kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/export.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/list.h>
  30
  31 #include <trace/events/block.h>
  32
  33 #include "trace_output.h"
  34
  35 #ifdef CONFIG_BLK_DEV_IO_TRACE
  36
  37 static unsigned int blktrace_seq __read_mostly = 1;
  38
  39 static struct trace_array *blk_tr;
  40 static bool blk_tracer_enabled __read_mostly;
  41
  42 static LIST_HEAD(running_trace_list);
  43 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  44
  45 /* Select an alternative, minimalistic output than the original one */
  46 #define TRACE_BLK_OPT_CLASSIC   0x1
  47
  48 static struct tracer_opt blk_tracer_opts[] = {
  49         /* Default disable the minimalistic output */
  50         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  51         { }
  52 };
  53
  54 static struct tracer_flags blk_tracer_flags = {
  55         .val  = 0,
  56         .opts = blk_tracer_opts,
  57 };
  58
  59 /* Global reference count of probes */
  60 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  61
  62 static void blk_register_tracepoints(void);
  63 static void blk_unregister_tracepoints(void);
  64
  65 /*
  66  * Send out a notify message.
  67  */
  68 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  69                        const void *data, size_t len)
  70 {
  71         struct blk_io_trace *t;
  72         struct ring_buffer_event *event = NULL;
  73         struct ring_buffer *buffer = NULL;
  74         int pc = 0;
  75         int cpu = smp_processor_id();
  76         bool blk_tracer = blk_tracer_enabled;
  77
  78         if (blk_tracer) {
  79                 buffer = blk_tr->trace_buffer.buffer;
  80                 pc = preempt_count();
  81                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  82                                                   sizeof(*t) + len,
  83                                                   0, pc);
  84                 if (!event)
  85                         return;
  86                 t = ring_buffer_event_data(event);
  87                 goto record_it;
  88         }
  89
  90         if (!bt->rchan)
  91                 return;
  92
  93         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  94         if (t) {
  95                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  96                 t->time = ktime_to_ns(ktime_get());
  97 record_it:
  98                 t->device = bt->dev;
  99                 t->action = action;
 100                 t->pid = pid;
 101                 t->cpu = cpu;
 102                 t->pdu_len = len;
 103                 memcpy((void *) t + sizeof(*t), data, len);
 104
 105                 if (blk_tracer)
 106                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 107         }
 108 }
 109
 110 /*
 111  * Send out a notify for this process, if we haven't done so since a trace
 112  * started
 113  */
 114 static void trace_note_tsk(struct task_struct *tsk)
 115 {
 116         unsigned long flags;
 117         struct blk_trace *bt;
 118
 119         tsk->btrace_seq = blktrace_seq;
 120         spin_lock_irqsave(&running_trace_lock, flags);
 121         list_for_each_entry(bt, &running_trace_list, running_list) {
 122                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 123                            sizeof(tsk->comm));
 124         }
 125         spin_unlock_irqrestore(&running_trace_lock, flags);
 126 }
 127
 128 static void trace_note_time(struct blk_trace *bt)
 129 {
 130         struct timespec now;
 131         unsigned long flags;
 132         u32 words[2];
 133
 134         getnstimeofday(&now);
 135         words[0] = now.tv_sec;
 136         words[1] = now.tv_nsec;
 137
 138         local_irq_save(flags);
 139         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 140         local_irq_restore(flags);
 141 }
 142
 143 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 144 {
 145         int n;
 146         va_list args;
 147         unsigned long flags;
 148         char *buf;
 149
 150         if (unlikely(bt->trace_state != Blktrace_running &&
 151                      !blk_tracer_enabled))
 152                 return;
 153
 154         /*
 155          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 156          * message to the trace.
 157          */
 158         if (!(bt->act_mask & BLK_TC_NOTIFY))
 159                 return;
 160
 161         local_irq_save(flags);
 162         buf = this_cpu_ptr(bt->msg_data);
 163         va_start(args, fmt);
 164         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 165         va_end(args);
 166
 167         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 168         local_irq_restore(flags);
 169 }
 170 EXPORT_SYMBOL_GPL(__trace_note_message);
 171
 172 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 173                          pid_t pid)
 174 {
 175         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 176                 return 1;
 177         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 178                 return 1;
 179         if (bt->pid && pid != bt->pid)
 180                 return 1;
 181
 182         return 0;
 183 }
 184
 185 /*
 186  * Data direction bit lookup
 187  */
 188 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 189                                  BLK_TC_ACT(BLK_TC_WRITE) };
 190
 191 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 192
 193 /* The ilog2() calls fall out because they're constant */
 194 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 195           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 196
 197 /*
 198  * The worker for the various blk_add_trace*() types. Fills out a
 199  * blk_io_trace structure and places it in a per-cpu subbuffer.
 200  */
 201 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 202                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 203 {
 204         struct task_struct *tsk = current;
 205         struct ring_buffer_event *event = NULL;
 206         struct ring_buffer *buffer = NULL;
 207         struct blk_io_trace *t;
 208         unsigned long flags = 0;
 209         unsigned long *sequence;
 210         pid_t pid;
 211         int cpu, pc = 0;
 212         bool blk_tracer = blk_tracer_enabled;
 213
 214         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 215                 return;
 216
 217         what |= ddir_act[rw & WRITE];
 218         what |= MASK_TC_BIT(rw, SYNC);
 219         what |= MASK_TC_BIT(rw, RAHEAD);
 220         what |= MASK_TC_BIT(rw, META);
 221         what |= MASK_TC_BIT(rw, DISCARD);
 222         what |= MASK_TC_BIT(rw, FLUSH);
 223         what |= MASK_TC_BIT(rw, FUA);
 224
 225         pid = tsk->pid;
 226         if (act_log_check(bt, what, sector, pid))
 227                 return;
 228         cpu = raw_smp_processor_id();
 229
 230         if (blk_tracer) {
 231                 tracing_record_cmdline(current);
 232
 233                 buffer = blk_tr->trace_buffer.buffer;
 234                 pc = preempt_count();
 235                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 236                                                   sizeof(*t) + pdu_len,
 237                                                   0, pc);
 238                 if (!event)
 239                         return;
 240                 t = ring_buffer_event_data(event);
 241                 goto record_it;
 242         }
 243
 244         if (unlikely(tsk->btrace_seq != blktrace_seq))
 245                 trace_note_tsk(tsk);
 246
 247         /*
 248          * A word about the locking here - we disable interrupts to reserve
 249          * some space in the relay per-cpu buffer, to prevent an irq
 250          * from coming in and stepping on our toes.
