start = 1; 660 case BLKTRACESTOP: 661 ret = blk_trace_startstop(q, start); 662 break; 663 case BLKTRACETEARDOWN: 664 ret = blk_trace_remove(q); 665 break; 666 default: 667 ret = -ENOTTY; 668 break; 669 } 670 671 mutex_unlock(&bdev->bd_mutex); 672 unlock_kernel(); 673 return ret; 674 } 675 676 /** 677 * blk_trace_shutdown: - stop and cleanup trace structures 678 * @q: the request queue associated with the device 679 * 680 **/ 681 void blk_trace_shutdown(struct request_queue *q) 682 { 683 if (q->blk_trace) { 684 blk_trace_startstop(q, 0); 685 blk_trace_remove(q); 686 } 687 } 688 689 /* 690 * blktrace probes 691 */ 692 693 /** 694 * blk_add_trace_rq - Add a trace for a request oriented action 695 * @q: queue the io is for 696 * @rq: the source request 697 * @what: the action 698 * 699 * Description: 700 * Records an action against a request. Will log the bio offset + size. 701 * 702 **/ 703 static void blk_add_trace_rq(struct request_queue *q, struct request *rq, 704 u32 what) 705 { 706 struct blk_trace *bt = q->blk_trace; 707 int rw = rq->cmd_flags & 0x03; 708 709 if (likely(!bt)) 710 return; 711 712 if (rq->cmd_flags & REQ_DISCARD) 713 rw |= REQ_DISCARD; 714 715 if (rq->cmd_flags & REQ_SECURE) 716 rw |= REQ_SECURE; 717 718 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { 719 what |= BLK_TC_ACT(BLK_TC_PC); 720 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw, 721 what, rq->errors, rq->cmd_len, rq->cmd); 722 } else { 723 what |= BLK_TC_ACT(BLK_TC_FS); 724 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw, 725 what, rq->errors, 0, NULL); 726 } 727 } 728 729 static void blk_add_trace_rq_abort(void *ignore, 730 struct request_queue *q, struct request *rq) 731 { 732 blk_add_trace_rq(q, rq, BLK_TA_ABORT); 733 } 734 735 static void blk_add_trace_rq_insert(void *ignore, 736 struct request_queue *q, struct request *rq) 737 { 738 blk_add_trace_rq(q, rq, BLK_TA_INSERT); 739 } 740 741 static void blk_add_trace_rq_issue(void *ignore, 742 struct request_queue *q, struct request *rq) 743 { 744 blk_add_trace_rq(q, rq, BLK_TA_ISSUE); 745 } 746 747 static void blk_add_trace_rq_requeue(void *ignore, 748 struct request_queue *q, 749 struct request *rq) 750 { 751 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE); 752 } 753 754 static void blk_add_trace_rq_complete(void *ignore, 755 struct request_queue *q, 756 struct request *rq) 757 { 758 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE); 759 } 760 761 /** 762 * blk_add_trace_bio - Add a trace for a bio oriented action 763 * @q: queue the io is for 764 * @bio: the source bio 765 * @what: the action 766 * 767 * Description: 768 * Records an action against a bio. Will log the bio offset + size. 769 * 770 **/ 771 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio, 772 u32 what) 773 { 774 struct blk_trace *bt = q->blk_trace; 775 776 if (likely(!bt)) 777 return; 778 779 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what, 780 !bio_flagged(bio, BIO_UPTODATE), 0, NULL); 781 } 782 783 static void blk_add_trace_bio_bounce(void *ignore, 784 struct request_queue *q, struct bio *bio) 785 { 786 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE); 787 } 788 789 static void blk_add_trace_bio_complete(void *ignore, 790 struct request_queue *q, struct bio *bio) 791 { 792 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE); 793 } 794 795 static void blk_add_trace_bio_backmerge(void *ignore, 796 struct request_queue *q, 797 struct bio *bio) 798 { 799 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE); 800 } 801 802 static void blk_add_trace_bio_frontmerge(void *ignore, 803 struct request_queue *q, 804 struct bio *bio) 805 { 806 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE); 807 } 808 809 static void blk_add_trace_bio_queue(void *ignore, 810 struct request_queue *q, struct bio *bio) 811 { 812 blk_add_trace_bio(q, bio, BLK_TA_QUEUE); 813 } 814 815 static void blk_add_trace_getrq(void *ignore, 816 struct request_queue *q, 817 struct bio *bio, int rw) 818 { 819 if (bio) 820 blk_add_trace_bio(q, bio, BLK_TA_GETRQ); 821 else { 822 struct blk_trace *bt = q->blk_trace; 823 824 if (bt) 825 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL); 826 } 827 } 828 829 830 static void blk_add_trace_sleeprq(void *ignore, 831 struct request_queue *q, 832 struct bio *bio, int rw) 833 { 834 if (bio) 835 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ); 836 else { 837 struct blk_trace *bt = q->blk_trace; 838 839 if (bt) 840 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ, 841 0, 0, NULL); 842 } 843 } 844 845 static void blk_add_trace_plug(void *ignore, struct request_queue *q) 846 { 847 struct blk_trace *bt = q->blk_trace; 848 849 if (bt) 850 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL); 851 } 852 853 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q) 854 { 855 struct blk_trace *bt = q->blk_trace; 856 857 if (bt) { 858 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE]; 859 __be64 rpdu = cpu_to_be64(pdu); 860 861 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0, 862 sizeof(rpdu), &rpdu); 863 } 864 } 865 866 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q) 867 { 868 struct blk_trace *bt = q->blk_trace; 869 870 if (bt) { 871 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE]; 872 __be64 rpdu = cpu_to_be64(pdu); 873 874 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0, 875 sizeof(rpdu), &rpdu); 876 } 877 } 878 879 static void blk_add_trace_split(void *ignore, 880 struct request_queue *q, struct bio *bio, 881 unsigned int pdu) 882 { 883 struct blk_trace *bt = q->blk_trace; 884 885 if (bt) { 886 __be64 rpdu = cpu_to_be64(pdu); 887 888 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, 889 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE), 890 sizeof(rpdu), &rpdu); 891 } 892 } 893 894 /** 895 * blk_add_trace_remap - Add a trace for a remap operation 896 * @ignore: trace callback data parameter (not used) 897 * @q: queue the io is for 898 * @bio: the source bio 899 * @dev: target device 900 * @from: source sector 901 * 902 * Description: 903 * Device mapper or raid target sometimes need to split a bio because 904 * it spans a stripe (or similar). Add a trace for that action. 905 * 906 **/ 907 static void blk_add_trace_remap(void *ignore, 908 struct request_queue *q, struct bio *bio, 909 dev_t dev, sector_t from) 910 { 911 struct blk_trace *bt = q->blk_trace; 912 struct blk_io_trace_remap r; 913 914 if (likely(!bt)) 915 return; 916 917 r.device_from = cpu_to_be32(dev); 918 r.device_to = cpu_to_be32(bio->bi_bdev->bd_dev); 919 r.sector_from = cpu_to_be64(from); 920 921 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, 922 BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE), 923 sizeof(r), &r); 924 } 925 926 /** 927 * blk_add_trace_rq_remap - Add a trace for a request-remap operation 928 * @ignore: trace callback data parameter (not used) 929 * @q: queue the io is for 930 * @rq: the source request 931 * @dev: target device 932 * @from: source sector 933 * 934 * Description: 935 * Device mapper remaps request to other devices. 936 * Add a trace for that action. 937 * 938 **/ 939 static void blk_add_trace_rq_remap(void *ignore, 940 struct request_queue *q, 941 struct request *rq, dev_t dev, 942 sector_t from) 943 { 944 struct blk_trace *bt = q->blk_trace; 945 struct blk_io_trace_remap r; 946 947 if (likely(!bt)) 948 return; 949 950 r.device_from = cpu_to_be32(dev); 951 r.device_to = cpu_to_be32(disk_devt(rq->rq_disk)); 952 r.sector_from = cpu_to_be64(from); 953 954 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 955 rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors, 956 sizeof(r), &r); 957 } 958 959 /** 960 * blk_add_driver_data - Add binary message with driver-specific data 961 * @q: queue the io is for 962 * @rq: io request 963 * @data: driver-specific data 964 * @len: length of driver-specific data 965 * 966 * Description: 967 * Some drivers might want to write driver-specific data per request. 