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/time.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <trace/events/block.h>
  30
  31 #include "trace_output.h"
  32
  33 #ifdef CONFIG_BLK_DEV_IO_TRACE
  34
  35 static unsigned int blktrace_seq __read_mostly = 1;
  36
  37 static struct trace_array *blk_tr;
  38 static bool blk_tracer_enabled __read_mostly;
  39
  40 /* Select an alternative, minimalistic output than the original one */
  41 #define TRACE_BLK_OPT_CLASSIC   0x1
  42
  43 static struct tracer_opt blk_tracer_opts[] = {
  44         /* Default disable the minimalistic output */
  45         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  46         { }
  47 };
  48
  49 static struct tracer_flags blk_tracer_flags = {
  50         .val  = 0,
  51         .opts = blk_tracer_opts,
  52 };
  53
  54 /* Global reference count of probes */
  55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  56
  57 static void blk_register_tracepoints(void);
  58 static void blk_unregister_tracepoints(void);
  59
  60 /*
  61  * Send out a notify message.
  62  */
  63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  64                        const void *data, size_t len)
  65 {
  66         struct blk_io_trace *t;
  67         struct ring_buffer_event *event = NULL;
  68         struct ring_buffer *buffer = NULL;
  69         int pc = 0;
  70         int cpu = smp_processor_id();
  71         bool blk_tracer = blk_tracer_enabled;
  72
  73         if (blk_tracer) {
  74                 buffer = blk_tr->buffer;
  75                 pc = preempt_count();
  76                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  77                                                   sizeof(*t) + len,
  78                                                   0, pc);
  79                 if (!event)
  80                         return;
  81                 t = ring_buffer_event_data(event);
  82                 goto record_it;
  83         }
  84
  85         if (!bt->rchan)
  86                 return;
  87
  88         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  89         if (t) {
  90                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  91                 t->time = ktime_to_ns(ktime_get());
  92 record_it:
  93                 t->device = bt->dev;
  94                 t->action = action;
  95                 t->pid = pid;
  96                 t->cpu = cpu;
  97                 t->pdu_len = len;
  98                 memcpy((void *) t + sizeof(*t), data, len);
  99
 100                 if (blk_tracer)
 101                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 102         }
 103 }
 104
 105 /*
 106  * Send out a notify for this process, if we haven't done so since a trace
 107  * started
 108  */
 109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
 110 {
 111         tsk->btrace_seq = blktrace_seq;
 112         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
 113 }
 114
 115 static void trace_note_time(struct blk_trace *bt)
 116 {
 117         struct timespec now;
 118         unsigned long flags;
 119         u32 words[2];
 120
 121         getnstimeofday(&now);
 122         words[0] = now.tv_sec;
 123         words[1] = now.tv_nsec;
 124
 125         local_irq_save(flags);
 126         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 127         local_irq_restore(flags);
 128 }
 129
 130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 131 {
 132         int n;
 133         va_list args;
 134         unsigned long flags;
 135         char *buf;
 136
 137         if (unlikely(bt->trace_state != Blktrace_running &&
 138                      !blk_tracer_enabled))
 139                 return;
 140
 141         /*
 142          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 143          * message to the trace.
 144          */
 145         if (!(bt->act_mask & BLK_TC_NOTIFY))
 146                 return;
 147
 148         local_irq_save(flags);
 149         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
 150         va_start(args, fmt);
 151         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 152         va_end(args);
 153
 154         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 155         local_irq_restore(flags);
 156 }
 157 EXPORT_SYMBOL_GPL(__trace_note_message);
 158
 159 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 160                          pid_t pid)
 161 {
 162         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 163                 return 1;
 164         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 165                 return 1;
 166         if (bt->pid && pid != bt->pid)
 167                 return 1;
 168
 169         return 0;
 170 }
 171
 172 /*
 173  * Data direction bit lookup
 174  */
 175 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 176                                  BLK_TC_ACT(BLK_TC_WRITE) };
 177
 178 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 179
 180 /* The ilog2() calls fall out because they're constant */
 181 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 182           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 183
 184 /*
 185  * The worker for the various blk_add_trace*() types. Fills out a
 186  * blk_io_trace structure and places it in a per-cpu subbuffer.
 187  */
 188 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 189                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 190 {
 191         struct task_struct *tsk = current;
 192         struct ring_buffer_event *event = NULL;
 193         struct ring_buffer *buffer = NULL;
 194         struct blk_io_trace *t;
 195         unsigned long flags = 0;
 196         unsigned long *sequence;
 197         pid_t pid;
 198         int cpu, pc = 0;
 199         bool blk_tracer = blk_tracer_enabled;
 200
 201         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 202                 return;
 203
 204         what |= ddir_act[rw & WRITE];
 205         what |= MASK_TC_BIT(rw, SYNC);
 206         what |= MASK_TC_BIT(rw, RAHEAD);
 207         what |= MASK_TC_BIT(rw, META);
 208         what |= MASK_TC_BIT(rw, DISCARD);
 209
 210         pid = tsk->pid;
 211         if (act_log_check(bt, what, sector, pid))
 212                 return;
 213         cpu = raw_smp_processor_id();
 214
 215         if (blk_tracer) {
 216                 tracing_record_cmdline(current);
 217
 218                 buffer = blk_tr->buffer;
 219                 pc = preempt_count();
 220                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 221                                                   sizeof(*t) + pdu_len,
 222                                                   0, pc);
 223                 if (!event)
 224                         return;
 225                 t = ring_buffer_event_data(event);
 226                 goto record_it;
 227         }
 228
 229         /*
 230          * A word about the locking here - we disable interrupts to reserve
 231          * some space in the relay per-cpu buffer, to prevent an irq
 232          * from coming in and stepping on our toes.
 233          */
 234         local_irq_save(flags);
 235
 236         if (unlikely(tsk->btrace_seq != blktrace_seq))
 237                 trace_note_tsk(bt, tsk);
 238
 239         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 240         if (t) {
 241                 sequence = per_cpu_ptr(bt->sequence, cpu);
 242
 243                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 244                 t->sequence = ++(*sequence);
 245                 t->time = ktime_to_ns(ktime_get());
 246 record_it:
 247                 /*
 248                  * These two are not needed in ftrace as they are in the
 249                  * generic trace_entry, filled by tracing_generic_entry_update,
 250                  * but for the trace_event->bin() synthesizer benefit we do it
 251                  * here too.
