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/time.h>
  26 #include <trace/block.h>
  27 #include <linux/uaccess.h>
  28 #include "trace_output.h"
  29
  30 static unsigned int blktrace_seq __read_mostly = 1;
  31
  32 static struct trace_array *blk_tr;
  33 static bool blk_tracer_enabled __read_mostly;
  34
  35 /* Select an alternative, minimalistic output than the original one */
  36 #define TRACE_BLK_OPT_CLASSIC   0x1
  37
  38 static struct tracer_opt blk_tracer_opts[] = {
  39         /* Default disable the minimalistic output */
  40         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  41         { }
  42 };
  43
  44 static struct tracer_flags blk_tracer_flags = {
  45         .val  = 0,
  46         .opts = blk_tracer_opts,
  47 };
  48
  49 /* Global reference count of probes */
  50 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  51
  52 static void blk_register_tracepoints(void);
  53 static void blk_unregister_tracepoints(void);
  54
  55 /*
  56  * Send out a notify message.
  57  */
  58 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  59                        const void *data, size_t len)
  60 {
  61         struct blk_io_trace *t;
  62
  63         if (!bt->rchan)
  64                 return;
  65
  66         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  67         if (t) {
  68                 const int cpu = smp_processor_id();
  69
  70                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  71                 t->time = ktime_to_ns(ktime_get());
  72                 t->device = bt->dev;
  73                 t->action = action;
  74                 t->pid = pid;
  75                 t->cpu = cpu;
  76                 t->pdu_len = len;
  77                 memcpy((void *) t + sizeof(*t), data, len);
  78         }
  79 }
  80
  81 /*
  82  * Send out a notify for this process, if we haven't done so since a trace
  83  * started
  84  */
  85 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
  86 {
  87         tsk->btrace_seq = blktrace_seq;
  88         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
  89 }
  90
  91 static void trace_note_time(struct blk_trace *bt)
  92 {
  93         struct timespec now;
  94         unsigned long flags;
  95         u32 words[2];
  96
  97         getnstimeofday(&now);
  98         words[0] = now.tv_sec;
  99         words[1] = now.tv_nsec;
 100
 101         local_irq_save(flags);
 102         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 103         local_irq_restore(flags);
 104 }
 105
 106 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 107 {
 108         int n;
 109         va_list args;
 110         unsigned long flags;
 111         char *buf;
 112
 113         if (blk_tracer_enabled) {
 114                 va_start(args, fmt);
 115                 ftrace_vprintk(fmt, args);
 116                 va_end(args);
 117                 return;
 118         }
 119
 120         if (!bt->msg_data)
 121                 return;
 122
 123         local_irq_save(flags);
 124         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 125         va_start(args, fmt);
 126         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 127         va_end(args);
 128
 129         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 130         local_irq_restore(flags);
 131 }
 132 EXPORT_SYMBOL_GPL(__trace_note_message);
 133
 134 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 135                          pid_t pid)
 136 {
 137         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 138                 return 1;
 139         if (sector < bt->start_lba || sector > bt->end_lba)
 140                 return 1;
 141         if (bt->pid && pid != bt->pid)
 142                 return 1;
 143
 144         return 0;
 145 }
 146
 147 /*
 148  * Data direction bit lookup
 149  */
 150 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 151                                  BLK_TC_ACT(BLK_TC_WRITE) };
 152
 153 /* The ilog2() calls fall out because they're constant */
 154 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 155           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 156
 157 /*
 158  * The worker for the various blk_add_trace*() types. Fills out a
 159  * blk_io_trace structure and places it in a per-cpu subbuffer.
 160  */
 161 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 162                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 163 {
 164         struct task_struct *tsk = current;
 165         struct ring_buffer_event *event = NULL;
 166         struct blk_io_trace *t;
 167         unsigned long flags = 0;
 168         unsigned long *sequence;
 169         pid_t pid;
 170         int cpu, pc = 0;
 171
 172         if (unlikely(bt->trace_state != Blktrace_running &&
 173                      !blk_tracer_enabled))
 174                 return;
 175
 176         what |= ddir_act[rw & WRITE];
 177         what |= MASK_TC_BIT(rw, BARRIER);
 178         what |= MASK_TC_BIT(rw, SYNCIO);
 179         what |= MASK_TC_BIT(rw, AHEAD);
 180         what |= MASK_TC_BIT(rw, META);
 181         what |= MASK_TC_BIT(rw, DISCARD);
 182
 183         pid = tsk->pid;
 184         if (unlikely(act_log_check(bt, what, sector, pid)))
 185                 return;
 186         cpu = raw_smp_processor_id();
 187
 188         if (blk_tracer_enabled) {
 189                 tracing_record_cmdline(current);
 190
 191                 pc = preempt_count();
 192                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 193                                                   sizeof(*t) + pdu_len,
 194                                                   0, pc);
 195                 if (!event)
 196                         return;
 197                 t = ring_buffer_event_data(event);
 198                 goto record_it;
 199         }
 200
 201         /*
 202          * A word about the locking here - we disable interrupts to reserve
 203          * some space in the relay per-cpu buffer, to prevent an irq
 204          * from coming in and stepping on our toes.
