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 #include <linux/list.h>
  30
  31 #include "../../block/blk.h"
  32
  33 #include <trace/events/block.h>
  34
  35 #include "trace_output.h"
  36
  37 #ifdef CONFIG_BLK_DEV_IO_TRACE
  38
  39 static unsigned int blktrace_seq __read_mostly = 1;
  40
  41 static struct trace_array *blk_tr;
  42 static bool blk_tracer_enabled __read_mostly;
  43
  44 static LIST_HEAD(running_trace_list);
  45 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  46
  47 /* Select an alternative, minimalistic output than the original one */
  48 #define TRACE_BLK_OPT_CLASSIC   0x1
  49
  50 static struct tracer_opt blk_tracer_opts[] = {
  51         /* Default disable the minimalistic output */
  52         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  53         { }
  54 };
  55
  56 static struct tracer_flags blk_tracer_flags = {
  57         .val  = 0,
  58         .opts = blk_tracer_opts,
  59 };
  60
  61 /* Global reference count of probes */
  62 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  63
  64 static void blk_register_tracepoints(void);
  65 static void blk_unregister_tracepoints(void);
  66
  67 /*
  68  * Send out a notify message.
  69  */
  70 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  71                        const void *data, size_t len)
  72 {
  73         struct blk_io_trace *t;
  74         struct ring_buffer_event *event = NULL;
  75         struct ring_buffer *buffer = NULL;
  76         int pc = 0;
  77         int cpu = smp_processor_id();
  78         bool blk_tracer = blk_tracer_enabled;
  79
  80         if (blk_tracer) {
  81                 buffer = blk_tr->trace_buffer.buffer;
  82                 pc = preempt_count();
  83                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84                                                   sizeof(*t) + len,
  85                                                   0, pc);
  86                 if (!event)
  87                         return;
  88                 t = ring_buffer_event_data(event);
  89                 goto record_it;
  90         }
  91
  92         if (!bt->rchan)
  93                 return;
  94
  95         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  96         if (t) {
  97                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  98                 t->time = ktime_to_ns(ktime_get());
  99 record_it:
 100                 t->device = bt->dev;
 101                 t->action = action;
 102                 t->pid = pid;
 103                 t->cpu = cpu;
 104                 t->pdu_len = len;
 105                 memcpy((void *) t + sizeof(*t), data, len);
 106
 107                 if (blk_tracer)
 108                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 109         }
 110 }
 111
 112 /*
 113  * Send out a notify for this process, if we haven't done so since a trace
 114  * started
 115  */
 116 static void trace_note_tsk(struct task_struct *tsk)
 117 {
 118         unsigned long flags;
 119         struct blk_trace *bt;
 120
 121         tsk->btrace_seq = blktrace_seq;
 122         spin_lock_irqsave(&running_trace_lock, flags);
 123         list_for_each_entry(bt, &running_trace_list, running_list) {
 124                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 125                            sizeof(tsk->comm));
 126         }
 127         spin_unlock_irqrestore(&running_trace_lock, flags);
 128 }
 129
 130 static void trace_note_time(struct blk_trace *bt)
 131 {
 132         struct timespec64 now;
 133         unsigned long flags;
 134         u32 words[2];
 135
 136         /* need to check user space to see if this breaks in y2038 or y2106 */
 137         ktime_get_real_ts64(&now);
 138         words[0] = (u32)now.tv_sec;
 139         words[1] = now.tv_nsec;
 140
 141         local_irq_save(flags);
 142         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 143         local_irq_restore(flags);
 144 }
 145
 146 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 147 {
 148         int n;
 149         va_list args;
 150         unsigned long flags;
 151         char *buf;
 152
 153         if (unlikely(bt->trace_state != Blktrace_running &&
 154                      !blk_tracer_enabled))
 155                 return;
 156
 157         /*
 158          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 159          * message to the trace.
 160          */
 161         if (!(bt->act_mask & BLK_TC_NOTIFY))
 162                 return;
 163
 164         local_irq_save(flags);
 165         buf = this_cpu_ptr(bt->msg_data);
 166         va_start(args, fmt);
 167         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 168         va_end(args);
 169
 170         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 171         local_irq_restore(flags);
 172 }
 173 EXPORT_SYMBOL_GPL(__trace_note_message);
 174
 175 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 176                          pid_t pid)
 177 {
 178         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 179                 return 1;
 180         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 181                 return 1;
 182         if (bt->pid && pid != bt->pid)
 183                 return 1;
 184
 185         return 0;
 186 }
 187
 188 /*
 189  * Data direction bit lookup
 190  */
 191 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 192                                  BLK_TC_ACT(BLK_TC_WRITE) };
 193
 194 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 195 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 196
 197 /* The ilog2() calls fall out because they're constant */
 198 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 199           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 200
 201 /*
 202  * The worker for the various blk_add_trace*() types. Fills out a
 203  * blk_io_trace structure and places it in a per-cpu subbuffer.
 204  */
 205 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 206                      int op, int op_flags, u32 what, int error, int pdu_len,
 207                      void *pdu_data)
 208 {
 209         struct task_struct *tsk = current;
 210         struct ring_buffer_event *event = NULL;
 211         struct ring_buffer *buffer = NULL;
 212         struct blk_io_trace *t;
 213         unsigned long flags = 0;
 214         unsigned long *sequence;
 215         pid_t pid;
 216         int cpu, pc = 0;
 217         bool blk_tracer = blk_tracer_enabled;
 218
 219         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 220                 return;
 221
 222         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 223         what |= MASK_TC_BIT(op_flags, SYNC);
 224         what |= MASK_TC_BIT(op_flags, RAHEAD);
 225         what |= MASK_TC_BIT(op_flags, META);
 226         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 227         what |= MASK_TC_BIT(op_flags, FUA);
 228         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 229                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 230         if (op == REQ_OP_FLUSH)
 231                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 232
 233         pid = tsk->pid;
 234         if (act_log_check(bt, what, sector, pid))
 235                 return;
 236         cpu = raw_smp_processor_id();
 237
 238         if (blk_tracer) {
 239                 tracing_record_cmdline(current);
 240
 241                 buffer = blk_tr->trace_buffer.buffer;
 242                 pc = preempt_count();
 243                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 244                                                   sizeof(*t) + pdu_len,
 245                                                   0, pc);
 246                 if (!event)
 247                         return;
 248                 t = ring_buffer_event_data(event);
 249                 goto record_it;
 250         }
 251
 252         if (unlikely(tsk->btrace_seq != blktrace_seq))
 253                 trace_note_tsk(tsk);
 254
 255         /*
 256          * A word about the locking here - we disable interrupts to reserve
 257          * some space in the relay per-cpu buffer, to prevent an irq
 258          * from coming in and stepping on our toes.
