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         struct ring_buffer_event *event = NULL;
  63         int pc = 0;
  64         int cpu = smp_processor_id();
  65         bool blk_tracer = blk_tracer_enabled;
  66
  67         if (blk_tracer) {
  68                 pc = preempt_count();
  69                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
  70                                                   sizeof(*t) + len,
  71                                                   0, pc);
  72                 if (!event)
  73                         return;
  74                 t = ring_buffer_event_data(event);
  75                 goto record_it;
  76         }
  77
  78         if (!bt->rchan)
  79                 return;
  80
  81         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  82         if (t) {
  83                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  84                 t->time = ktime_to_ns(ktime_get());
  85 record_it:
  86                 t->device = bt->dev;
  87                 t->action = action;
  88                 t->pid = pid;
  89                 t->cpu = cpu;
  90                 t->pdu_len = len;
  91                 memcpy((void *) t + sizeof(*t), data, len);
  92
  93                 if (blk_tracer)
  94                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
  95         }
  96 }
  97
  98 /*
  99  * Send out a notify for this process, if we haven't done so since a trace
 100  * started
 101  */
 102 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 103 {
 104         tsk->btrace_seq = blktrace_seq;
 105         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 106 }
 107
 108 static void trace_note_time(struct blk_trace *bt)
 109 {
 110         struct timespec now;
 111         unsigned long flags;
 112         u32 words[2];
 113
 114         getnstimeofday(&now);
 115         words[0] = now.tv_sec;
 116         words[1] = now.tv_nsec;
 117
 118         local_irq_save(flags);
 119         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 120         local_irq_restore(flags);
 121 }
 122
 123 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 124 {
 125         int n;
 126         va_list args;
 127         unsigned long flags;
 128         char *buf;
 129
 130         if (unlikely(bt->trace_state != Blktrace_running &&
 131                      !blk_tracer_enabled))
 132                 return;
 133
 134         local_irq_save(flags);
 135         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 136         va_start(args, fmt);
 137         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 138         va_end(args);
 139
 140         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 141         local_irq_restore(flags);
 142 }
 143 EXPORT_SYMBOL_GPL(__trace_note_message);
 144
 145 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 146                          pid_t pid)
 147 {
 148         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 149                 return 1;
 150         if (sector < bt->start_lba || sector > bt->end_lba)
 151                 return 1;
 152         if (bt->pid && pid != bt->pid)
 153                 return 1;
 154
 155         return 0;
 156 }
 157
 158 /*
 159  * Data direction bit lookup
 160  */
 161 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 162                                  BLK_TC_ACT(BLK_TC_WRITE) };
 163
 164 /* The ilog2() calls fall out because they're constant */
 165 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 166           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 167
 168 /*
 169  * The worker for the various blk_add_trace*() types. Fills out a
 170  * blk_io_trace structure and places it in a per-cpu subbuffer.
 171  */
 172 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 173                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 174 {
 175         struct task_struct *tsk = current;
 176         struct ring_buffer_event *event = NULL;
 177         struct blk_io_trace *t;
 178         unsigned long flags = 0;
 179         unsigned long *sequence;
 180         pid_t pid;
 181         int cpu, pc = 0;
 182         bool blk_tracer = blk_tracer_enabled;
 183
 184         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 185                 return;
 186
 187         what |= ddir_act[rw & WRITE];
 188         what |= MASK_TC_BIT(rw, BARRIER);
 189         what |= MASK_TC_BIT(rw, SYNCIO);
 190         what |= MASK_TC_BIT(rw, AHEAD);
 191         what |= MASK_TC_BIT(rw, META);
 192         what |= MASK_TC_BIT(rw, DISCARD);
 193
 194         pid = tsk->pid;
 195         if (unlikely(act_log_check(bt, what, sector, pid)))
 196                 return;
 197         cpu = raw_smp_processor_id();
 198
 199         if (blk_tracer) {
 200                 tracing_record_cmdline(current);
 201
 202                 pc = preempt_count();
 203                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 204                                                   sizeof(*t) + pdu_len,
 205                                                   0, pc);
 206                 if (!event)
 207                         return;
 208                 t = ring_buffer_event_data(event);
 209                 goto record_it;
 210         }
 211
 212         /*
 213          * A word about the locking here - we disable interrupts to reserve
 214          * some space in the relay per-cpu buffer, to prevent an irq
 215          * from coming in and stepping on our toes.
 216          */
 217         local_irq_save(flags);
 218
 219         if (unlikely(tsk->btrace_seq != blktrace_seq))
 220                 trace_note_tsk(bt, tsk);
 221
 222         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 223         if (t) {
 224                 sequence = per_cpu_ptr(bt->sequence, cpu);
 225
 226                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 227                 t->sequence = ++(*sequence);
 228                 t->time = ktime_to_ns(ktime_get());
 229 record_it:
 230                 /*
 231                  * These two are not needed in ftrace as they are in the
 232                  * generic trace_entry, filled by tracing_generic_entry_update,
 233                  * but for the trace_event->bin() synthesizer benefit we do it
 234                  * here too.
