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/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/export.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29
  30 #include <trace/events/block.h>
  31
  32 #include "trace_output.h"
  33
  34 #ifdef CONFIG_BLK_DEV_IO_TRACE
  35
  36 static unsigned int blktrace_seq __read_mostly = 1;
  37
  38 static struct trace_array *blk_tr;
  39 static bool blk_tracer_enabled __read_mostly;
  40
  41 /* Select an alternative, minimalistic output than the original one */
  42 #define TRACE_BLK_OPT_CLASSIC   0x1
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47         { }
  48 };
  49
  50 static struct tracer_flags blk_tracer_flags = {
  51         .val  = 0,
  52         .opts = blk_tracer_opts,
  53 };
  54
  55 /* Global reference count of probes */
  56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  57
  58 static void blk_register_tracepoints(void);
  59 static void blk_unregister_tracepoints(void);
  60
  61 /*
  62  * Send out a notify message.
  63  */
  64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  65                        const void *data, size_t len)
  66 {
  67         struct blk_io_trace *t;
  68         struct ring_buffer_event *event = NULL;
  69         struct ring_buffer *buffer = NULL;
  70         int pc = 0;
  71         int cpu = smp_processor_id();
  72         bool blk_tracer = blk_tracer_enabled;
  73
  74         if (blk_tracer) {
  75                 buffer = blk_tr->buffer;
  76                 pc = preempt_count();
  77                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  78                                                   sizeof(*t) + len,
  79                                                   0, pc);
  80                 if (!event)
  81                         return;
  82                 t = ring_buffer_event_data(event);
  83                 goto record_it;
  84         }
  85
  86         if (!bt->rchan)
  87                 return;
  88
  89         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  90         if (t) {
  91                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  92                 t->time = ktime_to_ns(ktime_get());
  93 record_it:
  94                 t->device = bt->dev;
  95                 t->action = action;
  96                 t->pid = pid;
  97                 t->cpu = cpu;
  98                 t->pdu_len = len;
  99                 memcpy((void *) t + sizeof(*t), data, len);
 100
 101                 if (blk_tracer)
 102                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 103         }
 104 }
 105
 106 /*
 107  * Send out a notify for this process, if we haven't done so since a trace
 108  * started
 109  */
 110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 111 {
 112         tsk->btrace_seq = blktrace_seq;
 113         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 114 }
 115
 116 static void trace_note_time(struct blk_trace *bt)
 117 {
 118         struct timespec now;
 119         unsigned long flags;
 120         u32 words[2];
 121
 122         getnstimeofday(&now);
 123         words[0] = now.tv_sec;
 124         words[1] = now.tv_nsec;
 125
 126         local_irq_save(flags);
 127         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 128         local_irq_restore(flags);
 129 }
 130
 131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 132 {
 133         int n;
 134         va_list args;
 135         unsigned long flags;
 136         char *buf;
 137
 138         if (unlikely(bt->trace_state != Blktrace_running &&
 139                      !blk_tracer_enabled))
 140                 return;
 141
 142         /*
 143          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 144          * message to the trace.
 145          */
 146         if (!(bt->act_mask & BLK_TC_NOTIFY))
 147                 return;
 148
 149         local_irq_save(flags);
 150         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 151         va_start(args, fmt);
 152         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 153         va_end(args);
 154
 155         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 156         local_irq_restore(flags);
 157 }
 158 EXPORT_SYMBOL_GPL(__trace_note_message);
 159
 160 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 161                          pid_t pid)
 162 {
 163         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 164                 return 1;
 165         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 166                 return 1;
 167         if (bt->pid && pid != bt->pid)
 168                 return 1;
 169
 170         return 0;
 171 }
 172
 173 /*
 174  * Data direction bit lookup
 175  */
 176 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 177                                  BLK_TC_ACT(BLK_TC_WRITE) };
 178
 179 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 180
 181 /* The ilog2() calls fall out because they're constant */
 182 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 183           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 184
 185 /*
 186  * The worker for the various blk_add_trace*() types. Fills out a
 187  * blk_io_trace structure and places it in a per-cpu subbuffer.
 188  */
 189 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 190                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 191 {
 192         struct task_struct *tsk = current;
 193         struct ring_buffer_event *event = NULL;
 194         struct ring_buffer *buffer = NULL;
 195         struct blk_io_trace *t;
 196         unsigned long flags = 0;
 197         unsigned long *sequence;
 198         pid_t pid;
 199         int cpu, pc = 0;
 200         bool blk_tracer = blk_tracer_enabled;
 201
 202         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 203                 return;
 204
 205         what |= ddir_act[rw & WRITE];
 206         what |= MASK_TC_BIT(rw, SYNC);
 207         what |= MASK_TC_BIT(rw, RAHEAD);
 208         what |= MASK_TC_BIT(rw, META);
 209         what |= MASK_TC_BIT(rw, DISCARD);
 210         what |= MASK_TC_BIT(rw, FLUSH);
 211         what |= MASK_TC_BIT(rw, FUA);
 212
 213         pid = tsk->pid;
 214         if (act_log_check(bt, what, sector, pid))
 215                 return;
 216         cpu = raw_smp_processor_id();
 217
 218         if (blk_tracer) {
 219                 tracing_record_cmdline(current);
 220
 221                 buffer = blk_tr->buffer;
 222                 pc = preempt_count();
 223                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 224                                                   sizeof(*t) + pdu_len,
 225                                                   0, pc);
 226                 if (!event)
 227                         return;
 228                 t = ring_buffer_event_data(event);
 229                 goto record_it;
 230         }
 231
 232         /*
 233          * A word about the locking here - we disable interrupts to reserve
 234          * some space in the relay per-cpu buffer, to prevent an irq
 235          * from coming in and stepping on our toes.
 236          */
 237         local_irq_save(flags);
 238
 239         if (unlikely(tsk->btrace_seq != blktrace_seq))
 240                 trace_note_tsk(bt, tsk);
 241
 242         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 243         if (t) {
 244                 sequence = per_cpu_ptr(bt->sequence, cpu);
 245
 246                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 247                 t->sequence = ++(*sequence);
 248                 t->time = ktime_to_ns(ktime_get());
 249 record_it:
 250                 /*
 251                  * These two are not needed in ftrace as they are in the
 252                  * generic trace_entry, filled by tracing_generic_entry_update,
 253                  * but for the trace_event->bin() synthesizer benefit we do it
 254                  * here too.
