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/time.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <trace/events/block.h>
  30
  31 #include "trace_output.h"
  32
  33 #ifdef CONFIG_BLK_DEV_IO_TRACE
  34
  35 static unsigned int blktrace_seq __read_mostly = 1;
  36
  37 static struct trace_array *blk_tr;
  38 static bool blk_tracer_enabled __read_mostly;
  39
  40 /* Select an alternative, minimalistic output than the original one */
  41 #define TRACE_BLK_OPT_CLASSIC   0x1
  42
  43 static struct tracer_opt blk_tracer_opts[] = {
  44         /* Default disable the minimalistic output */
  45         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  46         { }
  47 };
  48
  49 static struct tracer_flags blk_tracer_flags = {
  50         .val  = 0,
  51         .opts = blk_tracer_opts,
  52 };
  53
  54 /* Global reference count of probes */
  55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  56
  57 static void blk_register_tracepoints(void);
  58 static void blk_unregister_tracepoints(void);
  59
  60 /*
  61  * Send out a notify message.
  62  */
  63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  64                        const void *data, size_t len)
  65 {
  66         struct blk_io_trace *t;
  67         struct ring_buffer_event *event = NULL;
  68         struct ring_buffer *buffer = NULL;
  69         int pc = 0;
  70         int cpu = smp_processor_id();
  71         bool blk_tracer = blk_tracer_enabled;
  72
  73         if (blk_tracer) {
  74                 buffer = blk_tr->buffer;
  75                 pc = preempt_count();
  76                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  77                                                   sizeof(*t) + len,
  78                                                   0, pc);
  79                 if (!event)
  80                         return;
  81                 t = ring_buffer_event_data(event);
  82                 goto record_it;
  83         }
  84
  85         if (!bt->rchan)
  86                 return;
  87
  88         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  89         if (t) {
  90                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  91                 t->time = ktime_to_ns(ktime_get());
  92 record_it:
  93                 t->device = bt->dev;
  94                 t->action = action;
  95                 t->pid = pid;
  96                 t->cpu = cpu;
  97                 t->pdu_len = len;
  98                 memcpy((void *) t + sizeof(*t), data, len);
  99
 100                 if (blk_tracer)
 101                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 102         }
 103 }
 104
 105 /*
 106  * Send out a notify for this process, if we haven't done so since a trace
 107  * started
 108  */
 109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 110 {
 111         tsk->btrace_seq = blktrace_seq;
 112         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 113 }
 114
 115 static void trace_note_time(struct blk_trace *bt)
 116 {
 117         struct timespec now;
 118         unsigned long flags;
 119         u32 words[2];
 120
 121         getnstimeofday(&now);
 122         words[0] = now.tv_sec;
 123         words[1] = now.tv_nsec;
 124
 125         local_irq_save(flags);
 126         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 127         local_irq_restore(flags);
 128 }
 129
 130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 131 {
 132         int n;
 133         va_list args;
 134         unsigned long flags;
 135         char *buf;
 136
 137         if (unlikely(bt->trace_state != Blktrace_running &&
 138                      !blk_tracer_enabled))
 139                 return;
 140
 141         /*
 142          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 143          * message to the trace.
 144          */
 145         if (!(bt->act_mask & BLK_TC_NOTIFY))
 146                 return;
 147
 148         local_irq_save(flags);
 149         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 150         va_start(args, fmt);
 151         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 152         va_end(args);
 153
 154         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 155         local_irq_restore(flags);
 156 }
 157 EXPORT_SYMBOL_GPL(__trace_note_message);
 158
 159 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 160                          pid_t pid)
 161 {
 162         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 163                 return 1;
 164         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 165                 return 1;
 166         if (bt->pid && pid != bt->pid)
 167                 return 1;
 168
 169         return 0;
 170 }
 171
 172 /*
 173  * Data direction bit lookup
 174  */
 175 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 176                                  BLK_TC_ACT(BLK_TC_WRITE) };
 177
 178 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 179
 180 /* The ilog2() calls fall out because they're constant */
 181 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 182           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 183
 184 /*
 185  * The worker for the various blk_add_trace*() types. Fills out a
 186  * blk_io_trace structure and places it in a per-cpu subbuffer.
 187  */
 188 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 189                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 190 {
 191         struct task_struct *tsk = current;
 192         struct ring_buffer_event *event = NULL;
 193         struct ring_buffer *buffer = NULL;
 194         struct blk_io_trace *t;
 195         unsigned long flags = 0;
 196         unsigned long *sequence;
 197         pid_t pid;
 198         int cpu, pc = 0;
 199         bool blk_tracer = blk_tracer_enabled;
 200
 201         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 202                 return;
 203
 204         what |= ddir_act[rw & WRITE];
 205         what |= MASK_TC_BIT(rw, SYNC);
 206         what |= MASK_TC_BIT(rw, RAHEAD);
 207         what |= MASK_TC_BIT(rw, META);
 208         what |= MASK_TC_BIT(rw, DISCARD);
 209         what |= MASK_TC_BIT(rw, FLUSH);
 210         what |= MASK_TC_BIT(rw, FUA);
 211
 212         pid = tsk->pid;
 213         if (act_log_check(bt, what, sector, pid))
 214                 return;
 215         cpu = raw_smp_processor_id();
 216
 217         if (blk_tracer) {
 218                 tracing_record_cmdline(current);
 219
 220                 buffer = blk_tr->buffer;
 221                 pc = preempt_count();
 222                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 223                                                   sizeof(*t) + pdu_len,
 224                                                   0, pc);
 225                 if (!event)
 226                         return;
 227                 t = ring_buffer_event_data(event);
 228                 goto record_it;
 229         }
 230
 231         /*
 232          * A word about the locking here - we disable interrupts to reserve
 233          * some space in the relay per-cpu buffer, to prevent an irq
 234          * from coming in and stepping on our toes.