 252                  */
 253                 t->cpu = cpu;
 254                 t->pid = pid;
 255
 256                 t->sector = sector;
 257                 t->bytes = bytes;
 258                 t->action = what;
 259                 t->device = bt->dev;
 260                 t->error = error;
 261                 t->pdu_len = pdu_len;
 262
 263                 if (pdu_len)
 264                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 265
 266                 if (blk_tracer) {
 267                         trace_buffer_unlock_commit(buffer, event, 0, pc);
 268                         return;
 269                 }
 270         }
 271
 272         local_irq_restore(flags);
 273 }
 274
 275 static struct dentry *blk_tree_root;
 276 static DEFINE_MUTEX(blk_tree_mutex);
 277
 278 static void blk_trace_free(struct blk_trace *bt)
 279 {
 280         debugfs_remove(bt->msg_file);
 281         debugfs_remove(bt->dropped_file);
 282         relay_close(bt->rchan);
 283         debugfs_remove(bt->dir);
 284         free_percpu(bt->sequence);
 285         free_percpu(bt->msg_data);
 286         kfree(bt);
 287 }
 288
 289 static void blk_trace_cleanup(struct blk_trace *bt)
 290 {
 291         blk_trace_free(bt);
 292         if (atomic_dec_and_test(&blk_probes_ref))
 293                 blk_unregister_tracepoints();
 294 }
 295
 296 int blk_trace_remove(struct request_queue *q)
 297 {
 298         struct blk_trace *bt;
 299
 300         bt = xchg(&q->blk_trace, NULL);
 301         if (!bt)
 302                 return -EINVAL;
 303
 304         if (bt->trace_state != Blktrace_running)
 305                 blk_trace_cleanup(bt);
 306
 307         return 0;
 308 }
 309 EXPORT_SYMBOL_GPL(blk_trace_remove);
 310
 311 static int blk_dropped_open(struct inode *inode, struct file *filp)
 312 {
 313         filp->private_data = inode->i_private;
 314
 315         return 0;
 316 }
 317
 318 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 319                                 size_t count, loff_t *ppos)
 320 {
 321         struct blk_trace *bt = filp->private_data;
 322         char buf[16];
 323
 324         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 325
 326         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 327 }
 328
 329 static const struct file_operations blk_dropped_fops = {
 330         .owner =        THIS_MODULE,
 331         .open =         blk_dropped_open,
 332         .read =         blk_dropped_read,
 333         .llseek =       default_llseek,
 334 };
 335
 336 static int blk_msg_open(struct inode *inode, struct file *filp)
 337 {
 338         filp->private_data = inode->i_private;
 339
 340         return 0;
 341 }
 342
 343 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 344                                 size_t count, loff_t *ppos)
 345 {
 346         char *msg;
 347         struct blk_trace *bt;
 348
 349         if (count >= BLK_TN_MAX_MSG)
 350                 return -EINVAL;
 351
 352         msg = kmalloc(count + 1, GFP_KERNEL);
 353         if (msg == NULL)
 354                 return -ENOMEM;
 355
 356         if (copy_from_user(msg, buffer, count)) {
 357                 kfree(msg);
 358                 return -EFAULT;
 359         }
 360
 361         msg[count] = '\0';
 362         bt = filp->private_data;
 363         __trace_note_message(bt, "%s", msg);
 364         kfree(msg);
 365
 366         return count;
 367 }
 368
 369 static const struct file_operations blk_msg_fops = {
 370         .owner =        THIS_MODULE,
 371         .open =         blk_msg_open,
 372         .write =        blk_msg_write,
 373         .llseek =       noop_llseek,
 374 };
 375
 376 /*
 377  * Keep track of how many times we encountered a full subbuffer, to aid
 378  * the user space app in telling how many lost events there were.
 379  */
 380 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 381                                      void *prev_subbuf, size_t prev_padding)
 382 {
 383         struct blk_trace *bt;
 384
 385         if (!relay_buf_full(buf))
 386                 return 1;
 387
 388         bt = buf->chan->private_data;
 389         atomic_inc(&bt->dropped);
 390         return 0;
 391 }
 392
 393 static int blk_remove_buf_file_callback(struct dentry *dentry)
 394 {
 395         debugfs_remove(dentry);
 396
 397         return 0;
 398 }
 399
 400 static struct dentry *blk_create_buf_file_callback(const char *filename,
 401                                                    struct dentry *parent,
 402                                                    int mode,
 403                                                    struct rchan_buf *buf,
 404                                                    int *is_global)
 405 {
 406         return debugfs_create_file(filename, mode, parent, buf,
 407                                         &relay_file_operations);
 408 }
 409
 410 static struct rchan_callbacks blk_relay_callbacks = {
 411         .subbuf_start           = blk_subbuf_start_callback,
 412         .create_buf_file        = blk_create_buf_file_callback,
 413         .remove_buf_file        = blk_remove_buf_file_callback,
 414 };
 415
 416 static void blk_trace_setup_lba(struct blk_trace *bt,
 417                                 struct block_device *bdev)
 418 {
 419         struct hd_struct *part = NULL;
 420
 421         if (bdev)
 422                 part = bdev->bd_part;
 423
 424         if (part) {
 425                 bt->start_lba = part->start_sect;
 426                 bt->end_lba = part->start_sect + part->nr_sects;
 427         } else {
 428                 bt->start_lba = 0;
 429                 bt->end_lba = -1ULL;
 430         }
 431 }
 432
 433 /*
 434  * Setup everything required to start tracing
 435  */
 436 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 437                        struct block_device *bdev,
 438                        struct blk_user_trace_setup *buts)
 439 {
 440         struct blk_trace *old_bt, *bt = NULL;
 441         struct dentry *dir = NULL;
 442         int ret, i;
 443
 444         if (!buts->buf_size || !buts->buf_nr)
 445                 return -EINVAL;
 446
 447         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 448         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 449
 450         /*
 451          * some device names have larger paths - convert the slashes
 452          * to underscores for this to work as expected
 453          */
 454         for (i = 0; i < strlen(buts->name); i++)
 455                 if (buts->name[i] == '/')
 456                         buts->name[i] = '_';
 457
 458         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 459         if (!bt)
 460                 return -ENOMEM;
 461
 462         ret = -ENOMEM;
 463         bt->sequence = alloc_percpu(unsigned long);
 464         if (!bt->sequence)
 465                 goto err;
 466
 467         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 468         if (!bt->msg_data)
 469                 goto err;
 470
 471         ret = -ENOENT;
 472
 473         mutex_lock(&blk_tree_mutex);
 474         if (!blk_tree_root) {
 475                 blk_tree_root = debugfs_create_dir("block", NULL);
 476                 if (!blk_tree_root) {
