kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/time.h>
  26 #include <trace/block.h>
  27 #include <linux/uaccess.h>
  28 #include "trace_output.h"
  29
  30 static unsigned int blktrace_seq __read_mostly = 1;
  31
  32 static struct trace_array *blk_tr;
  33 static bool blk_tracer_enabled __read_mostly;
  34
  35 /* Select an alternative, minimalistic output than the original one */
  36 #define TRACE_BLK_OPT_CLASSIC   0x1
  37
  38 static struct tracer_opt blk_tracer_opts[] = {
  39         /* Default disable the minimalistic output */
  40         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  41         { }
  42 };
  43
  44 static struct tracer_flags blk_tracer_flags = {
  45         .val  = 0,
  46         .opts = blk_tracer_opts,
  47 };
  48
  49 /* Global reference count of probes */
  50 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  51
  52 static void blk_register_tracepoints(void);
  53 static void blk_unregister_tracepoints(void);
  54
  55 /*
  56  * Send out a notify message.
  57  */
  58 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  59                        const void *data, size_t len)
  60 {
  61         struct blk_io_trace *t;
  62
  63         if (!bt->rchan)
  64                 return;
  65
  66         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  67         if (t) {
  68                 const int cpu = smp_processor_id();
  69
  70                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  71                 t->time = ktime_to_ns(ktime_get());
  72                 t->device = bt->dev;
  73                 t->action = action;
  74                 t->pid = pid;
  75                 t->cpu = cpu;
  76                 t->pdu_len = len;
  77                 memcpy((void *) t + sizeof(*t), data, len);
  78         }
  79 }
  80
  81 /*
  82  * Send out a notify for this process, if we haven't done so since a trace
  83  * started
  84  */
  85 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
  86 {
  87         tsk->btrace_seq = blktrace_seq;
  88         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
  89 }
  90
  91 static void trace_note_time(struct blk_trace *bt)
  92 {
  93         struct timespec now;
  94         unsigned long flags;
  95         u32 words[2];
  96
  97         getnstimeofday(&now);
  98         words[0] = now.tv_sec;
  99         words[1] = now.tv_nsec;
 100
 101         local_irq_save(flags);
 102         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 103         local_irq_restore(flags);
 104 }
 105
 106 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 107 {
 108         int n;
 109         va_list args;
 110         unsigned long flags;
 111         char *buf;
 112
 113         if (blk_tr) {
 114                 va_start(args, fmt);
 115                 ftrace_vprintk(fmt, args);
 116                 va_end(args);
 117                 return;
 118         }
 119
 120         if (!bt->msg_data)
 121                 return;
 122
 123         local_irq_save(flags);
 124         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 125         va_start(args, fmt);
 126         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 127         va_end(args);
 128
 129         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 130         local_irq_restore(flags);
 131 }
 132 EXPORT_SYMBOL_GPL(__trace_note_message);
 133
 134 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 135                          pid_t pid)
 136 {
 137         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 138                 return 1;
 139         if (sector < bt->start_lba || sector > bt->end_lba)
 140                 return 1;
 141         if (bt->pid && pid != bt->pid)
 142                 return 1;
 143
 144         return 0;
 145 }
 146
 147 /*
 148  * Data direction bit lookup
 149  */
 150 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 151                                  BLK_TC_ACT(BLK_TC_WRITE) };
 152
 153 /* The ilog2() calls fall out because they're constant */
 154 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 155           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 156
 157 /*
 158  * The worker for the various blk_add_trace*() types. Fills out a
 159  * blk_io_trace structure and places it in a per-cpu subbuffer.
 160  */
 161 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 162                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 163 {
 164         struct task_struct *tsk = current;
 165         struct ring_buffer_event *event = NULL;
 166         struct blk_io_trace *t;
 167         unsigned long flags = 0;
 168         unsigned long *sequence;
 169         pid_t pid;
 170         int cpu, pc = 0;
 171
 172         if (unlikely(bt->trace_state != Blktrace_running ||
 173                      !blk_tracer_enabled))
 174                 return;
 175
 176         what |= ddir_act[rw & WRITE];
 177         what |= MASK_TC_BIT(rw, BARRIER);
 178         what |= MASK_TC_BIT(rw, SYNCIO);
 179         what |= MASK_TC_BIT(rw, AHEAD);
 180         what |= MASK_TC_BIT(rw, META);
 181         what |= MASK_TC_BIT(rw, DISCARD);
 182
 183         pid = tsk->pid;
 184         if (unlikely(act_log_check(bt, what, sector, pid)))
 185                 return;
 186         cpu = raw_smp_processor_id();
 187
 188         if (blk_tr) {
 189                 tracing_record_cmdline(current);
 190
 191                 pc = preempt_count();
 192                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 193                                                   sizeof(*t) + pdu_len,
 194                                                   0, pc);
 195                 if (!event)
 196                         return;
 197                 t = ring_buffer_event_data(event);
 198                 goto record_it;
 199         }
 200
 201         /*
 202          * A word about the locking here - we disable interrupts to reserve
 203          * some space in the relay per-cpu buffer, to prevent an irq
 204          * from coming in and stepping on our toes.