 235          */
 236         local_irq_save(flags);
 237
 238         if (unlikely(tsk->btrace_seq != blktrace_seq))
 239                 trace_note_tsk(bt, tsk);
 240
 241         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 242         if (t) {
 243                 sequence = per_cpu_ptr(bt->sequence, cpu);
 244
 245                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 246                 t->sequence = ++(*sequence);
 247                 t->time = ktime_to_ns(ktime_get());
 248 record_it:
 249                 /*
 250                  * These two are not needed in ftrace as they are in the
 251                  * generic trace_entry, filled by tracing_generic_entry_update,
 252                  * but for the trace_event->bin() synthesizer benefit we do it
 253                  * here too.
 254                  */
 255                 t->cpu = cpu;
 256                 t->pid = pid;
 257
 258                 t->sector = sector;
 259                 t->bytes = bytes;
 260                 t->action = what;
 261                 t->device = bt->dev;
 262                 t->error = error;
 263                 t->pdu_len = pdu_len;
 264
 265                 if (pdu_len)
 266                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 267
 268                 if (blk_tracer) {
 269                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 270                         return;
 271                 }
 272         }
 273
 274         local_irq_restore(flags);
 275 }
 276
 277 static struct dentry *blk_tree_root;
 278 static DEFINE_MUTEX(blk_tree_mutex);
 279
 280 static void blk_trace_free(struct blk_trace *bt)
 281 {
 282         debugfs_remove(bt->msg_file);
 283         debugfs_remove(bt->dropped_file);
 284         relay_close(bt->rchan);
 285         debugfs_remove(bt->dir);
 286         free_percpu(bt->sequence);
 287         free_percpu(bt->msg_data);
 288         kfree(bt);
 289 }
 290
 291 static void blk_trace_cleanup(struct blk_trace *bt)
 292 {
 293         blk_trace_free(bt);
 294         if (atomic_dec_and_test(&blk_probes_ref))
 295                 blk_unregister_tracepoints();
 296 }
 297
 298 int blk_trace_remove(struct request_queue *q)
 299 {
 300         struct blk_trace *bt;
 301
 302         bt = xchg(&q->blk_trace, NULL);
 303         if (!bt)
 304                 return -EINVAL;
 305
 306         if (bt->trace_state != Blktrace_running)
 307                 blk_trace_cleanup(bt);
 308
 309         return 0;
 310 }
 311 EXPORT_SYMBOL_GPL(blk_trace_remove);
 312
 313 static int blk_dropped_open(struct inode *inode, struct file *filp)
 314 {
 315         filp->private_data = inode->i_private;
 316
 317         return 0;
 318 }
 319
 320 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 321                                 size_t count, loff_t *ppos)
 322 {
 323         struct blk_trace *bt = filp->private_data;
 324         char buf[16];
 325
 326         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 327
 328         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 329 }
 330
 331 static const struct file_operations blk_dropped_fops = {
 332         .owner =        THIS_MODULE,
 333         .open =         blk_dropped_open,
 334         .read =         blk_dropped_read,
 335         .llseek =       default_llseek,
 336 };
 337
 338 static int blk_msg_open(struct inode *inode, struct file *filp)
 339 {
 340         filp->private_data = inode->i_private;
 341
 342         return 0;
 343 }
 344
 345 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 346                                 size_t count, loff_t *ppos)
 347 {
 348         char *msg;
 349         struct blk_trace *bt;
 350
 351         if (count >= BLK_TN_MAX_MSG)
 352                 return -EINVAL;
 353
 354         msg = kmalloc(count + 1, GFP_KERNEL);
 355         if (msg == NULL)
 356                 return -ENOMEM;
 357
 358         if (copy_from_user(msg, buffer, count)) {
 359                 kfree(msg);
 360                 return -EFAULT;
 361         }
 362
 363         msg[count] = '\0';
 364         bt = filp->private_data;
 365         __trace_note_message(bt, "%s", msg);
 366         kfree(msg);
 367
 368         return count;
 369 }
 370
 371 static const struct file_operations blk_msg_fops = {
 372         .owner =        THIS_MODULE,
 373         .open =         blk_msg_open,
 374         .write =        blk_msg_write,
 375         .llseek =       noop_llseek,
 376 };
 377
 378 /*
 379  * Keep track of how many times we encountered a full subbuffer, to aid
 380  * the user space app in telling how many lost events there were.
