git.oblomov.eu Git - linux-2.6/blob - 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         int pc = 0;
  69         int cpu = smp_processor_id();
  70         bool blk_tracer = blk_tracer_enabled;
  71
  72         if (blk_tracer) {
  73                 pc = preempt_count();
  74                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
  75                                                   sizeof(*t) + len,
  76                                                   0, pc);
  77                 if (!event)
  78                         return;
  79                 t = ring_buffer_event_data(event);
  80                 goto record_it;
  81         }
  82
  83         if (!bt->rchan)
  84                 return;
  85
  86         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  87         if (t) {
  88                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  89                 t->time = ktime_to_ns(ktime_get());
  90 record_it:
  91                 t->device = bt->dev;
  92                 t->action = action;
  93                 t->pid = pid;
  94                 t->cpu = cpu;
  95                 t->pdu_len = len;
  96                 memcpy((void *) t + sizeof(*t), data, len);
  97
  98                 if (blk_tracer)
  99                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 100         }
 101 }
 102
 103 /*
 104  * Send out a notify for this process, if we haven't done so since a trace
 105  * started
 106  */
 107 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 108 {
 109         tsk->btrace_seq = blktrace_seq;
 110         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 111 }
 112
 113 static void trace_note_time(struct blk_trace *bt)
 114 {
 115         struct timespec now;
 116         unsigned long flags;
 117         u32 words[2];
 118
 119         getnstimeofday(&now);
 120         words[0] = now.tv_sec;
 121         words[1] = now.tv_nsec;
 122
 123         local_irq_save(flags);
 124         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 125         local_irq_restore(flags);
 126 }
 127
 128 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 129 {
 130         int n;
 131         va_list args;
 132         unsigned long flags;
 133         char *buf;
 134
 135         if (unlikely(bt->trace_state != Blktrace_running &&
 136                      !blk_tracer_enabled))
 137                 return;
 138
 139         local_irq_save(flags);
 140         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 141         va_start(args, fmt);
 142         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 143         va_end(args);
 144
 145         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 146         local_irq_restore(flags);
 147 }
 148 EXPORT_SYMBOL_GPL(__trace_note_message);
 149
 150 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 151                          pid_t pid)
 152 {
 153         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 154                 return 1;
 155         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 156                 return 1;
 157         if (bt->pid && pid != bt->pid)
 158                 return 1;
 159
 160         return 0;
 161 }
 162
 163 /*
 164  * Data direction bit lookup
 165  */
 166 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 167                                  BLK_TC_ACT(BLK_TC_WRITE) };
 168
 169 /* The ilog2() calls fall out because they're constant */
 170 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 171           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 172
 173 /*
 174  * The worker for the various blk_add_trace*() types. Fills out a
 175  * blk_io_trace structure and places it in a per-cpu subbuffer.
 176  */
 177 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 178                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 179 {
 180         struct task_struct *tsk = current;
 181         struct ring_buffer_event *event = NULL;
 182         struct blk_io_trace *t;
 183         unsigned long flags = 0;
 184         unsigned long *sequence;
 185         pid_t pid;
 186         int cpu, pc = 0;
 187         bool blk_tracer = blk_tracer_enabled;
 188
 189         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 190                 return;
 191
 192         what |= ddir_act[rw & WRITE];
 193         what |= MASK_TC_BIT(rw, BARRIER);
 194         what |= MASK_TC_BIT(rw, SYNCIO);
 195         what |= MASK_TC_BIT(rw, AHEAD);
 196         what |= MASK_TC_BIT(rw, META);
 197         what |= MASK_TC_BIT(rw, DISCARD);
 198
 199         pid = tsk->pid;
 200         if (act_log_check(bt, what, sector, pid))
 201                 return;
 202         cpu = raw_smp_processor_id();
 203
 204         if (blk_tracer) {
 205                 tracing_record_cmdline(current);
 206
 207                 pc = preempt_count();
 208                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 209                                                   sizeof(*t) + pdu_len,
 210                                                   0, pc);
 211                 if (!event)
 212                         return;
 213                 t = ring_buffer_event_data(event);
 214                 goto record_it;
 215         }
 216
 217         /*
 218          * A word about the locking here - we disable interrupts to reserve
 219          * some space in the relay per-cpu buffer, to prevent an irq
 220          * from coming in and stepping on our toes.
 221          */
 222         local_irq_save(flags);
 223
 224         if (unlikely(tsk->btrace_seq != blktrace_seq))
 225                 trace_note_tsk(bt, tsk);
 226
 227         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 228         if (t) {
 229                 sequence = per_cpu_ptr(bt->sequence, cpu);
 230
 231                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 232                 t->sequence = ++(*sequence);
 233                 t->time = ktime_to_ns(ktime_get());
 234 record_it:
 235                 /*
 236                  * These two are not needed in ftrace as they are in the
 237                  * generic trace_entry, filled by tracing_generic_entry_update,
 238                  * but for the trace_event->bin() synthesizer benefit we do it
 239                  * here too.