 251          */
 252         local_irq_save(flags);
 253         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 254         if (t) {
 255                 sequence = per_cpu_ptr(bt->sequence, cpu);
 256
 257                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 258                 t->sequence = ++(*sequence);
 259                 t->time = ktime_to_ns(ktime_get());
 260 record_it:
 261                 /*
 262                  * These two are not needed in ftrace as they are in the
 263                  * generic trace_entry, filled by tracing_generic_entry_update,
 264                  * but for the trace_event->bin() synthesizer benefit we do it
 265                  * here too.
 266                  */
 267                 t->cpu = cpu;
 268                 t->pid = pid;
 269
 270                 t->sector = sector;
 271                 t->bytes = bytes;
 272                 t->action = what;
 273                 t->device = bt->dev;
 274                 t->error = error;
 275                 t->pdu_len = pdu_len;
 276
 277                 if (pdu_len)
 278                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 279
 280                 if (blk_tracer) {
 281                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 282                         return;
 283                 }
 284         }
 285
 286         local_irq_restore(flags);
 287 }
 288
 289 static struct dentry *blk_tree_root;
 290 static DEFINE_MUTEX(blk_tree_mutex);
 291
 292 static void blk_trace_free(struct blk_trace *bt)
 293 {
 294         debugfs_remove(bt->msg_file);
 295         debugfs_remove(bt->dropped_file);
 296         relay_close(bt->rchan);
 297         debugfs_remove(bt->dir);
 298         free_percpu(bt->sequence);
 299         free_percpu(bt->msg_data);
 300         kfree(bt);
 301 }
 302
 303 static void blk_trace_cleanup(struct blk_trace *bt)
 304 {
 305         blk_trace_free(bt);
 306         if (atomic_dec_and_test(&blk_probes_ref))
 307                 blk_unregister_tracepoints();
 308 }
 309
 310 int blk_trace_remove(struct request_queue *q)
 311 {
 312         struct blk_trace *bt;
 313
 314         bt = xchg(&q->blk_trace, NULL);
 315         if (!bt)
 316                 return -EINVAL;
 317
 318         if (bt->trace_state != Blktrace_running)
 319                 blk_trace_cleanup(bt);
 320
 321         return 0;
 322 }
 323 EXPORT_SYMBOL_GPL(blk_trace_remove);
 324
 325 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 326                                 size_t count, loff_t *ppos)
 327 {
 328         struct blk_trace *bt = filp->private_data;
 329         char buf[16];
 330
 331         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 332
 333         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 334 }
 335
 336 static const struct file_operations blk_dropped_fops = {
 337         .owner =        THIS_MODULE,
 338         .open =         simple_open,
 339         .read =         blk_dropped_read,
 340         .llseek =       default_llseek,
 341 };
 342
 343 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 344                                 size_t count, loff_t *ppos)
 345 {
 346         char *msg;
 347         struct blk_trace *bt;
 348
 349         if (count >= BLK_TN_MAX_MSG)
 350                 return -EINVAL;
 351
 352         msg = kmalloc(count + 1, GFP_KERNEL);
 353         if (msg == NULL)
 354                 return -ENOMEM;
 355
 356         if (copy_from_user(msg, buffer, count)) {
 357                 kfree(msg);
 358                 return -EFAULT;
 359         }
 360
 361         msg[count] = '\0';
 362         bt = filp->private_data;
 363         __trace_note_message(bt, "%s", msg);
 364         kfree(msg);
 365
 366         return count;
 367 }
 368
 369 static const struct file_operations blk_msg_fops = {
 370         .owner =        THIS_MODULE,
 371         .open =         simple_open,
 372         .write =        blk_msg_write,
 373         .llseek =       noop_llseek,
 374 };
 375
 376 /*
 377  * Keep track of how many times we encountered a full subbuffer, to aid
 378  * the user space app in telling how many lost events there were.
 379  */
 380 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 381                                      void *prev_subbuf, size_t prev_padding)
 382 {
 383         struct blk_trace *bt;
 384
 385         if (!relay_buf_full(buf))
 386                 return 1;
 387
 388         bt = buf->chan->private_data;
 389         atomic_inc(&bt->dropped);
 390         return 0;
 391 }
 392
 393 static int blk_remove_buf_file_callback(struct dentry *dentry)
 394 {
 395         debugfs_remove(dentry);
 396
 397         return 0;
 398 }
 399
 400 static struct dentry *blk_create_buf_file_callback(const char *filename,
 401                                                    struct dentry *parent,
 402                                                    umode_t mode,
 403                                                    struct rchan_buf *buf,
 404                                                    int *is_global)
 405 {
 406         return debugfs_create_file(filename, mode, parent, buf,
 407                                         &relay_file_operations);
 408 }
 409
 410 static struct rchan_callbacks blk_relay_callbacks = {
 411         .subbuf_start           = blk_subbuf_start_callback,
 412         .create_buf_file        = blk_create_buf_file_callback,
 413         .remove_buf_file        = blk_remove_buf_file_callback,
 414 };
 415
 416 static void blk_trace_setup_lba(struct blk_trace *bt,
 417                                 struct block_device *bdev)
 418 {
 419         struct hd_struct *part = NULL;
 420
 421         if (bdev)
 422                 part = bdev->bd_part;
 423
 424         if (part) {
 425                 bt->start_lba = part->start_sect;
 426                 bt->end_lba = part->start_sect + part->nr_sects;
 427         } else {
 428                 bt->start_lba = 0;
 429                 bt->end_lba = -1ULL;
 430         }
 431 }
 432
 433 /*
 434  * Setup everything required to start tracing
 435  */
 436 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 437                        struct block_device *bdev,
 438                        struct blk_user_trace_setup *buts)
 439 {
 440         struct blk_trace *bt = NULL;
 441         struct dentry *dir = NULL;
 442         int ret;
 443
 444         if (!buts->buf_size || !buts->buf_nr)
 445                 return -EINVAL;
 446
 447         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 448         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 449
 450         /*