 205          */
 206         local_irq_save(flags);
 207
 208         if (unlikely(tsk->btrace_seq != blktrace_seq))
 209                 trace_note_tsk(bt, tsk);
 210
 211         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 212         if (t) {
 213                 sequence = per_cpu_ptr(bt->sequence, cpu);
 214
 215                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 216                 t->sequence = ++(*sequence);
 217                 t->time = ktime_to_ns(ktime_get());
 218 record_it:
 219                 /*
 220                  * These two are not needed in ftrace as they are in the
 221                  * generic trace_entry, filled by tracing_generic_entry_update,
 222                  * but for the trace_event->bin() synthesizer benefit we do it
 223                  * here too.
 224                  */
 225                 t->cpu = cpu;
 226                 t->pid = pid;
 227
 228                 t->sector = sector;
 229                 t->bytes = bytes;
 230                 t->action = what;
 231                 t->device = bt->dev;
 232                 t->error = error;
 233                 t->pdu_len = pdu_len;
 234
 235                 if (pdu_len)
 236                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 237
 238                 if (blk_tracer_enabled) {
 239                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 240                         return;
 241                 }
 242         }
 243
 244         local_irq_restore(flags);
 245 }
 246
 247 static struct dentry *blk_tree_root;
 248 static DEFINE_MUTEX(blk_tree_mutex);
 249
 250 static void blk_trace_free(struct blk_trace *bt)
 251 {
 252         debugfs_remove(bt->msg_file);
 253         debugfs_remove(bt->dropped_file);
 254         relay_close(bt->rchan);
 255         free_percpu(bt->sequence);
 256         free_percpu(bt->msg_data);
 257         kfree(bt);
 258 }
 259
 260 static void blk_trace_cleanup(struct blk_trace *bt)
 261 {
 262         blk_trace_free(bt);
 263         if (atomic_dec_and_test(&blk_probes_ref))
 264                 blk_unregister_tracepoints();
 265 }
 266
 267 int blk_trace_remove(struct request_queue *q)
 268 {
 269         struct blk_trace *bt;
 270
 271         bt = xchg(&q->blk_trace, NULL);
 272         if (!bt)
 273                 return -EINVAL;
 274
 275         if (bt->trace_state != Blktrace_running)
 276                 blk_trace_cleanup(bt);
 277
 278         return 0;
 279 }
 280 EXPORT_SYMBOL_GPL(blk_trace_remove);
 281
 282 static int blk_dropped_open(struct inode *inode, struct file *filp)
 283 {
 284         filp->private_data = inode->i_private;
 285
 286         return 0;
 287 }
 288
 289 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 290                                 size_t count, loff_t *ppos)
 291 {
 292         struct blk_trace *bt = filp->private_data;
 293         char buf[16];
 294
 295         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 296
 297         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 298 }
 299
 300 static const struct file_operations blk_dropped_fops = {
 301         .owner =        THIS_MODULE,
 302         .open =         blk_dropped_open,
 303         .read =         blk_dropped_read,
 304 };
 305
 306 static int blk_msg_open(struct inode *inode, struct file *filp)
 307 {
 308         filp->private_data = inode->i_private;
 309
 310         return 0;
 311 }
 312
 313 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 314                                 size_t count, loff_t *ppos)
 315 {
 316         char *msg;
 317         struct blk_trace *bt;
 318
 319         if (count > BLK_TN_MAX_MSG)
 320                 return -EINVAL;
 321
 322         msg = kmalloc(count, GFP_KERNEL);
 323         if (msg == NULL)
 324                 return -ENOMEM;
 325
 326         if (copy_from_user(msg, buffer, count)) {
 327                 kfree(msg);
 328                 return -EFAULT;
 329         }
 330
 331         bt = filp->private_data;
 332         __trace_note_message(bt, "%s", msg);
 333         kfree(msg);
 334
 335         return count;
 336 }
 337
 338 static const struct file_operations blk_msg_fops = {
 339         .owner =        THIS_MODULE,
 340         .open =         blk_msg_open,
 341         .write =        blk_msg_write,
 342 };
 343
 344 /*
 345  * Keep track of how many times we encountered a full subbuffer, to aid
 346  * the user space app in telling how many lost events there were.
 347  */
 348 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 349                                      void *prev_subbuf, size_t prev_padding)
 350 {
 351         struct blk_trace *bt;
 352
 353         if (!relay_buf_full(buf))
 354                 return 1;
 355
 356         bt = buf->chan->private_data;
 357         atomic_inc(&bt->dropped);
 358         return 0;
 359 }
 360
 361 static int blk_remove_buf_file_callback(struct dentry *dentry)
 362 {
 363         struct dentry *parent = dentry->d_parent;
 364         debugfs_remove(dentry);
 365
 366         /*
 367         * this will fail for all but the last file, but that is ok. what we
 368         * care about is the top level buts->name directory going away, when
 369         * the last trace file is gone. Then we don't have to rmdir() that
 370         * manually on trace stop, so it nicely solves the issue with
 371         * force killing of running traces.
