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
  30 #include <trace/events/block.h>
  31
  32 #include "trace_output.h"
  33
  34 #ifdef CONFIG_BLK_DEV_IO_TRACE
  35
  36 static unsigned int blktrace_seq __read_mostly = 1;
  37
  38 static struct trace_array *blk_tr;
  39 static bool blk_tracer_enabled __read_mostly;
  40
  41 /* Select an alternative, minimalistic output than the original one */
  42 #define TRACE_BLK_OPT_CLASSIC   0x1
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47         { }
  48 };
  49
  50 static struct tracer_flags blk_tracer_flags = {
  51         .val  = 0,
  52         .opts = blk_tracer_opts,
  53 };
  54
  55 /* Global reference count of probes */
  56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  57
  58 static void blk_register_tracepoints(void);
  59 static void blk_unregister_tracepoints(void);
  60
  61 /*
  62  * Send out a notify message.
  63  */
  64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  65                        const void *data, size_t len)
  66 {
  67         struct blk_io_trace *t;
  68         struct ring_buffer_event *event = NULL;
  69         struct ring_buffer *buffer = NULL;
  70         int pc = 0;
  71         int cpu = smp_processor_id();
  72         bool blk_tracer = blk_tracer_enabled;
  73
  74         if (blk_tracer) {
  75                 buffer = blk_tr->buffer;
  76                 pc = preempt_count();
  77                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  78                                                   sizeof(*t) + len,
  79                                                   0, pc);
  80                 if (!event)
  81                         return;
  82                 t = ring_buffer_event_data(event);
  83                 goto record_it;
  84         }
  85
  86         if (!bt->rchan)
  87                 return;
  88
  89         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  90         if (t) {
  91                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  92                 t->time = ktime_to_ns(ktime_get());
  93 record_it:
  94                 t->device = bt->dev;
  95                 t->action = action;
  96                 t->pid = pid;
  97                 t->cpu = cpu;
  98                 t->pdu_len = len;
  99                 memcpy((void *) t + sizeof(*t), data, len);
 100
 101                 if (blk_tracer)
 102                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 103         }
 104 }
 105
 106 /*
 107  * Send out a notify for this process, if we haven't done so since a trace
 108  * started
 109  */
 110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 111 {
 112         tsk->btrace_seq = blktrace_seq;
 113         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 114 }
 115
 116 static void trace_note_time(struct blk_trace *bt)
 117 {
 118         struct timespec now;
 119         unsigned long flags;
 120         u32 words[2];
 121
 122         getnstimeofday(&now);
 123         words[0] = now.tv_sec;
 124         words[1] = now.tv_nsec;
 125
 126         local_irq_save(flags);
 127         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 128         local_irq_restore(flags);
 129 }
 130
 131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 132 {
 133         int n;
 134         va_list args;
 135         unsigned long flags;
 136         char *buf;
 137
 138         if (unlikely(bt->trace_state != Blktrace_running &&
 139                      !blk_tracer_enabled))
 140                 return;
 141
 142         /*
 143          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 144          * message to the trace.
 145          */
 146         if (!(bt->act_mask & BLK_TC_NOTIFY))
 147                 return;
 148
 149         local_irq_save(flags);
 150         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 151         va_start(args, fmt);
 152         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 153         va_end(args);
 154
 155         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 156         local_irq_restore(flags);
 157 }
 158 EXPORT_SYMBOL_GPL(__trace_note_message);
 159
 160 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 161                          pid_t pid)
 162 {
 163         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 164                 return 1;
 165         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 166                 return 1;
 167         if (bt->pid && pid != bt->pid)
 168                 return 1;
 169
 170         return 0;
 171 }
 172
 173 /*
 174  * Data direction bit lookup
 175  */
 176 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 177                                  BLK_TC_ACT(BLK_TC_WRITE) };
 178
 179 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 180
 181 /* The ilog2() calls fall out because they're constant */
 182 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 183           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 184
 185 /*
 186  * The worker for the various blk_add_trace*() types. Fills out a
 187  * blk_io_trace structure and places it in a per-cpu subbuffer.
 188  */
 189 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 190                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 191 {
 192         struct task_struct *tsk = current;
 193         struct ring_buffer_event *event = NULL;
 194         struct ring_buffer *buffer = NULL;
 195         struct blk_io_trace *t;
 196         unsigned long flags = 0;
 197         unsigned long *sequence;
 198         pid_t pid;
 199         int cpu, pc = 0;
 200         bool blk_tracer = blk_tracer_enabled;
 201
 202         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 203                 return;
 204
 205         what |= ddir_act[rw & WRITE];
 206         what |= MASK_TC_BIT(rw, SYNC);
 207         what |= MASK_TC_BIT(rw, RAHEAD);
 208         what |= MASK_TC_BIT(rw, META);
 209         what |= MASK_TC_BIT(rw, DISCARD);
 210         what |= MASK_TC_BIT(rw, FLUSH);
 211         what |= MASK_TC_BIT(rw, FUA);
 212
 213         pid = tsk->pid;
 214         if (act_log_check(bt, what, sector, pid))
 215                 return;
 216         cpu = raw_smp_processor_id();
 217
 218         if (blk_tracer) {
 219                 tracing_record_cmdline(current);
 220
 221                 buffer = blk_tr->buffer;
 222                 pc = preempt_count();
 223                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 224                                                   sizeof(*t) + pdu_len,
 225                                                   0, pc);
 226                 if (!event)
 227                         return;
 228                 t = ring_buffer_event_data(event);
 229                 goto record_it;
 230         }
 231
 232         /*
 233          * A word about the locking here - we disable interrupts to reserve
 234          * some space in the relay per-cpu buffer, to prevent an irq
 235          * from coming in and stepping on our toes.
 236          */
 237         local_irq_save(flags);
 238
 239         if (unlikely(tsk->btrace_seq != blktrace_seq))
 240                 trace_note_tsk(bt, tsk);
 241
 242         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 243         if (t) {
 244                 sequence = per_cpu_ptr(bt->sequence, cpu);
 245
 246                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 247                 t->sequence = ++(*sequence);
 248                 t->time = ktime_to_ns(ktime_get());
 249 record_it:
 250                 /*
 251                  * These two are not needed in ftrace as they are in the
 252                  * generic trace_entry, filled by tracing_generic_entry_update,
 253                  * but for the trace_event->bin() synthesizer benefit we do it
 254                  * here too.
