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 <linux/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),
 152                                          BLK_TC_ACT(BLK_TC_WRITE) };
 153
 154 /* The ilog2() calls fall out because they're constant */
 155 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 156           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 157
 158 /*
 159  * The worker for the various blk_add_trace*() types. Fills out a
 160  * blk_io_trace structure and places it in a per-cpu subbuffer.
 161  */
 162 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 163                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 164 {
 165         struct task_struct *tsk = current;
 166         struct ring_buffer_event *event = NULL;
 167         struct blk_io_trace *t;
 168         unsigned long flags;
 169         unsigned long *sequence;
 170         pid_t pid;
 171         int cpu, pc = 0;
 172
 173         if (unlikely(bt->trace_state != Blktrace_running ||
 174                      !blk_tracer_enabled))
 175                 return;
 176
 177         what |= ddir_act[rw & WRITE];
 178         what |= MASK_TC_BIT(rw, BARRIER);
 179         what |= MASK_TC_BIT(rw, SYNC);
 180         what |= MASK_TC_BIT(rw, AHEAD);
 181         what |= MASK_TC_BIT(rw, META);
 182         what |= MASK_TC_BIT(rw, DISCARD);
 183
 184         pid = tsk->pid;
 185         if (unlikely(act_log_check(bt, what, sector, pid)))
 186                 return;
 187         cpu = raw_smp_processor_id();
 188
 189         if (blk_tr) {
 190                 struct trace_entry *ent;
 191                 tracing_record_cmdline(current);
 192
 193                 event = ring_buffer_lock_reserve(blk_tr->buffer,
 194                                                  sizeof(*t) + pdu_len, &flags);
 195                 if (!event)
 196                         return;
 197
 198                 ent = ring_buffer_event_data(event);
 199                 t = (struct blk_io_trace *)ent;
 200                 pc = preempt_count();
 201                 tracing_generic_entry_update(ent, 0, pc);
 202                 ent->type = TRACE_BLK;
 203                 goto record_it;
 204         }
 205
 206         /*
 207          * A word about the locking here - we disable interrupts to reserve
 208          * some space in the relay per-cpu buffer, to prevent an irq
 209          * from coming in and stepping on our toes.
 210          */
 211         local_irq_save(flags);
 212
 213         if (unlikely(tsk->btrace_seq != blktrace_seq))
 214                 trace_note_tsk(bt, tsk);
 215
 216         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 217         if (t) {
 218                 sequence = per_cpu_ptr(bt->sequence, cpu);
 219
 220                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 221                 t->sequence = ++(*sequence);
 222                 t->time = ktime_to_ns(ktime_get());
 223 record_it:
 224                 /*
 225                  * These two are not needed in ftrace as they are in the
 226                  * generic trace_entry, filled by tracing_generic_entry_update,
 227                  * but for the trace_event->bin() synthesizer benefit we do it
 228                  * here too.
 229                  */
 230                 t->cpu = cpu;
 231                 t->pid = pid;
 232
 233                 t->sector = sector;
 234                 t->bytes = bytes;
 235                 t->action = what;
 236                 t->device = bt->dev;
 237                 t->error = error;
 238                 t->pdu_len = pdu_len;
 239
 240                 if (pdu_len)
 241                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 242
 243                 if (blk_tr) {
 244                         ring_buffer_unlock_commit(blk_tr->buffer, event, flags);
 245                         if (pid != 0 &&
 246                             !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC) &&
 247                             (trace_flags & TRACE_ITER_STACKTRACE) != 0)
 248                                 __trace_stack(blk_tr, NULL, flags, 5, pc);
 249                         trace_wake_up();
 250                         return;
 251                 }
 252         }
 253
 254         local_irq_restore(flags);
 255 }
 256
 257 static struct dentry *blk_tree_root;
 258 static DEFINE_MUTEX(blk_tree_mutex);
 259
 260 static void blk_trace_cleanup(struct blk_trace *bt)
 261 {
 262         debugfs_remove(bt->msg_file);
 263         debugfs_remove(bt->dropped_file);
 264         relay_close(bt->rchan);
 265         free_percpu(bt->sequence);
 266         free_percpu(bt->msg_data);
 267         kfree(bt);
 268         mutex_lock(&blk_probe_mutex);
 269         if (atomic_dec_and_test(&blk_probes_ref))
 270                 blk_unregister_tracepoints();
 271         mutex_unlock(&blk_probe_mutex);
 272 }
 273
 274 int blk_trace_remove(struct request_queue *q)
 275 {
 276         struct blk_trace *bt;
 277
 278         bt = xchg(&q->blk_trace, NULL);
 279         if (!bt)
 280                 return -EINVAL;
 281
 282         if (bt->trace_state == Blktrace_setup ||
 283             bt->trace_state == Blktrace_stopped)
 284                 blk_trace_cleanup(bt);
 285
 286         return 0;
 287 }
 288 EXPORT_SYMBOL_GPL(blk_trace_remove);
 289
 290 static int blk_dropped_open(struct inode *inode, struct file *filp)
 291 {
 292         filp->private_data = inode->i_private;
 293
 294         return 0;
 295 }
 296