 240                  */
 241                 t->cpu = cpu;
 242                 t->pid = pid;
 243
 244                 t->sector = sector;
 245                 t->bytes = bytes;
 246                 t->action = what;
 247                 t->device = bt->dev;
 248                 t->error = error;
 249                 t->pdu_len = pdu_len;
 250
 251                 if (pdu_len)
 252                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 253
 254                 if (blk_tracer) {
 255                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 256                         return;
 257                 }
 258         }
 259
 260         local_irq_restore(flags);
 261 }
 262
 263 static struct dentry *blk_tree_root;
 264 static DEFINE_MUTEX(blk_tree_mutex);
 265
 266 static void blk_trace_free(struct blk_trace *bt)
 267 {
 268         debugfs_remove(bt->msg_file);
 269         debugfs_remove(bt->dropped_file);
 270         debugfs_remove(bt->dir);
 271         relay_close(bt->rchan);
 272         free_percpu(bt->sequence);
 273         free_percpu(bt->msg_data);
 274         kfree(bt);
 275 }
 276
 277 static void blk_trace_cleanup(struct blk_trace *bt)
 278 {
 279         blk_trace_free(bt);
 280         if (atomic_dec_and_test(&blk_probes_ref))
 281                 blk_unregister_tracepoints();
 282 }
 283
 284 int blk_trace_remove(struct request_queue *q)
 285 {
 286         struct blk_trace *bt;
 287
 288         bt = xchg(&q->blk_trace, NULL);
 289         if (!bt)
 290                 return -EINVAL;
 291
 292         if (bt->trace_state != Blktrace_running)
 293                 blk_trace_cleanup(bt);
 294
 295         return 0;
 296 }
 297 EXPORT_SYMBOL_GPL(blk_trace_remove);
 298
 299 static int blk_dropped_open(struct inode *inode, struct file *filp)
 300 {
 301         filp->private_data = inode->i_private;
 302
 303         return 0;
 304 }
 305
 306 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 307                                 size_t count, loff_t *ppos)
 308 {
 309         struct blk_trace *bt = filp->private_data;
 310         char buf[16];
 311
 312         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 313
 314         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 315 }
 316
 317 static const struct file_operations blk_dropped_fops = {
 318         .owner =        THIS_MODULE,
 319         .open =         blk_dropped_open,
 320         .read =         blk_dropped_read,
 321 };
 322
 323 static int blk_msg_open(struct inode *inode, struct file *filp)
 324 {
 325         filp->private_data = inode->i_private;
 326
 327         return 0;
 328 }
 329
 330 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 331                                 size_t count, loff_t *ppos)
 332 {
 333         char *msg;
 334         struct blk_trace *bt;
 335
 336         if (count >= BLK_TN_MAX_MSG)
 337                 return -EINVAL;
 338
 339         msg = kmalloc(count + 1, GFP_KERNEL);
 340         if (msg == NULL)
 341                 return -ENOMEM;
 342
 343         if (copy_from_user(msg, buffer, count)) {
 344                 kfree(msg);
 345                 return -EFAULT;
 346         }
 347
 348         msg[count] = '\0';
 349         bt = filp->private_data;
 350         __trace_note_message(bt, "%s", msg);
 351         kfree(msg);
 352
 353         return count;
 354 }
 355
 356 static const struct file_operations blk_msg_fops = {
 357         .owner =        THIS_MODULE,
 358         .open =         blk_msg_open,
 359         .write =        blk_msg_write,
 360 };
 361
 362 /*
 363  * Keep track of how many times we encountered a full subbuffer, to aid
 364  * the user space app in telling how many lost events there were.
 365  */
 366 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 367                                      void *prev_subbuf, size_t prev_padding)
 368 {
 369         struct blk_trace *bt;
 370
 371         if (!relay_buf_full(buf))
 372                 return 1;
 373
 374         bt = buf->chan->private_data;
 375         atomic_inc(&bt->dropped);
 376         return 0;
 377 }
 378
 379 static int blk_remove_buf_file_callback(struct dentry *dentry)
 380 {
 381         struct dentry *parent = dentry->d_parent;
 382         debugfs_remove(dentry);
 383
 384         /*
 385         * this will fail for all but the last file, but that is ok. what we
 386         * care about is the top level buts->name directory going away, when
 387         * the last trace file is gone. Then we don't have to rmdir() that
 388         * manually on trace stop, so it nicely solves the issue with
 389         * force killing of running traces.
 390         */
 391
 392         debugfs_remove(parent);
 393         return 0;
 394 }
 395
 396 static struct dentry *blk_create_buf_file_callback(const char *filename,
 397                                                    struct dentry *parent,
 398                                                    int mode,
 399                                                    struct rchan_buf *buf,
 400                                                    int *is_global)
 401 {
 402         return debugfs_create_file(filename, mode, parent, buf,
 403                                         &relay_file_operations);
 404 }
 405
 406 static struct rchan_callbacks blk_relay_callbacks = {
 407         .subbuf_start           = blk_subbuf_start_callback,
 408         .create_buf_file        = blk_create_buf_file_callback,
 409         .remove_buf_file        = blk_remove_buf_file_callback,
 410 };
 411
 412 static void blk_trace_setup_lba(struct blk_trace *bt,
 413                                 struct block_device *bdev)
 414 {
 415         struct hd_struct *part = NULL;
 416
 417         if (bdev)
 418                 part = bdev->bd_part;
 419
 420         if (part) {
 421                 bt->start_lba = part->start_sect;
 422                 bt->end_lba = part->start_sect + part->nr_sects;
 423         } else {
 424                 bt->start_lba = 0;
 425                 bt->end_lba = -1ULL;
 426         }
 427 }
 428
 429 /*
 430  * Setup everything required to start tracing
 431  */
 432 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 433                        struct block_device *bdev,
 434                        struct blk_user_trace_setup *buts)
 435 {
 436         struct blk_trace *old_bt, *bt = NULL;
 437         struct dentry *dir = NULL;
 438         int ret, i;
 439
 440         if (!buts->buf_size || !buts->buf_nr)
 441                 return -EINVAL;
 442
 443         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 444         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 445