 477                         mutex_unlock(&blk_tree_mutex);
 478                         goto err;
 479                 }
 480         }
 481         mutex_unlock(&blk_tree_mutex);
 482
 483         dir = debugfs_create_dir(buts->name, blk_tree_root);
 484
 485         if (!dir)
 486                 goto err;
 487
 488         bt->dir = dir;
 489         bt->dev = dev;
 490         atomic_set(&bt->dropped, 0);
 491
 492         ret = -EIO;
 493         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 494                                                &blk_dropped_fops);
 495         if (!bt->dropped_file)
 496                 goto err;
 497
 498         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 499         if (!bt->msg_file)
 500                 goto err;
 501
 502         bt->rchan = relay_open("trace", dir, buts->buf_size,
 503                                 buts->buf_nr, &blk_relay_callbacks, bt);
 504         if (!bt->rchan)
 505                 goto err;
 506
 507         bt->act_mask = buts->act_mask;
 508         if (!bt->act_mask)
 509                 bt->act_mask = (u16) -1;
 510
 511         blk_trace_setup_lba(bt, bdev);
 512
 513         /* overwrite with user settings */
 514         if (buts->start_lba)
 515                 bt->start_lba = buts->start_lba;
 516         if (buts->end_lba)
 517                 bt->end_lba = buts->end_lba;
 518
 519         bt->pid = buts->pid;
 520         bt->trace_state = Blktrace_setup;
 521
 522         ret = -EBUSY;
 523         old_bt = xchg(&q->blk_trace, bt);
 524         if (old_bt) {
 525                 (void) xchg(&q->blk_trace, old_bt);
 526                 goto err;
 527         }
 528
 529         if (atomic_inc_return(&blk_probes_ref) == 1)
 530                 blk_register_tracepoints();
 531
 532         return 0;
 533 err:
 534         blk_trace_free(bt);
 535         return ret;
 536 }
 537
 538 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 539                     struct block_device *bdev,
 540                     char __user *arg)
 541 {
 542         struct blk_user_trace_setup buts;
 543         int ret;
 544
 545         ret = copy_from_user(&buts, arg, sizeof(buts));
 546         if (ret)
 547                 return -EFAULT;
 548
 549         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 550         if (ret)
 551                 return ret;
 552
 553         if (copy_to_user(arg, &buts, sizeof(buts))) {
 554                 blk_trace_remove(q);
 555                 return -EFAULT;
 556         }
 557         return 0;
 558 }
 559 EXPORT_SYMBOL_GPL(blk_trace_setup);
 560
 561 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 562 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 563                                   dev_t dev, struct block_device *bdev,
 564                                   char __user *arg)
 565 {
 566         struct blk_user_trace_setup buts;
 567         struct compat_blk_user_trace_setup cbuts;
 568         int ret;
 569
 570         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 571                 return -EFAULT;
 572
 573         buts = (struct blk_user_trace_setup) {
 574                 .act_mask = cbuts.act_mask,
 575                 .buf_size = cbuts.buf_size,
 576                 .buf_nr = cbuts.buf_nr,
 577                 .start_lba = cbuts.start_lba,
 578                 .end_lba = cbuts.end_lba,
 579                 .pid = cbuts.pid,
 580         };
 581         memcpy(&buts.name, &cbuts.name, 32);
 582
 583         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 584         if (ret)
 585                 return ret;
 586
 587         if (copy_to_user(arg, &buts.name, 32)) {
 588                 blk_trace_remove(q);
 589                 return -EFAULT;
 590         }
 591
 592         return 0;
 593 }
 594 #endif
 595
 596 int blk_trace_startstop(struct request_queue *q, int start)
 597 {
 598         int ret;
 599         struct blk_trace *bt = q->blk_trace;
 600
 601         if (bt == NULL)
 602                 return -EINVAL;
 603
 604         /*
 605          * For starting a trace, we can transition from a setup or stopped
 606          * trace. For stopping a trace, the state must be running
 607          */
 608         ret = -EINVAL;
 609         if (start) {
 610                 if (bt->trace_state == Blktrace_setup ||
 611                     bt->trace_state == Blktrace_stopped) {
 612                         blktrace_seq++;
 613                         smp_mb();
 614                         bt->trace_state = Blktrace_running;
 615
 616                         trace_note_time(bt);
 617                         ret = 0;
 618                 }
 619         } else {
 620                 if (bt->trace_state == Blktrace_running) {
 621                         bt->trace_state = Blktrace_stopped;
 622                         relay_flush(bt->rchan);
 623                         ret = 0;
 624                 }
 625         }
 626
 627         return ret;
 628 }
 629 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 630
 631 /**
 632  * blk_trace_ioctl: - handle the ioctls associated with tracing
 633  * @bdev:       the block device
 634  * @cmd:        the ioctl cmd
 635  * @arg:        the argument data, if any
 636  *
 637  **/
 638 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 639 {
 640         struct request_queue *q;
 641         int ret, start = 0;
 642         char b[BDEVNAME_SIZE];
 643
 644         q = bdev_get_queue(bdev);
 645         if (!q)
 646                 return -ENXIO;
 647
 648         mutex_lock(&bdev->bd_mutex);
 649
 650         switch (cmd) {
 651         case BLKTRACESETUP:
 652                 bdevname(bdev, b);
 653                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 654                 break;
 655 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 656         case BLKTRACESETUP32:
 657                 bdevname(bdev, b);
 658                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 659                 break;
 660 #endif
 661         case BLKTRACESTART:
 662                 start = 1;
 663         case BLKTRACESTOP:
 664                 ret = blk_trace_startstop(q, start);
 665                 break;
 666         case BLKTRACETEARDOWN:
 667                 ret = blk_trace_remove(q);
 668                 break;
 669         default:
 670                 ret = -ENOTTY;
 671                 break;
 672         }
 673
 674         mutex_unlock(&bdev->bd_mutex);
 675         return ret;
 676 }
 677
 678 /**
 679  * blk_trace_shutdown: - stop and cleanup trace structures
 680  * @q:    the request queue associated with the device
 681  *
 682  **/
 683 void blk_trace_shutdown(struct request_queue *q)
 684 {
 685         if (q->blk_trace) {
 686                 blk_trace_startstop(q, 0);
 687                 blk_trace_remove(q);
 688         }
 689 }
 690
 691 /*
 692  * blktrace probes
 693  */
 694
 695 /**
 696  * blk_add_trace_rq - Add a trace for a request oriented action
 697  * @q:          queue the io is for
 698  * @rq:         the source request
 699  * @what:       the action
 700  *
 701  * Description:
 702  *     Records an action against a request. Will log the bio offset + size.