 235                  */
 236                 t->cpu = cpu;
 237                 t->pid = pid;
 238
 239                 t->sector = sector;
 240                 t->bytes = bytes;
 241                 t->action = what;
 242                 t->device = bt->dev;
 243                 t->error = error;
 244                 t->pdu_len = pdu_len;
 245
 246                 if (pdu_len)
 247                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 248
 249                 if (blk_tracer) {
 250                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 251                         return;
 252                 }
 253         }
 254
 255         local_irq_restore(flags);
 256 }
 257
 258 static struct dentry *blk_tree_root;
 259 static DEFINE_MUTEX(blk_tree_mutex);
 260
 261 static void blk_trace_free(struct blk_trace *bt)
 262 {
 263         debugfs_remove(bt->msg_file);
 264         debugfs_remove(bt->dropped_file);
 265         relay_close(bt->rchan);
 266         free_percpu(bt->sequence);
 267         free_percpu(bt->msg_data);
 268         kfree(bt);
 269 }
 270
 271 static void blk_trace_cleanup(struct blk_trace *bt)
 272 {
 273         blk_trace_free(bt);
 274         if (atomic_dec_and_test(&blk_probes_ref))
 275                 blk_unregister_tracepoints();
 276 }
 277
 278 int blk_trace_remove(struct request_queue *q)
 279 {
 280         struct blk_trace *bt;
 281
 282         bt = xchg(&q->blk_trace, NULL);
 283         if (!bt)
 284                 return -EINVAL;
 285
 286         if (bt->trace_state != Blktrace_running)
 287                 blk_trace_cleanup(bt);
 288
 289         return 0;
 290 }
 291 EXPORT_SYMBOL_GPL(blk_trace_remove);
 292
 293 static int blk_dropped_open(struct inode *inode, struct file *filp)
 294 {
 295         filp->private_data = inode->i_private;
 296
 297         return 0;
 298 }
 299
 300 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 301                                 size_t count, loff_t *ppos)
 302 {
 303         struct blk_trace *bt = filp->private_data;
 304         char buf[16];
 305
 306         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 307
 308         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 309 }
 310
 311 static const struct file_operations blk_dropped_fops = {
 312         .owner =        THIS_MODULE,
 313         .open =         blk_dropped_open,
 314         .read =         blk_dropped_read,
 315 };
 316
 317 static int blk_msg_open(struct inode *inode, struct file *filp)
 318 {
 319         filp->private_data = inode->i_private;
 320
 321         return 0;
 322 }
 323
 324 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 325                                 size_t count, loff_t *ppos)
 326 {
 327         char *msg;
 328         struct blk_trace *bt;
 329
 330         if (count >= BLK_TN_MAX_MSG)
 331                 return -EINVAL;
 332
 333         msg = kmalloc(count + 1, GFP_KERNEL);
 334         if (msg == NULL)
 335                 return -ENOMEM;
 336
 337         if (copy_from_user(msg, buffer, count)) {
 338                 kfree(msg);
 339                 return -EFAULT;
 340         }
 341
 342         msg[count] = '\0';
 343         bt = filp->private_data;
 344         __trace_note_message(bt, "%s", msg);
 345         kfree(msg);
 346
 347         return count;
 348 }
 349
 350 static const struct file_operations blk_msg_fops = {
 351         .owner =        THIS_MODULE,
 352         .open =         blk_msg_open,
 353         .write =        blk_msg_write,
 354 };
 355
 356 /*
 357  * Keep track of how many times we encountered a full subbuffer, to aid
 358  * the user space app in telling how many lost events there were.
 359  */
 360 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 361                                      void *prev_subbuf, size_t prev_padding)
 362 {
 363         struct blk_trace *bt;
 364
 365         if (!relay_buf_full(buf))
 366                 return 1;
 367
 368         bt = buf->chan->private_data;
 369         atomic_inc(&bt->dropped);
 370         return 0;
 371 }
 372
 373 static int blk_remove_buf_file_callback(struct dentry *dentry)
 374 {
 375         struct dentry *parent = dentry->d_parent;
 376         debugfs_remove(dentry);
 377
 378         /*
 379         * this will fail for all but the last file, but that is ok. what we
 380         * care about is the top level buts->name directory going away, when
 381         * the last trace file is gone. Then we don't have to rmdir() that
 382         * manually on trace stop, so it nicely solves the issue with
 383         * force killing of running traces.