 446         /*
 447          * some device names have larger paths - convert the slashes
 448          * to underscores for this to work as expected
 449          */
 450         for (i = 0; i < strlen(buts->name); i++)
 451                 if (buts->name[i] == '/')
 452                         buts->name[i] = '_';
 453
 454         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 455         if (!bt)
 456                 return -ENOMEM;
 457
 458         ret = -ENOMEM;
 459         bt->sequence = alloc_percpu(unsigned long);
 460         if (!bt->sequence)
 461                 goto err;
 462
 463         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 464         if (!bt->msg_data)
 465                 goto err;
 466
 467         ret = -ENOENT;
 468
 469         mutex_lock(&blk_tree_mutex);
 470         if (!blk_tree_root) {
 471                 blk_tree_root = debugfs_create_dir("block", NULL);
 472                 if (!blk_tree_root) {
 473                         mutex_unlock(&blk_tree_mutex);
 474                         goto err;
 475                 }
 476         }
 477         mutex_unlock(&blk_tree_mutex);
 478
 479         dir = debugfs_create_dir(buts->name, blk_tree_root);
 480
 481         if (!dir)
 482                 goto err;
 483
 484         bt->dir = dir;
 485         bt->dev = dev;
 486         atomic_set(&bt->dropped, 0);
 487
 488         ret = -EIO;
 489         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 490                                                &blk_dropped_fops);
 491         if (!bt->dropped_file)
 492                 goto err;
 493
 494         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 495         if (!bt->msg_file)
 496                 goto err;
 497
 498         bt->rchan = relay_open("trace", dir, buts->buf_size,
 499                                 buts->buf_nr, &blk_relay_callbacks, bt);
 500         if (!bt->rchan)
 501                 goto err;
 502
 503         bt->act_mask = buts->act_mask;
 504         if (!bt->act_mask)
 505                 bt->act_mask = (u16) -1;
 506
 507         blk_trace_setup_lba(bt, bdev);
 508
 509         /* overwrite with user settings */
 510         if (buts->start_lba)
 511                 bt->start_lba = buts->start_lba;
 512         if (buts->end_lba)
 513                 bt->end_lba = buts->end_lba;
 514
 515         bt->pid = buts->pid;
 516         bt->trace_state = Blktrace_setup;
 517
 518         ret = -EBUSY;
 519         old_bt = xchg(&q->blk_trace, bt);
 520         if (old_bt) {
 521                 (void) xchg(&q->blk_trace, old_bt);
 522                 goto err;
 523         }
 524
 525         if (atomic_inc_return(&blk_probes_ref) == 1)
 526                 blk_register_tracepoints();
 527
 528         return 0;
 529 err:
 530         blk_trace_free(bt);
 531         return ret;
 532 }
 533
 534 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 535                     struct block_device *bdev,
 536                     char __user *arg)
 537 {
 538         struct blk_user_trace_setup buts;
 539         int ret;
 540
 541         ret = copy_from_user(&buts, arg, sizeof(buts));
 542         if (ret)
 543                 return -EFAULT;
 544
 545         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 546         if (ret)
 547                 return ret;
 548
 549         if (copy_to_user(arg, &buts, sizeof(buts)))
 550                 return -EFAULT;
 551
 552         return 0;
 553 }
 554 EXPORT_SYMBOL_GPL(blk_trace_setup);
 555
 556 int blk_trace_startstop(struct request_queue *q, int start)
 557 {
 558         int ret;
 559         struct blk_trace *bt = q->blk_trace;
 560
 561         if (bt == NULL)
 562                 return -EINVAL;
 563
 564         /*
 565          * For starting a trace, we can transition from a setup or stopped
 566          * trace. For stopping a trace, the state must be running
 567          */
 568         ret = -EINVAL;
 569         if (start) {
 570                 if (bt->trace_state == Blktrace_setup ||
 571                     bt->trace_state == Blktrace_stopped) {
 572                         blktrace_seq++;
 573                         smp_mb();
 574                         bt->trace_state = Blktrace_running;
 575
 576                         trace_note_time(bt);
 577                         ret = 0;
 578                 }
 579         } else {
 580                 if (bt->trace_state == Blktrace_running) {
 581                         bt->trace_state = Blktrace_stopped;
 582                         relay_flush(bt->rchan);
 583                         ret = 0;
 584                 }
 585         }
 586
 587         return ret;
 588 }
 589 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 590
 591 /**
 592  * blk_trace_ioctl: - handle the ioctls associated with tracing
 593  * @bdev:       the block device
 594  * @cmd:        the ioctl cmd
 595  * @arg:        the argument data, if any
 596  *
 597  **/
 598 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 599 {
 600         struct request_queue *q;
 601         int ret, start = 0;
 602         char b[BDEVNAME_SIZE];
 603
 604         q = bdev_get_queue(bdev);
 605         if (!q)
 606                 return -ENXIO;
 607
 608         mutex_lock(&bdev->bd_mutex);
 609
 610         switch (cmd) {
 611         case BLKTRACESETUP:
 612                 bdevname(bdev, b);
 613                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 614                 break;
 615         case BLKTRACESTART:
 616                 start = 1;
 617         case BLKTRACESTOP:
 618                 ret = blk_trace_startstop(q, start);
 619                 break;
 620         case BLKTRACETEARDOWN:
 621                 ret = blk_trace_remove(q);
 622                 break;
 623         default:
 624                 ret = -ENOTTY;
 625                 break;
 626         }
 627
 628         mutex_unlock(&bdev->bd_mutex);
 629         return ret;
 630 }
 631
 632 /**
 633  * blk_trace_shutdown: - stop and cleanup trace structures
 634  * @q:    the request queue associated with the device
 635  *
 636  **/
 637 void blk_trace_shutdown(struct request_queue *q)
 638 {
 639         if (q->blk_trace) {
 640                 blk_trace_startstop(q, 0);
 641                 blk_trace_remove(q);
 642         }
 643 }
 644
 645 /*
 646  * blktrace probes
 647  */
 648
 649 /**
 650  * blk_add_trace_rq - Add a trace for a request oriented action
 651  * @q:          queue the io is for
 652  * @rq:         the source request
 653  * @what:       the action
 654  *
 655  * Description:
 656  *     Records an action against a request. Will log the bio offset + size.