968 * 969 **/ 970 void blk_add_driver_data(struct request_queue *q, 971 struct request *rq, 972 void *data, size_t len) 973 { 974 struct blk_trace *bt = q->blk_trace; 975 976 if (likely(!bt)) 977 return; 978 979 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) 980 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 981 BLK_TA_DRV_DATA, rq->errors, len, data); 982 else 983 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 984 BLK_TA_DRV_DATA, rq->errors, len, data); 985 } 986 EXPORT_SYMBOL_GPL(blk_add_driver_data); 987 988 static void blk_register_tracepoints(void) 989 { 990 int ret; 991 992 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL); 993 WARN_ON(ret); 994 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL); 995 WARN_ON(ret); 996 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL); 997 WARN_ON(ret); 998 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL); 999 WARN_ON(ret); 1000 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL); 1001 WARN_ON(ret); 1002 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL); 1003 WARN_ON(ret); 1004 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL); 1005 WARN_ON(ret); 1006 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL); 1007 WARN_ON(ret); 1008 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL); 1009 WARN_ON(ret); 1010 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL); 1011 WARN_ON(ret); 1012 ret = register_trace_block_getrq(blk_add_trace_getrq, NULL); 1013 WARN_ON(ret); 1014 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL); 1015 WARN_ON(ret); 1016 ret = register_trace_block_plug(blk_add_trace_plug, NULL); 1017 WARN_ON(ret); 1018 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL); 1019 WARN_ON(ret); 1020 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL); 1021 WARN_ON(ret); 1022 ret = register_trace_block_split(blk_add_trace_split, NULL); 1023 WARN_ON(ret); 1024 ret = register_trace_block_remap(blk_add_trace_remap, NULL); 1025 WARN_ON(ret); 1026 ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL); 1027 WARN_ON(ret); 1028 } 1029 1030 static void blk_unregister_tracepoints(void) 1031 { 1032 unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL); 1033 unregister_trace_block_remap(blk_add_trace_remap, NULL); 1034 unregister_trace_block_split(blk_add_trace_split, NULL); 1035 unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL); 1036 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, 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 1058 static 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_DISCARD) 1069 rwbs[i++] = 'D'; 1070 else if (tc & BLK_TC_WRITE) 1071 rwbs[i++] = 'W'; 1072 else if (t->bytes) 1073 rwbs[i++] = 'R'; 1074 else 1075 rwbs[i++] = 'N'; 1076 1077 if (tc & BLK_TC_AHEAD) 1078 rwbs[i++] = 'A'; 1079 if (tc & BLK_TC_BARRIER) 1080 rwbs[i++] = 'B'; 1081 if (tc & BLK_TC_SYNC) 1082 rwbs[i++] = 'S'; 1083 if (tc & BLK_TC_META) 1084 rwbs[i++] = 'M'; 1085 out: 1086 rwbs[i] = '\0'; 1087 } 1088 1089 static inline 1090 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent) 1091 { 1092 return (const struct blk_io_trace *)ent; 1093 } 1094 1095 static inline const void *pdu_start(const struct trace_entry *ent) 1096 { 1097 return te_blk_io_trace(ent) + 1; 1098 } 1099 1100 static inline u32 t_action(const struct trace_entry *ent) 1101 { 1102 return te_blk_io_trace(ent)->action; 1103 } 1104 1105 static inline u32 t_bytes(const struct trace_entry *ent) 1106 { 1107 return te_blk_io_trace(ent)->bytes; 1108 } 1109 1110 static inline u32 t_sec(const struct trace_entry *ent) 1111 { 1112 return te_blk_io_trace(ent)->bytes >> 9; 1113 } 1114 1115 static inline unsigned