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