block/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 <asm/uaccess.h>
  28 #include <../kernel/trace/trace_output.h>
  29
  30 static unsigned int blktrace_seq __read_mostly = 1;
  31
  32 static struct trace_array *blk_tr;
  33 static int __read_mostly  blk_tracer_enabled;
  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 DEFINE_MUTEX(blk_probe_mutex);
  51 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  52
  53 static int blk_register_tracepoints(void);
  54 static void blk_unregister_tracepoints(void);
  55
  56 /*
  57  * Send out a notify message.
  58  */
  59 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  60                        const void *data, size_t len)
  61 {
  62         struct blk_io_trace *t;
  63
  64         if (!bt->rchan)
  65                 return;
  66
  67         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  68         if (t) {
  69                 const int cpu = smp_processor_id();
  70
  71                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  72                 t->time = ktime_to_ns(ktime_get());
  73                 t->device = bt->dev;
  74                 t->action = action;
  75                 t->pid = pid;
  76                 t->cpu = cpu;
  77                 t->pdu_len = len;
  78                 memcpy((void *) t + sizeof(*t), data, len);
  79         }
  80 }
  81
  82 /*
  83  * Send out a notify for this process, if we haven't done so since a trace
  84  * started
  85  */
  86 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
  87 {
  88         tsk->btrace_seq = blktrace_seq;
  89         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
  90 }
  91
  92 static void trace_note_time(struct blk_trace *bt)
  93 {
  94         struct timespec now;
  95         unsigned long flags;
  96         u32 words[2];
  97
  98         getnstimeofday(&now);
  99         words[0] = now.tv_sec;
 100         words[1] = now.tv_nsec;
 101
 102         local_irq_save(flags);
 103         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 104         local_irq_restore(flags);
 105 }
 106
 107 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 108 {
 109         int n;
 110         va_list args;
 111         unsigned long flags;
 112         char *buf;
 113
 114         if (blk_tr) {
 115                 va_start(args, fmt);
 116                 ftrace_vprintk(fmt, args);
 117                 va_end(args);
 118                 return;
 119         }
 120
 121         if (!bt->msg_data)
 122                 return;
 123
 124         local_irq_save(flags);
 125         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 126         va_start(args, fmt);
 127         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 128         va_end(args);
 129
 130         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 131         local_irq_restore(flags);
 132 }
 133 EXPORT_SYMBOL_GPL(__trace_note_message);
 134
 135 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 136                          pid_t pid)
 137 {
 138         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 139                 return 1;
 140         if (sector < bt->start_lba || sector > bt->end_lba)
 141                 return 1;
 142         if (bt->pid && pid != bt->pid)
 143                 return 1;
 144
 145         return 0;
 146 }
 147
 148 /*
 149  * Data direction bit lookup
 150  */
 151 static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), 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;
 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, SYNC);
 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_tr) {
 189                 struct trace_entry *ent;
 190                 tracing_record_cmdline(current);
 191
 192                 event = ring_buffer_lock_reserve(blk_tr->buffer,
 193                                                  sizeof(*t) + pdu_len, &flags);
 194                 if (!event)
 195                         return;
 196
 197                 ent = ring_buffer_event_data(event);
 198                 t = (struct blk_io_trace *)ent;
 199                 pc = preempt_count();
 200                 tracing_generic_entry_update(ent, 0, pc);
 201                 ent->type = TRACE_BLK;
 202                 goto record_it;
 203         }
 204
 205         /*
 206          * A word about the locking here - we disable interrupts to reserve
 207          * some space in the relay per-cpu buffer, to prevent an irq
 208          * from coming in and stepping on our toes.
 209          */
 210         local_irq_save(flags);
 211
 212         if (unlikely(tsk->btrace_seq != blktrace_seq))
 213                 trace_note_tsk(bt, tsk);
 214
 215         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 216         if (t) {
 217                 sequence = per_cpu_ptr(bt->sequence, cpu);
 218
 219                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 220                 t->sequence = ++(*sequence);
 221                 t->time = ktime_to_ns(ktime_get());
 222 record_it:
 223                 /*
 224                  * These two are not needed in ftrace as they are in the
 225                  * generic trace_entry, filled by tracing_generic_entry_update,
 226                  * but for the trace_event->bin() synthesizer benefit we do it
 227                  * here too.
