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