 255                  */
 256                 t->cpu = cpu;
 257                 t->pid = pid;
 258
 259                 t->sector = sector;
 260                 t->bytes = bytes;
 261                 t->action = what;
 262                 t->device = bt->dev;
 263                 t->error = error;
 264                 t->pdu_len = pdu_len;
 265
 266                 if (pdu_len)
 267                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 268
 269                 if (blk_tracer) {
 270                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 271                         return;
 272                 }
 273         }
 274
 275         local_irq_restore(flags);
 276 }
 277
 278 static struct dentry *blk_tree_root;
 279 static DEFINE_MUTEX(blk_tree_mutex);
 280
 281 static void blk_trace_free(struct blk_trace *bt)
 282 {
 283         debugfs_remove(bt->msg_file);
 284         debugfs_remove(bt->dropped_file);
 285         relay_close(bt->rchan);
 286         debugfs_remove(bt->dir);
 287         free_percpu(bt->sequence);
 288         free_percpu(bt->msg_data);
 289         kfree(bt);
 290 }
 291
 292 static void blk_trace_cleanup(struct blk_trace *bt)
 293 {
 294         blk_trace_free(bt);
 295         if (atomic_dec_and_test(&blk_probes_ref))
 296                 blk_unregister_tracepoints();
 297 }
 298
 299 static int __blk_trace_remove(struct request_queue *q)
 300 {
 301         struct blk_trace *bt;
 302
 303         bt = xchg(&q->blk_trace, NULL);
 304         if (!bt)
 305                 return -EINVAL;
 306
 307         if (bt->trace_state != Blktrace_running)
 308                 blk_trace_cleanup(bt);
 309
 310         return 0;
 311 }
 312
 313 int blk_trace_remove(struct request_queue *q)
 314 {
 315         int ret;
 316
 317         mutex_lock(&q->blk_trace_mutex);
 318         ret = __blk_trace_remove(q);
 319         mutex_unlock(&q->blk_trace_mutex);
 320
 321         return ret;
 322 }
 323 EXPORT_SYMBOL_GPL(blk_trace_remove);
 324
 325 static int blk_dropped_open(struct inode *inode, struct file *filp)
 326 {
 327         filp->private_data = inode->i_private;
 328
 329         return 0;
 330 }
 331
 332 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 333                                 size_t count, loff_t *ppos)
 334 {
 335         struct blk_trace *bt = filp->private_data;
 336         char buf[16];
 337
 338         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 339
 340         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 341 }
 342
 343 static const struct file_operations blk_dropped_fops = {
 344         .owner =        THIS_MODULE,
 345         .open =         blk_dropped_open,
 346         .read =         blk_dropped_read,
 347         .llseek =       default_llseek,
 348 };
 349
 350 static int blk_msg_open(struct inode *inode, struct file *filp)
 351 {
 352         filp->private_data = inode->i_private;
 353
 354         return 0;
 355 }
 356
 357 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 358                                 size_t count, loff_t *ppos)
 359 {
 360         char *msg;
 361         struct blk_trace *bt;
 362
 363         if (count >= BLK_TN_MAX_MSG)
 364                 return -EINVAL;
 365
 366         msg = kmalloc(count + 1, GFP_KERNEL);
 367         if (msg == NULL)
 368                 return -ENOMEM;
 369
 370         if (copy_from_user(msg, buffer, count)) {
 371                 kfree(msg);
 372                 return -EFAULT;
 373         }
 374
 375         msg[count] = '\0';
 376         bt = filp->private_data;
 377         __trace_note_message(bt, "%s", msg);
 378         kfree(msg);
 379
 380         return count;
 381 }
 382
 383 static const struct file_operations blk_msg_fops = {
 384         .owner =        THIS_MODULE,
 385         .open =         blk_msg_open,
 386         .write =        blk_msg_write,
 387         .llseek =       noop_llseek,
 388 };
 389
 390 /*
 391  * Keep track of how many times we encountered a full subbuffer, to aid
 392  * the user space app in telling how many lost events there were.
 393  */
 394 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 395                                      void *prev_subbuf, size_t prev_padding)
 396 {
 397         struct blk_trace *bt;
 398
 399         if (!relay_buf_full(buf))
 400                 return 1;
 401
 402         bt = buf->chan->private_data;
 403         atomic_inc(&bt->dropped);
 404         return 0;
 405 }
 406
 407 static int blk_remove_buf_file_callback(struct dentry *dentry)
 408 {
 409         debugfs_remove(dentry);
 410
 411         return 0;
 412 }
 413
 414 static struct dentry *blk_create_buf_file_callback(const char *filename,
 415                                                    struct dentry *parent,
 416                                                    umode_t mode,
 417                                                    struct rchan_buf *buf,
 418                                                    int *is_global)
 419 {
 420         return debugfs_create_file(filename, mode, parent, buf,
 421                                         &relay_file_operations);
 422 }
 423
 424 static struct rchan_callbacks blk_relay_callbacks = {
 425         .subbuf_start           = blk_subbuf_start_callback,
 426         .create_buf_file        = blk_create_buf_file_callback,
 427         .remove_buf_file        = blk_remove_buf_file_callback,
 428 };
 429
 430 static void blk_trace_setup_lba(struct blk_trace *bt,
 431                                 struct block_device *bdev)
 432 {
 433         struct hd_struct *part = NULL;
 434
 435         if (bdev)
 436                 part = bdev->bd_part;
 437
 438         if (part) {
 439                 bt->start_lba = part->start_sect;
 440                 bt->end_lba = part->start_sect + part->nr_sects;
 441         } else {
 442                 bt->start_lba = 0;
 443                 bt->end_lba = -1ULL;
 444         }
 445 }
 446
 447 /*
 448  * Setup everything required to start tracing
 449  */
 450 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 451                        struct block_device *bdev,
 452                        struct blk_user_trace_setup *buts)
 453 {
 454         struct blk_trace *old_bt, *bt = NULL;
 455         struct dentry *dir = NULL;
 456         int ret, i;
 457
 458         if (!buts->buf_size || !buts->buf_nr)
 459                 return -EINVAL;
 460
 461         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 462         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 463
 464         /*
 465          * some device names have larger paths - convert the slashes
 466          * to underscores for this to work as expected
 467          */