 228                  */
 229                 t->cpu = cpu;
 230                 t->pid = pid;
 231
 232                 t->sector = sector;
 233                 t->bytes = bytes;
 234                 t->action = what;
 235                 t->device = bt->dev;
 236                 t->error = error;
 237                 t->pdu_len = pdu_len;
 238
 239                 if (pdu_len)
 240                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 241
 242                 if (blk_tr) {
 243                         ring_buffer_unlock_commit(blk_tr->buffer, event, flags);
 244                         if (pid != 0 &&
 245                             !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC) &&
 246                             (trace_flags & TRACE_ITER_STACKTRACE) != 0)
 247                                 __trace_stack(blk_tr, NULL, flags, 5, pc);
 248                         trace_wake_up();
 249                         return;
 250                 }
 251         }
 252
 253         local_irq_restore(flags);
 254 }
 255
 256 static struct dentry *blk_tree_root;
 257 static DEFINE_MUTEX(blk_tree_mutex);
 258
 259 static void blk_trace_cleanup(struct blk_trace *bt)
 260 {
 261         debugfs_remove(bt->msg_file);
 262         debugfs_remove(bt->dropped_file);
 263         relay_close(bt->rchan);
 264         free_percpu(bt->sequence);
 265         free_percpu(bt->msg_data);
 266         kfree(bt);
 267         mutex_lock(&blk_probe_mutex);
 268         if (atomic_dec_and_test(&blk_probes_ref))
 269                 blk_unregister_tracepoints();
 270         mutex_unlock(&blk_probe_mutex);
 271 }
 272
 273 int blk_trace_remove(struct request_queue *q)
 274 {
 275         struct blk_trace *bt;
 276
 277         bt = xchg(&q->blk_trace, NULL);
 278         if (!bt)
 279                 return -EINVAL;
 280
 281         if (bt->trace_state == Blktrace_setup ||
 282             bt->trace_state == Blktrace_stopped)
 283                 blk_trace_cleanup(bt);
 284
 285         return 0;
 286 }
 287 EXPORT_SYMBOL_GPL(blk_trace_remove);
 288
 289 static int blk_dropped_open(struct inode *inode, struct file *filp)
 290 {
 291         filp->private_data = inode->i_private;
 292
 293         return 0;
 294 }
 295
 296 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 297                                 size_t count, loff_t *ppos)
 298 {
 299         struct blk_trace *bt = filp->private_data;
 300         char buf[16];
 301
 302         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 303
 304         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 305 }
 306
 307 static const struct file_operations blk_dropped_fops = {
 308         .owner =        THIS_MODULE,
 309         .open =         blk_dropped_open,
 310         .read =         blk_dropped_read,
 311 };
 312
 313 static int blk_msg_open(struct inode *inode, struct file *filp)
 314 {
 315         filp->private_data = inode->i_private;
 316
 317         return 0;
 318 }
 319
 320 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 321                                 size_t count, loff_t *ppos)
 322 {
 323         char *msg;
 324         struct blk_trace *bt;
 325
 326         if (count > BLK_TN_MAX_MSG)
 327                 return -EINVAL;
 328
 329         msg = kmalloc(count, GFP_KERNEL);
 330         if (msg == NULL)
 331                 return -ENOMEM;
 332
 333         if (copy_from_user(msg, buffer, count)) {
 334                 kfree(msg);
 335                 return -EFAULT;
 336         }
 337
 338         bt = filp->private_data;
 339         __trace_note_message(bt, "%s", msg);
 340         kfree(msg);
 341
 342         return count;
 343 }
 344
 345 static const struct file_operations blk_msg_fops = {
 346         .owner =        THIS_MODULE,
 347         .open =         blk_msg_open,
 348         .write =        blk_msg_write,
 349 };
 350
 351 /*
 352  * Keep track of how many times we encountered a full subbuffer, to aid
 353  * the user space app in telling how many lost events there were.
 354  */
 355 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 356                                      void *prev_subbuf, size_t prev_padding)
 357 {
 358         struct blk_trace *bt;
 359
 360         if (!relay_buf_full(buf))
 361                 return 1;
 362
 363         bt = buf->chan->private_data;
 364         atomic_inc(&bt->dropped);
 365         return 0;
 366 }
 367
 368 static int blk_remove_buf_file_callback(struct dentry *dentry)
 369 {
 370         struct dentry *parent = dentry->d_parent;
 371         debugfs_remove(dentry);
 372
 373         /*
 374         * this will fail for all but the last file, but that is ok. what we
 375         * care about is the top level buts->name directory going away, when
 376         * the last trace file is gone. Then we don't have to rmdir() that
 377         * manually on trace stop, so it nicely solves the issue with
 378         * force killing of running traces.
 379         */
 380
 381         debugfs_remove(parent);
 382         return 0;
 383 }
 384
 385 static struct dentry *blk_create_buf_file_callback(const char *filename,
 386                                                    struct dentry *parent,
 387                                                    int mode,
 388                                                    struct rchan_buf *buf,
 389                                                    int *is_global)
 390 {
 391         return debugfs_create_file(filename, mode, parent, buf,
 392                                         &relay_file_operations);
 393 }
 394
 395 static struct rchan_callbacks blk_relay_callbacks = {
 396         .subbuf_start           = blk_subbuf_start_callback,
 397         .create_buf_file        = blk_create_buf_file_callback,
 398         .remove_buf_file        = blk_remove_buf_file_callback,
 399 };
 400
 401 /*
 402  * Setup everything required to start tracing
 403  */
 404 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 405                         struct blk_user_trace_setup *buts)
 406 {
 407         struct blk_trace *old_bt, *bt = NULL;
 408         struct dentry *dir = NULL;
 409         int ret, i;
 410
 411         if (!buts->buf_size || !buts->buf_nr)
 412                 return -EINVAL;
 413
 414         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 415         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 416
 417         /*
 418          * some device names have larger paths - convert the slashes
 419          * to underscores for this to work as expected
 420          */
 421         for (i = 0; i < strlen(buts->name); i++)
 422                 if (buts->name[i] == '/')