 255                  */
 256                 t->cpu = cpu;
 257                 t->pid = pid;
 258
 259                 t->sector = sector;
 260                 t->bytes = bytes;
 261                 t->action = what;
 262                 t->device = bt->dev;
 263                 t->error = error;
 264                 t->pdu_len = pdu_len;
 265
 266                 if (pdu_len)
 267                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 268
 269                 if (blk_tracer) {
 270                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 271                         return;
 272                 }
 273         }
 274
 275         local_irq_restore(flags);
 276 }
 277
 278 static struct dentry *blk_tree_root;
 279 static DEFINE_MUTEX(blk_tree_mutex);
 280
 281 static void blk_trace_free(struct blk_trace *bt)
 282 {
 283         debugfs_remove(bt->msg_file);
 284         debugfs_remove(bt->dropped_file);
 285         relay_close(bt->rchan);
 286         debugfs_remove(bt->dir);
 287         free_percpu(bt->sequence);
 288         free_percpu(bt->msg_data);
 289         kfree(bt);
 290 }
 291
 292 static void blk_trace_cleanup(struct blk_trace *bt)
 293 {
 294         blk_trace_free(bt);
 295         if (atomic_dec_and_test(&blk_probes_ref))
 296                 blk_unregister_tracepoints();
 297 }
 298
 299 int blk_trace_remove(struct request_queue *q)
 300 {
 301         struct blk_trace *bt;
 302
 303         bt = xchg(&q->blk_trace, NULL);
 304         if (!bt)
 305                 return -EINVAL;
 306
 307         if (bt->trace_state != Blktrace_running)
 308                 blk_trace_cleanup(bt);
 309
 310         return 0;
 311 }
 312 EXPORT_SYMBOL_GPL(blk_trace_remove);
 313
 314 static int blk_dropped_open(struct inode *inode, struct file *filp)
 315 {
 316         filp->private_data = inode->i_private;
 317
 318         return 0;
 319 }
 320
 321 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 322                                 size_t count, loff_t *ppos)
 323 {
 324         struct blk_trace *bt = filp->private_data;
 325         char buf[16];
 326
 327         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 328
 329         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 330 }
 331
 332 static const struct file_operations blk_dropped_fops = {
 333         .owner =        THIS_MODULE,
 334         .open =         blk_dropped_open,
 335         .read =         blk_dropped_read,
 336         .llseek =       default_llseek,
 337 };
 338
 339 static int blk_msg_open(struct inode *inode, struct file *filp)
 340 {
 341         filp->private_data = inode->i_private;
 342
 343         return 0;
 344 }
 345
 346 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 347                                 size_t count, loff_t *ppos)
 348 {
 349         char *msg;
 350         struct blk_trace *bt;
 351
 352         if (count >= BLK_TN_MAX_MSG)
 353                 return -EINVAL;
 354
 355         msg = kmalloc(count + 1, GFP_KERNEL);
 356         if (msg == NULL)
 357                 return -ENOMEM;
 358
 359         if (copy_from_user(msg, buffer, count)) {
 360                 kfree(msg);
 361                 return -EFAULT;
 362         }
 363
 364         msg[count] = '\0';
 365         bt = filp->private_data;
 366         __trace_note_message(bt, "%s", msg);
 367         kfree(msg);
 368
 369         return count;
 370 }
 371
 372 static const struct file_operations blk_msg_fops = {
 373         .owner =        THIS_MODULE,
 374         .open =         blk_msg_open,
 375         .write =        blk_msg_write,
 376         .llseek =       noop_llseek,
 377 };
 378
 379 /*
 380  * Keep track of how many times we encountered a full subbuffer, to aid
 381  * the user space app in telling how many lost events there were.
 382  */
 383 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 384                                      void *prev_subbuf, size_t prev_padding)
 385 {
 386         struct blk_trace *bt;
 387
 388         if (!relay_buf_full(buf))
 389                 return 1;
 390
 391         bt = buf->chan->private_data;
 392         atomic_inc(&bt->dropped);
 393         return 0;
 394 }
 395
 396 static int blk_remove_buf_file_callback(struct dentry *dentry)
 397 {
 398         debugfs_remove(dentry);
 399
 400         return 0;
 401 }
 402
 403 static struct dentry *blk_create_buf_file_callback(const char *filename,
 404                                                    struct dentry *parent,
 405                                                    int mode,
 406                                                    struct rchan_buf *buf,
 407                                                    int *is_global)
 408 {
 409         return debugfs_create_file(filename, mode, parent, buf,
 410                                         &relay_file_operations);
 411 }
 412
 413 static struct rchan_callbacks blk_relay_callbacks = {
 414         .subbuf_start           = blk_subbuf_start_callback,
 415         .create_buf_file        = blk_create_buf_file_callback,
 416         .remove_buf_file        = blk_remove_buf_file_callback,
 417 };
 418
 419 static void blk_trace_setup_lba(struct blk_trace *bt,
 420                                 struct block_device *bdev)
 421 {
 422         struct hd_struct *part = NULL;
 423
 424         if (bdev)
 425                 part = bdev->bd_part;
 426
 427         if (part) {
 428                 bt->start_lba = part->start_sect;
 429                 bt->end_lba = part->start_sect + part->nr_sects;
 430         } else {
 431                 bt->start_lba = 0;
 432                 bt->end_lba = -1ULL;
 433         }
 434 }
 435
 436 /*
 437  * Setup everything required to start tracing
 438  */
 439 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 440                        struct block_device *bdev,
 441                        struct blk_user_trace_setup *buts)
 442 {
 443         struct blk_trace *old_bt, *bt = NULL;
 444         struct dentry *dir = NULL;
 445         int ret, i;
 446
 447         if (!buts->buf_size || !buts->buf_nr)
 448                 return -EINVAL;
 449
 450         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 451         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 452
 453         /*
 454          * some device names have larger paths - convert the slashes
 455          * to underscores for this to work as expected
 456          */
 457         for (i = 0; i < strlen(buts->name); i++)
 458                 if (buts->name[i] == '/')
 459                         buts->name[i] = '_';
 460
 461         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 462         if (!bt)
 463                 return -ENOMEM;
 464
 465         ret = -ENOMEM;
 466         bt->sequence = alloc_percpu(unsigned long);
 467         if (!bt->sequence)
 468                 goto err;
 469
 470         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 471         if (!bt->msg_data)
 472                 goto err;
 473
 474         ret = -ENOENT;
 475
 476         mutex_lock(&blk_tree_mutex);
 477         if (!blk_tree_root) {
 478                 blk_tree_root = debugfs_create_dir("block", NULL);
 479                 if (!blk_tree_root) {
