kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29
  30 #include <trace/events/block.h>
  31
  32 #include "trace_output.h"
  33
  34 #ifdef CONFIG_BLK_DEV_IO_TRACE
  35
  36 static unsigned int blktrace_seq __read_mostly = 1;
  37
  38 static struct trace_array *blk_tr;
  39 static bool blk_tracer_enabled __read_mostly;
  40
  41 /* Select an alternative, minimalistic output than the original one */
  42 #define TRACE_BLK_OPT_CLASSIC   0x1
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47         { }
  48 };
  49
  50 static struct tracer_flags blk_tracer_flags = {
  51         .val  = 0,
  52         .opts = blk_tracer_opts,
  53 };
  54
  55 /* Global reference count of probes */
  56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  57
  58 static void blk_register_tracepoints(void);
  59 static void blk_unregister_tracepoints(void);
  60
  61 /*
  62  * Send out a notify message.
  63  */
  64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  65                        const void *data, size_t len)
  66 {
  67         struct blk_io_trace *t;
  68         struct ring_buffer_event *event = NULL;
  69         struct ring_buffer *buffer = NULL;
  70         int pc = 0;
  71         int cpu = smp_processor_id();
  72         bool blk_tracer = blk_tracer_enabled;
  73
  74         if (blk_tracer) {
  75                 buffer = blk_tr->buffer;
  76                 pc = preempt_count();
  77                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  78                                                   sizeof(*t) + len,
  79                                                   0, pc);
  80                 if (!event)
  81                         return;
  82                 t = ring_buffer_event_data(event);
  83                 goto record_it;
  84         }
  85
  86         if (!bt->rchan)
  87                 return;
  88
  89         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  90         if (t) {
  91                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  92                 t->time = ktime_to_ns(ktime_get());
  93 record_it:
  94                 t->device = bt->dev;
  95                 t->action = action;
  96                 t->pid = pid;
  97                 t->cpu = cpu;
  98                 t->pdu_len = len;
  99                 memcpy((void *) t + sizeof(*t), data, len);
 100
 101                 if (blk_tracer)
 102                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 103         }
 104 }
 105
 106 /*
 107  * Send out a notify for this process, if we haven't done so since a trace
 108  * started
 109  */
 110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 111 {
 112         tsk->btrace_seq = blktrace_seq;
 113         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 114 }
 115
 116 static void trace_note_time(struct blk_trace *bt)
 117 {
 118         struct timespec now;
 119         unsigned long flags;
 120         u32 words[2];
 121
 122         getnstimeofday(&now);
 123         words[0] = now.tv_sec;
 124         words[1] = now.tv_nsec;
 125
 126         local_irq_save(flags);
 127         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 128         local_irq_restore(flags);
 129 }
 130
 131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 132 {
 133         int n;
 134         va_list args;
 135         unsigned long flags;
 136         char *buf;
 137
 138         if (unlikely(bt->trace_state != Blktrace_running &&
 139                      !blk_tracer_enabled))
 140                 return;
 141
 142         local_irq_save(flags);
 143         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 144         va_start(args, fmt);
 145         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 146         va_end(args);
 147
 148         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 149         local_irq_restore(flags);
 150 }
 151 EXPORT_SYMBOL_GPL(__trace_note_message);
 152
 153 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 154                          pid_t pid)
 155 {
 156         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 157                 return 1;
 158         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 159                 return 1;
 160         if (bt->pid && pid != bt->pid)
 161                 return 1;
 162
 163         return 0;
 164 }
 165
 166 /*
 167  * Data direction bit lookup
 168  */
 169 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 170                                  BLK_TC_ACT(BLK_TC_WRITE) };
 171
 172 #define BLK_TC_HARDBARRIER      BLK_TC_BARRIER
 173 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 174
 175 /* The ilog2() calls fall out because they're constant */
 176 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 177           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 178
 179 /*
 180  * The worker for the various blk_add_trace*() types. Fills out a
 181  * blk_io_trace structure and places it in a per-cpu subbuffer.
 182  */
 183 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 184                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 185 {
 186         struct task_struct *tsk = current;
 187         struct ring_buffer_event *event = NULL;
 188         struct ring_buffer *buffer = NULL;
 189         struct blk_io_trace *t;
 190         unsigned long flags = 0;
 191         unsigned long *sequence;
 192         pid_t pid;
 193         int cpu, pc = 0;
 194         bool blk_tracer = blk_tracer_enabled;
 195
 196         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 197                 return;
 198
 199         what |= ddir_act[rw & WRITE];
 200         what |= MASK_TC_BIT(rw, HARDBARRIER);
 201         what |= MASK_TC_BIT(rw, SYNC);
 202         what |= MASK_TC_BIT(rw, RAHEAD);
 203         what |= MASK_TC_BIT(rw, META);
 204         what |= MASK_TC_BIT(rw, DISCARD);
 205
 206         pid = tsk->pid;
 207         if (act_log_check(bt, what, sector, pid))
 208                 return;
 209         cpu = raw_smp_processor_id();
 210
 211         if (blk_tracer) {
 212                 tracing_record_cmdline(current);
 213
 214                 buffer = blk_tr->buffer;
 215                 pc = preempt_count();
 216                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 217                                                   sizeof(*t) + pdu_len,
 218                                                   0, pc);
 219                 if (!event)
 220                         return;
 221                 t = ring_buffer_event_data(event);
 222                 goto record_it;
 223         }
 224
 225         /*
 226          * A word about the locking here - we disable interrupts to reserve
 227          * some space in the relay per-cpu buffer, to prevent an irq
 228          * from coming in and stepping on our toes.
 229          */
 230         local_irq_save(flags);
 231
 232         if (unlikely(tsk->btrace_seq != blktrace_seq))
 233                 trace_note_tsk(bt, tsk);
 234
 235         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 236         if (t) {
 237                 sequence = per_cpu_ptr(bt->sequence, cpu);
 238
 239                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 240                 t->sequence = ++(*sequence);
 241                 t->time = ktime_to_ns(ktime_get());
 242 record_it:
 243                 /*
 244                  * These two are not needed in ftrace as they are in the
 245                  * generic trace_entry, filled by tracing_generic_entry_update,
 246                  * but for the trace_event->bin() synthesizer benefit we do it
 247                  * here too.