 423                         buts->name[i] = '_';
 424
 425         ret = -ENOMEM;
 426         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 427         if (!bt)
 428                 goto err;
 429
 430         bt->sequence = alloc_percpu(unsigned long);
 431         if (!bt->sequence)
 432                 goto err;
 433
 434         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 435         if (!bt->msg_data)
 436                 goto err;
 437
 438         ret = -ENOENT;
 439
 440         if (!blk_tree_root) {
 441                 blk_tree_root = debugfs_create_dir("block", NULL);
 442                 if (!blk_tree_root)
 443                         return -ENOMEM;
 444         }
 445
 446         dir = debugfs_create_dir(buts->name, blk_tree_root);
 447
 448         if (!dir)
 449                 goto err;
 450
 451         bt->dir = dir;
 452         bt->dev = dev;
 453         atomic_set(&bt->dropped, 0);
 454
 455         ret = -EIO;
 456         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 457         if (!bt->dropped_file)
 458                 goto err;
 459
 460         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 461         if (!bt->msg_file)
 462                 goto err;
 463
 464         bt->rchan = relay_open("trace", dir, buts->buf_size,
 465                                 buts->buf_nr, &blk_relay_callbacks, bt);
 466         if (!bt->rchan)
 467                 goto err;
 468
 469         bt->act_mask = buts->act_mask;
 470         if (!bt->act_mask)
 471                 bt->act_mask = (u16) -1;
 472
 473         bt->start_lba = buts->start_lba;
 474         bt->end_lba = buts->end_lba;
 475         if (!bt->end_lba)
 476                 bt->end_lba = -1ULL;
 477
 478         bt->pid = buts->pid;
 479         bt->trace_state = Blktrace_setup;
 480
 481         mutex_lock(&blk_probe_mutex);
 482         if (atomic_add_return(1, &blk_probes_ref) == 1) {
 483                 ret = blk_register_tracepoints();
 484                 if (ret)
 485                         goto probe_err;
 486         }
 487         mutex_unlock(&blk_probe_mutex);
 488
 489         ret = -EBUSY;
 490         old_bt = xchg(&q->blk_trace, bt);
 491         if (old_bt) {
 492                 (void) xchg(&q->blk_trace, old_bt);
 493                 goto err;
 494         }
 495
 496         return 0;
 497 probe_err:
 498         atomic_dec(&blk_probes_ref);
 499         mutex_unlock(&blk_probe_mutex);
 500 err:
 501         if (bt) {
 502                 if (bt->msg_file)
 503                         debugfs_remove(bt->msg_file);
 504                 if (bt->dropped_file)
 505                         debugfs_remove(bt->dropped_file);
 506                 free_percpu(bt->sequence);
 507                 free_percpu(bt->msg_data);
 508                 if (bt->rchan)
 509                         relay_close(bt->rchan);
 510                 kfree(bt);
 511         }
 512         return ret;
 513 }
 514
 515 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 516                     char __user *arg)
 517 {
 518         struct blk_user_trace_setup buts;
 519         int ret;
 520
 521         ret = copy_from_user(&buts, arg, sizeof(buts));
 522         if (ret)
 523                 return -EFAULT;
 524
 525         ret = do_blk_trace_setup(q, name, dev, &buts);
 526         if (ret)
 527                 return ret;
 528
 529         if (copy_to_user(arg, &buts, sizeof(buts)))
 530                 return -EFAULT;
 531
 532         return 0;
 533 }
 534 EXPORT_SYMBOL_GPL(blk_trace_setup);
 535
 536 int blk_trace_startstop(struct request_queue *q, int start)
 537 {
 538         struct blk_trace *bt;
 539         int ret;
 540
 541         if ((bt = q->blk_trace) == NULL)
 542                 return -EINVAL;
 543
 544         /*
 545          * For starting a trace, we can transition from a setup or stopped
 546          * trace. For stopping a trace, the state must be running
 547          */
 548         ret = -EINVAL;
 549         if (start) {
 550                 if (bt->trace_state == Blktrace_setup ||
 551                     bt->trace_state == Blktrace_stopped) {
 552                         blktrace_seq++;
 553                         smp_mb();
 554                         bt->trace_state = Blktrace_running;
 555
 556                         trace_note_time(bt);
 557                         ret = 0;
 558                 }
 559         } else {
 560                 if (bt->trace_state == Blktrace_running) {
 561                         bt->trace_state = Blktrace_stopped;
 562                         relay_flush(bt->rchan);
 563                         ret = 0;
 564                 }
 565         }
 566
 567         return ret;
 568 }
 569 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 570
 571 /**
 572  * blk_trace_ioctl: - handle the ioctls associated with tracing
 573  * @bdev:       the block device
 574  * @cmd:        the ioctl cmd
 575  * @arg:        the argument data, if any
 576  *
 577  **/
 578 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 579 {
 580         struct request_queue *q;
 581         int ret, start = 0;
 582         char b[BDEVNAME_SIZE];
 583
 584         q = bdev_get_queue(bdev);
 585         if (!q)
 586                 return -ENXIO;
 587
 588         mutex_lock(&bdev->bd_mutex);
 589
 590         switch (cmd) {
 591         case BLKTRACESETUP:
 592                 bdevname(bdev, b);
 593                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 594                 break;
 595         case BLKTRACESTART:
 596                 start = 1;
 597         case BLKTRACESTOP:
 598                 ret = blk_trace_startstop(q, start);
 599                 break;
 600         case BLKTRACETEARDOWN:
 601                 ret = blk_trace_remove(q);
 602                 break;
 603         default:
 604                 ret = -ENOTTY;
 605                 break;
 606         }
 607
 608         mutex_unlock(&bdev->bd_mutex);
 609         return ret;
 610 }
 611
 612 /**
 613  * blk_trace_shutdown: - stop and cleanup trace structures
 614  * @q:    the request queue associated with the device
 615  *
 616  **/
 617 void blk_trace_shutdown(struct request_queue *q)
 618 {
 619         if (q->blk_trace) {
 620                 blk_trace_startstop(q, 0);
 621                 blk_trace_remove(q);
 622         }
 623 }
 624
 625 /*
 626  * blktrace probes
 627  */
 628
 629 /**
 630  * blk_add_trace_rq - Add a trace for a request oriented action
 631  * @q:          queue the io is for
 632  * @rq:         the source request
 633  * @what:       the action
 634  *
 635  * Description:
 636  *     Records an action against a request. Will log the bio offset + size.