 205          */
 206         local_irq_save(flags);
 207
 208         if (unlikely(tsk->btrace_seq != blktrace_seq))
 209                 trace_note_tsk(bt, tsk);
 210
 211         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 212         if (t) {
 213                 sequence = per_cpu_ptr(bt->sequence, cpu);
 214
 215                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 216                 t->sequence = ++(*sequence);
 217                 t->time = ktime_to_ns(ktime_get());
 218 record_it:
 219                 /*
 220                  * These two are not needed in ftrace as they are in the
 221                  * generic trace_entry, filled by tracing_generic_entry_update,
 222                  * but for the trace_event->bin() synthesizer benefit we do it
 223                  * here too.
 224                  */
 225                 t->cpu = cpu;
 226                 t->pid = pid;
 227
 228                 t->sector = sector;
 229                 t->bytes = bytes;
 230                 t->action = what;
 231                 t->device = bt->dev;
 232                 t->error = error;
 233                 t->pdu_len = pdu_len;
 234
 235                 if (pdu_len)
 236                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 237
 238                 if (blk_tr) {
 239                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 240                         return;
 241                 }
 242         }
 243
 244         local_irq_restore(flags);
 245 }
 246
 247 static struct dentry *blk_tree_root;
 248 static DEFINE_MUTEX(blk_tree_mutex);
 249
 250 static void blk_trace_cleanup(struct blk_trace *bt)
 251 {
 252         debugfs_remove(bt->msg_file);
 253         debugfs_remove(bt->dropped_file);
 254         relay_close(bt->rchan);
 255         free_percpu(bt->sequence);
 256         free_percpu(bt->msg_data);
 257         kfree(bt);
 258         if (atomic_dec_and_test(&blk_probes_ref))
 259                 blk_unregister_tracepoints();
 260 }
 261
 262 int blk_trace_remove(struct request_queue *q)
 263 {
 264         struct blk_trace *bt;
 265
 266         bt = xchg(&q->blk_trace, NULL);
 267         if (!bt)
 268                 return -EINVAL;
 269
 270         if (bt->trace_state == Blktrace_setup ||
 271             bt->trace_state == Blktrace_stopped)
 272                 blk_trace_cleanup(bt);
 273
 274         return 0;
 275 }
 276 EXPORT_SYMBOL_GPL(blk_trace_remove);
 277
 278 static int blk_dropped_open(struct inode *inode, struct file *filp)
 279 {
 280         filp->private_data = inode->i_private;
 281
 282         return 0;
 283 }
 284
 285 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 286                                 size_t count, loff_t *ppos)
 287 {
 288         struct blk_trace *bt = filp->private_data;
 289         char buf[16];
 290
 291         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 292
 293         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 294 }
 295
 296 static const struct file_operations blk_dropped_fops = {
 297         .owner =        THIS_MODULE,
 298         .open =         blk_dropped_open,
 299         .read =         blk_dropped_read,
 300 };
 301
 302 static int blk_msg_open(struct inode *inode, struct file *filp)
 303 {
 304         filp->private_data = inode->i_private;
 305
 306         return 0;
 307 }
 308
 309 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 310                                 size_t count, loff_t *ppos)
 311 {
 312         char *msg;
 313         struct blk_trace *bt;
 314
 315         if (count > BLK_TN_MAX_MSG)
 316                 return -EINVAL;
 317
 318         msg = kmalloc(count, GFP_KERNEL);
 319         if (msg == NULL)
 320                 return -ENOMEM;
 321
 322         if (copy_from_user(msg, buffer, count)) {
 323                 kfree(msg);
 324                 return -EFAULT;
 325         }
 326
 327         bt = filp->private_data;
 328         __trace_note_message(bt, "%s", msg);
 329         kfree(msg);
 330
 331         return count;
 332 }
 333
 334 static const struct file_operations blk_msg_fops = {
 335         .owner =        THIS_MODULE,
 336         .open =         blk_msg_open,
 337         .write =        blk_msg_write,
 338 };
 339
 340 /*
 341  * Keep track of how many times we encountered a full subbuffer, to aid
 342  * the user space app in telling how many lost events there were.
 343  */
 344 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 345                                      void *prev_subbuf, size_t prev_padding)
 346 {
 347         struct blk_trace *bt;
 348
 349         if (!relay_buf_full(buf))
 350                 return 1;
 351
 352         bt = buf->chan->private_data;
 353         atomic_inc(&bt->dropped);
 354         return 0;
 355 }
 356
 357 static int blk_remove_buf_file_callback(struct dentry *dentry)
 358 {
 359         struct dentry *parent = dentry->d_parent;
 360         debugfs_remove(dentry);
 361
 362         /*
 363         * this will fail for all but the last file, but that is ok. what we
 364         * care about is the top level buts->name directory going away, when
 365         * the last trace file is gone. Then we don't have to rmdir() that
 366         * manually on trace stop, so it nicely solves the issue with
 367         * force killing of running traces.