 381  */
 382 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 383                                      void *prev_subbuf, size_t prev_padding)
 384 {
 385         struct blk_trace *bt;
 386
 387         if (!relay_buf_full(buf))
 388                 return 1;
 389
 390         bt = buf->chan->private_data;
 391         atomic_inc(&bt->dropped);
 392         return 0;
 393 }
 394
 395 static int blk_remove_buf_file_callback(struct dentry *dentry)
 396 {
 397         debugfs_remove(dentry);
 398
 399         return 0;
 400 }
 401
 402 static struct dentry *blk_create_buf_file_callback(const char *filename,
 403                                                    struct dentry *parent,
 404                                                    int mode,
 405                                                    struct rchan_buf *buf,
 406                                                    int *is_global)
 407 {
 408         return debugfs_create_file(filename, mode, parent, buf,
 409                                         &relay_file_operations);
 410 }
 411
 412 static struct rchan_callbacks blk_relay_callbacks = {
 413         .subbuf_start           = blk_subbuf_start_callback,
 414         .create_buf_file        = blk_create_buf_file_callback,
 415         .remove_buf_file        = blk_remove_buf_file_callback,
 416 };
 417
 418 static void blk_trace_setup_lba(struct blk_trace *bt,
 419                                 struct block_device *bdev)
 420 {
 421         struct hd_struct *part = NULL;
 422
 423         if (bdev)
 424                 part = bdev->bd_part;
 425
 426         if (part) {
 427                 bt->start_lba = part->start_sect;
 428                 bt->end_lba = part->start_sect + part->nr_sects;
 429         } else {
 430                 bt->start_lba = 0;
 431                 bt->end_lba = -1ULL;
 432         }
 433 }
 434
 435 /*
 436  * Setup everything required to start tracing
 437  */
 438 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 439                        struct block_device *bdev,
 440                        struct blk_user_trace_setup *buts)
 441 {
 442         struct blk_trace *old_bt, *bt = NULL;
 443         struct dentry *dir = NULL;
 444         int ret, i;
 445
 446         if (!buts->buf_size || !buts->buf_nr)
 447                 return -EINVAL;
 448
 449         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 450         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 451
 452         /*
 453          * some device names have larger paths - convert the slashes
 454          * to underscores for this to work as expected
 455          */
 456         for (i = 0; i < strlen(buts->name); i++)
 457                 if (buts->name[i] == '/')
 458                         buts->name[i] = '_';
 459
 460         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 461         if (!bt)
 462                 return -ENOMEM;
 463
 464         ret = -ENOMEM;
 465         bt->sequence = alloc_percpu(unsigned long);
 466         if (!bt->sequence)
 467                 goto err;
 468
 469         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 470         if (!bt->msg_data)
 471                 goto err;
 472
 473         ret = -ENOENT;
 474
 475         mutex_lock(&blk_tree_mutex);
 476         if (!blk_tree_root) {
 477                 blk_tree_root = debugfs_create_dir("block", NULL);
 478                 if (!blk_tree_root) {
 479                         mutex_unlock(&blk_tree_mutex);
 480                         goto err;
 481                 }
 482         }
 483         mutex_unlock(&blk_tree_mutex);
 484
 485         dir = debugfs_create_dir(buts->name, blk_tree_root);
 486
 487         if (!dir)
 488                 goto err;
 489
 490         bt->dir = dir;
 491         bt->dev = dev;
 492         atomic_set(&bt->dropped, 0);
 493
 494         ret = -EIO;
 495         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 496                                                &blk_dropped_fops);
 497         if (!bt->dropped_file)
 498                 goto err;
 499
 500         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 501         if (!bt->msg_file)
 502                 goto err;
 503
 504         bt->rchan = relay_open("trace", dir, buts->buf_size,
 505                                 buts->buf_nr, &blk_relay_callbacks, bt);
 506         if (!bt->rchan)
 507                 goto err;
 508
 509         bt->act_mask = buts->act_mask;
 510         if (!bt->act_mask)
 511                 bt->act_mask = (u16) -1;
 512
 513         blk_trace_setup_lba(bt, bdev);
 514
 515         /* overwrite with user settings */
 516         if (buts->start_lba)
 517                 bt->start_lba = buts->start_lba;
 518         if (buts->end_lba)
 519                 bt->end_lba = buts->end_lba;
 520
 521         bt->pid = buts->pid;
 522         bt->trace_state = Blktrace_setup;
 523
 524         ret = -EBUSY;
 525         old_bt = xchg(&q->blk_trace, bt);
 526         if (old_bt) {
 527                 (void) xchg(&q->blk_trace, old_bt);
 528                 goto err;
 529         }
 530
 531         if (atomic_inc_return(&blk_probes_ref) == 1)
 532                 blk_register_tracepoints();
 533
 534         return 0;
 535 err:
 536         blk_trace_free(bt);
 537         return ret;
 538 }
 539
 540 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 541                     struct block_device *bdev,
 542                     char __user *arg)
 543 {
 544         struct blk_user_trace_setup buts;
 545         int ret;
 546
 547         ret = copy_from_user(&buts, arg, sizeof(buts));
 548         if (ret)
 549                 return -EFAULT;
 550
 551         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 552         if (ret)