 480                         mutex_unlock(&blk_tree_mutex);
 481                         goto err;
 482                 }
 483         }
 484         mutex_unlock(&blk_tree_mutex);
 485
 486         dir = debugfs_create_dir(buts->name, blk_tree_root);
 487
 488         if (!dir)
 489                 goto err;
 490
 491         bt->dir = dir;
 492         bt->dev = dev;
 493         atomic_set(&bt->dropped, 0);
 494
 495         ret = -EIO;
 496         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 497                                                &blk_dropped_fops);
 498         if (!bt->dropped_file)
 499                 goto err;
 500
 501         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 502         if (!bt->msg_file)
 503                 goto err;
 504
 505         bt->rchan = relay_open("trace", dir, buts->buf_size,
 506                                 buts->buf_nr, &blk_relay_callbacks, bt);
 507         if (!bt->rchan)
 508                 goto err;
 509
 510         bt->act_mask = buts->act_mask;
 511         if (!bt->act_mask)
 512                 bt->act_mask = (u16) -1;
 513
 514         blk_trace_setup_lba(bt, bdev);
 515
 516         /* overwrite with user settings */
 517         if (buts->start_lba)
 518                 bt->start_lba = buts->start_lba;
 519         if (buts->end_lba)
 520                 bt->end_lba = buts->end_lba;
 521
 522         bt->pid = buts->pid;
 523         bt->trace_state = Blktrace_setup;
 524
 525         ret = -EBUSY;
 526         old_bt = xchg(&q->blk_trace, bt);
 527         if (old_bt) {
 528                 (void) xchg(&q->blk_trace, old_bt);
 529                 goto err;
 530         }
 531
 532         if (atomic_inc_return(&blk_probes_ref) == 1)
 533                 blk_register_tracepoints();
 534
 535         return 0;
 536 err:
 537         blk_trace_free(bt);
 538         return ret;
 539 }
 540
 541 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 542                     struct block_device *bdev,
 543                     char __user *arg)
 544 {
 545         struct blk_user_trace_setup buts;
 546         int ret;
 547
 548         ret = copy_from_user(&buts, arg, sizeof(buts));
 549         if (ret)
 550                 return -EFAULT;
 551
 552         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 553         if (ret)
 554                 return ret;
 555
 556         if (copy_to_user(arg, &buts, sizeof(buts))) {
 557                 blk_trace_remove(q);
 558                 return -EFAULT;
 559         }
 560         return 0;
 561 }
 562 EXPORT_SYMBOL_GPL(blk_trace_setup);
 563
 564 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 565 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 566                                   dev_t dev, struct block_device *bdev,
 567                                   char __user *arg)
 568 {
 569         struct blk_user_trace_setup buts;
 570         struct compat_blk_user_trace_setup cbuts;
 571         int ret;
 572
 573         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 574                 return -EFAULT;
 575
 576         buts = (struct blk_user_trace_setup) {
 577                 .act_mask = cbuts.act_mask,
 578                 .buf_size = cbuts.buf_size,
 579                 .buf_nr = cbuts.buf_nr,
 580                 .start_lba = cbuts.start_lba,
 581                 .end_lba = cbuts.end_lba,
 582                 .pid = cbuts.pid,
 583         };
 584         memcpy(&buts.name, &cbuts.name, 32);
 585
 586         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 587         if (ret)
 588                 return ret;
 589
 590         if (copy_to_user(arg, &buts.name, 32)) {
 591                 blk_trace_remove(q);
 592                 return -EFAULT;
 593         }
 594
 595         return 0;
 596 }
 597 #endif
 598
 599 int blk_trace_startstop(struct request_queue *q, int start)
 600 {
 601         int ret;
 602         struct blk_trace *bt = q->blk_trace;
 603
 604         if (bt == NULL)
 605                 return -EINVAL;
 606
 607         /*
 608          * For starting a trace, we can transition from a setup or stopped
 609          * trace. For stopping a trace, the state must be running
 610          */
 611         ret = -EINVAL;
 612         if (start) {
 613                 if (bt->trace_state == Blktrace_setup ||
 614                     bt->trace_state == Blktrace_stopped) {
 615                         blktrace_seq++;
 616                         smp_mb();
 617                         bt->trace_state = Blktrace_running;
 618
 619                         trace_note_time(bt);
 620                         ret = 0;
 621                 }
 622         } else {
 623                 if (bt->trace_state == Blktrace_running) {
 624                         bt->trace_state = Blktrace_stopped;
 625                         relay_flush(bt->rchan);
 626                         ret = 0;
 627                 }
 628         }
 629
 630         return ret;
 631 }
 632 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 633
 634 /**
 635  * blk_trace_ioctl: - handle the ioctls associated with tracing
 636  * @bdev:       the block device
 637  * @cmd:        the ioctl cmd
 638  * @arg:        the argument data, if any
 639  *
 640  **/
 641 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 642 {
 643         struct request_queue *q;
 644         int ret, start = 0;
 645         char b[BDEVNAME_SIZE];
 646
 647         q = bdev_get_queue(bdev);
 648         if (!q)
 649                 return -ENXIO;
 650
 651         mutex_lock(&bdev->bd_mutex);
 652
 653         switch (cmd) {
 654         case BLKTRACESETUP:
 655                 bdevname(bdev, b);
 656                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 657                 break;
 658 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 659         case BLKTRACESETUP32:
 660                 bdevname(bdev, b);
 661                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 662                 break;
 663 #endif
 664         case BLKTRACESTART:
 665                 start = 1;
 666         case BLKTRACESTOP:
 667                 ret = blk_trace_startstop(q, start);
 668                 break;
 669         case BLKTRACETEARDOWN:
 670                 ret = blk_trace_remove(q);
 671                 break;
 672         default:
 673                 ret = -ENOTTY;
 674                 break;
 675         }
 676
 677         mutex_unlock(&bdev->bd_mutex);
 678         return ret;
 679 }
 680
 681 /**
 682  * blk_trace_shutdown: - stop and cleanup trace structures
 683  * @q:    the request queue associated with the device
 684  *
 685  **/
 686 void blk_trace_shutdown(struct request_queue *q)
 687 {
 688         if (q->blk_trace) {
 689                 blk_trace_startstop(q, 0);
 690                 blk_trace_remove(q);
 691         }
 692 }
 693
 694 /*
 695  * blktrace probes
 696  */
 697
 698 /**
 699  * blk_add_trace_rq - Add a trace for a request oriented action
 700  * @q:          queue the io is for
 701  * @rq:         the source request
 702  * @what:       the action
 703  *
 704  * Description:
 705  *     Records an action against a request. Will log the bio offset + size.