 384         */
 385
 386         debugfs_remove(parent);
 387         return 0;
 388 }
 389
 390 static struct dentry *blk_create_buf_file_callback(const char *filename,
 391                                                    struct dentry *parent,
 392                                                    int mode,
 393                                                    struct rchan_buf *buf,
 394                                                    int *is_global)
 395 {
 396         return debugfs_create_file(filename, mode, parent, buf,
 397                                         &relay_file_operations);
 398 }
 399
 400 static struct rchan_callbacks blk_relay_callbacks = {
 401         .subbuf_start           = blk_subbuf_start_callback,
 402         .create_buf_file        = blk_create_buf_file_callback,
 403         .remove_buf_file        = blk_remove_buf_file_callback,
 404 };
 405
 406 /*
 407  * Setup everything required to start tracing
 408  */
 409 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 410                         struct blk_user_trace_setup *buts)
 411 {
 412         struct blk_trace *old_bt, *bt = NULL;
 413         struct dentry *dir = NULL;
 414         int ret, i;
 415
 416         if (!buts->buf_size || !buts->buf_nr)
 417                 return -EINVAL;
 418
 419         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 420         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 421
 422         /*
 423          * some device names have larger paths - convert the slashes
 424          * to underscores for this to work as expected
 425          */
 426         for (i = 0; i < strlen(buts->name); i++)
 427                 if (buts->name[i] == '/')
 428                         buts->name[i] = '_';
 429
 430         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 431         if (!bt)
 432                 return -ENOMEM;
 433
 434         ret = -ENOMEM;
 435         bt->sequence = alloc_percpu(unsigned long);
 436         if (!bt->sequence)
 437                 goto err;
 438
 439         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 440         if (!bt->msg_data)
 441                 goto err;
 442
 443         ret = -ENOENT;
 444
 445         mutex_lock(&blk_tree_mutex);
 446         if (!blk_tree_root) {
 447                 blk_tree_root = debugfs_create_dir("block", NULL);
 448                 if (!blk_tree_root) {
 449                         mutex_unlock(&blk_tree_mutex);
 450                         goto err;
 451                 }
 452         }
 453         mutex_unlock(&blk_tree_mutex);
 454
 455         dir = debugfs_create_dir(buts->name, blk_tree_root);
 456
 457         if (!dir)
 458                 goto err;
 459
 460         bt->dir = dir;
 461         bt->dev = dev;
 462         atomic_set(&bt->dropped, 0);
 463
 464         ret = -EIO;
 465         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 466                                                &blk_dropped_fops);
 467         if (!bt->dropped_file)
 468                 goto err;
 469
 470         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 471         if (!bt->msg_file)
 472                 goto err;
 473
 474         bt->rchan = relay_open("trace", dir, buts->buf_size,
 475                                 buts->buf_nr, &blk_relay_callbacks, bt);
 476         if (!bt->rchan)
 477                 goto err;
 478
 479         bt->act_mask = buts->act_mask;
 480         if (!bt->act_mask)
 481                 bt->act_mask = (u16) -1;
 482
 483         bt->start_lba = buts->start_lba;
 484         bt->end_lba = buts->end_lba;
 485         if (!bt->end_lba)
 486                 bt->end_lba = -1ULL;
 487
 488         bt->pid = buts->pid;
 489         bt->trace_state = Blktrace_setup;
 490
 491         ret = -EBUSY;
 492         old_bt = xchg(&q->blk_trace, bt);
 493         if (old_bt) {
 494                 (void) xchg(&q->blk_trace, old_bt);
 495                 goto err;
 496         }
 497
 498         if (atomic_inc_return(&blk_probes_ref) == 1)
 499                 blk_register_tracepoints();
 500
 501         return 0;
 502 err:
 503         blk_trace_free(bt);
 504         return ret;
 505 }
 506
 507 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 508                     char __user *arg)
 509 {
 510         struct blk_user_trace_setup buts;
 511         int ret;
 512
 513         ret = copy_from_user(&buts, arg, sizeof(buts));
 514         if (ret)
 515                 return -EFAULT;
 516
 517         ret = do_blk_trace_setup(q, name, dev, &buts);
 518         if (ret)
 519                 return ret;
 520
 521         if (copy_to_user(arg, &buts, sizeof(buts)))
 522                 return -EFAULT;
 523
 524         return 0;
 525 }
 526 EXPORT_SYMBOL_GPL(blk_trace_setup);
 527
 528 int blk_trace_startstop(struct request_queue *q, int start)
 529 {
 530         int ret;
 531         struct blk_trace *bt = q->blk_trace;
 532
 533         if (bt == NULL)
 534                 return -EINVAL;
 535
 536         /*
 537          * For starting a trace, we can transition from a setup or stopped
 538          * trace. For stopping a trace, the state must be running
 539          */
 540         ret = -EINVAL;
 541         if (start) {
 542                 if (bt->trace_state == Blktrace_setup ||
 543                     bt->trace_state == Blktrace_stopped) {
 544                         blktrace_seq++;
 545                         smp_mb();
 546                         bt->trace_state = Blktrace_running;
 547
 548                         trace_note_time(bt);
 549                         ret = 0;
 550                 }
 551         } else {
 552                 if (bt->trace_state == Blktrace_running) {
 553                         bt->trace_state = Blktrace_stopped;
 554                         relay_flush(bt->rchan);
 555                         ret = 0;
 556                 }
 557         }
 558
 559         return ret;
 560 }
 561 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 562
 563 /**
 564  * blk_trace_ioctl: - handle the ioctls associated with tracing
 565  * @bdev:       the block device
 566  * @cmd:        the ioctl cmd
 567  * @arg:        the argument data, if any
 568  *
 569  **/
 570 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 571 {
 572         struct request_queue *q;
 573         int ret, start = 0;
 574         char b[BDEVNAME_SIZE];
 575
 576         q = bdev_get_queue(bdev);
 577         if (!q)
 578                 return -ENXIO;
 579
 580         mutex_lock(&bdev->bd_mutex);
 581
 582         switch (cmd) {
 583         case BLKTRACESETUP:
 584                 bdevname(bdev, b);
 585                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 586                 break;
 587         case BLKTRACESTART:
 588                 start = 1;
 589         case BLKTRACESTOP:
 590                 ret = blk_trace_startstop(q, start);
 591                 break;
 592         case BLKTRACETEARDOWN:
 593                 ret = blk_trace_remove(q);
 594                 break;
 595         default:
 596                 ret = -ENOTTY;
 597                 break;
 598         }
 599
 600         mutex_unlock(&bdev->bd_mutex);
 601         return ret;
 602 }
 603
 604 /**
 605  * blk_trace_shutdown: - stop and cleanup trace structures
 606  * @q:    the request queue associated with the device
 607  *
 608  **/
 609 void blk_trace_shutdown(struct request_queue *q)
 610 {
 611         if (q->blk_trace) {
 612                 blk_trace_startstop(q, 0);
 613                 blk_trace_remove(q);
 614         }
 615 }
 616
 617 /*
 618  * blktrace probes
 619  */
 620
 621 /**
 622  * blk_add_trace_rq - Add a trace for a request oriented action
 623  * @q:          queue the io is for
 624  * @rq:         the source request
 625  * @what:       the action
 626  *
 627  * Description:
 628  *     Records an action against a request. Will log the bio offset + size.
