kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <[email protected]>
   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         local_irq_save(flags);
 142         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 143         va_start(args, fmt);
 144         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 145         va_end(args);
 146
 147         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 148         local_irq_restore(flags);
 149 }
 150 EXPORT_SYMBOL_GPL(__trace_note_message);
 151
 152 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 153                          pid_t pid)
 154 {
 155         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 156                 return 1;
 157         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 158                 return 1;
 159         if (bt->pid && pid != bt->pid)
 160                 return 1;
 161
 162         return 0;
 163 }
 164
 165 /*
 166  * Data direction bit lookup
 167  */
 168 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 169                                  BLK_TC_ACT(BLK_TC_WRITE) };
 170
 171 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 172
 173 /* The ilog2() calls fall out because they're constant */
 174 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 175           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 176
 177 /*
 178  * The worker for the various blk_add_trace*() types. Fills out a
 179  * blk_io_trace structure and places it in a per-cpu subbuffer.
 180  */
 181 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 182                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 183 {
 184         struct task_struct *tsk = current;
 185         struct ring_buffer_event *event = NULL;
 186         struct ring_buffer *buffer = NULL;
 187         struct blk_io_trace *t;
 188         unsigned long flags = 0;
 189         unsigned long *sequence;
 190         pid_t pid;
 191         int cpu, pc = 0;
 192         bool blk_tracer = blk_tracer_enabled;
 193
 194         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 195                 return;
 196
 197         what |= ddir_act[rw & WRITE];
 198         what |= MASK_TC_BIT(rw, SYNC);
 199         what |= MASK_TC_BIT(rw, RAHEAD);
 200         what |= MASK_TC_BIT(rw, META);
 201         what |= MASK_TC_BIT(rw, DISCARD);
 202
 203         pid = tsk->pid;
 204         if (act_log_check(bt, what, sector, pid))
 205                 return;
 206         cpu = raw_smp_processor_id();
 207
 208         if (blk_tracer) {
 209                 tracing_record_cmdline(current);
 210
 211                 buffer = blk_tr->buffer;
 212                 pc = preempt_count();
 213                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 214                                                   sizeof(*t) + pdu_len,
 215                                                   0, pc);
 216                 if (!event)
 217                         return;
 218                 t = ring_buffer_event_data(event);
 219                 goto record_it;
 220         }
 221
 222         /*
 223          * A word about the locking here - we disable interrupts to reserve
 224          * some space in the relay per-cpu buffer, to prevent an irq
 225          * from coming in and stepping on our toes.
 226          */
 227         local_irq_save(flags);
 228
 229         if (unlikely(tsk->btrace_seq != blktrace_seq))
 230                 trace_note_tsk(bt, tsk);
 231
 232         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 233         if (t) {
 234                 sequence = per_cpu_ptr(bt->sequence, cpu);
 235
 236                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 237                 t->sequence = ++(*sequence);
 238                 t->time = ktime_to_ns(ktime_get());
 239 record_it:
 240                 /*
 241                  * These two are not needed in ftrace as they are in the
 242                  * generic trace_entry, filled by tracing_generic_entry_update,
 243                  * but for the trace_event->bin() synthesizer benefit we do it
 244                  * here too.
 245                  */
 246                 t->cpu = cpu;
 247                 t->pid = pid;
 248
 249                 t->sector = sector;
 250                 t->bytes = bytes;
 251                 t->action = what;
 252                 t->device = bt->dev;
 253                 t->error = error;
 254                 t->pdu_len = pdu_len;
 255
 256                 if (pdu_len)
 257                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 258
 259                 if (blk_tracer) {
 260                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 261                         return;
 262                 }
 263         }
 264
 265         local_irq_restore(flags);
 266 }
 267
 268 static struct dentry *blk_tree_root;
 269 static DEFINE_MUTEX(blk_tree_mutex);
 270
 271 static void blk_trace_free(struct blk_trace *bt)
 272 {
 273         debugfs_remove(bt->msg_file);
 274         debugfs_remove(bt->dropped_file);
 275         relay_close(bt->rchan);
 276         debugfs_remove(bt->dir);
 277         free_percpu(bt->sequence);
 278         free_percpu(bt->msg_data);
 279         kfree(bt);
 280 }
 281
 282 static void blk_trace_cleanup(struct blk_trace *bt)
 283 {
 284         blk_trace_free(bt);
 285         if (atomic_dec_and_test(&blk_probes_ref))
 286                 blk_unregister_tracepoints();
 287 }
 288
 289 int blk_trace_remove(struct request_queue *q)
 290 {
 291         struct blk_trace *bt;
 292
 293         bt = xchg(&q->blk_trace, NULL);
 294         if (!bt)
 295                 return -EINVAL;
 296
 297         if (bt->trace_state != Blktrace_running)
 298                 blk_trace_cleanup(bt);
 299
 300         return 0;
 301 }
 302 EXPORT_SYMBOL_GPL(blk_trace_remove);
 303
 304 static int blk_dropped_open(struct inode *inode, struct file *filp)
 305 {
 306         filp->private_data = inode->i_private;
 307
 308         return 0;
 309 }
 310
 311 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 312                                 size_t count, loff_t *ppos)
 313 {
 314         struct blk_trace *bt = filp->private_data;
 315         char buf[16];
 316
 317         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 318
 319         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 320 }
 321
 322 static const struct file_operations blk_dropped_fops = {
 323         .owner =        THIS_MODULE,
 324         .open =         blk_dropped_open,
 325         .read =         blk_dropped_read,
 326         .llseek =       default_llseek,
 327 };
 328
 329 static int blk_msg_open(struct inode *inode, struct file *filp)
 330 {
 331         filp->private_data = inode->i_private;
 332
 333         return 0;
 334 }
 335
 336 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 337                                 size_t count, loff_t *ppos)
 338 {
 339         char *msg;
 340         struct blk_trace *bt;
 341
 342         if (count >= BLK_TN_MAX_MSG)
 343                 return -EINVAL;
 344
 345         msg = kmalloc(count + 1, GFP_KERNEL);
 346         if (msg == NULL)
 347                 return -ENOMEM;
 348
 349         if (copy_from_user(msg, buffer, count)) {
 350                 kfree(msg);
 351                 return -EFAULT;
 352         }
 353
 354         msg[count] = '\0';
 355         bt = filp->private_data;
 356         __trace_note_message(bt, "%s", msg);
 357         kfree(msg);
 358
 359         return count;
 360 }
 361
 362 static const struct file_operations blk_msg_fops = {
 363         .owner =        THIS_MODULE,
 364         .open =         blk_msg_open,
 365         .write =        blk_msg_write,
 366         .llseek =       noop_llseek,
 367 };
 368
 369 /*
 370  * Keep track of how many times we encountered a full subbuffer, to aid
 371  * the user space app in telling how many lost events there were.
