git.oblomov.eu Git - linux-2.6/blob - kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/debugfs.h>
  25 #include <linux/time.h>
  26 #include <trace/block.h>
  27 #include <linux/uaccess.h>
  28 #include "trace_output.h"
  29
  30 static unsigned int blktrace_seq __read_mostly = 1;
  31
  32 static struct trace_array *blk_tr;
  33 static bool blk_tracer_enabled __read_mostly;
  34
  35 /* Select an alternative, minimalistic output than the original one */
  36 #define TRACE_BLK_OPT_CLASSIC   0x1
  37
  38 static struct tracer_opt blk_tracer_opts[] = {
  39         /* Default disable the minimalistic output */
  40         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  41         { }
  42 };
  43
  44 static struct tracer_flags blk_tracer_flags = {
  45         .val  = 0,
  46         .opts = blk_tracer_opts,
  47 };
  48
  49 /* Global reference count of probes */
  50 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  51
  52 static void blk_register_tracepoints(void);
  53 static void blk_unregister_tracepoints(void);
  54
  55 /*
  56  * Send out a notify message.
  57  */
  58 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  59                        const void *data, size_t len)
  60 {
  61         struct blk_io_trace *t;
  62         struct ring_buffer_event *event = NULL;
  63         int pc = 0;
  64         int cpu = smp_processor_id();
  65         bool blk_tracer = blk_tracer_enabled;
  66
  67         if (blk_tracer) {
  68                 pc = preempt_count();
  69                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
  70                                                   sizeof(*t) + len,
  71                                                   0, pc);
  72                 if (!event)
  73                         return;
  74                 t = ring_buffer_event_data(event);
  75                 goto record_it;
  76         }
  77
  78         if (!bt->rchan)
  79                 return;
  80
  81         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  82         if (t) {
  83                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  84                 t->time = ktime_to_ns(ktime_get());
  85 record_it:
  86                 t->device = bt->dev;
  87                 t->action = action;
  88                 t->pid = pid;
  89                 t->cpu = cpu;
  90                 t->pdu_len = len;
  91                 memcpy((void *) t + sizeof(*t), data, len);
  92
  93                 if (blk_tracer)
  94                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
  95         }
  96 }
  97
  98 /*
  99  * Send out a notify for this process, if we haven't done so since a trace
 100  * started
 101  */
 102 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 103 {
 104         tsk->btrace_seq = blktrace_seq;
 105         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 106 }
 107
 108 static void trace_note_time(struct blk_trace *bt)
 109 {
 110         struct timespec now;
 111         unsigned long flags;
 112         u32 words[2];
 113
 114         getnstimeofday(&now);
 115         words[0] = now.tv_sec;
 116         words[1] = now.tv_nsec;
 117
 118         local_irq_save(flags);
 119         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 120         local_irq_restore(flags);
 121 }
 122
 123 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 124 {
 125         int n;
 126         va_list args;
 127         unsigned long flags;
 128         char *buf;
 129
 130         if (unlikely(bt->trace_state != Blktrace_running &&
 131                      !blk_tracer_enabled))
 132                 return;
 133
 134         local_irq_save(flags);
 135         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 136         va_start(args, fmt);
 137         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 138         va_end(args);
 139
 140         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 141         local_irq_restore(flags);
 142 }
 143 EXPORT_SYMBOL_GPL(__trace_note_message);
 144
 145 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 146                          pid_t pid)
 147 {
 148         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 149                 return 1;
 150         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 151                 return 1;
 152         if (bt->pid && pid != bt->pid)
 153                 return 1;
 154
 155         return 0;
 156 }
 157
 158 /*
 159  * Data direction bit lookup
 160  */
 161 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 162                                  BLK_TC_ACT(BLK_TC_WRITE) };
 163
 164 /* The ilog2() calls fall out because they're constant */
 165 #define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
 166           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
 167
 168 /*
 169  * The worker for the various blk_add_trace*() types. Fills out a
 170  * blk_io_trace structure and places it in a per-cpu subbuffer.
 171  */
 172 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 173                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 174 {
 175         struct task_struct *tsk = current;
 176         struct ring_buffer_event *event = NULL;
 177         struct blk_io_trace *t;
 178         unsigned long flags = 0;
 179         unsigned long *sequence;
 180         pid_t pid;
 181         int cpu, pc = 0;
 182         bool blk_tracer = blk_tracer_enabled;
 183
 184         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 185                 return;
 186
 187         what |= ddir_act[rw & WRITE];
 188         what |= MASK_TC_BIT(rw, BARRIER);
 189         what |= MASK_TC_BIT(rw, SYNCIO);
 190         what |= MASK_TC_BIT(rw, AHEAD);
 191         what |= MASK_TC_BIT(rw, META);
 192         what |= MASK_TC_BIT(rw, DISCARD);
 193
 194         pid = tsk->pid;
 195         if (act_log_check(bt, what, sector, pid))
 196                 return;
 197         cpu = raw_smp_processor_id();
 198
 199         if (blk_tracer) {
 200                 tracing_record_cmdline(current);
 201
 202                 pc = preempt_count();
 203                 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
 204                                                   sizeof(*t) + pdu_len,
 205                                                   0, pc);
 206                 if (!event)
 207                         return;
 208                 t = ring_buffer_event_data(event);
 209                 goto record_it;
 210         }
 211
 212         /*
 213          * A word about the locking here - we disable interrupts to reserve
 214          * some space in the relay per-cpu buffer, to prevent an irq
 215          * from coming in and stepping on our toes.
 216          */
 217         local_irq_save(flags);
 218
 219         if (unlikely(tsk->btrace_seq != blktrace_seq))
 220                 trace_note_tsk(bt, tsk);
 221
 222         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 223         if (t) {
 224                 sequence = per_cpu_ptr(bt->sequence, cpu);
 225
 226                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 227                 t->sequence = ++(*sequence);
 228                 t->time = ktime_to_ns(ktime_get());
 229 record_it:
 230                 /*
 231                  * These two are not needed in ftrace as they are in the
 232                  * generic trace_entry, filled by tracing_generic_entry_update,
 233                  * but for the trace_event->bin() synthesizer benefit we do it
 234                  * here too.