 468         for (i = 0; i < strlen(buts->name); i++)
 469                 if (buts->name[i] == '/')
 470                         buts->name[i] = '_';
 471
 472         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 473         if (!bt)
 474                 return -ENOMEM;
 475
 476         ret = -ENOMEM;
 477         bt->sequence = alloc_percpu(unsigned long);
 478         if (!bt->sequence)
 479                 goto err;
 480
 481         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 482         if (!bt->msg_data)
 483                 goto err;
 484
 485         ret = -ENOENT;
 486
 487         mutex_lock(&blk_tree_mutex);
 488         if (!blk_tree_root) {
 489                 blk_tree_root = debugfs_create_dir("block", NULL);
 490                 if (!blk_tree_root) {
 491                         mutex_unlock(&blk_tree_mutex);
 492                         goto err;
 493                 }
 494         }
 495         mutex_unlock(&blk_tree_mutex);
 496
 497         dir = debugfs_create_dir(buts->name, blk_tree_root);
 498
 499         if (!dir)
 500                 goto err;
 501
 502         bt->dir = dir;
 503         bt->dev = dev;
 504         atomic_set(&bt->dropped, 0);
 505
 506         ret = -EIO;
 507         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 508                                                &blk_dropped_fops);
 509         if (!bt->dropped_file)
 510                 goto err;
 511
 512         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 513         if (!bt->msg_file)
 514                 goto err;
 515
 516         bt->rchan = relay_open("trace", dir, buts->buf_size,
 517                                 buts->buf_nr, &blk_relay_callbacks, bt);
 518         if (!bt->rchan)
 519                 goto err;
 520
 521         bt->act_mask = buts->act_mask;
 522         if (!bt->act_mask)
 523                 bt->act_mask = (u16) -1;
 524
 525         blk_trace_setup_lba(bt, bdev);
 526
 527         /* overwrite with user settings */
 528         if (buts->start_lba)
 529                 bt->start_lba = buts->start_lba;
 530         if (buts->end_lba)
 531                 bt->end_lba = buts->end_lba;
 532
 533         bt->pid = buts->pid;
 534         bt->trace_state = Blktrace_setup;
 535
 536         ret = -EBUSY;
 537         old_bt = xchg(&q->blk_trace, bt);
 538         if (old_bt) {
 539                 (void) xchg(&q->blk_trace, old_bt);
 540                 goto err;
 541         }
 542
 543         if (atomic_inc_return(&blk_probes_ref) == 1)
 544                 blk_register_tracepoints();
 545
 546         return 0;
 547 err:
 548         blk_trace_free(bt);
 549         return ret;
 550 }
 551
 552 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 553                              struct block_device *bdev, char __user *arg)
 554 {
 555         struct blk_user_trace_setup buts;
 556         int ret;
 557
 558         ret = copy_from_user(&buts, arg, sizeof(buts));
 559         if (ret)
 560                 return -EFAULT;
 561
 562         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 563         if (ret)
 564                 return ret;
 565
 566         if (copy_to_user(arg, &buts, sizeof(buts))) {
 567                 blk_trace_remove(q);
 568                 return -EFAULT;
 569         }
 570         return 0;
 571 }
 572
 573 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 574                     struct block_device *bdev,
 575                     char __user *arg)
 576 {
 577         int ret;
 578
 579         mutex_lock(&q->blk_trace_mutex);
 580         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 581         mutex_unlock(&q->blk_trace_mutex);
 582
 583         return ret;
 584 }
 585 EXPORT_SYMBOL_GPL(blk_trace_setup);
 586
 587 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 588 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 589                                   dev_t dev, struct block_device *bdev,
 590                                   char __user *arg)
 591 {
 592         struct blk_user_trace_setup buts;
 593         struct compat_blk_user_trace_setup cbuts;
 594         int ret;
 595
 596         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 597                 return -EFAULT;
 598
 599         buts = (struct blk_user_trace_setup) {
 600                 .act_mask = cbuts.act_mask,
 601                 .buf_size = cbuts.buf_size,
 602                 .buf_nr = cbuts.buf_nr,
 603                 .start_lba = cbuts.start_lba,
 604                 .end_lba = cbuts.end_lba,
 605                 .pid = cbuts.pid,
 606         };
 607         memcpy(&buts.name, &cbuts.name, 32);
 608
 609         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 610         if (ret)
 611                 return ret;
 612
 613         if (copy_to_user(arg, &buts.name, 32)) {
 614                 blk_trace_remove(q);
 615                 return -EFAULT;
 616         }
 617
 618         return 0;
 619 }
 620 #endif
 621
 622 static int __blk_trace_startstop(struct request_queue *q, int start)
 623 {
 624         int ret;
 625         struct blk_trace *bt = q->blk_trace;
 626
 627         if (bt == NULL)
 628                 return -EINVAL;
 629
 630         /*
 631          * For starting a trace, we can transition from a setup or stopped
 632          * trace. For stopping a trace, the state must be running
 633          */
 634         ret = -EINVAL;
 635         if (start) {
 636                 if (bt->trace_state == Blktrace_setup ||
 637                     bt->trace_state == Blktrace_stopped) {
 638                         blktrace_seq++;
 639                         smp_mb();
 640                         bt->trace_state = Blktrace_running;
 641
 642                         trace_note_time(bt);
 643                         ret = 0;
 644                 }
 645         } else {
 646                 if (bt->trace_state == Blktrace_running) {
 647                         bt->trace_state = Blktrace_stopped;
 648                         relay_flush(bt->rchan);
 649                         ret = 0;
 650                 }
 651         }
 652
 653         return ret;
 654 }
 655
 656 int blk_trace_startstop(struct request_queue *q, int start)
 657 {
 658         int ret;
 659
 660         mutex_lock(&q->blk_trace_mutex);
 661         ret = __blk_trace_startstop(q, start);
 662         mutex_unlock(&q->blk_trace_mutex);
 663
 664         return ret;
 665 }
 666 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 667
 668 /*
 669  * When reading or writing the blktrace sysfs files, the references to the
 670  * opened sysfs or device files should prevent the underlying block device
 671  * from being removed. So no further delete protection is really needed.