 372  */
 373 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 374                                      void *prev_subbuf, size_t prev_padding)
 375 {
 376         struct blk_trace *bt;
 377
 378         if (!relay_buf_full(buf))
 379                 return 1;
 380
 381         bt = buf->chan->private_data;
 382         atomic_inc(&bt->dropped);
 383         return 0;
 384 }
 385
 386 static int blk_remove_buf_file_callback(struct dentry *dentry)
 387 {
 388         debugfs_remove(dentry);
 389
 390         return 0;
 391 }
 392
 393 static struct dentry *blk_create_buf_file_callback(const char *filename,
 394                                                    struct dentry *parent,
 395                                                    int mode,
 396                                                    struct rchan_buf *buf,
 397                                                    int *is_global)
 398 {
 399         return debugfs_create_file(filename, mode, parent, buf,
 400                                         &relay_file_operations);
 401 }
 402
 403 static struct rchan_callbacks blk_relay_callbacks = {
 404         .subbuf_start           = blk_subbuf_start_callback,
 405         .create_buf_file        = blk_create_buf_file_callback,
 406         .remove_buf_file        = blk_remove_buf_file_callback,
 407 };
 408
 409 static void blk_trace_setup_lba(struct blk_trace *bt,
 410                                 struct block_device *bdev)
 411 {
 412         struct hd_struct *part = NULL;
 413
 414         if (bdev)
 415                 part = bdev->bd_part;
 416
 417         if (part) {
 418                 bt->start_lba = part->start_sect;
 419                 bt->end_lba = part->start_sect + part->nr_sects;
 420         } else {
 421                 bt->start_lba = 0;
 422                 bt->end_lba = -1ULL;
 423         }
 424 }
 425
 426 /*
 427  * Setup everything required to start tracing
 428  */
 429 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 430                        struct block_device *bdev,
 431                        struct blk_user_trace_setup *buts)
 432 {
 433         struct blk_trace *old_bt, *bt = NULL;
 434         struct dentry *dir = NULL;
 435         int ret, i;
 436
 437         if (!buts->buf_size || !buts->buf_nr)
 438                 return -EINVAL;
 439
 440         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 441         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 442
 443         /*
 444          * some device names have larger paths - convert the slashes
 445          * to underscores for this to work as expected
 446          */
 447         for (i = 0; i < strlen(buts->name); i++)
 448                 if (buts->name[i] == '/')
 449                         buts->name[i] = '_';
 450
 451         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 452         if (!bt)
 453                 return -ENOMEM;
 454
 455         ret = -ENOMEM;
 456         bt->sequence = alloc_percpu(unsigned long);
 457         if (!bt->sequence)
 458                 goto err;
 459
 460         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 461         if (!bt->msg_data)
 462                 goto err;
 463
 464         ret = -ENOENT;
 465
 466         mutex_lock(&blk_tree_mutex);
 467         if (!blk_tree_root) {
 468                 blk_tree_root = debugfs_create_dir("block", NULL);
 469                 if (!blk_tree_root) {
 470                         mutex_unlock(&blk_tree_mutex);
 471                         goto err;
 472                 }
 473         }
 474         mutex_unlock(&blk_tree_mutex);
 475
 476         dir = debugfs_create_dir(buts->name, blk_tree_root);
 477
 478         if (!dir)
 479                 goto err;
 480
 481         bt->dir = dir;
 482         bt->dev = dev;
 483         atomic_set(&bt->dropped, 0);
 484
 485         ret = -EIO;
 486         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 487                                                &blk_dropped_fops);
 488         if (!bt->dropped_file)
 489                 goto err;
 490
 491         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 492         if (!bt->msg_file)
 493                 goto err;
 494
 495         bt->rchan = relay_open("trace", dir, buts->buf_size,
 496                                 buts->buf_nr, &blk_relay_callbacks, bt);
 497         if (!bt->rchan)
 498                 goto err;
 499
 500         bt->act_mask = buts->act_mask;
 501         if (!bt->act_mask)
 502                 bt->act_mask = (u16) -1;
 503
 504         blk_trace_setup_lba(bt, bdev);
 505
 506         /* overwrite with user settings */
 507         if (buts->start_lba)
 508                 bt->start_lba = buts->start_lba;
 509         if (buts->end_lba)
 510                 bt->end_lba = buts->end_lba;
 511
 512         bt->pid = buts->pid;
 513         bt->trace_state = Blktrace_setup;
 514
 515         ret = -EBUSY;
 516         old_bt = xchg(&q->blk_trace, bt);
 517         if (old_bt) {
 518                 (void) xchg(&q->blk_trace, old_bt);
 519                 goto err;
 520         }
 521
 522         if (atomic_inc_return(&blk_probes_ref) == 1)
 523                 blk_register_tracepoints();
 524
 525         return 0;
 526 err:
 527         blk_trace_free(bt);
 528         return ret;
 529 }
 530
 531 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 532                     struct block_device *bdev,
 533                     char __user *arg)
 534 {
 535         struct blk_user_trace_setup buts;
 536         int ret;
 537
 538         ret = copy_from_user(&buts, arg, sizeof(buts));
 539         if (ret)
 540                 return -EFAULT;
 541
 542         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 543         if (ret)
 544                 return ret;
 545
 546         if (copy_to_user(arg, &buts, sizeof(buts))) {
 547                 blk_trace_remove(q);
 548                 return -EFAULT;
 549         }
 550         return 0;
 551 }
 552 EXPORT_SYMBOL_GPL(blk_trace_setup);
 553
 554 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 555 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 556                                   dev_t dev, struct block_device *bdev,
 557                                   char __user *arg)
 558 {
 559         struct blk_user_trace_setup buts;
 560         struct compat_blk_user_trace_setup cbuts;
 561         int ret;
 562
 563         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 564                 return -EFAULT;
 565
 566         buts = (struct blk_user_trace_setup) {