 259          */
 260         local_irq_save(flags);
 261         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 262         if (t) {
 263                 sequence = per_cpu_ptr(bt->sequence, cpu);
 264
 265                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 266                 t->sequence = ++(*sequence);
 267                 t->time = ktime_to_ns(ktime_get());
 268 record_it:
 269                 /*
 270                  * These two are not needed in ftrace as they are in the
 271                  * generic trace_entry, filled by tracing_generic_entry_update,
 272                  * but for the trace_event->bin() synthesizer benefit we do it
 273                  * here too.
 274                  */
 275                 t->cpu = cpu;
 276                 t->pid = pid;
 277
 278                 t->sector = sector;
 279                 t->bytes = bytes;
 280                 t->action = what;
 281                 t->device = bt->dev;
 282                 t->error = error;
 283                 t->pdu_len = pdu_len;
 284
 285                 if (pdu_len)
 286                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 287
 288                 if (blk_tracer) {
 289                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 290                         return;
 291                 }
 292         }
 293
 294         local_irq_restore(flags);
 295 }
 296
 297 static void blk_trace_free(struct blk_trace *bt)
 298 {
 299         debugfs_remove(bt->msg_file);
 300         debugfs_remove(bt->dropped_file);
 301         relay_close(bt->rchan);
 302         debugfs_remove(bt->dir);
 303         free_percpu(bt->sequence);
 304         free_percpu(bt->msg_data);
 305         kfree(bt);
 306 }
 307
 308 static void blk_trace_cleanup(struct blk_trace *bt)
 309 {
 310         blk_trace_free(bt);
 311         if (atomic_dec_and_test(&blk_probes_ref))
 312                 blk_unregister_tracepoints();
 313 }
 314
 315 int blk_trace_remove(struct request_queue *q)
 316 {
 317         struct blk_trace *bt;
 318
 319         bt = xchg(&q->blk_trace, NULL);
 320         if (!bt)
 321                 return -EINVAL;
 322
 323         if (bt->trace_state != Blktrace_running)
 324                 blk_trace_cleanup(bt);
 325
 326         return 0;
 327 }
 328 EXPORT_SYMBOL_GPL(blk_trace_remove);
 329
 330 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 331                                 size_t count, loff_t *ppos)
 332 {
 333         struct blk_trace *bt = filp->private_data;
 334         char buf[16];
 335
 336         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 337
 338         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 339 }
 340
 341 static const struct file_operations blk_dropped_fops = {
 342         .owner =        THIS_MODULE,
 343         .open =         simple_open,
 344         .read =         blk_dropped_read,
 345         .llseek =       default_llseek,
 346 };
 347
 348 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 349                                 size_t count, loff_t *ppos)
 350 {
 351         char *msg;
 352         struct blk_trace *bt;
 353
 354         if (count >= BLK_TN_MAX_MSG)
 355                 return -EINVAL;
 356
 357         msg = memdup_user_nul(buffer, count);
 358         if (IS_ERR(msg))
 359                 return PTR_ERR(msg);
 360
 361         bt = filp->private_data;
 362         __trace_note_message(bt, "%s", msg);
 363         kfree(msg);
 364
 365         return count;
 366 }
 367
 368 static const struct file_operations blk_msg_fops = {
 369         .owner =        THIS_MODULE,
 370         .open =         simple_open,
 371         .write =        blk_msg_write,
 372         .llseek =       noop_llseek,
 373 };
 374
 375 /*
 376  * Keep track of how many times we encountered a full subbuffer, to aid
 377  * the user space app in telling how many lost events there were.
 378  */
 379 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 380                                      void *prev_subbuf, size_t prev_padding)
 381 {
 382         struct blk_trace *bt;
 383
 384         if (!relay_buf_full(buf))
 385                 return 1;
 386
 387         bt = buf->chan->private_data;
 388         atomic_inc(&bt->dropped);
 389         return 0;
 390 }
 391
 392 static int blk_remove_buf_file_callback(struct dentry *dentry)
 393 {
 394         debugfs_remove(dentry);
 395
 396         return 0;
 397 }
 398
 399 static struct dentry *blk_create_buf_file_callback(const char *filename,
 400                                                    struct dentry *parent,
 401                                                    umode_t mode,
 402                                                    struct rchan_buf *buf,
 403                                                    int *is_global)
 404 {
 405         return debugfs_create_file(filename, mode, parent, buf,
 406                                         &relay_file_operations);
 407 }
 408
 409 static struct rchan_callbacks blk_relay_callbacks = {
 410         .subbuf_start           = blk_subbuf_start_callback,
 411         .create_buf_file        = blk_create_buf_file_callback,
 412         .remove_buf_file        = blk_remove_buf_file_callback,
 413 };
 414
 415 static void blk_trace_setup_lba(struct blk_trace *bt,
 416                                 struct block_device *bdev)
 417 {
 418         struct hd_struct *part = NULL;
 419
 420         if (bdev)
 421                 part = bdev->bd_part;
 422
 423         if (part) {
 424                 bt->start_lba = part->start_sect;
 425                 bt->end_lba = part->start_sect + part->nr_sects;
 426         } else {
 427                 bt->start_lba = 0;
 428                 bt->end_lba = -1ULL;
 429         }
 430 }
 431
 432 /*
 433  * Setup everything required to start tracing
 434  */
 435 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 436                               struct block_device *bdev,
 437                               struct blk_user_trace_setup *buts)
 438 {
 439         struct blk_trace *bt = NULL;
 440         struct dentry *dir = NULL;
 441         int ret;
 442
 443         if (!buts->buf_size || !buts->buf_nr)
 444                 return -EINVAL;
 445
 446         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 447         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 448
 449         /*
 450          * some device names have larger paths - convert the slashes
 451          * to underscores for this to work as expected
 452          */
 453         strreplace(buts->name, '/', '_');
 454
 455         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 456         if (!bt)
 457                 return -ENOMEM;
 458
 459         ret = -ENOMEM;
 460         bt->sequence = alloc_percpu(unsigned long);
 461         if (!bt->sequence)
 462                 goto err;
 463
 464         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 465         if (!bt->msg_data)
 466                 goto err;
 467
 468         ret = -ENOENT;
 469
 470         if (!blk_debugfs_root)
 471                 goto err;
 472
 473         dir = debugfs_lookup(buts->name, blk_debugfs_root);
 474         if (!dir)
 475                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 476         if (!dir)
 477                 goto err;