 703  *
 704  **/
 705 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 706                              u32 what)
 707 {
 708         struct blk_trace *bt = q->blk_trace;
 709
 710         if (likely(!bt))
 711                 return;
 712
 713         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 714                 what |= BLK_TC_ACT(BLK_TC_PC);
 715                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
 716                                 what, rq->errors, rq->cmd_len, rq->cmd);
 717         } else  {
 718                 what |= BLK_TC_ACT(BLK_TC_FS);
 719                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 720                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 721         }
 722 }
 723
 724 static void blk_add_trace_rq_abort(void *ignore,
 725                                    struct request_queue *q, struct request *rq)
 726 {
 727         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
 728 }
 729
 730 static void blk_add_trace_rq_insert(void *ignore,
 731                                     struct request_queue *q, struct request *rq)
 732 {
 733         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
 734 }
 735
 736 static void blk_add_trace_rq_issue(void *ignore,
 737                                    struct request_queue *q, struct request *rq)
 738 {
 739         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
 740 }
 741
 742 static void blk_add_trace_rq_requeue(void *ignore,
 743                                      struct request_queue *q,
 744                                      struct request *rq)
 745 {
 746         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
 747 }
 748
 749 static void blk_add_trace_rq_complete(void *ignore,
 750                                       struct request_queue *q,
 751                                       struct request *rq)
 752 {
 753         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
 754 }
 755
 756 /**
 757  * blk_add_trace_bio - Add a trace for a bio oriented action
 758  * @q:          queue the io is for
 759  * @bio:        the source bio
 760  * @what:       the action
 761  * @error:      error, if any
 762  *
 763  * Description:
 764  *     Records an action against a bio. Will log the bio offset + size.
 765  *
 766  **/
 767 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 768                               u32 what, int error)
 769 {
 770         struct blk_trace *bt = q->blk_trace;
 771
 772         if (likely(!bt))
 773                 return;
 774
 775         if (!error && !bio_flagged(bio, BIO_UPTODATE))
 776                 error = EIO;
 777
 778         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
 779                         error, 0, NULL);
 780 }
 781
 782 static void blk_add_trace_bio_bounce(void *ignore,
 783                                      struct request_queue *q, struct bio *bio)
 784 {
 785         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 786 }
 787
 788 static void blk_add_trace_bio_complete(void *ignore,
 789                                        struct request_queue *q, struct bio *bio,
 790                                        int error)
 791 {
 792         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 793 }
 794
 795 static void blk_add_trace_bio_backmerge(void *ignore,
 796                                         struct request_queue *q,
 797                                         struct bio *bio)
 798 {
 799         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 800 }
 801
 802 static void blk_add_trace_bio_frontmerge(void *ignore,
 803                                          struct request_queue *q,
 804                                          struct bio *bio)
 805 {
 806         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 807 }
 808
 809 static void blk_add_trace_bio_queue(void *ignore,
 810                                     struct request_queue *q, struct bio *bio)
 811 {
 812         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 813 }
 814
 815 static void blk_add_trace_getrq(void *ignore,
 816                                 struct request_queue *q,
 817                                 struct bio *bio, int rw)
 818 {
 819         if (bio)
 820                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 821         else {
 822                 struct blk_trace *bt = q->blk_trace;
 823
 824                 if (bt)
 825                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
 826         }
 827 }
 828
 829
 830 static void blk_add_trace_sleeprq(void *ignore,
 831                                   struct request_queue *q,
 832                                   struct bio *bio, int rw)
 833 {
 834         if (bio)
 835                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 836         else {
 837                 struct blk_trace *bt = q->blk_trace;
 838
 839                 if (bt)
 840                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 841                                         0, 0, NULL);
 842         }
 843 }
 844
 845 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 846 {
 847         struct blk_trace *bt = q->blk_trace;
 848
 849         if (bt)
 850                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 851 }
 852
 853 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 854                                     unsigned int depth, bool explicit)
 855 {
 856         struct blk_trace *bt = q->blk_trace;
 857
 858         if (bt) {
 859                 __be64 rpdu = cpu_to_be64(depth);
 860                 u32 what;
 861
 862                 if (explicit)
 863                         what = BLK_TA_UNPLUG_IO;
 864                 else
 865                         what = BLK_TA_UNPLUG_TIMER;
 866
 867                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 868         }
 869 }
 870
 871 static void blk_add_trace_split(void *ignore,
 872                                 struct request_queue *q, struct bio *bio,
 873                                 unsigned int pdu)
 874 {
 875         struct blk_trace *bt = q->blk_trace;
 876
 877         if (bt) {
 878                 __be64 rpdu = cpu_to_be64(pdu);
 879
 880                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 881                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
 882                                 sizeof(rpdu), &rpdu);
 883         }
 884 }
 885
 886 /**
 887  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 888  * @ignore:     trace callback data parameter (not used)
 889  * @q:          queue the io is for
 890  * @bio:        the source bio
 891  * @dev:        target device
 892  * @from:       source sector
 893  *
 894  * Description:
 895  *     Device mapper or raid target sometimes need to split a bio because
 896  *     it spans a stripe (or similar). Add a trace for that action.
 897  *
 898  **/
 899 static void blk_add_trace_bio_remap(void *ignore,
 900                                     struct request_queue *q, struct bio *bio,
 901                                     dev_t dev, sector_t from)
 902 {
 903         struct blk_trace *bt = q->blk_trace;
 904         struct blk_io_trace_remap r;
 905
 906         if (likely(!bt))
 907                 return;
 908
 909         r.device_from = cpu_to_be32(dev);
 910         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 911         r.sector_from = cpu_to_be64(from);
 912
 913         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
 914                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
 915                         sizeof(r), &r);
 916 }
 917
 918 /**
 919  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 920  * @ignore:     trace callback data parameter (not used)
 921  * @q:          queue the io is for
 922  * @rq:         the source request
 923  * @dev:        target device
 924  * @from:       source sector
 925  *
 926  * Description:
 927  *     Device mapper remaps request to other devices.