 567                 .act_mask = cbuts.act_mask,
 568                 .buf_size = cbuts.buf_size,
 569                 .buf_nr = cbuts.buf_nr,
 570                 .start_lba = cbuts.start_lba,
 571                 .end_lba = cbuts.end_lba,
 572                 .pid = cbuts.pid,
 573         };
 574         memcpy(&buts.name, &cbuts.name, 32);
 575
 576         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 577         if (ret)
 578                 return ret;
 579
 580         if (copy_to_user(arg, &buts.name, 32)) {
 581                 blk_trace_remove(q);
 582                 return -EFAULT;
 583         }
 584
 585         return 0;
 586 }
 587 #endif
 588
 589 int blk_trace_startstop(struct request_queue *q, int start)
 590 {
 591         int ret;
 592         struct blk_trace *bt = q->blk_trace;
 593
 594         if (bt == NULL)
 595                 return -EINVAL;
 596
 597         /*
 598          * For starting a trace, we can transition from a setup or stopped
 599          * trace. For stopping a trace, the state must be running
 600          */
 601         ret = -EINVAL;
 602         if (start) {
 603                 if (bt->trace_state == Blktrace_setup ||
 604                     bt->trace_state == Blktrace_stopped) {
 605                         blktrace_seq++;
 606                         smp_mb();
 607                         bt->trace_state = Blktrace_running;
 608
 609                         trace_note_time(bt);
 610                         ret = 0;
 611                 }
 612         } else {
 613                 if (bt->trace_state == Blktrace_running) {
 614                         bt->trace_state = Blktrace_stopped;
 615                         relay_flush(bt->rchan);
 616                         ret = 0;
 617                 }
 618         }
 619
 620         return ret;
 621 }
 622 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 623
 624 /**
 625  * blk_trace_ioctl: - handle the ioctls associated with tracing
 626  * @bdev:       the block device
 627  * @cmd:        the ioctl cmd
 628  * @arg:        the argument data, if any
 629  *
 630  **/
 631 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 632 {
 633         struct request_queue *q;
 634         int ret, start = 0;
 635         char b[BDEVNAME_SIZE];
 636
 637         q = bdev_get_queue(bdev);
 638         if (!q)
 639                 return -ENXIO;
 640
 641         mutex_lock(&bdev->bd_mutex);
 642
 643         switch (cmd) {
 644         case BLKTRACESETUP:
 645                 bdevname(bdev, b);
 646                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 647                 break;
 648 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 649         case BLKTRACESETUP32:
 650                 bdevname(bdev, b);
 651                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 652                 break;
 653 #endif
 654         case BLKTRACESTART:
 655                 start = 1;
 656         case BLKTRACESTOP:
 657                 ret = blk_trace_startstop(q, start);
 658                 break;
 659         case BLKTRACETEARDOWN:
 660                 ret = blk_trace_remove(q);
 661                 break;
 662         default:
 663                 ret = -ENOTTY;
 664                 break;
 665         }
 666
 667         mutex_unlock(&bdev->bd_mutex);
 668         return ret;
 669 }
 670
 671 /**
 672  * blk_trace_shutdown: - stop and cleanup trace structures
 673  * @q:    the request queue associated with the device
 674  *
 675  **/
 676 void blk_trace_shutdown(struct request_queue *q)
 677 {
 678         if (q->blk_trace) {
 679                 blk_trace_startstop(q, 0);
 680                 blk_trace_remove(q);
 681         }
 682 }
 683
 684 /*
 685  * blktrace probes
 686  */
 687
 688 /**
 689  * blk_add_trace_rq - Add a trace for a request oriented action
 690  * @q:          queue the io is for
 691  * @rq:         the source request
 692  * @what:       the action
 693  *
 694  * Description:
 695  *     Records an action against a request. Will log the bio offset + size.
 696  *
 697  **/
 698 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 699                                     u32 what)
 700 {
 701         struct blk_trace *bt = q->blk_trace;
 702         int rw = rq->cmd_flags & 0x03;
 703
 704         if (likely(!bt))
 705                 return;
 706
 707         if (rq->cmd_flags & REQ_DISCARD)
 708                 rw |= REQ_DISCARD;
 709
 710         if (rq->cmd_flags & REQ_SECURE)
 711                 rw |= REQ_SECURE;
 712
 713         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 714                 what |= BLK_TC_ACT(BLK_TC_PC);
 715                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
 716                                 what, rq->errors, rq->cmd_len, rq->cmd);
 717         } else  {
 718                 what |= BLK_TC_ACT(BLK_TC_FS);
 719                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
 720                                 what, rq->errors, 0, NULL);
 721         }
 722 }
 723
 724 static void blk_add_trace_rq_abort(void *ignore,
 725                                    struct request_queue *q, struct request *rq)
 726 {
 727         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 728 }
 729
 730 static void blk_add_trace_rq_insert(void *ignore,
 731                                     struct request_queue *q, struct request *rq)
 732 {
 733         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 734 }
 735
 736 static void blk_add_trace_rq_issue(void *ignore,
 737                                    struct request_queue *q, struct request *rq)
 738 {
 739         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 740 }
 741
 742 static void blk_add_trace_rq_requeue(void *ignore,
 743                                      struct request_queue *q,
 744                                      struct request *rq)
 745 {
 746         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 747 }
 748
 749 static void blk_add_trace_rq_complete(void *ignore,
 750                                       struct request_queue *q,
 751                                       struct request *rq)
 752 {
 753         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 754 }
 755
 756 /**
 757  * blk_add_trace_bio - Add a trace for a bio oriented action
 758  * @q:          queue the io is for
 759  * @bio:        the source bio
 760  * @what:       the action
 761  * @error:      error, if any
 762  *
 763  * Description:
 764  *     Records an action against a bio. Will log the bio offset + size.
