kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/time.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <trace/events/block.h>
  30
  31 #include "trace_output.h"
  32
  33 #ifdef CONFIG_BLK_DEV_IO_TRACE
  34
  35 static unsigned int blktrace_seq __read_mostly = 1;
  36
  37 static struct trace_array *blk_tr;
  38 static bool blk_tracer_enabled __read_mostly;
  39
  40 /* Select an alternative, minimalistic output than the original one */
  41 #define TRACE_BLK_OPT_CLASSIC   0x1
  42
  43 static struct tracer_opt blk_tracer_opts[] = {
  44         /* Default disable the minimalistic output */
  45         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  46         { }
  47 };
  48
  49 static struct tracer_flags blk_tracer_flags = {
  50         .val  = 0,
  51         .opts = blk_tracer_opts,
  52 };
  53
  54 /* Global reference count of probes */
  55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  56
  57 static void blk_register_tracepoints(void);
  58 static void blk_unregister_tracepoints(void);
  59
  60 /*
  61  * Send out a notify message.
  62  */
  63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  64                        const void *data, size_t len)
  65 {
  66         struct blk_io_trace *t;
  67         struct ring_buffer_event *event = NULL;
  68         struct ring_buffer *buffer = NULL;
  69         int pc = 0;
  70         int cpu = smp_processor_id();
  71         bool blk_tracer = blk_tracer_enabled;
  72
  73         if (blk_tracer) {
  74                 buffer = blk_tr->buffer;
  75                 pc = preempt_count();
  76                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  77                                                   sizeof(*t) + len,
  78                                                   0, pc);
  79                 if (!event)
  80                         return;
  81                 t = ring_buffer_event_data(event);
  82                 goto record_it;
  83         }
  84
  85         if (!bt->rchan)
  86                 return;
  87
  88         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  89         if (t) {
  90                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  91                 t->time = ktime_to_ns(ktime_get());
  92 record_it:
  93                 t->device = bt->dev;
  94                 t->action = action;
  95                 t->pid = pid;
  96                 t->cpu = cpu;
  97                 t->pdu_len = len;
  98                 memcpy((void *) t + sizeof(*t), data, len);
  99
 100                 if (blk_tracer)
 101                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 102         }
 103 }
 104
 105 /*
 106  * Send out a notify for this process, if we haven't done so since a trace
 107  * started
 108  */
 109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 110 {
 111         tsk->btrace_seq = blktrace_seq;
 112         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 113 }
 114
 115 static void trace_note_time(struct blk_trace *bt)
 116 {
 117         struct timespec now;
 118         unsigned long flags;
 119         u32 words[2];
 120
 121         getnstimeofday(&now);
 122         words[0] = now.tv_sec;
 123         words[1] = now.tv_nsec;
 124
 125         local_irq_save(flags);
 126         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 127         local_irq_restore(flags);
 128 }
 129
 130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 131 {
 132         int n;
 133         va_list args;
 134         unsigned long flags;
 135         char *buf;
 136
 137         if (unlikely(bt->trace_state != Blktrace_running &&
 138                      !blk_tracer_enabled))
 139                 return;
 140
 141         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  *
 762  * Description:
 763  *     Records an action against a bio. Will log the bio offset + size.
 764  *
 765  **/
 766 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 767                                      u32 what)
 768 {
 769         struct blk_trace *bt = q->blk_trace;
 770
 771         if (likely(!bt))
 772                 return;
 773
 774         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 775                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 776 }
 777
 778 static void blk_add_trace_bio_bounce(void *ignore,
 779                                      struct request_queue *q, struct bio *bio)
 780 {
 781         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 782 }
 783
 784 static void blk_add_trace_bio_complete(void *ignore,
 785                                        struct request_queue *q, struct bio *bio)
 786 {
 787         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 788 }
 789
 790 static void blk_add_trace_bio_backmerge(void *ignore,
 791                                         struct request_queue *q,
 792                                         struct bio *bio)
 793 {
 794         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 795 }
 796
 797 static void blk_add_trace_bio_frontmerge(void *ignore,
 798                                          struct request_queue *q,
 799                                          struct bio *bio)
 800 {
 801         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 802 }
 803
 804 static void blk_add_trace_bio_queue(void *ignore,
 805                                     struct request_queue *q, struct bio *bio)
 806 {
 807         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 808 }
 809
 810 static void blk_add_trace_getrq(void *ignore,
 811                                 struct request_queue *q,
 812                                 struct bio *bio, int rw)
 813 {
 814         if (bio)
 815                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 816         else {
 817                 struct blk_trace *bt = q->blk_trace;
 818
 819                 if (bt)
 820                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 821         }
 822 }
 823
 824
 825 static void blk_add_trace_sleeprq(void *ignore,
 826                                   struct request_queue *q,