 629  *
 630  **/
 631 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 632                                     u32 what)
 633 {
 634         struct blk_trace *bt = q->blk_trace;
 635         int rw = rq->cmd_flags & 0x03;
 636
 637         if (likely(!bt))
 638                 return;
 639
 640         if (blk_discard_rq(rq))
 641                 rw |= (1 << BIO_RW_DISCARD);
 642
 643         if (blk_pc_request(rq)) {
 644                 what |= BLK_TC_ACT(BLK_TC_PC);
 645                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 646                                 rq->cmd_len, rq->cmd);
 647         } else  {
 648                 what |= BLK_TC_ACT(BLK_TC_FS);
 649                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 650                                 rw, what, rq->errors, 0, NULL);
 651         }
 652 }
 653
 654 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 655 {
 656         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 657 }
 658
 659 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 660 {
 661         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 662 }
 663
 664 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 665 {
 666         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 667 }
 668
 669 static void blk_add_trace_rq_requeue(struct request_queue *q,
 670                                      struct request *rq)
 671 {
 672         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 673 }
 674
 675 static void blk_add_trace_rq_complete(struct request_queue *q,
 676                                       struct request *rq)
 677 {
 678         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 679 }
 680
 681 /**
 682  * blk_add_trace_bio - Add a trace for a bio oriented action
 683  * @q:          queue the io is for
 684  * @bio:        the source bio
 685  * @what:       the action
 686  *
 687  * Description:
 688  *     Records an action against a bio. Will log the bio offset + size.
 689  *
 690  **/
 691 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 692                                      u32 what)
 693 {
 694         struct blk_trace *bt = q->blk_trace;
 695
 696         if (likely(!bt))
 697                 return;
 698
 699         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 700                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 701 }
 702
 703 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 704 {
 705         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 706 }
 707
 708 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 709 {
 710         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 711 }
 712
 713 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 714                                         struct bio *bio)
 715 {
 716         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 717 }
 718
 719 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 720                                          struct bio *bio)
 721 {
 722         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 723 }
 724
 725 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 726 {
 727         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 728 }
 729
 730 static void blk_add_trace_getrq(struct request_queue *q,
 731                                 struct bio *bio, int rw)
 732 {
 733         if (bio)
 734                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 735         else {
 736                 struct blk_trace *bt = q->blk_trace;
 737
 738                 if (bt)
 739                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 740         }
 741 }
 742
 743
 744 static void blk_add_trace_sleeprq(struct request_queue *q,
 745                                   struct bio *bio, int rw)
 746 {
 747         if (bio)
 748                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 749         else {
 750                 struct blk_trace *bt = q->blk_trace;
 751
 752                 if (bt)
 753                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 754                                         0, 0, NULL);
 755         }
 756 }
 757
 758 static void blk_add_trace_plug(struct request_queue *q)
 759 {
 760         struct blk_trace *bt = q->blk_trace;
 761
 762         if (bt)
 763                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 764 }
 765
 766 static void blk_add_trace_unplug_io(struct request_queue *q)
 767 {
 768         struct blk_trace *bt = q->blk_trace;
 769
 770         if (bt) {
 771                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 772                 __be64 rpdu = cpu_to_be64(pdu);
 773
 774                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 775                                 sizeof(rpdu), &rpdu);
 776         }
 777 }
 778
 779 static void blk_add_trace_unplug_timer(struct request_queue *q)
 780 {
 781         struct blk_trace *bt = q->blk_trace;
 782
 783         if (bt) {
 784                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 785                 __be64 rpdu = cpu_to_be64(pdu);
 786
 787                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 788                                 sizeof(rpdu), &rpdu);
 789         }
 790 }
 791
 792 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 793                                 unsigned int pdu)
 794 {
 795         struct blk_trace *bt = q->blk_trace;
 796
 797         if (bt) {
 798                 __be64 rpdu = cpu_to_be64(pdu);
 799
 800                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 801                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 802                                 sizeof(rpdu), &rpdu);
 803         }
 804 }
 805
 806 /**
 807  * blk_add_trace_remap - Add a trace for a remap operation
 808  * @q:          queue the io is for
 809  * @bio:        the source bio
 810  * @dev:        target device
 811  * @from:       source sector
 812  * @to:         target sector
 813  *
 814  * Description:
 815  *     Device mapper or raid target sometimes need to split a bio because
 816  *     it spans a stripe (or similar). Add a trace for that action.