 765  *
 766  **/
 767 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 768                               u32 what, int error)
 769 {
 770         struct blk_trace *bt = q->blk_trace;
 771
 772         if (likely(!bt))
 773                 return;
 774
 775         if (!error && !bio_flagged(bio, BIO_UPTODATE))
 776                 error = EIO;
 777
 778         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 779                         error, 0, NULL);
 780 }
 781
 782 static void blk_add_trace_bio_bounce(void *ignore,
 783                                      struct request_queue *q, struct bio *bio)
 784 {
 785         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 786 }
 787
 788 static void blk_add_trace_bio_complete(void *ignore,
 789                                        struct request_queue *q, struct bio *bio,
 790                                        int error)
 791 {
 792         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 793 }
 794
 795 static void blk_add_trace_bio_backmerge(void *ignore,
 796                                         struct request_queue *q,
 797                                         struct bio *bio)
 798 {
 799         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 800 }
 801
 802 static void blk_add_trace_bio_frontmerge(void *ignore,
 803                                          struct request_queue *q,
 804                                          struct bio *bio)
 805 {
 806         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 807 }
 808
 809 static void blk_add_trace_bio_queue(void *ignore,
 810                                     struct request_queue *q, struct bio *bio)
 811 {
 812         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 813 }
 814
 815 static void blk_add_trace_getrq(void *ignore,
 816                                 struct request_queue *q,
 817                                 struct bio *bio, int rw)
 818 {
 819         if (bio)
 820                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 821         else {
 822                 struct blk_trace *bt = q->blk_trace;
 823
 824                 if (bt)
 825                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 826         }
 827 }
 828
 829
 830 static void blk_add_trace_sleeprq(void *ignore,
 831                                   struct request_queue *q,
 832                                   struct bio *bio, int rw)
 833 {
 834         if (bio)
 835                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 836         else {
 837                 struct blk_trace *bt = q->blk_trace;
 838
 839                 if (bt)
 840                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 841                                         0, 0, NULL);
 842         }
 843 }
 844
 845 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 846 {
 847         struct blk_trace *bt = q->blk_trace;
 848
 849         if (bt)
 850                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 851 }
 852
 853 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
 854 {
 855         struct blk_trace *bt = q->blk_trace;
 856
 857         if (bt) {
 858                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 859                 __be64 rpdu = cpu_to_be64(pdu);
 860
 861                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 862                                 sizeof(rpdu), &rpdu);
 863         }
 864 }
 865
 866 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
 867 {
 868         struct blk_trace *bt = q->blk_trace;
 869
 870         if (bt) {
 871                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 872                 __be64 rpdu = cpu_to_be64(pdu);
 873
 874                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 875                                 sizeof(rpdu), &rpdu);
 876         }
 877 }
 878
 879 static void blk_add_trace_split(void *ignore,
 880                                 struct request_queue *q, struct bio *bio,
 881                                 unsigned int pdu)
 882 {
 883         struct blk_trace *bt = q->blk_trace;
 884
 885         if (bt) {
 886                 __be64 rpdu = cpu_to_be64(pdu);
 887
 888                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 889                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 890                                 sizeof(rpdu), &rpdu);
 891         }
 892 }
 893
 894 /**
 895  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 896  * @ignore:     trace callback data parameter (not used)
 897  * @q:          queue the io is for
 898  * @bio:        the source bio
 899  * @dev:        target device
 900  * @from:       source sector
 901  *
 902  * Description:
 903  *     Device mapper or raid target sometimes need to split a bio because
 904  *     it spans a stripe (or similar). Add a trace for that action.
 905  *
 906  **/
 907 static void blk_add_trace_bio_remap(void *ignore,
 908                                     struct request_queue *q, struct bio *bio,
 909                                     dev_t dev, sector_t from)
 910 {
 911         struct blk_trace *bt = q->blk_trace;
 912         struct blk_io_trace_remap r;
 913
 914         if (likely(!bt))
 915                 return;
 916
 917         r.device_from = cpu_to_be32(dev);
 918         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 919         r.sector_from = cpu_to_be64(from);
 920
 921         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 922                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 923                         sizeof(r), &r);
 924 }
 925
 926 /**
 927  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 928  * @ignore:     trace callback data parameter (not used)
 929  * @q:          queue the io is for
 930  * @rq:         the source request
 931  * @dev:        target device
 932  * @from:       source sector
 933  *
 934  * Description:
 935  *     Device mapper remaps request to other devices.
 936  *     Add a trace for that action.
 937  *
 938  **/
 939 static void blk_add_trace_rq_remap(void *ignore,
 940                                    struct request_queue *q,
 941                                    struct request *rq, dev_t dev,
 942                                    sector_t from)
 943 {
 944         struct blk_trace *bt = q->blk_trace;
 945         struct blk_io_trace_remap r;
 946
 947         if (likely(!bt))
 948                 return;
 949
 950         r.device_from = cpu_to_be32(dev);
 951         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 952         r.sector_from = cpu_to_be64(from);
 953
 954         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 955                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 956                         sizeof(r), &r);
 957 }
 958
 959 /**
 960  * blk_add_driver_data - Add binary message with driver-specific data
 961  * @q:          queue the io is for
 962  * @rq:         io request
 963  * @data:       driver-specific data
 964  * @len:        length of driver-specific data
 965  *
 966  * Description:
 967  *     Some drivers might want to write driver-specific data per request.