 235                  */
 236                 t->cpu = cpu;
 237                 t->pid = pid;
 238
 239                 t->sector = sector;
 240                 t->bytes = bytes;
 241                 t->action = what;
 242                 t->device = bt->dev;
 243                 t->error = error;
 244                 t->pdu_len = pdu_len;
 245
 246                 if (pdu_len)
 247                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 248
 249                 if (blk_tracer) {
 250                         trace_buffer_unlock_commit(blk_tr, event, 0, pc);
 251                         return;
 252                 }
 253         }
 254
 255         local_irq_restore(flags);
 256 }
 257
 258 static struct dentry *blk_tree_root;
 259 static DEFINE_MUTEX(blk_tree_mutex);
 260
 261 static void blk_trace_free(struct blk_trace *bt)
 262 {
 263         debugfs_remove(bt->msg_file);
 264         debugfs_remove(bt->dropped_file);
 265         debugfs_remove(bt->dir);
 266         relay_close(bt->rchan);
 267         free_percpu(bt->sequence);
 268         free_percpu(bt->msg_data);
 269         kfree(bt);
 270 }
 271
 272 static void blk_trace_cleanup(struct blk_trace *bt)
 273 {
 274         blk_trace_free(bt);
 275         if (atomic_dec_and_test(&blk_probes_ref))
 276                 blk_unregister_tracepoints();
 277 }
 278
 279 int blk_trace_remove(struct request_queue *q)
 280 {
 281         struct blk_trace *bt;
 282
 283         bt = xchg(&q->blk_trace, NULL);
 284         if (!bt)
 285                 return -EINVAL;
 286
 287         if (bt->trace_state != Blktrace_running)
 288                 blk_trace_cleanup(bt);
 289
 290         return 0;
 291 }
 292 EXPORT_SYMBOL_GPL(blk_trace_remove);
 293
 294 static int blk_dropped_open(struct inode *inode, struct file *filp)
 295 {
 296         filp->private_data = inode->i_private;
 297
 298         return 0;
 299 }
 300
 301 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 302                                 size_t count, loff_t *ppos)
 303 {
 304         struct blk_trace *bt = filp->private_data;
 305         char buf[16];
 306
 307         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 308
 309         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 310 }
 311
 312 static const struct file_operations blk_dropped_fops = {
 313         .owner =        THIS_MODULE,
 314         .open =         blk_dropped_open,
 315         .read =         blk_dropped_read,
 316 };
 317
 318 static int blk_msg_open(struct inode *inode, struct file *filp)
 319 {
 320         filp->private_data = inode->i_private;
 321
 322         return 0;
 323 }
 324
 325 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 326                                 size_t count, loff_t *ppos)
 327 {
 328         char *msg;
 329         struct blk_trace *bt;
 330
 331         if (count >= BLK_TN_MAX_MSG)
 332                 return -EINVAL;
 333
 334         msg = kmalloc(count + 1, GFP_KERNEL);
 335         if (msg == NULL)
 336                 return -ENOMEM;
 337
 338         if (copy_from_user(msg, buffer, count)) {
 339                 kfree(msg);
 340                 return -EFAULT;
 341         }
 342
 343         msg[count] = '\0';
 344         bt = filp->private_data;
 345         __trace_note_message(bt, "%s", msg);
 346         kfree(msg);
 347
 348         return count;
 349 }
 350
 351 static const struct file_operations blk_msg_fops = {
 352         .owner =        THIS_MODULE,
 353         .open =         blk_msg_open,
 354         .write =        blk_msg_write,
 355 };
 356
 357 /*
 358  * Keep track of how many times we encountered a full subbuffer, to aid
 359  * the user space app in telling how many lost events there were.
 360  */
 361 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 362                                      void *prev_subbuf, size_t prev_padding)
 363 {
 364         struct blk_trace *bt;
 365
 366         if (!relay_buf_full(buf))
 367                 return 1;
 368
 369         bt = buf->chan->private_data;
 370         atomic_inc(&bt->dropped);
 371         return 0;
 372 }
 373
 374 static int blk_remove_buf_file_callback(struct dentry *dentry)
 375 {
 376         struct dentry *parent = dentry->d_parent;
 377         debugfs_remove(dentry);
 378
 379         /*
 380         * this will fail for all but the last file, but that is ok. what we
 381         * care about is the top level buts->name directory going away, when
 382         * the last trace file is gone. Then we don't have to rmdir() that
 383         * manually on trace stop, so it nicely solves the issue with
 384         * force killing of running traces.