long long t_sector(const struct trace_entry *ent) 1116 { 1117 return te_blk_io_trace(ent)->sector; 1118 } 1119 1120 static inline __u16 t_error(const struct trace_entry *ent) 1121 { 1122 return te_blk_io_trace(ent)->error; 1123 } 1124 1125 static __u64 get_pdu_int(const struct trace_entry *ent) 1126 { 1127 const __u64 *val = pdu_start(ent); 1128 return be64_to_cpu(*val); 1129 } 1130 1131 static void get_pdu_remap(const struct trace_entry *ent, 1132 struct blk_io_trace_remap *r) 1133 { 1134 const struct blk_io_trace_remap *__r = pdu_start(ent); 1135 __u64 sector_from = __r->sector_from; 1136 1137 r->device_from = be32_to_cpu(__r->device_from); 1138 r->device_to = be32_to_cpu(__r->device_to); 1139 r->sector_from = be64_to_cpu(sector_from); 1140 } 1141 1142 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act); 1143 1144 static int blk_log_action_classic(struct trace_iterator *iter, const char *act) 1145 { 1146 char rwbs[6]; 1147 unsigned long long ts = iter->ts; 1148 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC); 1149 unsigned secs = (unsigned long)ts; 1150 const struct blk_io_trace *t = te_blk_io_trace(iter->ent); 1151 1152 fill_rwbs(rwbs, t); 1153 1154 return trace_seq_printf(&iter->seq, 1155 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ", 1156 MAJOR(t->device), MINOR(t->device), iter->cpu, 1157 secs, nsec_rem, iter->ent->pid, act, rwbs); 1158 } 1159 1160 static int blk_log_action(struct trace_iterator *iter, const char *act) 1161 { 1162 char rwbs[6]; 1163 const struct blk_io_trace *t = te_blk_io_trace(iter->ent); 1164 1165 fill_rwbs(rwbs, t); 1166 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ", 1167 MAJOR(t->device), MINOR(t->device), act, rwbs); 1168 } 1169 1170 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent) 1171 { 1172 const unsigned char *pdu_buf; 1173 int pdu_len; 1174 int i, end, ret; 1175 1176 pdu_buf = pdu_start(ent); 1177 pdu_len = te_blk_io_trace(ent)->pdu_len; 1178 1179 if (!pdu_len) 1180 return 1; 1181 1182 /* find the last zero that needs to be printed */ 1183 for (end = pdu_len - 1; end >= 0; end--) 1184 if (pdu_buf[end]) 1185 break; 1186 end++; 1187 1188 if (!trace_seq_putc(s, '(')) 1189 return 0; 1190 1191 for (i = 0; i < pdu_len; i++) { 1192 1193 ret = trace_seq_printf(s, "%s%02x", 1194 i == 0 ? "" : " ", pdu_buf[i]); 1195 if (!ret) 1196 return ret; 1197 1198 /* 1199 * stop when the rest is just zeroes and indicate so 1200 * with a ".." appended 1201 */ 1202 if (i == end && end != pdu_len - 1) 1203 return trace_seq_puts(s, " ..) "); 1204 } 1205 1206 return trace_seq_puts(s, ") "); 1207 } 1208 1209 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent) 1210 { 1211 char cmd[TASK_COMM_LEN]; 1212 1213 trace_find_cmdline(ent->pid, cmd); 1214 1215 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { 1216 int ret; 1217 1218 ret = trace_seq_printf(s, "%u ", t_bytes(ent)); 1219 if (!ret) 1220 return 0; 1221 ret = blk_log_dump_pdu(s, ent); 1222 if (!ret) 1223 return 0; 1224 return trace_seq_printf(s, "[%s]\n", cmd); 1225 } else { 1226 if (t_sec(ent)) 1227 return trace_seq_printf(s, "%llu + %u [%s]\n", 1228 t_sector(ent), t_sec(ent), cmd); 1229 return trace_seq_printf(s, "[%s]\n", cmd); 1230 } 1231 } 1232 1233 static int blk_log_with_error(struct trace_seq *s, 1234 const struct trace_entry *ent) 1235 { 1236 if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { 1237 int ret; 1238 1239 ret = blk_log_dump_pdu(s, ent); 1240 if (ret) 1241 return trace_seq_printf(s, "[%d]\n", t_error(ent)); 1242 return 0; 1243 } else { 1244 if (t_sec(ent)) 1245 return trace_seq_printf(s, "%llu + %u [%d]\n", 1246 t_sector(ent), 1247 t_sec(ent), t_error(ent)); 1248 return trace_seq_printf(s, "%llu [%d]\n", 1249 t_sector(ent), t_error(ent)); 1250 } 1251 } 1252 1253 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent) 1254 { 1255 struct blk_io_trace_remap r = { .device_from = 0, }; 1256 1257 get_pdu_remap(ent, &r); 1258 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n", 1259 t_sector(ent), t_sec(ent), 1260 MAJOR(r.device_from), MINOR(r.device_from), 1261 (unsigned long long)r.sector_from); 1262 } 1263 1264 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent) 1265 { 1266 char cmd[TASK_COMM_LEN]; 1267 1268 trace_find_cmdline(ent->pid, cmd); 1269 1270 return trace_seq_printf(s, "[%s]\n", cmd); 1271 } 1272 1273 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent) 1274 { 1275 char cmd[TASK_COMM_LEN]; 1276 1277 trace_find_cmdline(ent->pid, cmd); 1278 1279 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent)); 1280 } 1281 1282 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent) 1283 { 1284 char cmd[TASK_COMM_LEN]; 1285 1286 trace_find_cmdline(ent->pid, cmd); 1287 1288 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent), 1289 get_pdu_int(ent), cmd); 1290 } 1291 1292 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent) 1293 { 1294 int ret; 1295 const struct blk_io_trace *t = te_blk_io_trace(ent); 1296 1297 ret = trace_seq_putmem(s, t + 1, t->pdu_len); 1298 if (ret) 1299 return trace_seq_putc(s, '\n'); 1300 return ret; 1301 } 1302 1303 /* 1304 * struct tracer operations 1305 */ 1306 1307 static void blk_tracer_print_header(struct seq_file *m) 1308 { 1309 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) 1310 return; 1311 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n" 1312 "# | | | | | |\n"); 1313 } 1314 1315 static void blk_tracer_start(struct trace_array *tr) 1316 { 1317 blk_tracer_enabled = true; 1318 } 1319 1320 static int blk_tracer_init(struct trace_array *tr) 1321 { 1322 blk_tr = tr; 1323 blk_tracer_start(tr); 1324 return 0; 1325 } 1326 1327 static void blk_tracer_stop(struct trace_array *tr) 1328 { 1329 blk_tracer_enabled = false; 1330 } 1331 1332 static void blk_tracer_reset(struct trace_array *tr) 1333 { 1334 blk_tracer_stop(tr); 1335 } 1336 1337 static const struct { 1338 const char *act[2]; 1339 int (*print)(struct trace_seq *s, const struct trace_entry *ent); 1340 } what2act[] = { 1341 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic }, 1342 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic }, 1343 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic }, 1344 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic }, 1345 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic }, 1346 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error }, 1347 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic }, 1348 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error }, 1349 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug }, 1350 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug }, 1351 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug }, 1352 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic }, 1353 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split }, 1354 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic }, 1355 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap }, 1356 }; 1357 1358 static enum print_line_t print_one_line(struct trace_iterator *iter, 1359 bool classic) 1360 { 1361 struct trace_seq *s = &iter->seq; 1362 const struct blk_io_trace *t; 1363 u16 what; 1364 int ret; 1365 bool long_act; 1366 blk_log_action_t *log_action; 1367 1368 t = te_blk_io_trace(iter->ent); 1369 what = t->action & ((1 << BLK_TC_SHIFT) - 1); 1370 long_act = !!