 706  *
 707  **/
 708 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 709                              u32 what)
 710 {
 711         struct blk_trace *bt = q->blk_trace;
 712
 713         if (likely(!bt))
 714                 return;
 715
 716         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 717                 what |= BLK_TC_ACT(BLK_TC_PC);
 718                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
 719                                 what, rq->errors, rq->cmd_len, rq->cmd);
 720         } else  {
 721                 what |= BLK_TC_ACT(BLK_TC_FS);
 722                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 723                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 724         }
 725 }
 726
 727 static void blk_add_trace_rq_abort(void *ignore,
 728                                    struct request_queue *q, struct request *rq)
 729 {
 730         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 731 }
 732
 733 static void blk_add_trace_rq_insert(void *ignore,
 734                                     struct request_queue *q, struct request *rq)
 735 {
 736         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 737 }
 738
 739 static void blk_add_trace_rq_issue(void *ignore,
 740                                    struct request_queue *q, struct request *rq)
 741 {
 742         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 743 }
 744
 745 static void blk_add_trace_rq_requeue(void *ignore,
 746                                      struct request_queue *q,
 747                                      struct request *rq)
 748 {
 749         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 750 }
 751
 752 static void blk_add_trace_rq_complete(void *ignore,
 753                                       struct request_queue *q,
 754                                       struct request *rq)
 755 {
 756         blk_add_trace_rq(q, rq, 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         if (!error && !bio_flagged(bio, BIO_UPTODATE))
 779                 error = EIO;
 780
 781         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 782                         error, 0, NULL);
 783 }
 784
 785 static void blk_add_trace_bio_bounce(void *ignore,
 786                                      struct request_queue *q, struct bio *bio)
 787 {
 788         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 789 }
 790
 791 static void blk_add_trace_bio_complete(void *ignore,
 792                                        struct request_queue *q, struct bio *bio,
 793                                        int error)
 794 {
 795         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 796 }
 797
 798 static void blk_add_trace_bio_backmerge(void *ignore,
 799                                         struct request_queue *q,
 800                                         struct bio *bio)
 801 {
 802         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 803 }
 804
 805 static void blk_add_trace_bio_frontmerge(void *ignore,
 806                                          struct request_queue *q,
 807                                          struct bio *bio)
 808 {
 809         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 810 }
 811
 812 static void blk_add_trace_bio_queue(void *ignore,
 813                                     struct request_queue *q, struct bio *bio)
 814 {
 815         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 816 }
 817
 818 static void blk_add_trace_getrq(void *ignore,
 819                                 struct request_queue *q,
 820                                 struct bio *bio, int rw)
 821 {
 822         if (bio)
 823                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 824         else {
 825                 struct blk_trace *bt = q->blk_trace;
 826
 827                 if (bt)
 828                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 829         }
 830 }
 831
 832
 833 static void blk_add_trace_sleeprq(void *ignore,
 834                                   struct request_queue *q,
 835                                   struct bio *bio, int rw)
 836 {
 837         if (bio)
 838                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 839         else {
 840                 struct blk_trace *bt = q->blk_trace;
 841
 842                 if (bt)
 843                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 844                                         0, 0, NULL);
 845         }
 846 }
 847
 848 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 849 {
 850         struct blk_trace *bt = q->blk_trace;
 851
 852         if (bt)
 853                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 854 }
 855
 856 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 857                                     unsigned int depth, bool explicit)
 858 {
 859         struct blk_trace *bt = q->blk_trace;
 860
 861         if (bt) {
 862                 __be64 rpdu = cpu_to_be64(depth);
 863                 u32 what;
 864
 865                 if (explicit)
 866                         what = BLK_TA_UNPLUG_IO;
 867                 else
 868                         what = BLK_TA_UNPLUG_TIMER;
 869
 870                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 871         }
 872 }
 873
 874 static void blk_add_trace_split(void *ignore,
 875                                 struct request_queue *q, struct bio *bio,
 876                                 unsigned int pdu)
 877 {
 878         struct blk_trace *bt = q->blk_trace;
 879
 880         if (bt) {
 881                 __be64 rpdu = cpu_to_be64(pdu);
 882
 883                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 884                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 885                                 sizeof(rpdu), &rpdu);
 886         }
 887 }
 888
 889 /**
 890  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 891  * @ignore:     trace callback data parameter (not used)
 892  * @q:          queue the io is for
 893  * @bio:        the source bio
 894  * @dev:        target device
 895  * @from:       source sector
 896  *
 897  * Description:
 898  *     Device mapper or raid target sometimes need to split a bio because
 899  *     it spans a stripe (or similar). Add a trace for that action.
 900  *
 901  **/
 902 static void blk_add_trace_bio_remap(void *ignore,
 903                                     struct request_queue *q, struct bio *bio,
 904                                     dev_t dev, sector_t from)
 905 {
 906         struct blk_trace *bt = q->blk_trace;
 907         struct blk_io_trace_remap r;
 908
 909         if (likely(!bt))
 910                 return;
 911
 912         r.device_from = cpu_to_be32(dev);
 913         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 914         r.sector_from = cpu_to_be64(from);
 915
 916         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 917                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 918                         sizeof(r), &r);
 919 }
 920
 921 /**
 922  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 923  * @ignore:     trace callback data parameter (not used)
 924  * @q:          queue the io is for
 925  * @rq:         the source request
 926  * @dev:        target device
 927  * @from:       source sector
 928  *
 929  * Description:
 930  *     Device mapper remaps request to other devices.
 931  *     Add a trace for that action.