 672  */
 673
 674 /**
 675  * blk_trace_ioctl: - handle the ioctls associated with tracing
 676  * @bdev:       the block device
 677  * @cmd:        the ioctl cmd
 678  * @arg:        the argument data, if any
 679  *
 680  **/
 681 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 682 {
 683         struct request_queue *q;
 684         int ret, start = 0;
 685         char b[BDEVNAME_SIZE];
 686
 687         q = bdev_get_queue(bdev);
 688         if (!q)
 689                 return -ENXIO;
 690
 691         mutex_lock(&q->blk_trace_mutex);
 692
 693         switch (cmd) {
 694         case BLKTRACESETUP:
 695                 bdevname(bdev, b);
 696                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 697                 break;
 698 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 699         case BLKTRACESETUP32:
 700                 bdevname(bdev, b);
 701                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 702                 break;
 703 #endif
 704         case BLKTRACESTART:
 705                 start = 1;
 706         case BLKTRACESTOP:
 707                 ret = __blk_trace_startstop(q, start);
 708                 break;
 709         case BLKTRACETEARDOWN:
 710                 ret = __blk_trace_remove(q);
 711                 break;
 712         default:
 713                 ret = -ENOTTY;
 714                 break;
 715         }
 716
 717         mutex_unlock(&q->blk_trace_mutex);
 718         return ret;
 719 }
 720
 721 /**
 722  * blk_trace_shutdown: - stop and cleanup trace structures
 723  * @q:    the request queue associated with the device
 724  *
 725  **/
 726 void blk_trace_shutdown(struct request_queue *q)
 727 {
 728         mutex_lock(&q->blk_trace_mutex);
 729
 730         if (q->blk_trace) {
 731                 __blk_trace_startstop(q, 0);
 732                 __blk_trace_remove(q);
 733         }
 734
 735         mutex_unlock(&q->blk_trace_mutex);
 736 }
 737
 738 /*
 739  * blktrace probes
 740  */
 741
 742 /**
 743  * blk_add_trace_rq - Add a trace for a request oriented action
 744  * @q:          queue the io is for
 745  * @rq:         the source request
 746  * @nr_bytes:   number of completed bytes
 747  * @what:       the action
 748  *
 749  * Description:
 750  *     Records an action against a request. Will log the bio offset + size.
 751  *
 752  **/
 753 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 754                              unsigned int nr_bytes, u32 what)
 755 {
 756         struct blk_trace *bt = q->blk_trace;
 757
 758         if (likely(!bt))
 759                 return;
 760
 761         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 762                 what |= BLK_TC_ACT(BLK_TC_PC);
 763                 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
 764                                 what, rq->errors, rq->cmd_len, rq->cmd);
 765         } else  {
 766                 what |= BLK_TC_ACT(BLK_TC_FS);
 767                 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
 768                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 769         }
 770 }
 771
 772 static void blk_add_trace_rq_abort(void *ignore,
 773                                    struct request_queue *q, struct request *rq)
 774 {
 775         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
 776 }
 777
 778 static void blk_add_trace_rq_insert(void *ignore,
 779                                     struct request_queue *q, struct request *rq)
 780 {
 781         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
 782 }
 783
 784 static void blk_add_trace_rq_issue(void *ignore,
 785                                    struct request_queue *q, struct request *rq)
 786 {
 787         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
 788 }
 789
 790 static void blk_add_trace_rq_requeue(void *ignore,
 791                                      struct request_queue *q,
 792                                      struct request *rq)
 793 {
 794         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 795 }
 796
 797 static void blk_add_trace_rq_complete(void *ignore,
 798                                       struct request_queue *q,
 799                                       struct request *rq,
 800                                       unsigned int nr_bytes)
 801 {
 802         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 803 }
 804
 805 /**
 806  * blk_add_trace_bio - Add a trace for a bio oriented action
 807  * @q:          queue the io is for
 808  * @bio:        the source bio
 809  * @what:       the action
 810  * @error:      error, if any
 811  *
 812  * Description:
 813  *     Records an action against a bio. Will log the bio offset + size.
 814  *
 815  **/
 816 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 817                               u32 what, int error)
 818 {
 819         struct blk_trace *bt = q->blk_trace;
 820
 821         if (likely(!bt))
 822                 return;
 823
 824         if (!error && !bio_flagged(bio, BIO_UPTODATE))
 825                 error = EIO;
 826
 827         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 828                         error, 0, NULL);
 829 }
 830
 831 static void blk_add_trace_bio_bounce(void *ignore,
 832                                      struct request_queue *q, struct bio *bio)
 833 {
 834         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 835 }
 836
 837 static void blk_add_trace_bio_complete(void *ignore,
 838                                        struct request_queue *q, struct bio *bio,
 839                                        int error)
 840 {
 841         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 842 }
 843
 844 static void blk_add_trace_bio_backmerge(void *ignore,
 845                                         struct request_queue *q,
 846                                         struct bio *bio)
 847 {
 848         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 849 }
 850
 851 static void blk_add_trace_bio_frontmerge(void *ignore,
 852                                          struct request_queue *q,
 853                                          struct bio *bio)
 854 {
 855         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 856 }
 857
 858 static void blk_add_trace_bio_queue(void *ignore,
 859                                     struct request_queue *q, struct bio *bio)
 860 {
 861         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 862 }
 863
 864 static void blk_add_trace_getrq(void *ignore,
 865                                 struct request_queue *q,
 866                                 struct bio *bio, int rw)
 867 {
 868         if (bio)
 869                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 870         else {
 871                 struct blk_trace *bt = q->blk_trace;
 872
 873                 if (bt)
 874                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 875         }
 876 }
 877
 878
 879 static void blk_add_trace_sleeprq(void *ignore,
 880                                   struct request_queue *q,
 881                                   struct bio *bio, int rw)
 882 {
 883         if (bio)
 884                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 885         else {
 886                 struct blk_trace *bt = q->blk_trace;
 887
 888                 if (bt)
 889                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 890                                         0, 0, NULL);
 891         }
 892 }
 893
 894 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 895 {
 896         struct blk_trace *bt = q->blk_trace;
 897
 898         if (bt)
 899                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 900 }
 901
 902 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 903                                     unsigned int depth, bool explicit)
 904 {
 905         struct blk_trace *bt = q->blk_trace;
 906
 907         if (bt) {
 908                 __be64 rpdu = cpu_to_be64(depth);
 909                 u32 what;
 910
 911                 if (explicit)
 912                         what = BLK_TA_UNPLUG_IO;
 913                 else
 914                         what = BLK_TA_UNPLUG_TIMER;
 915
 916                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 917         }
 918 }
 919
 920 static void blk_add_trace_split(void *ignore,
 921                                 struct request_queue *q, struct bio *bio,
 922                                 unsigned int pdu)
 923 {
 924         struct blk_trace *bt = q->blk_trace;
 925
 926         if (bt) {
 927                 __be64 rpdu = cpu_to_be64(pdu);
 928
 929                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 930                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 931                                 sizeof(rpdu), &rpdu);
 932         }
 933 }
 934
 935 /**
 936  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 937  * @ignore:     trace callback data parameter (not used)
 938  * @q:          queue the io is for
 939  * @bio:        the source bio
 940  * @dev:        target device
 941  * @from:       source sector
 942  *
 943  * Description:
 944  *     Device mapper or raid target sometimes need to split a bio because
 945  *     it spans a stripe (or similar). Add a trace for that action.