 657  *
 658  **/
 659 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 660                                     u32 what)
 661 {
 662         struct blk_trace *bt = q->blk_trace;
 663         int rw = rq->cmd_flags & 0x03;
 664
 665         if (likely(!bt))
 666                 return;
 667
 668         if (blk_discard_rq(rq))
 669                 rw |= (1 << BIO_RW_DISCARD);
 670
 671         if (blk_pc_request(rq)) {
 672                 what |= BLK_TC_ACT(BLK_TC_PC);
 673                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
 674                                 what, rq->errors, rq->cmd_len, rq->cmd);
 675         } else  {
 676                 what |= BLK_TC_ACT(BLK_TC_FS);
 677                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
 678                                 what, rq->errors, 0, NULL);
 679         }
 680 }
 681
 682 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 683 {
 684         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 685 }
 686
 687 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 688 {
 689         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 690 }
 691
 692 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 693 {
 694         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 695 }
 696
 697 static void blk_add_trace_rq_requeue(struct request_queue *q,
 698                                      struct request *rq)
 699 {
 700         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 701 }
 702
 703 static void blk_add_trace_rq_complete(struct request_queue *q,
 704                                       struct request *rq)
 705 {
 706         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 707 }
 708
 709 /**
 710  * blk_add_trace_bio - Add a trace for a bio oriented action
 711  * @q:          queue the io is for
 712  * @bio:        the source bio
 713  * @what:       the action
 714  *
 715  * Description:
 716  *     Records an action against a bio. Will log the bio offset + size.
 717  *
 718  **/
 719 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 720                                      u32 what)
 721 {
 722         struct blk_trace *bt = q->blk_trace;
 723
 724         if (likely(!bt))
 725                 return;
 726
 727         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 728                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 729 }
 730
 731 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 732 {
 733         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 734 }
 735
 736 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 737 {
 738         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 739 }
 740
 741 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 742                                         struct bio *bio)
 743 {
 744         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 745 }
 746
 747 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 748                                          struct bio *bio)
 749 {
 750         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 751 }
 752
 753 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 754 {
 755         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 756 }
 757
 758 static void blk_add_trace_getrq(struct request_queue *q,
 759                                 struct bio *bio, int rw)
 760 {
 761         if (bio)
 762                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 763         else {
 764                 struct blk_trace *bt = q->blk_trace;
 765
 766                 if (bt)
 767                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 768         }
 769 }
 770
 771
 772 static void blk_add_trace_sleeprq(struct request_queue *q,
 773                                   struct bio *bio, int rw)
 774 {
 775         if (bio)
 776                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 777         else {
 778                 struct blk_trace *bt = q->blk_trace;
 779
 780                 if (bt)
 781                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 782                                         0, 0, NULL);
 783         }
 784 }
 785
 786 static void blk_add_trace_plug(struct request_queue *q)
 787 {
 788         struct blk_trace *bt = q->blk_trace;
 789
 790         if (bt)
 791                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 792 }
 793
 794 static void blk_add_trace_unplug_io(struct request_queue *q)
 795 {
 796         struct blk_trace *bt = q->blk_trace;
 797
 798         if (bt) {
 799                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 800                 __be64 rpdu = cpu_to_be64(pdu);
 801
 802                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 803                                 sizeof(rpdu), &rpdu);
 804         }
 805 }
 806
 807 static void blk_add_trace_unplug_timer(struct request_queue *q)
 808 {
 809         struct blk_trace *bt = q->blk_trace;
 810
 811         if (bt) {
 812                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 813                 __be64 rpdu = cpu_to_be64(pdu);
 814
 815                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 816                                 sizeof(rpdu), &rpdu);
 817         }
 818 }
 819
 820 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 821                                 unsigned int pdu)
 822 {
 823         struct blk_trace *bt = q->blk_trace;
 824
 825         if (bt) {
 826                 __be64 rpdu = cpu_to_be64(pdu);
 827
 828                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 829                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 830                                 sizeof(rpdu), &rpdu);
 831         }
 832 }
 833
 834 /**
 835  * blk_add_trace_remap - Add a trace for a remap operation
 836  * @q:          queue the io is for
 837  * @bio:        the source bio
 838  * @dev:        target device
 839  * @from:       source sector
 840  *
 841  * Description:
 842  *     Device mapper or raid target sometimes need to split a bio because
 843  *     it spans a stripe (or similar). Add a trace for that action.
 844  *
 845  **/
 846 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 847                                        dev_t dev, sector_t from)
 848 {
 849         struct blk_trace *bt = q->blk_trace;
 850         struct blk_io_trace_remap r;
 851
 852         if (likely(!bt))
 853                 return;
 854
 855         r.device_from = cpu_to_be32(dev);
 856         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 857         r.sector_from = cpu_to_be64(from);
 858
 859         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 860                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 861                         sizeof(r), &r);
 862 }
 863
 864 /**
 865  * blk_add_driver_data - Add binary message with driver-specific data
 866  * @q:          queue the io is for
 867  * @rq:         io request
 868  * @data:       driver-specific data
 869  * @len:        length of driver-specific data
 870  *
 871  * Description:
 872  *     Some drivers might want to write driver-specific data per request.