 368         */
 369
 370         debugfs_remove(parent);
 371         return 0;
 372 }
 373
 374 static struct dentry *blk_create_buf_file_callback(const char *filename,
 375                                                    struct dentry *parent,
 376                                                    int mode,
 377                                                    struct rchan_buf *buf,
 378                                                    int *is_global)
 379 {
 380         return debugfs_create_file(filename, mode, parent, buf,
 381                                         &relay_file_operations);
 382 }
 383
 384 static struct rchan_callbacks blk_relay_callbacks = {
 385         .subbuf_start           = blk_subbuf_start_callback,
 386         .create_buf_file        = blk_create_buf_file_callback,
 387         .remove_buf_file        = blk_remove_buf_file_callback,
 388 };
 389
 390 /*
 391  * Setup everything required to start tracing
 392  */
 393 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 394                         struct blk_user_trace_setup *buts)
 395 {
 396         struct blk_trace *old_bt, *bt = NULL;
 397         struct dentry *dir = NULL;
 398         int ret, i;
 399
 400         if (!buts->buf_size || !buts->buf_nr)
 401                 return -EINVAL;
 402
 403         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 404         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 405
 406         /*
 407          * some device names have larger paths - convert the slashes
 408          * to underscores for this to work as expected
 409          */
 410         for (i = 0; i < strlen(buts->name); i++)
 411                 if (buts->name[i] == '/')
 412                         buts->name[i] = '_';
 413
 414         ret = -ENOMEM;
 415         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 416         if (!bt)
 417                 goto err;
 418
 419         bt->sequence = alloc_percpu(unsigned long);
 420         if (!bt->sequence)
 421                 goto err;
 422
 423         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 424         if (!bt->msg_data)
 425                 goto err;
 426
 427         ret = -ENOENT;
 428
 429         mutex_lock(&blk_tree_mutex);
 430         if (!blk_tree_root) {
 431                 blk_tree_root = debugfs_create_dir("block", NULL);
 432                 if (!blk_tree_root) {
 433                         mutex_unlock(&blk_tree_mutex);
 434                         goto err;
 435                 }
 436         }
 437         mutex_unlock(&blk_tree_mutex);
 438
 439         dir = debugfs_create_dir(buts->name, blk_tree_root);
 440
 441         if (!dir)
 442                 goto err;
 443
 444         bt->dir = dir;
 445         bt->dev = dev;
 446         atomic_set(&bt->dropped, 0);
 447
 448         ret = -EIO;
 449         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 450                                                &blk_dropped_fops);
 451         if (!bt->dropped_file)
 452                 goto err;
 453
 454         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 455         if (!bt->msg_file)
 456                 goto err;
 457
 458         bt->rchan = relay_open("trace", dir, buts->buf_size,
 459                                 buts->buf_nr, &blk_relay_callbacks, bt);
 460         if (!bt->rchan)
 461                 goto err;
 462
 463         bt->act_mask = buts->act_mask;
 464         if (!bt->act_mask)
 465                 bt->act_mask = (u16) -1;
 466
 467         bt->start_lba = buts->start_lba;
 468         bt->end_lba = buts->end_lba;
 469         if (!bt->end_lba)
 470                 bt->end_lba = -1ULL;
 471
 472         bt->pid = buts->pid;
 473         bt->trace_state = Blktrace_setup;
 474
 475         ret = -EBUSY;
 476         old_bt = xchg(&q->blk_trace, bt);
 477         if (old_bt) {
 478                 (void) xchg(&q->blk_trace, old_bt);
 479                 goto err;
 480         }
 481
 482         if (atomic_add_return(1, &blk_probes_ref) == 1)
 483                 blk_register_tracepoints();
 484
 485         return 0;
 486 err:
 487         if (bt) {
 488                 if (bt->msg_file)
 489                         debugfs_remove(bt->msg_file);
 490                 if (bt->dropped_file)
 491                         debugfs_remove(bt->dropped_file);
 492                 free_percpu(bt->sequence);
 493                 free_percpu(bt->msg_data);
 494                 if (bt->rchan)
 495                         relay_close(bt->rchan);
 496                 kfree(bt);
 497         }
 498         return ret;
 499 }
 500
 501 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 502                     char __user *arg)
 503 {
 504         struct blk_user_trace_setup buts;
 505         int ret;
 506
 507         ret = copy_from_user(&buts, arg, sizeof(buts));
 508         if (ret)
 509                 return -EFAULT;
 510
 511         ret = do_blk_trace_setup(q, name, dev, &buts);
 512         if (ret)
 513                 return ret;
 514
 515         if (copy_to_user(arg, &buts, sizeof(buts)))
 516                 return -EFAULT;
 517
 518         return 0;
 519 }
 520 EXPORT_SYMBOL_GPL(blk_trace_setup);
 521
 522 int blk_trace_startstop(struct request_queue *q, int start)
 523 {
 524         int ret;
 525         struct blk_trace *bt = q->blk_trace;
 526
 527         if (bt == NULL)
 528                 return -EINVAL;
 529
 530         /*
 531          * For starting a trace, we can transition from a setup or stopped
 532          * trace. For stopping a trace, the state must be running
 533          */
 534         ret = -EINVAL;
 535         if (start) {
 536                 if (bt->trace_state == Blktrace_setup ||
 537                     bt->trace_state == Blktrace_stopped) {
 538                         blktrace_seq++;
 539                         smp_mb();
 540                         bt->trace_state = Blktrace_running;
 541
 542                         trace_note_time(bt);
 543                         ret = 0;
 544                 }
 545         } else {
 546                 if (bt->trace_state == Blktrace_running) {
 547                         bt->trace_state = Blktrace_stopped;
 548                         relay_flush(bt->rchan);
 549                         ret = 0;
 550                 }
 551         }
 552
 553         return ret;
 554 }
 555 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 556
 557 /**
 558  * blk_trace_ioctl: - handle the ioctls associated with tracing
 559  * @bdev:       the block device
 560  * @cmd:        the ioctl cmd
 561  * @arg:        the argument data, if any
 562  *
 563  **/
 564 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 565 {
 566         struct request_queue *q;
 567         int ret, start = 0;
 568         char b[BDEVNAME_SIZE];
 569
 570         q = bdev_get_queue(bdev);
 571         if (!q)
 572                 return -ENXIO;
 573
 574         mutex_lock(&bdev->bd_mutex);
 575
 576         switch (cmd) {
 577         case BLKTRACESETUP:
 578                 bdevname(bdev, b);
 579                 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
 580                 break;
 581         case BLKTRACESTART:
 582                 start = 1;
 583         case BLKTRACESTOP:
 584                 ret = blk_trace_startstop(q, start);
 585                 break;
 586         case BLKTRACETEARDOWN:
 587                 ret = blk_trace_remove(q);
 588                 break;
 589         default:
 590                 ret = -ENOTTY;
 591                 break;
 592         }
 593
 594         mutex_unlock(&bdev->bd_mutex);
 595         return ret;
 596 }
 597
 598 /**
 599  * blk_trace_shutdown: - stop and cleanup trace structures
 600  * @q:    the request queue associated with the device
 601  *
 602  **/
 603 void blk_trace_shutdown(struct request_queue *q)
 604 {
 605         if (q->blk_trace) {
 606                 blk_trace_startstop(q, 0);
 607                 blk_trace_remove(q);
 608         }
 609 }
 610
 611 /*
 612  * blktrace probes
 613  */
 614
 615 /**
 616  * blk_add_trace_rq - Add a trace for a request oriented action
 617  * @q:          queue the io is for
 618  * @rq:         the source request
 619  * @what:       the action
 620  *
 621  * Description:
 622  *     Records an action against a request. Will log the bio offset + size.