 372         */
 373
 374         debugfs_remove(parent);
 375         return 0;
 376 }
 377
 378 static struct dentry *blk_create_buf_file_callback(const char *filename,
 379                                                    struct dentry *parent,
 380                                                    int mode,
 381                                                    struct rchan_buf *buf,
 382                                                    int *is_global)
 383 {
 384         return debugfs_create_file(filename, mode, parent, buf,
 385                                         &relay_file_operations);
 386 }
 387
 388 static struct rchan_callbacks blk_relay_callbacks = {
 389         .subbuf_start           = blk_subbuf_start_callback,
 390         .create_buf_file        = blk_create_buf_file_callback,
 391         .remove_buf_file        = blk_remove_buf_file_callback,
 392 };
 393
 394 /*
 395  * Setup everything required to start tracing
 396  */
 397 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 398                         struct blk_user_trace_setup *buts)
 399 {
 400         struct blk_trace *old_bt, *bt = NULL;
 401         struct dentry *dir = NULL;
 402         int ret, i;
 403
 404         if (!buts->buf_size || !buts->buf_nr)
 405                 return -EINVAL;
 406
 407         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 408         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 409
 410         /*
 411          * some device names have larger paths - convert the slashes
 412          * to underscores for this to work as expected
 413          */
 414         for (i = 0; i < strlen(buts->name); i++)
 415                 if (buts->name[i] == '/')
 416                         buts->name[i] = '_';
 417
 418         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 419         if (!bt)
 420                 return -ENOMEM;
 421
 422         ret = -ENOMEM;
 423         bt->sequence = alloc_percpu(unsigned long);
 424         if (!bt->sequence)
 425                 goto err;
 426
 427         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 428         if (!bt->msg_data)
 429                 goto err;
 430
 431         ret = -ENOENT;
 432
 433         mutex_lock(&blk_tree_mutex);
 434         if (!blk_tree_root) {
 435                 blk_tree_root = debugfs_create_dir("block", NULL);
 436                 if (!blk_tree_root) {
 437                         mutex_unlock(&blk_tree_mutex);
 438                         goto err;
 439                 }
 440         }
 441         mutex_unlock(&blk_tree_mutex);
 442
 443         dir = debugfs_create_dir(buts->name, blk_tree_root);
 444
 445         if (!dir)
 446                 goto err;
 447
 448         bt->dir = dir;
 449         bt->dev = dev;
 450         atomic_set(&bt->dropped, 0);
 451
 452         ret = -EIO;
 453         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 454                                                &blk_dropped_fops);
 455         if (!bt->dropped_file)
 456                 goto err;
 457
 458         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 459         if (!bt->msg_file)
 460                 goto err;
 461
 462         bt->rchan = relay_open("trace", dir, buts->buf_size,
 463                                 buts->buf_nr, &blk_relay_callbacks, bt);
 464         if (!bt->rchan)
 465                 goto err;
 466
 467         bt->act_mask = buts->act_mask;
 468         if (!bt->act_mask)
 469                 bt->act_mask = (u16) -1;
 470
 471         bt->start_lba = buts->start_lba;
 472         bt->end_lba = buts->end_lba;
 473         if (!bt->end_lba)
 474                 bt->end_lba = -1ULL;
 475
 476         bt->pid = buts->pid;
 477         bt->trace_state = Blktrace_setup;
 478
 479         ret = -EBUSY;
 480         old_bt = xchg(&q->blk_trace, bt);
 481         if (old_bt) {
 482                 (void) xchg(&q->blk_trace, old_bt);
 483                 goto err;
 484         }
 485
 486         if (atomic_inc_return(&blk_probes_ref) == 1)
 487                 blk_register_tracepoints();
 488
 489         return 0;
 490 err:
 491         blk_trace_free(bt);
 492         return ret;
 493 }
 494
 495 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 496                     char __user *arg)
 497 {
 498         struct blk_user_trace_setup buts;
 499         int ret;
 500
 501         ret = copy_from_user(&buts, arg, sizeof(buts));
 502         if (ret)
 503                 return -EFAULT;
 504
 505         ret = do_blk_trace_setup(q, name, dev, &buts);
 506         if (ret)
 507                 return ret;
 508
 509         if (copy_to_user(arg, &buts, sizeof(buts)))
 510                 return -EFAULT;
 511
 512         return 0;
 513 }
 514 EXPORT_SYMBOL_GPL(blk_trace_setup);
 515
 516 int blk_trace_startstop(struct request_queue *q, int start)
 517 {
 518         int ret;
 519         struct blk_trace *bt = q->blk_trace;
 520
 521         if (bt == NULL)
 522                 return -EINVAL;
 523
 524         /*
 525          * For starting a trace, we can transition from a setup or stopped
 526          * trace. For stopping a trace, the state must be running
 527          */
 528         ret = -EINVAL;
 529         if (start) {
 530                 if (bt->trace_state == Blktrace_setup ||
 531                     bt->trace_state == Blktrace_stopped) {
 532                         blktrace_seq++;
 533                         smp_mb();
 534                         bt->trace_state = Blktrace_running;
 535
 536                         trace_note_time(bt);
 537                         ret = 0;
 538                 }
 539         } else {
 540                 if (bt->trace_state == Blktrace_running) {
 541                         bt->trace_state = Blktrace_stopped;
 542                         relay_flush(bt->rchan);
 543                         ret = 0;
 544                 }
 545         }
 546
 547         return ret;
 548 }
 549 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 550
 551 /**
 552  * blk_trace_ioctl: - handle the ioctls associated with tracing
 553  * @bdev:       the block device
 554  * @cmd:        the ioctl cmd
 555  * @arg:        the argument data, if any
 556  *
 557  **/
 558 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 559 {
 560         struct request_queue *q;
 561         int ret, start = 0;
 562         char b[BDEVNAME_SIZE];
 563
 564         q = bdev_get_queue(bdev);
 565         if (!q)
 566                 return -ENXIO;
 567
 568         mutex_lock(&bdev->bd_mutex);
 569
 570         switch (cmd) {
 571         case BLKTRACESETUP:
 572                 bdevname(bdev, b);
 573                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 574                 break;
 575         case BLKTRACESTART:
 576                 start = 1;
 577         case BLKTRACESTOP:
 578                 ret = blk_trace_startstop(q, start);
 579                 break;
 580         case BLKTRACETEARDOWN:
 581                 ret = blk_trace_remove(q);
 582                 break;
 583         default:
 584                 ret = -ENOTTY;
 585                 break;
 586         }
 587
 588         mutex_unlock(&bdev->bd_mutex);
 589         return ret;
 590 }
 591
 592 /**
 593  * blk_trace_shutdown: - stop and cleanup trace structures
 594  * @q:    the request queue associated with the device
 595  *
 596  **/
 597 void blk_trace_shutdown(struct request_queue *q)
 598 {
 599         if (q->blk_trace) {
 600                 blk_trace_startstop(q, 0);
 601                 blk_trace_remove(q);
 602         }
 603 }
 604
 605 /*
 606  * blktrace probes
 607  */
 608
 609 /**
 610  * blk_add_trace_rq - Add a trace for a request oriented action
 611  * @q:          queue the io is for
 612  * @rq:         the source request
 613  * @what:       the action
 614  *
 615  * Description:
 616  *     Records an action against a request. Will log the bio offset + size.
 617  *
 618  **/
 619 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 620                                     u32 what)
 621 {
 622         struct blk_trace *bt = q->blk_trace;
 623         int rw = rq->cmd_flags & 0x03;
 624
 625         if (likely(!bt))
 626                 return;
 627
 628         if (blk_discard_rq(rq))
 629                 rw |= (1 << BIO_RW_DISCARD);
 630
 631         if (blk_pc_request(rq)) {
 632                 what |= BLK_TC_ACT(BLK_TC_PC);
 633                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 634                                 sizeof(rq->cmd), rq->cmd);
 635         } else  {
 636                 what |= BLK_TC_ACT(BLK_TC_FS);
 637                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 638                                 rw, what, rq->errors, 0, NULL);
 639         }
 640 }
 641
 642 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 643 {
 644         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 645 }
 646
 647 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 648 {
 649         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 650 }
 651
 652 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 653 {
 654         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 655 }
 656
 657 static void blk_add_trace_rq_requeue(struct request_queue *q,
 658                                      struct request *rq)
 659 {
 660         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 661 }
 662
 663 static void blk_add_trace_rq_complete(struct request_queue *q,
 664                                       struct request *rq)
 665 {
 666         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 667 }
 668
 669 /**
 670  * blk_add_trace_bio - Add a trace for a bio oriented action
 671  * @q:          queue the io is for
 672  * @bio:        the source bio
 673  * @what:       the action
 674  *
 675  * Description:
 676  *     Records an action against a bio. Will log the bio offset + size.
 677  *
 678  **/
 679 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 680                                      u32 what)
 681 {
 682         struct blk_trace *bt = q->blk_trace;
 683
 684         if (likely(!bt))
 685                 return;
 686
 687         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 688                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 689 }
 690
 691 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 692 {
 693         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 694 }
 695
 696 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 697 {
 698         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 699 }
 700
 701 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 702                                         struct bio *bio)
 703 {
 704         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 705 }
 706
 707 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 708                                          struct bio *bio)
 709 {
 710         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 711 }
 712
 713 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 714 {
 715         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 716 }
 717
 718 static void blk_add_trace_getrq(struct request_queue *q,
 719                                 struct bio *bio, int rw)
 720 {
 721         if (bio)
 722                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 723         else {
 724                 struct blk_trace *bt = q->blk_trace;
 725
 726                 if (bt)
 727                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 728         }
 729 }
 730
 731
 732 static void blk_add_trace_sleeprq(struct request_queue *q,
 733                                   struct bio *bio, int rw)
 734 {
 735         if (bio)
 736                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 737         else {
 738                 struct blk_trace *bt = q->blk_trace;
 739
 740                 if (bt)
 741                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 742                                         0, 0, NULL);
 743         }
 744 }
 745
 746 static void blk_add_trace_plug(struct request_queue *q)
 747 {
 748         struct blk_trace *bt = q->blk_trace;
 749
 750         if (bt)
 751                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 752 }
 753
 754 static void blk_add_trace_unplug_io(struct request_queue *q)
 755 {
 756         struct blk_trace *bt = q->blk_trace;
 757
 758         if (bt) {
 759                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 760                 __be64 rpdu = cpu_to_be64(pdu);
 761
 762                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 763                                 sizeof(rpdu), &rpdu);
 764         }
 765 }
 766
 767 static void blk_add_trace_unplug_timer(struct request_queue *q)
 768 {
 769         struct blk_trace *bt = q->blk_trace;
 770
 771         if (bt) {
 772                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 773                 __be64 rpdu = cpu_to_be64(pdu);
 774
 775                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 776                                 sizeof(rpdu), &rpdu);
 777         }
 778 }
 779
 780 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 781                                 unsigned int pdu)
 782 {
 783         struct blk_trace *bt = q->blk_trace;
 784
 785         if (bt) {
 786                 __be64 rpdu = cpu_to_be64(pdu);
 787
 788                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 789                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 790                                 sizeof(rpdu), &rpdu);
 791         }
 792 }
 793
 794 /**
 795  * blk_add_trace_remap - Add a trace for a remap operation
 796  * @q:          queue the io is for
 797  * @bio:        the source bio
 798  * @dev:        target device
 799  * @from:       source sector
 800  * @to:         target sector
 801  *
 802  * Description:
 803  *     Device mapper or raid target sometimes need to split a bio because
 804  *     it spans a stripe (or similar). Add a trace for that action.