(trace_flags & TRACE_ITER_VERBOSE); 1371 log_action = classic ? &blk_log_action_classic : &blk_log_action; 1372 1373 if (t->action == BLK_TN_MESSAGE) { 1374 ret = log_action(iter, long_act ? "message" : "m"); 1375 if (ret) 1376 ret = blk_log_msg(s, iter->ent); 1377 goto out; 1378 } 1379 1380 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act))) 1381 ret = trace_seq_printf(s, "Unknown action %x\n", what); 1382 else { 1383 ret = log_action(iter, what2act[what].act[long_act]); 1384 if (ret) 1385 ret = what2act[what].print(s, iter->ent); 1386 } 1387 out: 1388 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE; 1389 } 1390 1391 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter, 1392 int flags, struct trace_event *event) 1393 { 1394 return print_one_line(iter, false); 1395 } 1396 1397 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter) 1398 { 1399 struct trace_seq *s = &iter->seq; 1400 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent; 1401 const int offset = offsetof(struct blk_io_trace, sector); 1402 struct blk_io_trace old = { 1403 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION, 1404 .time = iter->ts, 1405 }; 1406 1407 if (!trace_seq_putmem(s, &old, offset)) 1408 return 0; 1409 return trace_seq_putmem(s, &t->sector, 1410 sizeof(old) - offset + t->pdu_len); 1411 } 1412 1413 static enum print_line_t 1414 blk_trace_event_print_binary(struct trace_iterator *iter, int flags, 1415 struct trace_event *event) 1416 { 1417 return blk_trace_synthesize_old_trace(iter) ? 1418 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE; 1419 } 1420 1421 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter) 1422 { 1423 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) 1424 return TRACE_TYPE_UNHANDLED; 1425 1426 return print_one_line(iter, true); 1427 } 1428 1429 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set) 1430 { 1431 /* don't output context-info for blk_classic output */ 1432 if (bit == TRACE_BLK_OPT_CLASSIC) { 1433 if (set) 1434 trace_flags &= ~TRACE_ITER_CONTEXT_INFO; 1435 else 1436 trace_flags |= TRACE_ITER_CONTEXT_INFO; 1437 } 1438 return 0; 1439 } 1440 1441 static struct tracer blk_tracer __read_mostly = { 1442 .name = "blk", 1443 .init = blk_tracer_init, 1444 .reset = blk_tracer_reset, 1445 .start = blk_tracer_start, 1446 .stop = blk_tracer_stop, 1447 .print_header = blk_tracer_print_header, 1448 .print_line = blk_tracer_print_line, 1449 .flags = &blk_tracer_flags, 1450 .set_flag = blk_tracer_set_flag, 1451 }; 1452 1453 static struct trace_event_functions trace_blk_event_funcs = { 1454 .trace = blk_trace_event_print, 1455 .binary = blk_trace_event_print_binary, 1456 }; 1457 1458 static struct trace_event trace_blk_event = { 1459 .type = TRACE_BLK, 1460 .funcs = &trace_blk_event_funcs, 1461 }; 1462 1463 static int __init init_blk_tracer(void) 1464 { 1465 if (!register_ftrace_event(&trace_blk_event)) { 1466 pr_warning("Warning: could not register block events\n"); 1467 return 1; 1468 } 1469 1470 if (register_tracer(&blk_tracer) != 0) { 1471 pr_warning("Warning: could not register the block tracer\n"); 1472 unregister_ftrace_event(&trace_blk_event); 1473 return 1; 1474 } 1475 1476 return 0; 1477 } 1478 1479 device_initcall(init_blk_tracer); 1480 1481 static int blk_trace_remove_queue(struct request_queue *q) 1482 { 1483 struct blk_trace *bt; 1484 1485 bt = xchg(&q->blk_trace, NULL); 1486 if (bt == NULL) 1487 return -EINVAL; 1488 1489 if (atomic_dec_and_test(&blk_probes_ref)) 1490 blk_unregister_tracepoints(); 1491 1492 blk_trace_free(bt); 1493 return 0; 1494 } 1495 1496 /* 1497 * Setup everything required to start tracing 1498 */ 1499 static int blk_trace_setup_queue(struct request_queue *q, 1500 struct block_device *bdev) 1501 { 1502 struct blk_trace *old_bt, *bt = NULL; 1503 int ret = -ENOMEM; 1504 1505 bt = kzalloc(sizeof(*bt), GFP_KERNEL); 1506 if (!bt) 1507 return -ENOMEM; 1508 1509 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); 1510 if (!bt->msg_data) 1511 goto free_bt; 1512 1513 bt->dev = bdev->bd_dev; 1514 bt->act_mask = (u16)-1; 1515 1516 blk_trace_setup_lba(bt, bdev); 1517 1518 old_bt = xchg(&q->blk_trace, bt); 1519 if (old_bt != NULL) { 1520 (void)xchg(&q->blk_trace, old_bt); 1521 ret = -EBUSY; 1522 goto free_bt; 1523 } 1524 1525 if (atomic_inc_return(&blk_probes_ref) == 