 385         */
 386
 387         debugfs_remove(parent);
 388         return 0;
 389 }
 390
 391 static struct dentry *blk_create_buf_file_callback(const char *filename,
 392                                                    struct dentry *parent,
 393                                                    int mode,
 394                                                    struct rchan_buf *buf,
 395                                                    int *is_global)
 396 {
 397         return debugfs_create_file(filename, mode, parent, buf,
 398                                         &relay_file_operations);
 399 }
 400
 401 static struct rchan_callbacks blk_relay_callbacks = {
 402         .subbuf_start           = blk_subbuf_start_callback,
 403         .create_buf_file        = blk_create_buf_file_callback,
 404         .remove_buf_file        = blk_remove_buf_file_callback,
 405 };
 406
 407 static void blk_trace_setup_lba(struct blk_trace *bt,
 408                                 struct block_device *bdev)
 409 {
 410         struct hd_struct *part = NULL;
 411
 412         if (bdev)
 413                 part = bdev->bd_part;
 414
 415         if (part) {
 416                 bt->start_lba = part->start_sect;
 417                 bt->end_lba = part->start_sect + part->nr_sects;
 418         } else {
 419                 bt->start_lba = 0;
 420                 bt->end_lba = -1ULL;
 421         }
 422 }
 423
 424 /*
 425  * Setup everything required to start tracing
 426  */
 427 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 428                        struct block_device *bdev,
 429                        struct blk_user_trace_setup *buts)
 430 {
 431         struct blk_trace *old_bt, *bt = NULL;
 432         struct dentry *dir = NULL;
 433         int ret, i;
 434
 435         if (!buts->buf_size || !buts->buf_nr)
 436                 return -EINVAL;
 437
 438         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 439         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 440
 441         /*
 442          * some device names have larger paths - convert the slashes
 443          * to underscores for this to work as expected
 444          */
 445         for (i = 0; i < strlen(buts->name); i++)
 446                 if (buts->name[i] == '/')
 447                         buts->name[i] = '_';
 448
 449         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 450         if (!bt)
 451                 return -ENOMEM;
 452
 453         ret = -ENOMEM;
 454         bt->sequence = alloc_percpu(unsigned long);
 455         if (!bt->sequence)
 456                 goto err;
 457
 458         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 459         if (!bt->msg_data)
 460                 goto err;
 461
 462         ret = -ENOENT;
 463
 464         mutex_lock(&blk_tree_mutex);
 465         if (!blk_tree_root) {
 466                 blk_tree_root = debugfs_create_dir("block", NULL);
 467                 if (!blk_tree_root) {
 468                         mutex_unlock(&blk_tree_mutex);
 469                         goto err;
 470                 }
 471         }
 472         mutex_unlock(&blk_tree_mutex);
 473
 474         dir = debugfs_create_dir(buts->name, blk_tree_root);
 475
 476         if (!dir)
 477                 goto err;
 478
 479         bt->dir = dir;
 480         bt->dev = dev;
 481         atomic_set(&bt->dropped, 0);
 482
 483         ret = -EIO;
 484         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 485                                                &blk_dropped_fops);
 486         if (!bt->dropped_file)
 487                 goto err;
 488
 489         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 490         if (!bt->msg_file)
 491                 goto err;
 492
 493         bt->rchan = relay_open("trace", dir, buts->buf_size,
 494                                 buts->buf_nr, &blk_relay_callbacks, bt);
 495         if (!bt->rchan)
 496                 goto err;
 497
 498         bt->act_mask = buts->act_mask;
 499         if (!bt->act_mask)
 500                 bt->act_mask = (u16) -1;
 501
 502         blk_trace_setup_lba(bt, bdev);
 503
 504         /* overwrite with user settings */
 505         if (buts->start_lba)
 506                 bt->start_lba = buts->start_lba;
 507         if (buts->end_lba)
 508                 bt->end_lba = buts->end_lba;
 509
 510         bt->pid = buts->pid;
 511         bt->trace_state = Blktrace_setup;
 512
 513         ret = -EBUSY;
 514         old_bt = xchg(&q->blk_trace, bt);
 515         if (old_bt) {
 516                 (void) xchg(&q->blk_trace, old_bt);
 517                 goto err;
 518         }
 519
 520         if (atomic_inc_return(&blk_probes_ref) == 1)
 521                 blk_register_tracepoints();
 522
 523         return 0;
 524 err:
 525         blk_trace_free(bt);
 526         return ret;
 527 }
 528
 529 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 530                     struct block_device *bdev,
 531                     char __user *arg)
 532 {
 533         struct blk_user_trace_setup buts;
 534         int ret;
 535
 536         ret = copy_from_user(&buts, arg, sizeof(buts));
 537         if (ret)
 538                 return -EFAULT;
 539
 540         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 541         if (ret)
 542                 return ret;
 543
 544         if (copy_to_user(arg, &buts, sizeof(buts)))
 545                 return -EFAULT;
 546
 547         return 0;
 548 }
 549 EXPORT_SYMBOL_GPL(blk_trace_setup);
 550
 551 int blk_trace_startstop(struct request_queue *q, int start)
 552 {
 553         int ret;
 554         struct blk_trace *bt = q->blk_trace;
 555
 556         if (bt == NULL)
 557                 return -EINVAL;
 558
 559         /*
 560          * For starting a trace, we can transition from a setup or stopped
 561          * trace. For stopping a trace, the state must be running
 562          */
 563         ret = -EINVAL;
 564         if (start) {
 565                 if (bt->trace_state == Blktrace_setup ||
 566                     bt->trace_state == Blktrace_stopped) {
 567                         blktrace_seq++;
 568                         smp_mb();
 569                         bt->trace_state = Blktrace_running;
 570
 571                         trace_note_time(bt);
 572                         ret = 0;
 573                 }
 574         } else {
 575                 if (bt->trace_state == Blktrace_running) {
 576                         bt->trace_state = Blktrace_stopped;
 577                         relay_flush(bt->rchan);
 578                         ret = 0;
 579                 }
 580         }
 581
 582         return ret;
 583 }
 584 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 585
 586 /**
 587  * blk_trace_ioctl: - handle the ioctls associated with tracing
 588  * @bdev:       the block device
 589  * @cmd:        the ioctl cmd
 590  * @arg:        the argument data, if any
 591  *
 592  **/
 593 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 594 {
 595         struct request_queue *q;
 596         int ret, start = 0;
 597         char b[BDEVNAME_SIZE];
 598
 599         q = bdev_get_queue(bdev);
 600         if (!q)
 601                 return -ENXIO;
 602
 603         mutex_lock(&bdev->bd_mutex);
 604
 605         switch (cmd) {
 606         case BLKTRACESETUP:
 607                 bdevname(bdev, b);
 608                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 609                 break;
 610         case BLKTRACESTART:
 611                 start = 1;
 612         case BLKTRACESTOP:
 613                 ret = blk_trace_startstop(q, start);
 614                 break;
 615         case BLKTRACETEARDOWN:
 616                 ret = blk_trace_remove(q);
 617                 break;
 618         default:
 619                 ret = -ENOTTY;
 620                 break;
 621         }
 622
 623         mutex_unlock(&bdev->bd_mutex);
 624         return ret;
 625 }
 626
 627 /**
 628  * blk_trace_shutdown: - stop and cleanup trace structures
 629  * @q:    the request queue associated with the device
 630  *
 631  **/
 632 void blk_trace_shutdown(struct request_queue *q)
 633 {
 634         if (q->blk_trace) {
 635                 blk_trace_startstop(q, 0);
 636                 blk_trace_remove(q);
 637         }
 638 }
 639
 640 /*
 641  * blktrace probes
 642  */
 643
 644 /**
 645  * blk_add_trace_rq - Add a trace for a request oriented action
 646  * @q:          queue the io is for
 647  * @rq:         the source request
 648  * @what:       the action
 649  *
 650  * Description:
 651  *     Records an action against a request. Will log the bio offset + size.