 553                 return ret;
 554
 555         if (copy_to_user(arg, &buts, sizeof(buts))) {
 556                 blk_trace_remove(q);
 557                 return -EFAULT;
 558         }
 559         return 0;
 560 }
 561 EXPORT_SYMBOL_GPL(blk_trace_setup);
 562
 563 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 564 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 565                                   dev_t dev, struct block_device *bdev,
 566                                   char __user *arg)
 567 {
 568         struct blk_user_trace_setup buts;
 569         struct compat_blk_user_trace_setup cbuts;
 570         int ret;
 571
 572         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 573                 return -EFAULT;
 574
 575         buts = (struct blk_user_trace_setup) {
 576                 .act_mask = cbuts.act_mask,
 577                 .buf_size = cbuts.buf_size,
 578                 .buf_nr = cbuts.buf_nr,
 579                 .start_lba = cbuts.start_lba,
 580                 .end_lba = cbuts.end_lba,
 581                 .pid = cbuts.pid,
 582         };
 583         memcpy(&buts.name, &cbuts.name, 32);
 584
 585         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 586         if (ret)
 587                 return ret;
 588
 589         if (copy_to_user(arg, &buts.name, 32)) {
 590                 blk_trace_remove(q);
 591                 return -EFAULT;
 592         }
 593
 594         return 0;
 595 }
 596 #endif
 597
 598 int blk_trace_startstop(struct request_queue *q, int start)
 599 {
 600         int ret;
 601         struct blk_trace *bt = q->blk_trace;
 602
 603         if (bt == NULL)
 604                 return -EINVAL;
 605
 606         /*
 607          * For starting a trace, we can transition from a setup or stopped
 608          * trace. For stopping a trace, the state must be running
 609          */
 610         ret = -EINVAL;
 611         if (start) {
 612                 if (bt->trace_state == Blktrace_setup ||
 613                     bt->trace_state == Blktrace_stopped) {
 614                         blktrace_seq++;
 615                         smp_mb();
 616                         bt->trace_state = Blktrace_running;
 617
 618                         trace_note_time(bt);
 619                         ret = 0;
 620                 }
 621         } else {
 622                 if (bt->trace_state == Blktrace_running) {
 623                         bt->trace_state = Blktrace_stopped;
 624                         relay_flush(bt->rchan);
 625                         ret = 0;
 626                 }
 627         }
 628
 629         return ret;
 630 }
 631 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 632
 633 /**
 634  * blk_trace_ioctl: - handle the ioctls associated with tracing
 635  * @bdev:       the block device
 636  * @cmd:        the ioctl cmd
 637  * @arg:        the argument data, if any
 638  *
 639  **/
 640 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 641 {
 642         struct request_queue *q;
 643         int ret, start = 0;
 644         char b[BDEVNAME_SIZE];
 645
 646         q = bdev_get_queue(bdev);
 647         if (!q)
 648                 return -ENXIO;
 649
 650         mutex_lock(&bdev->bd_mutex);
 651
 652         switch (cmd) {
 653         case BLKTRACESETUP:
 654                 bdevname(bdev, b);
 655                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 656                 break;
 657 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 658         case BLKTRACESETUP32:
 659                 bdevname(bdev, b);
 660                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 661                 break;
 662 #endif
 663         case BLKTRACESTART:
 664                 start = 1;
 665         case BLKTRACESTOP:
 666                 ret = blk_trace_startstop(q, start);
 667                 break;
 668         case BLKTRACETEARDOWN:
 669                 ret = blk_trace_remove(q);
 670                 break;
 671         default:
 672                 ret = -ENOTTY;
 673                 break;
 674         }
 675
 676         mutex_unlock(&bdev->bd_mutex);
 677         return ret;
 678 }
 679
 680 /**
 681  * blk_trace_shutdown: - stop and cleanup trace structures
 682  * @q:    the request queue associated with the device
 683  *
 684  **/
 685 void blk_trace_shutdown(struct request_queue *q)
 686 {
 687         if (q->blk_trace) {
 688                 blk_trace_startstop(q, 0);
 689                 blk_trace_remove(q);
 690         }
 691 }
 692
 693 /*
 694  * blktrace probes
 695  */
 696
 697 /**
 698  * blk_add_trace_rq - Add a trace for a request oriented action
 699  * @q:          queue the io is for
 700  * @rq:         the source request
 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                              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, blk_rq_bytes(rq), 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), blk_rq_bytes(rq),
 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_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_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_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_TA_REQUEUE);
 749 }
 750
 751 static void blk_add_trace_rq_complete(void *ignore,
 752                                       struct request_queue *q,
 753                                       struct request *rq)
 754 {
 755         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 756 }
 757
 758 /**
 759  * blk_add_trace_bio - Add a trace for a bio oriented action
 760  * @q:          queue the io is for
 761  * @bio:        the source bio
 762  * @what:       the action
 763  * @error:      error, if any
 764  *
 765  * Description:
 766  *     Records an action against a bio. Will log the bio offset + size.