 805  *
 806  **/
 807 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 808                                        dev_t dev, sector_t from, sector_t to)
 809 {
 810         struct blk_trace *bt = q->blk_trace;
 811         struct blk_io_trace_remap r;
 812
 813         if (likely(!bt))
 814                 return;
 815
 816         r.device = cpu_to_be32(dev);
 817         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 818         r.sector = cpu_to_be64(to);
 819
 820         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 821                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 822 }
 823
 824 /**
 825  * blk_add_driver_data - Add binary message with driver-specific data
 826  * @q:          queue the io is for
 827  * @rq:         io request
 828  * @data:       driver-specific data
 829  * @len:        length of driver-specific data
 830  *
 831  * Description:
 832  *     Some drivers might want to write driver-specific data per request.
 833  *
 834  **/
 835 void blk_add_driver_data(struct request_queue *q,
 836                          struct request *rq,
 837                          void *data, size_t len)
 838 {
 839         struct blk_trace *bt = q->blk_trace;
 840
 841         if (likely(!bt))
 842                 return;
 843
 844         if (blk_pc_request(rq))
 845                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 846                                 rq->errors, len, data);
 847         else
 848                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 849                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 850 }
 851 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 852
 853 static void blk_register_tracepoints(void)
 854 {
 855         int ret;
 856
 857         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 858         WARN_ON(ret);
 859         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 860         WARN_ON(ret);
 861         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 862         WARN_ON(ret);
 863         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 864         WARN_ON(ret);
 865         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 866         WARN_ON(ret);
 867         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 868         WARN_ON(ret);
 869         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 870         WARN_ON(ret);
 871         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 872         WARN_ON(ret);
 873         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 874         WARN_ON(ret);
 875         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 876         WARN_ON(ret);
 877         ret = register_trace_block_getrq(blk_add_trace_getrq);
 878         WARN_ON(ret);
 879         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 880         WARN_ON(ret);
 881         ret = register_trace_block_plug(blk_add_trace_plug);
 882         WARN_ON(ret);
 883         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 884         WARN_ON(ret);
 885         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 886         WARN_ON(ret);
 887         ret = register_trace_block_split(blk_add_trace_split);
 888         WARN_ON(ret);
 889         ret = register_trace_block_remap(blk_add_trace_remap);
 890         WARN_ON(ret);
 891 }
 892
 893 static void blk_unregister_tracepoints(void)
 894 {
 895         unregister_trace_block_remap(blk_add_trace_remap);
 896         unregister_trace_block_split(blk_add_trace_split);
 897         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 898         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 899         unregister_trace_block_plug(blk_add_trace_plug);
 900         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 901         unregister_trace_block_getrq(blk_add_trace_getrq);
 902         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 903         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 904         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 905         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 906         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 907         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 908         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 909         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 910         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 911         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 912
 913         tracepoint_synchronize_unregister();
 914 }
 915
 916 /*
 917  * struct blk_io_tracer formatting routines
 918  */
 919
 920 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 921 {
 922         int i = 0;
 923         int tc = t->action >> BLK_TC_SHIFT;
 924
 925         if (tc & BLK_TC_DISCARD)
 926                 rwbs[i++] = 'D';
 927         else if (tc & BLK_TC_WRITE)
 928                 rwbs[i++] = 'W';
 929         else if (t->bytes)
 930                 rwbs[i++] = 'R';
 931         else
 932                 rwbs[i++] = 'N';
 933
 934         if (tc & BLK_TC_AHEAD)
 935                 rwbs[i++] = 'A';
 936         if (tc & BLK_TC_BARRIER)
 937                 rwbs[i++] = 'B';
 938         if (tc & BLK_TC_SYNC)
 939                 rwbs[i++] = 'S';
 940         if (tc & BLK_TC_META)
 941                 rwbs[i++] = 'M';
 942
 943         rwbs[i] = '\0';
 944 }
 945
 946 static inline
 947 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 948 {
 949         return (const struct blk_io_trace *)ent;
 950 }
 951
 952 static inline const void *pdu_start(const struct trace_entry *ent)
 953 {
 954         return te_blk_io_trace(ent) + 1;
 955 }
 956
 957 static inline u32 t_sec(const struct trace_entry *ent)
 958 {
 959         return te_blk_io_trace(ent)->bytes >> 9;
 960 }
 961
 962 static inline unsigned long long t_sector(const struct trace_entry *ent)
 963 {
 964         return te_blk_io_trace(ent)->sector;
 965 }
 966
 967 static inline __u16 t_error(const struct trace_entry *ent)
 968 {
 969         return te_blk_io_trace(ent)->error;
 970 }
 971
 972 static __u64 get_pdu_int(const struct trace_entry *ent)
 973 {
 974         const __u64 *val = pdu_start(ent);
 975         return be64_to_cpu(*val);
 976 }
 977
 978 static void get_pdu_remap(const struct trace_entry *ent,
 979                           struct blk_io_trace_remap *r)
 980 {
 981         const struct blk_io_trace_remap *__r = pdu_start(ent);
 982         __u64 sector = __r->sector;
 983
 984         r->device = be32_to_cpu(__r->device);
 985         r->device_from = be32_to_cpu(__r->device_from);
 986         r->sector = be64_to_cpu(sector);
 987 }
 988
 989 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
 990
 991 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
 992 {
 993         char rwbs[6];
 994         unsigned long long ts  = iter->ts;
 995         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
 996         unsigned secs          = (unsigned long)ts;