 873  *
 874  **/
 875 void blk_add_driver_data(struct request_queue *q,
 876                          struct request *rq,
 877                          void *data, size_t len)
 878 {
 879         struct blk_trace *bt = q->blk_trace;
 880
 881         if (likely(!bt))
 882                 return;
 883
 884         if (blk_pc_request(rq))
 885                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 886                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 887         else
 888                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 889                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 890 }
 891 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 892
 893 static void blk_register_tracepoints(void)
 894 {
 895         int ret;
 896
 897         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 898         WARN_ON(ret);
 899         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 900         WARN_ON(ret);
 901         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 902         WARN_ON(ret);
 903         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 904         WARN_ON(ret);
 905         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 906         WARN_ON(ret);
 907         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 908         WARN_ON(ret);
 909         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 910         WARN_ON(ret);
 911         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 912         WARN_ON(ret);
 913         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 914         WARN_ON(ret);
 915         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 916         WARN_ON(ret);
 917         ret = register_trace_block_getrq(blk_add_trace_getrq);
 918         WARN_ON(ret);
 919         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 920         WARN_ON(ret);
 921         ret = register_trace_block_plug(blk_add_trace_plug);
 922         WARN_ON(ret);
 923         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 924         WARN_ON(ret);
 925         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 926         WARN_ON(ret);
 927         ret = register_trace_block_split(blk_add_trace_split);
 928         WARN_ON(ret);
 929         ret = register_trace_block_remap(blk_add_trace_remap);
 930         WARN_ON(ret);
 931 }
 932
 933 static void blk_unregister_tracepoints(void)
 934 {
 935         unregister_trace_block_remap(blk_add_trace_remap);
 936         unregister_trace_block_split(blk_add_trace_split);
 937         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 938         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 939         unregister_trace_block_plug(blk_add_trace_plug);
 940         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 941         unregister_trace_block_getrq(blk_add_trace_getrq);
 942         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 943         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 944         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 945         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 946         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 947         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 948         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 949         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 950         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 951         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 952
 953         tracepoint_synchronize_unregister();
 954 }
 955
 956 /*
 957  * struct blk_io_tracer formatting routines
 958  */
 959
 960 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 961 {
 962         int i = 0;
 963         int tc = t->action >> BLK_TC_SHIFT;
 964
 965         if (t->action == BLK_TN_MESSAGE) {
 966                 rwbs[i++] = 'N';
 967                 goto out;
 968         }
 969
 970         if (tc & BLK_TC_DISCARD)
 971                 rwbs[i++] = 'D';
 972         else if (tc & BLK_TC_WRITE)
 973                 rwbs[i++] = 'W';
 974         else if (t->bytes)
 975                 rwbs[i++] = 'R';
 976         else
 977                 rwbs[i++] = 'N';
 978
 979         if (tc & BLK_TC_AHEAD)
 980                 rwbs[i++] = 'A';
 981         if (tc & BLK_TC_BARRIER)
 982                 rwbs[i++] = 'B';
 983         if (tc & BLK_TC_SYNC)
 984                 rwbs[i++] = 'S';
 985         if (tc & BLK_TC_META)
 986                 rwbs[i++] = 'M';
 987 out:
 988         rwbs[i] = '\0';
 989 }
 990
 991 static inline
 992 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 993 {
 994         return (const struct blk_io_trace *)ent;
 995 }
 996
 997 static inline const void *pdu_start(const struct trace_entry *ent)
 998 {
 999         return te_blk_io_trace(ent) + 1;
1000 }
1001
1002 static inline u32 t_action(const struct trace_entry *ent)
1003 {
1004         return te_blk_io_trace(ent)->action;
1005 }
1006
1007 static inline u32 t_bytes(const struct trace_entry *ent)
1008 {
1009         return te_blk_io_trace(ent)->bytes;
1010 }
1011
1012 static inline u32 t_sec(const struct trace_entry *ent)
1013 {
1014         return te_blk_io_trace(ent)->bytes >> 9;
1015 }
1016
1017 static inline unsigned long long t_sector(const struct trace_entry *ent)
1018 {
1019         return te_blk_io_trace(ent)->sector;
1020 }
1021
1022 static inline __u16 t_error(const struct trace_entry *ent)
1023 {
1024         return te_blk_io_trace(ent)->error;
1025 }
1026
1027 static __u64 get_pdu_int(const struct trace_entry *ent)
1028 {
1029         const __u64 *val = pdu_start(ent);
1030         return be64_to_cpu(*val);
1031 }
1032
1033 static void get_pdu_remap(const struct trace_entry *ent,
1034                           struct blk_io_trace_remap *r)
1035 {
1036         const struct blk_io_trace_remap *__r = pdu_start(ent);
1037         __u64 sector_from = __r->sector_from;
1038
1039         r->device_from = be32_to_cpu(__r->device_from);
1040         r->device_to   = be32_to_cpu(__r->device_to);
1041         r->sector_from = be64_to_cpu(sector_from);
1042 }
1043
1044 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1045
1046 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1047 {
1048         char rwbs[6];
1049         unsigned long long ts  = iter->ts;
1050         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1051         unsigned secs          = (unsigned long)ts;
1052         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1053
1054         fill_rwbs(rwbs, t);
1055
1056         return trace_seq_printf(&iter->seq,
1057                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1058                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1059                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1060 }
1061
1062 static int blk_log_action(struct trace_iterator *iter, const char *act)
1063 {
1064         char rwbs[6];
1065         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1066
1067         fill_rwbs(rwbs, t);
1068         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1069                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1070 }
1071
1072 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1073 {
1074         const unsigned char *pdu_buf;
1075         int pdu_len;
1076         int i, end, ret;
1077
1078         pdu_buf = pdu_start(ent);
1079         pdu_len = te_blk_io_trace(ent)->pdu_len;
1080
1081         if (!pdu_len)
1082                 return 1;
1083
1084         /* find the last zero that needs to be printed */
1085         for (end = pdu_len - 1; end >= 0; end--)
1086                 if (pdu_buf[end])
1087                         break;
1088         end++;
1089
1090         if (!trace_seq_putc(s, '('))
1091                 return 0;
1092
1093         for (i = 0; i < pdu_len; i++) {
1094
1095                 ret = trace_seq_printf(s, "%s%02x",
1096                                        i == 0 ? "" : " ", pdu_buf[i]);