 297 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 298                                 size_t count, loff_t *ppos)
 299 {
 300         struct blk_trace *bt = filp->private_data;
 301         char buf[16];
 302
 303         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 304
 305         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 306 }
 307
 308 static const struct file_operations blk_dropped_fops = {
 309         .owner =        THIS_MODULE,
 310         .open =         blk_dropped_open,
 311         .read =         blk_dropped_read,
 312 };
 313
 314 static int blk_msg_open(struct inode *inode, struct file *filp)
 315 {
 316         filp->private_data = inode->i_private;
 317
 318         return 0;
 319 }
 320
 321 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 322                                 size_t count, loff_t *ppos)
 323 {
 324         char *msg;
 325         struct blk_trace *bt;
 326
 327         if (count > BLK_TN_MAX_MSG)
 328                 return -EINVAL;
 329
 330         msg = kmalloc(count, GFP_KERNEL);
 331         if (msg == NULL)
 332                 return -ENOMEM;
 333
 334         if (copy_from_user(msg, buffer, count)) {
 335                 kfree(msg);
 336                 return -EFAULT;
 337         }
 338
 339         bt = filp->private_data;
 340         __trace_note_message(bt, "%s", msg);
 341         kfree(msg);
 342
 343         return count;
 344 }
 345
 346 static const struct file_operations blk_msg_fops = {
 347         .owner =        THIS_MODULE,
 348         .open =         blk_msg_open,
 349         .write =        blk_msg_write,
 350 };
 351
 352 /*
 353  * Keep track of how many times we encountered a full subbuffer, to aid
 354  * the user space app in telling how many lost events there were.
 355  */
 356 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 357                                      void *prev_subbuf, size_t prev_padding)
 358 {
 359         struct blk_trace *bt;
 360
 361         if (!relay_buf_full(buf))
 362                 return 1;
 363
 364         bt = buf->chan->private_data;
 365         atomic_inc(&bt->dropped);
 366         return 0;
 367 }
 368
 369 static int blk_remove_buf_file_callback(struct dentry *dentry)
 370 {
 371         struct dentry *parent = dentry->d_parent;
 372         debugfs_remove(dentry);
 373
 374         /*
 375         * this will fail for all but the last file, but that is ok. what we
 376         * care about is the top level buts->name directory going away, when
 377         * the last trace file is gone. Then we don't have to rmdir() that
 378         * manually on trace stop, so it nicely solves the issue with
 379         * force killing of running traces.
 380         */
 381
 382         debugfs_remove(parent);
 383         return 0;
 384 }
 385
 386 static struct dentry *blk_create_buf_file_callback(const char *filename,
 387                                                    struct dentry *parent,
 388                                                    int mode,
 389                                                    struct rchan_buf *buf,
 390                                                    int *is_global)
 391 {
 392         return debugfs_create_file(filename, mode, parent, buf,
 393                                         &relay_file_operations);
 394 }
 395
 396 static struct rchan_callbacks blk_relay_callbacks = {
 397         .subbuf_start           = blk_subbuf_start_callback,
 398         .create_buf_file        = blk_create_buf_file_callback,
 399         .remove_buf_file        = blk_remove_buf_file_callback,
 400 };
 401
 402 /*
 403  * Setup everything required to start tracing
 404  */
 405 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 406                         struct blk_user_trace_setup *buts)
 407 {
 408         struct blk_trace *old_bt, *bt = NULL;
 409         struct dentry *dir = NULL;
 410         int ret, i;
 411
 412         if (!buts->buf_size || !buts->buf_nr)
 413                 return -EINVAL;
 414
 415         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 416         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 417
 418         /*
 419          * some device names have larger paths - convert the slashes
 420          * to underscores for this to work as expected
 421          */
 422         for (i = 0; i < strlen(buts->name); i++)
 423                 if (buts->name[i] == '/')
 424                         buts->name[i] = '_';
 425
 426         ret = -ENOMEM;
 427         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 428         if (!bt)
 429                 goto err;
 430
 431         bt->sequence = alloc_percpu(unsigned long);
 432         if (!bt->sequence)
 433                 goto err;
 434
 435         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
 436         if (!bt->msg_data)
 437                 goto err;
 438
 439         ret = -ENOENT;
 440
 441         if (!blk_tree_root) {
 442                 blk_tree_root = debugfs_create_dir("block", NULL);
 443                 if (!blk_tree_root)
 444                         return -ENOMEM;
 445         }
 446
 447         dir = debugfs_create_dir(buts->name, blk_tree_root);
 448
 449         if (!dir)
 450                 goto err;
 451
 452         bt->dir = dir;
 453         bt->dev = dev;
 454         atomic_set(&bt->dropped, 0);
 455
 456         ret = -EIO;
 457         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 458                                                &blk_dropped_fops);