 248                  */
 249                 t->cpu = cpu;
 250                 t->pid = pid;
 251
 252                 t->sector = sector;
 253                 t->bytes = bytes;
 254                 t->action = what;
 255                 t->device = bt->dev;
 256                 t->error = error;
 257                 t->pdu_len = pdu_len;
 258
 259                 if (pdu_len)
 260                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 261
 262                 if (blk_tracer) {
 263                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 264                         return;
 265                 }
 266         }
 267
 268         local_irq_restore(flags);
 269 }
 270
 271 static struct dentry *blk_tree_root;
 272 static DEFINE_MUTEX(blk_tree_mutex);
 273
 274 static void blk_trace_free(struct blk_trace *bt)
 275 {
 276         debugfs_remove(bt->msg_file);
 277         debugfs_remove(bt->dropped_file);
 278         relay_close(bt->rchan);
 279         debugfs_remove(bt->dir);
 280         free_percpu(bt->sequence);
 281         free_percpu(bt->msg_data);
 282         kfree(bt);
 283 }
 284
 285 static void blk_trace_cleanup(struct blk_trace *bt)
 286 {
 287         blk_trace_free(bt);
 288         if (atomic_dec_and_test(&blk_probes_ref))
 289                 blk_unregister_tracepoints();
 290 }
 291
 292 int blk_trace_remove(struct request_queue *q)
 293 {
 294         struct blk_trace *bt;
 295
 296         bt = xchg(&q->blk_trace, NULL);
 297         if (!bt)
 298                 return -EINVAL;
 299
 300         if (bt->trace_state != Blktrace_running)
 301                 blk_trace_cleanup(bt);
 302
 303         return 0;
 304 }
 305 EXPORT_SYMBOL_GPL(blk_trace_remove);
 306
 307 static int blk_dropped_open(struct inode *inode, struct file *filp)
 308 {
 309         filp->private_data = inode->i_private;
 310
 311         return 0;
 312 }
 313
 314 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 315                                 size_t count, loff_t *ppos)
 316 {
 317         struct blk_trace *bt = filp->private_data;
 318         char buf[16];
 319
 320         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 321
 322         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 323 }
 324
 325 static const struct file_operations blk_dropped_fops = {
 326         .owner =        THIS_MODULE,
 327         .open =         blk_dropped_open,
 328         .read =         blk_dropped_read,
 329 };
 330
 331 static int blk_msg_open(struct inode *inode, struct file *filp)
 332 {
 333         filp->private_data = inode->i_private;
 334
 335         return 0;
 336 }
 337
 338 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 339                                 size_t count, loff_t *ppos)
 340 {
 341         char *msg;
 342         struct blk_trace *bt;
 343
 344         if (count >= BLK_TN_MAX_MSG)
 345                 return -EINVAL;
 346
 347         msg = kmalloc(count + 1, GFP_KERNEL);
 348         if (msg == NULL)
 349                 return -ENOMEM;
 350
 351         if (copy_from_user(msg, buffer, count)) {
 352                 kfree(msg);
 353                 return -EFAULT;
 354         }
 355
 356         msg[count] = '\0';
 357         bt = filp->private_data;
 358         __trace_note_message(bt, "%s", msg);
 359         kfree(msg);
 360
 361         return count;
 362 }
 363
 364 static const struct file_operations blk_msg_fops = {
 365         .owner =        THIS_MODULE,
 366         .open =         blk_msg_open,
 367         .write =        blk_msg_write,
 368 };
 369
 370 /*
 371  * Keep track of how many times we encountered a full subbuffer, to aid
 372  * the user space app in telling how many lost events there were.
 373  */
 374 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 375                                      void *prev_subbuf, size_t prev_padding)
 376 {
 377         struct blk_trace *bt;
 378
 379         if (!relay_buf_full(buf))
 380                 return 1;
 381
 382         bt = buf->chan->private_data;
 383         atomic_inc(&bt->dropped);
 384         return 0;
 385 }
 386
 387 static int blk_remove_buf_file_callback(struct dentry *dentry)
 388 {
 389         debugfs_remove(dentry);
 390
 391         return 0;
 392 }
 393
 394 static struct dentry *blk_create_buf_file_callback(const char *filename,
 395                                                    struct dentry *parent,
 396                                                    int mode,
 397                                                    struct rchan_buf *buf,
 398                                                    int *is_global)
 399 {
 400         return debugfs_create_file(filename, mode, parent, buf,
 401                                         &relay_file_operations);
 402 }
 403
 404 static struct rchan_callbacks blk_relay_callbacks = {
 405         .subbuf_start           = blk_subbuf_start_callback,
 406         .create_buf_file        = blk_create_buf_file_callback,
 407         .remove_buf_file        = blk_remove_buf_file_callback,
 408 };
 409
 410 static void blk_trace_setup_lba(struct blk_trace *bt,
 411                                 struct block_device *bdev)
 412 {
 413         struct hd_struct *part = NULL;
 414
 415         if (bdev)
 416                 part = bdev->bd_part;
 417
 418         if (part) {
 419                 bt->start_lba = part->start_sect;
 420                 bt->end_lba = part->start_sect + part->nr_sects;
 421         } else {
 422                 bt->start_lba = 0;
 423                 bt->end_lba = -1ULL;
 424         }
 425 }
 426
 427 /*
 428  * Setup everything required to start tracing
 429  */
 430 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 431                        struct block_device *bdev,
 432                        struct blk_user_trace_setup *buts)
 433 {
 434         struct blk_trace *old_bt, *bt = NULL;
 435         struct dentry *dir = NULL;
 436         int ret, i;
 437
 438         if (!buts->buf_size || !buts->buf_nr)
 439                 return -EINVAL;
 440
 441         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 442         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 443
 444         /*
 445          * some device names have larger paths - convert the slashes
 446          * to underscores for this to work as expected
 447          */
 448         for (i = 0; i < strlen(buts->name); i++)
 449                 if (buts->name[i] == '/')