 652  *
 653  **/
 654 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 655                                     u32 what)
 656 {
 657         struct blk_trace *bt = q->blk_trace;
 658         int rw = rq->cmd_flags & 0x03;
 659
 660         if (likely(!bt))
 661                 return;
 662
 663         if (blk_discard_rq(rq))
 664                 rw |= (1 << BIO_RW_DISCARD);
 665
 666         if (blk_pc_request(rq)) {
 667                 what |= BLK_TC_ACT(BLK_TC_PC);
 668                 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
 669                                 rq->cmd_len, rq->cmd);
 670         } else  {
 671                 what |= BLK_TC_ACT(BLK_TC_FS);
 672                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 673                                 rw, what, rq->errors, 0, NULL);
 674         }
 675 }
 676
 677 static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
 678 {
 679         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 680 }
 681
 682 static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
 683 {
 684         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 685 }
 686
 687 static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
 688 {
 689         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 690 }
 691
 692 static void blk_add_trace_rq_requeue(struct request_queue *q,
 693                                      struct request *rq)
 694 {
 695         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 696 }
 697
 698 static void blk_add_trace_rq_complete(struct request_queue *q,
 699                                       struct request *rq)
 700 {
 701         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 702 }
 703
 704 /**
 705  * blk_add_trace_bio - Add a trace for a bio oriented action
 706  * @q:          queue the io is for
 707  * @bio:        the source bio
 708  * @what:       the action
 709  *
 710  * Description:
 711  *     Records an action against a bio. Will log the bio offset + size.
 712  *
 713  **/
 714 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 715                                      u32 what)
 716 {
 717         struct blk_trace *bt = q->blk_trace;
 718
 719         if (likely(!bt))
 720                 return;
 721
 722         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 723                         !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
 724 }
 725
 726 static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
 727 {
 728         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
 729 }
 730
 731 static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
 732 {
 733         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
 734 }
 735
 736 static void blk_add_trace_bio_backmerge(struct request_queue *q,
 737                                         struct bio *bio)
 738 {
 739         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
 740 }
 741
 742 static void blk_add_trace_bio_frontmerge(struct request_queue *q,
 743                                          struct bio *bio)
 744 {
 745         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
 746 }
 747
 748 static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
 749 {
 750         blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
 751 }
 752
 753 static void blk_add_trace_getrq(struct request_queue *q,
 754                                 struct bio *bio, int rw)
 755 {
 756         if (bio)
 757                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
 758         else {
 759                 struct blk_trace *bt = q->blk_trace;
 760
 761                 if (bt)
 762                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 763         }
 764 }
 765
 766
 767 static void blk_add_trace_sleeprq(struct request_queue *q,
 768                                   struct bio *bio, int rw)
 769 {
 770         if (bio)
 771                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
 772         else {
 773                 struct blk_trace *bt = q->blk_trace;
 774
 775                 if (bt)
 776                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 777                                         0, 0, NULL);
 778         }
 779 }
 780
 781 static void blk_add_trace_plug(struct request_queue *q)
 782 {
 783         struct blk_trace *bt = q->blk_trace;
 784
 785         if (bt)
 786                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 787 }
 788
 789 static void blk_add_trace_unplug_io(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_IO, 0,
 798                                 sizeof(rpdu), &rpdu);
 799         }
 800 }
 801
 802 static void blk_add_trace_unplug_timer(struct request_queue *q)
 803 {
 804         struct blk_trace *bt = q->blk_trace;
 805
 806         if (bt) {
 807                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
 808                 __be64 rpdu = cpu_to_be64(pdu);
 809
 810                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
 811                                 sizeof(rpdu), &rpdu);
 812         }
 813 }
 814
 815 static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
 816                                 unsigned int pdu)
 817 {
 818         struct blk_trace *bt = q->blk_trace;
 819
 820         if (bt) {
 821                 __be64 rpdu = cpu_to_be64(pdu);
 822
 823                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 824                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 825                                 sizeof(rpdu), &rpdu);
 826         }
 827 }
 828
 829 /**
 830  * blk_add_trace_remap - Add a trace for a remap operation
 831  * @q:          queue the io is for
 832  * @bio:        the source bio
 833  * @dev:        target device
 834  * @from:       source sector
 835  *
 836  * Description:
 837  *     Device mapper or raid target sometimes need to split a bio because
 838  *     it spans a stripe (or similar). Add a trace for that action.
 839  *
 840  **/
 841 static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
 842                                        dev_t dev, sector_t from)
 843 {
 844         struct blk_trace *bt = q->blk_trace;
 845         struct blk_io_trace_remap r;
 846
 847         if (likely(!bt))
 848                 return;
 849
 850         r.device_from = cpu_to_be32(dev);
 851         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 852         r.sector_from = cpu_to_be64(from);
 853
 854         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 855                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 856                         sizeof(r), &r);
 857 }
 858
 859 /**
 860  * blk_add_driver_data - Add binary message with driver-specific data
 861  * @q:          queue the io is for
 862  * @rq:         io request
 863  * @data:       driver-specific data
 864  * @len:        length of driver-specific data
 865  *
 866  * Description:
 867  *     Some drivers might want to write driver-specific data per request.