 459         if (!bt->dropped_file)
 460                 goto err;
 461
 462         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 463         if (!bt->msg_file)
 464                 goto err;
 465
 466         bt->rchan = relay_open("trace", dir, buts->buf_size,
 467                                 buts->buf_nr, &blk_relay_callbacks, bt);
 468         if (!bt->rchan)
 469                 goto err;
 470
 471         bt->act_mask = buts->act_mask;
 472         if (!bt->act_mask)
 473                 bt->act_mask = (u16) -1;
 474
 475         bt->start_lba = buts->start_lba;
 476         bt->end_lba = buts->end_lba;
 477         if (!bt->end_lba)
 478                 bt->end_lba = -1ULL;
 479
 480         bt->pid = buts->pid;
 481         bt->trace_state = Blktrace_setup;
 482
 483         mutex_lock(&blk_probe_mutex);
 484         if (atomic_add_return(1, &blk_probes_ref) == 1) {
 485                 ret = blk_register_tracepoints();
 486                 if (ret)
 487                         goto probe_err;
 488         }
 489         mutex_unlock(&blk_probe_mutex);
 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         return 0;
 499 probe_err:
 500         atomic_dec(&blk_probes_ref);
 501         mutex_unlock(&blk_probe_mutex);
 502 err:
 503         if (bt) {
 504                 if (bt->msg_file)
 505                         debugfs_remove(bt->msg_file);
 506                 if (bt->dropped_file)
 507                         debugfs_remove(bt->dropped_file);
 508                 free_percpu(bt->sequence);
 509                 free_percpu(bt->msg_data);
 510                 if (bt->rchan)
 511                         relay_close(bt->rchan);
 512                 kfree(bt);
 513         }
 514         return ret;
 515 }
 516
 517 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 518                     char __user *arg)
 519 {
 520         struct blk_user_trace_setup buts;
 521         int ret;
 522
 523         ret = copy_from_user(&buts, arg, sizeof(buts));
 524         if (ret)
 525                 return -EFAULT;
 526
 527         ret = do_blk_trace_setup(q, name, dev, &buts);
 528         if (ret)
 529                 return ret;
 530
 531         if (copy_to_user(arg, &buts, sizeof(buts)))
 532                 return -EFAULT;
 533
 534         return 0;
 535 }
 536 EXPORT_SYMBOL_GPL(blk_trace_setup);
 537
 538 int blk_trace_startstop(struct request_queue *q, int start)
 539 {
 540         int ret;
 541         struct blk_trace *bt = q->blk_trace;
 542
 543         if (bt == NULL)
 544                 return -EINVAL;
 545
 546         /*
 547          * For starting a trace, we can transition from a setup or stopped
 548          * trace. For stopping a trace, the state must be running
 549          */
 550         ret = -EINVAL;
 551         if (start) {
 552                 if (bt->trace_state == Blktrace_setup ||
 553                     bt->trace_state == Blktrace_stopped) {
 554                         blktrace_seq++;
 555                         smp_mb();
 556                         bt->trace_state = Blktrace_running;
 557
 558                         trace_note_time(bt);
 559                         ret = 0;
 560                 }
 561         } else {
 562                 if (bt->trace_state == Blktrace_running) {
 563                         bt->trace_state = Blktrace_stopped;
 564                         relay_flush(bt->rchan);
 565                         ret = 0;
 566                 }
 567         }
 568
 569         return ret;
 570 }
 571 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 572
 573 /**
 574  * blk_trace_ioctl: - handle the ioctls associated with tracing
 575  * @bdev:       the block device
 576  * @cmd:        the ioctl cmd
 577  * @arg:        the argument data, if any
 578  *
 579  **/
 580 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 581 {
 582         struct request_queue *q;
 583         int ret, start = 0;
 584         char b[BDEVNAME_SIZE];
 585
 586         q = bdev_get_queue(bdev);
 587         if (!q)
 588                 return -ENXIO;
 589
 590         mutex_lock(&bdev->bd_mutex);
 591
 592         switch (cmd) {
 593         case BLKTRACESETUP:
 594                 bdevname(bdev, b);
 595                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 596                 break;
 597         case BLKTRACESTART:
 598                 start = 1;
 599         case BLKTRACESTOP:
 600                 ret = blk_trace_startstop(q, start);
 601                 break;
 602         case BLKTRACETEARDOWN:
 603                 ret = blk_trace_remove(q);
 604                 break;
 605         default:
 606                 ret = -ENOTTY;
 607                 break;
 608         }
 609
 610         mutex_unlock(&bdev->bd_mutex);
 611         return ret;
 612 }
 613
 614 /**
 615  * blk_trace_shutdown: - stop and cleanup trace structures
 616  * @q:    the request queue associated with the device
 617  *
 618  **/
 619 void blk_trace_shutdown(struct request_queue *q)
 620 {
 621         if (q->blk_trace) {
 622                 blk_trace_startstop(q, 0);
 623                 blk_trace_remove(q);
 624         }
 625 }
 626
 627 /*
 628  * blktrace probes
 629  */
 630
 631 /**
 632  * blk_add_trace_rq - Add a trace for a request oriented action
 633  * @q:          queue the io is for
 634  * @rq:         the source request
 635  * @what:       the action
 636  *
 637  * Description:
 638  *     Records an action against a request. Will log the bio offset + size.