 637  *
 638  **/
 639 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 640                                     u32 what)
 641 {
 642         struct blk_trace *bt = q->blk_trace;
 643         int rw = rq->cmd_flags & 0x03;
 644
 645         if (likely(!bt))
 646                 return;
 647
 648         if (blk_discard_rq(rq))
 649                 rw |= (1 << BIO_RW_DISCARD);
 650
 651         if (blk_pc_request(rq)) {
 652                 what |= BLK_TC_ACT(BLK_TC_PC);
 653                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 654                                 sizeof(rq->cmd), rq->cmd);
 655         } else  {
 656                 what |= BLK_TC_ACT(BLK_TC_FS);
 657                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 658                                 rw, what, rq->errors, 0, NULL);
 659         }
 660 }
 661
 662 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 663 {
 664         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 665 }
 666
 667 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 668 {
 669         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 670 }
 671
 672 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 673 {
 674         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 675 }
 676
 677 static void blk_add_trace_rq_requeue(struct request_queue *q, struct request *rq)
 678 {
 679         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 680 }
 681
 682 static void blk_add_trace_rq_complete(struct request_queue *q, struct request *rq)
 683 {
 684         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 685 }
 686
 687 /**
 688  * blk_add_trace_bio - Add a trace for a bio oriented action
 689  * @q:          queue the io is for
 690  * @bio:        the source bio
 691  * @what:       the action
 692  *
 693  * Description:
 694  *     Records an action against a bio. Will log the bio offset + size.
 695  *
 696  **/
 697 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 698                                      u32 what)
 699 {
 700         struct blk_trace *bt = q->blk_trace;
 701
 702         if (likely(!bt))
 703                 return;
 704
 705         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 706                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 707 }
 708
 709 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 710 {
 711         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 712 }
 713
 714 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 715 {
 716         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 717 }
 718
 719 static void blk_add_trace_bio_backmerge(struct request_queue *q, struct bio *bio)
 720 {
 721         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 722 }
 723
 724 static void blk_add_trace_bio_frontmerge(struct request_queue *q, struct bio *bio)
 725 {
 726         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 727 }
 728
 729 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 730 {
 731         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 732 }
 733
 734 static void blk_add_trace_getrq(struct request_queue *q, struct bio *bio, int rw)
 735 {
 736         if (bio)
 737                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 738         else {
 739                 struct blk_trace *bt = q->blk_trace;
 740
 741                 if (bt)
 742                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 743         }
 744 }
 745
 746
 747 static void blk_add_trace_sleeprq(struct request_queue *q, struct bio *bio, int rw)
 748 {
 749         if (bio)
 750                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 751         else {
 752                 struct blk_trace *bt = q->blk_trace;
 753
 754                 if (bt)
 755                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ, 0, 0, NULL);
 756         }
 757 }
 758
 759 static void blk_add_trace_plug(struct request_queue *q)
 760 {
 761         struct blk_trace *bt = q->blk_trace;
 762
 763         if (bt)
 764                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 765 }
 766
 767 static void blk_add_trace_unplug_io(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_IO, 0,
 776                                 sizeof(rpdu), &rpdu);
 777         }
 778 }
 779
 780 static void blk_add_trace_unplug_timer(struct request_queue *q)
 781 {
 782         struct blk_trace *bt = q->blk_trace;
 783
 784         if (bt) {
 785                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 786                 __be64 rpdu = cpu_to_be64(pdu);
 787
 788                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 789                                 sizeof(rpdu), &rpdu);
 790         }
 791 }
 792
 793 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 794                                 unsigned int pdu)
 795 {
 796         struct blk_trace *bt = q->blk_trace;
 797
 798         if (bt) {
 799                 __be64 rpdu = cpu_to_be64(pdu);
 800
 801                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 802                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 803                                 sizeof(rpdu), &rpdu);
 804         }
 805 }
 806
 807 /**
 808  * blk_add_trace_remap - Add a trace for a remap operation
 809  * @q:          queue the io is for
 810  * @bio:        the source bio
 811  * @dev:        target device
 812  * @from:       source sector
 813  * @to:         target sector
 814  *
 815  * Description:
 816  *     Device mapper or raid target sometimes need to split a bio because
 817  *     it spans a stripe (or similar). Add a trace for that action.
 818  *
 819  **/
 820 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 821                                        dev_t dev, sector_t from, sector_t to)
 822 {
 823         struct blk_trace *bt = q->blk_trace;
 824         struct blk_io_trace_remap r;
 825
 826         if (likely(!bt))
 827                 return;
 828
 829         r.device = cpu_to_be32(dev);
 830         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 831         r.sector = cpu_to_be64(to);
 832
 833         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 834                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 835 }
 836
 837 /**
 838  * blk_add_driver_data - Add binary message with driver-specific data
 839  * @q:          queue the io is for
 840  * @rq:         io request
 841  * @data:       driver-specific data
 842  * @len:        length of driver-specific data
 843  *
 844  * Description:
 845  *     Some drivers might want to write driver-specific data per request.