 827                                   struct bio *bio, int rw)
 828 {
 829         if (bio)
 830                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 831         else {
 832                 struct blk_trace *bt = q->blk_trace;
 833
 834                 if (bt)
 835                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 836                                         0, 0, NULL);
 837         }
 838 }
 839
 840 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 841 {
 842         struct blk_trace *bt = q->blk_trace;
 843
 844         if (bt)
 845                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 846 }
 847
 848 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
 849 {
 850         struct blk_trace *bt = q->blk_trace;
 851
 852         if (bt) {
 853                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 854                 __be64 rpdu = cpu_to_be64(pdu);
 855
 856                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 857                                 sizeof(rpdu), &rpdu);
 858         }
 859 }
 860
 861 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
 862 {
 863         struct blk_trace *bt = q->blk_trace;
 864
 865         if (bt) {
 866                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 867                 __be64 rpdu = cpu_to_be64(pdu);
 868
 869                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 870                                 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_remap - Add a trace for a 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_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_timer(blk_add_trace_unplug_timer, NULL);
1014         WARN_ON(ret);
1015         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1016         WARN_ON(ret);
1017         ret = register_trace_block_split(blk_add_trace_split, NULL);
1018         WARN_ON(ret);
1019         ret = register_trace_block_remap(blk_add_trace_remap, NULL);
1020         WARN_ON(ret);
1021         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1022         WARN_ON(ret);
1023 }
1024
1025 static void blk_unregister_tracepoints(void)
1026 {
1027         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1028         unregister_trace_block_remap(blk_add_trace_remap, NULL);
1029         unregister_trace_block_split(blk_add_trace_split, NULL);
1030         unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1031         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
1032         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1033         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1034         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1035         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1036         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1037         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1038         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1039         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1040         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1041         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1042         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1043         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1044         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1045
1046         tracepoint_synchronize_unregister();
1047 }
1048
1049 /*
1050  * struct blk_io_tracer formatting routines
1051  */
1052
1053 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1054 {
1055         int i = 0;
1056         int tc = t->action >> BLK_TC_SHIFT;
1057
1058         if (t->action == BLK_TN_MESSAGE) {
1059                 rwbs[i++] = 'N';
1060                 goto out;
1061         }
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_AHEAD)
1073                 rwbs[i++] = 'A';
1074         if (tc & BLK_TC_BARRIER)
1075                 rwbs[i++] = 'B';
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[6];
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[6];
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_BARRIER,       "barrier"       },
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 };
1583
1584 static int blk_trace_str2mask(const char *str)
1585 {
1586         int i;
1587         int mask = 0;
1588         char *buf, *s, *token;
1589
1590         buf = kstrdup(str, GFP_KERNEL);
1591         if (buf == NULL)
1592                 return -ENOMEM;
1593         s = strstrip(buf);
1594
1595         while (1) {
1596                 token = strsep(&s, ",");
1597                 if (token == NULL)
1598                         break;
1599
1600                 if (*token == '\0')
1601                         continue;
1602
1603                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1604                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1605                                 mask |= mask_maps[i].mask;
1606                                 break;
1607                         }
1608                 }
1609                 if (i == ARRAY_SIZE(mask_maps)) {
1610                         mask = -EINVAL;
1611                         break;
1612                 }
1613         }
1614         kfree(buf);
1615
1616         return mask;
1617 }
1618
1619 static ssize_t blk_trace_mask2str(char *buf, int mask)
1620 {
1621         int i;
1622         char *p = buf;
1623
1624         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1625                 if (mask & mask_maps[i].mask) {
1626                         p += sprintf(p, "%s%s",
1627                                     (p == buf) ? "" : ",", mask_maps[i].str);
1628                 }
1629         }
1630         *p++ = '\n';
1631
1632         return p - buf;
1633 }
1634
1635 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1636 {
1637         if (bdev->bd_disk == NULL)
1638                 return NULL;
1639
1640         return bdev_get_queue(bdev);
1641 }
1642
1643 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1644                                          struct device_attribute *attr,