 450                         buts->name[i] = '_';
 451
 452         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 453         if (!bt)
 454                 return -ENOMEM;
 455
 456         ret = -ENOMEM;
 457         bt->sequence = alloc_percpu(unsigned long);
 458         if (!bt->sequence)
 459                 goto err;
 460
 461         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 462         if (!bt->msg_data)
 463                 goto err;
 464
 465         ret = -ENOENT;
 466
 467         mutex_lock(&blk_tree_mutex);
 468         if (!blk_tree_root) {
 469                 blk_tree_root = debugfs_create_dir("block", NULL);
 470                 if (!blk_tree_root) {
 471                         mutex_unlock(&blk_tree_mutex);
 472                         goto err;
 473                 }
 474         }
 475         mutex_unlock(&blk_tree_mutex);
 476
 477         dir = debugfs_create_dir(buts->name, blk_tree_root);
 478
 479         if (!dir)
 480                 goto err;
 481
 482         bt->dir = dir;
 483         bt->dev = dev;
 484         atomic_set(&bt->dropped, 0);
 485
 486         ret = -EIO;
 487         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 488                                                &blk_dropped_fops);
 489         if (!bt->dropped_file)
 490                 goto err;
 491
 492         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 493         if (!bt->msg_file)
 494                 goto err;
 495
 496         bt->rchan = relay_open("trace", dir, buts->buf_size,
 497                                 buts->buf_nr, &blk_relay_callbacks, bt);
 498         if (!bt->rchan)
 499                 goto err;
 500
 501         bt->act_mask = buts->act_mask;
 502         if (!bt->act_mask)
 503                 bt->act_mask = (u16) -1;
 504
 505         blk_trace_setup_lba(bt, bdev);
 506
 507         /* overwrite with user settings */
 508         if (buts->start_lba)
 509                 bt->start_lba = buts->start_lba;
 510         if (buts->end_lba)
 511                 bt->end_lba = buts->end_lba;
 512
 513         bt->pid = buts->pid;
 514         bt->trace_state = Blktrace_setup;
 515
 516         ret = -EBUSY;
 517         old_bt = xchg(&q->blk_trace, bt);
 518         if (old_bt) {
 519                 (void) xchg(&q->blk_trace, old_bt);
 520                 goto err;
 521         }
 522
 523         if (atomic_inc_return(&blk_probes_ref) == 1)
 524                 blk_register_tracepoints();
 525
 526         return 0;
 527 err:
 528         blk_trace_free(bt);
 529         return ret;
 530 }
 531
 532 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 533                     struct block_device *bdev,
 534                     char __user *arg)
 535 {
 536         struct blk_user_trace_setup buts;
 537         int ret;
 538
 539         ret = copy_from_user(&buts, arg, sizeof(buts));
 540         if (ret)
 541                 return -EFAULT;
 542
 543         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 544         if (ret)
 545                 return ret;
 546
 547         if (copy_to_user(arg, &buts, sizeof(buts))) {
 548                 blk_trace_remove(q);
 549                 return -EFAULT;
 550         }
 551         return 0;
 552 }
 553 EXPORT_SYMBOL_GPL(blk_trace_setup);
 554
 555 int blk_trace_startstop(struct request_queue *q, int start)
 556 {
 557         int ret;
 558         struct blk_trace *bt = q->blk_trace;
 559
 560         if (bt == NULL)
 561                 return -EINVAL;
 562
 563         /*
 564          * For starting a trace, we can transition from a setup or stopped
 565          * trace. For stopping a trace, the state must be running
 566          */
 567         ret = -EINVAL;
 568         if (start) {
 569                 if (bt->trace_state == Blktrace_setup ||
 570                     bt->trace_state == Blktrace_stopped) {
 571                         blktrace_seq++;
 572                         smp_mb();
 573                         bt->trace_state = Blktrace_running;
 574
 575                         trace_note_time(bt);
 576                         ret = 0;
 577                 }
 578         } else {
 579                 if (bt->trace_state == Blktrace_running) {
 580                         bt->trace_state = Blktrace_stopped;
 581                         relay_flush(bt->rchan);
 582                         ret = 0;
 583                 }
 584         }
 585
 586         return ret;
 587 }
 588 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 589
 590 /**
 591  * blk_trace_ioctl: - handle the ioctls associated with tracing
 592  * @bdev:       the block device
 593  * @cmd:        the ioctl cmd
 594  * @arg:        the argument data, if any
 595  *
 596  **/
 597 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 598 {
 599         struct request_queue *q;
 600         int ret, start = 0;
 601         char b[BDEVNAME_SIZE];
 602
 603         q = bdev_get_queue(bdev);
 604         if (!q)
 605                 return -ENXIO;
 606
 607         mutex_lock(&bdev->bd_mutex);
 608
 609         switch (cmd) {
 610         case BLKTRACESETUP:
 611                 bdevname(bdev, b);
 612                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 613                 break;
 614         case BLKTRACESTART:
 615                 start = 1;
 616         case BLKTRACESTOP:
 617                 ret = blk_trace_startstop(q, start);
 618                 break;
 619         case BLKTRACETEARDOWN:
 620                 ret = blk_trace_remove(q);
 621                 break;
 622         default:
 623                 ret = -ENOTTY;
 624                 break;
 625         }
 626
 627         mutex_unlock(&bdev->bd_mutex);
 628         return ret;
 629 }
 630
 631 /**
 632  * blk_trace_shutdown: - stop and cleanup trace structures
 633  * @q:    the request queue associated with the device
 634  *
 635  **/
 636 void blk_trace_shutdown(struct request_queue *q)
 637 {
 638         if (q->blk_trace) {
 639                 blk_trace_startstop(q, 0);
 640                 blk_trace_remove(q);
 641         }
 642 }
 643
 644 /*
 645  * blktrace probes
 646  */
 647
 648 /**
 649  * blk_add_trace_rq - Add a trace for a request oriented action
 650  * @q:          queue the io is for
 651  * @rq:         the source request
 652  * @what:       the action
 653  *
 654  * Description:
 655  *     Records an action against a request. Will log the bio offset + size.