 997         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
 998
 999         fill_rwbs(rwbs, t);
1000
1001         return trace_seq_printf(&iter->seq,
1002                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1003                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1004                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1005 }
1006
1007 static int blk_log_action(struct trace_iterator *iter, const char *act)
1008 {
1009         char rwbs[6];
1010         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1011
1012         fill_rwbs(rwbs, t);
1013         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1014                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1015 }
1016
1017 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1018 {
1019         char cmd[TASK_COMM_LEN];
1020
1021         trace_find_cmdline(ent->pid, cmd);
1022
1023         if (t_sec(ent))
1024                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1025                                         t_sector(ent), t_sec(ent), cmd);
1026         return trace_seq_printf(s, "[%s]\n", cmd);
1027 }
1028
1029 static int blk_log_with_error(struct trace_seq *s,
1030                               const struct trace_entry *ent)
1031 {
1032         if (t_sec(ent))
1033                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1034                                         t_sec(ent), t_error(ent));
1035         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1036 }
1037
1038 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1039 {
1040         struct blk_io_trace_remap r = { .device = 0, };
1041
1042         get_pdu_remap(ent, &r);
1043         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1044                                t_sector(ent),
1045                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1046                                (unsigned long long)r.sector);
1047 }
1048
1049 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1050 {
1051         char cmd[TASK_COMM_LEN];
1052
1053         trace_find_cmdline(ent->pid, cmd);
1054
1055         return trace_seq_printf(s, "[%s]\n", cmd);
1056 }
1057
1058 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1059 {
1060         char cmd[TASK_COMM_LEN];
1061
1062         trace_find_cmdline(ent->pid, cmd);
1063
1064         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1065 }
1066
1067 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1068 {
1069         char cmd[TASK_COMM_LEN];
1070
1071         trace_find_cmdline(ent->pid, cmd);
1072
1073         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1074                                 get_pdu_int(ent), cmd);
1075 }
1076
1077 /*
1078  * struct tracer operations
1079  */
1080
1081 static void blk_tracer_print_header(struct seq_file *m)
1082 {
1083         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1084                 return;
1085         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1086                     "#  |     |     |           |   |   |\n");
1087 }
1088
1089 static void blk_tracer_start(struct trace_array *tr)
1090 {
1091         blk_tracer_enabled = true;
1092         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1093 }
1094
1095 static int blk_tracer_init(struct trace_array *tr)
1096 {
1097         blk_tr = tr;
1098         blk_tracer_start(tr);
1099         return 0;
1100 }
1101
1102 static void blk_tracer_stop(struct trace_array *tr)
1103 {
1104         blk_tracer_enabled = false;
1105         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1106 }
1107
1108 static void blk_tracer_reset(struct trace_array *tr)
1109 {
1110         blk_tracer_stop(tr);
1111 }
1112
1113 static const struct {
1114         const char *act[2];
1115         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1116 } what2act[] = {
1117         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1118         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1119         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1120         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1121         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1122         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1123         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1124         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1125         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1126         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1127         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1128         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1129         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1130         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1131         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1132 };
1133
1134 static enum print_line_t print_one_line(struct trace_iterator *iter,
1135                                         bool classic)
1136 {
1137         struct trace_seq *s = &iter->seq;
1138         const struct blk_io_trace *t;
1139         u16 what;
1140         int ret;
1141         bool long_act;
1142         blk_log_action_t *log_action;
1143
1144         t          = te_blk_io_trace(iter->ent);
1145         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1146         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1147         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1148
1149         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1150                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1151         else {
1152                 ret = log_action(iter, what2act[what].act[long_act]);
1153                 if (ret)
1154                         ret = what2act[what].print(s, iter->ent);
1155         }
1156
1157         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1158 }
1159
1160 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1161                                                int flags)
1162 {
1163         if (!trace_print_context(iter))
1164                 return TRACE_TYPE_PARTIAL_LINE;
1165
1166         return print_one_line(iter, false);
1167 }
1168
1169 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1170 {
1171         struct trace_seq *s = &iter->seq;
1172         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1173         const int offset = offsetof(struct blk_io_trace, sector);
1174         struct blk_io_trace old = {
1175                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1176                 .time     = iter->ts,
1177         };
1178
1179         if (!trace_seq_putmem(s, &old, offset))
1180                 return 0;
1181         return trace_seq_putmem(s, &t->sector,
1182                                 sizeof(old) - offset + t->pdu_len);
1183 }
1184
1185 static enum print_line_t
1186 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1187 {
1188         return blk_trace_synthesize_old_trace(iter) ?