 639  *
 640  **/
 641 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 642                                     u32 what)
 643 {
 644         struct blk_trace *bt = q->blk_trace;
 645         int rw = rq->cmd_flags & 0x03;
 646
 647         if (likely(!bt))
 648                 return;
 649
 650         if (blk_discard_rq(rq))
 651                 rw |= (1 << BIO_RW_DISCARD);
 652
 653         if (blk_pc_request(rq)) {
 654                 what |= BLK_TC_ACT(BLK_TC_PC);
 655                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 656                                 sizeof(rq->cmd), rq->cmd);
 657         } else  {
 658                 what |= BLK_TC_ACT(BLK_TC_FS);
 659                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 660                                 rw, what, rq->errors, 0, NULL);
 661         }
 662 }
 663
 664 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 665 {
 666         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 667 }
 668
 669 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 670 {
 671         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 672 }
 673
 674 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 675 {
 676         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 677 }
 678
 679 static void blk_add_trace_rq_requeue(struct request_queue *q,
 680                                      struct request *rq)
 681 {
 682         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 683 }
 684
 685 static void blk_add_trace_rq_complete(struct request_queue *q,
 686                                       struct request *rq)
 687 {
 688         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 689 }
 690
 691 /**
 692  * blk_add_trace_bio - Add a trace for a bio oriented action
 693  * @q:          queue the io is for
 694  * @bio:        the source bio
 695  * @what:       the action
 696  *
 697  * Description:
 698  *     Records an action against a bio. Will log the bio offset + size.
 699  *
 700  **/
 701 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 702                                      u32 what)
 703 {
 704         struct blk_trace *bt = q->blk_trace;
 705
 706         if (likely(!bt))
 707                 return;
 708
 709         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 710                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 711 }
 712
 713 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 714 {
 715         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 716 }
 717
 718 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 719 {
 720         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 721 }
 722
 723 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 724                                         struct bio *bio)
 725 {
 726         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 727 }
 728
 729 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 730                                          struct bio *bio)
 731 {
 732         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 733 }
 734
 735 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 736 {
 737         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 738 }
 739
 740 static void blk_add_trace_getrq(struct request_queue *q,
 741                                 struct bio *bio, int rw)
 742 {
 743         if (bio)
 744                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 745         else {
 746                 struct blk_trace *bt = q->blk_trace;
 747
 748                 if (bt)
 749                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 750         }
 751 }
 752
 753
 754 static void blk_add_trace_sleeprq(struct request_queue *q,
 755                                   struct bio *bio, int rw)
 756 {
 757         if (bio)
 758                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 759         else {
 760                 struct blk_trace *bt = q->blk_trace;
 761
 762                 if (bt)
 763                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 764                                         0, 0, NULL);
 765         }
 766 }
 767
 768 static void blk_add_trace_plug(struct request_queue *q)
 769 {
 770         struct blk_trace *bt = q->blk_trace;
 771
 772         if (bt)
 773                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 774 }
 775
 776 static void blk_add_trace_unplug_io(struct request_queue *q)
 777 {
 778         struct blk_trace *bt = q->blk_trace;
 779
 780         if (bt) {
 781                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 782                 __be64 rpdu = cpu_to_be64(pdu);
 783
 784                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 785                                 sizeof(rpdu), &rpdu);
 786         }
 787 }
 788
 789 static void blk_add_trace_unplug_timer(struct request_queue *q)
 790 {
 791         struct blk_trace *bt = q->blk_trace;
 792
 793         if (bt) {
 794                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 795                 __be64 rpdu = cpu_to_be64(pdu);
 796
 797                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 798                                 sizeof(rpdu), &rpdu);
 799         }
 800 }
 801
 802 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 803                                 unsigned int pdu)
 804 {
 805         struct blk_trace *bt = q->blk_trace;
 806
 807         if (bt) {
 808                 __be64 rpdu = cpu_to_be64(pdu);
 809
 810                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 811                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 812                                 sizeof(rpdu), &rpdu);
 813         }
 814 }
 815
 816 /**
 817  * blk_add_trace_remap - Add a trace for a remap operation
 818  * @q:          queue the io is for
 819  * @bio:        the source bio
 820  * @dev:        target device
 821  * @from:       source sector
 822  * @to:         target sector
 823  *
 824  * Description:
 825  *     Device mapper or raid target sometimes need to split a bio because
 826  *     it spans a stripe (or similar). Add a trace for that action.
 827  *
 828  **/
 829 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 830                                        dev_t dev, sector_t from, sector_t to)
 831 {
 832         struct blk_trace *bt = q->blk_trace;
 833         struct blk_io_trace_remap r;
 834
 835         if (likely(!bt))
 836                 return;
 837
 838         r.device = cpu_to_be32(dev);
 839         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 840         r.sector = cpu_to_be64(to);
 841
 842         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 843                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 844 }
 845
 846 /**
 847  * blk_add_driver_data - Add binary message with driver-specific data
 848  * @q:          queue the io is for
 849  * @rq:         io request
 850  * @data:       driver-specific data
 851  * @len:        length of driver-specific data
 852  *
 853  * Description:
 854  *     Some drivers might want to write driver-specific data per request.