 767  *
 768  **/
 769 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 770                               u32 what, int error)
 771 {
 772         struct blk_trace *bt = q->blk_trace;
 773
 774         if (likely(!bt))
 775                 return;
 776
 777         if (!error && !bio_flagged(bio, BIO_UPTODATE))
 778                 error = EIO;
 779
 780         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 781                         error, 0, NULL);
 782 }
 783
 784 static void blk_add_trace_bio_bounce(void *ignore,
 785                                      struct request_queue *q, struct bio *bio)
 786 {
 787         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 788 }
 789
 790 static void blk_add_trace_bio_complete(void *ignore,
 791                                        struct request_queue *q, struct bio *bio,
 792                                        int error)
 793 {
 794         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 795 }
 796
 797 static void blk_add_trace_bio_backmerge(void *ignore,
 798                                         struct request_queue *q,
 799                                         struct bio *bio)
 800 {
 801         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 802 }
 803
 804 static void blk_add_trace_bio_frontmerge(void *ignore,
 805                                          struct request_queue *q,
 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_sector, bio->bi_size, bio->bi_rw,
 883                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 884                                 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_sector, bio->bi_size, bio->bi_rw,
 916                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 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 int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1137
1138 static int 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         return 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 int 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         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1161                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1162 }
1163
1164 static int 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, ret;
1169
1170         pdu_buf = pdu_start(ent);
1171         pdu_len = te_blk_io_trace(ent)->pdu_len;
1172
1173         if (!pdu_len)
1174                 return 1;
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         if (!trace_seq_putc(s, '('))
1183                 return 0;
1184
1185         for (i = 0; i < pdu_len; i++) {
1186
1187                 ret = trace_seq_printf(s, "%s%02x",
1188                                        i == 0 ? "" : " ", pdu_buf[i]);
1189                 if (!ret)
1190                         return ret;
1191
1192                 /*
1193                  * stop when the rest is just zeroes and indicate so
1194                  * with a ".." appended
1195                  */
1196                 if (i == end && end != pdu_len - 1)
1197                         return trace_seq_puts(s, " ..) ");
1198         }
1199
1200         return trace_seq_puts(s, ") ");
1201 }
1202
1203 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1204 {
1205         char cmd[TASK_COMM_LEN];
1206
1207         trace_find_cmdline(ent->pid, cmd);
1208
1209         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1210                 int ret;
1211
1212                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1213                 if (!ret)
1214                         return 0;
1215                 ret = blk_log_dump_pdu(s, ent);
1216                 if (!ret)
1217                         return 0;
1218                 return trace_seq_printf(s, "[%s]\n", cmd);
1219         } else {
1220                 if (t_sec(ent))
1221                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1222                                                 t_sector(ent), t_sec(ent), cmd);
1223                 return trace_seq_printf(s, "[%s]\n", cmd);
1224         }
1225 }
1226
1227 static int blk_log_with_error(struct trace_seq *s,
1228                               const struct trace_entry *ent)
1229 {
1230         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1231                 int ret;
1232
1233                 ret = blk_log_dump_pdu(s, ent);
1234                 if (ret)
1235                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1236                 return 0;
1237         } else {
1238                 if (t_sec(ent))
1239                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1240                                                 t_sector(ent),
1241                                                 t_sec(ent), t_error(ent));
1242                 return trace_seq_printf(s, "%llu [%d]\n",
1243                                         t_sector(ent), t_error(ent));
1244         }
1245 }
1246
1247 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1248 {
1249         struct blk_io_trace_remap r = { .device_from = 0, };
1250
1251         get_pdu_remap(ent, &r);
1252         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1253                                 t_sector(ent), t_sec(ent),
1254                                 MAJOR(r.device_from), MINOR(r.device_from),
1255                                 (unsigned long long)r.sector_from);
1256 }
1257
1258 static int blk_log_plug(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         return trace_seq_printf(s, "[%s]\n", cmd);
1265 }
1266
1267 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1268 {
1269         char cmd[TASK_COMM_LEN];
1270
1271         trace_find_cmdline(ent->pid, cmd);
1272
1273         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1274 }
1275
1276 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1277 {
1278         char cmd[TASK_COMM_LEN];
1279
1280         trace_find_cmdline(ent->pid, cmd);
1281
1282         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1283                                 get_pdu_int(ent), cmd);
1284 }
1285
1286 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1287 {