1645                                          char *buf)
1646 {
1647         struct hd_struct *p = dev_to_part(dev);
1648         struct request_queue *q;
1649         struct block_device *bdev;
1650         ssize_t ret = -ENXIO;
1651
1652         bdev = bdget(part_devt(p));
1653         if (bdev == NULL)
1654                 goto out;
1655
1656         q = blk_trace_get_queue(bdev);
1657         if (q == NULL)
1658                 goto out_bdput;
1659
1660         mutex_lock(&bdev->bd_mutex);
1661
1662         if (attr == &dev_attr_enable) {
1663                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1664                 goto out_unlock_bdev;
1665         }
1666
1667         if (q->blk_trace == NULL)
1668                 ret = sprintf(buf, "disabled\n");
1669         else if (attr == &dev_attr_act_mask)
1670                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1671         else if (attr == &dev_attr_pid)
1672                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1673         else if (attr == &dev_attr_start_lba)
1674                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1675         else if (attr == &dev_attr_end_lba)
1676                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1677
1678 out_unlock_bdev:
1679         mutex_unlock(&bdev->bd_mutex);
1680 out_bdput:
1681         bdput(bdev);
1682 out:
1683         return ret;
1684 }
1685
1686 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1687                                           struct device_attribute *attr,
1688                                           const char *buf, size_t count)
1689 {
1690         struct block_device *bdev;
1691         struct request_queue *q;
1692         struct hd_struct *p;
1693         u64 value;
1694         ssize_t ret = -EINVAL;
1695
1696         if (count == 0)
1697                 goto out;
1698
1699         if (attr == &dev_attr_act_mask) {
1700                 if (sscanf(buf, "%llx", &value) != 1) {
1701                         /* Assume it is a list of trace category names */
1702                         ret = blk_trace_str2mask(buf);
1703                         if (ret < 0)
1704                                 goto out;
1705                         value = ret;
1706                 }
1707         } else if (sscanf(buf, "%llu", &value) != 1)
1708                 goto out;
1709
1710         ret = -ENXIO;
1711
1712         p = dev_to_part(dev);
1713         bdev = bdget(part_devt(p));
1714         if (bdev == NULL)
1715                 goto out;
1716
1717         q = blk_trace_get_queue(bdev);
1718         if (q == NULL)
1719                 goto out_bdput;
1720
1721         mutex_lock(&bdev->bd_mutex);
1722
1723         if (attr == &dev_attr_enable) {
1724                 if (value)
1725                         ret = blk_trace_setup_queue(q, bdev);
1726                 else
1727                         ret = blk_trace_remove_queue(q);
1728                 goto out_unlock_bdev;
1729         }
1730
1731         ret = 0;
1732         if (q->blk_trace == NULL)
1733                 ret = blk_trace_setup_queue(q, bdev);
1734
1735         if (ret == 0) {
1736                 if (attr == &dev_attr_act_mask)
1737                         q->blk_trace->act_mask = value;
1738                 else if (attr == &dev_attr_pid)
1739                         q->blk_trace->pid = value;
1740                 else if (attr == &dev_attr_start_lba)
1741                         q->blk_trace->start_lba = value;
1742                 else if (attr == &dev_attr_end_lba)
1743                         q->blk_trace->end_lba = value;
1744         }
1745
1746 out_unlock_bdev:
1747         mutex_unlock(&bdev->bd_mutex);
1748 out_bdput:
1749         bdput(bdev);
1750 out:
1751         return ret ? ret : count;
1752 }
1753
1754 int blk_trace_init_sysfs(struct device *dev)
1755 {
1756         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1757 }
1758
1759 void blk_trace_remove_sysfs(struct device *dev)
1760 {
1761         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1762 }
1763
1764 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1765
1766 #ifdef CONFIG_EVENT_TRACING
1767
1768 void blk_dump_cmd(char *buf, struct request *rq)
1769 {
1770         int i, end;
1771         int len = rq->cmd_len;
1772         unsigned char *cmd = rq->cmd;
1773
1774         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1775                 buf[0] = '\0';
1776                 return;
1777         }
1778
1779         for (end = len - 1; end >= 0; end--)
1780                 if (cmd[end])
1781                         break;
1782         end++;
1783
1784         for (i = 0; i < len; i++) {
1785                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1786                 if (i == end && end != len - 1) {
1787                         sprintf(buf, " ..");
1788                         break;
1789                 }
1790         }
1791 }
1792
1793 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1794 {
1795         int i = 0;
1796
1797         if (rw & WRITE)
1798                 rwbs[i++] = 'W';
1799         else if (rw & REQ_DISCARD)
1800                 rwbs[i++] = 'D';
1801         else if (bytes)
1802                 rwbs[i++] = 'R';
1803         else
1804                 rwbs[i++] = 'N';
1805
1806         if (rw & REQ_RAHEAD)
1807                 rwbs[i++] = 'A';
1808         if (rw & REQ_SYNC)
1809                 rwbs[i++] = 'S';
1810         if (rw & REQ_META)
1811                 rwbs[i++] = 'M';
1812         if (rw & REQ_SECURE)
1813                 rwbs[i++] = 'E';
1814
1815         rwbs[i] = '\0';
1816 }
1817
1818 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1819 {
1820         int rw = rq->cmd_flags & 0x03;
1821         int bytes;
1822
1823         if (rq->cmd_flags & REQ_DISCARD)
1824                 rw |= REQ_DISCARD;
1825
1826         if (rq->cmd_flags & REQ_SECURE)
1827                 rw |= REQ_SECURE;
1828
1829         bytes = blk_rq_bytes(rq);
1830
1831         blk_fill_rwbs(rwbs, rw, bytes);
1832 }
1833
1834 #endif /* CONFIG_EVENT_TRACING */
1835