1) 1526 blk_register_tracepoints(); 1527 return 0; 1528 1529 free_bt: 1530 blk_trace_free(bt); 1531 return ret; 1532 } 1533 1534 /* 1535 * sysfs interface to enable and configure tracing 1536 */ 1537 1538 static ssize_t sysfs_blk_trace_attr_show(struct device *dev, 1539 struct device_attribute *attr, 1540 char *buf); 1541 static ssize_t sysfs_blk_trace_attr_store(struct device *dev, 1542 struct device_attribute *attr, 1543 const char *buf, size_t count); 1544 #define BLK_TRACE_DEVICE_ATTR(_name) \ 1545 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \ 1546 sysfs_blk_trace_attr_show, \ 1547 sysfs_blk_trace_attr_store) 1548 1549 static BLK_TRACE_DEVICE_ATTR(enable); 1550 static BLK_TRACE_DEVICE_ATTR(act_mask); 1551 static BLK_TRACE_DEVICE_ATTR(pid); 1552 static BLK_TRACE_DEVICE_ATTR(start_lba); 1553 static BLK_TRACE_DEVICE_ATTR(end_lba); 1554 1555 static struct attribute *blk_trace_attrs[] = { 1556 &dev_attr_enable.attr, 1557 &dev_attr_act_mask.attr, 1558 &dev_attr_pid.attr, 1559 &dev_attr_start_lba.attr, 1560 &dev_attr_end_lba.attr, 1561 NULL 1562 }; 1563 1564 struct attribute_group blk_trace_attr_group = { 1565 .name = "trace", 1566 .attrs = blk_trace_attrs, 1567 }; 1568 1569 static const struct { 1570 int mask; 1571 const char *str; 1572 } mask_maps[] = { 1573 { BLK_TC_READ, "read" }, 1574 { BLK_TC_WRITE, "write" }, 1575 { BLK_TC_BARRIER, "barrier" }, 1576 { BLK_TC_SYNC, "sync" }, 1577 { BLK_TC_QUEUE, "queue" }, 1578 { BLK_TC_REQUEUE, "requeue" }, 1579 { BLK_TC_ISSUE, "issue" }, 1580 { BLK_TC_COMPLETE, "complete" }, 1581 { BLK_TC_FS, "fs" }, 1582 { BLK_TC_PC, "pc" }, 1583 { BLK_TC_AHEAD, "ahead" }, 1584 { BLK_TC_META, "meta" }, 1585 { BLK_TC_DISCARD, "discard" }, 1586 { BLK_TC_DRV_DATA, "drv_data" }, 1587 }; 1588 1589 static int blk_trace_str2mask(const char *str) 1590 { 1591 int i; 1592 int mask = 0; 1593 char *buf, *s, *token; 1594 1595 buf = kstrdup(str, GFP_KERNEL); 1596 if (buf == NULL) 1597 return -ENOMEM; 1598 s = strstrip(buf); 1599 1600 while (1) { 1601 token = strsep(&s, ","); 1602 if (token == NULL) 1603 break; 1604 1605 if (*token == '\0') 1606 continue; 1607 1608 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { 1609 if (strcasecmp(token, mask_maps[i].str) == 0) { 1610 mask |= mask_maps[i].mask; 1611 break; 1612 } 1613 } 1614 if (i == ARRAY_SIZE(mask_maps)) { 1615 mask = -EINVAL; 1616 break; 1617 } 1618 } 1619 kfree(buf); 1620 1621 return mask; 1622 } 1623 1624 static ssize_t blk_trace_mask2str(char *buf, int mask) 1625 { 1626 int i; 1627 char *p = buf; 1628 1629 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { 1630 if (mask & mask_maps[i].mask) { 1631 p += sprintf(p, "%s%s", 1632 (p == buf) ? "" : ",", mask_maps[i].str); 1633 } 1634 } 1635 *p++ = '\n'; 1636 1637 return p - buf; 1638 } 1639 1640 static struct request_queue *blk_trace_get_queue(struct block_device *bdev) 1641 { 1642 if (bdev->bd_disk == NULL) 1643 return NULL; 1644 1645 return bdev_get_queue(bdev); 1646 } 1647 1648 static ssize_t sysfs_blk_trace_attr_show(struct device *dev, 1649 struct device_attribute *attr, 1650 char *buf) 1651 { 1652 struct hd_struct *p = dev_to_part(dev); 1653 struct request_queue *q; 1654 struct block_device *bdev; 1655 ssize_t ret = -ENXIO; 1656 1657 lock_kernel(); 1658 bdev = bdget(part_devt(p)); 1659 if (bdev == NULL) 1660 goto out_unlock_kernel; 1661 1662 q = blk_trace_get_queue(bdev); 1663 if (q == NULL) 1664 goto out_bdput; 1665 1666 mutex_lock(&bdev->bd_mutex); 1667 1668 if (attr == &dev_attr_enable) { 1669 ret = sprintf(buf, "%u\n", !!q->blk_trace); 1670 goto out_unlock_bdev; 1671 } 1672 1673 if (q->blk_trace == NULL) 1674 ret = sprintf(buf, "disabled\n"); 1675 else if (attr == &dev_attr_act_mask) 1676 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask); 1677 else if (attr == &dev_attr_pid) 1678 ret = sprintf(buf, "%u\n", q->blk_trace->pid); 1679 else if (attr == &dev_attr_start_lba) 1680 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba); 1681 else if (attr == &dev_attr_end_lba) 1682 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba); 1683 1684 out_unlock_bdev: 1685 mutex_unlock(&bdev->bd_mutex); 1686 out_bdput: 1687 bdput(bdev); 1688 out_unlock_kernel: 1689 unlock_kernel(); 1690 return ret; 1691 } 1692 1693 static ssize_t sysfs_blk_trace_attr_store(struct device *dev, 1694 struct device_attribute *attr, 1695 const char *buf, size_t count) 1696 { 1697 struct block_device *bdev; 1698 struct request_queue *q; 1699 struct hd_struct *p; 1700 u64 value; 1701 ssize_t ret = -EINVAL; 1702 1703 if (count == 0) 1704 goto out; 1705 1706 if (attr == &dev_attr_act_mask) { 1707 if (sscanf(buf, "%llx", &value) != 1) { 1708 /* Assume it is a list of trace category names */ 1709 ret = blk_trace_str2mask(buf); 1710 if (ret < 0) 1711 goto out; 1712 value = ret; 1713 } 1714 } else if (sscanf(buf, "%llu", &value) != 1) 1715 goto out; 1716 1717 ret = -ENXIO; 1718 1719 lock_kernel(); 1720 p = dev_to_part(dev); 1721 bdev = bdget(part_devt(p)); 1722 if (bdev == NULL) 1723 goto out_unlock_kernel; 1724 1725 q = blk_trace_get_queue(bdev); 1726 if (q == NULL) 1727 goto out_bdput; 1728 1729 mutex_lock(&bdev->bd_mutex); 1730 1731 if (attr == &dev_attr_enable) { 1732 if (value) 1733 ret = blk_trace_setup_queue(q, bdev); 1734 else 1735 ret = blk_trace_remove_queue(q); 1736 goto out_unlock_bdev; 1737 } 1738 1739 ret = 0; 1740 if (q->blk_trace == NULL) 1741 ret = blk_trace_setup_queue(q, bdev); 1742 1743 if (ret == 0) { 1744 if (attr == &dev_attr_act_mask) 1745 q->blk_trace->act_mask = value; 1746 else if (attr == &dev_attr_pid) 1747 q->blk_trace->pid = value; 1748 else if (attr == &dev_attr_start_lba) 1749 q->blk_trace->start_lba = value; 1750 else if (attr == &dev_attr_end_lba) 1751 q->blk_trace->end_lba = value; 1752 } 1753 1754 out_unlock_bdev: 1755 mutex_unlock(&bdev->bd_mutex); 1756 out_bdput: 1757 bdput(bdev); 1758 out_unlock_kernel: 1759 unlock_kernel(); 1760 out: 1761 return ret ? ret : count; 1762 } 1763 1764 int blk_trace_init_sysfs(struct device *dev) 1765 { 1766 return sysfs_create_group(&dev->kobj, &blk_trace_attr_group); 1767 } 1768 1769 void blk_trace_remove_sysfs(struct device *dev) 1770 { 1771 sysfs_remove_group(&dev->kobj, &blk_trace_attr_group); 1772 } 1773 1774 #endif /* CONFIG_BLK_DEV_IO_TRACE */ 1775 1776 #ifdef CONFIG_EVENT_TRACING 1777 1778 void blk_dump_cmd(char *buf, struct request *rq) 1779 { 1780 int i, end; 1781 int len = rq->cmd_len; 1782 unsigned char *cmd = rq->cmd; 1783 1784 if (rq->cmd_type != REQ_TYPE_BLOCK_PC) { 1785 buf[0] = '\0'; 1786 return; 1787 } 1788 1789 for (end = len - 1; end >= 0; end--) 1790 if (cmd[end]) 1791 break; 1792 end++; 1793 1794 for (i = 0; i < len; i++) { 1795 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]); 1796 if (i == end && end != len - 1) { 1797 sprintf(buf, " .."); 1798 break; 1799 } 1800 } 1801 } 1802 1803 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes) 1804 { 1805 int i = 0; 1806 1807 if (rw & WRITE) 1808 rwbs[i++] = 'W'; 1809 else if (rw & REQ_DISCARD) 1810 rwbs[i++] = 'D'; 1811 else if (bytes) 1812 rwbs[i++] = 'R'; 1813 else 1814 rwbs[i++] = 'N'; 1815 1816 if (rw & REQ_RAHEAD) 1817 rwbs[i++] = 'A'; 1818 if (rw & REQ_HARDBARRIER) 1819 rwbs[i++] = 'B'; 1820 if (rw & REQ_SYNC) 1821 rwbs[i++] = 'S'; 1822 if (rw & REQ_META) 1823 rwbs[i++] = 'M'; 1824 if (rw & REQ_SECURE) 1825 rwbs[i++] = 'E'; 1826 1827 rwbs[i] = '\0'; 1828 } 1829 1830 void blk_fill_rwbs_rq(char *rwbs, struct request *rq) 1831 { 1832 int rw = rq->cmd_flags & 0x03; 1833 int bytes; 1834 1835 if (rq->cmd_flags & REQ_DISCARD) 1836 rw |= REQ_DISCARD; 1837 1838 if (rq->cmd_flags & REQ_SECURE) 1839 rw |= REQ_SECURE; 1840 1841 bytes = blk_rq_bytes(rq); 1842 1843 blk_fill_rwbs(rwbs, rw, bytes); 1844 } 1845 1846 #endif /* CONFIG_EVENT_TRACING */ 1847 1848