1288         int ret;
1289         const struct blk_io_trace *t = te_blk_io_trace(ent);
1290
1291         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1292         if (ret)
1293                 return trace_seq_putc(s, '\n');
1294         return ret;
1295 }
1296
1297 /*
1298  * struct tracer operations
1299  */
1300
1301 static void blk_tracer_print_header(struct seq_file *m)
1302 {
1303         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1304                 return;
1305         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1306                     "#  |     |     |           |   |   |\n");
1307 }
1308
1309 static void blk_tracer_start(struct trace_array *tr)
1310 {
1311         blk_tracer_enabled = true;
1312 }
1313
1314 static int blk_tracer_init(struct trace_array *tr)
1315 {
1316         blk_tr = tr;
1317         blk_tracer_start(tr);
1318         return 0;
1319 }
1320
1321 static void blk_tracer_stop(struct trace_array *tr)
1322 {
1323         blk_tracer_enabled = false;
1324 }
1325
1326 static void blk_tracer_reset(struct trace_array *tr)
1327 {
1328         blk_tracer_stop(tr);
1329 }
1330
1331 static const struct {
1332         const char *act[2];
1333         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1334 } what2act[] = {
1335         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1336         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1337         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1338         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1339         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1340         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1341         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1342         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1343         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1344         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1345         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1346         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1347         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1348         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1349         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1350 };
1351
1352 static enum print_line_t print_one_line(struct trace_iterator *iter,
1353                                         bool classic)
1354 {
1355         struct trace_seq *s = &iter->seq;
1356         const struct blk_io_trace *t;
1357         u16 what;
1358         int ret;
1359         bool long_act;
1360         blk_log_action_t *log_action;
1361
1362         t          = te_blk_io_trace(iter->ent);
1363         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1364         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1365         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1366
1367         if (t->action == BLK_TN_MESSAGE) {
1368                 ret = log_action(iter, long_act ? "message" : "m");
1369                 if (ret)
1370                         ret = blk_log_msg(s, iter->ent);
1371                 goto out;
1372         }
1373
1374         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1375                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1376         else {
1377                 ret = log_action(iter, what2act[what].act[long_act]);
1378                 if (ret)
1379                         ret = what2act[what].print(s, iter->ent);
1380         }
1381 out:
1382         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1383 }
1384
1385 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1386                                                int flags, struct trace_event *event)
1387 {
1388         return print_one_line(iter, false);
1389 }
1390
1391 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1392 {
1393         struct trace_seq *s = &iter->seq;
1394         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1395         const int offset = offsetof(struct blk_io_trace, sector);
1396         struct blk_io_trace old = {
1397                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1398                 .time     = iter->ts,
1399         };
1400
1401         if (!trace_seq_putmem(s, &old, offset))
1402                 return 0;
1403         return trace_seq_putmem(s, &t->sector,
1404                                 sizeof(old) - offset + t->pdu_len);
1405 }
1406
1407 static enum print_line_t
1408 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1409                              struct trace_event *event)
1410 {
1411         return blk_trace_synthesize_old_trace(iter) ?
1412                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1413 }
1414
1415 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1416 {
1417         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1418                 return TRACE_TYPE_UNHANDLED;
1419
1420         return print_one_line(iter, true);
1421 }
1422
1423 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1424 {
1425         /* don't output context-info for blk_classic output */
1426         if (bit == TRACE_BLK_OPT_CLASSIC) {
1427                 if (set)
1428                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1429                 else
1430                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1431         }
1432         return 0;
1433 }
1434
1435 static struct tracer blk_tracer __read_mostly = {
1436         .name           = "blk",
1437         .init           = blk_tracer_init,
1438         .reset          = blk_tracer_reset,
1439         .start          = blk_tracer_start,
1440         .stop           = blk_tracer_stop,
1441         .print_header   = blk_tracer_print_header,
1442         .print_line     = blk_tracer_print_line,
1443         .flags          = &blk_tracer_flags,
1444         .set_flag       = blk_tracer_set_flag,
1445 };
1446
1447 static struct trace_event_functions trace_blk_event_funcs = {
1448         .trace          = blk_trace_event_print,
1449         .binary         = blk_trace_event_print_binary,