 623  *
 624  **/
 625 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 626                                     u32 what)
 627 {
 628         struct blk_trace *bt = q->blk_trace;
 629         int rw = rq->cmd_flags & 0x03;
 630
 631         if (likely(!bt))
 632                 return;
 633
 634         if (blk_discard_rq(rq))
 635                 rw |= (1 << BIO_RW_DISCARD);
 636
 637         if (blk_pc_request(rq)) {
 638                 what |= BLK_TC_ACT(BLK_TC_PC);
 639                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 640                                 sizeof(rq->cmd), rq->cmd);
 641         } else  {
 642                 what |= BLK_TC_ACT(BLK_TC_FS);
 643                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 644                                 rw, what, rq->errors, 0, NULL);
 645         }
 646 }
 647
 648 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 649 {
 650         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 651 }
 652
 653 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 654 {
 655         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 656 }
 657
 658 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 659 {
 660         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 661 }
 662
 663 static void blk_add_trace_rq_requeue(struct request_queue *q,
 664                                      struct request *rq)
 665 {
 666         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 667 }
 668
 669 static void blk_add_trace_rq_complete(struct request_queue *q,
 670                                       struct request *rq)
 671 {
 672         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 673 }
 674
 675 /**
 676  * blk_add_trace_bio - Add a trace for a bio oriented action
 677  * @q:          queue the io is for
 678  * @bio:        the source bio
 679  * @what:       the action
 680  *
 681  * Description:
 682  *     Records an action against a bio. Will log the bio offset + size.
 683  *
 684  **/
 685 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 686                                      u32 what)
 687 {
 688         struct blk_trace *bt = q->blk_trace;
 689
 690         if (likely(!bt))
 691                 return;
 692
 693         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 694                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 695 }
 696
 697 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 698 {
 699         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 700 }
 701
 702 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 703 {
 704         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 705 }
 706
 707 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 708                                         struct bio *bio)
 709 {
 710         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 711 }
 712
 713 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 714                                          struct bio *bio)
 715 {
 716         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 717 }
 718
 719 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 720 {
 721         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 722 }
 723
 724 static void blk_add_trace_getrq(struct request_queue *q,
 725                                 struct bio *bio, int rw)
 726 {
 727         if (bio)
 728                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 729         else {
 730                 struct blk_trace *bt = q->blk_trace;
 731
 732                 if (bt)
 733                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 734         }
 735 }
 736
 737
 738 static void blk_add_trace_sleeprq(struct request_queue *q,
 739                                   struct bio *bio, int rw)
 740 {
 741         if (bio)
 742                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 743         else {
 744                 struct blk_trace *bt = q->blk_trace;
 745
 746                 if (bt)
 747                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 748                                         0, 0, NULL);
 749         }
 750 }
 751
 752 static void blk_add_trace_plug(struct request_queue *q)
 753 {
 754         struct blk_trace *bt = q->blk_trace;
 755
 756         if (bt)
 757                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 758 }
 759
 760 static void blk_add_trace_unplug_io(struct request_queue *q)
 761 {
 762         struct blk_trace *bt = q->blk_trace;
 763
 764         if (bt) {
 765                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 766                 __be64 rpdu = cpu_to_be64(pdu);
 767
 768                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
 769                                 sizeof(rpdu), &rpdu);
 770         }
 771 }
 772
 773 static void blk_add_trace_unplug_timer(struct request_queue *q)
 774 {
 775         struct blk_trace *bt = q->blk_trace;
 776
 777         if (bt) {
 778                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 779                 __be64 rpdu = cpu_to_be64(pdu);
 780
 781                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 782                                 sizeof(rpdu), &rpdu);
 783         }
 784 }
 785
 786 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 787                                 unsigned int pdu)
 788 {
 789         struct blk_trace *bt = q->blk_trace;
 790
 791         if (bt) {
 792                 __be64 rpdu = cpu_to_be64(pdu);
 793
 794                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 795                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 796                                 sizeof(rpdu), &rpdu);
 797         }
 798 }
 799
 800 /**
 801  * blk_add_trace_remap - Add a trace for a remap operation
 802  * @q:          queue the io is for
 803  * @bio:        the source bio
 804  * @dev:        target device
 805  * @from:       source sector
 806  * @to:         target sector
 807  *
 808  * Description:
 809  *     Device mapper or raid target sometimes need to split a bio because
 810  *     it spans a stripe (or similar). Add a trace for that action.