 946  *
 947  **/
 948 static void blk_add_trace_bio_remap(void *ignore,
 949                                     struct request_queue *q, struct bio *bio,
 950                                     dev_t dev, sector_t from)
 951 {
 952         struct blk_trace *bt = q->blk_trace;
 953         struct blk_io_trace_remap r;
 954
 955         if (likely(!bt))
 956                 return;
 957
 958         r.device_from = cpu_to_be32(dev);
 959         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 960         r.sector_from = cpu_to_be64(from);
 961
 962         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 963                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 964                         sizeof(r), &r);
 965 }
 966
 967 /**
 968  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 969  * @ignore:     trace callback data parameter (not used)
 970  * @q:          queue the io is for
 971  * @rq:         the source request
 972  * @dev:        target device
 973  * @from:       source sector
 974  *
 975  * Description:
 976  *     Device mapper remaps request to other devices.
 977  *     Add a trace for that action.
 978  *
 979  **/
 980 static void blk_add_trace_rq_remap(void *ignore,
 981                                    struct request_queue *q,
 982                                    struct request *rq, dev_t dev,
 983                                    sector_t from)
 984 {
 985         struct blk_trace *bt = q->blk_trace;
 986         struct blk_io_trace_remap r;
 987
 988         if (likely(!bt))
 989                 return;
 990
 991         r.device_from = cpu_to_be32(dev);
 992         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 993         r.sector_from = cpu_to_be64(from);
 994
 995         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 996                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 997                         sizeof(r), &r);
 998 }
 999
1000 /**
1001  * blk_add_driver_data - Add binary message with driver-specific data
1002  * @q:          queue the io is for
1003  * @rq:         io request
1004  * @data:       driver-specific data
1005  * @len:        length of driver-specific data
1006  *
1007  * Description:
1008  *     Some drivers might want to write driver-specific data per request.
1009  *
1010  **/
1011 void blk_add_driver_data(struct request_queue *q,
1012                          struct request *rq,
1013                          void *data, size_t len)
1014 {
1015         struct blk_trace *bt = q->blk_trace;
1016
1017         if (likely(!bt))
1018                 return;
1019
1020         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1021                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
1022                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1023         else
1024                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
1025                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1026 }
1027 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1028
1029 static void blk_register_tracepoints(void)
1030 {
1031         int ret;
1032
1033         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1034         WARN_ON(ret);
1035         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1036         WARN_ON(ret);
1037         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1038         WARN_ON(ret);
1039         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1040         WARN_ON(ret);
1041         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1042         WARN_ON(ret);
1043         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1044         WARN_ON(ret);
1045         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1046         WARN_ON(ret);
1047         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1048         WARN_ON(ret);
1049         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1050         WARN_ON(ret);
1051         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1052         WARN_ON(ret);
1053         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1054         WARN_ON(ret);
1055         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1056         WARN_ON(ret);
1057         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1058         WARN_ON(ret);
1059         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1060         WARN_ON(ret);
1061         ret = register_trace_block_split(blk_add_trace_split, NULL);
1062         WARN_ON(ret);
1063         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1064         WARN_ON(ret);
1065         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1066         WARN_ON(ret);
1067 }
1068
1069 static void blk_unregister_tracepoints(void)
1070 {
1071         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1072         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1073         unregister_trace_block_split(blk_add_trace_split, NULL);
1074         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1075         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1076         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1077         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1078         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1079         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1080         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1081         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1082         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1083         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1084         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1085         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1086         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1087         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1088
1089         tracepoint_synchronize_unregister();
1090 }
1091
1092 /*
1093  * struct blk_io_tracer formatting routines
1094  */
1095
1096 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1097 {
1098         int i = 0;
1099         int tc = t->action >> BLK_TC_SHIFT;
1100
1101         if (t->action == BLK_TN_MESSAGE) {
1102                 rwbs[i++] = 'N';
1103                 goto out;
1104         }
1105
1106         if (tc & BLK_TC_FLUSH)
1107                 rwbs[i++] = 'F';
1108
1109         if (tc & BLK_TC_DISCARD)
1110                 rwbs[i++] = 'D';
1111         else if (tc & BLK_TC_WRITE)
1112                 rwbs[i++] = 'W';
1113         else if (t->bytes)
1114                 rwbs[i++] = 'R';
1115         else
1116                 rwbs[i++] = 'N';
1117
1118         if (tc & BLK_TC_FUA)
1119                 rwbs[i++] = 'F';
1120         if (tc & BLK_TC_AHEAD)
1121                 rwbs[i++] = 'A';
1122         if (tc & BLK_TC_SYNC)
1123                 rwbs[i++] = 'S';
1124         if (tc & BLK_TC_META)
1125                 rwbs[i++] = 'M';
1126 out:
1127         rwbs[i] = '\0';
1128 }
1129
1130 static inline
1131 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1132 {
1133         return (const struct blk_io_trace *)ent;
1134 }
1135
1136 static inline const void *pdu_start(const struct trace_entry *ent)
1137 {
1138         return te_blk_io_trace(ent) + 1;
1139 }
1140
1141 static inline u32 t_action(const struct trace_entry *ent)
1142 {
1143         return te_blk_io_trace(ent)->action;
1144 }
1145
1146 static inline u32 t_bytes(const struct trace_entry *ent)
1147 {
1148         return te_blk_io_trace(ent)->bytes;
1149 }
1150
1151 static inline u32 t_sec(const struct trace_entry *ent)
1152 {
1153         return te_blk_io_trace(ent)->bytes >> 9;
1154 }
1155
1156 static inline unsigned long long t_sector(const struct trace_entry *ent)
1157 {
1158         return te_blk_io_trace(ent)->sector;
1159 }
1160
1161 static inline __u16 t_error(const struct trace_entry *ent)
1162 {
1163         return te_blk_io_trace(ent)->error;
1164 }
1165
1166 static __u64 get_pdu_int(const struct trace_entry *ent)
1167 {
1168         const __u64 *val = pdu_start(ent);
1169         return be64_to_cpu(*val);
1170 }
1171
1172 static void get_pdu_remap(const struct trace_entry *ent,
1173                           struct blk_io_trace_remap *r)
1174 {
1175         const struct blk_io_trace_remap *__r = pdu_start(ent);
1176         __u64 sector_from = __r->sector_from;
1177
1178         r->device_from = be32_to_cpu(__r->device_from);
1179         r->device_to   = be32_to_cpu(__r->device_to);
1180         r->sector_from = be64_to_cpu(sector_from);
1181 }
1182
1183 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1184
1185 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1186 {
1187         char rwbs[RWBS_LEN];
1188         unsigned long long ts  = iter->ts;
1189         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1190         unsigned secs          = (unsigned long)ts;
1191         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1192
1193         fill_rwbs(rwbs, t);
1194
1195         return trace_seq_printf(&iter->seq,
1196                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1197                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1198                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1199 }
1200