1450 };
1451
1452 static struct trace_event trace_blk_event = {
1453         .type           = TRACE_BLK,
1454         .funcs          = &trace_blk_event_funcs,
1455 };
1456
1457 static int __init init_blk_tracer(void)
1458 {
1459         if (!register_ftrace_event(&trace_blk_event)) {
1460                 pr_warning("Warning: could not register block events\n");
1461                 return 1;
1462         }
1463
1464         if (register_tracer(&blk_tracer) != 0) {
1465                 pr_warning("Warning: could not register the block tracer\n");
1466                 unregister_ftrace_event(&trace_blk_event);
1467                 return 1;
1468         }
1469
1470         return 0;
1471 }
1472
1473 device_initcall(init_blk_tracer);
1474
1475 static int blk_trace_remove_queue(struct request_queue *q)
1476 {
1477         struct blk_trace *bt;
1478
1479         bt = xchg(&q->blk_trace, NULL);
1480         if (bt == NULL)
1481                 return -EINVAL;
1482
1483         if (atomic_dec_and_test(&blk_probes_ref))
1484                 blk_unregister_tracepoints();
1485
1486         blk_trace_free(bt);
1487         return 0;
1488 }
1489
1490 /*
1491  * Setup everything required to start tracing
1492  */
1493 static int blk_trace_setup_queue(struct request_queue *q,
1494                                  struct block_device *bdev)
1495 {
1496         struct blk_trace *old_bt, *bt = NULL;
1497         int ret = -ENOMEM;
1498
1499         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1500         if (!bt)
1501                 return -ENOMEM;
1502
1503         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1504         if (!bt->msg_data)
1505                 goto free_bt;
1506
1507         bt->dev = bdev->bd_dev;
1508         bt->act_mask = (u16)-1;
1509
1510         blk_trace_setup_lba(bt, bdev);
1511
1512         old_bt = xchg(&q->blk_trace, bt);
1513         if (old_bt != NULL) {
1514                 (void)xchg(&q->blk_trace, old_bt);
1515                 ret = -EBUSY;
1516                 goto free_bt;
1517         }
1518
1519         if (atomic_inc_return(&blk_probes_ref) == 1)
1520                 blk_register_tracepoints();
1521         return 0;
1522
1523 free_bt:
1524         blk_trace_free(bt);
1525         return ret;
1526 }
1527
1528 /*
1529  * sysfs interface to enable and configure tracing
1530  */
1531
1532 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1533                                          struct device_attribute *attr,
1534                                          char *buf);
1535 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1536                                           struct device_attribute *attr,
1537                                           const char *buf, size_t count);
1538 #define BLK_TRACE_DEVICE_ATTR(_name) \
1539         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1540                     sysfs_blk_trace_attr_show, \
1541                     sysfs_blk_trace_attr_store)
1542
1543 static BLK_TRACE_DEVICE_ATTR(enable);
1544 static BLK_TRACE_DEVICE_ATTR(act_mask);
1545 static BLK_TRACE_DEVICE_ATTR(pid);
1546 static BLK_TRACE_DEVICE_ATTR(start_lba);
1547 static BLK_TRACE_DEVICE_ATTR(end_lba);
1548
1549 static struct attribute *blk_trace_attrs[] = {
1550         &dev_attr_enable.attr,
1551         &dev_attr_act_mask.attr,
1552         &dev_attr_pid.attr,
1553         &dev_attr_start_lba.attr,
1554         &dev_attr_end_lba.attr,
1555         NULL
1556 };
1557
1558 struct attribute_group blk_trace_attr_group = {
1559         .name  = "trace",
1560         .attrs = blk_trace_attrs,
1561 };
1562
1563 static const struct {
1564         int mask;
1565         const char *str;
1566 } mask_maps[] = {
1567         { BLK_TC_READ,          "read"          },
1568         { BLK_TC_WRITE,         "write"         },
1569         { BLK_TC_FLUSH,         "flush"         },
1570         { BLK_TC_SYNC,          "sync"          },
1571         { BLK_TC_QUEUE,         "queue"         },
1572         { BLK_TC_REQUEUE,       "requeue"       },
1573         { BLK_TC_ISSUE,         "issue"         },
1574         { BLK_TC_COMPLETE,      "complete"      },
1575         { BLK_TC_FS,            "fs"            },
1576         { BLK_TC_PC,            "pc"            },
1577         { BLK_TC_AHEAD,         "ahead"         },
1578         { BLK_TC_META,          "meta"          },
1579         { BLK_TC_DISCARD,       "discard"       },
1580         { BLK_TC_DRV_DATA,      "drv_data"      },
1581         { BLK_TC_FUA,           "fua"           },
1582 };
1583
1584 static int blk_trace_str2mask(const char *str)
1585 {
1586         int i;
1587         int mask = 0;
1588         char *buf, *s, *token;
1589
1590         buf = kstrdup(str, GFP_KERNEL);
1591         if (buf == NULL)
1592                 return -ENOMEM;
1593         s = strstrip(buf);
1594
1595         while (1) {
1596                 token = strsep(&s, ",");
1597                 if (token == NULL)
1598                         break;
1599
1600                 if (*token == '\0')
1601                         continue;
1602
1603                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1604                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1605                                 mask |= mask_maps[i].mask;
1606                                 break;
1607                         }
1608                 }
1609                 if (i == ARRAY_SIZE(mask_maps)) {
1610                         mask = -EINVAL;
1611                         break;
1612                 }
1613         }
1614         kfree(buf);
1615
1616         return mask;
1617 }
1618
1619 static ssize_t blk_trace_mask2str(char *buf, int mask)
1620 {
1621         int i;
1622         char *p = buf;
1623
1624         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1625                 if (mask & mask_maps[i].mask) {
1626                         p += sprintf(p, "%s%s",
1627                                     (p == buf) ? "" : ",", mask_maps[i].str);
1628                 }
1629         }
1630         *p++ = '\n';
1631
1632         return p - buf;
1633 }
1634