 451          * some device names have larger paths - convert the slashes
 452          * to underscores for this to work as expected
 453          */
 454         strreplace(buts->name, '/', '_');
 455
 456         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 457         if (!bt)
 458                 return -ENOMEM;
 459
 460         ret = -ENOMEM;
 461         bt->sequence = alloc_percpu(unsigned long);
 462         if (!bt->sequence)
 463                 goto err;
 464
 465         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 466         if (!bt->msg_data)
 467                 goto err;
 468
 469         ret = -ENOENT;
 470
 471         mutex_lock(&blk_tree_mutex);
 472         if (!blk_tree_root) {
 473                 blk_tree_root = debugfs_create_dir("block", NULL);
 474                 if (!blk_tree_root) {
 475                         mutex_unlock(&blk_tree_mutex);
 476                         goto err;
 477                 }
 478         }
 479         mutex_unlock(&blk_tree_mutex);
 480
 481         dir = debugfs_create_dir(buts->name, blk_tree_root);
 482
 483         if (!dir)
 484                 goto err;
 485
 486         bt->dir = dir;
 487         bt->dev = dev;
 488         atomic_set(&bt->dropped, 0);
 489         INIT_LIST_HEAD(&bt->running_list);
 490
 491         ret = -EIO;
 492         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 493                                                &blk_dropped_fops);
 494         if (!bt->dropped_file)
 495                 goto err;
 496
 497         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 498         if (!bt->msg_file)
 499                 goto err;
 500
 501         bt->rchan = relay_open("trace", dir, buts->buf_size,
 502                                 buts->buf_nr, &blk_relay_callbacks, bt);
 503         if (!bt->rchan)
 504                 goto err;
 505
 506         bt->act_mask = buts->act_mask;
 507         if (!bt->act_mask)
 508                 bt->act_mask = (u16) -1;
 509
 510         blk_trace_setup_lba(bt, bdev);
 511
 512         /* overwrite with user settings */
 513         if (buts->start_lba)
 514                 bt->start_lba = buts->start_lba;
 515         if (buts->end_lba)
 516                 bt->end_lba = buts->end_lba;
 517
 518         bt->pid = buts->pid;
 519         bt->trace_state = Blktrace_setup;
 520
 521         ret = -EBUSY;
 522         if (cmpxchg(&q->blk_trace, NULL, bt))
 523                 goto err;
 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
 578         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 579         if (ret)
 580                 return ret;
 581
 582         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 583                 blk_trace_remove(q);
 584                 return -EFAULT;
 585         }
 586
 587         return 0;
 588 }
 589 #endif
 590
 591 int blk_trace_startstop(struct request_queue *q, int start)
 592 {
 593         int ret;
 594         struct blk_trace *bt = q->blk_trace;
 595
 596         if (bt == NULL)
 597                 return -EINVAL;
 598
 599         /*
 600          * For starting a trace, we can transition from a setup or stopped
 601          * trace. For stopping a trace, the state must be running
 602          */
 603         ret = -EINVAL;
 604         if (start) {
 605                 if (bt->trace_state == Blktrace_setup ||
 606                     bt->trace_state == Blktrace_stopped) {
 607                         blktrace_seq++;
 608                         smp_mb();
 609                         bt->trace_state = Blktrace_running;
 610                         spin_lock_irq(&running_trace_lock);
 611                         list_add(&bt->running_list, &running_trace_list);
 612                         spin_unlock_irq(&running_trace_lock);
 613
 614                         trace_note_time(bt);
 615                         ret = 0;
 616                 }
 617         } else {
 618                 if (bt->trace_state == Blktrace_running) {
 619                         bt->trace_state = Blktrace_stopped;
 620                         spin_lock_irq(&running_trace_lock);
 621                         list_del_init(&bt->running_list);
 622                         spin_unlock_irq(&running_trace_lock);
 623                         relay_flush(bt->rchan);
 624                         ret = 0;
 625                 }
 626         }
 627
 628         return ret;
 629 }
 630 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 631
 632 /**
 633  * blk_trace_ioctl: - handle the ioctls associated with tracing
 634  * @bdev:       the block device
 635  * @cmd:        the ioctl cmd
 636  * @arg:        the argument data, if any
 637  *
 638  **/
 639 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 640 {
 641         struct request_queue *q;
 642         int ret, start = 0;
 643         char b[BDEVNAME_SIZE];
 644
 645         q = bdev_get_queue(bdev);
 646         if (!q)
 647                 return -ENXIO;
 648
 649         mutex_lock(&bdev->bd_mutex);
 650
 651         switch (cmd) {
 652         case BLKTRACESETUP:
 653                 bdevname(bdev, b);
 654                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 655                 break;
 656 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 657         case BLKTRACESETUP32:
 658                 bdevname(bdev, b);
 659                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 660                 break;
 661 #endif
 662         case BLKTRACESTART:
 663                 start = 1;
 664         case BLKTRACESTOP:
 665                 ret = blk_trace_startstop(q, start);
 666                 break;
 667         case BLKTRACETEARDOWN:
 668                 ret = blk_trace_remove(q);
 669                 break;
 670         default:
 671                 ret = -ENOTTY;
 672                 break;
 673         }
 674
 675         mutex_unlock(&bdev->bd_mutex);
 676         return ret;
 677 }
 678
 679 /**
 680  * blk_trace_shutdown: - stop and cleanup trace structures
 681  * @q:    the request queue associated with the device
 682  *
 683  **/
 684 void blk_trace_shutdown(struct request_queue *q)
 685 {
 686         if (q->blk_trace) {
 687                 blk_trace_startstop(q, 0);
 688                 blk_trace_remove(q);
 689         }
 690 }
 691
 692 /*
 693  * blktrace probes
 694  */
 695
 696 /**
 697  * blk_add_trace_rq - Add a trace for a request oriented action
 698  * @q:          queue the io is for
 699  * @rq:         the source request
 700  * @nr_bytes:   number of completed bytes
 701  * @what:       the action
 702  *
 703  * Description:
 704  *     Records an action against a request. Will log the bio offset + size.