1201 static int blk_log_action(struct trace_iterator *iter, const char *act)
1202 {
1203         char rwbs[RWBS_LEN];
1204         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1205
1206         fill_rwbs(rwbs, t);
1207         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1208                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1209 }
1210
1211 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1212 {
1213         const unsigned char *pdu_buf;
1214         int pdu_len;
1215         int i, end, ret;
1216
1217         pdu_buf = pdu_start(ent);
1218         pdu_len = te_blk_io_trace(ent)->pdu_len;
1219
1220         if (!pdu_len)
1221                 return 1;
1222
1223         /* find the last zero that needs to be printed */
1224         for (end = pdu_len - 1; end >= 0; end--)
1225                 if (pdu_buf[end])
1226                         break;
1227         end++;
1228
1229         if (!trace_seq_putc(s, '('))
1230                 return 0;
1231
1232         for (i = 0; i < pdu_len; i++) {
1233
1234                 ret = trace_seq_printf(s, "%s%02x",
1235                                        i == 0 ? "" : " ", pdu_buf[i]);
1236                 if (!ret)
1237                         return ret;
1238
1239                 /*
1240                  * stop when the rest is just zeroes and indicate so
1241                  * with a ".." appended
1242                  */
1243                 if (i == end && end != pdu_len - 1)
1244                         return trace_seq_puts(s, " ..) ");
1245         }
1246
1247         return trace_seq_puts(s, ") ");
1248 }
1249
1250 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1251 {
1252         char cmd[TASK_COMM_LEN];
1253
1254         trace_find_cmdline(ent->pid, cmd);
1255
1256         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1257                 int ret;
1258
1259                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1260                 if (!ret)
1261                         return 0;
1262                 ret = blk_log_dump_pdu(s, ent);
1263                 if (!ret)
1264                         return 0;
1265                 return trace_seq_printf(s, "[%s]\n", cmd);
1266         } else {
1267                 if (t_sec(ent))
1268                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1269                                                 t_sector(ent), t_sec(ent), cmd);
1270                 return trace_seq_printf(s, "[%s]\n", cmd);
1271         }
1272 }
1273
1274 static int blk_log_with_error(struct trace_seq *s,
1275                               const struct trace_entry *ent)
1276 {
1277         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1278                 int ret;
1279
1280                 ret = blk_log_dump_pdu(s, ent);
1281                 if (ret)
1282                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1283                 return 0;
1284         } else {
1285                 if (t_sec(ent))
1286                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1287                                                 t_sector(ent),
1288                                                 t_sec(ent), t_error(ent));
1289                 return trace_seq_printf(s, "%llu [%d]\n",
1290                                         t_sector(ent), t_error(ent));
1291         }
1292 }
1293
1294 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1295 {
1296         struct blk_io_trace_remap r = { .device_from = 0, };
1297
1298         get_pdu_remap(ent, &r);
1299         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1300                                 t_sector(ent), t_sec(ent),
1301                                 MAJOR(r.device_from), MINOR(r.device_from),
1302                                 (unsigned long long)r.sector_from);
1303 }
1304
1305 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1306 {
1307         char cmd[TASK_COMM_LEN];
1308
1309         trace_find_cmdline(ent->pid, cmd);
1310
1311         return trace_seq_printf(s, "[%s]\n", cmd);
1312 }
1313
1314 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1315 {
1316         char cmd[TASK_COMM_LEN];
1317
1318         trace_find_cmdline(ent->pid, cmd);
1319
1320         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1321 }
1322
1323 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1324 {
1325         char cmd[TASK_COMM_LEN];
1326
1327         trace_find_cmdline(ent->pid, cmd);
1328
1329         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1330                                 get_pdu_int(ent), cmd);
1331 }
1332
1333 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1334 {
1335         int ret;
1336         const struct blk_io_trace *t = te_blk_io_trace(ent);
1337
1338         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1339         if (ret)
1340                 return trace_seq_putc(s, '\n');
1341         return ret;
1342 }
1343
1344 /*
1345  * struct tracer operations
1346  */
1347
1348 static void blk_tracer_print_header(struct seq_file *m)
1349 {
1350         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1351                 return;
1352         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1353                     "#  |     |     |           |   |   |\n");
1354 }
1355
1356 static void blk_tracer_start(struct trace_array *tr)
1357 {
1358         blk_tracer_enabled = true;
1359 }
1360
1361 static int blk_tracer_init(struct trace_array *tr)
1362 {
1363         blk_tr = tr;
1364         blk_tracer_start(tr);
1365         return 0;
1366 }
1367
1368 static void blk_tracer_stop(struct trace_array *tr)
1369 {
1370         blk_tracer_enabled = false;
1371 }
1372
1373 static void blk_tracer_reset(struct trace_array *tr)
1374 {
1375         blk_tracer_stop(tr);
1376 }
1377
1378 static const struct {
1379         const char *act[2];
1380         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1381 } what2act[] = {
1382         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1383         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1384         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1385         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1386         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1387         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1388         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1389         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1390         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1391         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1392         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1393         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1394         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1395         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1396         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1397 };
1398
1399 static enum print_line_t print_one_line(struct trace_iterator *iter,
1400                                         bool classic)
1401 {
1402         struct trace_seq *s = &iter->seq;
1403         const struct blk_io_trace *t;
1404         u16 what;
1405         int ret;
1406         bool long_act;
1407         blk_log_action_t *log_action;
1408
1409         t          = te_blk_io_trace(iter->ent);
1410         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1411         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1412         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1413
1414         if (t->action == BLK_TN_MESSAGE) {
1415                 ret = log_action(iter, long_act ? "message" : "m");
1416                 if (ret)
1417                         ret = blk_log_msg(s, iter->ent);
1418                 goto out;
1419         }
1420
1421         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1422                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1423         else {
1424                 ret = log_action(iter, what2act[what].act[long_act]);
1425                 if (ret)
1426                         ret = what2act[what].print(s, iter->ent);
1427         }
1428 out:
1429         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1430 }
1431
1432 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1433                                                int flags, struct trace_event *event)
1434 {
1435         return print_one_line(iter, false);
1436 }
1437
1438 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1439 {
1440         struct trace_seq *s = &iter->seq;
1441         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1442         const int offset = offsetof(struct blk_io_trace, sector);
1443         struct blk_io_trace old = {
1444                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1445                 .time     = iter->ts,
1446         };
1447
1448         if (!trace_seq_putmem(s, &old, offset))
1449                 return 0;
1450         return trace_seq_putmem(s, &t->sector,
1451                                 sizeof(old) - offset + t->pdu_len);
1452 }
1453
1454 static enum print_line_t
1455 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1456                              struct trace_event *event)
1457 {
1458         return blk_trace_synthesize_old_trace(iter) ?