1097                 if (!ret)
1098                         return ret;
1099
1100                 /*
1101                  * stop when the rest is just zeroes and indicate so
1102                  * with a ".." appended
1103                  */
1104                 if (i == end && end != pdu_len - 1)
1105                         return trace_seq_puts(s, " ..) ");
1106         }
1107
1108         return trace_seq_puts(s, ") ");
1109 }
1110
1111 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1112 {
1113         char cmd[TASK_COMM_LEN];
1114
1115         trace_find_cmdline(ent->pid, cmd);
1116
1117         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1118                 int ret;
1119
1120                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1121                 if (!ret)
1122                         return 0;
1123                 ret = blk_log_dump_pdu(s, ent);
1124                 if (!ret)
1125                         return 0;
1126                 return trace_seq_printf(s, "[%s]\n", cmd);
1127         } else {
1128                 if (t_sec(ent))
1129                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1130                                                 t_sector(ent), t_sec(ent), cmd);
1131                 return trace_seq_printf(s, "[%s]\n", cmd);
1132         }
1133 }
1134
1135 static int blk_log_with_error(struct trace_seq *s,
1136                               const struct trace_entry *ent)
1137 {
1138         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1139                 int ret;
1140
1141                 ret = blk_log_dump_pdu(s, ent);
1142                 if (ret)
1143                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1144                 return 0;
1145         } else {
1146                 if (t_sec(ent))
1147                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1148                                                 t_sector(ent),
1149                                                 t_sec(ent), t_error(ent));
1150                 return trace_seq_printf(s, "%llu [%d]\n",
1151                                         t_sector(ent), t_error(ent));
1152         }
1153 }
1154
1155 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1156 {
1157         struct blk_io_trace_remap r = { .device_from = 0, };
1158
1159         get_pdu_remap(ent, &r);
1160         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1161                                 t_sector(ent), t_sec(ent),
1162                                 MAJOR(r.device_from), MINOR(r.device_from),
1163                                 (unsigned long long)r.sector_from);
1164 }
1165
1166 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1167 {
1168         char cmd[TASK_COMM_LEN];
1169
1170         trace_find_cmdline(ent->pid, cmd);
1171
1172         return trace_seq_printf(s, "[%s]\n", cmd);
1173 }
1174
1175 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1176 {
1177         char cmd[TASK_COMM_LEN];
1178
1179         trace_find_cmdline(ent->pid, cmd);
1180
1181         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1182 }
1183
1184 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1185 {
1186         char cmd[TASK_COMM_LEN];
1187
1188         trace_find_cmdline(ent->pid, cmd);
1189
1190         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1191                                 get_pdu_int(ent), cmd);
1192 }
1193
1194 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1195 {
1196         int ret;
1197         const struct blk_io_trace *t = te_blk_io_trace(ent);
1198
1199         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1200         if (ret)
1201                 return trace_seq_putc(s, '\n');
1202         return ret;
1203 }
1204
1205 /*
1206  * struct tracer operations
1207  */
1208
1209 static void blk_tracer_print_header(struct seq_file *m)
1210 {
1211         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1212                 return;
1213         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1214                     "#  |     |     |           |   |   |\n");
1215 }
1216
1217 static void blk_tracer_start(struct trace_array *tr)
1218 {
1219         blk_tracer_enabled = true;
1220 }
1221
1222 static int blk_tracer_init(struct trace_array *tr)
1223 {
1224         blk_tr = tr;
1225         blk_tracer_start(tr);
1226         return 0;
1227 }
1228
1229 static void blk_tracer_stop(struct trace_array *tr)
1230 {
1231         blk_tracer_enabled = false;
1232 }
1233
1234 static void blk_tracer_reset(struct trace_array *tr)
1235 {
1236         blk_tracer_stop(tr);
1237 }
1238
1239 static const struct {
1240         const char *act[2];
1241         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1242 } what2act[] = {
1243         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1244         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1245         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1246         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1247         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1248         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1249         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1250         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1251         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1252         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1253         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1254         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1255         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1256         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1257         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1258 };
1259
1260 static enum print_line_t print_one_line(struct trace_iterator *iter,
1261                                         bool classic)
1262 {
1263         struct trace_seq *s = &iter->seq;
1264         const struct blk_io_trace *t;
1265         u16 what;
1266         int ret;
1267         bool long_act;
1268         blk_log_action_t *log_action;
1269
1270         t          = te_blk_io_trace(iter->ent);
1271         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1272         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1273         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1274
1275         if (t->action == BLK_TN_MESSAGE) {
1276                 ret = log_action(iter, long_act ? "message" : "m");
1277                 if (ret)
1278                         ret = blk_log_msg(s, iter->ent);
1279                 goto out;
1280         }
1281
1282         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1283                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1284         else {
1285                 ret = log_action(iter, what2act[what].act[long_act]);
1286                 if (ret)
1287                         ret = what2act[what].print(s, iter->ent);
1288         }
1289 out:
1290         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1291 }
1292
1293 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1294                                                int flags)
1295 {
1296         return print_one_line(iter, false);
1297 }
1298
1299 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1300 {
1301         struct trace_seq *s = &iter->seq;
1302         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1303         const int offset = offsetof(struct blk_io_trace, sector);
1304         struct blk_io_trace old = {
1305                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1306                 .time     = iter->ts,
1307         };
1308
1309         if (!trace_seq_putmem(s, &old, offset))
1310                 return 0;
1311         return trace_seq_putmem(s, &t->sector,
1312                                 sizeof(old) - offset + t->pdu_len);
1313 }
1314
1315 static enum print_line_t
1316 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1317 {
1318         return blk_trace_synthesize_old_trace(iter) ?