 705  *
 706  **/
 707 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 708                              unsigned int nr_bytes, u32 what)
 709 {
 710         struct blk_trace *bt = q->blk_trace;
 711
 712         if (likely(!bt))
 713                 return;
 714
 715         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 716                 what |= BLK_TC_ACT(BLK_TC_PC);
 717                 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
 718                                 what, rq->errors, rq->cmd_len, rq->cmd);
 719         } else  {
 720                 what |= BLK_TC_ACT(BLK_TC_FS);
 721                 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
 722                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 723         }
 724 }
 725
 726 static void blk_add_trace_rq_abort(void *ignore,
 727                                    struct request_queue *q, struct request *rq)
 728 {
 729         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
 730 }
 731
 732 static void blk_add_trace_rq_insert(void *ignore,
 733                                     struct request_queue *q, struct request *rq)
 734 {
 735         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
 736 }
 737
 738 static void blk_add_trace_rq_issue(void *ignore,
 739                                    struct request_queue *q, struct request *rq)
 740 {
 741         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
 742 }
 743
 744 static void blk_add_trace_rq_requeue(void *ignore,
 745                                      struct request_queue *q,
 746                                      struct request *rq)
 747 {
 748         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 749 }
 750
 751 static void blk_add_trace_rq_complete(void *ignore,
 752                                       struct request_queue *q,
 753                                       struct request *rq,
 754                                       unsigned int nr_bytes)
 755 {
 756         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 757 }
 758
 759 /**
 760  * blk_add_trace_bio - Add a trace for a bio oriented action
 761  * @q:          queue the io is for
 762  * @bio:        the source bio
 763  * @what:       the action
 764  * @error:      error, if any
 765  *
 766  * Description:
 767  *     Records an action against a bio. Will log the bio offset + size.
 768  *
 769  **/
 770 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 771                               u32 what, int error)
 772 {
 773         struct blk_trace *bt = q->blk_trace;
 774
 775         if (likely(!bt))
 776                 return;
 777
 778         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 779                         bio->bi_rw, what, error, 0, NULL);
 780 }
 781
 782 static void blk_add_trace_bio_bounce(void *ignore,
 783                                      struct request_queue *q, struct bio *bio)
 784 {
 785         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 786 }
 787
 788 static void blk_add_trace_bio_complete(void *ignore,
 789                                        struct request_queue *q, struct bio *bio,
 790                                        int error)
 791 {
 792         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 793 }
 794
 795 static void blk_add_trace_bio_backmerge(void *ignore,
 796                                         struct request_queue *q,
 797                                         struct request *rq,
 798                                         struct bio *bio)
 799 {
 800         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 801 }
 802
 803 static void blk_add_trace_bio_frontmerge(void *ignore,
 804                                          struct request_queue *q,
 805                                          struct request *rq,
 806                                          struct bio *bio)
 807 {
 808         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 809 }
 810
 811 static void blk_add_trace_bio_queue(void *ignore,
 812                                     struct request_queue *q, struct bio *bio)
 813 {
 814         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 815 }
 816
 817 static void blk_add_trace_getrq(void *ignore,
 818                                 struct request_queue *q,
 819                                 struct bio *bio, int rw)
 820 {
 821         if (bio)
 822                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 823         else {
 824                 struct blk_trace *bt = q->blk_trace;
 825
 826                 if (bt)
 827                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 828         }
 829 }
 830
 831
 832 static void blk_add_trace_sleeprq(void *ignore,
 833                                   struct request_queue *q,
 834                                   struct bio *bio, int rw)
 835 {
 836         if (bio)
 837                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 838         else {
 839                 struct blk_trace *bt = q->blk_trace;
 840
 841                 if (bt)
 842                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 843                                         0, 0, NULL);
 844         }
 845 }
 846
 847 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 848 {
 849         struct blk_trace *bt = q->blk_trace;
 850
 851         if (bt)
 852                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 853 }
 854
 855 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 856                                     unsigned int depth, bool explicit)
 857 {
 858         struct blk_trace *bt = q->blk_trace;
 859
 860         if (bt) {
 861                 __be64 rpdu = cpu_to_be64(depth);
 862                 u32 what;
 863
 864                 if (explicit)
 865                         what = BLK_TA_UNPLUG_IO;
 866                 else
 867                         what = BLK_TA_UNPLUG_TIMER;
 868
 869                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 870         }
 871 }
 872
 873 static void blk_add_trace_split(void *ignore,
 874                                 struct request_queue *q, struct bio *bio,
 875                                 unsigned int pdu)
 876 {
 877         struct blk_trace *bt = q->blk_trace;
 878
 879         if (bt) {
 880                 __be64 rpdu = cpu_to_be64(pdu);
 881
 882                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 883                                 bio->bi_iter.bi_size, bio->bi_rw, BLK_TA_SPLIT,
 884                                 bio->bi_error, sizeof(rpdu), &rpdu);
 885         }
 886 }
 887
 888 /**
 889  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 890  * @ignore:     trace callback data parameter (not used)
 891  * @q:          queue the io is for
 892  * @bio:        the source bio
 893  * @dev:        target device
 894  * @from:       source sector
 895  *
 896  * Description:
 897  *     Device mapper or raid target sometimes need to split a bio because
 898  *     it spans a stripe (or similar). Add a trace for that action.
 899  *
 900  **/
 901 static void blk_add_trace_bio_remap(void *ignore,
 902                                     struct request_queue *q, struct bio *bio,
 903                                     dev_t dev, sector_t from)
 904 {
 905         struct blk_trace *bt = q->blk_trace;
 906         struct blk_io_trace_remap r;
 907
 908         if (likely(!bt))
 909                 return;
 910
 911         r.device_from = cpu_to_be32(dev);
 912         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 913         r.sector_from = cpu_to_be64(from);
 914
 915         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 916                         bio->bi_rw, BLK_TA_REMAP, bio->bi_error,
 917                         sizeof(r), &r);
 918 }
 919
 920 /**
 921  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 922  * @ignore:     trace callback data parameter (not used)
 923  * @q:          queue the io is for
 924  * @rq:         the source request
 925  * @dev:        target device
 926  * @from:       source sector
 927  *
 928  * Description:
 929  *     Device mapper remaps request to other devices.
 930  *     Add a trace for that action.