 656  *
 657  **/
 658 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 659                                     u32 what)
 660 {
 661         struct blk_trace *bt = q->blk_trace;
 662         int rw = rq->cmd_flags & 0x03;
 663
 664         if (likely(!bt))
 665                 return;
 666
 667         if (rq->cmd_flags & REQ_DISCARD)
 668                 rw |= REQ_DISCARD;
 669
 670         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 671                 what |= BLK_TC_ACT(BLK_TC_PC);
 672                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rw,
 673                                 what, rq->errors, rq->cmd_len, rq->cmd);
 674         } else  {
 675                 what |= BLK_TC_ACT(BLK_TC_FS);
 676                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), rw,
 677                                 what, rq->errors, 0, NULL);
 678         }
 679 }
 680
 681 static void blk_add_trace_rq_abort(void *ignore,
 682                                    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(void *ignore,
 688                                     struct request_queue *q, struct request *rq)
 689 {
 690         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 691 }
 692
 693 static void blk_add_trace_rq_issue(void *ignore,
 694                                    struct request_queue *q, struct request *rq)
 695 {
 696         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 697 }
 698
 699 static void blk_add_trace_rq_requeue(void *ignore,
 700                                      struct request_queue *q,
 701                                      struct request *rq)
 702 {
 703         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 704 }
 705
 706 static void blk_add_trace_rq_complete(void *ignore,
 707                                       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(void *ignore,
 736                                      struct request_queue *q, struct bio *bio)
 737 {
 738         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 739 }
 740
 741 static void blk_add_trace_bio_complete(void *ignore,
 742                                        struct request_queue *q, struct bio *bio)
 743 {
 744         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 745 }
 746
 747 static void blk_add_trace_bio_backmerge(void *ignore,
 748                                         struct request_queue *q,
 749                                         struct bio *bio)
 750 {
 751         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 752 }
 753
 754 static void blk_add_trace_bio_frontmerge(void *ignore,
 755                                          struct request_queue *q,
 756                                          struct bio *bio)
 757 {
 758         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 759 }
 760
 761 static void blk_add_trace_bio_queue(void *ignore,
 762                                     struct request_queue *q, struct bio *bio)
 763 {
 764         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 765 }
 766
 767 static void blk_add_trace_getrq(void *ignore,
 768                                 struct request_queue *q,
 769                                 struct bio *bio, int rw)
 770 {
 771         if (bio)
 772                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 773         else {
 774                 struct blk_trace *bt = q->blk_trace;
 775
 776                 if (bt)
 777                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 778         }
 779 }
 780
 781
 782 static void blk_add_trace_sleeprq(void *ignore,
 783                                   struct request_queue *q,
 784                                   struct bio *bio, int rw)
 785 {
 786         if (bio)
 787                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 788         else {
 789                 struct blk_trace *bt = q->blk_trace;
 790
 791                 if (bt)
 792                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 793                                         0, 0, NULL);
 794         }
 795 }
 796
 797 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 798 {
 799         struct blk_trace *bt = q->blk_trace;
 800
 801         if (bt)
 802                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 803 }
 804
 805 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
 806 {
 807         struct blk_trace *bt = q->blk_trace;
 808
 809         if (bt) {
 810                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 811                 __be64 rpdu = cpu_to_be64(pdu);
 812
 813                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 814                                 sizeof(rpdu), &rpdu);
 815         }
 816 }
 817
 818 static void blk_add_trace_unplug_timer(void *ignore, struct request_queue *q)
 819 {
 820         struct blk_trace *bt = q->blk_trace;
 821
 822         if (bt) {
 823                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 824                 __be64 rpdu = cpu_to_be64(pdu);
 825
 826                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 827                                 sizeof(rpdu), &rpdu);
 828         }
 829 }
 830
 831 static void blk_add_trace_split(void *ignore,
 832                                 struct request_queue *q, struct bio *bio,
 833                                 unsigned int pdu)
 834 {
 835         struct blk_trace *bt = q->blk_trace;
 836
 837         if (bt) {
 838                 __be64 rpdu = cpu_to_be64(pdu);
 839
 840                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 841                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 842                                 sizeof(rpdu), &rpdu);
 843         }
 844 }
 845
 846 /**
 847  * blk_add_trace_remap - Add a trace for a remap operation
 848  * @ignore:     trace callback data parameter (not used)
 849  * @q:          queue the io is for
 850  * @bio:        the source bio
 851  * @dev:        target device
 852  * @from:       source sector
 853  *
 854  * Description:
 855  *     Device mapper or raid target sometimes need to split a bio because
 856  *     it spans a stripe (or similar). Add a trace for that action.
 857  *
 858  **/
 859 static void blk_add_trace_remap(void *ignore,
 860                                 struct request_queue *q, struct bio *bio,
 861                                 dev_t dev, sector_t from)
 862 {
 863         struct blk_trace *bt = q->blk_trace;
 864         struct blk_io_trace_remap r;
 865
 866         if (likely(!bt))
 867                 return;
 868
 869         r.device_from = cpu_to_be32(dev);
 870         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 871         r.sector_from = cpu_to_be64(from);
 872
 873         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 874                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 875                         sizeof(r), &r);
 876 }
 877
 878 /**
 879  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 880  * @ignore:     trace callback data parameter (not used)
 881  * @q:          queue the io is for
 882  * @rq:         the source request
 883  * @dev:        target device
 884  * @from:       source sector
 885  *
 886  * Description:
 887  *     Device mapper remaps request to other devices.
 888  *     Add a trace for that action.
 889  *
 890  **/
 891 static void blk_add_trace_rq_remap(void *ignore,
 892                                    struct request_queue *q,
 893                                    struct request *rq, dev_t dev,
 894                                    sector_t from)
 895 {
 896         struct blk_trace *bt = q->blk_trace;
 897         struct blk_io_trace_remap r;
 898
 899         if (likely(!bt))
 900                 return;
 901
 902         r.device_from = cpu_to_be32(dev);
 903         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 904         r.sector_from = cpu_to_be64(from);
 905
 906         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 907                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 908                         sizeof(r), &r);
 909 }
 910
 911 /**
 912  * blk_add_driver_data - Add binary message with driver-specific data
 913  * @q:          queue the io is for
 914  * @rq:         io request
 915  * @data:       driver-specific data
 916  * @len:        length of driver-specific data
 917  *
 918  * Description:
 919  *     Some drivers might want to write driver-specific data per request.