 932  *
 933  **/
 934 static void blk_add_trace_rq_remap(void *ignore,
 935                                    struct request_queue *q,
 936                                    struct request *rq, dev_t dev,
 937                                    sector_t from)
 938 {
 939         struct blk_trace *bt = q->blk_trace;
 940         struct blk_io_trace_remap r;
 941
 942         if (likely(!bt))
 943                 return;
 944
 945         r.device_from = cpu_to_be32(dev);
 946         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 947         r.sector_from = cpu_to_be64(from);
 948
 949         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 950                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 951                         sizeof(r), &r);
 952 }
 953
 954 /**
 955  * blk_add_driver_data - Add binary message with driver-specific data
 956  * @q:          queue the io is for
 957  * @rq:         io request
 958  * @data:       driver-specific data
 959  * @len:        length of driver-specific data
 960  *
 961  * Description:
 962  *     Some drivers might want to write driver-specific data per request.
 963  *
 964  **/
 965 void blk_add_driver_data(struct request_queue *q,
 966                          struct request *rq,
 967                          void *data, size_t len)
 968 {
 969         struct blk_trace *bt = q->blk_trace;
 970
 971         if (likely(!bt))
 972                 return;
 973
 974         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 975                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 976                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 977         else
 978                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 979                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 980 }
 981 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 982
 983 static void blk_register_tracepoints(void)
 984 {
 985         int ret;
 986
 987         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 988         WARN_ON(ret);
 989         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 990         WARN_ON(ret);
 991         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 992         WARN_ON(ret);
 993         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 994         WARN_ON(ret);
 995         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 996         WARN_ON(ret);
 997         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 998         WARN_ON(ret);
 999         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1000         WARN_ON(ret);
1001         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1002         WARN_ON(ret);
1003         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1004         WARN_ON(ret);
1005         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1006         WARN_ON(ret);
1007         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1008         WARN_ON(ret);
1009         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1010         WARN_ON(ret);
1011         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1012         WARN_ON(ret);
1013         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1014         WARN_ON(ret);
1015         ret = register_trace_block_split(blk_add_trace_split, NULL);
1016         WARN_ON(ret);
1017         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1018         WARN_ON(ret);
1019         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1020         WARN_ON(ret);
1021 }
1022
1023 static void blk_unregister_tracepoints(void)
1024 {
1025         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1026         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1027         unregister_trace_block_split(blk_add_trace_split, NULL);
1028         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1029         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1030         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1031         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1032         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1033         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1034         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1035         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1036         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1037         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1038         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1039         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1040         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1041         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1042
1043         tracepoint_synchronize_unregister();
1044 }
1045
1046 /*
1047  * struct blk_io_tracer formatting routines
1048  */
1049
1050 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1051 {
1052         int i = 0;
1053         int tc = t->action >> BLK_TC_SHIFT;
1054
1055         if (t->action == BLK_TN_MESSAGE) {
1056                 rwbs[i++] = 'N';
1057                 goto out;
1058         }
1059
1060         if (tc & BLK_TC_FLUSH)
1061                 rwbs[i++] = 'F';
1062
1063         if (tc & BLK_TC_DISCARD)
1064                 rwbs[i++] = 'D';
1065         else if (tc & BLK_TC_WRITE)
1066                 rwbs[i++] = 'W';
1067         else if (t->bytes)
1068                 rwbs[i++] = 'R';
1069         else
1070                 rwbs[i++] = 'N';
1071
1072         if (tc & BLK_TC_FUA)
1073                 rwbs[i++] = 'F';
1074         if (tc & BLK_TC_AHEAD)
1075                 rwbs[i++] = 'A';
1076         if (tc & BLK_TC_SYNC)
1077                 rwbs[i++] = 'S';
1078         if (tc & BLK_TC_META)
1079                 rwbs[i++] = 'M';
1080 out:
1081         rwbs[i] = '\0';
1082 }
1083
1084 static inline
1085 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1086 {
1087         return (const struct blk_io_trace *)ent;
1088 }
1089
1090 static inline const void *pdu_start(const struct trace_entry *ent)
1091 {
1092         return te_blk_io_trace(ent) + 1;
1093 }
1094
1095 static inline u32 t_action(const struct trace_entry *ent)
1096 {
1097         return te_blk_io_trace(ent)->action;
1098 }
1099
1100 static inline u32 t_bytes(const struct trace_entry *ent)
1101 {
1102         return te_blk_io_trace(ent)->bytes;
1103 }
1104
1105 static inline u32 t_sec(const struct trace_entry *ent)
1106 {
1107         return te_blk_io_trace(ent)->bytes >> 9;
1108 }
1109
1110 static inline unsigned long long t_sector(const struct trace_entry *ent)
1111 {
1112         return te_blk_io_trace(ent)->sector;
1113 }
1114
1115 static inline __u16 t_error(const struct trace_entry *ent)
1116 {
1117         return te_blk_io_trace(ent)->error;
1118 }
1119
1120 static __u64 get_pdu_int(const struct trace_entry *ent)
1121 {
1122         const __u64 *val = pdu_start(ent);
1123         return be64_to_cpu(*val);
1124 }
1125
1126 static void get_pdu_remap(const struct trace_entry *ent,
1127                           struct blk_io_trace_remap *r)
1128 {
1129         const struct blk_io_trace_remap *__r = pdu_start(ent);
1130         __u64 sector_from = __r->sector_from;
1131
1132         r->device_from = be32_to_cpu(__r->device_from);
1133         r->device_to   = be32_to_cpu(__r->device_to);
1134         r->sector_from = be64_to_cpu(sector_from);
1135 }
1136
1137 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1138
1139 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1140 {
1141         char rwbs[RWBS_LEN];
1142         unsigned long long ts  = iter->ts;
1143         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1144         unsigned secs          = (unsigned long)ts;
1145         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1146
1147         fill_rwbs(rwbs, t);
1148
1149         return trace_seq_printf(&iter->seq,
1150                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1151                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1152                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1153 }
1154
1155 static int blk_log_action(struct trace_iterator *iter, const char *act)
1156 {
1157         char rwbs[RWBS_LEN];
1158         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1159
1160         fill_rwbs(rwbs, t);
1161         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1162                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1163 }
1164
1165 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1166 {
1167         const unsigned char *pdu_buf;
1168         int pdu_len;
1169         int i, end, ret;
1170
1171         pdu_buf = pdu_start(ent);