 931  *
 932  **/
 933 static void blk_add_trace_rq_remap(void *ignore,
 934                                    struct request_queue *q,
 935                                    struct request *rq, dev_t dev,
 936                                    sector_t from)
 937 {
 938         struct blk_trace *bt = q->blk_trace;
 939         struct blk_io_trace_remap r;
 940
 941         if (likely(!bt))
 942                 return;
 943
 944         r.device_from = cpu_to_be32(dev);
 945         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 946         r.sector_from = cpu_to_be64(from);
 947
 948         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 949                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 950                         sizeof(r), &r);
 951 }
 952
 953 /**
 954  * blk_add_driver_data - Add binary message with driver-specific data
 955  * @q:          queue the io is for
 956  * @rq:         io request
 957  * @data:       driver-specific data
 958  * @len:        length of driver-specific data
 959  *
 960  * Description:
 961  *     Some drivers might want to write driver-specific data per request.
 962  *
 963  **/
 964 void blk_add_driver_data(struct request_queue *q,
 965                          struct request *rq,
 966                          void *data, size_t len)
 967 {
 968         struct blk_trace *bt = q->blk_trace;
 969
 970         if (likely(!bt))
 971                 return;
 972
 973         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 974                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 975                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 976         else
 977                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 978                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 979 }
 980 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 981
 982 static void blk_register_tracepoints(void)
 983 {
 984         int ret;
 985
 986         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 987         WARN_ON(ret);
 988         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 989         WARN_ON(ret);
 990         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 991         WARN_ON(ret);
 992         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 993         WARN_ON(ret);
 994         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 995         WARN_ON(ret);
 996         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 997         WARN_ON(ret);
 998         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 999         WARN_ON(ret);
1000         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1001         WARN_ON(ret);
1002         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1003         WARN_ON(ret);
1004         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1005         WARN_ON(ret);
1006         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1007         WARN_ON(ret);
1008         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1009         WARN_ON(ret);
1010         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1011         WARN_ON(ret);
1012         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1013         WARN_ON(ret);
1014         ret = register_trace_block_split(blk_add_trace_split, NULL);
1015         WARN_ON(ret);
1016         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1017         WARN_ON(ret);
1018         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1019         WARN_ON(ret);
1020 }
1021
1022 static void blk_unregister_tracepoints(void)
1023 {
1024         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1025         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1026         unregister_trace_block_split(blk_add_trace_split, NULL);
1027         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1028         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1029         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1030         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1031         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1032         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1033         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1034         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1035         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1036         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1037         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1038         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1039         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1040         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1041
1042         tracepoint_synchronize_unregister();
1043 }
1044
1045 /*
1046  * struct blk_io_tracer formatting routines
1047  */
1048
1049 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1050 {
1051         int i = 0;
1052         int tc = t->action >> BLK_TC_SHIFT;
1053
1054         if (t->action == BLK_TN_MESSAGE) {
1055                 rwbs[i++] = 'N';
1056                 goto out;
1057         }
1058
1059         if (tc & BLK_TC_FLUSH)
1060                 rwbs[i++] = 'F';
1061
1062         if (tc & BLK_TC_DISCARD)
1063                 rwbs[i++] = 'D';
1064         else if (tc & BLK_TC_WRITE)
1065                 rwbs[i++] = 'W';
1066         else if (t->bytes)
1067                 rwbs[i++] = 'R';
1068         else
1069                 rwbs[i++] = 'N';
1070
1071         if (tc & BLK_TC_FUA)
1072                 rwbs[i++] = 'F';
1073         if (tc & BLK_TC_AHEAD)
1074                 rwbs[i++] = 'A';
1075         if (tc & BLK_TC_SYNC)
1076                 rwbs[i++] = 'S';
1077         if (tc & BLK_TC_META)
1078                 rwbs[i++] = 'M';
1079 out:
1080         rwbs[i] = '\0';
1081 }
1082
1083 static inline
1084 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1085 {
1086         return (const struct blk_io_trace *)ent;
1087 }
1088
1089 static inline const void *pdu_start(const struct trace_entry *ent)
1090 {
1091         return te_blk_io_trace(ent) + 1;
1092 }
1093
1094 static inline u32 t_action(const struct trace_entry *ent)
1095 {
1096         return te_blk_io_trace(ent)->action;
1097 }
1098
1099 static inline u32 t_bytes(const struct trace_entry *ent)
1100 {
1101         return te_blk_io_trace(ent)->bytes;
1102 }
1103
1104 static inline u32 t_sec(const struct trace_entry *ent)
1105 {
1106         return te_blk_io_trace(ent)->bytes >> 9;
1107 }
1108
1109 static inline unsigned long long t_sector(const struct trace_entry *ent)
1110 {
1111         return te_blk_io_trace(ent)->sector;
1112 }
1113
1114 static inline __u16 t_error(const struct trace_entry *ent)
1115 {
1116         return te_blk_io_trace(ent)->error;
1117 }
1118
1119 static __u64 get_pdu_int(const struct trace_entry *ent)
1120 {
1121         const __u64 *val = pdu_start(ent);
1122         return be64_to_cpu(*val);
1123 }
1124
1125 static void get_pdu_remap(const struct trace_entry *ent,
1126                           struct blk_io_trace_remap *r)
1127 {
1128         const struct blk_io_trace_remap *__r = pdu_start(ent);
1129         __u64 sector_from = __r->sector_from;
1130
1131         r->device_from = be32_to_cpu(__r->device_from);
1132         r->device_to   = be32_to_cpu(__r->device_to);
1133         r->sector_from = be64_to_cpu(sector_from);
1134 }
1135
1136 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1137
1138 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1139 {
1140         char rwbs[RWBS_LEN];