 478
 479         bt->dev = dev;
 480         atomic_set(&bt->dropped, 0);
 481         INIT_LIST_HEAD(&bt->running_list);
 482
 483         ret = -EIO;
 484         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 485                                                &blk_dropped_fops);
 486         if (!bt->dropped_file)
 487                 goto err;
 488
 489         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 490         if (!bt->msg_file)
 491                 goto err;
 492
 493         bt->rchan = relay_open("trace", dir, buts->buf_size,
 494                                 buts->buf_nr, &blk_relay_callbacks, bt);
 495         if (!bt->rchan)
 496                 goto err;
 497
 498         bt->act_mask = buts->act_mask;
 499         if (!bt->act_mask)
 500                 bt->act_mask = (u16) -1;
 501
 502         blk_trace_setup_lba(bt, bdev);
 503
 504         /* overwrite with user settings */
 505         if (buts->start_lba)
 506                 bt->start_lba = buts->start_lba;
 507         if (buts->end_lba)
 508                 bt->end_lba = buts->end_lba;
 509
 510         bt->pid = buts->pid;
 511         bt->trace_state = Blktrace_setup;
 512
 513         ret = -EBUSY;
 514         if (cmpxchg(&q->blk_trace, NULL, bt))
 515                 goto err;
 516
 517         if (atomic_inc_return(&blk_probes_ref) == 1)
 518                 blk_register_tracepoints();
 519
 520         ret = 0;
 521 err:
 522         if (dir && !bt->dir)
 523                 dput(dir);
 524         if (ret)
 525                 blk_trace_free(bt);
 526         return ret;
 527 }
 528
 529 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 530                     struct block_device *bdev,
 531                     char __user *arg)
 532 {
 533         struct blk_user_trace_setup buts;
 534         int ret;
 535
 536         ret = copy_from_user(&buts, arg, sizeof(buts));
 537         if (ret)
 538                 return -EFAULT;
 539
 540         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 541         if (ret)
 542                 return ret;
 543
 544         if (copy_to_user(arg, &buts, sizeof(buts))) {
 545                 blk_trace_remove(q);
 546                 return -EFAULT;
 547         }
 548         return 0;
 549 }
 550 EXPORT_SYMBOL_GPL(blk_trace_setup);
 551
 552 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 553 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 554                                   dev_t dev, struct block_device *bdev,
 555                                   char __user *arg)
 556 {
 557         struct blk_user_trace_setup buts;
 558         struct compat_blk_user_trace_setup cbuts;
 559         int ret;
 560
 561         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 562                 return -EFAULT;
 563
 564         buts = (struct blk_user_trace_setup) {
 565                 .act_mask = cbuts.act_mask,
 566                 .buf_size = cbuts.buf_size,
 567                 .buf_nr = cbuts.buf_nr,
 568                 .start_lba = cbuts.start_lba,
 569                 .end_lba = cbuts.end_lba,
 570                 .pid = cbuts.pid,
 571         };
 572
 573         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 574         if (ret)
 575                 return ret;
 576
 577         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 578                 blk_trace_remove(q);
 579                 return -EFAULT;
 580         }
 581
 582         return 0;
 583 }
 584 #endif
 585
 586 int blk_trace_startstop(struct request_queue *q, int start)
 587 {
 588         int ret;
 589         struct blk_trace *bt = q->blk_trace;
 590
 591         if (bt == NULL)
 592                 return -EINVAL;
 593
 594         /*
 595          * For starting a trace, we can transition from a setup or stopped
 596          * trace. For stopping a trace, the state must be running
 597          */
 598         ret = -EINVAL;
 599         if (start) {
 600                 if (bt->trace_state == Blktrace_setup ||
 601                     bt->trace_state == Blktrace_stopped) {
 602                         blktrace_seq++;
 603                         smp_mb();
 604                         bt->trace_state = Blktrace_running;
 605                         spin_lock_irq(&running_trace_lock);
 606                         list_add(&bt->running_list, &running_trace_list);
 607                         spin_unlock_irq(&running_trace_lock);
 608
 609                         trace_note_time(bt);
 610                         ret = 0;
 611                 }
 612         } else {
 613                 if (bt->trace_state == Blktrace_running) {
 614                         bt->trace_state = Blktrace_stopped;
 615                         spin_lock_irq(&running_trace_lock);
 616                         list_del_init(&bt->running_list);
 617                         spin_unlock_irq(&running_trace_lock);
 618                         relay_flush(bt->rchan);
 619                         ret = 0;
 620                 }
 621         }
 622
 623         return ret;
 624 }
 625 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 626
 627 /**
 628  * blk_trace_ioctl: - handle the ioctls associated with tracing
 629  * @bdev:       the block device
 630  * @cmd:        the ioctl cmd
 631  * @arg:        the argument data, if any
 632  *
 633  **/
 634 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 635 {
 636         struct request_queue *q;
 637         int ret, start = 0;
 638         char b[BDEVNAME_SIZE];
 639
 640         q = bdev_get_queue(bdev);
 641         if (!q)
 642                 return -ENXIO;
 643
 644         mutex_lock(&bdev->bd_mutex);
 645
 646         switch (cmd) {
 647         case BLKTRACESETUP:
 648                 bdevname(bdev, b);
 649                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 650                 break;
 651 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 652         case BLKTRACESETUP32:
 653                 bdevname(bdev, b);
 654                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 655                 break;
 656 #endif
 657         case BLKTRACESTART:
 658                 start = 1;
 659         case BLKTRACESTOP:
 660                 ret = blk_trace_startstop(q, start);
 661                 break;
 662         case BLKTRACETEARDOWN:
 663                 ret = blk_trace_remove(q);
 664                 break;
 665         default:
 666                 ret = -ENOTTY;
 667                 break;
 668         }
 669
 670         mutex_unlock(&bdev->bd_mutex);
 671         return ret;
 672 }
 673
 674 /**
 675  * blk_trace_shutdown: - stop and cleanup trace structures
 676  * @q:    the request queue associated with the device
 677  *
 678  **/
 679 void blk_trace_shutdown(struct request_queue *q)
 680 {
 681         if (q->blk_trace) {
 682                 blk_trace_startstop(q, 0);
 683                 blk_trace_remove(q);
 684         }
 685 }
 686
 687 /*
 688  * blktrace probes
 689  */
 690
 691 /**
 692  * blk_add_trace_rq - Add a trace for a request oriented action
 693  * @rq:         the source request
 694  * @error:      return status to log
 695  * @nr_bytes:   number of completed bytes
 696  * @what:       the action
 697  *
 698  * Description:
 699  *     Records an action against a request. Will log the bio offset + size.