 811  *
 812  **/
 813 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 814                                        dev_t dev, sector_t from, sector_t to)
 815 {
 816         struct blk_trace *bt = q->blk_trace;
 817         struct blk_io_trace_remap r;
 818
 819         if (likely(!bt))
 820                 return;
 821
 822         r.device = cpu_to_be32(dev);
 823         r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
 824         r.sector = cpu_to_be64(to);
 825
 826         __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
 827                         !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
 828 }
 829
 830 /**
 831  * blk_add_driver_data - Add binary message with driver-specific data
 832  * @q:          queue the io is for
 833  * @rq:         io request
 834  * @data:       driver-specific data
 835  * @len:        length of driver-specific data
 836  *
 837  * Description:
 838  *     Some drivers might want to write driver-specific data per request.
 839  *
 840  **/
 841 void blk_add_driver_data(struct request_queue *q,
 842                          struct request *rq,
 843                          void *data, size_t len)
 844 {
 845         struct blk_trace *bt = q->blk_trace;
 846
 847         if (likely(!bt))
 848                 return;
 849
 850         if (blk_pc_request(rq))
 851                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 852                                 rq->errors, len, data);
 853         else
 854                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 855                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 856 }
 857 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 858
 859 static void blk_register_tracepoints(void)
 860 {
 861         int ret;
 862
 863         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 864         WARN_ON(ret);
 865         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 866         WARN_ON(ret);
 867         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 868         WARN_ON(ret);
 869         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 870         WARN_ON(ret);
 871         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 872         WARN_ON(ret);
 873         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 874         WARN_ON(ret);
 875         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 876         WARN_ON(ret);
 877         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 878         WARN_ON(ret);
 879         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 880         WARN_ON(ret);
 881         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 882         WARN_ON(ret);
 883         ret = register_trace_block_getrq(blk_add_trace_getrq);
 884         WARN_ON(ret);
 885         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 886         WARN_ON(ret);
 887         ret = register_trace_block_plug(blk_add_trace_plug);
 888         WARN_ON(ret);
 889         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 890         WARN_ON(ret);
 891         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 892         WARN_ON(ret);
 893         ret = register_trace_block_split(blk_add_trace_split);
 894         WARN_ON(ret);
 895         ret = register_trace_block_remap(blk_add_trace_remap);
 896         WARN_ON(ret);
 897 }
 898
 899 static void blk_unregister_tracepoints(void)
 900 {
 901         unregister_trace_block_remap(blk_add_trace_remap);
 902         unregister_trace_block_split(blk_add_trace_split);
 903         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 904         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 905         unregister_trace_block_plug(blk_add_trace_plug);
 906         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 907         unregister_trace_block_getrq(blk_add_trace_getrq);
 908         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 909         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 910         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 911         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 912         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 913         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 914         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 915         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 916         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 917         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 918
 919         tracepoint_synchronize_unregister();
 920 }
 921
 922 /*
 923  * struct blk_io_tracer formatting routines
 924  */
 925
 926 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 927 {
 928         int i = 0;
 929         int tc = t->action >> BLK_TC_SHIFT;
 930
 931         if (tc & BLK_TC_DISCARD)
 932                 rwbs[i++] = 'D';
 933         else if (tc & BLK_TC_WRITE)
 934                 rwbs[i++] = 'W';
 935         else if (t->bytes)
 936                 rwbs[i++] = 'R';
 937         else
 938                 rwbs[i++] = 'N';
 939
 940         if (tc & BLK_TC_AHEAD)
 941                 rwbs[i++] = 'A';
 942         if (tc & BLK_TC_BARRIER)
 943                 rwbs[i++] = 'B';
 944         if (tc & BLK_TC_SYNC)
 945                 rwbs[i++] = 'S';
 946         if (tc & BLK_TC_META)
 947                 rwbs[i++] = 'M';
 948
 949         rwbs[i] = '\0';
 950 }
 951
 952 static inline
 953 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 954 {
 955         return (const struct blk_io_trace *)ent;
 956 }
 957
 958 static inline const void *pdu_start(const struct trace_entry *ent)
 959 {
 960         return te_blk_io_trace(ent) + 1;
 961 }
 962
 963 static inline u32 t_sec(const struct trace_entry *ent)
 964 {
 965         return te_blk_io_trace(ent)->bytes >> 9;
 966 }
 967
 968 static inline unsigned long long t_sector(const struct trace_entry *ent)
 969 {
 970         return te_blk_io_trace(ent)->sector;
 971 }
 972
 973 static inline __u16 t_error(const struct trace_entry *ent)
 974 {
 975         return te_blk_io_trace(ent)->error;
 976 }
 977
 978 static __u64 get_pdu_int(const struct trace_entry *ent)
 979 {
 980         const __u64 *val = pdu_start(ent);
 981         return be64_to_cpu(*val);
 982 }
 983
 984 static void get_pdu_remap(const struct trace_entry *ent,
 985                           struct blk_io_trace_remap *r)
 986 {
 987         const struct blk_io_trace_remap *__r = pdu_start(ent);
 988         __u64 sector = __r->sector;
 989
 990         r->device = be32_to_cpu(__r->device);
 991         r->device_from = be32_to_cpu(__r->device_from);
 992         r->sector = be64_to_cpu(sector);
 993 }
 994
 995 static int blk_log_action_iter(struct trace_iterator *iter, const char *act)
 996 {
 997         char rwbs[6];
 998         unsigned long long ts  = ns2usecs(iter->ts);
 999         unsigned long usec_rem = do_div(ts, USEC_PER_SEC);
1000         unsigned secs          = (unsigned long)ts;