 920  *
 921  **/
 922 void blk_add_driver_data(struct request_queue *q,
 923                          struct request *rq,
 924                          void *data, size_t len)
 925 {
 926         struct blk_trace *bt = q->blk_trace;
 927
 928         if (likely(!bt))
 929                 return;
 930
 931         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 932                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 933                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 934         else
 935                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 936                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 937 }
 938 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 939
 940 static void blk_register_tracepoints(void)
 941 {
 942         int ret;
 943
 944         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 945         WARN_ON(ret);
 946         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 947         WARN_ON(ret);
 948         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 949         WARN_ON(ret);
 950         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 951         WARN_ON(ret);
 952         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 953         WARN_ON(ret);
 954         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 955         WARN_ON(ret);
 956         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 957         WARN_ON(ret);
 958         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 959         WARN_ON(ret);
 960         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
 961         WARN_ON(ret);
 962         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
 963         WARN_ON(ret);
 964         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
 965         WARN_ON(ret);
 966         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
 967         WARN_ON(ret);
 968         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
 969         WARN_ON(ret);
 970         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
 971         WARN_ON(ret);
 972         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
 973         WARN_ON(ret);
 974         ret = register_trace_block_split(blk_add_trace_split, NULL);
 975         WARN_ON(ret);
 976         ret = register_trace_block_remap(blk_add_trace_remap, NULL);
 977         WARN_ON(ret);
 978         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
 979         WARN_ON(ret);
 980 }
 981
 982 static void blk_unregister_tracepoints(void)
 983 {
 984         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
 985         unregister_trace_block_remap(blk_add_trace_remap, NULL);
 986         unregister_trace_block_split(blk_add_trace_split, NULL);
 987         unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
 988         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer, NULL);
 989         unregister_trace_block_plug(blk_add_trace_plug, NULL);
 990         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
 991         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
 992         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
 993         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
 994         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 995         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 996         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 997         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 998         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 999         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1000         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1001         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1002
1003         tracepoint_synchronize_unregister();
1004 }
1005
1006 /*
1007  * struct blk_io_tracer formatting routines
1008  */
1009
1010 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1011 {
1012         int i = 0;
1013         int tc = t->action >> BLK_TC_SHIFT;
1014
1015         if (t->action == BLK_TN_MESSAGE) {
1016                 rwbs[i++] = 'N';
1017                 goto out;
1018         }
1019
1020         if (tc & BLK_TC_DISCARD)
1021                 rwbs[i++] = 'D';
1022         else if (tc & BLK_TC_WRITE)
1023                 rwbs[i++] = 'W';
1024         else if (t->bytes)
1025                 rwbs[i++] = 'R';
1026         else
1027                 rwbs[i++] = 'N';
1028
1029         if (tc & BLK_TC_AHEAD)
1030                 rwbs[i++] = 'A';
1031         if (tc & BLK_TC_BARRIER)
1032                 rwbs[i++] = 'B';
1033         if (tc & BLK_TC_SYNC)
1034                 rwbs[i++] = 'S';
1035         if (tc & BLK_TC_META)
1036                 rwbs[i++] = 'M';
1037 out:
1038         rwbs[i] = '\0';
1039 }
1040
1041 static inline
1042 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1043 {
1044         return (const struct blk_io_trace *)ent;
1045 }
1046
1047 static inline const void *pdu_start(const struct trace_entry *ent)
1048 {
1049         return te_blk_io_trace(ent) + 1;
1050 }
1051
1052 static inline u32 t_action(const struct trace_entry *ent)
1053 {
1054         return te_blk_io_trace(ent)->action;
1055 }
1056
1057 static inline u32 t_bytes(const struct trace_entry *ent)
1058 {
1059         return te_blk_io_trace(ent)->bytes;
1060 }
1061
1062 static inline u32 t_sec(const struct trace_entry *ent)
1063 {
1064         return te_blk_io_trace(ent)->bytes >> 9;
1065 }
1066
1067 static inline unsigned long long t_sector(const struct trace_entry *ent)
1068 {
1069         return te_blk_io_trace(ent)->sector;
1070 }
1071
1072 static inline __u16 t_error(const struct trace_entry *ent)
1073 {
1074         return te_blk_io_trace(ent)->error;
1075 }
1076
1077 static __u64 get_pdu_int(const struct trace_entry *ent)
1078 {
1079         const __u64 *val = pdu_start(ent);
1080         return be64_to_cpu(*val);
1081 }
1082
1083 static void get_pdu_remap(const struct trace_entry *ent,
1084                           struct blk_io_trace_remap *r)
1085 {
1086         const struct blk_io_trace_remap *__r = pdu_start(ent);
1087         __u64 sector_from = __r->sector_from;
1088
1089         r->device_from = be32_to_cpu(__r->device_from);
1090         r->device_to   = be32_to_cpu(__r->device_to);
1091         r->sector_from = be64_to_cpu(sector_from);
1092 }
1093
1094 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1095
1096 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1097 {
1098         char rwbs[6];
1099         unsigned long long ts  = iter->ts;
1100         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1101         unsigned secs          = (unsigned long)ts;
1102         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1103
1104         fill_rwbs(rwbs, t);
1105
1106         return trace_seq_printf(&iter->seq,
1107                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1108                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1109                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1110 }
1111
1112 static int blk_log_action(struct trace_iterator *iter, const char *act)
1113 {
1114         char rwbs[6];
1115         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1116
1117         fill_rwbs(rwbs, t);
1118         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1119                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1120 }
1121
1122 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1123 {
1124         const unsigned char *pdu_buf;
1125         int pdu_len;
1126         int i, end, ret;
1127
1128         pdu_buf = pdu_start(ent);
1129         pdu_len = te_blk_io_trace(ent)->pdu_len;
1130
1131         if (!pdu_len)
1132                 return 1;
1133
1134         /* find the last zero that needs to be printed */
1135         for (end = pdu_len - 1; end >= 0; end--)
1136                 if (pdu_buf[end])
1137                         break;
1138         end++;
1139
1140         if (!trace_seq_putc(s, '('))
1141                 return 0;
1142
1143         for (i = 0; i < pdu_len; i++) {
1144
1145                 ret = trace_seq_printf(s, "%s%02x",