 928  *     Add a trace for that action.
 929  *
 930  **/
 931 static void blk_add_trace_rq_remap(void *ignore,
 932                                    struct request_queue *q,
 933                                    struct request *rq, dev_t dev,
 934                                    sector_t from)
 935 {
 936         struct blk_trace *bt = q->blk_trace;
 937         struct blk_io_trace_remap r;
 938
 939         if (likely(!bt))
 940                 return;
 941
 942         r.device_from = cpu_to_be32(dev);
 943         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 944         r.sector_from = cpu_to_be64(from);
 945
 946         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 947                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
 948                         sizeof(r), &r);
 949 }
 950
 951 /**
 952  * blk_add_driver_data - Add binary message with driver-specific data
 953  * @q:          queue the io is for
 954  * @rq:         io request
 955  * @data:       driver-specific data
 956  * @len:        length of driver-specific data
 957  *
 958  * Description:
 959  *     Some drivers might want to write driver-specific data per request.
 960  *
 961  **/
 962 void blk_add_driver_data(struct request_queue *q,
 963                          struct request *rq,
 964                          void *data, size_t len)
 965 {
 966         struct blk_trace *bt = q->blk_trace;
 967
 968         if (likely(!bt))
 969                 return;
 970
 971         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 972                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
 973                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 974         else
 975                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
 976                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 977 }
 978 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 979
 980 static void blk_register_tracepoints(void)
 981 {
 982         int ret;
 983
 984         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 985         WARN_ON(ret);
 986         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 987         WARN_ON(ret);
 988         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 989         WARN_ON(ret);
 990         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 991         WARN_ON(ret);
 992         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 993         WARN_ON(ret);
 994         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 995         WARN_ON(ret);
 996         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 997         WARN_ON(ret);
 998         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 999         WARN_ON(ret);
1000         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1001         WARN_ON(ret);
1002         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1003         WARN_ON(ret);
1004         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1005         WARN_ON(ret);
1006         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1007         WARN_ON(ret);
1008         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1009         WARN_ON(ret);
1010         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1011         WARN_ON(ret);
1012         ret = register_trace_block_split(blk_add_trace_split, NULL);
1013         WARN_ON(ret);
1014         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1015         WARN_ON(ret);
1016         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1017         WARN_ON(ret);
1018 }
1019
1020 static void blk_unregister_tracepoints(void)
1021 {
1022         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1023         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1024         unregister_trace_block_split(blk_add_trace_split, NULL);
1025         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1026         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1027         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1028         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1029         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1030         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1031         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1032         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1033         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1034         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1035         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1036         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1037         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1038         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1039
1040         tracepoint_synchronize_unregister();
1041 }
1042
1043 /*
1044  * struct blk_io_tracer formatting routines
1045  */
1046
1047 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1048 {
1049         int i = 0;
1050         int tc = t->action >> BLK_TC_SHIFT;
1051
1052         if (t->action == BLK_TN_MESSAGE) {
1053                 rwbs[i++] = 'N';
1054                 goto out;
1055         }
1056
1057         if (tc & BLK_TC_DISCARD)
1058                 rwbs[i++] = 'D';
1059         else if (tc & BLK_TC_WRITE)
1060                 rwbs[i++] = 'W';
1061         else if (t->bytes)
1062                 rwbs[i++] = 'R';
1063         else
1064                 rwbs[i++] = 'N';
1065
1066         if (tc & BLK_TC_AHEAD)
1067                 rwbs[i++] = 'A';
1068         if (tc & BLK_TC_BARRIER)
1069                 rwbs[i++] = 'B';
1070         if (tc & BLK_TC_SYNC)
1071                 rwbs[i++] = 'S';
1072         if (tc & BLK_TC_META)
1073                 rwbs[i++] = 'M';
1074 out:
1075         rwbs[i] = '\0';
1076 }
1077
1078 static inline
1079 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1080 {
1081         return (const struct blk_io_trace *)ent;
1082 }
1083
1084 static inline const void *pdu_start(const struct trace_entry *ent)
1085 {
1086         return te_blk_io_trace(ent) + 1;
1087 }
1088
1089 static inline u32 t_action(const struct trace_entry *ent)
1090 {
1091         return te_blk_io_trace(ent)->action;
1092 }
1093
1094 static inline u32 t_bytes(const struct trace_entry *ent)
1095 {
1096         return te_blk_io_trace(ent)->bytes;
1097 }
1098
1099 static inline u32 t_sec(const struct trace_entry *ent)
1100 {
1101         return te_blk_io_trace(ent)->bytes >> 9;
1102 }
1103
1104 static inline unsigned long long t_sector(const struct trace_entry *ent)
1105 {
1106         return te_blk_io_trace(ent)->sector;
1107 }
1108
1109 static inline __u16 t_error(const struct trace_entry *ent)
1110 {
1111         return te_blk_io_trace(ent)->error;
1112 }
1113
1114 static __u64 get_pdu_int(const struct trace_entry *ent)
1115 {
1116         const __u64 *val = pdu_start(ent);
1117         return be64_to_cpu(*val);
1118 }
1119
1120 static void get_pdu_remap(const struct trace_entry *ent,
1121                           struct blk_io_trace_remap *r)
1122 {
1123         const struct blk_io_trace_remap *__r = pdu_start(ent);
1124         __u64 sector_from = __r->sector_from;
1125
1126         r->device_from = be32_to_cpu(__r->device_from);
1127         r->device_to   = be32_to_cpu(__r->device_to);
1128         r->sector_from = be64_to_cpu(sector_from);
1129 }
1130
1131 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1132
1133 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1134 {
1135         char rwbs[6];
1136         unsigned long long ts  = iter->ts;
1137         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1138         unsigned secs          = (unsigned long)ts;
1139         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1140
1141         fill_rwbs(rwbs, t);
1142
1143         return trace_seq_printf(&iter->seq,
1144                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1145                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1146                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1147 }
1148
1149 static int blk_log_action(struct trace_iterator *iter, const char *act)
1150 {
1151         char rwbs[6];
1152         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1153
1154         fill_rwbs(rwbs, t);
1155         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1156                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1157 }
1158
1159 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1160 {
1161         const unsigned char *pdu_buf;
1162         int pdu_len;
1163         int i, end, ret;
1164
1165         pdu_buf = pdu_start(ent);
1166         pdu_len = te_blk_io_trace(ent)->pdu_len;