1001         const struct trace_entry *ent = iter->ent;
1002         const struct blk_io_trace *t = (const struct blk_io_trace *)ent;
1003
1004         fill_rwbs(rwbs, t);
1005
1006         return trace_seq_printf(&iter->seq,
1007                                 "%3d,%-3d %2d %5d.%06lu %5u %2s %3s ",
1008                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1009                                 secs, usec_rem, ent->pid, act, rwbs);
1010 }
1011
1012 static int blk_log_action_seq(struct trace_seq *s, const struct blk_io_trace *t,
1013                               const char *act)
1014 {
1015         char rwbs[6];
1016         fill_rwbs(rwbs, t);
1017         return trace_seq_printf(s, "%3d,%-3d %2s %3s ",
1018                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1019 }
1020
1021 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1022 {
1023         char cmd[TASK_COMM_LEN];
1024
1025         trace_find_cmdline(ent->pid, cmd);
1026
1027         if (t_sec(ent))
1028                 return trace_seq_printf(s, "%llu + %u [%s]\n",
1029                                         t_sector(ent), t_sec(ent), cmd);
1030         return trace_seq_printf(s, "[%s]\n", cmd);
1031 }
1032
1033 static int blk_log_with_error(struct trace_seq *s,
1034                               const struct trace_entry *ent)
1035 {
1036         if (t_sec(ent))
1037                 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1038                                         t_sec(ent), t_error(ent));
1039         return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1040 }
1041
1042 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1043 {
1044         struct blk_io_trace_remap r = { .device = 0, };
1045
1046         get_pdu_remap(ent, &r);
1047         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1048                                t_sector(ent),
1049                                t_sec(ent), MAJOR(r.device), MINOR(r.device),
1050                                (unsigned long long)r.sector);
1051 }
1052
1053 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1054 {
1055         char cmd[TASK_COMM_LEN];
1056
1057         trace_find_cmdline(ent->pid, cmd);
1058
1059         return trace_seq_printf(s, "[%s]\n", cmd);
1060 }
1061
1062 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1063 {
1064         char cmd[TASK_COMM_LEN];
1065
1066         trace_find_cmdline(ent->pid, cmd);
1067
1068         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1069 }
1070
1071 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1072 {
1073         char cmd[TASK_COMM_LEN];
1074
1075         trace_find_cmdline(ent->pid, cmd);
1076
1077         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1078                                 get_pdu_int(ent), cmd);
1079 }
1080
1081 /*
1082  * struct tracer operations
1083  */
1084
1085 static void blk_tracer_print_header(struct seq_file *m)
1086 {
1087         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1088                 return;
1089         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1090                     "#  |     |     |           |   |   |\n");
1091 }
1092
1093 static void blk_tracer_start(struct trace_array *tr)
1094 {
1095         if (atomic_add_return(1, &blk_probes_ref) == 1)
1096                 blk_register_tracepoints();
1097         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1098 }
1099
1100 static int blk_tracer_init(struct trace_array *tr)
1101 {
1102         blk_tr = tr;
1103         blk_tracer_start(tr);
1104         blk_tracer_enabled = true;
1105         return 0;
1106 }
1107
1108 static void blk_tracer_stop(struct trace_array *tr)
1109 {
1110         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1111         if (atomic_dec_and_test(&blk_probes_ref))
1112                 blk_unregister_tracepoints();
1113 }
1114
1115 static void blk_tracer_reset(struct trace_array *tr)
1116 {
1117         if (!atomic_read(&blk_probes_ref))
1118                 return;
1119
1120         blk_tracer_enabled = false;
1121         blk_tracer_stop(tr);
1122 }
1123
1124 static const struct {
1125         const char *act[2];
1126         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1127 } what2act[] = {
1128         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1129         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1130         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1131         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1132         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1133         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1134         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1135         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1136         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1137         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1138         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1139         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1140         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1141         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1142         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1143 };
1144
1145 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1146                                                int flags)
1147 {
1148         struct trace_seq *s = &iter->seq;
1149         const struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1150         const u16 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1151         int ret;
1152
1153         if (!trace_print_context(iter))
1154                 return TRACE_TYPE_PARTIAL_LINE;
1155
1156         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1157                 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1158         else {
1159                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1160                 ret = blk_log_action_seq(s, t, what2act[what].act[long_act]);
1161                 if (ret)
1162                         ret = what2act[what].print(s, iter->ent);
1163         }
1164
1165         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1166 }
1167
1168 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1169 {
1170         struct trace_seq *s = &iter->seq;
1171         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1172         const int offset = offsetof(struct blk_io_trace, sector);
1173         struct blk_io_trace old = {
1174                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1175                 .time     = iter->ts,
1176         };
1177
1178         if (!trace_seq_putmem(s, &old, offset))
1179                 return 0;
1180         return trace_seq_putmem(s, &t->sector,
1181                                 sizeof(old) - offset + t->pdu_len);
1182 }
1183
1184 static enum print_line_t
1185 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1186 {
1187         return blk_trace_synthesize_old_trace(iter) ?