1319                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1320 }
1321
1322 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1323 {
1324         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1325                 return TRACE_TYPE_UNHANDLED;
1326
1327         return print_one_line(iter, true);
1328 }
1329
1330 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1331 {
1332         /* don't output context-info for blk_classic output */
1333         if (bit == TRACE_BLK_OPT_CLASSIC) {
1334                 if (set)
1335                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1336                 else
1337                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1338         }
1339         return 0;
1340 }
1341
1342 static struct tracer blk_tracer __read_mostly = {
1343         .name           = "blk",
1344         .init           = blk_tracer_init,
1345         .reset          = blk_tracer_reset,
1346         .start          = blk_tracer_start,
1347         .stop           = blk_tracer_stop,
1348         .print_header   = blk_tracer_print_header,
1349         .print_line     = blk_tracer_print_line,
1350         .flags          = &blk_tracer_flags,
1351         .set_flag       = blk_tracer_set_flag,
1352 };
1353
1354 static struct trace_event trace_blk_event = {
1355         .type           = TRACE_BLK,
1356         .trace          = blk_trace_event_print,
1357         .binary         = blk_trace_event_print_binary,
1358 };
1359
1360 static int __init init_blk_tracer(void)
1361 {
1362         if (!register_ftrace_event(&trace_blk_event)) {
1363                 pr_warning("Warning: could not register block events\n");
1364                 return 1;
1365         }
1366
1367         if (register_tracer(&blk_tracer) != 0) {
1368                 pr_warning("Warning: could not register the block tracer\n");
1369                 unregister_ftrace_event(&trace_blk_event);
1370                 return 1;
1371         }
1372
1373         return 0;
1374 }
1375
1376 device_initcall(init_blk_tracer);
1377
1378 static int blk_trace_remove_queue(struct request_queue *q)
1379 {
1380         struct blk_trace *bt;
1381
1382         bt = xchg(&q->blk_trace, NULL);
1383         if (bt == NULL)
1384                 return -EINVAL;
1385
1386         if (atomic_dec_and_test(&blk_probes_ref))
1387                 blk_unregister_tracepoints();
1388
1389         blk_trace_free(bt);
1390         return 0;
1391 }
1392
1393 /*
1394  * Setup everything required to start tracing
1395  */
1396 static int blk_trace_setup_queue(struct request_queue *q,
1397                                  struct block_device *bdev)
1398 {
1399         struct blk_trace *old_bt, *bt = NULL;
1400         int ret = -ENOMEM;
1401
1402         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1403         if (!bt)
1404                 return -ENOMEM;
1405
1406         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1407         if (!bt->msg_data)
1408                 goto free_bt;
1409
1410         bt->dev = bdev->bd_dev;
1411         bt->act_mask = (u16)-1;
1412
1413         blk_trace_setup_lba(bt, bdev);
1414
1415         old_bt = xchg(&q->blk_trace, bt);
1416         if (old_bt != NULL) {
1417                 (void)xchg(&q->blk_trace, old_bt);
1418                 ret = -EBUSY;
1419                 goto free_bt;
1420         }
1421
1422         if (atomic_inc_return(&blk_probes_ref) == 1)
1423                 blk_register_tracepoints();
1424         return 0;
1425
1426 free_bt:
1427         blk_trace_free(bt);
1428         return ret;
1429 }
1430
1431 /*
1432  * sysfs interface to enable and configure tracing
1433  */
1434
1435 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1436                                          struct device_attribute *attr,
1437                                          char *buf);
1438 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1439                                           struct device_attribute *attr,
1440                                           const char *buf, size_t count);
1441 #define BLK_TRACE_DEVICE_ATTR(_name) \
1442         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1443                     sysfs_blk_trace_attr_show, \
1444                     sysfs_blk_trace_attr_store)
1445
1446 static BLK_TRACE_DEVICE_ATTR(enable);
1447 static BLK_TRACE_DEVICE_ATTR(act_mask);
1448 static BLK_TRACE_DEVICE_ATTR(pid);
1449 static BLK_TRACE_DEVICE_ATTR(start_lba);
1450 static BLK_TRACE_DEVICE_ATTR(end_lba);
1451
1452 static struct attribute *blk_trace_attrs[] = {
1453         &dev_attr_enable.attr,
1454         &dev_attr_act_mask.attr,
1455         &dev_attr_pid.attr,
1456         &dev_attr_start_lba.attr,
1457         &dev_attr_end_lba.attr,
1458         NULL
1459 };
1460
1461 struct attribute_group blk_trace_attr_group = {
1462         .name  = "trace",
1463         .attrs = blk_trace_attrs,
1464 };
1465
1466 static const struct {
1467         int mask;
1468         const char *str;
1469 } mask_maps[] = {
1470         { BLK_TC_READ,          "read"          },
1471         { BLK_TC_WRITE,         "write"         },
1472         { BLK_TC_BARRIER,       "barrier"       },
1473         { BLK_TC_SYNC,          "sync"          },
1474         { BLK_TC_QUEUE,         "queue"         },
1475         { BLK_TC_REQUEUE,       "requeue"       },
1476         { BLK_TC_ISSUE,         "issue"         },
1477         { BLK_TC_COMPLETE,      "complete"      },
1478         { BLK_TC_FS,            "fs"            },
1479         { BLK_TC_PC,            "pc"            },
1480         { BLK_TC_AHEAD,         "ahead"         },
1481         { BLK_TC_META,          "meta"          },
1482         { BLK_TC_DISCARD,       "discard"       },
1483         { BLK_TC_DRV_DATA,      "drv_data"      },
1484 };
1485
1486 static int blk_trace_str2mask(const char *str)
1487 {
1488         int i;
1489         int mask = 0;
1490         char *buf, *s, *token;
1491
1492         buf = kstrdup(str, GFP_KERNEL);
1493         if (buf == NULL)
1494                 return -ENOMEM;
1495         s = strstrip(buf);
1496
1497         while (1) {
1498                 token = strsep(&s, ",");
1499                 if (token == NULL)
1500                         break;
1501
1502                 if (*token == '\0')
1503                         continue;
1504
1505                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1506                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1507                                 mask |= mask_maps[i].mask;
1508                                 break;
1509                         }
1510                 }
1511                 if (i == ARRAY_SIZE(mask_maps)) {
1512                         mask = -EINVAL;
1513                         break;
1514                 }
1515         }
1516         kfree(buf);
1517
1518         return mask;
1519 }