1167
1168         if (!pdu_len)
1169                 return 1;
1170
1171         /* find the last zero that needs to be printed */
1172         for (end = pdu_len - 1; end >= 0; end--)
1173                 if (pdu_buf[end])
1174                         break;
1175         end++;
1176
1177         if (!trace_seq_putc(s, '('))
1178                 return 0;
1179
1180         for (i = 0; i < pdu_len; i++) {
1181
1182                 ret = trace_seq_printf(s, "%s%02x",
1183                                        i == 0 ? "" : " ", pdu_buf[i]);
1184                 if (!ret)
1185                         return ret;
1186
1187                 /*
1188                  * stop when the rest is just zeroes and indicate so
1189                  * with a ".." appended
1190                  */
1191                 if (i == end && end != pdu_len - 1)
1192                         return trace_seq_puts(s, " ..) ");
1193         }
1194
1195         return trace_seq_puts(s, ") ");
1196 }
1197
1198 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1199 {
1200         char cmd[TASK_COMM_LEN];
1201
1202         trace_find_cmdline(ent->pid, cmd);
1203
1204         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1205                 int ret;
1206
1207                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1208                 if (!ret)
1209                         return 0;
1210                 ret = blk_log_dump_pdu(s, ent);
1211                 if (!ret)
1212                         return 0;
1213                 return trace_seq_printf(s, "[%s]\n", cmd);
1214         } else {
1215                 if (t_sec(ent))
1216                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1217                                                 t_sector(ent), t_sec(ent), cmd);
1218                 return trace_seq_printf(s, "[%s]\n", cmd);
1219         }
1220 }
1221
1222 static int blk_log_with_error(struct trace_seq *s,
1223                               const struct trace_entry *ent)
1224 {
1225         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1226                 int ret;
1227
1228                 ret = blk_log_dump_pdu(s, ent);
1229                 if (ret)
1230                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1231                 return 0;
1232         } else {
1233                 if (t_sec(ent))
1234                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1235                                                 t_sector(ent),
1236                                                 t_sec(ent), t_error(ent));
1237                 return trace_seq_printf(s, "%llu [%d]\n",
1238                                         t_sector(ent), t_error(ent));
1239         }
1240 }
1241
1242 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1243 {
1244         struct blk_io_trace_remap r = { .device_from = 0, };
1245
1246         get_pdu_remap(ent, &r);
1247         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1248                                 t_sector(ent), t_sec(ent),
1249                                 MAJOR(r.device_from), MINOR(r.device_from),
1250                                 (unsigned long long)r.sector_from);
1251 }
1252
1253 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1254 {
1255         char cmd[TASK_COMM_LEN];
1256
1257         trace_find_cmdline(ent->pid, cmd);
1258
1259         return trace_seq_printf(s, "[%s]\n", cmd);
1260 }
1261
1262 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1263 {
1264         char cmd[TASK_COMM_LEN];
1265
1266         trace_find_cmdline(ent->pid, cmd);
1267
1268         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1269 }
1270
1271 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1272 {
1273         char cmd[TASK_COMM_LEN];
1274
1275         trace_find_cmdline(ent->pid, cmd);
1276
1277         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1278                                 get_pdu_int(ent), cmd);
1279 }
1280
1281 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1282 {
1283         int ret;
1284         const struct blk_io_trace *t = te_blk_io_trace(ent);
1285
1286         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1287         if (ret)
1288                 return trace_seq_putc(s, '\n');
1289         return ret;
1290 }
1291
1292 /*
1293  * struct tracer operations
1294  */
1295
1296 static void blk_tracer_print_header(struct seq_file *m)
1297 {
1298         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1299                 return;
1300         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1301                     "#  |     |     |           |   |   |\n");
1302 }
1303
1304 static void blk_tracer_start(struct trace_array *tr)
1305 {
1306         blk_tracer_enabled = true;
1307 }
1308
1309 static int blk_tracer_init(struct trace_array *tr)
1310 {
1311         blk_tr = tr;
1312         blk_tracer_start(tr);
1313         return 0;
1314 }
1315
1316 static void blk_tracer_stop(struct trace_array *tr)
1317 {
1318         blk_tracer_enabled = false;
1319 }
1320
1321 static void blk_tracer_reset(struct trace_array *tr)
1322 {
1323         blk_tracer_stop(tr);
1324 }
1325
1326 static const struct {
1327         const char *act[2];
1328         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1329 } what2act[] = {
1330         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1331         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1332         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1333         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1334         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1335         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1336         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1337         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1338         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1339         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1340         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1341         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1342         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1343         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1344         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1345 };
1346
1347 static enum print_line_t print_one_line(struct trace_iterator *iter,
1348                                         bool classic)
1349 {
1350         struct trace_seq *s = &iter->seq;
1351         const struct blk_io_trace *t;
1352         u16 what;
1353         int ret;
1354         bool long_act;
1355         blk_log_action_t *log_action;
1356
1357         t          = te_blk_io_trace(iter->ent);
1358         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1359         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1360         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1361
1362         if (t->action == BLK_TN_MESSAGE) {
1363                 ret = log_action(iter, long_act ? "message" : "m");
1364                 if (ret)
1365                         ret = blk_log_msg(s, iter->ent);
1366                 goto out;
1367         }
1368
1369         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1370                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1371         else {
1372                 ret = log_action(iter, what2act[what].act[long_act]);
1373                 if (ret)
1374                         ret = what2act[what].print(s, iter->ent);
1375         }
1376 out:
1377         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1378 }
1379
1380 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1381                                                int flags, struct trace_event *event)
1382 {
1383         return print_one_line(iter, false);
1384 }
1385
1386 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1387 {
1388         struct trace_seq *s = &iter->seq;
1389         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1390         const int offset = offsetof(struct blk_io_trace, sector);
1391         struct blk_io_trace old = {
1392                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1393                 .time     = iter->ts,
1394         };
1395
1396         if (!trace_seq_putmem(s, &old, offset))
1397                 return 0;
1398         return trace_seq_putmem(s, &t->sector,
1399                                 sizeof(old) - offset + t->pdu_len);
1400 }
1401
1402 static enum print_line_t
1403 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1404                              struct trace_event *event)
1405 {
1406         return blk_trace_synthesize_old_trace(iter) ?