 846  *
 847  **/
 848 void blk_add_driver_data(struct request_queue *q,
 849                          struct request *rq,
 850                          void *data, size_t len)
 851 {
 852         struct blk_trace *bt = q->blk_trace;
 853
 854         if (likely(!bt))
 855                 return;
 856
 857         if (blk_pc_request(rq))
 858                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 859                                 rq->errors, len, data);
 860         else
 861                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 862                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 863 }
 864 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 865
 866 static int blk_register_tracepoints(void)
 867 {
 868         int ret;
 869
 870         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 871         WARN_ON(ret);
 872         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 873         WARN_ON(ret);
 874         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 875         WARN_ON(ret);
 876         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 877         WARN_ON(ret);
 878         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 879         WARN_ON(ret);
 880         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 881         WARN_ON(ret);
 882         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 883         WARN_ON(ret);
 884         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 885         WARN_ON(ret);
 886         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 887         WARN_ON(ret);
 888         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 889         WARN_ON(ret);
 890         ret = register_trace_block_getrq(blk_add_trace_getrq);
 891         WARN_ON(ret);
 892         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 893         WARN_ON(ret);
 894         ret = register_trace_block_plug(blk_add_trace_plug);
 895         WARN_ON(ret);
 896         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 897         WARN_ON(ret);
 898         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 899         WARN_ON(ret);
 900         ret = register_trace_block_split(blk_add_trace_split);
 901         WARN_ON(ret);
 902         ret = register_trace_block_remap(blk_add_trace_remap);
 903         WARN_ON(ret);
 904         return 0;
 905 }
 906
 907 static void blk_unregister_tracepoints(void)
 908 {
 909         unregister_trace_block_remap(blk_add_trace_remap);
 910         unregister_trace_block_split(blk_add_trace_split);
 911         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 912         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 913         unregister_trace_block_plug(blk_add_trace_plug);
 914         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 915         unregister_trace_block_getrq(blk_add_trace_getrq);
 916         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 917         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 918         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 919         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 920         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 921         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 922         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 923         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 924         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 925         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 926
 927         tracepoint_synchronize_unregister();
 928 }
 929
 930 /*
 931  * struct blk_io_tracer formatting routines
 932  */
 933
 934 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 935 {
 936         int i = 0;
 937
 938         if (t->action & BLK_TC_DISCARD)
 939                 rwbs[i++] = 'D';
 940         else if (t->action & BLK_TC_WRITE)
 941                 rwbs[i++] = 'W';
 942         else if (t->bytes)
 943                 rwbs[i++] = 'R';
 944         else
 945                 rwbs[i++] = 'N';
 946
 947         if (t->action & BLK_TC_AHEAD)
 948                 rwbs[i++] = 'A';
 949         if (t->action & BLK_TC_BARRIER)
 950                 rwbs[i++] = 'B';
 951         if (t->action & BLK_TC_SYNC)
 952                 rwbs[i++] = 'S';
 953         if (t->action & BLK_TC_META)
 954                 rwbs[i++] = 'M';
 955
 956         rwbs[i] = '\0';
 957 }
 958
 959 static inline
 960 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 961 {
 962         return (const struct blk_io_trace *)ent;
 963 }
 964
 965 static inline const void *pdu_start(const struct trace_entry *ent)
 966 {
 967         return te_blk_io_trace(ent) + 1;
 968 }
 969
 970 static inline u32 t_sec(const struct trace_entry *ent)
 971 {
 972         return te_blk_io_trace(ent)->bytes >> 9;
 973 }
 974
 975 static inline unsigned long long t_sector(const struct trace_entry *ent)
 976 {
 977         return te_blk_io_trace(ent)->sector;
 978 }
 979
 980 static inline __u16 t_error(const struct trace_entry *ent)
 981 {
 982         return te_blk_io_trace(ent)->sector;
 983 }
 984
 985 static __u64 get_pdu_int(const struct trace_entry *ent)
 986 {
 987         const __u64 *val = pdu_start(ent);
 988         return be64_to_cpu(*val);
 989 }
 990
 991 static void get_pdu_remap(const struct trace_entry *ent,
 992                           struct blk_io_trace_remap *r)
 993 {
 994         const struct blk_io_trace_remap *__r = pdu_start(ent);
 995         __u64 sector = __r->sector;
 996
 997         r->device = be32_to_cpu(__r->device);
 998         r->device_from = be32_to_cpu(__r->device_from);
 999         r->sector = be64_to_cpu(sector);
1000 }
1001
1002 static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
1003 {
1004         char rwbs[6];
1005         unsigned long long ts  = ns2usecs(iter->ts);
1006         unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
1007         unsigned secs          = (unsigned long)ts;
1008         const struct trace_entry *ent = iter->ent;
1009         const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
1010
1011         fill_rwbs(rwbs, t);
1012
1013         return trace_seq_printf(&iter->seq,
1014                                 "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
1015                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1016                                 secs, usec_rem, ent->pid, act, rwbs);
1017 }
1018
1019 static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
1020                               const char *act)
1021 {
1022         char rwbs[6];