1146                                        i == 0 ? "" : " ", pdu_buf[i]);
1147                 if (!ret)
1148                         return ret;
1149
1150                 /*
1151                  * stop when the rest is just zeroes and indicate so
1152                  * with a ".." appended
1153                  */
1154                 if (i == end && end != pdu_len - 1)
1155                         return trace_seq_puts(s, " ..) ");
1156         }
1157
1158         return trace_seq_puts(s, ") ");
1159 }
1160
1161 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1162 {
1163         char cmd[TASK_COMM_LEN];
1164
1165         trace_find_cmdline(ent->pid, cmd);
1166
1167         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1168                 int ret;
1169
1170                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1171                 if (!ret)
1172                         return 0;
1173                 ret = blk_log_dump_pdu(s, ent);
1174                 if (!ret)
1175                         return 0;
1176                 return trace_seq_printf(s, "[%s]\n", cmd);
1177         } else {
1178                 if (t_sec(ent))
1179                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1180                                                 t_sector(ent), t_sec(ent), cmd);
1181                 return trace_seq_printf(s, "[%s]\n", cmd);
1182         }
1183 }
1184
1185 static int blk_log_with_error(struct trace_seq *s,
1186                               const struct trace_entry *ent)
1187 {
1188         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1189                 int ret;
1190
1191                 ret = blk_log_dump_pdu(s, ent);
1192                 if (ret)
1193                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1194                 return 0;
1195         } else {
1196                 if (t_sec(ent))
1197                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1198                                                 t_sector(ent),
1199                                                 t_sec(ent), t_error(ent));
1200                 return trace_seq_printf(s, "%llu [%d]\n",
1201                                         t_sector(ent), t_error(ent));
1202         }
1203 }
1204
1205 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1206 {
1207         struct blk_io_trace_remap r = { .device_from = 0, };
1208
1209         get_pdu_remap(ent, &r);
1210         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1211                                 t_sector(ent), t_sec(ent),
1212                                 MAJOR(r.device_from), MINOR(r.device_from),
1213                                 (unsigned long long)r.sector_from);
1214 }
1215
1216 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1217 {
1218         char cmd[TASK_COMM_LEN];
1219
1220         trace_find_cmdline(ent->pid, cmd);
1221
1222         return trace_seq_printf(s, "[%s]\n", cmd);
1223 }
1224
1225 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1226 {
1227         char cmd[TASK_COMM_LEN];
1228
1229         trace_find_cmdline(ent->pid, cmd);
1230
1231         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1232 }
1233
1234 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1235 {
1236         char cmd[TASK_COMM_LEN];
1237
1238         trace_find_cmdline(ent->pid, cmd);
1239
1240         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1241                                 get_pdu_int(ent), cmd);
1242 }
1243
1244 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1245 {
1246         int ret;
1247         const struct blk_io_trace *t = te_blk_io_trace(ent);
1248
1249         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1250         if (ret)
1251                 return trace_seq_putc(s, '\n');
1252         return ret;
1253 }
1254
1255 /*
1256  * struct tracer operations
1257  */
1258
1259 static void blk_tracer_print_header(struct seq_file *m)
1260 {
1261         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1262                 return;
1263         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1264                     "#  |     |     |           |   |   |\n");
1265 }
1266
1267 static void blk_tracer_start(struct trace_array *tr)
1268 {
1269         blk_tracer_enabled = true;
1270 }
1271
1272 static int blk_tracer_init(struct trace_array *tr)
1273 {
1274         blk_tr = tr;
1275         blk_tracer_start(tr);
1276         return 0;
1277 }
1278
1279 static void blk_tracer_stop(struct trace_array *tr)
1280 {
1281         blk_tracer_enabled = false;
1282 }
1283
1284 static void blk_tracer_reset(struct trace_array *tr)
1285 {
1286         blk_tracer_stop(tr);
1287 }
1288
1289 static const struct {
1290         const char *act[2];
1291         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1292 } what2act[] = {
1293         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1294         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1295         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1296         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1297         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1298         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1299         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1300         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1301         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1302         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1303         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1304         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1305         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1306         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1307         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1308 };
1309
1310 static enum print_line_t print_one_line(struct trace_iterator *iter,
1311                                         bool classic)
1312 {
1313         struct trace_seq *s = &iter->seq;
1314         const struct blk_io_trace *t;
1315         u16 what;
1316         int ret;
1317         bool long_act;
1318         blk_log_action_t *log_action;
1319
1320         t          = te_blk_io_trace(iter->ent);
1321         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1322         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1323         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1324
1325         if (t->action == BLK_TN_MESSAGE) {
1326                 ret = log_action(iter, long_act ? "message" : "m");
1327                 if (ret)
1328                         ret = blk_log_msg(s, iter->ent);
1329                 goto out;
1330         }
1331
1332         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1333                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1334         else {
1335                 ret = log_action(iter, what2act[what].act[long_act]);
1336                 if (ret)
1337                         ret = what2act[what].print(s, iter->ent);
1338         }
1339 out:
1340         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1341 }
1342
1343 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1344                                                int flags, struct trace_event *event)
1345 {
1346         return print_one_line(iter, false);
1347 }
1348
1349 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1350 {
1351         struct trace_seq *s = &iter->seq;
1352         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1353         const int offset = offsetof(struct blk_io_trace, sector);
1354         struct blk_io_trace old = {
1355                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1356                 .time     = iter->ts,
1357         };
1358
1359         if (!trace_seq_putmem(s, &old, offset))
1360                 return 0;
1361         return trace_seq_putmem(s, &t->sector,
1362                                 sizeof(old) - offset + t->pdu_len);
1363 }
1364
1365 static enum print_line_t
1366 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1367                              struct trace_event *event)
1368 {
1369         return blk_trace_synthesize_old_trace(iter) ?