 817  *
 818  **/
 819 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 820                                        dev_t dev, sector_t from, sector_t to)
 821 {
 822         struct blk_trace *bt = q->blk_trace;
 823         struct blk_io_trace_remap r;
 824
 825         if (likely(!bt))
 826                 return;
 827
 828         r.device = cpu_to_be32(dev);
 829         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 830         r.sector = cpu_to_be64(to);
 831
 832         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 833                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 834 }
 835
 836 /**
 837  * blk_add_driver_data - Add binary message with driver-specific data
 838  * @q:          queue the io is for
 839  * @rq:         io request
 840  * @data:       driver-specific data
 841  * @len:        length of driver-specific data
 842  *
 843  * Description:
 844  *     Some drivers might want to write driver-specific data per request.
 845  *
 846  **/
 847 void blk_add_driver_data(struct request_queue *q,
 848                          struct request *rq,
 849                          void *data, size_t len)
 850 {
 851         struct blk_trace *bt = q->blk_trace;
 852
 853         if (likely(!bt))
 854                 return;
 855
 856         if (blk_pc_request(rq))
 857                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 858                                 rq->errors, len, data);
 859         else
 860                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 861                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 862 }
 863 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 864
 865 static void blk_register_tracepoints(void)
 866 {
 867         int ret;
 868
 869         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 870         WARN_ON(ret);
 871         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 872         WARN_ON(ret);
 873         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 874         WARN_ON(ret);
 875         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 876         WARN_ON(ret);
 877         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 878         WARN_ON(ret);
 879         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 880         WARN_ON(ret);
 881         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 882         WARN_ON(ret);
 883         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 884         WARN_ON(ret);
 885         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 886         WARN_ON(ret);
 887         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 888         WARN_ON(ret);
 889         ret = register_trace_block_getrq(blk_add_trace_getrq);
 890         WARN_ON(ret);
 891         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 892         WARN_ON(ret);
 893         ret = register_trace_block_plug(blk_add_trace_plug);
 894         WARN_ON(ret);
 895         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 896         WARN_ON(ret);
 897         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 898         WARN_ON(ret);
 899         ret = register_trace_block_split(blk_add_trace_split);
 900         WARN_ON(ret);
 901         ret = register_trace_block_remap(blk_add_trace_remap);
 902         WARN_ON(ret);
 903 }
 904
 905 static void blk_unregister_tracepoints(void)
 906 {
 907         unregister_trace_block_remap(blk_add_trace_remap);
 908         unregister_trace_block_split(blk_add_trace_split);
 909         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 910         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 911         unregister_trace_block_plug(blk_add_trace_plug);
 912         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 913         unregister_trace_block_getrq(blk_add_trace_getrq);
 914         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 915         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 916         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 917         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 918         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 919         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 920         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 921         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 922         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 923         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 924
 925         tracepoint_synchronize_unregister();
 926 }
 927
 928 /*
 929  * struct blk_io_tracer formatting routines
 930  */
 931
 932 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 933 {
 934         int i = 0;
 935         int tc = t->action >> BLK_TC_SHIFT;
 936
 937         if (t->action == BLK_TN_MESSAGE) {
 938                 rwbs[i++] = 'N';
 939                 goto out;
 940         }
 941
 942         if (tc & BLK_TC_DISCARD)
 943                 rwbs[i++] = 'D';
 944         else if (tc & BLK_TC_WRITE)
 945                 rwbs[i++] = 'W';
 946         else if (t->bytes)
 947                 rwbs[i++] = 'R';
 948         else
 949                 rwbs[i++] = 'N';
 950
 951         if (tc & BLK_TC_AHEAD)
 952                 rwbs[i++] = 'A';
 953         if (tc & BLK_TC_BARRIER)
 954                 rwbs[i++] = 'B';
 955         if (tc & BLK_TC_SYNC)
 956                 rwbs[i++] = 'S';
 957         if (tc & BLK_TC_META)
 958                 rwbs[i++] = 'M';
 959 out:
 960         rwbs[i] = '\0';
 961 }
 962
 963 static inline
 964 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 965 {
 966         return (const struct blk_io_trace *)ent;
 967 }
 968
 969 static inline const void *pdu_start(const struct trace_entry *ent)
 970 {
 971         return te_blk_io_trace(ent) + 1;
 972 }
 973
 974 static inline u32 t_sec(const struct trace_entry *ent)
 975 {
 976         return te_blk_io_trace(ent)->bytes >> 9;
 977 }
 978
 979 static inline unsigned long long t_sector(const struct trace_entry *ent)
 980 {
 981         return te_blk_io_trace(ent)->sector;
 982 }
 983
 984 static inline __u16 t_error(const struct trace_entry *ent)
 985 {
 986         return te_blk_io_trace(ent)->error;
 987 }
 988
 989 static __u64 get_pdu_int(const struct trace_entry *ent)
 990 {
 991         const __u64 *val = pdu_start(ent);
 992         return be64_to_cpu(*val);
 993 }
 994
 995 static void get_pdu_remap(const struct trace_entry *ent,
 996                           struct blk_io_trace_remap *r)
 997 {
 998         const struct blk_io_trace_remap *__r = pdu_start(ent);
 999         __u64 sector = __r->sector;
1000
1001         r->device = be32_to_cpu(__r->device);
1002         r->device_from = be32_to_cpu(__r->device_from);
1003         r->sector = be64_to_cpu(sector);
1004 }
1005
1006 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1007
1008 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1009 {
1010         char rwbs[6];
1011         unsigned long long ts  = iter->ts;
1012         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1013         unsigned secs          = (unsigned long)ts;
1014         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1015
1016         fill_rwbs(rwbs, t);
1017
1018         return trace_seq_printf(&iter->seq,