 700  *
 701  **/
 702 static void blk_add_trace_rq(struct request *rq, int error,
 703                              unsigned int nr_bytes, u32 what)
 704 {
 705         struct blk_trace *bt = rq->q->blk_trace;
 706
 707         if (likely(!bt))
 708                 return;
 709
 710         if (blk_rq_is_passthrough(rq))
 711                 what |= BLK_TC_ACT(BLK_TC_PC);
 712         else
 713                 what |= BLK_TC_ACT(BLK_TC_FS);
 714
 715         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 716                         rq->cmd_flags, what, error, 0, NULL);
 717 }
 718
 719 static void blk_add_trace_rq_insert(void *ignore,
 720                                     struct request_queue *q, struct request *rq)
 721 {
 722         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT);
 723 }
 724
 725 static void blk_add_trace_rq_issue(void *ignore,
 726                                    struct request_queue *q, struct request *rq)
 727 {
 728         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE);
 729 }
 730
 731 static void blk_add_trace_rq_requeue(void *ignore,
 732                                      struct request_queue *q,
 733                                      struct request *rq)
 734 {
 735         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 736 }
 737
 738 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 739                         int error, unsigned int nr_bytes)
 740 {
 741         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE);
 742 }
 743
 744 /**
 745  * blk_add_trace_bio - Add a trace for a bio oriented action
 746  * @q:          queue the io is for
 747  * @bio:        the source bio
 748  * @what:       the action
 749  * @error:      error, if any
 750  *
 751  * Description:
 752  *     Records an action against a bio. Will log the bio offset + size.
 753  *
 754  **/
 755 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 756                               u32 what, int error)
 757 {
 758         struct blk_trace *bt = q->blk_trace;
 759
 760         if (likely(!bt))
 761                 return;
 762
 763         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 764                         bio_op(bio), bio->bi_opf, what, error, 0, NULL);
 765 }
 766
 767 static void blk_add_trace_bio_bounce(void *ignore,
 768                                      struct request_queue *q, struct bio *bio)
 769 {
 770         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 771 }
 772
 773 static void blk_add_trace_bio_complete(void *ignore,
 774                                        struct request_queue *q, struct bio *bio,
 775                                        int error)
 776 {
 777         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 778 }
 779
 780 static void blk_add_trace_bio_backmerge(void *ignore,
 781                                         struct request_queue *q,
 782                                         struct request *rq,
 783                                         struct bio *bio)
 784 {
 785         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 786 }
 787
 788 static void blk_add_trace_bio_frontmerge(void *ignore,
 789                                          struct request_queue *q,
 790                                          struct request *rq,
 791                                          struct bio *bio)
 792 {
 793         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 794 }
 795
 796 static void blk_add_trace_bio_queue(void *ignore,
 797                                     struct request_queue *q, struct bio *bio)
 798 {
 799         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 800 }
 801
 802 static void blk_add_trace_getrq(void *ignore,
 803                                 struct request_queue *q,
 804                                 struct bio *bio, int rw)
 805 {
 806         if (bio)
 807                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 808         else {
 809                 struct blk_trace *bt = q->blk_trace;
 810
 811                 if (bt)
 812                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 813                                         NULL);
 814         }
 815 }
 816
 817
 818 static void blk_add_trace_sleeprq(void *ignore,
 819                                   struct request_queue *q,
 820                                   struct bio *bio, int rw)
 821 {
 822         if (bio)
 823                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 824         else {
 825                 struct blk_trace *bt = q->blk_trace;
 826
 827                 if (bt)
 828                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 829                                         0, 0, NULL);
 830         }
 831 }
 832
 833 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 834 {
 835         struct blk_trace *bt = q->blk_trace;
 836
 837         if (bt)
 838                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 839 }
 840
 841 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 842                                     unsigned int depth, bool explicit)
 843 {
 844         struct blk_trace *bt = q->blk_trace;
 845
 846         if (bt) {
 847                 __be64 rpdu = cpu_to_be64(depth);
 848                 u32 what;
 849
 850                 if (explicit)
 851                         what = BLK_TA_UNPLUG_IO;
 852                 else
 853                         what = BLK_TA_UNPLUG_TIMER;
 854
 855                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 856         }
 857 }
 858
 859 static void blk_add_trace_split(void *ignore,
 860                                 struct request_queue *q, struct bio *bio,
 861                                 unsigned int pdu)
 862 {
 863         struct blk_trace *bt = q->blk_trace;
 864
 865         if (bt) {
 866                 __be64 rpdu = cpu_to_be64(pdu);
 867
 868                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 869                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 870                                 BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
 871                                 &rpdu);
 872         }
 873 }
 874
 875 /**
 876  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 877  * @ignore:     trace callback data parameter (not used)
 878  * @q:          queue the io is for
 879  * @bio:        the source bio
 880  * @dev:        target device
 881  * @from:       source sector
 882  *
 883  * Description:
 884  *     Device mapper or raid target sometimes need to split a bio because
 885  *     it spans a stripe (or similar). Add a trace for that action.
 886  *
 887  **/
 888 static void blk_add_trace_bio_remap(void *ignore,
 889                                     struct request_queue *q, struct bio *bio,
 890                                     dev_t dev, sector_t from)
 891 {
 892         struct blk_trace *bt = q->blk_trace;
 893         struct blk_io_trace_remap r;
 894
 895         if (likely(!bt))
 896                 return;
 897
 898         r.device_from = cpu_to_be32(dev);
 899         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 900         r.sector_from = cpu_to_be64(from);
 901
 902         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 903                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_error,
 904                         sizeof(r), &r);
 905 }
 906
 907 /**
 908  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 909  * @ignore:     trace callback data parameter (not used)
 910  * @q:          queue the io is for
 911  * @rq:         the source request
 912  * @dev:        target device
 913  * @from:       source sector
 914  *
 915  * Description:
 916  *     Device mapper remaps request to other devices.