 968  *
 969  **/
 970 void blk_add_driver_data(struct request_queue *q,
 971                          struct request *rq,
 972                          void *data, size_t len)
 973 {
 974         struct blk_trace *bt = q->blk_trace;
 975
 976         if (likely(!bt))
 977                 return;
 978
 979         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 980                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 981                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 982         else
 983                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 984                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 985 }
 986 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 987
 988 static void blk_register_tracepoints(void)
 989 {
 990         int ret;
 991
 992         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 993         WARN_ON(ret);
 994         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 995         WARN_ON(ret);
 996         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 997         WARN_ON(ret);
 998         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 999         WARN_ON(ret);
1000         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1001         WARN_ON(ret);
1002         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1003         WARN_ON(ret);
1004         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1005         WARN_ON(ret);
1006         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1007         WARN_ON(ret);
1008         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1009         WARN_ON(ret);
1010         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1011         WARN_ON(ret);
1012         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1013         WARN_ON(ret);
1014         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1015         WARN_ON(ret);
1016         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1017         WARN_ON(ret);
1018         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1019         WARN_ON(ret);
1020         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1021         WARN_ON(ret);
1022         ret = register_trace_block_split(blk_add_trace_split, NULL);
1023         WARN_ON(ret);
1024         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1025         WARN_ON(ret);
1026         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1027         WARN_ON(ret);
1028 }
1029
1030 static void blk_unregister_tracepoints(void)
1031 {
1032         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1033         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1034         unregister_trace_block_split(blk_add_trace_split, NULL);
1035         unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1036         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1037         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1038         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1039         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1040         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1041         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1042         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1043         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1044         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1045         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1046         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1047         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1048         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1049         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1050
1051         tracepoint_synchronize_unregister();
1052 }
1053
1054 /*
1055  * struct blk_io_tracer formatting routines
1056  */
1057
1058 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1059 {
1060         int i = 0;
1061         int tc = t->action >> BLK_TC_SHIFT;
1062
1063         if (t->action == BLK_TN_MESSAGE) {
1064                 rwbs[i++] = 'N';
1065                 goto out;
1066         }
1067
1068         if (tc & BLK_TC_DISCARD)
1069                 rwbs[i++] = 'D';
1070         else if (tc & BLK_TC_WRITE)
1071                 rwbs[i++] = 'W';
1072         else if (t->bytes)
1073                 rwbs[i++] = 'R';
1074         else
1075                 rwbs[i++] = 'N';
1076
1077         if (tc & BLK_TC_AHEAD)
1078                 rwbs[i++] = 'A';
1079         if (tc & BLK_TC_BARRIER)
1080                 rwbs[i++] = 'B';
1081         if (tc & BLK_TC_SYNC)
1082                 rwbs[i++] = 'S';
1083         if (tc & BLK_TC_META)
1084                 rwbs[i++] = 'M';
1085 out:
1086         rwbs[i] = '\0';
1087 }
1088
1089 static inline
1090 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1091 {
1092         return (const struct blk_io_trace *)ent;
1093 }
1094
1095 static inline const void *pdu_start(const struct trace_entry *ent)
1096 {
1097         return te_blk_io_trace(ent) + 1;
1098 }
1099
1100 static inline u32 t_action(const struct trace_entry *ent)
1101 {
1102         return te_blk_io_trace(ent)->action;
1103 }
1104
1105 static inline u32 t_bytes(const struct trace_entry *ent)
1106 {
1107         return te_blk_io_trace(ent)->bytes;
1108 }
1109
1110 static inline u32 t_sec(const struct trace_entry *ent)
1111 {
1112         return te_blk_io_trace(ent)->bytes >> 9;
1113 }
1114
1115 static inline unsigned long long t_sector(const struct trace_entry *ent)
1116 {
1117         return te_blk_io_trace(ent)->sector;
1118 }
1119
1120 static inline __u16 t_error(const struct trace_entry *ent)
1121 {
1122         return te_blk_io_trace(ent)->error;
1123 }
1124
1125 static __u64 get_pdu_int(const struct trace_entry *ent)
1126 {
1127         const __u64 *val = pdu_start(ent);
1128         return be64_to_cpu(*val);
1129 }
1130
1131 static void get_pdu_remap(const struct trace_entry *ent,
1132                           struct blk_io_trace_remap *r)
1133 {
1134         const struct blk_io_trace_remap *__r = pdu_start(ent);
1135         __u64 sector_from = __r->sector_from;
1136
1137         r->device_from = be32_to_cpu(__r->device_from);
1138         r->device_to   = be32_to_cpu(__r->device_to);
1139         r->sector_from = be64_to_cpu(sector_from);
1140 }
1141
1142 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1143
1144 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1145 {
1146         char rwbs[6];
1147         unsigned long long ts  = iter->ts;
1148         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1149         unsigned secs          = (unsigned long)ts;
1150         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1151
1152         fill_rwbs(rwbs, t);
1153
1154         return trace_seq_printf(&iter->seq,
1155                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1156                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1157                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1158 }
1159
1160 static int blk_log_action(struct trace_iterator *iter, const char *act)
1161 {
1162         char rwbs[6];
1163         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1164
1165         fill_rwbs(rwbs, t);
1166         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1167                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1168 }
1169
1170 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1171 {
1172         const unsigned char *pdu_buf;
1173         int pdu_len;
1174         int i, end, ret;
1175
1176         pdu_buf = pdu_start(ent);
1177         pdu_len = te_blk_io_trace(ent)->pdu_len;
1178
1179         if (!pdu_len)
1180                 return 1;
1181
1182         /* find the last zero that needs to be printed */
1183         for (end = pdu_len - 1; end >= 0; end--)
1184                 if (pdu_buf[end])
1185                         break;
1186         end++;
1187
1188         if (!trace_seq_putc(s, '('))
1189                 return 0;
1190
1191         for (i = 0; i < pdu_len; i++) {
1192
1193                 ret = trace_seq_printf(s, "%s%02x",