 868  *
 869  **/
 870 void blk_add_driver_data(struct request_queue *q,
 871                          struct request *rq,
 872                          void *data, size_t len)
 873 {
 874         struct blk_trace *bt = q->blk_trace;
 875
 876         if (likely(!bt))
 877                 return;
 878
 879         if (blk_pc_request(rq))
 880                 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
 881                                 rq->errors, len, data);
 882         else
 883                 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
 884                                 0, BLK_TA_DRV_DATA, rq->errors, len, data);
 885 }
 886 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 887
 888 static void blk_register_tracepoints(void)
 889 {
 890         int ret;
 891
 892         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
 893         WARN_ON(ret);
 894         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
 895         WARN_ON(ret);
 896         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
 897         WARN_ON(ret);
 898         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 899         WARN_ON(ret);
 900         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
 901         WARN_ON(ret);
 902         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 903         WARN_ON(ret);
 904         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
 905         WARN_ON(ret);
 906         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 907         WARN_ON(ret);
 908         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 909         WARN_ON(ret);
 910         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
 911         WARN_ON(ret);
 912         ret = register_trace_block_getrq(blk_add_trace_getrq);
 913         WARN_ON(ret);
 914         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
 915         WARN_ON(ret);
 916         ret = register_trace_block_plug(blk_add_trace_plug);
 917         WARN_ON(ret);
 918         ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 919         WARN_ON(ret);
 920         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
 921         WARN_ON(ret);
 922         ret = register_trace_block_split(blk_add_trace_split);
 923         WARN_ON(ret);
 924         ret = register_trace_block_remap(blk_add_trace_remap);
 925         WARN_ON(ret);
 926 }
 927
 928 static void blk_unregister_tracepoints(void)
 929 {
 930         unregister_trace_block_remap(blk_add_trace_remap);
 931         unregister_trace_block_split(blk_add_trace_split);
 932         unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
 933         unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
 934         unregister_trace_block_plug(blk_add_trace_plug);
 935         unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
 936         unregister_trace_block_getrq(blk_add_trace_getrq);
 937         unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
 938         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
 939         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
 940         unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
 941         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
 942         unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
 943         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
 944         unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
 945         unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
 946         unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
 947
 948         tracepoint_synchronize_unregister();
 949 }
 950
 951 /*
 952  * struct blk_io_tracer formatting routines
 953  */
 954
 955 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
 956 {
 957         int i = 0;
 958         int tc = t->action >> BLK_TC_SHIFT;
 959
 960         if (t->action == BLK_TN_MESSAGE) {
 961                 rwbs[i++] = 'N';
 962                 goto out;
 963         }
 964
 965         if (tc & BLK_TC_DISCARD)
 966                 rwbs[i++] = 'D';
 967         else if (tc & BLK_TC_WRITE)
 968                 rwbs[i++] = 'W';
 969         else if (t->bytes)
 970                 rwbs[i++] = 'R';
 971         else
 972                 rwbs[i++] = 'N';
 973
 974         if (tc & BLK_TC_AHEAD)
 975                 rwbs[i++] = 'A';
 976         if (tc & BLK_TC_BARRIER)
 977                 rwbs[i++] = 'B';
 978         if (tc & BLK_TC_SYNC)
 979                 rwbs[i++] = 'S';
 980         if (tc & BLK_TC_META)
 981                 rwbs[i++] = 'M';
 982 out:
 983         rwbs[i] = '\0';
 984 }
 985
 986 static inline
 987 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
 988 {
 989         return (const struct blk_io_trace *)ent;
 990 }
 991
 992 static inline const void *pdu_start(const struct trace_entry *ent)
 993 {
 994         return te_blk_io_trace(ent) + 1;
 995 }
 996
 997 static inline u32 t_action(const struct trace_entry *ent)
 998 {
 999         return te_blk_io_trace(ent)->action;
1000 }
1001
1002 static inline u32 t_bytes(const struct trace_entry *ent)
1003 {
1004         return te_blk_io_trace(ent)->bytes;
1005 }
1006
1007 static inline u32 t_sec(const struct trace_entry *ent)
1008 {
1009         return te_blk_io_trace(ent)->bytes >> 9;
1010 }
1011
1012 static inline unsigned long long t_sector(const struct trace_entry *ent)
1013 {
1014         return te_blk_io_trace(ent)->sector;
1015 }
1016
1017 static inline __u16 t_error(const struct trace_entry *ent)
1018 {
1019         return te_blk_io_trace(ent)->error;
1020 }
1021
1022 static __u64 get_pdu_int(const struct trace_entry *ent)
1023 {
1024         const __u64 *val = pdu_start(ent);
1025         return be64_to_cpu(*val);
1026 }
1027
1028 static void get_pdu_remap(const struct trace_entry *ent,
1029                           struct blk_io_trace_remap *r)
1030 {
1031         const struct blk_io_trace_remap *__r = pdu_start(ent);
1032         __u64 sector_from = __r->sector_from;
1033
1034         r->device_from = be32_to_cpu(__r->device_from);
1035         r->device_to   = be32_to_cpu(__r->device_to);
1036         r->sector_from = be64_to_cpu(sector_from);
1037 }
1038
1039 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1040
1041 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1042 {
1043         char rwbs[6];
1044         unsigned long long ts  = iter->ts;
1045         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1046         unsigned secs          = (unsigned long)ts;
1047         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1048
1049         fill_rwbs(rwbs, t);
1050
1051         return trace_seq_printf(&iter->seq,
1052                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1053                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1054                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1055 }
1056
1057 static int blk_log_action(struct trace_iterator *iter, const char *act)
1058 {
1059         char rwbs[6];
1060         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1061
1062         fill_rwbs(rwbs, t);
1063         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1064                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1065 }
1066
1067 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1068 {
1069         const unsigned char *pdu_buf;
1070         int pdu_len;
1071         int i, end, ret;
1072
1073         pdu_buf = pdu_start(ent);
1074         pdu_len = te_blk_io_trace(ent)->pdu_len;
1075
1076         if (!pdu_len)
1077                 return 1;
1078
1079         /* find the last zero that needs to be printed */
1080         for (end = pdu_len - 1; end >= 0; end--)
1081                 if (pdu_buf[end])
1082                         break;
1083         end++;
1084
1085         if (!trace_seq_putc(s, '('))
1086                 return 0;
1087
1088         for (i = 0; i < pdu_len; i++) {
1089
1090                 ret = trace_seq_printf(s, "%s%02x",
1091                                        i == 0 ? "" : " ", pdu_buf[i]);