1023         fill_rwbs(rwbs, t);
1024         return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
1025                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1026 }
1027
1028 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1029 {
1030         const char *cmd = trace_find_cmdline(ent->pid);
1031
1032         if (t_sec(ent))
1033                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1034                                         t_sector(ent), t_sec(ent), cmd);
1035         return trace_seq_printf(s, "[%s]\n", cmd);
1036 }
1037
1038 static int blk_log_with_error(struct trace_seq *s,
1039                               const struct trace_entry *ent)
1040 {
1041         if (t_sec(ent))
1042                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1043                                         t_sec(ent), t_error(ent));
1044         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1045 }
1046
1047 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1048 {
1049         struct blk_io_trace_remap r = { .device = 0, };
1050
1051         get_pdu_remap(ent, &r);
1052         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1053                                t_sector(ent),
1054                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1055                                (unsigned long long)r.sector);
1056 }
1057
1058 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1059 {
1060         return trace_seq_printf(s, "[%s]\n", trace_find_cmdline(ent->pid));
1061 }
1062
1063 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1064 {
1065         return trace_seq_printf(s, "[%s] %llu\n", trace_find_cmdline(ent->pid),
1066                                 get_pdu_int(ent));
1067 }
1068
1069 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1070 {
1071         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1072                                 get_pdu_int(ent), trace_find_cmdline(ent->pid));
1073 }
1074
1075 /*
1076  * struct tracer operations
1077  */
1078
1079 static void blk_tracer_print_header(struct seq_file *m)
1080 {
1081         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1082                 return;
1083         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1084                     "#  |     |     |           |   |   |\n");
1085 }
1086
1087 static void blk_tracer_start(struct trace_array *tr)
1088 {
1089         tracing_reset_online_cpus(tr);
1090
1091         mutex_lock(&blk_probe_mutex);
1092         if (atomic_add_return(1, &blk_probes_ref) == 1)
1093                 if (blk_register_tracepoints())
1094                         atomic_dec(&blk_probes_ref);
1095         mutex_unlock(&blk_probe_mutex);
1096         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1097 }
1098
1099 static int blk_tracer_init(struct trace_array *tr)
1100 {
1101         blk_tr = tr;
1102         blk_tracer_start(tr);
1103         mutex_lock(&blk_probe_mutex);
1104         blk_tracer_enabled++;
1105         mutex_unlock(&blk_probe_mutex);
1106         return 0;
1107 }
1108
1109 static void blk_tracer_stop(struct trace_array *tr)
1110 {
1111         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1112         mutex_lock(&blk_probe_mutex);
1113         if (atomic_dec_and_test(&blk_probes_ref))
1114                 blk_unregister_tracepoints();
1115         mutex_unlock(&blk_probe_mutex);
1116 }
1117
1118 static void blk_tracer_reset(struct trace_array *tr)
1119 {
1120         if (!atomic_read(&blk_probes_ref))
1121                 return;
1122
1123         mutex_lock(&blk_probe_mutex);
1124         blk_tracer_enabled--;
1125         WARN_ON(blk_tracer_enabled < 0);
1126         mutex_unlock(&blk_probe_mutex);
1127
1128         blk_tracer_stop(tr);
1129 }
1130
1131 static struct {
1132         const char *act[2];
1133         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1134 } what2act[] __read_mostly = {
1135         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1136         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1137         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1138         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1139         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1140         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1141         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1142         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1143         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1144         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1145         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1146         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1147         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1148         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1149         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1150 };
1151
1152 static int blk_trace_event_print(struct trace_iterator *iter, int flags)
1153 {
1154         struct trace_seq *s = &iter->seq;
1155         const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1156         const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1157         int ret;
1158
1159         if (trace_print_context(iter))
1160                 return TRACE_TYPE_PARTIAL_LINE;
1161
1162         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1163                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1164         else {
1165                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1166                 ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
1167                 if (ret)
1168                         ret = what2act[what].print(s, iter->ent);
1169         }
1170
1171         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1172 }
1173
1174 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1175 {
1176         struct trace_seq *s = &iter->seq;
1177         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1178         const int offset = offsetof(struct blk_io_trace, sector);
1179         struct blk_io_trace old = {
1180                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1181                 .time     = ns2usecs(iter->ts),
1182         };
1183
1184         if (!trace_seq_putmem(s, &old, offset))
1185                 return 0;
1186         return trace_seq_putmem(s, &t->sector,
1187                                 sizeof(old) - offset + t->pdu_len);
1188 }
1189
1190 static int blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1191 {
1192         return blk_trace_synthesize_old_trace(iter) ?