1459                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1460 }
1461
1462 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1463 {
1464         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1465                 return TRACE_TYPE_UNHANDLED;
1466
1467         return print_one_line(iter, true);
1468 }
1469
1470 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1471 {
1472         /* don't output context-info for blk_classic output */
1473         if (bit == TRACE_BLK_OPT_CLASSIC) {
1474                 if (set)
1475                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1476                 else
1477                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1478         }
1479         return 0;
1480 }
1481
1482 static struct tracer blk_tracer __read_mostly = {
1483         .name           = "blk",
1484         .init           = blk_tracer_init,
1485         .reset          = blk_tracer_reset,
1486         .start          = blk_tracer_start,
1487         .stop           = blk_tracer_stop,
1488         .print_header   = blk_tracer_print_header,
1489         .print_line     = blk_tracer_print_line,
1490         .flags          = &blk_tracer_flags,
1491         .set_flag       = blk_tracer_set_flag,
1492 };
1493
1494 static struct trace_event_functions trace_blk_event_funcs = {
1495         .trace          = blk_trace_event_print,
1496         .binary         = blk_trace_event_print_binary,
1497 };
1498
1499 static struct trace_event trace_blk_event = {
1500         .type           = TRACE_BLK,
1501         .funcs          = &trace_blk_event_funcs,
1502 };
1503
1504 static int __init init_blk_tracer(void)
1505 {
1506         if (!register_ftrace_event(&trace_blk_event)) {
1507                 pr_warning("Warning: could not register block events\n");
1508                 return 1;
1509         }
1510
1511         if (register_tracer(&blk_tracer) != 0) {
1512                 pr_warning("Warning: could not register the block tracer\n");
1513                 unregister_ftrace_event(&trace_blk_event);
1514                 return 1;
1515         }
1516
1517         return 0;
1518 }
1519
1520 device_initcall(init_blk_tracer);
1521
1522 static int blk_trace_remove_queue(struct request_queue *q)
1523 {
1524         struct blk_trace *bt;
1525
1526         bt = xchg(&q->blk_trace, NULL);
1527         if (bt == NULL)
1528                 return -EINVAL;
1529
1530         if (atomic_dec_and_test(&blk_probes_ref))
1531                 blk_unregister_tracepoints();
1532
1533         blk_trace_free(bt);
1534         return 0;
1535 }
1536
1537 /*
1538  * Setup everything required to start tracing
1539  */
1540 static int blk_trace_setup_queue(struct request_queue *q,
1541                                  struct block_device *bdev)
1542 {
1543         struct blk_trace *old_bt, *bt = NULL;
1544         int ret = -ENOMEM;
1545
1546         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1547         if (!bt)
1548                 return -ENOMEM;
1549
1550         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1551         if (!bt->msg_data)
1552                 goto free_bt;
1553
1554         bt->dev = bdev->bd_dev;
1555         bt->act_mask = (u16)-1;
1556
1557         blk_trace_setup_lba(bt, bdev);
1558
1559         old_bt = xchg(&q->blk_trace, bt);
1560         if (old_bt != NULL) {
1561                 (void)xchg(&q->blk_trace, old_bt);
1562                 ret = -EBUSY;
1563                 goto free_bt;
1564         }
1565
1566         if (atomic_inc_return(&blk_probes_ref) == 1)
1567                 blk_register_tracepoints();
1568         return 0;
1569
1570 free_bt:
1571         blk_trace_free(bt);
1572         return ret;
1573 }
1574
1575 /*
1576  * sysfs interface to enable and configure tracing
1577  */
1578
1579 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1580                                          struct device_attribute *attr,
1581                                          char *buf);
1582 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1583                                           struct device_attribute *attr,
1584                                           const char *buf, size_t count);
1585 #define BLK_TRACE_DEVICE_ATTR(_name) \
1586         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1587                     sysfs_blk_trace_attr_show, \
1588                     sysfs_blk_trace_attr_store)
1589
1590 static BLK_TRACE_DEVICE_ATTR(enable);
1591 static BLK_TRACE_DEVICE_ATTR(act_mask);
1592 static BLK_TRACE_DEVICE_ATTR(pid);
1593 static BLK_TRACE_DEVICE_ATTR(start_lba);
1594 static BLK_TRACE_DEVICE_ATTR(end_lba);
1595
1596 static struct attribute *blk_trace_attrs[] = {
1597         &dev_attr_enable.attr,
1598         &dev_attr_act_mask.attr,
1599         &dev_attr_pid.attr,
1600         &dev_attr_start_lba.attr,
1601         &dev_attr_end_lba.attr,
1602         NULL
1603 };
1604
1605 struct attribute_group blk_trace_attr_group = {
1606         .name  = "trace",
1607         .attrs = blk_trace_attrs,
1608 };
1609
1610 static const struct {
1611         int mask;
1612         const char *str;
1613 } mask_maps[] = {
1614         { BLK_TC_READ,          "read"          },
1615         { BLK_TC_WRITE,         "write"         },
1616         { BLK_TC_FLUSH,         "flush"         },
1617         { BLK_TC_SYNC,          "sync"          },
1618         { BLK_TC_QUEUE,         "queue"         },
1619         { BLK_TC_REQUEUE,       "requeue"       },
1620         { BLK_TC_ISSUE,         "issue"         },
1621         { BLK_TC_COMPLETE,      "complete"      },
1622         { BLK_TC_FS,            "fs"            },
1623         { BLK_TC_PC,            "pc"            },
1624         { BLK_TC_AHEAD,         "ahead"         },
1625         { BLK_TC_META,          "meta"          },
1626         { BLK_TC_DISCARD,       "discard"       },
1627         { BLK_TC_DRV_DATA,      "drv_data"      },
1628         { BLK_TC_FUA,           "fua"           },
1629 };
1630
1631 static int blk_trace_str2mask(const char *str)
1632 {
1633         int i;
1634         int mask = 0;
1635         char *buf, *s, *token;
1636
1637         buf = kstrdup(str, GFP_KERNEL);
1638         if (buf == NULL)
1639                 return -ENOMEM;
1640         s = strstrip(buf);
1641
1642         while (1) {
1643                 token = strsep(&s, ",");
1644                 if (token == NULL)
1645                         break;
1646
1647                 if (*token == '\0')
1648                         continue;
1649
1650                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1651                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1652                                 mask |= mask_maps[i].mask;
1653                                 break;
1654                         }
1655                 }