1407                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1408 }
1409
1410 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1411 {
1412         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1413                 return TRACE_TYPE_UNHANDLED;
1414
1415         return print_one_line(iter, true);
1416 }
1417
1418 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1419 {
1420         /* don't output context-info for blk_classic output */
1421         if (bit == TRACE_BLK_OPT_CLASSIC) {
1422                 if (set)
1423                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1424                 else
1425                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1426         }
1427         return 0;
1428 }
1429
1430 static struct tracer blk_tracer __read_mostly = {
1431         .name           = "blk",
1432         .init           = blk_tracer_init,
1433         .reset          = blk_tracer_reset,
1434         .start          = blk_tracer_start,
1435         .stop           = blk_tracer_stop,
1436         .print_header   = blk_tracer_print_header,
1437         .print_line     = blk_tracer_print_line,
1438         .flags          = &blk_tracer_flags,
1439         .set_flag       = blk_tracer_set_flag,
1440 };
1441
1442 static struct trace_event_functions trace_blk_event_funcs = {
1443         .trace          = blk_trace_event_print,
1444         .binary         = blk_trace_event_print_binary,
1445 };
1446
1447 static struct trace_event trace_blk_event = {
1448         .type           = TRACE_BLK,
1449         .funcs          = &trace_blk_event_funcs,
1450 };
1451
1452 static int __init init_blk_tracer(void)
1453 {
1454         if (!register_ftrace_event(&trace_blk_event)) {
1455                 pr_warning("Warning: could not register block events\n");
1456                 return 1;
1457         }
1458
1459         if (register_tracer(&blk_tracer) != 0) {
1460                 pr_warning("Warning: could not register the block tracer\n");
1461                 unregister_ftrace_event(&trace_blk_event);
1462                 return 1;
1463         }
1464
1465         return 0;
1466 }
1467
1468 device_initcall(init_blk_tracer);
1469
1470 static int blk_trace_remove_queue(struct request_queue *q)
1471 {
1472         struct blk_trace *bt;
1473
1474         bt = xchg(&q->blk_trace, NULL);
1475         if (bt == NULL)
1476                 return -EINVAL;
1477
1478         if (atomic_dec_and_test(&blk_probes_ref))
1479                 blk_unregister_tracepoints();
1480
1481         blk_trace_free(bt);
1482         return 0;
1483 }
1484
1485 /*
1486  * Setup everything required to start tracing
1487  */
1488 static int blk_trace_setup_queue(struct request_queue *q,
1489                                  struct block_device *bdev)
1490 {
1491         struct blk_trace *old_bt, *bt = NULL;
1492         int ret = -ENOMEM;
1493
1494         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1495         if (!bt)
1496                 return -ENOMEM;
1497
1498         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1499         if (!bt->msg_data)
1500                 goto free_bt;
1501
1502         bt->dev = bdev->bd_dev;
1503         bt->act_mask = (u16)-1;
1504
1505         blk_trace_setup_lba(bt, bdev);
1506
1507         old_bt = xchg(&q->blk_trace, bt);
1508         if (old_bt != NULL) {
1509                 (void)xchg(&q->blk_trace, old_bt);
1510                 ret = -EBUSY;
1511                 goto free_bt;
1512         }
1513
1514         if (atomic_inc_return(&blk_probes_ref) == 1)
1515                 blk_register_tracepoints();
1516         return 0;
1517
1518 free_bt:
1519         blk_trace_free(bt);
1520         return ret;
1521 }
1522
1523 /*
1524  * sysfs interface to enable and configure tracing
1525  */
1526
1527 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1528                                          struct device_attribute *attr,
1529                                          char *buf);
1530 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1531                                           struct device_attribute *attr,
1532                                           const char *buf, size_t count);
1533 #define BLK_TRACE_DEVICE_ATTR(_name) \
1534         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1535                     sysfs_blk_trace_attr_show, \
1536                     sysfs_blk_trace_attr_store)
1537
1538 static BLK_TRACE_DEVICE_ATTR(enable);
1539 static BLK_TRACE_DEVICE_ATTR(act_mask);
1540 static BLK_TRACE_DEVICE_ATTR(pid);
1541 static BLK_TRACE_DEVICE_ATTR(start_lba);
1542 static BLK_TRACE_DEVICE_ATTR(end_lba);
1543
1544 static struct attribute *blk_trace_attrs[] = {
1545         &dev_attr_enable.attr,
1546         &dev_attr_act_mask.attr,
1547         &dev_attr_pid.attr,
1548         &dev_attr_start_lba.attr,
1549         &dev_attr_end_lba.attr,
1550         NULL
1551 };
1552
1553 struct attribute_group blk_trace_attr_group = {
1554         .name  = "trace",
1555         .attrs = blk_trace_attrs,
1556 };
1557
1558 static const struct {
1559         int mask;
1560         const char *str;
1561 } mask_maps[] = {
1562         { BLK_TC_READ,          "read"          },
1563         { BLK_TC_WRITE,         "write"         },
1564         { BLK_TC_BARRIER,       "barrier"       },
1565         { BLK_TC_SYNC,          "sync"          },
1566         { BLK_TC_QUEUE,         "queue"         },
1567         { BLK_TC_REQUEUE,       "requeue"       },
1568         { BLK_TC_ISSUE,         "issue"         },
1569         { BLK_TC_COMPLETE,      "complete"      },
1570         { BLK_TC_FS,            "fs"            },
1571         { BLK_TC_PC,            "pc"            },
1572         { BLK_TC_AHEAD,         "ahead"         },
1573         { BLK_TC_META,          "meta"          },
1574         { BLK_TC_DISCARD,       "discard"       },
1575         { BLK_TC_DRV_DATA,      "drv_data"      },
1576 };
1577
1578 static int blk_trace_str2mask(const char *str)
1579 {
1580         int i;
1581         int mask = 0;
1582         char *buf, *s, *token;
1583
1584         buf = kstrdup(str, GFP_KERNEL);
1585         if (buf == NULL)
1586                 return -ENOMEM;
1587         s = strstrip(buf);
1588
1589         while (1) {
1590                 token = strsep(&s, ",");
1591                 if (token == NULL)
1592                         break;
1593
1594                 if (*token == '\0')
1595                         continue;
1596
1597                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1598                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1599                                 mask |= mask_maps[i].mask;
1600                                 break;
1601                         }