1092                 if (!ret)
1093                         return ret;
1094
1095                 /*
1096                  * stop when the rest is just zeroes and indicate so
1097                  * with a ".." appended
1098                  */
1099                 if (i == end && end != pdu_len - 1)
1100                         return trace_seq_puts(s, " ..) ");
1101         }
1102
1103         return trace_seq_puts(s, ") ");
1104 }
1105
1106 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1107 {
1108         char cmd[TASK_COMM_LEN];
1109
1110         trace_find_cmdline(ent->pid, cmd);
1111
1112         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1113                 int ret;
1114
1115                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1116                 if (!ret)
1117                         return 0;
1118                 ret = blk_log_dump_pdu(s, ent);
1119                 if (!ret)
1120                         return 0;
1121                 return trace_seq_printf(s, "[%s]\n", cmd);
1122         } else {
1123                 if (t_sec(ent))
1124                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1125                                                 t_sector(ent), t_sec(ent), cmd);
1126                 return trace_seq_printf(s, "[%s]\n", cmd);
1127         }
1128 }
1129
1130 static int blk_log_with_error(struct trace_seq *s,
1131                               const struct trace_entry *ent)
1132 {
1133         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1134                 int ret;
1135
1136                 ret = blk_log_dump_pdu(s, ent);
1137                 if (ret)
1138                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1139                 return 0;
1140         } else {
1141                 if (t_sec(ent))
1142                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1143                                                 t_sector(ent),
1144                                                 t_sec(ent), t_error(ent));
1145                 return trace_seq_printf(s, "%llu [%d]\n",
1146                                         t_sector(ent), t_error(ent));
1147         }
1148 }
1149
1150 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1151 {
1152         struct blk_io_trace_remap r = { .device_from = 0, };
1153
1154         get_pdu_remap(ent, &r);
1155         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1156                                 t_sector(ent), t_sec(ent),
1157                                 MAJOR(r.device_from), MINOR(r.device_from),
1158                                 (unsigned long long)r.sector_from);
1159 }
1160
1161 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1162 {
1163         char cmd[TASK_COMM_LEN];
1164
1165         trace_find_cmdline(ent->pid, cmd);
1166
1167         return trace_seq_printf(s, "[%s]\n", cmd);
1168 }
1169
1170 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1171 {
1172         char cmd[TASK_COMM_LEN];
1173
1174         trace_find_cmdline(ent->pid, cmd);
1175
1176         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1177 }
1178
1179 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1180 {
1181         char cmd[TASK_COMM_LEN];
1182
1183         trace_find_cmdline(ent->pid, cmd);
1184
1185         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1186                                 get_pdu_int(ent), cmd);
1187 }
1188
1189 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1190 {
1191         int ret;
1192         const struct blk_io_trace *t = te_blk_io_trace(ent);
1193
1194         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1195         if (ret)
1196                 return trace_seq_putc(s, '\n');
1197         return ret;
1198 }
1199
1200 /*
1201  * struct tracer operations
1202  */
1203
1204 static void blk_tracer_print_header(struct seq_file *m)
1205 {
1206         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1207                 return;
1208         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1209                     "#  |     |     |           |   |   |\n");
1210 }
1211
1212 static void blk_tracer_start(struct trace_array *tr)
1213 {
1214         blk_tracer_enabled = true;
1215 }
1216
1217 static int blk_tracer_init(struct trace_array *tr)
1218 {
1219         blk_tr = tr;
1220         blk_tracer_start(tr);
1221         return 0;
1222 }
1223
1224 static void blk_tracer_stop(struct trace_array *tr)
1225 {
1226         blk_tracer_enabled = false;
1227 }
1228
1229 static void blk_tracer_reset(struct trace_array *tr)
1230 {
1231         blk_tracer_stop(tr);
1232 }
1233
1234 static const struct {
1235         const char *act[2];
1236         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1237 } what2act[] = {
1238         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1239         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1240         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1241         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1242         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1243         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1244         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1245         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1246         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1247         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1248         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1249         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1250         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1251         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1252         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1253 };
1254
1255 static enum print_line_t print_one_line(struct trace_iterator *iter,
1256                                         bool classic)
1257 {
1258         struct trace_seq *s = &iter->seq;
1259         const struct blk_io_trace *t;
1260         u16 what;
1261         int ret;
1262         bool long_act;
1263         blk_log_action_t *log_action;
1264
1265         t          = te_blk_io_trace(iter->ent);
1266         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1267         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1268         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1269
1270         if (t->action == BLK_TN_MESSAGE) {
1271                 ret = log_action(iter, long_act ? "message" : "m");
1272                 if (ret)
1273                         ret = blk_log_msg(s, iter->ent);
1274                 goto out;
1275         }
1276
1277         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1278                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1279         else {
1280                 ret = log_action(iter, what2act[what].act[long_act]);
1281                 if (ret)
1282                         ret = what2act[what].print(s, iter->ent);
1283         }
1284 out:
1285         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1286 }
1287
1288 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1289                                                int flags)
1290 {
1291         return print_one_line(iter, false);
1292 }
1293
1294 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1295 {
1296         struct trace_seq *s = &iter->seq;
1297         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1298         const int offset = offsetof(struct blk_io_trace, sector);
1299         struct blk_io_trace old = {
1300                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1301                 .time     = iter->ts,
1302         };
1303
1304         if (!trace_seq_putmem(s, &old, offset))
1305                 return 0;
1306         return trace_seq_putmem(s, &t->sector,
1307                                 sizeof(old) - offset + t->pdu_len);
1308 }
1309
1310 static enum print_line_t
1311 blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1312 {
1313         return blk_trace_synthesize_old_trace(iter) ?