1194                                        i == 0 ? "" : " ", pdu_buf[i]);
1195                 if (!ret)
1196                         return ret;
1197
1198                 /*
1199                  * stop when the rest is just zeroes and indicate so
1200                  * with a ".." appended
1201                  */
1202                 if (i == end && end != pdu_len - 1)
1203                         return trace_seq_puts(s, " ..) ");
1204         }
1205
1206         return trace_seq_puts(s, ") ");
1207 }
1208
1209 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1210 {
1211         char cmd[TASK_COMM_LEN];
1212
1213         trace_find_cmdline(ent->pid, cmd);
1214
1215         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1216                 int ret;
1217
1218                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1219                 if (!ret)
1220                         return 0;
1221                 ret = blk_log_dump_pdu(s, ent);
1222                 if (!ret)
1223                         return 0;
1224                 return trace_seq_printf(s, "[%s]\n", cmd);
1225         } else {
1226                 if (t_sec(ent))
1227                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1228                                                 t_sector(ent), t_sec(ent), cmd);
1229                 return trace_seq_printf(s, "[%s]\n", cmd);
1230         }
1231 }
1232
1233 static int blk_log_with_error(struct trace_seq *s,
1234                               const struct trace_entry *ent)
1235 {
1236         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1237                 int ret;
1238
1239                 ret = blk_log_dump_pdu(s, ent);
1240                 if (ret)
1241                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1242                 return 0;
1243         } else {
1244                 if (t_sec(ent))
1245                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1246                                                 t_sector(ent),
1247                                                 t_sec(ent), t_error(ent));
1248                 return trace_seq_printf(s, "%llu [%d]\n",
1249                                         t_sector(ent), t_error(ent));
1250         }
1251 }
1252
1253 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1254 {
1255         struct blk_io_trace_remap r = { .device_from = 0, };
1256
1257         get_pdu_remap(ent, &r);
1258         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1259                                 t_sector(ent), t_sec(ent),
1260                                 MAJOR(r.device_from), MINOR(r.device_from),
1261                                 (unsigned long long)r.sector_from);
1262 }
1263
1264 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1265 {
1266         char cmd[TASK_COMM_LEN];
1267
1268         trace_find_cmdline(ent->pid, cmd);
1269
1270         return trace_seq_printf(s, "[%s]\n", cmd);
1271 }
1272
1273 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1274 {
1275         char cmd[TASK_COMM_LEN];
1276
1277         trace_find_cmdline(ent->pid, cmd);
1278
1279         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1280 }
1281
1282 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1283 {
1284         char cmd[TASK_COMM_LEN];
1285
1286         trace_find_cmdline(ent->pid, cmd);
1287
1288         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1289                                 get_pdu_int(ent), cmd);
1290 }
1291
1292 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1293 {
1294         int ret;
1295         const struct blk_io_trace *t = te_blk_io_trace(ent);
1296
1297         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1298         if (ret)
1299                 return trace_seq_putc(s, '\n');
1300         return ret;
1301 }
1302
1303 /*
1304  * struct tracer operations
1305  */
1306
1307 static void blk_tracer_print_header(struct seq_file *m)
1308 {
1309         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1310                 return;
1311         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1312                     "#  |     |     |           |   |   |\n");
1313 }
1314
1315 static void blk_tracer_start(struct trace_array *tr)
1316 {
1317         blk_tracer_enabled = true;
1318 }
1319
1320 static int blk_tracer_init(struct trace_array *tr)
1321 {
1322         blk_tr = tr;
1323         blk_tracer_start(tr);
1324         return 0;
1325 }
1326
1327 static void blk_tracer_stop(struct trace_array *tr)
1328 {
1329         blk_tracer_enabled = false;
1330 }
1331
1332 static void blk_tracer_reset(struct trace_array *tr)
1333 {
1334         blk_tracer_stop(tr);
1335 }
1336
1337 static const struct {
1338         const char *act[2];
1339         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1340 } what2act[] = {
1341         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1342         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1343         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1344         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1345         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1346         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1347         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1348         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1349         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1350         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1351         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1352         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1353         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1354         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1355         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1356 };
1357
1358 static enum print_line_t print_one_line(struct trace_iterator *iter,
1359                                         bool classic)
1360 {
1361         struct trace_seq *s = &iter->seq;
1362         const struct blk_io_trace *t;
1363         u16 what;
1364         int ret;
1365         bool long_act;
1366         blk_log_action_t *log_action;
1367
1368         t          = te_blk_io_trace(iter->ent);
1369         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1370         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1371         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1372
1373         if (t->action == BLK_TN_MESSAGE) {
1374                 ret = log_action(iter, long_act ? "message" : "m");
1375                 if (ret)
1376                         ret = blk_log_msg(s, iter->ent);
1377                 goto out;
1378         }
1379
1380         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1381                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1382         else {
1383                 ret = log_action(iter, what2act[what].act[long_act]);
1384                 if (ret)
1385                         ret = what2act[what].print(s, iter->ent);
1386         }
1387 out:
1388         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1389 }
1390
1391 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1392                                                int flags, struct trace_event *event)
1393 {
1394         return print_one_line(iter, false);
1395 }
1396
1397 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1398 {
1399         struct trace_seq *s = &iter->seq;
1400         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1401         const int offset = offsetof(struct blk_io_trace, sector);
1402         struct blk_io_trace old = {
1403                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1404                 .time     = iter->ts,
1405         };
1406
1407         if (!trace_seq_putmem(s, &old, offset))
1408                 return 0;
1409         return trace_seq_putmem(s, &t->sector,
1410                                 sizeof(old) - offset + t->pdu_len);
1411 }
1412
1413 static enum print_line_t
1414 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1415                              struct trace_event *event)
1416 {
1417         return blk_trace_synthesize_old_trace(iter) ?