1188                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1189 }
1190
1191 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1192 {
1193         const struct blk_io_trace *t;
1194         u16 what;
1195         int ret;
1196
1197         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1198                 return TRACE_TYPE_UNHANDLED;
1199
1200         t = (const struct blk_io_trace *)iter->ent;
1201         what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1202
1203         if (unlikely(what == 0 || what > ARRAY_SIZE(what2act)))
1204                 ret = trace_seq_printf(&iter->seq, "Bad pc action %x\n", what);
1205         else {
1206                 const bool long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1207                 ret = blk_log_action_iter(iter, what2act[what].act[long_act]);
1208                 if (ret)
1209                         ret = what2act[what].print(&iter->seq, iter->ent);
1210         }
1211
1212         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1213 }
1214
1215 static struct tracer blk_tracer __read_mostly = {
1216         .name           = "blk",
1217         .init           = blk_tracer_init,
1218         .reset          = blk_tracer_reset,
1219         .start          = blk_tracer_start,
1220         .stop           = blk_tracer_stop,
1221         .print_header   = blk_tracer_print_header,
1222         .print_line     = blk_tracer_print_line,
1223         .flags          = &blk_tracer_flags,
1224 };
1225
1226 static struct trace_event trace_blk_event = {
1227         .type           = TRACE_BLK,
1228         .trace          = blk_trace_event_print,
1229         .binary         = blk_trace_event_print_binary,
1230 };
1231
1232 static int __init init_blk_tracer(void)
1233 {
1234         if (!register_ftrace_event(&trace_blk_event)) {
1235                 pr_warning("Warning: could not register block events\n");
1236                 return 1;
1237         }
1238
1239         if (register_tracer(&blk_tracer) != 0) {
1240                 pr_warning("Warning: could not register the block tracer\n");
1241                 unregister_ftrace_event(&trace_blk_event);
1242                 return 1;
1243         }
1244
1245         return 0;
1246 }
1247
1248 device_initcall(init_blk_tracer);
1249
1250 static int blk_trace_remove_queue(struct request_queue *q)
1251 {
1252         struct blk_trace *bt;
1253
1254         bt = xchg(&q->blk_trace, NULL);
1255         if (bt == NULL)
1256                 return -EINVAL;
1257
1258         kfree(bt);
1259         return 0;
1260 }
1261
1262 /*
1263  * Setup everything required to start tracing
1264  */
1265 static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1266 {
1267         struct blk_trace *old_bt, *bt = NULL;
1268
1269         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1270         if (!bt)
1271                 return -ENOMEM;
1272
1273         bt->dev = dev;
1274         bt->act_mask = (u16)-1;
1275         bt->end_lba = -1ULL;
1276         bt->trace_state = Blktrace_running;
1277
1278         old_bt = xchg(&q->blk_trace, bt);
1279         if (old_bt != NULL) {
1280                 (void)xchg(&q->blk_trace, old_bt);
1281                 kfree(bt);
1282                 return -EBUSY;
1283         }
1284
1285         return 0;
1286 }
1287
1288 /*
1289  * sysfs interface to enable and configure tracing
1290  */
1291
1292 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1293                                          struct device_attribute *attr,
1294                                          char *buf);
1295 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1296                                           struct device_attribute *attr,
1297                                           const char *buf, size_t count);
1298 #define BLK_TRACE_DEVICE_ATTR(_name) \
1299         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1300                     sysfs_blk_trace_attr_show, \
1301                     sysfs_blk_trace_attr_store)
1302
1303 static BLK_TRACE_DEVICE_ATTR(enable);
1304 static BLK_TRACE_DEVICE_ATTR(act_mask);
1305 static BLK_TRACE_DEVICE_ATTR(pid);
1306 static BLK_TRACE_DEVICE_ATTR(start_lba);
1307 static BLK_TRACE_DEVICE_ATTR(end_lba);
1308
1309 static struct attribute *blk_trace_attrs[] = {
1310         &dev_attr_enable.attr,
1311         &dev_attr_act_mask.attr,
1312         &dev_attr_pid.attr,
1313         &dev_attr_start_lba.attr,
1314         &dev_attr_end_lba.attr,
1315         NULL
1316 };
1317
1318 struct attribute_group blk_trace_attr_group = {
1319         .name  = "trace",
1320         .attrs = blk_trace_attrs,
1321 };
1322
1323 static const struct {
1324         int mask;
1325         const char *str;
1326 } mask_maps[] = {
1327         { BLK_TC_READ,          "read"          },
1328         { BLK_TC_WRITE,         "write"         },
1329         { BLK_TC_BARRIER,       "barrier"       },
1330         { BLK_TC_SYNC,          "sync"          },
1331         { BLK_TC_QUEUE,         "queue"         },
1332         { BLK_TC_REQUEUE,       "requeue"       },
1333         { BLK_TC_ISSUE,         "issue"         },
1334         { BLK_TC_COMPLETE,      "complete"      },
1335         { BLK_TC_FS,            "fs"            },
1336         { BLK_TC_PC,            "pc"            },
1337         { BLK_TC_AHEAD,         "ahead"         },
1338         { BLK_TC_META,          "meta"          },
1339         { BLK_TC_DISCARD,       "discard"       },
1340         { BLK_TC_DRV_DATA,      "drv_data"      },
1341 };
1342
1343 static int blk_trace_str2mask(const char *str)
1344 {
1345         int i;
1346         int mask = 0;
1347         char *s, *token;
1348
1349         s = kstrdup(str, GFP_KERNEL);