1314                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1315 }
1316
1317 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1318 {
1319         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1320                 return TRACE_TYPE_UNHANDLED;
1321
1322         return print_one_line(iter, true);
1323 }
1324
1325 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1326 {
1327         /* don't output context-info for blk_classic output */
1328         if (bit == TRACE_BLK_OPT_CLASSIC) {
1329                 if (set)
1330                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1331                 else
1332                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1333         }
1334         return 0;
1335 }
1336
1337 static struct tracer blk_tracer __read_mostly = {
1338         .name           = "blk",
1339         .init           = blk_tracer_init,
1340         .reset          = blk_tracer_reset,
1341         .start          = blk_tracer_start,
1342         .stop           = blk_tracer_stop,
1343         .print_header   = blk_tracer_print_header,
1344         .print_line     = blk_tracer_print_line,
1345         .flags          = &blk_tracer_flags,
1346         .set_flag       = blk_tracer_set_flag,
1347 };
1348
1349 static struct trace_event trace_blk_event = {
1350         .type           = TRACE_BLK,
1351         .trace          = blk_trace_event_print,
1352         .binary         = blk_trace_event_print_binary,
1353 };
1354
1355 static int __init init_blk_tracer(void)
1356 {
1357         if (!register_ftrace_event(&trace_blk_event)) {
1358                 pr_warning("Warning: could not register block events\n");
1359                 return 1;
1360         }
1361
1362         if (register_tracer(&blk_tracer) != 0) {
1363                 pr_warning("Warning: could not register the block tracer\n");
1364                 unregister_ftrace_event(&trace_blk_event);
1365                 return 1;
1366         }
1367
1368         return 0;
1369 }
1370
1371 device_initcall(init_blk_tracer);
1372
1373 static int blk_trace_remove_queue(struct request_queue *q)
1374 {
1375         struct blk_trace *bt;
1376
1377         bt = xchg(&q->blk_trace, NULL);
1378         if (bt == NULL)
1379                 return -EINVAL;
1380
1381         if (atomic_dec_and_test(&blk_probes_ref))
1382                 blk_unregister_tracepoints();
1383
1384         blk_trace_free(bt);
1385         return 0;
1386 }
1387
1388 /*
1389  * Setup everything required to start tracing
1390  */
1391 static int blk_trace_setup_queue(struct request_queue *q,
1392                                  struct block_device *bdev)
1393 {
1394         struct blk_trace *old_bt, *bt = NULL;
1395         int ret = -ENOMEM;
1396
1397         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1398         if (!bt)
1399                 return -ENOMEM;
1400
1401         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1402         if (!bt->msg_data)
1403                 goto free_bt;
1404
1405         bt->dev = bdev->bd_dev;
1406         bt->act_mask = (u16)-1;
1407
1408         blk_trace_setup_lba(bt, bdev);
1409
1410         old_bt = xchg(&q->blk_trace, bt);
1411         if (old_bt != NULL) {
1412                 (void)xchg(&q->blk_trace, old_bt);
1413                 ret = -EBUSY;
1414                 goto free_bt;
1415         }
1416
1417         if (atomic_inc_return(&blk_probes_ref) == 1)
1418                 blk_register_tracepoints();
1419         return 0;
1420
1421 free_bt:
1422         blk_trace_free(bt);
1423         return ret;
1424 }
1425
1426 /*
1427  * sysfs interface to enable and configure tracing
1428  */
1429
1430 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1431                                          struct device_attribute *attr,
1432                                          char *buf);
1433 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1434                                           struct device_attribute *attr,
1435                                           const char *buf, size_t count);
1436 #define BLK_TRACE_DEVICE_ATTR(_name) \
1437         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1438                     sysfs_blk_trace_attr_show, \
1439                     sysfs_blk_trace_attr_store)
1440
1441 static BLK_TRACE_DEVICE_ATTR(enable);
1442 static BLK_TRACE_DEVICE_ATTR(act_mask);
1443 static BLK_TRACE_DEVICE_ATTR(pid);
1444 static BLK_TRACE_DEVICE_ATTR(start_lba);
1445 static BLK_TRACE_DEVICE_ATTR(end_lba);
1446
1447 static struct attribute *blk_trace_attrs[] = {
1448         &dev_attr_enable.attr,
1449         &dev_attr_act_mask.attr,
1450         &dev_attr_pid.attr,
1451         &dev_attr_start_lba.attr,
1452         &dev_attr_end_lba.attr,
1453         NULL
1454 };
1455
1456 struct attribute_group blk_trace_attr_group = {
1457         .name  = "trace",
1458         .attrs = blk_trace_attrs,
1459 };
1460
1461 static const struct {
1462         int mask;
1463         const char *str;
1464 } mask_maps[] = {
1465         { BLK_TC_READ,          "read"          },
1466         { BLK_TC_WRITE,         "write"         },
1467         { BLK_TC_BARRIER,       "barrier"       },
1468         { BLK_TC_SYNC,          "sync"          },
1469         { BLK_TC_QUEUE,         "queue"         },
1470         { BLK_TC_REQUEUE,       "requeue"       },
1471         { BLK_TC_ISSUE,         "issue"         },
1472         { BLK_TC_COMPLETE,      "complete"      },
1473         { BLK_TC_FS,            "fs"            },
1474         { BLK_TC_PC,            "pc"            },
1475         { BLK_TC_AHEAD,         "ahead"         },
1476         { BLK_TC_META,          "meta"          },
1477         { BLK_TC_DISCARD,       "discard"       },
1478         { BLK_TC_DRV_DATA,      "drv_data"      },
1479 };
1480
1481 static int blk_trace_str2mask(const char *str)
1482 {
1483         int i;
1484         int mask = 0;