1189                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1190 }
1191
1192 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1193 {
1194         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1195                 return TRACE_TYPE_UNHANDLED;
1196
1197         return print_one_line(iter, true);
1198 }
1199
1200 static struct tracer blk_tracer __read_mostly = {
1201         .name           = "blk",
1202         .init           = blk_tracer_init,
1203         .reset          = blk_tracer_reset,
1204         .start          = blk_tracer_start,
1205         .stop           = blk_tracer_stop,
1206         .print_header   = blk_tracer_print_header,
1207         .print_line     = blk_tracer_print_line,
1208         .flags          = &blk_tracer_flags,
1209 };
1210
1211 static struct trace_event trace_blk_event = {
1212         .type           = TRACE_BLK,
1213         .trace          = blk_trace_event_print,
1214         .binary         = blk_trace_event_print_binary,
1215 };
1216
1217 static int __init init_blk_tracer(void)
1218 {
1219         if (!register_ftrace_event(&trace_blk_event)) {
1220                 pr_warning("Warning: could not register block events\n");
1221                 return 1;
1222         }
1223
1224         if (register_tracer(&blk_tracer) != 0) {
1225                 pr_warning("Warning: could not register the block tracer\n");
1226                 unregister_ftrace_event(&trace_blk_event);
1227                 return 1;
1228         }
1229
1230         return 0;
1231 }
1232
1233 device_initcall(init_blk_tracer);
1234
1235 static int blk_trace_remove_queue(struct request_queue *q)
1236 {
1237         struct blk_trace *bt;
1238
1239         bt = xchg(&q->blk_trace, NULL);
1240         if (bt == NULL)
1241                 return -EINVAL;
1242
1243         if (atomic_dec_and_test(&blk_probes_ref))
1244                 blk_unregister_tracepoints();
1245
1246         blk_trace_free(bt);
1247         return 0;
1248 }
1249
1250 /*
1251  * Setup everything required to start tracing
1252  */
1253 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1254 {
1255         struct blk_trace *old_bt, *bt = NULL;
1256
1257         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1258         if (!bt)
1259                 return -ENOMEM;
1260
1261         bt->dev = dev;
1262         bt->act_mask = (u16)-1;
1263         bt->end_lba = -1ULL;
1264
1265         old_bt = xchg(&q->blk_trace, bt);
1266         if (old_bt != NULL) {
1267                 (void)xchg(&q->blk_trace, old_bt);
1268                 kfree(bt);
1269                 return -EBUSY;
1270         }
1271
1272         if (atomic_inc_return(&blk_probes_ref) == 1)
1273                 blk_register_tracepoints();
1274
1275         return 0;
1276 }
1277
1278 /*
1279  * sysfs interface to enable and configure tracing
1280  */
1281
1282 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1283                                          struct device_attribute *attr,
1284                                          char *buf);
1285 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1286                                           struct device_attribute *attr,
1287                                           const char *buf, size_t count);
1288 #define BLK_TRACE_DEVICE_ATTR(_name) \
1289         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1290                     sysfs_blk_trace_attr_show, \
1291                     sysfs_blk_trace_attr_store)
1292
1293 static BLK_TRACE_DEVICE_ATTR(enable);
1294 static BLK_TRACE_DEVICE_ATTR(act_mask);
1295 static BLK_TRACE_DEVICE_ATTR(pid);
1296 static BLK_TRACE_DEVICE_ATTR(start_lba);
1297 static BLK_TRACE_DEVICE_ATTR(end_lba);
1298
1299 static struct attribute *blk_trace_attrs[] = {
1300         &dev_attr_enable.attr,
1301         &dev_attr_act_mask.attr,
1302         &dev_attr_pid.attr,
1303         &dev_attr_start_lba.attr,
1304         &dev_attr_end_lba.attr,
1305         NULL
1306 };
1307
1308 struct attribute_group blk_trace_attr_group = {
1309         .name  = "trace",
1310         .attrs = blk_trace_attrs,
1311 };
1312
1313 static const struct {
1314         int mask;
1315         const char *str;
1316 } mask_maps[] = {
1317         { BLK_TC_READ,          "read"          },
1318         { BLK_TC_WRITE,         "write"         },
1319         { BLK_TC_BARRIER,       "barrier"       },
1320         { BLK_TC_SYNC,          "sync"          },
1321         { BLK_TC_QUEUE,         "queue"         },
1322         { BLK_TC_REQUEUE,       "requeue"       },
1323         { BLK_TC_ISSUE,         "issue"         },
1324         { BLK_TC_COMPLETE,      "complete"      },
1325         { BLK_TC_FS,            "fs"            },
1326         { BLK_TC_PC,            "pc"            },
1327         { BLK_TC_AHEAD,         "ahead"         },
1328         { BLK_TC_META,          "meta"          },
1329         { BLK_TC_DISCARD,       "discard"       },
1330         { BLK_TC_DRV_DATA,      "drv_data"      },
1331 };
1332
1333 static int blk_trace_str2mask(const char *str)
1334 {
1335         int i;
1336         int mask = 0;
1337         char *s, *token;
1338
1339         s = kstrdup(str, GFP_KERNEL);
1340         if (s == NULL)
1341                 return -ENOMEM;
1342         s = strstrip(s);
1343
1344         while (1) {
1345                 token = strsep(&s, ",");
1346                 if (token == NULL)
1347                         break;
1348