1193                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1194 }
1195
1196 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1197 {
1198         const struct blk_io_trace *t;
1199         u16 what;
1200         int ret;
1201
1202         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1203                 return TRACE_TYPE_UNHANDLED;
1204
1205         t = (const struct blk_io_trace *)iter->ent;
1206         what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1207
1208         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1209                 ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
1210         else {
1211                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1212                 ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
1213                 if (ret)
1214                         ret = what2act[what].print(&iter->seq, iter->ent);
1215         }
1216
1217         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1218 }
1219
1220 static struct tracer blk_tracer __read_mostly = {
1221         .name           = "blk",
1222         .init           = blk_tracer_init,
1223         .reset          = blk_tracer_reset,
1224         .start          = blk_tracer_start,
1225         .stop           = blk_tracer_stop,
1226         .print_header   = blk_tracer_print_header,
1227         .print_line     = blk_tracer_print_line,
1228         .flags          = &blk_tracer_flags,
1229 };
1230
1231 static struct trace_event trace_blk_event = {
1232         .type           = TRACE_BLK,
1233         .trace          = blk_trace_event_print,
1234         .latency_trace  = blk_trace_event_print,
1235         .raw            = trace_nop_print,
1236         .hex            = trace_nop_print,
1237         .binary         = blk_trace_event_print_binary,
1238 };
1239
1240 static int __init init_blk_tracer(void)
1241 {
1242         if (!register_ftrace_event(&trace_blk_event)) {
1243                 pr_warning("Warning: could not register block events\n");
1244                 return 1;
1245         }
1246
1247         if (register_tracer(&blk_tracer) != 0) {
1248                 pr_warning("Warning: could not register the block tracer\n");
1249                 unregister_ftrace_event(&trace_blk_event);
1250                 return 1;
1251         }
1252
1253         return 0;
1254 }
1255
1256 device_initcall(init_blk_tracer);
1257
1258 static int blk_trace_remove_queue(struct request_queue *q)
1259 {
1260         struct blk_trace *bt;
1261
1262         bt = xchg(&q->blk_trace, NULL);
1263         if (bt == NULL)
1264                 return -EINVAL;
1265
1266         kfree(bt);
1267         return 0;
1268 }
1269
1270 /*
1271  * Setup everything required to start tracing
1272  */
1273 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1274 {
1275         struct blk_trace *old_bt, *bt = NULL;
1276         int ret;
1277
1278         ret = -ENOMEM;
1279         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1280         if (!bt)
1281                 goto err;
1282
1283         bt->dev = dev;
1284         bt->act_mask = (u16)-1;
1285         bt->end_lba = -1ULL;
1286         bt->trace_state = Blktrace_running;
1287
1288         old_bt = xchg(&q->blk_trace, bt);
1289         if (old_bt != NULL) {
1290                 (void)xchg(&q->blk_trace, old_bt);
1291                 kfree(bt);
1292                 ret = -EBUSY;
1293         }
1294         return 0;
1295 err:
1296         return ret;
1297 }
1298
1299 /*
1300  * sysfs interface to enable and configure tracing
1301  */
1302
1303 static ssize_t sysfs_blk_trace_enable_show(struct device *dev,
1304                                            struct device_attribute *attr,
1305                                            char *buf)
1306 {
1307         struct hd_struct *p = dev_to_part(dev);
1308         struct block_device *bdev;
1309         ssize_t ret = -ENXIO;
1310
1311         lock_kernel();
1312         bdev = bdget(part_devt(p));
1313         if (bdev != NULL) {
1314                 struct request_queue *q = bdev_get_queue(bdev);
1315
1316                 if (q != NULL) {
1317                         mutex_lock(&bdev->bd_mutex);
1318                         ret = sprintf(buf, "%u\n", !!q->blk_trace);
1319                         mutex_unlock(&bdev->bd_mutex);
1320                 }
1321
1322                 bdput(bdev);
1323         }
1324
1325         unlock_kernel();
1326         return ret;
1327 }
1328
1329 static ssize_t sysfs_blk_trace_enable_store(struct device *dev,
1330                                             struct device_attribute *attr,
1331                                             const char *buf, size_t count)
1332 {
1333         struct block_device *bdev;
1334         struct request_queue *q;
1335         struct hd_struct *p;
1336         int value;
1337         ssize_t ret = -ENXIO;
1338
1339         if (count == 0 || sscanf(buf, "%d", &value) != 1)
1340                 goto out;
1341
1342         lock_kernel();
1343         p = dev_to_part(dev);
1344         bdev = bdget(part_devt(p));
1345         if (bdev == NULL)
1346                 goto out_unlock_kernel;
1347
1348         q = bdev_get_queue(bdev);
1349         if (q == NULL)
1350                 goto out_bdput;
1351
1352         mutex_lock(&bdev->bd_mutex);
1353         if (value)
1354                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1355         else
1356                 ret = blk_trace_remove_queue(q);
1357         mutex_unlock(&bdev->bd_mutex);
1358
1359         if (ret == 0)
1360                 ret = count;
1361 out_bdput:
1362         bdput(bdev);
1363 out_unlock_kernel:
1364         unlock_kernel();
1365 out:
1366         return ret;
1367 }
1368
1369 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1370                                          struct device_attribute *attr,
1371                                          char *buf);
1372 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1373                                           struct device_attribute *attr,
1374                                           const char *buf, size_t count);