1019                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1020                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1021                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1022 }
1023
1024 static int blk_log_action(struct trace_iterator *iter, const char *act)
1025 {
1026         char rwbs[6];
1027         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1028
1029         fill_rwbs(rwbs, t);
1030         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1031                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1032 }
1033
1034 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1035 {
1036         char cmd[TASK_COMM_LEN];
1037
1038         trace_find_cmdline(ent->pid, cmd);
1039
1040         if (t_sec(ent))
1041                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1042                                         t_sector(ent), t_sec(ent), cmd);
1043         return trace_seq_printf(s, "[%s]\n", cmd);
1044 }
1045
1046 static int blk_log_with_error(struct trace_seq *s,
1047                               const struct trace_entry *ent)
1048 {
1049         if (t_sec(ent))
1050                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1051                                         t_sec(ent), t_error(ent));
1052         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1053 }
1054
1055 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1056 {
1057         struct blk_io_trace_remap r = { .device = 0, };
1058
1059         get_pdu_remap(ent, &r);
1060         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1061                                t_sector(ent),
1062                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1063                                (unsigned long long)r.sector);
1064 }
1065
1066 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1067 {
1068         char cmd[TASK_COMM_LEN];
1069
1070         trace_find_cmdline(ent->pid, cmd);
1071
1072         return trace_seq_printf(s, "[%s]\n", cmd);
1073 }
1074
1075 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1076 {
1077         char cmd[TASK_COMM_LEN];
1078
1079         trace_find_cmdline(ent->pid, cmd);
1080
1081         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1082 }
1083
1084 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1085 {
1086         char cmd[TASK_COMM_LEN];
1087
1088         trace_find_cmdline(ent->pid, cmd);
1089
1090         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1091                                 get_pdu_int(ent), cmd);
1092 }
1093
1094 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1095 {
1096         int ret;
1097         const struct blk_io_trace *t = te_blk_io_trace(ent);
1098
1099         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1100         if (ret)
1101                 return trace_seq_putc(s, '\n');
1102         return ret;
1103 }
1104
1105 /*
1106  * struct tracer operations
1107  */
1108
1109 static void blk_tracer_print_header(struct seq_file *m)
1110 {
1111         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1112                 return;
1113         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1114                     "#  |     |     |           |   |   |\n");
1115 }
1116
1117 static void blk_tracer_start(struct trace_array *tr)
1118 {
1119         blk_tracer_enabled = true;
1120         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1121 }
1122
1123 static int blk_tracer_init(struct trace_array *tr)
1124 {
1125         blk_tr = tr;
1126         blk_tracer_start(tr);
1127         return 0;
1128 }
1129
1130 static void blk_tracer_stop(struct trace_array *tr)
1131 {
1132         blk_tracer_enabled = false;
1133         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1134 }
1135
1136 static void blk_tracer_reset(struct trace_array *tr)
1137 {
1138         blk_tracer_stop(tr);
1139 }
1140
1141 static const struct {
1142         const char *act[2];
1143         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1144 } what2act[] = {
1145         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1146         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1147         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1148         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1149         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1150         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1151         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1152         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1153         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1154         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1155         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1156         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1157         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1158         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1159         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1160 };
1161
1162 static enum print_line_t print_one_line(struct trace_iterator *iter,
1163                                         bool classic)
1164 {
1165         struct trace_seq *s = &iter->seq;
1166         const struct blk_io_trace *t;
1167         u16 what;
1168         int ret;
1169         bool long_act;
1170         blk_log_action_t *log_action;
1171
1172         t          = te_blk_io_trace(iter->ent);
1173         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1174         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1175         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1176
1177         if (t->action == BLK_TN_MESSAGE) {
1178                 ret = log_action(iter, long_act ? "message" : "m");
1179                 if (ret)
1180                         ret = blk_log_msg(s, iter->ent);
1181                 goto out;
1182         }
1183
1184         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1185                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1186         else {
1187                 ret = log_action(iter, what2act[what].act[long_act]);
1188                 if (ret)
1189                         ret = what2act[what].print(s, iter->ent);
1190         }
1191 out:
1192         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1193 }
1194
1195 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1196                                                int flags)
1197 {
1198         if (!trace_print_context(iter))
1199                 return TRACE_TYPE_PARTIAL_LINE;
1200
1201         return print_one_line(iter, false);
1202 }
1203
1204 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1205 {
1206         struct trace_seq *s = &iter->seq;
1207         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1208         const int offset = offsetof(struct blk_io_trace, sector);
1209         struct blk_io_trace old = {
1210                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1211                 .time     = iter->ts,
1212         };
1213
1214         if (!trace_seq_putmem(s, &old, offset))
1215                 return 0;
1216         return trace_seq_putmem(s, &t->sector,
1217                                 sizeof(old) - offset + t->pdu_len);
1218 }
1219
1220 static enum print_line_t
1221 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1222 {
1223         return blk_trace_synthesize_old_trace(iter) ?