1370                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1371 }
1372
1373 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1374 {
1375         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1376                 return TRACE_TYPE_UNHANDLED;
1377
1378         return print_one_line(iter, true);
1379 }
1380
1381 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1382 {
1383         /* don't output context-info for blk_classic output */
1384         if (bit == TRACE_BLK_OPT_CLASSIC) {
1385                 if (set)
1386                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1387                 else
1388                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1389         }
1390         return 0;
1391 }
1392
1393 static struct tracer blk_tracer __read_mostly = {
1394         .name           = "blk",
1395         .init           = blk_tracer_init,
1396         .reset          = blk_tracer_reset,
1397         .start          = blk_tracer_start,
1398         .stop           = blk_tracer_stop,
1399         .print_header   = blk_tracer_print_header,
1400         .print_line     = blk_tracer_print_line,
1401         .flags          = &blk_tracer_flags,
1402         .set_flag       = blk_tracer_set_flag,
1403 };
1404
1405 static struct trace_event_functions trace_blk_event_funcs = {
1406         .trace          = blk_trace_event_print,
1407         .binary         = blk_trace_event_print_binary,
1408 };
1409
1410 static struct trace_event trace_blk_event = {
1411         .type           = TRACE_BLK,
1412         .funcs          = &trace_blk_event_funcs,
1413 };
1414
1415 static int __init init_blk_tracer(void)
1416 {
1417         if (!register_ftrace_event(&trace_blk_event)) {
1418                 pr_warning("Warning: could not register block events\n");
1419                 return 1;
1420         }
1421
1422         if (register_tracer(&blk_tracer) != 0) {
1423                 pr_warning("Warning: could not register the block tracer\n");
1424                 unregister_ftrace_event(&trace_blk_event);
1425                 return 1;
1426         }
1427
1428         return 0;
1429 }
1430
1431 device_initcall(init_blk_tracer);
1432
1433 static int blk_trace_remove_queue(struct request_queue *q)
1434 {
1435         struct blk_trace *bt;
1436
1437         bt = xchg(&q->blk_trace, NULL);
1438         if (bt == NULL)
1439                 return -EINVAL;
1440
1441         if (atomic_dec_and_test(&blk_probes_ref))
1442                 blk_unregister_tracepoints();
1443
1444         blk_trace_free(bt);
1445         return 0;
1446 }
1447
1448 /*
1449  * Setup everything required to start tracing
1450  */
1451 static int blk_trace_setup_queue(struct request_queue *q,
1452                                  struct block_device *bdev)
1453 {
1454         struct blk_trace *old_bt, *bt = NULL;
1455         int ret = -ENOMEM;
1456
1457         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1458         if (!bt)
1459                 return -ENOMEM;
1460
1461         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1462         if (!bt->msg_data)
1463                 goto free_bt;
1464
1465         bt->dev = bdev->bd_dev;
1466         bt->act_mask = (u16)-1;
1467
1468         blk_trace_setup_lba(bt, bdev);
1469
1470         old_bt = xchg(&q->blk_trace, bt);
1471         if (old_bt != NULL) {
1472                 (void)xchg(&q->blk_trace, old_bt);
1473                 ret = -EBUSY;
1474                 goto free_bt;
1475         }
1476
1477         if (atomic_inc_return(&blk_probes_ref) == 1)
1478                 blk_register_tracepoints();
1479         return 0;
1480
1481 free_bt:
1482         blk_trace_free(bt);
1483         return ret;
1484 }
1485
1486 /*
1487  * sysfs interface to enable and configure tracing
1488  */
1489
1490 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1491                                          struct device_attribute *attr,
1492                                          char *buf);
1493 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1494                                           struct device_attribute *attr,
1495                                           const char *buf, size_t count);
1496 #define BLK_TRACE_DEVICE_ATTR(_name) \
1497         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1498                     sysfs_blk_trace_attr_show, \
1499                     sysfs_blk_trace_attr_store)
1500
1501 static BLK_TRACE_DEVICE_ATTR(enable);
1502 static BLK_TRACE_DEVICE_ATTR(act_mask);
1503 static BLK_TRACE_DEVICE_ATTR(pid);
1504 static BLK_TRACE_DEVICE_ATTR(start_lba);
1505 static BLK_TRACE_DEVICE_ATTR(end_lba);
1506
1507 static struct attribute *blk_trace_attrs[] = {
1508         &dev_attr_enable.attr,
1509         &dev_attr_act_mask.attr,
1510         &dev_attr_pid.attr,
1511         &dev_attr_start_lba.attr,
1512         &dev_attr_end_lba.attr,
1513         NULL
1514 };
1515
1516 struct attribute_group blk_trace_attr_group = {
1517         .name  = "trace",
1518         .attrs = blk_trace_attrs,
1519 };
1520
1521 static const struct {
1522         int mask;
1523         const char *str;
1524 } mask_maps[] = {
1525         { BLK_TC_READ,          "read"          },
1526         { BLK_TC_WRITE,         "write"         },
1527         { BLK_TC_BARRIER,       "barrier"       },
1528         { BLK_TC_SYNC,          "sync"          },
1529         { BLK_TC_QUEUE,         "queue"         },
1530         { BLK_TC_REQUEUE,       "requeue"       },
1531         { BLK_TC_ISSUE,         "issue"         },
1532         { BLK_TC_COMPLETE,      "complete"      },
1533         { BLK_TC_FS,            "fs"            },
1534         { BLK_TC_PC,            "pc"            },
1535         { BLK_TC_AHEAD,         "ahead"         },
1536         { BLK_TC_META,          "meta"          },
1537         { BLK_TC_DISCARD,       "discard"       },
1538         { BLK_TC_DRV_DATA,      "drv_data"      },
1539 };
1540
1541 static int blk_trace_str2mask(const char *str)
1542 {
1543         int i;
1544         int mask = 0;
1545         char *buf, *s, *token;
1546
1547         buf = kstrdup(str, GFP_KERNEL);
1548         if (buf == NULL)
1549                 return -ENOMEM;
1550         s = strstrip(buf);
1551
1552         while (1) {
1553                 token = strsep(&s, ",");
1554                 if (token == NULL)
1555                         break;
1556
1557                 if (*token == '\0')
1558                         continue;
1559
1560                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1561                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1562                                 mask |= mask_maps[i].mask;
1563                                 break;
1564                         }
1565                 }
1566                 if (i == ARRAY_SIZE(mask_maps)) {
1567                         mask = -EINVAL;
1568                         break;
1569                 }
1570         }
1571         kfree(buf);
1572
1573         return mask;