1375 #define BLK_TRACE_DEVICE_ATTR(_name) \
1376         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1377                     sysfs_blk_trace_attr_show, \
1378                     sysfs_blk_trace_attr_store)
1379
1380 static DEVICE_ATTR(enable, S_IRUGO | S_IWUSR,
1381                    sysfs_blk_trace_enable_show, sysfs_blk_trace_enable_store);
1382 static BLK_TRACE_DEVICE_ATTR(act_mask);
1383 static BLK_TRACE_DEVICE_ATTR(pid);
1384 static BLK_TRACE_DEVICE_ATTR(start_lba);
1385 static BLK_TRACE_DEVICE_ATTR(end_lba);
1386
1387 static struct attribute *blk_trace_attrs[] = {
1388         &dev_attr_enable.attr,
1389         &dev_attr_act_mask.attr,
1390         &dev_attr_pid.attr,
1391         &dev_attr_start_lba.attr,
1392         &dev_attr_end_lba.attr,
1393         NULL
1394 };
1395
1396 struct attribute_group blk_trace_attr_group = {
1397         .name  = "trace",
1398         .attrs = blk_trace_attrs,
1399 };
1400
1401 static int blk_str2act_mask(const char *str)
1402 {
1403         int mask = 0;
1404         char *copy = kstrdup(str, GFP_KERNEL), *s;
1405
1406         if (copy == NULL)
1407                 return -ENOMEM;
1408
1409         s = strstrip(copy);
1410
1411         while (1) {
1412                 char *sep = strchr(s, ',');
1413
1414                 if (sep != NULL)
1415                         *sep = '\0';
1416
1417                 if (strcasecmp(s, "barrier") == 0)
1418                         mask |= BLK_TC_BARRIER;
1419                 else if (strcasecmp(s, "complete") == 0)
1420                         mask |= BLK_TC_COMPLETE;
1421                 else if (strcasecmp(s, "fs") == 0)
1422                         mask |= BLK_TC_FS;
1423                 else if (strcasecmp(s, "issue") == 0)
1424                         mask |= BLK_TC_ISSUE;
1425                 else if (strcasecmp(s, "pc") == 0)
1426                         mask |= BLK_TC_PC;
1427                 else if (strcasecmp(s, "queue") == 0)
1428                         mask |= BLK_TC_QUEUE;
1429                 else if (strcasecmp(s, "read") == 0)
1430                         mask |= BLK_TC_READ;
1431                 else if (strcasecmp(s, "requeue") == 0)
1432                         mask |= BLK_TC_REQUEUE;
1433                 else if (strcasecmp(s, "sync") == 0)
1434                         mask |= BLK_TC_SYNC;
1435                 else if (strcasecmp(s, "write") == 0)
1436                         mask |= BLK_TC_WRITE;
1437
1438                 if (sep == NULL)
1439                         break;
1440
1441                 s = sep + 1;
1442         }
1443         kfree(copy);
1444
1445         return mask;
1446 }
1447
1448 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1449                                          struct device_attribute *attr,
1450                                          char *buf)
1451 {
1452         struct hd_struct *p = dev_to_part(dev);
1453         struct request_queue *q;
1454         struct block_device *bdev;
1455         ssize_t ret = -ENXIO;
1456
1457         lock_kernel();
1458         bdev = bdget(part_devt(p));
1459         if (bdev == NULL)
1460                 goto out_unlock_kernel;
1461
1462         q = bdev_get_queue(bdev);
1463         if (q == NULL)
1464                 goto out_bdput;
1465         mutex_lock(&bdev->bd_mutex);
1466         if (q->blk_trace == NULL)
1467                 ret = sprintf(buf, "disabled\n");
1468         else if (attr == &dev_attr_act_mask)
1469                 ret = sprintf(buf, "%#x\n", q->blk_trace->act_mask);
1470         else if (attr == &dev_attr_pid)
1471                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1472         else if (attr == &dev_attr_start_lba)
1473                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1474         else if (attr == &dev_attr_end_lba)
1475                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1476         mutex_unlock(&bdev->bd_mutex);
1477 out_bdput:
1478         bdput(bdev);
1479 out_unlock_kernel:
1480         unlock_kernel();
1481         return ret;
1482 }
1483
1484 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1485                                           struct device_attribute *attr,
1486                                           const char *buf, size_t count)
1487 {
1488         struct block_device *bdev;
1489         struct request_queue *q;
1490         struct hd_struct *p;
1491         u64 value;
1492         ssize_t ret = -ENXIO;
1493
1494         if (count == 0)
1495                 goto out;
1496
1497         if (attr == &dev_attr_act_mask) {
1498                 if (sscanf(buf, "%llx", &value) != 1) {
1499                         /* Assume it is a list of trace category names */
1500                         value = blk_str2act_mask(buf);
1501                         if (value < 0)
1502                                 goto out;
1503                 }
1504         } else if (sscanf(buf, "%llu", &value) != 1)
1505                 goto out;
1506
1507         lock_kernel();
1508         p = dev_to_part(dev);
1509         bdev = bdget(part_devt(p));
1510         if (bdev == NULL)
1511                 goto out_unlock_kernel;
1512
1513         q = bdev_get_queue(bdev);
1514         if (q == NULL)
1515                 goto out_bdput;
1516
1517         mutex_lock(&bdev->bd_mutex);
1518         ret = 0;
1519         if (q->blk_trace == NULL)
1520                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1521
1522         if (ret == 0) {
1523                 if (attr == &dev_attr_act_mask)
1524                         q->blk_trace->act_mask = value;
1525                 else if (attr == &dev_attr_pid)
1526                         q->blk_trace->pid = value;
1527                 else if (attr == &dev_attr_start_lba)
1528                         q->blk_trace->start_lba = value;
1529                 else if (attr == &dev_attr_end_lba)
1530                         q->blk_trace->end_lba = value;
1531                 ret = count;
1532         }
1533         mutex_unlock(&bdev->bd_mutex);
1534 out_bdput:
1535         bdput(bdev);
1536 out_unlock_kernel:
1537         unlock_kernel();
1538 out:
1539         return ret;
1540 }