1141         unsigned long long ts  = iter->ts;
1142         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1143         unsigned secs          = (unsigned long)ts;
1144         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1145
1146         fill_rwbs(rwbs, t);
1147
1148         trace_seq_printf(&iter->seq,
1149                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1150                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1151                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1152 }
1153
1154 static void blk_log_action(struct trace_iterator *iter, const char *act)
1155 {
1156         char rwbs[RWBS_LEN];
1157         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1158
1159         fill_rwbs(rwbs, t);
1160         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1161                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1162 }
1163
1164 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1165 {
1166         const unsigned char *pdu_buf;
1167         int pdu_len;
1168         int i, end;
1169
1170         pdu_buf = pdu_start(ent);
1171         pdu_len = te_blk_io_trace(ent)->pdu_len;
1172
1173         if (!pdu_len)
1174                 return;
1175
1176         /* find the last zero that needs to be printed */
1177         for (end = pdu_len - 1; end >= 0; end--)
1178                 if (pdu_buf[end])
1179                         break;
1180         end++;
1181
1182         trace_seq_putc(s, '(');
1183
1184         for (i = 0; i < pdu_len; i++) {
1185
1186                 trace_seq_printf(s, "%s%02x",
1187                                  i == 0 ? "" : " ", pdu_buf[i]);
1188
1189                 /*
1190                  * stop when the rest is just zeroes and indicate so
1191                  * with a ".." appended
1192                  */
1193                 if (i == end && end != pdu_len - 1) {
1194                         trace_seq_puts(s, " ..) ");
1195                         return;
1196                 }
1197         }
1198
1199         trace_seq_puts(s, ") ");
1200 }
1201
1202 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1203 {
1204         char cmd[TASK_COMM_LEN];
1205
1206         trace_find_cmdline(ent->pid, cmd);
1207
1208         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1209                 trace_seq_printf(s, "%u ", t_bytes(ent));
1210                 blk_log_dump_pdu(s, ent);
1211                 trace_seq_printf(s, "[%s]\n", cmd);
1212         } else {
1213                 if (t_sec(ent))
1214                         trace_seq_printf(s, "%llu + %u [%s]\n",
1215                                                 t_sector(ent), t_sec(ent), cmd);
1216                 else
1217                         trace_seq_printf(s, "[%s]\n", cmd);
1218         }
1219 }
1220
1221 static void blk_log_with_error(struct trace_seq *s,
1222                               const struct trace_entry *ent)
1223 {
1224         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1225                 blk_log_dump_pdu(s, ent);
1226                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1227         } else {
1228                 if (t_sec(ent))
1229                         trace_seq_printf(s, "%llu + %u [%d]\n",
1230                                          t_sector(ent),
1231                                          t_sec(ent), t_error(ent));
1232                 else
1233                         trace_seq_printf(s, "%llu [%d]\n",
1234                                          t_sector(ent), t_error(ent));
1235         }
1236 }
1237
1238 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1239 {
1240         struct blk_io_trace_remap r = { .device_from = 0, };
1241
1242         get_pdu_remap(ent, &r);
1243         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1244                          t_sector(ent), t_sec(ent),
1245                          MAJOR(r.device_from), MINOR(r.device_from),
1246                          (unsigned long long)r.sector_from);
1247 }
1248
1249 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1250 {
1251         char cmd[TASK_COMM_LEN];
1252
1253         trace_find_cmdline(ent->pid, cmd);
1254
1255         trace_seq_printf(s, "[%s]\n", cmd);
1256 }
1257
1258 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1259 {
1260         char cmd[TASK_COMM_LEN];
1261
1262         trace_find_cmdline(ent->pid, cmd);
1263
1264         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1265 }
1266
1267 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1268 {
1269         char cmd[TASK_COMM_LEN];
1270
1271         trace_find_cmdline(ent->pid, cmd);
1272
1273         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1274                          get_pdu_int(ent), cmd);
1275 }
1276
1277 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1278 {
1279         const struct blk_io_trace *t = te_blk_io_trace(ent);
1280
1281         trace_seq_putmem(s, t + 1, t->pdu_len);
1282         trace_seq_putc(s, '\n');
1283 }
1284
1285 /*
1286  * struct tracer operations
1287  */
1288
1289 static void blk_tracer_print_header(struct seq_file *m)
1290 {
1291         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1292                 return;
1293         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1294                     "#  |     |     |           |   |   |\n");
1295 }
1296
1297 static void blk_tracer_start(struct trace_array *tr)
1298 {
1299         blk_tracer_enabled = true;
1300 }
1301
1302 static int blk_tracer_init(struct trace_array *tr)
1303 {
1304         blk_tr = tr;
1305         blk_tracer_start(tr);
1306         return 0;
1307 }
1308
1309 static void blk_tracer_stop(struct trace_array *tr)
1310 {
1311         blk_tracer_enabled = false;
1312 }
1313
1314 static void blk_tracer_reset(struct trace_array *tr)
1315 {
1316         blk_tracer_stop(tr);
1317 }
1318
1319 static const struct {
1320         const char *act[2];
1321         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1322 } what2act[] = {
1323         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1324         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1325         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1326         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1327         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1328         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1329         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1330         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1331         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1332         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1333         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1334         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1335         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1336         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1337         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1338 };
1339
1340 static enum print_line_t print_one_line(struct trace_iterator *iter,
1341                                         bool classic)
1342 {
1343         struct trace_seq *s = &iter->seq;
1344         const struct blk_io_trace *t;
1345         u16 what;
1346         bool long_act;
1347         blk_log_action_t *log_action;
1348
1349         t          = te_blk_io_trace(iter->ent);
1350         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1351         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1352         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1353