 855  *
 856  **/
 857 void blk_add_driver_data(struct request_queue *q,
 858                          struct request *rq,
 859                          void *data, size_t len)
 860 {
 861         struct blk_trace *bt = q->blk_trace;
 862
 863         if (likely(!bt))
 864                 return;
 865
 866         if (blk_pc_request(rq))
 867                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 868                                 rq->errors, len, data);
 869         else
 870                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 871                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 872 }
 873 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 874
 875 static int blk_register_tracepoints(void)
 876 {
 877         int ret;
 878
 879         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 880         WARN_ON(ret);
 881         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 882         WARN_ON(ret);
 883         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 884         WARN_ON(ret);
 885         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 886         WARN_ON(ret);
 887         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 888         WARN_ON(ret);
 889         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 890         WARN_ON(ret);
 891         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 892         WARN_ON(ret);
 893         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 894         WARN_ON(ret);
 895         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 896         WARN_ON(ret);
 897         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 898         WARN_ON(ret);
 899         ret = register_trace_block_getrq(blk_add_trace_getrq);
 900         WARN_ON(ret);
 901         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 902         WARN_ON(ret);
 903         ret = register_trace_block_plug(blk_add_trace_plug);
 904         WARN_ON(ret);
 905         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 906         WARN_ON(ret);
 907         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 908         WARN_ON(ret);
 909         ret = register_trace_block_split(blk_add_trace_split);
 910         WARN_ON(ret);
 911         ret = register_trace_block_remap(blk_add_trace_remap);
 912         WARN_ON(ret);
 913         return 0;
 914 }
 915
 916 static void blk_unregister_tracepoints(void)
 917 {
 918         unregister_trace_block_remap(blk_add_trace_remap);
 919         unregister_trace_block_split(blk_add_trace_split);
 920         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 921         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 922         unregister_trace_block_plug(blk_add_trace_plug);
 923         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 924         unregister_trace_block_getrq(blk_add_trace_getrq);
 925         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 926         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 927         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 928         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 929         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 930         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 931         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 932         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 933         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 934         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 935
 936         tracepoint_synchronize_unregister();
 937 }
 938
 939 /*
 940  * struct blk_io_tracer formatting routines
 941  */
 942
 943 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 944 {
 945         int i = 0;
 946
 947         if (t->action & BLK_TC_DISCARD)
 948                 rwbs[i++] = 'D';
 949         else if (t->action & BLK_TC_WRITE)
 950                 rwbs[i++] = 'W';
 951         else if (t->bytes)
 952                 rwbs[i++] = 'R';
 953         else
 954                 rwbs[i++] = 'N';
 955
 956         if (t->action & BLK_TC_AHEAD)
 957                 rwbs[i++] = 'A';
 958         if (t->action & BLK_TC_BARRIER)
 959                 rwbs[i++] = 'B';
 960         if (t->action & BLK_TC_SYNC)
 961                 rwbs[i++] = 'S';
 962         if (t->action & BLK_TC_META)
 963                 rwbs[i++] = 'M';
 964
 965         rwbs[i] = '\0';
 966 }
 967
 968 static inline
 969 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 970 {
 971         return (const struct blk_io_trace *)ent;
 972 }
 973
 974 static inline const void *pdu_start(const struct trace_entry *ent)
 975 {
 976         return te_blk_io_trace(ent) + 1;
 977 }
 978
 979 static inline u32 t_sec(const struct trace_entry *ent)
 980 {
 981         return te_blk_io_trace(ent)->bytes >> 9;
 982 }
 983
 984 static inline unsigned long long t_sector(const struct trace_entry *ent)
 985 {
 986         return te_blk_io_trace(ent)->sector;
 987 }
 988
 989 static inline __u16 t_error(const struct trace_entry *ent)
 990 {
 991         return te_blk_io_trace(ent)->sector;
 992 }
 993
 994 static __u64 get_pdu_int(const struct trace_entry *ent)
 995 {
 996         const __u64 *val = pdu_start(ent);
 997         return be64_to_cpu(*val);
 998 }
 999
1000 static void get_pdu_remap(const struct trace_entry *ent,
1001                           struct blk_io_trace_remap *r)
1002 {
1003         const struct blk_io_trace_remap *__r = pdu_start(ent);
1004         __u64 sector = __r->sector;
1005
1006         r->device = be32_to_cpu(__r->device);
1007         r->device_from = be32_to_cpu(__r->device_from);
1008         r->sector = be64_to_cpu(sector);
1009 }
1010
1011 static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
1012 {
1013         char rwbs[6];
1014         unsigned long long ts  = ns2usecs(iter->ts);
1015         unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
1016         unsigned secs          = (unsigned long)ts;
1017         const struct trace_entry *ent = iter->ent;
1018         const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
1019
1020         fill_rwbs(rwbs, t);
1021
1022         return trace_seq_printf(&iter->seq,
1023                                 "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
1024                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1025                                 secs, usec_rem, ent->pid, act, rwbs);
1026 }
1027
1028 static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
1029                               const char *act)
1030 {
1031         char rwbs[6];
1032         fill_rwbs(rwbs, t);