 917  *     Add a trace for that action.
 918  *
 919  **/
 920 static void blk_add_trace_rq_remap(void *ignore,
 921                                    struct request_queue *q,
 922                                    struct request *rq, dev_t dev,
 923                                    sector_t from)
 924 {
 925         struct blk_trace *bt = q->blk_trace;
 926         struct blk_io_trace_remap r;
 927
 928         if (likely(!bt))
 929                 return;
 930
 931         r.device_from = cpu_to_be32(dev);
 932         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 933         r.sector_from = cpu_to_be64(from);
 934
 935         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 936                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
 937                         sizeof(r), &r);
 938 }
 939
 940 /**
 941  * blk_add_driver_data - Add binary message with driver-specific data
 942  * @q:          queue the io is for
 943  * @rq:         io request
 944  * @data:       driver-specific data
 945  * @len:        length of driver-specific data
 946  *
 947  * Description:
 948  *     Some drivers might want to write driver-specific data per request.
 949  *
 950  **/
 951 void blk_add_driver_data(struct request_queue *q,
 952                          struct request *rq,
 953                          void *data, size_t len)
 954 {
 955         struct blk_trace *bt = q->blk_trace;
 956
 957         if (likely(!bt))
 958                 return;
 959
 960         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
 961                                 BLK_TA_DRV_DATA, 0, len, data);
 962 }
 963 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 964
 965 static void blk_register_tracepoints(void)
 966 {
 967         int ret;
 968
 969         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 970         WARN_ON(ret);
 971         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 972         WARN_ON(ret);
 973         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 974         WARN_ON(ret);
 975         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 976         WARN_ON(ret);
 977         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 978         WARN_ON(ret);
 979         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 980         WARN_ON(ret);
 981         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 982         WARN_ON(ret);
 983         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
 984         WARN_ON(ret);
 985         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
 986         WARN_ON(ret);
 987         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
 988         WARN_ON(ret);
 989         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
 990         WARN_ON(ret);
 991         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
 992         WARN_ON(ret);
 993         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
 994         WARN_ON(ret);
 995         ret = register_trace_block_split(blk_add_trace_split, NULL);
 996         WARN_ON(ret);
 997         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
 998         WARN_ON(ret);
 999         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1000         WARN_ON(ret);
1001 }
1002
1003 static void blk_unregister_tracepoints(void)
1004 {
1005         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1006         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1007         unregister_trace_block_split(blk_add_trace_split, NULL);
1008         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1009         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1010         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1011         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1012         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1013         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1014         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1015         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1016         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1017         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1018         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1019         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1020         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1021
1022         tracepoint_synchronize_unregister();
1023 }
1024
1025 /*
1026  * struct blk_io_tracer formatting routines
1027  */
1028
1029 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1030 {
1031         int i = 0;
1032         int tc = t->action >> BLK_TC_SHIFT;
1033
1034         if (t->action == BLK_TN_MESSAGE) {
1035                 rwbs[i++] = 'N';
1036                 goto out;
1037         }
1038
1039         if (tc & BLK_TC_FLUSH)
1040                 rwbs[i++] = 'F';
1041
1042         if (tc & BLK_TC_DISCARD)
1043                 rwbs[i++] = 'D';
1044         else if (tc & BLK_TC_WRITE)
1045                 rwbs[i++] = 'W';
1046         else if (t->bytes)
1047                 rwbs[i++] = 'R';
1048         else
1049                 rwbs[i++] = 'N';
1050
1051         if (tc & BLK_TC_FUA)
1052                 rwbs[i++] = 'F';
1053         if (tc & BLK_TC_AHEAD)
1054                 rwbs[i++] = 'A';
1055         if (tc & BLK_TC_SYNC)
1056                 rwbs[i++] = 'S';
1057         if (tc & BLK_TC_META)
1058                 rwbs[i++] = 'M';
1059 out:
1060         rwbs[i] = '\0';
1061 }
1062
1063 static inline
1064 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1065 {
1066         return (const struct blk_io_trace *)ent;
1067 }
1068
1069 static inline const void *pdu_start(const struct trace_entry *ent)
1070 {
1071         return te_blk_io_trace(ent) + 1;
1072 }
1073
1074 static inline u32 t_action(const struct trace_entry *ent)
1075 {
1076         return te_blk_io_trace(ent)->action;
1077 }
1078
1079 static inline u32 t_bytes(const struct trace_entry *ent)
1080 {
1081         return te_blk_io_trace(ent)->bytes;
1082 }
1083
1084 static inline u32 t_sec(const struct trace_entry *ent)
1085 {
1086         return te_blk_io_trace(ent)->bytes >> 9;
1087 }
1088
1089 static inline unsigned long long t_sector(const struct trace_entry *ent)
1090 {
1091         return te_blk_io_trace(ent)->sector;
1092 }
1093
1094 static inline __u16 t_error(const struct trace_entry *ent)
1095 {
1096         return te_blk_io_trace(ent)->error;
1097 }
1098
1099 static __u64 get_pdu_int(const struct trace_entry *ent)
1100 {
1101         const __u64 *val = pdu_start(ent);
1102         return be64_to_cpu(*val);
1103 }
1104
1105 static void get_pdu_remap(const struct trace_entry *ent,
1106                           struct blk_io_trace_remap *r)
1107 {
1108         const struct blk_io_trace_remap *__r = pdu_start(ent);
1109         __u64 sector_from = __r->sector_from;
1110
1111         r->device_from = be32_to_cpu(__r->device_from);
1112         r->device_to   = be32_to_cpu(__r->device_to);
1113         r->sector_from = be64_to_cpu(sector_from);
1114 }
1115
1116 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1117
1118 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1119 {
1120         char rwbs[RWBS_LEN];
1121         unsigned long long ts  = iter->ts;
1122         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1123         unsigned secs          = (unsigned long)ts;
1124         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1125