1172         pdu_len = te_blk_io_trace(ent)->pdu_len;
1173
1174         if (!pdu_len)
1175                 return 1;
1176
1177         /* find the last zero that needs to be printed */
1178         for (end = pdu_len - 1; end >= 0; end--)
1179                 if (pdu_buf[end])
1180                         break;
1181         end++;
1182
1183         if (!trace_seq_putc(s, '('))
1184                 return 0;
1185
1186         for (i = 0; i < pdu_len; i++) {
1187
1188                 ret = trace_seq_printf(s, "%s%02x",
1189                                        i == 0 ? "" : " ", pdu_buf[i]);
1190                 if (!ret)
1191                         return ret;
1192
1193                 /*
1194                  * stop when the rest is just zeroes and indicate so
1195                  * with a ".." appended
1196                  */
1197                 if (i == end && end != pdu_len - 1)
1198                         return trace_seq_puts(s, " ..) ");
1199         }
1200
1201         return trace_seq_puts(s, ") ");
1202 }
1203
1204 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1205 {
1206         char cmd[TASK_COMM_LEN];
1207
1208         trace_find_cmdline(ent->pid, cmd);
1209
1210         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1211                 int ret;
1212
1213                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1214                 if (!ret)
1215                         return 0;
1216                 ret = blk_log_dump_pdu(s, ent);
1217                 if (!ret)
1218                         return 0;
1219                 return trace_seq_printf(s, "[%s]\n", cmd);
1220         } else {
1221                 if (t_sec(ent))
1222                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1223                                                 t_sector(ent), t_sec(ent), cmd);
1224                 return trace_seq_printf(s, "[%s]\n", cmd);
1225         }
1226 }
1227
1228 static int blk_log_with_error(struct trace_seq *s,
1229                               const struct trace_entry *ent)
1230 {
1231         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1232                 int ret;
1233
1234                 ret = blk_log_dump_pdu(s, ent);
1235                 if (ret)
1236                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1237                 return 0;
1238         } else {
1239                 if (t_sec(ent))
1240                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1241                                                 t_sector(ent),
1242                                                 t_sec(ent), t_error(ent));
1243                 return trace_seq_printf(s, "%llu [%d]\n",
1244                                         t_sector(ent), t_error(ent));
1245         }
1246 }
1247
1248 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1249 {
1250         struct blk_io_trace_remap r = { .device_from = 0, };
1251
1252         get_pdu_remap(ent, &r);
1253         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1254                                 t_sector(ent), t_sec(ent),
1255                                 MAJOR(r.device_from), MINOR(r.device_from),
1256                                 (unsigned long long)r.sector_from);
1257 }
1258
1259 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1260 {
1261         char cmd[TASK_COMM_LEN];
1262
1263         trace_find_cmdline(ent->pid, cmd);
1264
1265         return trace_seq_printf(s, "[%s]\n", cmd);
1266 }
1267
1268 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1269 {
1270         char cmd[TASK_COMM_LEN];
1271
1272         trace_find_cmdline(ent->pid, cmd);
1273
1274         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1275 }
1276
1277 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1278 {
1279         char cmd[TASK_COMM_LEN];
1280
1281         trace_find_cmdline(ent->pid, cmd);
1282
1283         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1284                                 get_pdu_int(ent), cmd);
1285 }
1286
1287 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1288 {
1289         int ret;
1290         const struct blk_io_trace *t = te_blk_io_trace(ent);
1291
1292         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1293         if (ret)
1294                 return trace_seq_putc(s, '\n');
1295         return ret;
1296 }
1297
1298 /*
1299  * struct tracer operations
1300  */
1301
1302 static void blk_tracer_print_header(struct seq_file *m)
1303 {
1304         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1305                 return;
1306         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1307                     "#  |     |     |           |   |   |\n");
1308 }
1309
1310 static void blk_tracer_start(struct trace_array *tr)
1311 {
1312         blk_tracer_enabled = true;
1313 }
1314
1315 static int blk_tracer_init(struct trace_array *tr)
1316 {
1317         blk_tr = tr;
1318         blk_tracer_start(tr);
1319         return 0;
1320 }
1321
1322 static void blk_tracer_stop(struct trace_array *tr)
1323 {
1324         blk_tracer_enabled = false;
1325 }
1326
1327 static void blk_tracer_reset(struct trace_array *tr)
1328 {
1329         blk_tracer_stop(tr);
1330 }
1331
1332 static const struct {
1333         const char *act[2];
1334         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1335 } what2act[] = {
1336         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1337         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1338         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1339         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1340         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1341         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1342         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1343         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1344         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1345         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1346         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1347         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1348         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1349         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1350         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1351 };
1352
1353 static enum print_line_t print_one_line(struct trace_iterator *iter,
1354                                         bool classic)
1355 {
1356         struct trace_seq *s = &iter->seq;
1357         const struct blk_io_trace *t;
1358         u16 what;
1359         int ret;
1360         bool long_act;
1361         blk_log_action_t *log_action;
1362
1363         t          = te_blk_io_trace(iter->ent);
1364         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1365         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1366         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1367
1368         if (t->action == BLK_TN_MESSAGE) {
1369                 ret = log_action(iter, long_act ? "message" : "m");
1370                 if (ret)
1371                         ret = blk_log_msg(s, iter->ent);
1372                 goto out;
1373         }
1374
1375         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1376                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1377         else {
1378                 ret = log_action(iter, what2act[what].act[long_act]);
1379                 if (ret)
1380                         ret = what2act[what].print(s, iter->ent);
1381         }
1382 out:
1383         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1384 }
1385
1386 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1387                                                int flags, struct trace_event *event)
1388 {
1389         return print_one_line(iter, false);
1390 }
1391
1392 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1393 {
1394         struct trace_seq *s = &iter->seq;
1395         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1396         const int offset = offsetof(struct blk_io_trace, sector);
1397         struct blk_io_trace old = {
1398                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1399                 .time     = iter->ts,
1400         };
1401
1402         if (!trace_seq_putmem(s, &old, offset))
1403                 return 0;
1404         return trace_seq_putmem(s, &t->sector,
1405                                 sizeof(old) - offset + t->pdu_len);
1406 }
1407
1408 static enum print_line_t
1409 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1410                              struct trace_event *event)
1411 {
1412         return blk_trace_synthesize_old_trace(iter) ?