1350         if (s == NULL)
1351                 return -ENOMEM;
1352         s = strstrip(s);
1353
1354         while (1) {
1355                 token = strsep(&s, ",");
1356                 if (token == NULL)
1357                         break;
1358
1359                 if (*token == '\0')
1360                         continue;
1361
1362                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1363                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1364                                 mask |= mask_maps[i].mask;
1365                                 break;
1366                         }
1367                 }
1368                 if (i == ARRAY_SIZE(mask_maps)) {
1369                         mask = -EINVAL;
1370                         break;
1371                 }
1372         }
1373         kfree(s);
1374
1375         return mask;
1376 }
1377
1378 static ssize_t blk_trace_mask2str(char *buf, int mask)
1379 {
1380         int i;
1381         char *p = buf;
1382
1383         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1384                 if (mask & mask_maps[i].mask) {
1385                         p += sprintf(p, "%s%s",
1386                                     (p == buf) ? "" : ",", mask_maps[i].str);
1387                 }
1388         }
1389         *p++ = '\n';
1390
1391         return p - buf;
1392 }
1393
1394 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1395 {
1396         if (bdev->bd_disk == NULL)
1397                 return NULL;
1398
1399         return bdev_get_queue(bdev);
1400 }
1401
1402 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1403                                          struct device_attribute *attr,
1404                                          char *buf)
1405 {
1406         struct hd_struct *p = dev_to_part(dev);
1407         struct request_queue *q;
1408         struct block_device *bdev;
1409         ssize_t ret = -ENXIO;
1410
1411         lock_kernel();
1412         bdev = bdget(part_devt(p));
1413         if (bdev == NULL)
1414                 goto out_unlock_kernel;
1415
1416         q = blk_trace_get_queue(bdev);
1417         if (q == NULL)
1418                 goto out_bdput;
1419
1420         mutex_lock(&bdev->bd_mutex);
1421
1422         if (attr == &dev_attr_enable) {
1423                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1424                 goto out_unlock_bdev;
1425         }
1426
1427         if (q->blk_trace == NULL)
1428                 ret = sprintf(buf, "disabled\n");
1429         else if (attr == &dev_attr_act_mask)
1430                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1431         else if (attr == &dev_attr_pid)
1432                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1433         else if (attr == &dev_attr_start_lba)
1434                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1435         else if (attr == &dev_attr_end_lba)
1436                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1437
1438 out_unlock_bdev:
1439         mutex_unlock(&bdev->bd_mutex);
1440 out_bdput:
1441         bdput(bdev);
1442 out_unlock_kernel:
1443         unlock_kernel();
1444         return ret;
1445 }
1446
1447 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1448                                           struct device_attribute *attr,
1449                                           const char *buf, size_t count)
1450 {
1451         struct block_device *bdev;
1452         struct request_queue *q;
1453         struct hd_struct *p;
1454         u64 value;
1455         ssize_t ret = -EINVAL;
1456
1457         if (count == 0)
1458                 goto out;
1459
1460         if (attr == &dev_attr_act_mask) {
1461                 if (sscanf(buf, "%llx", &value) != 1) {
1462                         /* Assume it is a list of trace category names */
1463                         ret = blk_trace_str2mask(buf);
1464                         if (ret < 0)
1465                                 goto out;
1466                         value = ret;
1467                 }
1468         } else if (sscanf(buf, "%llu", &value) != 1)
1469                 goto out;
1470
1471         ret = -ENXIO;
1472
1473         lock_kernel();
1474         p = dev_to_part(dev);
1475         bdev = bdget(part_devt(p));
1476         if (bdev == NULL)
1477                 goto out_unlock_kernel;
1478
1479         q = blk_trace_get_queue(bdev);
1480         if (q == NULL)
1481                 goto out_bdput;
1482
1483         mutex_lock(&bdev->bd_mutex);
1484
1485         if (attr == &dev_attr_enable) {
1486                 if (value)
1487                         ret = blk_trace_setup_queue(q, bdev->bd_dev);
1488                 else
1489                         ret = blk_trace_remove_queue(q);
1490                 goto out_unlock_bdev;
1491         }
1492
1493         ret = 0;
1494         if (q->blk_trace == NULL)
1495                 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1496
1497         if (ret == 0) {
1498                 if (attr == &dev_attr_act_mask)
1499                         q->blk_trace->act_mask = value;
1500                 else if (attr == &dev_attr_pid)
1501                         q->blk_trace->pid = value;
1502                 else if (attr == &dev_attr_start_lba)
1503                         q->blk_trace->start_lba = value;
1504                 else if (attr == &dev_attr_end_lba)
1505                         q->blk_trace->end_lba = value;
1506         }
1507
1508 out_unlock_bdev:
1509         mutex_unlock(&bdev->bd_mutex);
1510 out_bdput:
1511         bdput(bdev);
1512 out_unlock_kernel:
1513         unlock_kernel();
1514 out:
1515         return ret ? ret : count;
1516 }
1517