1224                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1225 }
1226
1227 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1228 {
1229         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1230                 return TRACE_TYPE_UNHANDLED;
1231
1232         return print_one_line(iter, true);
1233 }
1234
1235 static struct tracer blk_tracer __read_mostly = {
1236         .name           = "blk",
1237         .init           = blk_tracer_init,
1238         .reset          = blk_tracer_reset,
1239         .start          = blk_tracer_start,
1240         .stop           = blk_tracer_stop,
1241         .print_header   = blk_tracer_print_header,
1242         .print_line     = blk_tracer_print_line,
1243         .flags          = &blk_tracer_flags,
1244 };
1245
1246 static struct trace_event trace_blk_event = {
1247         .type           = TRACE_BLK,
1248         .trace          = blk_trace_event_print,
1249         .binary         = blk_trace_event_print_binary,
1250 };
1251
1252 static int __init init_blk_tracer(void)
1253 {
1254         if (!register_ftrace_event(&trace_blk_event)) {
1255                 pr_warning("Warning: could not register block events\n");
1256                 return 1;
1257         }
1258
1259         if (register_tracer(&blk_tracer) != 0) {
1260                 pr_warning("Warning: could not register the block tracer\n");
1261                 unregister_ftrace_event(&trace_blk_event);
1262                 return 1;
1263         }
1264
1265         return 0;
1266 }
1267
1268 device_initcall(init_blk_tracer);
1269
1270 static int blk_trace_remove_queue(struct request_queue *q)
1271 {
1272         struct blk_trace *bt;
1273
1274         bt = xchg(&q->blk_trace, NULL);
1275         if (bt == NULL)
1276                 return -EINVAL;
1277
1278         if (atomic_dec_and_test(&blk_probes_ref))
1279                 blk_unregister_tracepoints();
1280
1281         blk_trace_free(bt);
1282         return 0;
1283 }
1284
1285 /*
1286  * Setup everything required to start tracing
1287  */
1288 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1289 {
1290         struct blk_trace *old_bt, *bt = NULL;
1291         int ret = -ENOMEM;
1292
1293         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1294         if (!bt)
1295                 return -ENOMEM;
1296
1297         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1298         if (!bt->msg_data)
1299                 goto free_bt;
1300
1301         bt->dev = dev;
1302         bt->act_mask = (u16)-1;
1303         bt->end_lba = -1ULL;
1304
1305         old_bt = xchg(&q->blk_trace, bt);
1306         if (old_bt != NULL) {
1307                 (void)xchg(&q->blk_trace, old_bt);
1308                 ret = -EBUSY;
1309                 goto free_bt;
1310         }
1311
1312         if (atomic_inc_return(&blk_probes_ref) == 1)
1313                 blk_register_tracepoints();
1314         return 0;
1315
1316 free_bt:
1317         blk_trace_free(bt);
1318         return ret;
1319 }
1320
1321 /*
1322  * sysfs interface to enable and configure tracing
1323  */
1324
1325 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1326                                          struct device_attribute *attr,
1327                                          char *buf);
1328 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1329                                           struct device_attribute *attr,
1330                                           const char *buf, size_t count);
1331 #define BLK_TRACE_DEVICE_ATTR(_name) \
1332         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1333                     sysfs_blk_trace_attr_show, \
1334                     sysfs_blk_trace_attr_store)
1335
1336 static BLK_TRACE_DEVICE_ATTR(enable);
1337 static BLK_TRACE_DEVICE_ATTR(act_mask);
1338 static BLK_TRACE_DEVICE_ATTR(pid);
1339 static BLK_TRACE_DEVICE_ATTR(start_lba);
1340 static BLK_TRACE_DEVICE_ATTR(end_lba);
1341
1342 static struct attribute *blk_trace_attrs[] = {
1343         &dev_attr_enable.attr,
1344         &dev_attr_act_mask.attr,
1345         &dev_attr_pid.attr,
1346         &dev_attr_start_lba.attr,
1347         &dev_attr_end_lba.attr,
1348         NULL
1349 };
1350
1351 struct attribute_group blk_trace_attr_group = {
1352         .name  = "trace",
1353         .attrs = blk_trace_attrs,
1354 };
1355
1356 static const struct {
1357         int mask;
1358         const char *str;
1359 } mask_maps[] = {
1360         { BLK_TC_READ,          "read"          },
1361         { BLK_TC_WRITE,         "write"         },
1362         { BLK_TC_BARRIER,       "barrier"       },
1363         { BLK_TC_SYNC,          "sync"          },
1364         { BLK_TC_QUEUE,         "queue"         },
1365         { BLK_TC_REQUEUE,       "requeue"       },
1366         { BLK_TC_ISSUE,         "issue"         },
1367         { BLK_TC_COMPLETE,      "complete"      },
1368         { BLK_TC_FS,            "fs"            },
1369         { BLK_TC_PC,            "pc"            },
1370         { BLK_TC_AHEAD,         "ahead"         },
1371         { BLK_TC_META,          "meta"          },
1372         { BLK_TC_DISCARD,       "discard"       },
1373         { BLK_TC_DRV_DATA,      "drv_data"      },
1374 };
1375
1376 static int blk_trace_str2mask(const char *str)
1377 {
1378         int i;
1379         int mask = 0;
1380         char *buf, *s, *token;
1381
1382         buf = kstrdup(str, GFP_KERNEL);
1383         if (buf == NULL)
1384                 return -ENOMEM;
1385         s = strstrip(buf);
1386
1387         while (1) {