1413                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1414 }
1415
1416 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1417 {
1418         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1419                 return TRACE_TYPE_UNHANDLED;
1420
1421         return print_one_line(iter, true);
1422 }
1423
1424 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1425 {
1426         /* don't output context-info for blk_classic output */
1427         if (bit == TRACE_BLK_OPT_CLASSIC) {
1428                 if (set)
1429                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1430                 else
1431                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1432         }
1433         return 0;
1434 }
1435
1436 static struct tracer blk_tracer __read_mostly = {
1437         .name           = "blk",
1438         .init           = blk_tracer_init,
1439         .reset          = blk_tracer_reset,
1440         .start          = blk_tracer_start,
1441         .stop           = blk_tracer_stop,
1442         .print_header   = blk_tracer_print_header,
1443         .print_line     = blk_tracer_print_line,
1444         .flags          = &blk_tracer_flags,
1445         .set_flag       = blk_tracer_set_flag,
1446 };
1447
1448 static struct trace_event_functions trace_blk_event_funcs = {
1449         .trace          = blk_trace_event_print,
1450         .binary         = blk_trace_event_print_binary,
1451 };
1452
1453 static struct trace_event trace_blk_event = {
1454         .type           = TRACE_BLK,
1455         .funcs          = &trace_blk_event_funcs,
1456 };
1457
1458 static int __init init_blk_tracer(void)
1459 {
1460         if (!register_ftrace_event(&trace_blk_event)) {
1461                 pr_warning("Warning: could not register block events\n");
1462                 return 1;
1463         }
1464
1465         if (register_tracer(&blk_tracer) != 0) {
1466                 pr_warning("Warning: could not register the block tracer\n");
1467                 unregister_ftrace_event(&trace_blk_event);
1468                 return 1;
1469         }
1470
1471         return 0;
1472 }
1473
1474 device_initcall(init_blk_tracer);
1475
1476 static int blk_trace_remove_queue(struct request_queue *q)
1477 {
1478         struct blk_trace *bt;
1479
1480         bt = xchg(&q->blk_trace, NULL);
1481         if (bt == NULL)
1482                 return -EINVAL;
1483
1484         if (atomic_dec_and_test(&blk_probes_ref))
1485                 blk_unregister_tracepoints();
1486
1487         blk_trace_free(bt);
1488         return 0;
1489 }
1490
1491 /*
1492  * Setup everything required to start tracing
1493  */
1494 static int blk_trace_setup_queue(struct request_queue *q,
1495                                  struct block_device *bdev)
1496 {
1497         struct blk_trace *old_bt, *bt = NULL;
1498         int ret = -ENOMEM;
1499
1500         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1501         if (!bt)
1502                 return -ENOMEM;
1503
1504         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1505         if (!bt->msg_data)
1506                 goto free_bt;
1507
1508         bt->dev = bdev->bd_dev;
1509         bt->act_mask = (u16)-1;
1510
1511         blk_trace_setup_lba(bt, bdev);
1512
1513         old_bt = xchg(&q->blk_trace, bt);
1514         if (old_bt != NULL) {
1515                 (void)xchg(&q->blk_trace, old_bt);
1516                 ret = -EBUSY;
1517                 goto free_bt;
1518         }
1519
1520         if (atomic_inc_return(&blk_probes_ref) == 1)
1521                 blk_register_tracepoints();
1522         return 0;
1523
1524 free_bt:
1525         blk_trace_free(bt);
1526         return ret;
1527 }
1528
1529 /*
1530  * sysfs interface to enable and configure tracing
1531  */
1532
1533 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1534                                          struct device_attribute *attr,
1535                                          char *buf);
1536 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1537                                           struct device_attribute *attr,
1538                                           const char *buf, size_t count);
1539 #define BLK_TRACE_DEVICE_ATTR(_name) \
1540         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1541                     sysfs_blk_trace_attr_show, \
1542                     sysfs_blk_trace_attr_store)
1543
1544 static BLK_TRACE_DEVICE_ATTR(enable);
1545 static BLK_TRACE_DEVICE_ATTR(act_mask);
1546 static BLK_TRACE_DEVICE_ATTR(pid);
1547 static BLK_TRACE_DEVICE_ATTR(start_lba);
1548 static BLK_TRACE_DEVICE_ATTR(end_lba);
1549
1550 static struct attribute *blk_trace_attrs[] = {
1551         &dev_attr_enable.attr,
1552         &dev_attr_act_mask.attr,
1553         &dev_attr_pid.attr,
1554         &dev_attr_start_lba.attr,
1555         &dev_attr_end_lba.attr,
1556         NULL
1557 };
1558
1559 struct attribute_group blk_trace_attr_group = {
1560         .name  = "trace",
1561         .attrs = blk_trace_attrs,
1562 };
1563
1564 static const struct {
1565         int mask;
1566         const char *str;
1567 } mask_maps[] = {
1568         { BLK_TC_READ,          "read"          },
1569         { BLK_TC_WRITE,         "write"         },
1570         { BLK_TC_FLUSH,         "flush"         },
1571         { BLK_TC_SYNC,          "sync"          },
1572         { BLK_TC_QUEUE,         "queue"         },
1573         { BLK_TC_REQUEUE,       "requeue"       },
1574         { BLK_TC_ISSUE,         "issue"         },
1575         { BLK_TC_COMPLETE,      "complete"      },
1576         { BLK_TC_FS,            "fs"            },
1577         { BLK_TC_PC,            "pc"            },
1578         { BLK_TC_AHEAD,         "ahead"         },
1579         { BLK_TC_META,          "meta"          },
1580         { BLK_TC_DISCARD,       "discard"       },
1581         { BLK_TC_DRV_DATA,      "drv_data"      },
1582         { BLK_TC_FUA,           "fua"           },
1583 };
1584
1585 static int blk_trace_str2mask(const char *str)
1586 {
1587         int i;
1588         int mask = 0;
1589         char *buf, *s, *token;
1590
1591         buf = kstrdup(str, GFP_KERNEL);
1592         if (buf == NULL)
1593                 return -ENOMEM;
1594         s = strstrip(buf);
1595
1596         while (1) {
1597                 token = strsep(&s, ",");
1598                 if (token == NULL)
1599                         break;
1600
1601                 if (*token == '\0')
1602                         continue;
1603
1604                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1605                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1606                                 mask |= mask_maps[i].mask;
1607                                 break;
1608                         }
1609                 }