1354         if (t->action == BLK_TN_MESSAGE) {
1355                 log_action(iter, long_act ? "message" : "m");
1356                 blk_log_msg(s, iter->ent);
1357         }
1358
1359         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1360                 trace_seq_printf(s, "Unknown action %x\n", what);
1361         else {
1362                 log_action(iter, what2act[what].act[long_act]);
1363                 what2act[what].print(s, iter->ent);
1364         }
1365
1366         return trace_handle_return(s);
1367 }
1368
1369 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1370                                                int flags, struct trace_event *event)
1371 {
1372         return print_one_line(iter, false);
1373 }
1374
1375 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1376 {
1377         struct trace_seq *s = &iter->seq;
1378         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1379         const int offset = offsetof(struct blk_io_trace, sector);
1380         struct blk_io_trace old = {
1381                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1382                 .time     = iter->ts,
1383         };
1384
1385         trace_seq_putmem(s, &old, offset);
1386         trace_seq_putmem(s, &t->sector,
1387                          sizeof(old) - offset + t->pdu_len);
1388 }
1389
1390 static enum print_line_t
1391 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1392                              struct trace_event *event)
1393 {
1394         blk_trace_synthesize_old_trace(iter);
1395
1396         return trace_handle_return(&iter->seq);
1397 }
1398
1399 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1400 {
1401         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1402                 return TRACE_TYPE_UNHANDLED;
1403
1404         return print_one_line(iter, true);
1405 }
1406
1407 static int
1408 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1409 {
1410         /* don't output context-info for blk_classic output */
1411         if (bit == TRACE_BLK_OPT_CLASSIC) {
1412                 if (set)
1413                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1414                 else
1415                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1416         }
1417         return 0;
1418 }
1419
1420 static struct tracer blk_tracer __read_mostly = {
1421         .name           = "blk",
1422         .init           = blk_tracer_init,
1423         .reset          = blk_tracer_reset,
1424         .start          = blk_tracer_start,
1425         .stop           = blk_tracer_stop,
1426         .print_header   = blk_tracer_print_header,
1427         .print_line     = blk_tracer_print_line,
1428         .flags          = &blk_tracer_flags,
1429         .set_flag       = blk_tracer_set_flag,
1430 };
1431
1432 static struct trace_event_functions trace_blk_event_funcs = {
1433         .trace          = blk_trace_event_print,
1434         .binary         = blk_trace_event_print_binary,
1435 };
1436
1437 static struct trace_event trace_blk_event = {
1438         .type           = TRACE_BLK,
1439         .funcs          = &trace_blk_event_funcs,
1440 };
1441
1442 static int __init init_blk_tracer(void)
1443 {
1444         if (!register_trace_event(&trace_blk_event)) {
1445                 pr_warning("Warning: could not register block events\n");
1446                 return 1;
1447         }
1448
1449         if (register_tracer(&blk_tracer) != 0) {
1450                 pr_warning("Warning: could not register the block tracer\n");
1451                 unregister_trace_event(&trace_blk_event);
1452                 return 1;
1453         }
1454
1455         return 0;
1456 }
1457
1458 device_initcall(init_blk_tracer);
1459
1460 static int blk_trace_remove_queue(struct request_queue *q)
1461 {
1462         struct blk_trace *bt;
1463
1464         bt = xchg(&q->blk_trace, NULL);
1465         if (bt == NULL)
1466                 return -EINVAL;
1467
1468         if (atomic_dec_and_test(&blk_probes_ref))
1469                 blk_unregister_tracepoints();
1470
1471         blk_trace_free(bt);
1472         return 0;
1473 }
1474
1475 /*
1476  * Setup everything required to start tracing
1477  */
1478 static int blk_trace_setup_queue(struct request_queue *q,
1479                                  struct block_device *bdev)
1480 {
1481         struct blk_trace *bt = NULL;
1482         int ret = -ENOMEM;
1483
1484         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1485         if (!bt)
1486                 return -ENOMEM;
1487
1488         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1489         if (!bt->msg_data)
1490                 goto free_bt;
1491
1492         bt->dev = bdev->bd_dev;
1493         bt->act_mask = (u16)-1;
1494
1495         blk_trace_setup_lba(bt, bdev);
1496
1497         ret = -EBUSY;
1498         if (cmpxchg(&q->blk_trace, NULL, bt))
1499                 goto free_bt;
1500
1501         if (atomic_inc_return(&blk_probes_ref) == 1)
1502                 blk_register_tracepoints();
1503         return 0;
1504
1505 free_bt:
1506         blk_trace_free(bt);
1507         return ret;
1508 }
1509
1510 /*
1511  * sysfs interface to enable and configure tracing
1512  */
1513
1514 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1515                                          struct device_attribute *attr,
1516                                          char *buf);
1517 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1518                                           struct device_attribute *attr,
1519                                           const char *buf, size_t count);
1520 #define BLK_TRACE_DEVICE_ATTR(_name) \
1521         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1522                     sysfs_blk_trace_attr_show, \
1523                     sysfs_blk_trace_attr_store)
1524
1525 static BLK_TRACE_DEVICE_ATTR(enable);
1526 static BLK_TRACE_DEVICE_ATTR(act_mask);
1527 static BLK_TRACE_DEVICE_ATTR(pid);
1528 static BLK_TRACE_DEVICE_ATTR(start_lba);
1529 static BLK_TRACE_DEVICE_ATTR(end_lba);
1530
1531 static struct attribute *blk_trace_attrs[] = {
1532         &dev_attr_enable.attr,
1533         &dev_attr_act_mask.attr,
1534         &dev_attr_pid.attr,
1535         &dev_attr_start_lba.attr,
1536         &dev_attr_end_lba.attr,
1537         NULL
1538 };
1539
1540 struct attribute_group blk_trace_attr_group = {
1541         .name  = "trace",
1542         .attrs = blk_trace_attrs,
1543 };
1544
1545 static const struct {
1546         int mask;
1547         const char *str;
1548 } mask_maps[] = {
1549         { BLK_TC_READ,          "read"          },
1550         { BLK_TC_WRITE,         "write"         },
1551         { BLK_TC_FLUSH,         "flush"         },
1552         { BLK_TC_SYNC,          "sync"          },
1553         { BLK_TC_QUEUE,         "queue"         },
1554         { BLK_TC_REQUEUE,       "requeue"       },
1555         { BLK_TC_ISSUE,         "issue"         },
1556         { BLK_TC_COMPLETE,      "complete"      },
1557         { BLK_TC_FS,            "fs"            },
1558         { BLK_TC_PC,            "pc"            },
1559         { BLK_TC_AHEAD,         "ahead"         },
1560         { BLK_TC_META,          "meta"          },
1561         { BLK_TC_DISCARD,       "discard"       },
1562         { BLK_TC_DRV_DATA,      "drv_data"      },
1563         { BLK_TC_FUA,           "fua"           },
1564 };
1565
1566 static int blk_trace_str2mask(const char *str)
1567 {
1568         int i;
1569         int mask = 0;
1570         char *buf, *s, *token;
1571
1572         buf = kstrdup(str, GFP_KERNEL);