1485         char *buf, *s, *token;
1486
1487         buf = kstrdup(str, GFP_KERNEL);
1488         if (buf == NULL)
1489                 return -ENOMEM;
1490         s = strstrip(buf);
1491
1492         while (1) {
1493                 token = strsep(&s, ",");
1494                 if (token == NULL)
1495                         break;
1496
1497                 if (*token == '\0')
1498                         continue;
1499
1500                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1501                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1502                                 mask |= mask_maps[i].mask;
1503                                 break;
1504                         }
1505                 }
1506                 if (i == ARRAY_SIZE(mask_maps)) {
1507                         mask = -EINVAL;
1508                         break;
1509                 }
1510         }
1511         kfree(buf);
1512
1513         return mask;
1514 }
1515
1516 static ssize_t blk_trace_mask2str(char *buf, int mask)
1517 {
1518         int i;
1519         char *p = buf;
1520
1521         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1522                 if (mask & mask_maps[i].mask) {
1523                         p += sprintf(p, "%s%s",
1524                                     (p == buf) ? "" : ",", mask_maps[i].str);
1525                 }
1526         }
1527         *p++ = '\n';
1528
1529         return p - buf;
1530 }
1531
1532 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1533 {
1534         if (bdev->bd_disk == NULL)
1535                 return NULL;
1536
1537         return bdev_get_queue(bdev);
1538 }
1539
1540 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1541                                          struct device_attribute *attr,
1542                                          char *buf)
1543 {
1544         struct hd_struct *p = dev_to_part(dev);
1545         struct request_queue *q;
1546         struct block_device *bdev;
1547         ssize_t ret = -ENXIO;
1548
1549         lock_kernel();
1550         bdev = bdget(part_devt(p));
1551         if (bdev == NULL)
1552                 goto out_unlock_kernel;
1553
1554         q = blk_trace_get_queue(bdev);
1555         if (q == NULL)
1556                 goto out_bdput;
1557
1558         mutex_lock(&bdev->bd_mutex);
1559
1560         if (attr == &dev_attr_enable) {
1561                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1562                 goto out_unlock_bdev;
1563         }
1564
1565         if (q->blk_trace == NULL)
1566                 ret = sprintf(buf, "disabled\n");
1567         else if (attr == &dev_attr_act_mask)
1568                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1569         else if (attr == &dev_attr_pid)
1570                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1571         else if (attr == &dev_attr_start_lba)
1572                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1573         else if (attr == &dev_attr_end_lba)
1574                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1575
1576 out_unlock_bdev:
1577         mutex_unlock(&bdev->bd_mutex);
1578 out_bdput:
1579         bdput(bdev);
1580 out_unlock_kernel:
1581         unlock_kernel();
1582         return ret;
1583 }
1584
1585 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1586                                           struct device_attribute *attr,
1587                                           const char *buf, size_t count)
1588 {
1589         struct block_device *bdev;
1590         struct request_queue *q;
1591         struct hd_struct *p;
1592         u64 value;
1593         ssize_t ret = -EINVAL;
1594
1595         if (count == 0)
1596                 goto out;
1597
1598         if (attr == &dev_attr_act_mask) {
1599                 if (sscanf(buf, "%llx", &value) != 1) {
1600                         /* Assume it is a list of trace category names */
1601                         ret = blk_trace_str2mask(buf);
1602                         if (ret < 0)
1603                                 goto out;
1604                         value = ret;
1605                 }
1606         } else if (sscanf(buf, "%llu", &value) != 1)
1607                 goto out;
1608
1609         ret = -ENXIO;
1610
1611         lock_kernel();
1612         p = dev_to_part(dev);
1613         bdev = bdget(part_devt(p));
1614         if (bdev == NULL)
1615                 goto out_unlock_kernel;
1616
1617         q = blk_trace_get_queue(bdev);
1618         if (q == NULL)
1619                 goto out_bdput;
1620
1621         mutex_lock(&bdev->bd_mutex);
1622
1623         if (attr == &dev_attr_enable) {
1624                 if (value)
1625                         ret = blk_trace_setup_queue(q, bdev);
1626                 else
1627                         ret = blk_trace_remove_queue(q);
1628                 goto out_unlock_bdev;
1629         }
1630
1631         ret = 0;
1632         if (q->blk_trace == NULL)
1633                 ret = blk_trace_setup_queue(q, bdev);
1634
1635         if (ret == 0) {
1636                 if (attr == &dev_attr_act_mask)
1637                         q->blk_trace->act_mask = value;
1638                 else if (attr == &dev_attr_pid)
1639                         q->blk_trace->pid = value;
1640                 else if (attr == &dev_attr_start_lba)
1641                         q->blk_trace->start_lba = value;
1642                 else if (attr == &dev_attr_end_lba)
1643                         q->blk_trace->end_lba = value;
1644         }
1645
1646 out_unlock_bdev:
1647         mutex_unlock(&bdev->bd_mutex);
1648 out_bdput:
1649         bdput(bdev);
1650 out_unlock_kernel:
1651         unlock_kernel();
1652 out:
1653         return ret ? ret : count;
1654 }
1655
1656 int blk_trace_init_sysfs(struct device *dev)
1657 {
1658         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1659 }
1660