1033         return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
1034                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1035 }
1036
1037 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1038 {
1039         const char *cmd = trace_find_cmdline(ent->pid);
1040
1041         if (t_sec(ent))
1042                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1043                                         t_sector(ent), t_sec(ent), cmd);
1044         return trace_seq_printf(s, "[%s]\n", cmd);
1045 }
1046
1047 static int blk_log_with_error(struct trace_seq *s,
1048                               const struct trace_entry *ent)
1049 {
1050         if (t_sec(ent))
1051                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1052                                         t_sec(ent), t_error(ent));
1053         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1054 }
1055
1056 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1057 {
1058         struct blk_io_trace_remap r = { .device = 0, };
1059
1060         get_pdu_remap(ent, &r);
1061         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1062                                t_sector(ent),
1063                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1064                                (unsigned long long)r.sector);
1065 }
1066
1067 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1068 {
1069         return trace_seq_printf(s, "[%s]\n", trace_find_cmdline(ent->pid));
1070 }
1071
1072 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1073 {
1074         return trace_seq_printf(s, "[%s] %llu\n", trace_find_cmdline(ent->pid),
1075                                 get_pdu_int(ent));
1076 }
1077
1078 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1079 {
1080         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1081                                 get_pdu_int(ent), trace_find_cmdline(ent->pid));
1082 }
1083
1084 /*
1085  * struct tracer operations
1086  */
1087
1088 static void blk_tracer_print_header(struct seq_file *m)
1089 {
1090         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1091                 return;
1092         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1093                     "#  |     |     |           |   |   |\n");
1094 }
1095
1096 static void blk_tracer_start(struct trace_array *tr)
1097 {
1098         tracing_reset_online_cpus(tr);
1099
1100         mutex_lock(&blk_probe_mutex);
1101         if (atomic_add_return(1, &blk_probes_ref) == 1)
1102                 if (blk_register_tracepoints())
1103                         atomic_dec(&blk_probes_ref);
1104         mutex_unlock(&blk_probe_mutex);
1105         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1106 }
1107
1108 static int blk_tracer_init(struct trace_array *tr)
1109 {
1110         blk_tr = tr;
1111         blk_tracer_start(tr);
1112         mutex_lock(&blk_probe_mutex);
1113         blk_tracer_enabled++;
1114         mutex_unlock(&blk_probe_mutex);
1115         return 0;
1116 }
1117
1118 static void blk_tracer_stop(struct trace_array *tr)
1119 {
1120         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1121         mutex_lock(&blk_probe_mutex);
1122         if (atomic_dec_and_test(&blk_probes_ref))
1123                 blk_unregister_tracepoints();
1124         mutex_unlock(&blk_probe_mutex);
1125 }
1126
1127 static void blk_tracer_reset(struct trace_array *tr)
1128 {
1129         if (!atomic_read(&blk_probes_ref))
1130                 return;
1131
1132         mutex_lock(&blk_probe_mutex);
1133         blk_tracer_enabled--;
1134         WARN_ON(blk_tracer_enabled < 0);
1135         mutex_unlock(&blk_probe_mutex);
1136
1137         blk_tracer_stop(tr);
1138 }
1139
1140 static struct {
1141         const char *act[2];
1142         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1143 } what2act[] __read_mostly = {
1144         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1145         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1146         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1147         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1148         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1149         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1150         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1151         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1152         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1153         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1154         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1155         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1156         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1157         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1158         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1159 };
1160
1161 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1162                                                int flags)
1163 {
1164         struct trace_seq *s = &iter->seq;
1165         const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1166         const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1167         int ret;
1168
1169         if (!trace_print_context(iter))
1170                 return TRACE_TYPE_PARTIAL_LINE;
1171
1172         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1173                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1174         else {
1175                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1176                 ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
1177                 if (ret)
1178                         ret = what2act[what].print(s, iter->ent);
1179         }
1180
1181         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1182 }
1183
1184 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1185 {
1186         struct trace_seq *s = &iter->seq;
1187         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1188         const int offset = offsetof(struct blk_io_trace, sector);
1189         struct blk_io_trace old = {
1190                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1191                 .time     = ns2usecs(iter->ts),
1192         };
1193
1194         if (!trace_seq_putmem(s, &old, offset))
1195                 return 0;
1196         return trace_seq_putmem(s, &t->sector,
1197                                 sizeof(old) - offset + t->pdu_len);
1198 }
1199
1200 static enum print_line_t
1201 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1202 {
1203         return blk_trace_synthesize_old_trace(iter) ?