1126         fill_rwbs(rwbs, t);
1127
1128         trace_seq_printf(&iter->seq,
1129                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1130                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1131                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1132 }
1133
1134 static void blk_log_action(struct trace_iterator *iter, const char *act)
1135 {
1136         char rwbs[RWBS_LEN];
1137         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1138
1139         fill_rwbs(rwbs, t);
1140         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1141                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1142 }
1143
1144 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1145 {
1146         const unsigned char *pdu_buf;
1147         int pdu_len;
1148         int i, end;
1149
1150         pdu_buf = pdu_start(ent);
1151         pdu_len = te_blk_io_trace(ent)->pdu_len;
1152
1153         if (!pdu_len)
1154                 return;
1155
1156         /* find the last zero that needs to be printed */
1157         for (end = pdu_len - 1; end >= 0; end--)
1158                 if (pdu_buf[end])
1159                         break;
1160         end++;
1161
1162         trace_seq_putc(s, '(');
1163
1164         for (i = 0; i < pdu_len; i++) {
1165
1166                 trace_seq_printf(s, "%s%02x",
1167                                  i == 0 ? "" : " ", pdu_buf[i]);
1168
1169                 /*
1170                  * stop when the rest is just zeroes and indicate so
1171                  * with a ".." appended
1172                  */
1173                 if (i == end && end != pdu_len - 1) {
1174                         trace_seq_puts(s, " ..) ");
1175                         return;
1176                 }
1177         }
1178
1179         trace_seq_puts(s, ") ");
1180 }
1181
1182 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1183 {
1184         char cmd[TASK_COMM_LEN];
1185
1186         trace_find_cmdline(ent->pid, cmd);
1187
1188         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1189                 trace_seq_printf(s, "%u ", t_bytes(ent));
1190                 blk_log_dump_pdu(s, ent);
1191                 trace_seq_printf(s, "[%s]\n", cmd);
1192         } else {
1193                 if (t_sec(ent))
1194                         trace_seq_printf(s, "%llu + %u [%s]\n",
1195                                                 t_sector(ent), t_sec(ent), cmd);
1196                 else
1197                         trace_seq_printf(s, "[%s]\n", cmd);
1198         }
1199 }
1200
1201 static void blk_log_with_error(struct trace_seq *s,
1202                               const struct trace_entry *ent)
1203 {
1204         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1205                 blk_log_dump_pdu(s, ent);
1206                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1207         } else {
1208                 if (t_sec(ent))
1209                         trace_seq_printf(s, "%llu + %u [%d]\n",
1210                                          t_sector(ent),
1211                                          t_sec(ent), t_error(ent));
1212                 else
1213                         trace_seq_printf(s, "%llu [%d]\n",
1214                                          t_sector(ent), t_error(ent));
1215         }
1216 }
1217
1218 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1219 {
1220         struct blk_io_trace_remap r = { .device_from = 0, };
1221
1222         get_pdu_remap(ent, &r);
1223         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1224                          t_sector(ent), t_sec(ent),
1225                          MAJOR(r.device_from), MINOR(r.device_from),
1226                          (unsigned long long)r.sector_from);
1227 }
1228
1229 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1230 {
1231         char cmd[TASK_COMM_LEN];
1232
1233         trace_find_cmdline(ent->pid, cmd);
1234
1235         trace_seq_printf(s, "[%s]\n", cmd);
1236 }
1237
1238 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1239 {
1240         char cmd[TASK_COMM_LEN];
1241
1242         trace_find_cmdline(ent->pid, cmd);
1243
1244         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1245 }
1246
1247 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1248 {
1249         char cmd[TASK_COMM_LEN];
1250
1251         trace_find_cmdline(ent->pid, cmd);
1252
1253         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1254                          get_pdu_int(ent), cmd);
1255 }
1256
1257 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1258 {
1259         const struct blk_io_trace *t = te_blk_io_trace(ent);
1260
1261         trace_seq_putmem(s, t + 1, t->pdu_len);
1262         trace_seq_putc(s, '\n');
1263 }
1264
1265 /*
1266  * struct tracer operations
1267  */
1268
1269 static void blk_tracer_print_header(struct seq_file *m)
1270 {
1271         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1272                 return;
1273         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1274                     "#  |     |     |           |   |   |\n");
1275 }
1276
1277 static void blk_tracer_start(struct trace_array *tr)
1278 {
1279         blk_tracer_enabled = true;
1280 }
1281
1282 static int blk_tracer_init(struct trace_array *tr)
1283 {
1284         blk_tr = tr;
1285         blk_tracer_start(tr);
1286         return 0;
1287 }
1288
1289 static void blk_tracer_stop(struct trace_array *tr)
1290 {
1291         blk_tracer_enabled = false;
1292 }
1293
1294 static void blk_tracer_reset(struct trace_array *tr)
1295 {
1296         blk_tracer_stop(tr);
1297 }
1298
1299 static const struct {
1300         const char *act[2];
1301         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1302 } what2act[] = {
1303         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1304         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1305         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1306         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1307         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1308         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1309         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1310         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1311         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1312         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1313         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1314         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1315         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1316         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1317         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1318 };
1319
1320 static enum print_line_t print_one_line(struct trace_iterator *iter,
1321                                         bool classic)
1322 {
1323         struct trace_array *tr = iter->tr;
1324         struct trace_seq *s = &iter->seq;
1325         const struct blk_io_trace *t;
1326         u16 what;
1327         bool long_act;
1328         blk_log_action_t *log_action;
1329
1330         t          = te_blk_io_trace(iter->ent);
1331         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1332         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1333         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1334