1388                 token = strsep(&s, ",");
1389                 if (token == NULL)
1390                         break;
1391
1392                 if (*token == '\0')
1393                         continue;
1394
1395                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1396                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1397                                 mask |= mask_maps[i].mask;
1398                                 break;
1399                         }
1400                 }
1401                 if (i == ARRAY_SIZE(mask_maps)) {
1402                         mask = -EINVAL;
1403                         break;
1404                 }
1405         }
1406         kfree(buf);
1407
1408         return mask;
1409 }
1410
1411 static ssize_t blk_trace_mask2str(char *buf, int mask)
1412 {
1413         int i;
1414         char *p = buf;
1415
1416         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1417                 if (mask & mask_maps[i].mask) {
1418                         p += sprintf(p, "%s%s",
1419                                     (p == buf) ? "" : ",", mask_maps[i].str);
1420                 }
1421         }
1422         *p++ = '\n';
1423
1424         return p - buf;
1425 }
1426
1427 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1428 {
1429         if (bdev->bd_disk == NULL)
1430                 return NULL;
1431
1432         return bdev_get_queue(bdev);
1433 }
1434
1435 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1436                                          struct device_attribute *attr,
1437                                          char *buf)
1438 {
1439         struct hd_struct *p = dev_to_part(dev);
1440         struct request_queue *q;
1441         struct block_device *bdev;
1442         ssize_t ret = -ENXIO;
1443
1444         lock_kernel();
1445         bdev = bdget(part_devt(p));
1446         if (bdev == NULL)
1447                 goto out_unlock_kernel;
1448
1449         q = blk_trace_get_queue(bdev);
1450         if (q == NULL)
1451                 goto out_bdput;
1452
1453         mutex_lock(&bdev->bd_mutex);
1454
1455         if (attr == &dev_attr_enable) {
1456                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1457                 goto out_unlock_bdev;
1458         }
1459
1460         if (q->blk_trace == NULL)
1461                 ret = sprintf(buf, "disabled\n");
1462         else if (attr == &dev_attr_act_mask)
1463                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1464         else if (attr == &dev_attr_pid)
1465                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1466         else if (attr == &dev_attr_start_lba)
1467                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1468         else if (attr == &dev_attr_end_lba)
1469                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1470
1471 out_unlock_bdev:
1472         mutex_unlock(&bdev->bd_mutex);
1473 out_bdput:
1474         bdput(bdev);
1475 out_unlock_kernel:
1476         unlock_kernel();
1477         return ret;
1478 }
1479
1480 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1481                                           struct device_attribute *attr,
1482                                           const char *buf, size_t count)
1483 {
1484         struct block_device *bdev;
1485         struct request_queue *q;
1486         struct hd_struct *p;
1487         u64 value;
1488         ssize_t ret = -EINVAL;
1489
1490         if (count == 0)
1491                 goto out;
1492
1493         if (attr == &dev_attr_act_mask) {
1494                 if (sscanf(buf, "%llx", &value) != 1) {
1495                         /* Assume it is a list of trace category names */
1496                         ret = blk_trace_str2mask(buf);
1497                         if (ret < 0)
1498                                 goto out;
1499                         value = ret;
1500                 }
1501         } else if (sscanf(buf, "%llu", &value) != 1)
1502                 goto out;
1503
1504         ret = -ENXIO;
1505
1506         lock_kernel();
1507         p = dev_to_part(dev);
1508         bdev = bdget(part_devt(p));
1509         if (bdev == NULL)
1510                 goto out_unlock_kernel;
1511
1512         q = blk_trace_get_queue(bdev);
1513         if (q == NULL)
1514                 goto out_bdput;
1515
1516         mutex_lock(&bdev->bd_mutex);
1517
1518         if (attr == &dev_attr_enable) {
1519                 if (value)
1520                         ret = blk_trace_setup_queue(q, bdev->bd_dev);
1521                 else
1522                         ret = blk_trace_remove_queue(q);
1523                 goto out_unlock_bdev;
1524         }
1525
1526         ret = 0;
1527         if (q->blk_trace == NULL)
1528                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1529
1530         if (ret == 0) {
1531                 if (attr == &dev_attr_act_mask)
1532                         q->blk_trace->act_mask = value;
1533                 else if (attr == &dev_attr_pid)
1534                         q->blk_trace->pid = value;
1535                 else if (attr == &dev_attr_start_lba)
1536                         q->blk_trace->start_lba = value;
1537                 else if (attr == &dev_attr_end_lba)
1538                         q->blk_trace->end_lba = value;
1539         }
1540
1541 out_unlock_bdev:
1542         mutex_unlock(&bdev->bd_mutex);
1543 out_bdput:
1544         bdput(bdev);
1545 out_unlock_kernel:
1546         unlock_kernel();
1547 out:
1548         return ret ? ret : count;
1549 }
1550