1635 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1636 {
1637         if (bdev->bd_disk == NULL)
1638                 return NULL;
1639
1640         return bdev_get_queue(bdev);
1641 }
1642
1643 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1644                                          struct device_attribute *attr,
1645                                          char *buf)
1646 {
1647         struct hd_struct *p = dev_to_part(dev);
1648         struct request_queue *q;
1649         struct block_device *bdev;
1650         ssize_t ret = -ENXIO;
1651
1652         bdev = bdget(part_devt(p));
1653         if (bdev == NULL)
1654                 goto out;
1655
1656         q = blk_trace_get_queue(bdev);
1657         if (q == NULL)
1658                 goto out_bdput;
1659
1660         mutex_lock(&bdev->bd_mutex);
1661
1662         if (attr == &dev_attr_enable) {
1663                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1664                 goto out_unlock_bdev;
1665         }
1666
1667         if (q->blk_trace == NULL)
1668                 ret = sprintf(buf, "disabled\n");
1669         else if (attr == &dev_attr_act_mask)
1670                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1671         else if (attr == &dev_attr_pid)
1672                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1673         else if (attr == &dev_attr_start_lba)
1674                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1675         else if (attr == &dev_attr_end_lba)
1676                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1677
1678 out_unlock_bdev:
1679         mutex_unlock(&bdev->bd_mutex);
1680 out_bdput:
1681         bdput(bdev);
1682 out:
1683         return ret;
1684 }
1685
1686 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1687                                           struct device_attribute *attr,
1688                                           const char *buf, size_t count)
1689 {
1690         struct block_device *bdev;
1691         struct request_queue *q;
1692         struct hd_struct *p;
1693         u64 value;
1694         ssize_t ret = -EINVAL;
1695
1696         if (count == 0)
1697                 goto out;
1698
1699         if (attr == &dev_attr_act_mask) {
1700                 if (sscanf(buf, "%llx", &value) != 1) {
1701                         /* Assume it is a list of trace category names */
1702                         ret = blk_trace_str2mask(buf);
1703                         if (ret < 0)
1704                                 goto out;
1705                         value = ret;
1706                 }
1707         } else if (sscanf(buf, "%llu", &value) != 1)
1708                 goto out;
1709
1710         ret = -ENXIO;
1711
1712         p = dev_to_part(dev);
1713         bdev = bdget(part_devt(p));
1714         if (bdev == NULL)
1715                 goto out;
1716
1717         q = blk_trace_get_queue(bdev);
1718         if (q == NULL)
1719                 goto out_bdput;
1720
1721         mutex_lock(&bdev->bd_mutex);
1722
1723         if (attr == &dev_attr_enable) {
1724                 if (value)
1725                         ret = blk_trace_setup_queue(q, bdev);
1726                 else
1727                         ret = blk_trace_remove_queue(q);
1728                 goto out_unlock_bdev;
1729         }
1730
1731         ret = 0;
1732         if (q->blk_trace == NULL)
1733                 ret = blk_trace_setup_queue(q, bdev);
1734
1735         if (ret == 0) {
1736                 if (attr == &dev_attr_act_mask)
1737                         q->blk_trace->act_mask = value;
1738                 else if (attr == &dev_attr_pid)
1739                         q->blk_trace->pid = value;
1740                 else if (attr == &dev_attr_start_lba)
1741                         q->blk_trace->start_lba = value;
1742                 else if (attr == &dev_attr_end_lba)
1743                         q->blk_trace->end_lba = value;
1744         }
1745
1746 out_unlock_bdev:
1747         mutex_unlock(&bdev->bd_mutex);
1748 out_bdput:
1749         bdput(bdev);
1750 out:
1751         return ret ? ret : count;
1752 }
1753
1754 int blk_trace_init_sysfs(struct device *dev)
1755 {
1756         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1757 }
1758
1759 void blk_trace_remove_sysfs(struct device *dev)
1760 {
1761         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1762 }
1763
1764 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1765
1766 #ifdef CONFIG_EVENT_TRACING
1767
1768 void blk_dump_cmd(char *buf, struct request *rq)
1769 {
1770         int i, end;
1771         int len = rq->cmd_len;
1772         unsigned char *cmd = rq->cmd;
1773
1774         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1775                 buf[0] = '\0';
1776                 return;
1777         }
1778
1779         for (end = len - 1; end >= 0; end--)
1780                 if (cmd[end])
1781                         break;
1782         end++;
1783
1784         for (i = 0; i < len; i++) {
1785                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1786                 if (i == end && end != len - 1) {
1787                         sprintf(buf, " ..");
1788                         break;
1789                 }
1790         }
1791 }
1792
1793 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1794 {
1795         int i = 0;
1796
1797         if (rw & REQ_FLUSH)
1798                 rwbs[i++] = 'F';
1799
1800         if (rw & WRITE)
1801                 rwbs[i++] = 'W';
1802         else if (rw & REQ_DISCARD)
1803                 rwbs[i++] = 'D';
1804         else if (bytes)
1805                 rwbs[i++] = 'R';
1806         else
1807                 rwbs[i++] = 'N';
1808
1809         if (rw & REQ_FUA)
1810                 rwbs[i++] = 'F';
1811         if (rw & REQ_RAHEAD)
1812                 rwbs[i++] = 'A';
1813         if (rw & REQ_SYNC)
1814                 rwbs[i++] = 'S';
1815         if (rw & REQ_META)
1816                 rwbs[i++] = 'M';
1817         if (rw & REQ_SECURE)
1818                 rwbs[i++] = 'E';
1819
1820         rwbs[i] = '\0';
1821 }
1822
1823 #endif /* CONFIG_EVENT_TRACING */
1824