1574 }
1575
1576 static ssize_t blk_trace_mask2str(char *buf, int mask)
1577 {
1578         int i;
1579         char *p = buf;
1580
1581         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1582                 if (mask & mask_maps[i].mask) {
1583                         p += sprintf(p, "%s%s",
1584                                     (p == buf) ? "" : ",", mask_maps[i].str);
1585                 }
1586         }
1587         *p++ = '\n';
1588
1589         return p - buf;
1590 }
1591
1592 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1593 {
1594         if (bdev->bd_disk == NULL)
1595                 return NULL;
1596
1597         return bdev_get_queue(bdev);
1598 }
1599
1600 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1601                                          struct device_attribute *attr,
1602                                          char *buf)
1603 {
1604         struct hd_struct *p = dev_to_part(dev);
1605         struct request_queue *q;
1606         struct block_device *bdev;
1607         ssize_t ret = -ENXIO;
1608
1609         lock_kernel();
1610         bdev = bdget(part_devt(p));
1611         if (bdev == NULL)
1612                 goto out_unlock_kernel;
1613
1614         q = blk_trace_get_queue(bdev);
1615         if (q == NULL)
1616                 goto out_bdput;
1617
1618         mutex_lock(&bdev->bd_mutex);
1619
1620         if (attr == &dev_attr_enable) {
1621                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1622                 goto out_unlock_bdev;
1623         }
1624
1625         if (q->blk_trace == NULL)
1626                 ret = sprintf(buf, "disabled\n");
1627         else if (attr == &dev_attr_act_mask)
1628                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1629         else if (attr == &dev_attr_pid)
1630                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1631         else if (attr == &dev_attr_start_lba)
1632                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1633         else if (attr == &dev_attr_end_lba)
1634                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1635
1636 out_unlock_bdev:
1637         mutex_unlock(&bdev->bd_mutex);
1638 out_bdput:
1639         bdput(bdev);
1640 out_unlock_kernel:
1641         unlock_kernel();
1642         return ret;
1643 }
1644
1645 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1646                                           struct device_attribute *attr,
1647                                           const char *buf, size_t count)
1648 {
1649         struct block_device *bdev;
1650         struct request_queue *q;
1651         struct hd_struct *p;
1652         u64 value;
1653         ssize_t ret = -EINVAL;
1654
1655         if (count == 0)
1656                 goto out;
1657
1658         if (attr == &dev_attr_act_mask) {
1659                 if (sscanf(buf, "%llx", &value) != 1) {
1660                         /* Assume it is a list of trace category names */
1661                         ret = blk_trace_str2mask(buf);
1662                         if (ret < 0)
1663                                 goto out;
1664                         value = ret;
1665                 }
1666         } else if (sscanf(buf, "%llu", &value) != 1)
1667                 goto out;
1668
1669         ret = -ENXIO;
1670
1671         lock_kernel();
1672         p = dev_to_part(dev);
1673         bdev = bdget(part_devt(p));
1674         if (bdev == NULL)
1675                 goto out_unlock_kernel;
1676
1677         q = blk_trace_get_queue(bdev);
1678         if (q == NULL)
1679                 goto out_bdput;
1680
1681         mutex_lock(&bdev->bd_mutex);
1682
1683         if (attr == &dev_attr_enable) {
1684                 if (value)
1685                         ret = blk_trace_setup_queue(q, bdev);
1686                 else
1687                         ret = blk_trace_remove_queue(q);
1688                 goto out_unlock_bdev;
1689         }
1690
1691         ret = 0;
1692         if (q->blk_trace == NULL)
1693                 ret = blk_trace_setup_queue(q, bdev);
1694
1695         if (ret == 0) {
1696                 if (attr == &dev_attr_act_mask)
1697                         q->blk_trace->act_mask = value;
1698                 else if (attr == &dev_attr_pid)
1699                         q->blk_trace->pid = value;
1700                 else if (attr == &dev_attr_start_lba)
1701                         q->blk_trace->start_lba = value;
1702                 else if (attr == &dev_attr_end_lba)
1703                         q->blk_trace->end_lba = value;
1704         }
1705
1706 out_unlock_bdev:
1707         mutex_unlock(&bdev->bd_mutex);
1708 out_bdput:
1709         bdput(bdev);
1710 out_unlock_kernel:
1711         unlock_kernel();
1712 out:
1713         return ret ? ret : count;
1714 }
1715
1716 int blk_trace_init_sysfs(struct device *dev)
1717 {
1718         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1719 }
1720
1721 void blk_trace_remove_sysfs(struct device *dev)
1722 {
1723         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1724 }
1725
1726 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1727
1728 #ifdef CONFIG_EVENT_TRACING
1729
1730 void blk_dump_cmd(char *buf, struct request *rq)
1731 {
1732         int i, end;
1733         int len = rq->cmd_len;
1734         unsigned char *cmd = rq->cmd;
1735
1736         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1737                 buf[0] = '\0';
1738                 return;
1739         }
1740
1741         for (end = len - 1; end >= 0; end--)
1742                 if (cmd[end])
1743                         break;
1744         end++;
1745
1746         for (i = 0; i < len; i++) {
1747                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1748                 if (i == end && end != len - 1) {
1749                         sprintf(buf, " ..");
1750                         break;
1751                 }
1752         }
1753 }
1754
1755 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1756 {
1757         int i = 0;
1758
1759         if (rw & WRITE)
1760                 rwbs[i++] = 'W';
1761         else if (rw & REQ_DISCARD)
1762                 rwbs[i++] = 'D';
1763         else if (bytes)
1764                 rwbs[i++] = 'R';
1765         else
1766                 rwbs[i++] = 'N';
1767
1768         if (rw & REQ_RAHEAD)
1769                 rwbs[i++] = 'A';
1770         if (rw & REQ_HARDBARRIER)
1771                 rwbs[i++] = 'B';
1772         if (rw & REQ_SYNC)
1773                 rwbs[i++] = 'S';
1774         if (rw & REQ_META)
1775                 rwbs[i++] = 'M';
1776
1777         rwbs[i] = '\0';
1778 }
1779
1780 void blk_fill_rwbs_rq(char *rwbs, struct request *rq)
1781 {
1782         int rw = rq->cmd_flags & 0x03;
1783         int bytes;
1784
1785         if (rq->cmd_flags & REQ_DISCARD)
1786                 rw |= REQ_DISCARD;
1787
1788         bytes = blk_rq_bytes(rq);
1789
1790         blk_fill_rwbs(rwbs, rw, bytes);
1791 }
1792
1793 #endif /* CONFIG_EVENT_TRACING */
1794