1204                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1205 }
1206
1207 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1208 {
1209         const struct blk_io_trace *t;
1210         u16 what;
1211         int ret;
1212
1213         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1214                 return TRACE_TYPE_UNHANDLED;
1215
1216         t = (const struct blk_io_trace *)iter->ent;
1217         what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1218
1219         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1220                 ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
1221         else {
1222                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1223                 ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
1224                 if (ret)
1225                         ret = what2act[what].print(&iter->seq, iter->ent);
1226         }
1227
1228         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1229 }
1230
1231 static struct tracer blk_tracer __read_mostly = {
1232         .name           = "blk",
1233         .init           = blk_tracer_init,
1234         .reset          = blk_tracer_reset,
1235         .start          = blk_tracer_start,
1236         .stop           = blk_tracer_stop,
1237         .print_header   = blk_tracer_print_header,
1238         .print_line     = blk_tracer_print_line,
1239         .flags          = &blk_tracer_flags,
1240 };
1241
1242 static struct trace_event trace_blk_event = {
1243         .type           = TRACE_BLK,
1244         .trace          = blk_trace_event_print,
1245         .latency_trace  = blk_trace_event_print,
1246         .binary         = blk_trace_event_print_binary,
1247 };
1248
1249 static int __init init_blk_tracer(void)
1250 {
1251         if (!register_ftrace_event(&trace_blk_event)) {
1252                 pr_warning("Warning: could not register block events\n");
1253                 return 1;
1254         }
1255
1256         if (register_tracer(&blk_tracer) != 0) {
1257                 pr_warning("Warning: could not register the block tracer\n");
1258                 unregister_ftrace_event(&trace_blk_event);
1259                 return 1;
1260         }
1261
1262         return 0;
1263 }
1264
1265 device_initcall(init_blk_tracer);
1266
1267 static int blk_trace_remove_queue(struct request_queue *q)
1268 {
1269         struct blk_trace *bt;
1270
1271         bt = xchg(&q->blk_trace, NULL);
1272         if (bt == NULL)
1273                 return -EINVAL;
1274
1275         kfree(bt);
1276         return 0;
1277 }
1278
1279 /*
1280  * Setup everything required to start tracing
1281  */
1282 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1283 {
1284         struct blk_trace *old_bt, *bt = NULL;
1285         int ret;
1286
1287         ret = -ENOMEM;
1288         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1289         if (!bt)
1290                 goto err;
1291
1292         bt->dev = dev;
1293         bt->act_mask = (u16)-1;
1294         bt->end_lba = -1ULL;
1295         bt->trace_state = Blktrace_running;
1296
1297         old_bt = xchg(&q->blk_trace, bt);
1298         if (old_bt != NULL) {
1299                 (void)xchg(&q->blk_trace, old_bt);
1300                 kfree(bt);
1301                 ret = -EBUSY;
1302         }
1303         return 0;
1304 err:
1305         return ret;
1306 }
1307
1308 /*
1309  * sysfs interface to enable and configure tracing
1310  */
1311
1312 static ssize_t sysfs_blk_trace_enable_show(struct device *dev,
1313                                            struct device_attribute *attr,
1314                                            char *buf)
1315 {
1316         struct hd_struct *p = dev_to_part(dev);
1317         struct block_device *bdev;
1318         ssize_t ret = -ENXIO;
1319
1320         lock_kernel();
1321         bdev = bdget(part_devt(p));
1322         if (bdev != NULL) {
1323                 struct request_queue *q = bdev_get_queue(bdev);
1324
1325                 if (q != NULL) {
1326                         mutex_lock(&bdev->bd_mutex);
1327                         ret = sprintf(buf, "%u\n", !!q->blk_trace);
1328                         mutex_unlock(&bdev->bd_mutex);
1329                 }
1330
1331                 bdput(bdev);
1332         }
1333
1334         unlock_kernel();
1335         return ret;
1336 }
1337
1338 static ssize_t sysfs_blk_trace_enable_store(struct device *dev,
1339                                             struct device_attribute *attr,
1340                                             const char *buf, size_t count)
1341 {
1342         struct block_device *bdev;
1343         struct request_queue *q;
1344         struct hd_struct *p;
1345         int value;
1346         ssize_t ret = -ENXIO;
1347
1348         if (count == 0 || sscanf(buf, "%d", &value) != 1)
1349                 goto out;
1350
1351         lock_kernel();
1352         p = dev_to_part(dev);
1353         bdev = bdget(part_devt(p));
1354         if (bdev == NULL)
1355                 goto out_unlock_kernel;
1356
1357         q = bdev_get_queue(bdev);
1358         if (q == NULL)
1359                 goto out_bdput;
1360
1361         mutex_lock(&bdev->bd_mutex);
1362         if (value)
1363                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1364         else
1365                 ret = blk_trace_remove_queue(q);
1366         mutex_unlock(&bdev->bd_mutex);
1367
1368         if (ret == 0)
1369                 ret = count;
1370 out_bdput:
1371         bdput(bdev);
1372 out_unlock_kernel:
1373         unlock_kernel();
1374 out:
1375         return ret;
1376 }
1377
1378 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1379                                          struct device_attribute *attr,
1380                                          char *buf);
1381 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1382                                           struct device_attribute *attr,
1383                                           const char *buf, size_t count);
1384 #define BLK_TRACE_DEVICE_ATTR(_name) \