1418                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1419 }
1420
1421 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1422 {
1423         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1424                 return TRACE_TYPE_UNHANDLED;
1425
1426         return print_one_line(iter, true);
1427 }
1428
1429 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1430 {
1431         /* don't output context-info for blk_classic output */
1432         if (bit == TRACE_BLK_OPT_CLASSIC) {
1433                 if (set)
1434                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1435                 else
1436                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1437         }
1438         return 0;
1439 }
1440
1441 static struct tracer blk_tracer __read_mostly = {
1442         .name           = "blk",
1443         .init           = blk_tracer_init,
1444         .reset          = blk_tracer_reset,
1445         .start          = blk_tracer_start,
1446         .stop           = blk_tracer_stop,
1447         .print_header   = blk_tracer_print_header,
1448         .print_line     = blk_tracer_print_line,
1449         .flags          = &blk_tracer_flags,
1450         .set_flag       = blk_tracer_set_flag,
1451 };
1452
1453 static struct trace_event_functions trace_blk_event_funcs = {
1454         .trace          = blk_trace_event_print,
1455         .binary         = blk_trace_event_print_binary,
1456 };
1457
1458 static struct trace_event trace_blk_event = {
1459         .type           = TRACE_BLK,
1460         .funcs          = &trace_blk_event_funcs,
1461 };
1462
1463 static int __init init_blk_tracer(void)
1464 {
1465         if (!register_ftrace_event(&trace_blk_event)) {
1466                 pr_warning("Warning: could not register block events\n");
1467                 return 1;
1468         }
1469
1470         if (register_tracer(&blk_tracer) != 0) {
1471                 pr_warning("Warning: could not register the block tracer\n");
1472                 unregister_ftrace_event(&trace_blk_event);
1473                 return 1;
1474         }
1475
1476         return 0;
1477 }
1478
1479 device_initcall(init_blk_tracer);
1480
1481 static int blk_trace_remove_queue(struct request_queue *q)
1482 {
1483         struct blk_trace *bt;
1484
1485         bt = xchg(&q->blk_trace, NULL);
1486         if (bt == NULL)
1487                 return -EINVAL;
1488
1489         if (atomic_dec_and_test(&blk_probes_ref))
1490                 blk_unregister_tracepoints();
1491
1492         blk_trace_free(bt);
1493         return 0;
1494 }
1495
1496 /*
1497  * Setup everything required to start tracing
1498  */
1499 static int blk_trace_setup_queue(struct request_queue *q,
1500                                  struct block_device *bdev)
1501 {
1502         struct blk_trace *old_bt, *bt = NULL;
1503         int ret = -ENOMEM;
1504
1505         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1506         if (!bt)
1507                 return -ENOMEM;
1508
1509         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1510         if (!bt->msg_data)
1511                 goto free_bt;
1512
1513         bt->dev = bdev->bd_dev;
1514         bt->act_mask = (u16)-1;
1515
1516         blk_trace_setup_lba(bt, bdev);
1517
1518         old_bt = xchg(&q->blk_trace, bt);
1519         if (old_bt != NULL) {
1520                 (void)xchg(&q->blk_trace, old_bt);
1521                 ret = -EBUSY;
1522                 goto free_bt;
1523         }
1524
1525         if (atomic_inc_return(&blk_probes_ref) == 1)
1526                 blk_register_tracepoints();
1527         return 0;
1528
1529 free_bt:
1530         blk_trace_free(bt);
1531         return ret;
1532 }
1533
1534 /*
1535  * sysfs interface to enable and configure tracing
1536  */
1537
1538 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1539                                          struct device_attribute *attr,
1540                                          char *buf);
1541 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1542                                           struct device_attribute *attr,
1543                                           const char *buf, size_t count);
1544 #define BLK_TRACE_DEVICE_ATTR(_name) \
1545         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1546                     sysfs_blk_trace_attr_show, \
1547                     sysfs_blk_trace_attr_store)
1548
1549 static BLK_TRACE_DEVICE_ATTR(enable);
1550 static BLK_TRACE_DEVICE_ATTR(act_mask);
1551 static BLK_TRACE_DEVICE_ATTR(pid);
1552 static BLK_TRACE_DEVICE_ATTR(start_lba);
1553 static BLK_TRACE_DEVICE_ATTR(end_lba);
1554
1555 static struct attribute *blk_trace_attrs[] = {
1556         &dev_attr_enable.attr,
1557         &dev_attr_act_mask.attr,
1558         &dev_attr_pid.attr,
1559         &dev_attr_start_lba.attr,
1560         &dev_attr_end_lba.attr,
1561         NULL
1562 };
1563
1564 struct attribute_group blk_trace_attr_group = {
1565         .name  = "trace",
1566         .attrs = blk_trace_attrs,
1567 };
1568
1569 static const struct {
1570         int mask;
1571         const char *str;
1572 } mask_maps[] = {
1573         { BLK_TC_READ,          "read"          },
1574         { BLK_TC_WRITE,         "write"         },
1575         { BLK_TC_BARRIER,       "barrier"       },
1576         { BLK_TC_SYNC,          "sync"          },
1577         { BLK_TC_QUEUE,         "queue"         },
1578         { BLK_TC_REQUEUE,       "requeue"       },
1579         { BLK_TC_ISSUE,         "issue"         },
1580         { BLK_TC_COMPLETE,      "complete"      },
1581         { BLK_TC_FS,            "fs"            },
1582         { BLK_TC_PC,            "pc"            },
1583         { BLK_TC_AHEAD,         "ahead"         },
1584         { BLK_TC_META,          "meta"          },
1585         { BLK_TC_DISCARD,       "discard"       },
1586         { BLK_TC_DRV_DATA,      "drv_data"      },
1587 };
1588
1589 static int blk_trace_str2mask(const char *str)
1590 {
1591         int i;
1592         int mask = 0;
1593         char *buf, *s, *token;
1594
1595         buf = kstrdup(str, GFP_KERNEL);
1596         if (buf == NULL)
1597                 return -ENOMEM;
1598         s = strstrip(buf);
1599
1600         while (1) {
1601                 token = strsep(&s, ",");
1602                 if (token == NULL)
1603                         break;
1604
1605                 if (*token == '\0')
1606                         continue;
1607
1608                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1609                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1610                                 mask |= mask_maps[i].mask;
1611                                 break;
1612                         }
1613                 }
1614                 if (i == ARRAY_SIZE(mask_maps)) {