1656                 if (i == ARRAY_SIZE(mask_maps)) {
1657                         mask = -EINVAL;
1658                         break;
1659                 }
1660         }
1661         kfree(buf);
1662
1663         return mask;
1664 }
1665
1666 static ssize_t blk_trace_mask2str(char *buf, int mask)
1667 {
1668         int i;
1669         char *p = buf;
1670
1671         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1672                 if (mask & mask_maps[i].mask) {
1673                         p += sprintf(p, "%s%s",
1674                                     (p == buf) ? "" : ",", mask_maps[i].str);
1675                 }
1676         }
1677         *p++ = '\n';
1678
1679         return p - buf;
1680 }
1681
1682 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1683 {
1684         if (bdev->bd_disk == NULL)
1685                 return NULL;
1686
1687         return bdev_get_queue(bdev);
1688 }
1689
1690 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1691                                          struct device_attribute *attr,
1692                                          char *buf)
1693 {
1694         struct hd_struct *p = dev_to_part(dev);
1695         struct request_queue *q;
1696         struct block_device *bdev;
1697         ssize_t ret = -ENXIO;
1698
1699         bdev = bdget(part_devt(p));
1700         if (bdev == NULL)
1701                 goto out;
1702
1703         q = blk_trace_get_queue(bdev);
1704         if (q == NULL)
1705                 goto out_bdput;
1706
1707         mutex_lock(&q->blk_trace_mutex);
1708
1709         if (attr == &dev_attr_enable) {
1710                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1711                 goto out_unlock_bdev;
1712         }
1713
1714         if (q->blk_trace == NULL)
1715                 ret = sprintf(buf, "disabled\n");
1716         else if (attr == &dev_attr_act_mask)
1717                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1718         else if (attr == &dev_attr_pid)
1719                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1720         else if (attr == &dev_attr_start_lba)
1721                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1722         else if (attr == &dev_attr_end_lba)
1723                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1724
1725 out_unlock_bdev:
1726         mutex_unlock(&q->blk_trace_mutex);
1727 out_bdput:
1728         bdput(bdev);
1729 out:
1730         return ret;
1731 }
1732
1733 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1734                                           struct device_attribute *attr,
1735                                           const char *buf, size_t count)
1736 {
1737         struct block_device *bdev;
1738         struct request_queue *q;
1739         struct hd_struct *p;
1740         u64 value;
1741         ssize_t ret = -EINVAL;
1742
1743         if (count == 0)
1744                 goto out;
1745
1746         if (attr == &dev_attr_act_mask) {
1747                 if (sscanf(buf, "%llx", &value) != 1) {
1748                         /* Assume it is a list of trace category names */
1749                         ret = blk_trace_str2mask(buf);
1750                         if (ret < 0)
1751                                 goto out;
1752                         value = ret;
1753                 }
1754         } else if (sscanf(buf, "%llu", &value) != 1)
1755                 goto out;
1756
1757         ret = -ENXIO;
1758
1759         p = dev_to_part(dev);
1760         bdev = bdget(part_devt(p));
1761         if (bdev == NULL)
1762                 goto out;
1763
1764         q = blk_trace_get_queue(bdev);
1765         if (q == NULL)
1766                 goto out_bdput;
1767
1768         mutex_lock(&q->blk_trace_mutex);
1769
1770         if (attr == &dev_attr_enable) {
1771                 if (value)
1772                         ret = blk_trace_setup_queue(q, bdev);
1773                 else
1774                         ret = blk_trace_remove_queue(q);
1775                 goto out_unlock_bdev;
1776         }
1777
1778         ret = 0;
1779         if (q->blk_trace == NULL)
1780                 ret = blk_trace_setup_queue(q, bdev);
1781
1782         if (ret == 0) {
1783                 if (attr == &dev_attr_act_mask)
1784                         q->blk_trace->act_mask = value;
1785                 else if (attr == &dev_attr_pid)
1786                         q->blk_trace->pid = value;
1787                 else if (attr == &dev_attr_start_lba)
1788                         q->blk_trace->start_lba = value;
1789                 else if (attr == &dev_attr_end_lba)
1790                         q->blk_trace->end_lba = value;
1791         }
1792
1793 out_unlock_bdev:
1794         mutex_unlock(&q->blk_trace_mutex);
1795 out_bdput:
1796         bdput(bdev);
1797 out:
1798         return ret ? ret : count;
1799 }
1800
1801 int blk_trace_init_sysfs(struct device *dev)
1802 {
1803         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1804 }
1805
1806 void blk_trace_remove_sysfs(struct device *dev)
1807 {
1808         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1809 }
1810
1811 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1812
1813 #ifdef CONFIG_EVENT_TRACING
1814
1815 void blk_dump_cmd(char *buf, struct request *rq)
1816 {
1817         int i, end;
1818         int len = rq->cmd_len;
1819         unsigned char *cmd = rq->cmd;
1820
1821         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1822                 buf[0] = '\0';
1823                 return;
1824         }
1825
1826         for (end = len - 1; end >= 0; end--)
1827                 if (cmd[end])
1828                         break;
1829         end++;
1830
1831         for (i = 0; i < len; i++) {
1832                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1833                 if (i == end && end != len - 1) {
1834                         sprintf(buf, " ..");
1835                         break;
1836                 }
1837         }
1838 }
1839
1840 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1841 {
1842         int i = 0;
1843
1844         if (rw & REQ_FLUSH)
1845                 rwbs[i++] = 'F';
1846
1847         if (rw & WRITE)
1848                 rwbs[i++] = 'W';
1849         else if (rw & REQ_DISCARD)
1850                 rwbs[i++] = 'D';
1851         else if (bytes)
1852                 rwbs[i++] = 'R';
1853         else
1854                 rwbs[i++] = 'N';
1855
1856         if (rw & REQ_FUA)
1857                 rwbs[i++] = 'F';
1858         if (rw & REQ_RAHEAD)
1859                 rwbs[i++] = 'A';
1860         if (rw & REQ_SYNC)
1861                 rwbs[i++] = 'S';
1862         if (rw & REQ_META)
1863                 rwbs[i++] = 'M';
1864         if (rw & REQ_SECURE)
1865                 rwbs[i++] = 'E';
1866
1867         rwbs[i] = '\0';
1868 }
1869
1870 #endif /* CONFIG_EVENT_TRACING */
1871