1385         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1386                     sysfs_blk_trace_attr_show, \
1387                     sysfs_blk_trace_attr_store)
1388
1389 static DEVICE_ATTR(enable, S_IRUGO | S_IWUSR,
1390                    sysfs_blk_trace_enable_show, sysfs_blk_trace_enable_store);
1391 static BLK_TRACE_DEVICE_ATTR(act_mask);
1392 static BLK_TRACE_DEVICE_ATTR(pid);
1393 static BLK_TRACE_DEVICE_ATTR(start_lba);
1394 static BLK_TRACE_DEVICE_ATTR(end_lba);
1395
1396 static struct attribute *blk_trace_attrs[] = {
1397         &dev_attr_enable.attr,
1398         &dev_attr_act_mask.attr,
1399         &dev_attr_pid.attr,
1400         &dev_attr_start_lba.attr,
1401         &dev_attr_end_lba.attr,
1402         NULL
1403 };
1404
1405 struct attribute_group blk_trace_attr_group = {
1406         .name  = "trace",
1407         .attrs = blk_trace_attrs,
1408 };
1409
1410 static int blk_str2act_mask(const char *str)
1411 {
1412         int mask = 0;
1413         char *copy = kstrdup(str, GFP_KERNEL), *s;
1414
1415         if (copy == NULL)
1416                 return -ENOMEM;
1417
1418         s = strstrip(copy);
1419
1420         while (1) {
1421                 char *sep = strchr(s, ',');
1422
1423                 if (sep != NULL)
1424                         *sep = '\0';
1425
1426                 if (strcasecmp(s, "barrier") == 0)
1427                         mask |= BLK_TC_BARRIER;
1428                 else if (strcasecmp(s, "complete") == 0)
1429                         mask |= BLK_TC_COMPLETE;
1430                 else if (strcasecmp(s, "fs") == 0)
1431                         mask |= BLK_TC_FS;
1432                 else if (strcasecmp(s, "issue") == 0)
1433                         mask |= BLK_TC_ISSUE;
1434                 else if (strcasecmp(s, "pc") == 0)
1435                         mask |= BLK_TC_PC;
1436                 else if (strcasecmp(s, "queue") == 0)
1437                         mask |= BLK_TC_QUEUE;
1438                 else if (strcasecmp(s, "read") == 0)
1439                         mask |= BLK_TC_READ;
1440                 else if (strcasecmp(s, "requeue") == 0)
1441                         mask |= BLK_TC_REQUEUE;
1442                 else if (strcasecmp(s, "sync") == 0)
1443                         mask |= BLK_TC_SYNC;
1444                 else if (strcasecmp(s, "write") == 0)
1445                         mask |= BLK_TC_WRITE;
1446
1447                 if (sep == NULL)
1448                         break;
1449
1450                 s = sep + 1;
1451         }
1452         kfree(copy);
1453
1454         return mask;
1455 }
1456
1457 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1458                                          struct device_attribute *attr,
1459                                          char *buf)
1460 {
1461         struct hd_struct *p = dev_to_part(dev);
1462         struct request_queue *q;
1463         struct block_device *bdev;
1464         ssize_t ret = -ENXIO;
1465
1466         lock_kernel();
1467         bdev = bdget(part_devt(p));
1468         if (bdev == NULL)
1469                 goto out_unlock_kernel;
1470
1471         q = bdev_get_queue(bdev);
1472         if (q == NULL)
1473                 goto out_bdput;
1474         mutex_lock(&bdev->bd_mutex);
1475         if (q->blk_trace == NULL)
1476                 ret = sprintf(buf, "disabled\n");
1477         else if (attr == &dev_attr_act_mask)
1478                 ret = sprintf(buf, "%#x\n", q->blk_trace->act_mask);
1479         else if (attr == &dev_attr_pid)
1480                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1481         else if (attr == &dev_attr_start_lba)
1482                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1483         else if (attr == &dev_attr_end_lba)
1484                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1485         mutex_unlock(&bdev->bd_mutex);
1486 out_bdput:
1487         bdput(bdev);
1488 out_unlock_kernel:
1489         unlock_kernel();
1490         return ret;
1491 }
1492
1493 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1494                                           struct device_attribute *attr,
1495                                           const char *buf, size_t count)
1496 {
1497         struct block_device *bdev;
1498         struct request_queue *q;
1499         struct hd_struct *p;
1500         u64 value;
1501         ssize_t ret = -ENXIO;
1502
1503         if (count == 0)
1504                 goto out;
1505
1506         if (attr == &dev_attr_act_mask) {
1507                 if (sscanf(buf, "%llx", &value) != 1) {
1508                         /* Assume it is a list of trace category names */
1509                         value = blk_str2act_mask(buf);
1510                         if (value < 0)
1511                                 goto out;
1512                 }
1513         } else if (sscanf(buf, "%llu", &value) != 1)
1514                 goto out;
1515
1516         lock_kernel();
1517         p = dev_to_part(dev);
1518         bdev = bdget(part_devt(p));
1519         if (bdev == NULL)
1520                 goto out_unlock_kernel;
1521
1522         q = bdev_get_queue(bdev);
1523         if (q == NULL)
1524                 goto out_bdput;
1525
1526         mutex_lock(&bdev->bd_mutex);
1527         ret = 0;
1528         if (q->blk_trace == NULL)
1529                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1530
1531         if (ret == 0) {
1532                 if (attr == &dev_attr_act_mask)
1533                         q->blk_trace->act_mask = value;
1534                 else if (attr == &dev_attr_pid)
1535                         q->blk_trace->pid = value;
1536                 else if (attr == &dev_attr_start_lba)
1537                         q->blk_trace->start_lba = value;
1538                 else if (attr == &dev_attr_end_lba)
1539                         q->blk_trace->end_lba = value;
1540                 ret = count;
1541         }
1542         mutex_unlock(&bdev->bd_mutex);
1543 out_bdput:
1544         bdput(bdev);
1545 out_unlock_kernel:
1546         unlock_kernel();
1547 out:
1548         return ret;
1549 }