1615                         mask = -EINVAL;
1616                         break;
1617                 }
1618         }
1619         kfree(buf);
1620
1621         return mask;
1622 }
1623
1624 static ssize_t blk_trace_mask2str(char *buf, int mask)
1625 {
1626         int i;
1627         char *p = buf;
1628
1629         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1630                 if (mask & mask_maps[i].mask) {
1631                         p += sprintf(p, "%s%s",
1632                                     (p == buf) ? "" : ",", mask_maps[i].str);
1633                 }
1634         }
1635         *p++ = '\n';
1636
1637         return p - buf;
1638 }
1639
1640 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1641 {
1642         if (bdev->bd_disk == NULL)
1643                 return NULL;
1644
1645         return bdev_get_queue(bdev);
1646 }
1647
1648 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1649                                          struct device_attribute *attr,
1650                                          char *buf)
1651 {
1652         struct hd_struct *p = dev_to_part(dev);
1653         struct request_queue *q;
1654         struct block_device *bdev;
1655         ssize_t ret = -ENXIO;
1656
1657         bdev = bdget(part_devt(p));
1658         if (bdev == NULL)
1659                 goto out;
1660
1661         q = blk_trace_get_queue(bdev);
1662         if (q == NULL)
1663                 goto out_bdput;
1664
1665         mutex_lock(&bdev->bd_mutex);
1666
1667         if (attr == &dev_attr_enable) {
1668                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1669                 goto out_unlock_bdev;
1670         }
1671
1672         if (q->blk_trace == NULL)
1673                 ret = sprintf(buf, "disabled\n");
1674         else if (attr == &dev_attr_act_mask)
1675                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1676         else if (attr == &dev_attr_pid)
1677                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1678         else if (attr == &dev_attr_start_lba)
1679                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1680         else if (attr == &dev_attr_end_lba)
1681                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1682
1683 out_unlock_bdev:
1684         mutex_unlock(&bdev->bd_mutex);
1685 out_bdput:
1686         bdput(bdev);
1687 out:
1688         return ret;
1689 }
1690
1691 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1692                                           struct device_attribute *attr,
1693                                           const char *buf, size_t count)
1694 {
1695         struct block_device *bdev;
1696         struct request_queue *q;
1697         struct hd_struct *p;
1698         u64 value;
1699         ssize_t ret = -EINVAL;
1700
1701         if (count == 0)
1702                 goto out;
1703
1704         if (attr == &dev_attr_act_mask) {
1705                 if (sscanf(buf, "%llx", &value) != 1) {
1706                         /* Assume it is a list of trace category names */
1707                         ret = blk_trace_str2mask(buf);
1708                         if (ret < 0)
1709                                 goto out;
1710                         value = ret;
1711                 }
1712         } else if (sscanf(buf, "%llu", &value) != 1)
1713                 goto out;
1714
1715         ret = -ENXIO;
1716
1717         p = dev_to_part(dev);
1718         bdev = bdget(part_devt(p));
1719         if (bdev == NULL)
1720                 goto out;
1721
1722         q = blk_trace_get_queue(bdev);
1723         if (q == NULL)
1724                 goto out_bdput;
1725
1726         mutex_lock(&bdev->bd_mutex);
1727
1728         if (attr == &dev_attr_enable) {
1729                 if (value)
1730                         ret = blk_trace_setup_queue(q, bdev);
1731                 else
1732                         ret = blk_trace_remove_queue(q);
1733                 goto out_unlock_bdev;
1734         }
1735
1736         ret = 0;
1737         if (q->blk_trace == NULL)
1738                 ret = blk_trace_setup_queue(q, bdev);
1739
1740         if (ret == 0) {
1741                 if (attr == &dev_attr_act_mask)
1742                         q->blk_trace->act_mask = value;
1743                 else if (attr == &dev_attr_pid)
1744                         q->blk_trace->pid = value;
1745                 else if (attr == &dev_attr_start_lba)
1746                         q->blk_trace->start_lba = value;
1747                 else if (attr == &dev_attr_end_lba)
1748                         q->blk_trace->end_lba = value;
1749         }
1750
1751 out_unlock_bdev:
1752         mutex_unlock(&bdev->bd_mutex);
1753 out_bdput:
1754         bdput(bdev);
1755 out:
1756         return ret ? ret : count;
1757 }
1758
1759 int blk_trace_init_sysfs(struct device *dev)
1760 {
1761         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1762 }
1763
1764 void blk_trace_remove_sysfs(struct device *dev)
1765 {
1766         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1767 }
1768
1769 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1770
1771 #ifdef CONFIG_EVENT_TRACING
1772
1773 void blk_dump_cmd(char *buf, struct request *rq)
1774 {
1775         int i, end;
1776         int len = rq->cmd_len;
1777         unsigned char *cmd = rq->cmd;
1778
1779         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1780                 buf[0] = '\0';
1781                 return;
1782         }
1783
1784         for (end = len - 1; end >= 0; end--)
1785                 if (cmd[end])
1786                         break;
1787         end++;
1788
1789         for (i = 0; i < len; i++) {
1790                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1791                 if (i == end && end != len - 1) {
1792                         sprintf(buf, " ..");
1793                         break;
1794                 }
1795         }
1796 }
1797
1798 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1799 {
1800         int i = 0;
1801
1802         if (rw & WRITE)
1803                 rwbs[i++] = 'W';
1804         else if (rw & REQ_DISCARD)
1805                 rwbs[i++] = 'D';
1806         else if (bytes)
1807                 rwbs[i++] = 'R';
1808         else
1809                 rwbs[i++] = 'N';
1810
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 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1824 {
1825         int rw = rq->cmd_flags & 0x03;
1826         int bytes;
1827
1828         if (rq->cmd_flags & REQ_DISCARD)
1829                 rw |= REQ_DISCARD;
1830
1831         if (rq->cmd_flags & REQ_SECURE)
1832                 rw |= REQ_SECURE;
1833
1834         bytes = blk_rq_bytes(rq);
1835
1836         blk_fill_rwbs(rwbs, rw, bytes);
1837 }
1838
1839 #endif /* CONFIG_EVENT_TRACING */
1840