1573         if (buf == NULL)
1574                 return -ENOMEM;
1575         s = strstrip(buf);
1576
1577         while (1) {
1578                 token = strsep(&s, ",");
1579                 if (token == NULL)
1580                         break;
1581
1582                 if (*token == '\0')
1583                         continue;
1584
1585                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1586                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1587                                 mask |= mask_maps[i].mask;
1588                                 break;
1589                         }
1590                 }
1591                 if (i == ARRAY_SIZE(mask_maps)) {
1592                         mask = -EINVAL;
1593                         break;
1594                 }
1595         }
1596         kfree(buf);
1597
1598         return mask;
1599 }
1600
1601 static ssize_t blk_trace_mask2str(char *buf, int mask)
1602 {
1603         int i;
1604         char *p = buf;
1605
1606         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1607                 if (mask & mask_maps[i].mask) {
1608                         p += sprintf(p, "%s%s",
1609                                     (p == buf) ? "" : ",", mask_maps[i].str);
1610                 }
1611         }
1612         *p++ = '\n';
1613
1614         return p - buf;
1615 }
1616
1617 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1618 {
1619         if (bdev->bd_disk == NULL)
1620                 return NULL;
1621
1622         return bdev_get_queue(bdev);
1623 }
1624
1625 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1626                                          struct device_attribute *attr,
1627                                          char *buf)
1628 {
1629         struct hd_struct *p = dev_to_part(dev);
1630         struct request_queue *q;
1631         struct block_device *bdev;
1632         ssize_t ret = -ENXIO;
1633
1634         bdev = bdget(part_devt(p));
1635         if (bdev == NULL)
1636                 goto out;
1637
1638         q = blk_trace_get_queue(bdev);
1639         if (q == NULL)
1640                 goto out_bdput;
1641
1642         mutex_lock(&bdev->bd_mutex);
1643
1644         if (attr == &dev_attr_enable) {
1645                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1646                 goto out_unlock_bdev;
1647         }
1648
1649         if (q->blk_trace == NULL)
1650                 ret = sprintf(buf, "disabled\n");
1651         else if (attr == &dev_attr_act_mask)
1652                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1653         else if (attr == &dev_attr_pid)
1654                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1655         else if (attr == &dev_attr_start_lba)
1656                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1657         else if (attr == &dev_attr_end_lba)
1658                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1659
1660 out_unlock_bdev:
1661         mutex_unlock(&bdev->bd_mutex);
1662 out_bdput:
1663         bdput(bdev);
1664 out:
1665         return ret;
1666 }
1667
1668 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1669                                           struct device_attribute *attr,
1670                                           const char *buf, size_t count)
1671 {
1672         struct block_device *bdev;
1673         struct request_queue *q;
1674         struct hd_struct *p;
1675         u64 value;
1676         ssize_t ret = -EINVAL;
1677
1678         if (count == 0)
1679                 goto out;
1680
1681         if (attr == &dev_attr_act_mask) {
1682                 if (sscanf(buf, "%llx", &value) != 1) {
1683                         /* Assume it is a list of trace category names */
1684                         ret = blk_trace_str2mask(buf);
1685                         if (ret < 0)
1686                                 goto out;
1687                         value = ret;
1688                 }
1689         } else if (sscanf(buf, "%llu", &value) != 1)
1690                 goto out;
1691
1692         ret = -ENXIO;
1693
1694         p = dev_to_part(dev);
1695         bdev = bdget(part_devt(p));
1696         if (bdev == NULL)
1697                 goto out;
1698
1699         q = blk_trace_get_queue(bdev);
1700         if (q == NULL)
1701                 goto out_bdput;
1702
1703         mutex_lock(&bdev->bd_mutex);
1704
1705         if (attr == &dev_attr_enable) {
1706                 if (value)
1707                         ret = blk_trace_setup_queue(q, bdev);
1708                 else
1709                         ret = blk_trace_remove_queue(q);
1710                 goto out_unlock_bdev;
1711         }
1712
1713         ret = 0;
1714         if (q->blk_trace == NULL)
1715                 ret = blk_trace_setup_queue(q, bdev);
1716
1717         if (ret == 0) {
1718                 if (attr == &dev_attr_act_mask)
1719                         q->blk_trace->act_mask = value;
1720                 else if (attr == &dev_attr_pid)
1721                         q->blk_trace->pid = value;
1722                 else if (attr == &dev_attr_start_lba)
1723                         q->blk_trace->start_lba = value;
1724                 else if (attr == &dev_attr_end_lba)
1725                         q->blk_trace->end_lba = value;
1726         }
1727
1728 out_unlock_bdev:
1729         mutex_unlock(&bdev->bd_mutex);
1730 out_bdput:
1731         bdput(bdev);
1732 out:
1733         return ret ? ret : count;
1734 }
1735
1736 int blk_trace_init_sysfs(struct device *dev)
1737 {
1738         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1739 }
1740
1741 void blk_trace_remove_sysfs(struct device *dev)
1742 {
1743         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1744 }
1745
1746 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1747
1748 #ifdef CONFIG_EVENT_TRACING
1749
1750 void blk_dump_cmd(char *buf, struct request *rq)
1751 {
1752         int i, end;
1753         int len = rq->cmd_len;
1754         unsigned char *cmd = rq->cmd;
1755
1756         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1757                 buf[0] = '\0';
1758                 return;
1759         }
1760
1761         for (end = len - 1; end >= 0; end--)
1762                 if (cmd[end])
1763                         break;
1764         end++;
1765
1766         for (i = 0; i < len; i++) {
1767                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1768                 if (i == end && end != len - 1) {
1769                         sprintf(buf, " ..");
1770                         break;
1771                 }
1772         }
1773 }
1774
1775 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1776 {
1777         int i = 0;
1778
1779         if (rw & REQ_FLUSH)
1780                 rwbs[i++] = 'F';
1781
1782         if (rw & WRITE)
1783                 rwbs[i++] = 'W';
1784         else if (rw & REQ_DISCARD)
1785                 rwbs[i++] = 'D';
1786         else if (bytes)
1787                 rwbs[i++] = 'R';
1788         else
1789                 rwbs[i++] = 'N';
1790
1791         if (rw & REQ_FUA)
1792                 rwbs[i++] = 'F';
1793         if (rw & REQ_RAHEAD)
1794                 rwbs[i++] = 'A';
1795         if (rw & REQ_SYNC)
1796                 rwbs[i++] = 'S';
1797         if (rw & REQ_META)
1798                 rwbs[i++] = 'M';
1799         if (rw & REQ_SECURE)
1800                 rwbs[i++] = 'E';
1801
1802         rwbs[i] = '\0';
1803 }
1804 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1805
1806 #endif /* CONFIG_EVENT_TRACING */
1807