1335         if (t->action == BLK_TN_MESSAGE) {
1336                 log_action(iter, long_act ? "message" : "m");
1337                 blk_log_msg(s, iter->ent);
1338                 return trace_handle_return(s);
1339         }
1340
1341         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1342                 trace_seq_printf(s, "Unknown action %x\n", what);
1343         else {
1344                 log_action(iter, what2act[what].act[long_act]);
1345                 what2act[what].print(s, iter->ent);
1346         }
1347
1348         return trace_handle_return(s);
1349 }
1350
1351 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1352                                                int flags, struct trace_event *event)
1353 {
1354         return print_one_line(iter, false);
1355 }
1356
1357 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1358 {
1359         struct trace_seq *s = &iter->seq;
1360         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1361         const int offset = offsetof(struct blk_io_trace, sector);
1362         struct blk_io_trace old = {
1363                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1364                 .time     = iter->ts,
1365         };
1366
1367         trace_seq_putmem(s, &old, offset);
1368         trace_seq_putmem(s, &t->sector,
1369                          sizeof(old) - offset + t->pdu_len);
1370 }
1371
1372 static enum print_line_t
1373 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1374                              struct trace_event *event)
1375 {
1376         blk_trace_synthesize_old_trace(iter);
1377
1378         return trace_handle_return(&iter->seq);
1379 }
1380
1381 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1382 {
1383         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1384                 return TRACE_TYPE_UNHANDLED;
1385
1386         return print_one_line(iter, true);
1387 }
1388
1389 static int
1390 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1391 {
1392         /* don't output context-info for blk_classic output */
1393         if (bit == TRACE_BLK_OPT_CLASSIC) {
1394                 if (set)
1395                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1396                 else
1397                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1398         }
1399         return 0;
1400 }
1401
1402 static struct tracer blk_tracer __read_mostly = {
1403         .name           = "blk",
1404         .init           = blk_tracer_init,
1405         .reset          = blk_tracer_reset,
1406         .start          = blk_tracer_start,
1407         .stop           = blk_tracer_stop,
1408         .print_header   = blk_tracer_print_header,
1409         .print_line     = blk_tracer_print_line,
1410         .flags          = &blk_tracer_flags,
1411         .set_flag       = blk_tracer_set_flag,
1412 };
1413
1414 static struct trace_event_functions trace_blk_event_funcs = {
1415         .trace          = blk_trace_event_print,
1416         .binary         = blk_trace_event_print_binary,
1417 };
1418
1419 static struct trace_event trace_blk_event = {
1420         .type           = TRACE_BLK,
1421         .funcs          = &trace_blk_event_funcs,
1422 };
1423
1424 static int __init init_blk_tracer(void)
1425 {
1426         if (!register_trace_event(&trace_blk_event)) {
1427                 pr_warn("Warning: could not register block events\n");
1428                 return 1;
1429         }
1430
1431         if (register_tracer(&blk_tracer) != 0) {
1432                 pr_warn("Warning: could not register the block tracer\n");
1433                 unregister_trace_event(&trace_blk_event);
1434                 return 1;
1435         }
1436
1437         return 0;
1438 }
1439
1440 device_initcall(init_blk_tracer);
1441
1442 static int blk_trace_remove_queue(struct request_queue *q)
1443 {
1444         struct blk_trace *bt;
1445
1446         bt = xchg(&q->blk_trace, NULL);
1447         if (bt == NULL)
1448                 return -EINVAL;
1449
1450         if (atomic_dec_and_test(&blk_probes_ref))
1451                 blk_unregister_tracepoints();
1452
1453         blk_trace_free(bt);
1454         return 0;
1455 }
1456
1457 /*
1458  * Setup everything required to start tracing
1459  */
1460 static int blk_trace_setup_queue(struct request_queue *q,
1461                                  struct block_device *bdev)
1462 {
1463         struct blk_trace *bt = NULL;
1464         int ret = -ENOMEM;
1465
1466         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1467         if (!bt)
1468                 return -ENOMEM;
1469
1470         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1471         if (!bt->msg_data)
1472                 goto free_bt;
1473
1474         bt->dev = bdev->bd_dev;
1475         bt->act_mask = (u16)-1;
1476
1477         blk_trace_setup_lba(bt, bdev);
1478
1479         ret = -EBUSY;
1480         if (cmpxchg(&q->blk_trace, NULL, bt))
1481                 goto free_bt;
1482
1483         if (atomic_inc_return(&blk_probes_ref) == 1)
1484                 blk_register_tracepoints();
1485         return 0;
1486
1487 free_bt:
1488         blk_trace_free(bt);
1489         return ret;
1490 }
1491
1492 /*
1493  * sysfs interface to enable and configure tracing
1494  */
1495
1496 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1497                                          struct device_attribute *attr,
1498                                          char *buf);
1499 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1500                                           struct device_attribute *attr,
1501                                           const char *buf, size_t count);
1502 #define BLK_TRACE_DEVICE_ATTR(_name) \
1503         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1504                     sysfs_blk_trace_attr_show, \
1505                     sysfs_blk_trace_attr_store)
1506
1507 static BLK_TRACE_DEVICE_ATTR(enable);
1508 static BLK_TRACE_DEVICE_ATTR(act_mask);
1509 static BLK_TRACE_DEVICE_ATTR(pid);
1510 static BLK_TRACE_DEVICE_ATTR(start_lba);
1511 static BLK_TRACE_DEVICE_ATTR(end_lba);
1512
1513 static struct attribute *blk_trace_attrs[] = {
1514         &dev_attr_enable.attr,
1515         &dev_attr_act_mask.attr,
1516         &dev_attr_pid.attr,
1517         &dev_attr_start_lba.attr,
1518         &dev_attr_end_lba.attr,
1519         NULL
1520 };
1521
1522 struct attribute_group blk_trace_attr_group = {
1523         .name  = "trace",
1524         .attrs = blk_trace_attrs,
1525 };
1526
1527 static const struct {
1528         int mask;
1529         const char *str;
1530 } mask_maps[] = {
1531         { BLK_TC_READ,          "read"          },
1532         { BLK_TC_WRITE,         "write"         },
1533         { BLK_TC_FLUSH,         "flush"         },
1534         { BLK_TC_SYNC,          "sync"          },
1535         { BLK_TC_QUEUE,         "queue"         },
1536         { BLK_TC_REQUEUE,       "requeue"       },
1537         { BLK_TC_ISSUE,         "issue"         },
1538         { BLK_TC_COMPLETE,      "complete"      },
1539         { BLK_TC_FS,            "fs"            },
1540         { BLK_TC_PC,            "pc"            },
1541         { BLK_TC_NOTIFY,        "notify"        },
1542         { BLK_TC_AHEAD,         "ahead"         },
1543         { BLK_TC_META,          "meta"          },
1544         { BLK_TC_DISCARD,       "discard"       },
1545         { BLK_TC_DRV_DATA,      "drv_data"      },