1602                 }
1603                 if (i == ARRAY_SIZE(mask_maps)) {
1604                         mask = -EINVAL;
1605                         break;
1606                 }
1607         }
1608         kfree(buf);
1609
1610         return mask;
1611 }
1612
1613 static ssize_t blk_trace_mask2str(char *buf, int mask)
1614 {
1615         int i;
1616         char *p = buf;
1617
1618         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1619                 if (mask & mask_maps[i].mask) {
1620                         p += sprintf(p, "%s%s",
1621                                     (p == buf) ? "" : ",", mask_maps[i].str);
1622                 }
1623         }
1624         *p++ = '\n';
1625
1626         return p - buf;
1627 }
1628
1629 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1630 {
1631         if (bdev->bd_disk == NULL)
1632                 return NULL;
1633
1634         return bdev_get_queue(bdev);
1635 }
1636
1637 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1638                                          struct device_attribute *attr,
1639                                          char *buf)
1640 {
1641         struct hd_struct *p = dev_to_part(dev);
1642         struct request_queue *q;
1643         struct block_device *bdev;
1644         ssize_t ret = -ENXIO;
1645
1646         bdev = bdget(part_devt(p));
1647         if (bdev == NULL)
1648                 goto out;
1649
1650         q = blk_trace_get_queue(bdev);
1651         if (q == NULL)
1652                 goto out_bdput;
1653
1654         mutex_lock(&bdev->bd_mutex);
1655
1656         if (attr == &dev_attr_enable) {
1657                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1658                 goto out_unlock_bdev;
1659         }
1660
1661         if (q->blk_trace == NULL)
1662                 ret = sprintf(buf, "disabled\n");
1663         else if (attr == &dev_attr_act_mask)
1664                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1665         else if (attr == &dev_attr_pid)
1666                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1667         else if (attr == &dev_attr_start_lba)
1668                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1669         else if (attr == &dev_attr_end_lba)
1670                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1671
1672 out_unlock_bdev:
1673         mutex_unlock(&bdev->bd_mutex);
1674 out_bdput:
1675         bdput(bdev);
1676 out:
1677         return ret;
1678 }
1679
1680 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1681                                           struct device_attribute *attr,
1682                                           const char *buf, size_t count)
1683 {
1684         struct block_device *bdev;
1685         struct request_queue *q;
1686         struct hd_struct *p;
1687         u64 value;
1688         ssize_t ret = -EINVAL;
1689
1690         if (count == 0)
1691                 goto out;
1692
1693         if (attr == &dev_attr_act_mask) {
1694                 if (sscanf(buf, "%llx", &value) != 1) {
1695                         /* Assume it is a list of trace category names */
1696                         ret = blk_trace_str2mask(buf);
1697                         if (ret < 0)
1698                                 goto out;
1699                         value = ret;
1700                 }
1701         } else if (sscanf(buf, "%llu", &value) != 1)
1702                 goto out;
1703
1704         ret = -ENXIO;
1705
1706         p = dev_to_part(dev);
1707         bdev = bdget(part_devt(p));
1708         if (bdev == NULL)
1709                 goto out;
1710
1711         q = blk_trace_get_queue(bdev);
1712         if (q == NULL)
1713                 goto out_bdput;
1714
1715         mutex_lock(&bdev->bd_mutex);
1716
1717         if (attr == &dev_attr_enable) {
1718                 if (value)
1719                         ret = blk_trace_setup_queue(q, bdev);
1720                 else
1721                         ret = blk_trace_remove_queue(q);
1722                 goto out_unlock_bdev;
1723         }
1724
1725         ret = 0;
1726         if (q->blk_trace == NULL)
1727                 ret = blk_trace_setup_queue(q, bdev);
1728
1729         if (ret == 0) {
1730                 if (attr == &dev_attr_act_mask)
1731                         q->blk_trace->act_mask = value;
1732                 else if (attr == &dev_attr_pid)
1733                         q->blk_trace->pid = value;
1734                 else if (attr == &dev_attr_start_lba)
1735                         q->blk_trace->start_lba = value;
1736                 else if (attr == &dev_attr_end_lba)
1737                         q->blk_trace->end_lba = value;
1738         }
1739
1740 out_unlock_bdev:
1741         mutex_unlock(&bdev->bd_mutex);
1742 out_bdput:
1743         bdput(bdev);
1744 out:
1745         return ret ? ret : count;
1746 }
1747
1748 int blk_trace_init_sysfs(struct device *dev)
1749 {
1750         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1751 }
1752
1753 void blk_trace_remove_sysfs(struct device *dev)
1754 {
1755         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1756 }
1757
1758 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1759
1760 #ifdef CONFIG_EVENT_TRACING
1761
1762 void blk_dump_cmd(char *buf, struct request *rq)
1763 {
1764         int i, end;
1765         int len = rq->cmd_len;
1766         unsigned char *cmd = rq->cmd;
1767
1768         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1769                 buf[0] = '\0';
1770                 return;
1771         }
1772
1773         for (end = len - 1; end >= 0; end--)
1774                 if (cmd[end])
1775                         break;
1776         end++;
1777
1778         for (i = 0; i < len; i++) {
1779                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1780                 if (i == end && end != len - 1) {
1781                         sprintf(buf, " ..");
1782                         break;
1783                 }
1784         }
1785 }
1786
1787 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1788 {
1789         int i = 0;
1790
1791         if (rw & WRITE)
1792                 rwbs[i++] = 'W';
1793         else if (rw & REQ_DISCARD)
1794                 rwbs[i++] = 'D';
1795         else if (bytes)
1796                 rwbs[i++] = 'R';
1797         else
1798                 rwbs[i++] = 'N';
1799
1800         if (rw & REQ_RAHEAD)
1801                 rwbs[i++] = 'A';
1802         if (rw & REQ_SYNC)
1803                 rwbs[i++] = 'S';
1804         if (rw & REQ_META)
1805                 rwbs[i++] = 'M';
1806         if (rw & REQ_SECURE)
1807                 rwbs[i++] = 'E';
1808
1809         rwbs[i] = '\0';
1810 }
1811
1812 #endif /* CONFIG_EVENT_TRACING */
1813