1520
1521 static ssize_t blk_trace_mask2str(char *buf, int mask)
1522 {
1523         int i;
1524         char *p = buf;
1525
1526         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1527                 if (mask & mask_maps[i].mask) {
1528                         p += sprintf(p, "%s%s",
1529                                     (p == buf) ? "" : ",", mask_maps[i].str);
1530                 }
1531         }
1532         *p++ = '\n';
1533
1534         return p - buf;
1535 }
1536
1537 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1538 {
1539         if (bdev->bd_disk == NULL)
1540                 return NULL;
1541
1542         return bdev_get_queue(bdev);
1543 }
1544
1545 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1546                                          struct device_attribute *attr,
1547                                          char *buf)
1548 {
1549         struct hd_struct *p = dev_to_part(dev);
1550         struct request_queue *q;
1551         struct block_device *bdev;
1552         ssize_t ret = -ENXIO;
1553
1554         lock_kernel();
1555         bdev = bdget(part_devt(p));
1556         if (bdev == NULL)
1557                 goto out_unlock_kernel;
1558
1559         q = blk_trace_get_queue(bdev);
1560         if (q == NULL)
1561                 goto out_bdput;
1562
1563         mutex_lock(&bdev->bd_mutex);
1564
1565         if (attr == &dev_attr_enable) {
1566                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1567                 goto out_unlock_bdev;
1568         }
1569
1570         if (q->blk_trace == NULL)
1571                 ret = sprintf(buf, "disabled\n");
1572         else if (attr == &dev_attr_act_mask)
1573                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1574         else if (attr == &dev_attr_pid)
1575                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1576         else if (attr == &dev_attr_start_lba)
1577                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1578         else if (attr == &dev_attr_end_lba)
1579                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1580
1581 out_unlock_bdev:
1582         mutex_unlock(&bdev->bd_mutex);
1583 out_bdput:
1584         bdput(bdev);
1585 out_unlock_kernel:
1586         unlock_kernel();
1587         return ret;
1588 }
1589
1590 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1591                                           struct device_attribute *attr,
1592                                           const char *buf, size_t count)
1593 {
1594         struct block_device *bdev;
1595         struct request_queue *q;
1596         struct hd_struct *p;
1597         u64 value;
1598         ssize_t ret = -EINVAL;
1599
1600         if (count == 0)
1601                 goto out;
1602
1603         if (attr == &dev_attr_act_mask) {
1604                 if (sscanf(buf, "%llx", &value) != 1) {
1605                         /* Assume it is a list of trace category names */
1606                         ret = blk_trace_str2mask(buf);
1607                         if (ret < 0)
1608                                 goto out;
1609                         value = ret;
1610                 }
1611         } else if (sscanf(buf, "%llu", &value) != 1)
1612                 goto out;
1613
1614         ret = -ENXIO;
1615
1616         lock_kernel();
1617         p = dev_to_part(dev);
1618         bdev = bdget(part_devt(p));
1619         if (bdev == NULL)
1620                 goto out_unlock_kernel;
1621
1622         q = blk_trace_get_queue(bdev);
1623         if (q == NULL)
1624                 goto out_bdput;
1625
1626         mutex_lock(&bdev->bd_mutex);
1627
1628         if (attr == &dev_attr_enable) {
1629                 if (value)
1630                         ret = blk_trace_setup_queue(q, bdev);
1631                 else
1632                         ret = blk_trace_remove_queue(q);
1633                 goto out_unlock_bdev;
1634         }
1635
1636         ret = 0;
1637         if (q->blk_trace == NULL)
1638                 ret = blk_trace_setup_queue(q, bdev);
1639
1640         if (ret == 0) {
1641                 if (attr == &dev_attr_act_mask)
1642                         q->blk_trace->act_mask = value;
1643                 else if (attr == &dev_attr_pid)
1644                         q->blk_trace->pid = value;
1645                 else if (attr == &dev_attr_start_lba)
1646                         q->blk_trace->start_lba = value;
1647                 else if (attr == &dev_attr_end_lba)
1648                         q->blk_trace->end_lba = value;
1649         }
1650
1651 out_unlock_bdev:
1652         mutex_unlock(&bdev->bd_mutex);
1653 out_bdput:
1654         bdput(bdev);
1655 out_unlock_kernel:
1656         unlock_kernel();
1657 out:
1658         return ret ? ret : count;
1659 }
1660
1661 int blk_trace_init_sysfs(struct device *dev)
1662 {
1663         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1664 }
1665
1666 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1667
1668 #ifdef CONFIG_EVENT_TRACING
1669
1670 void blk_dump_cmd(char *buf, struct request *rq)
1671 {
1672         int i, end;
1673         int len = rq->cmd_len;
1674         unsigned char *cmd = rq->cmd;
1675
1676         if (!blk_pc_request(rq)) {
1677                 buf[0] = '\0';
1678                 return;
1679         }
1680
1681         for (end = len - 1; end >= 0; end--)
1682                 if (cmd[end])
1683                         break;
1684         end++;
1685
1686         for (i = 0; i < len; i++) {
1687                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1688                 if (i == end && end != len - 1) {
1689                         sprintf(buf, " ..");
1690                         break;
1691                 }
1692         }
1693 }
1694
1695 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1696 {
1697         int i = 0;
1698
1699         if (rw & WRITE)
1700                 rwbs[i++] = 'W';
1701         else if (rw & 1 << BIO_RW_DISCARD)
1702                 rwbs[i++] = 'D';
1703         else if (bytes)
1704                 rwbs[i++] = 'R';
1705         else
1706                 rwbs[i++] = 'N';
1707
1708         if (rw & 1 << BIO_RW_AHEAD)
1709                 rwbs[i++] = 'A';
1710         if (rw & 1 << BIO_RW_BARRIER)
1711                 rwbs[i++] = 'B';
1712         if (rw & 1 << BIO_RW_SYNCIO)
1713                 rwbs[i++] = 'S';
1714         if (rw & 1 << BIO_RW_META)
1715                 rwbs[i++] = 'M';
1716
1717         rwbs[i] = '\0';
1718 }
1719
1720 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1721 {
1722         int rw = rq->cmd_flags & 0x03;
1723         int bytes;
1724
1725         if (blk_discard_rq(rq))
1726                 rw |= (1 << BIO_RW_DISCARD);
1727
1728         bytes = blk_rq_bytes(rq);
1729
1730         blk_fill_rwbs(rwbs, rw, bytes);
1731 }
1732
1733 #endif /* CONFIG_EVENT_TRACING */
1734