1546         { BLK_TC_FUA,           "fua"           },
1547 };
1548
1549 static int blk_trace_str2mask(const char *str)
1550 {
1551         int i;
1552         int mask = 0;
1553         char *buf, *s, *token;
1554
1555         buf = kstrdup(str, GFP_KERNEL);
1556         if (buf == NULL)
1557                 return -ENOMEM;
1558         s = strstrip(buf);
1559
1560         while (1) {
1561                 token = strsep(&s, ",");
1562                 if (token == NULL)
1563                         break;
1564
1565                 if (*token == '\0')
1566                         continue;
1567
1568                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1569                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1570                                 mask |= mask_maps[i].mask;
1571                                 break;
1572                         }
1573                 }
1574                 if (i == ARRAY_SIZE(mask_maps)) {
1575                         mask = -EINVAL;
1576                         break;
1577                 }
1578         }
1579         kfree(buf);
1580
1581         return mask;
1582 }
1583
1584 static ssize_t blk_trace_mask2str(char *buf, int mask)
1585 {
1586         int i;
1587         char *p = buf;
1588
1589         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1590                 if (mask & mask_maps[i].mask) {
1591                         p += sprintf(p, "%s%s",
1592                                     (p == buf) ? "" : ",", mask_maps[i].str);
1593                 }
1594         }
1595         *p++ = '\n';
1596
1597         return p - buf;
1598 }
1599
1600 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1601 {
1602         if (bdev->bd_disk == NULL)
1603                 return NULL;
1604
1605         return bdev_get_queue(bdev);
1606 }
1607
1608 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1609                                          struct device_attribute *attr,
1610                                          char *buf)
1611 {
1612         struct hd_struct *p = dev_to_part(dev);
1613         struct request_queue *q;
1614         struct block_device *bdev;
1615         ssize_t ret = -ENXIO;
1616
1617         bdev = bdget(part_devt(p));
1618         if (bdev == NULL)
1619                 goto out;
1620
1621         q = blk_trace_get_queue(bdev);
1622         if (q == NULL)
1623                 goto out_bdput;
1624
1625         mutex_lock(&bdev->bd_mutex);
1626
1627         if (attr == &dev_attr_enable) {
1628                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1629                 goto out_unlock_bdev;
1630         }
1631
1632         if (q->blk_trace == NULL)
1633                 ret = sprintf(buf, "disabled\n");
1634         else if (attr == &dev_attr_act_mask)
1635                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1636         else if (attr == &dev_attr_pid)
1637                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1638         else if (attr == &dev_attr_start_lba)
1639                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1640         else if (attr == &dev_attr_end_lba)
1641                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1642
1643 out_unlock_bdev:
1644         mutex_unlock(&bdev->bd_mutex);
1645 out_bdput:
1646         bdput(bdev);
1647 out:
1648         return ret;
1649 }
1650
1651 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1652                                           struct device_attribute *attr,
1653                                           const char *buf, size_t count)
1654 {
1655         struct block_device *bdev;
1656         struct request_queue *q;
1657         struct hd_struct *p;
1658         u64 value;
1659         ssize_t ret = -EINVAL;
1660
1661         if (count == 0)
1662                 goto out;
1663
1664         if (attr == &dev_attr_act_mask) {
1665                 if (sscanf(buf, "%llx", &value) != 1) {
1666                         /* Assume it is a list of trace category names */
1667                         ret = blk_trace_str2mask(buf);
1668                         if (ret < 0)
1669                                 goto out;
1670                         value = ret;
1671                 }
1672         } else if (sscanf(buf, "%llu", &value) != 1)
1673                 goto out;
1674
1675         ret = -ENXIO;
1676
1677         p = dev_to_part(dev);
1678         bdev = bdget(part_devt(p));
1679         if (bdev == NULL)
1680                 goto out;
1681
1682         q = blk_trace_get_queue(bdev);
1683         if (q == NULL)
1684                 goto out_bdput;
1685
1686         mutex_lock(&bdev->bd_mutex);
1687
1688         if (attr == &dev_attr_enable) {
1689                 if (value)
1690                         ret = blk_trace_setup_queue(q, bdev);
1691                 else
1692                         ret = blk_trace_remove_queue(q);
1693                 goto out_unlock_bdev;
1694         }
1695
1696         ret = 0;
1697         if (q->blk_trace == NULL)
1698                 ret = blk_trace_setup_queue(q, bdev);
1699
1700         if (ret == 0) {
1701                 if (attr == &dev_attr_act_mask)
1702                         q->blk_trace->act_mask = value;
1703                 else if (attr == &dev_attr_pid)
1704                         q->blk_trace->pid = value;
1705                 else if (attr == &dev_attr_start_lba)
1706                         q->blk_trace->start_lba = value;
1707                 else if (attr == &dev_attr_end_lba)
1708                         q->blk_trace->end_lba = value;
1709         }
1710
1711 out_unlock_bdev:
1712         mutex_unlock(&bdev->bd_mutex);
1713 out_bdput:
1714         bdput(bdev);
1715 out:
1716         return ret ? ret : count;
1717 }
1718
1719 int blk_trace_init_sysfs(struct device *dev)
1720 {
1721         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1722 }
1723
1724 void blk_trace_remove_sysfs(struct device *dev)
1725 {
1726         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1727 }
1728
1729 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1730
1731 #ifdef CONFIG_EVENT_TRACING
1732
1733 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1734 {
1735         int i = 0;
1736
1737         if (op & REQ_PREFLUSH)
1738                 rwbs[i++] = 'F';
1739
1740         switch (op & REQ_OP_MASK) {
1741         case REQ_OP_WRITE:
1742         case REQ_OP_WRITE_SAME:
1743                 rwbs[i++] = 'W';
1744                 break;
1745         case REQ_OP_DISCARD:
1746                 rwbs[i++] = 'D';
1747                 break;
1748         case REQ_OP_SECURE_ERASE:
1749                 rwbs[i++] = 'D';
1750                 rwbs[i++] = 'E';
1751                 break;
1752         case REQ_OP_FLUSH:
1753                 rwbs[i++] = 'F';
1754                 break;
1755         case REQ_OP_READ:
1756                 rwbs[i++] = 'R';
1757                 break;
1758         default:
1759                 rwbs[i++] = 'N';
1760         }
1761
1762         if (op & REQ_FUA)
1763                 rwbs[i++] = 'F';
1764         if (op & REQ_RAHEAD)
1765                 rwbs[i++] = 'A';
1766         if (op & REQ_SYNC)
1767                 rwbs[i++] = 'S';
1768         if (op & REQ_META)
1769                 rwbs[i++] = 'M';
1770
1771         rwbs[i] = '\0';
1772 }
1773 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1774
1775 #endif /* CONFIG_EVENT_TRACING */
1776