1610                 if (i == ARRAY_SIZE(mask_maps)) {
1611                         mask = -EINVAL;
1612                         break;
1613                 }
1614         }
1615         kfree(buf);
1616
1617         return mask;
1618 }
1619
1620 static ssize_t blk_trace_mask2str(char *buf, int mask)
1621 {
1622         int i;
1623         char *p = buf;
1624
1625         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1626                 if (mask & mask_maps[i].mask) {
1627                         p += sprintf(p, "%s%s",
1628                                     (p == buf) ? "" : ",", mask_maps[i].str);
1629                 }
1630         }
1631         *p++ = '\n';
1632
1633         return p - buf;
1634 }
1635
1636 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1637 {
1638         if (bdev->bd_disk == NULL)
1639                 return NULL;
1640
1641         return bdev_get_queue(bdev);
1642 }
1643
1644 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1645                                          struct device_attribute *attr,
1646                                          char *buf)
1647 {
1648         struct hd_struct *p = dev_to_part(dev);
1649         struct request_queue *q;
1650         struct block_device *bdev;
1651         ssize_t ret = -ENXIO;
1652
1653         bdev = bdget(part_devt(p));
1654         if (bdev == NULL)
1655                 goto out;
1656
1657         q = blk_trace_get_queue(bdev);
1658         if (q == NULL)
1659                 goto out_bdput;
1660
1661         mutex_lock(&bdev->bd_mutex);
1662
1663         if (attr == &dev_attr_enable) {
1664                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1665                 goto out_unlock_bdev;
1666         }
1667
1668         if (q->blk_trace == NULL)
1669                 ret = sprintf(buf, "disabled\n");
1670         else if (attr == &dev_attr_act_mask)
1671                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1672         else if (attr == &dev_attr_pid)
1673                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1674         else if (attr == &dev_attr_start_lba)
1675                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1676         else if (attr == &dev_attr_end_lba)
1677                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1678
1679 out_unlock_bdev:
1680         mutex_unlock(&bdev->bd_mutex);
1681 out_bdput:
1682         bdput(bdev);
1683 out:
1684         return ret;
1685 }
1686
1687 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1688                                           struct device_attribute *attr,
1689                                           const char *buf, size_t count)
1690 {
1691         struct block_device *bdev;
1692         struct request_queue *q;
1693         struct hd_struct *p;
1694         u64 value;
1695         ssize_t ret = -EINVAL;
1696
1697         if (count == 0)
1698                 goto out;
1699
1700         if (attr == &dev_attr_act_mask) {
1701                 if (sscanf(buf, "%llx", &value) != 1) {
1702                         /* Assume it is a list of trace category names */
1703                         ret = blk_trace_str2mask(buf);
1704                         if (ret < 0)
1705                                 goto out;
1706                         value = ret;
1707                 }
1708         } else if (sscanf(buf, "%llu", &value) != 1)
1709                 goto out;
1710
1711         ret = -ENXIO;
1712
1713         p = dev_to_part(dev);
1714         bdev = bdget(part_devt(p));
1715         if (bdev == NULL)
1716                 goto out;
1717
1718         q = blk_trace_get_queue(bdev);
1719         if (q == NULL)
1720                 goto out_bdput;
1721
1722         mutex_lock(&bdev->bd_mutex);
1723
1724         if (attr == &dev_attr_enable) {
1725                 if (value)
1726                         ret = blk_trace_setup_queue(q, bdev);
1727                 else
1728                         ret = blk_trace_remove_queue(q);
1729                 goto out_unlock_bdev;
1730         }
1731
1732         ret = 0;
1733         if (q->blk_trace == NULL)
1734                 ret = blk_trace_setup_queue(q, bdev);
1735
1736         if (ret == 0) {
1737                 if (attr == &dev_attr_act_mask)
1738                         q->blk_trace->act_mask = value;
1739                 else if (attr == &dev_attr_pid)
1740                         q->blk_trace->pid = value;
1741                 else if (attr == &dev_attr_start_lba)
1742                         q->blk_trace->start_lba = value;
1743                 else if (attr == &dev_attr_end_lba)
1744                         q->blk_trace->end_lba = value;
1745         }
1746
1747 out_unlock_bdev:
1748         mutex_unlock(&bdev->bd_mutex);
1749 out_bdput:
1750         bdput(bdev);
1751 out:
1752         return ret ? ret : count;
1753 }
1754
1755 int blk_trace_init_sysfs(struct device *dev)
1756 {
1757         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1758 }
1759
1760 void blk_trace_remove_sysfs(struct device *dev)
1761 {
1762         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1763 }
1764
1765 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1766
1767 #ifdef CONFIG_EVENT_TRACING
1768
1769 void blk_dump_cmd(char *buf, struct request *rq)
1770 {
1771         int i, end;
1772         int len = rq->cmd_len;
1773         unsigned char *cmd = rq->cmd;
1774
1775         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1776                 buf[0] = '\0';
1777                 return;
1778         }
1779
1780         for (end = len - 1; end >= 0; end--)
1781                 if (cmd[end])
1782                         break;
1783         end++;
1784
1785         for (i = 0; i < len; i++) {
1786                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1787                 if (i == end && end != len - 1) {
1788                         sprintf(buf, " ..");
1789                         break;
1790                 }
1791         }
1792 }
1793
1794 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1795 {
1796         int i = 0;
1797
1798         if (rw & REQ_FLUSH)
1799                 rwbs[i++] = 'F';
1800
1801         if (rw & WRITE)
1802                 rwbs[i++] = 'W';
1803         else if (rw & REQ_DISCARD)
1804                 rwbs[i++] = 'D';
1805         else if (bytes)
1806                 rwbs[i++] = 'R';
1807         else
1808                 rwbs[i++] = 'N';
1809
1810         if (rw & REQ_FUA)
1811                 rwbs[i++] = 'F';
1812         if (rw & REQ_RAHEAD)
1813                 rwbs[i++] = 'A';
1814         if (rw & REQ_SYNC)
1815                 rwbs[i++] = 'S';
1816         if (rw & REQ_META)
1817                 rwbs[i++] = 'M';
1818         if (rw & REQ_SECURE)
1819                 rwbs[i++] = 'E';
1820
1821         rwbs[i] = '\0';
1822 }
1823
1824 #endif /* CONFIG_EVENT_TRACING */
1825