1349                 if (*token == '\0')
1350                         continue;
1351
1352                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1353                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1354                                 mask |= mask_maps[i].mask;
1355                                 break;
1356                         }
1357                 }
1358                 if (i == ARRAY_SIZE(mask_maps)) {
1359                         mask = -EINVAL;
1360                         break;
1361                 }
1362         }
1363         kfree(s);
1364
1365         return mask;
1366 }
1367
1368 static ssize_t blk_trace_mask2str(char *buf, int mask)
1369 {
1370         int i;
1371         char *p = buf;
1372
1373         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1374                 if (mask & mask_maps[i].mask) {
1375                         p += sprintf(p, "%s%s",
1376                                     (p == buf) ? "" : ",", mask_maps[i].str);
1377                 }
1378         }
1379         *p++ = '\n';
1380
1381         return p - buf;
1382 }
1383
1384 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1385 {
1386         if (bdev->bd_disk == NULL)
1387                 return NULL;
1388
1389         return bdev_get_queue(bdev);
1390 }
1391
1392 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1393                                          struct device_attribute *attr,
1394                                          char *buf)
1395 {
1396         struct hd_struct *p = dev_to_part(dev);
1397         struct request_queue *q;
1398         struct block_device *bdev;
1399         ssize_t ret = -ENXIO;
1400
1401         lock_kernel();
1402         bdev = bdget(part_devt(p));
1403         if (bdev == NULL)
1404                 goto out_unlock_kernel;
1405
1406         q = blk_trace_get_queue(bdev);
1407         if (q == NULL)
1408                 goto out_bdput;
1409
1410         mutex_lock(&bdev->bd_mutex);
1411
1412         if (attr == &dev_attr_enable) {
1413                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1414                 goto out_unlock_bdev;
1415         }
1416
1417         if (q->blk_trace == NULL)
1418                 ret = sprintf(buf, "disabled\n");
1419         else if (attr == &dev_attr_act_mask)
1420                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1421         else if (attr == &dev_attr_pid)
1422                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1423         else if (attr == &dev_attr_start_lba)
1424                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1425         else if (attr == &dev_attr_end_lba)
1426                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1427
1428 out_unlock_bdev:
1429         mutex_unlock(&bdev->bd_mutex);
1430 out_bdput:
1431         bdput(bdev);
1432 out_unlock_kernel:
1433         unlock_kernel();
1434         return ret;
1435 }
1436
1437 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1438                                           struct device_attribute *attr,
1439                                           const char *buf, size_t count)
1440 {
1441         struct block_device *bdev;
1442         struct request_queue *q;
1443         struct hd_struct *p;
1444         u64 value;
1445         ssize_t ret = -EINVAL;
1446
1447         if (count == 0)
1448                 goto out;
1449
1450         if (attr == &dev_attr_act_mask) {
1451                 if (sscanf(buf, "%llx", &value) != 1) {
1452                         /* Assume it is a list of trace category names */
1453                         ret = blk_trace_str2mask(buf);
1454                         if (ret < 0)
1455                                 goto out;
1456                         value = ret;
1457                 }
1458         } else if (sscanf(buf, "%llu", &value) != 1)
1459                 goto out;
1460
1461         ret = -ENXIO;
1462
1463         lock_kernel();
1464         p = dev_to_part(dev);
1465         bdev = bdget(part_devt(p));
1466         if (bdev == NULL)
1467                 goto out_unlock_kernel;
1468
1469         q = blk_trace_get_queue(bdev);
1470         if (q == NULL)
1471                 goto out_bdput;
1472
1473         mutex_lock(&bdev->bd_mutex);
1474
1475         if (attr == &dev_attr_enable) {
1476                 if (value)
1477                         ret = blk_trace_setup_queue(q, bdev->bd_dev);
1478                 else
1479                         ret = blk_trace_remove_queue(q);
1480                 goto out_unlock_bdev;
1481         }
1482
1483         ret = 0;
1484         if (q->blk_trace == NULL)
1485                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1486
1487         if (ret == 0) {
1488                 if (attr == &dev_attr_act_mask)
1489                         q->blk_trace->act_mask = value;
1490                 else if (attr == &dev_attr_pid)
1491                         q->blk_trace->pid = value;
1492                 else if (attr == &dev_attr_start_lba)
1493                         q->blk_trace->start_lba = value;
1494                 else if (attr == &dev_attr_end_lba)
1495                         q->blk_trace->end_lba = value;
1496         }
1497
1498 out_unlock_bdev:
1499         mutex_unlock(&bdev->bd_mutex);
1500 out_bdput:
1501         bdput(bdev);
1502 out_unlock_kernel:
1503         unlock_kernel();
1504 out:
1505         return ret ? ret : count;
1506 }
1507