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  * @q:          queue the io is for
 694  * @rq:         the source request
 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_queue *q, struct request *rq,
 703                              unsigned int nr_bytes, u32 what)
 704 {
 705         struct blk_trace *bt = 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, rq->errors, 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(q, rq, 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(q, rq, 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(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 736 }
 737
 738 static void blk_add_trace_rq_complete(void *ignore,
 739                                       struct request_queue *q,
 740                                       struct request *rq,
 741                                       unsigned int nr_bytes)
 742 {
 743         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 744 }
 745
 746 /**
 747  * blk_add_trace_bio - Add a trace for a bio oriented action
 748  * @q:          queue the io is for
 749  * @bio:        the source bio
 750  * @what:       the action
 751  * @error:      error, if any
 752  *
 753  * Description:
 754  *     Records an action against a bio. Will log the bio offset + size.
 755  *
 756  **/
 757 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 758                               u32 what, int error)
 759 {
 760         struct blk_trace *bt = q->blk_trace;
 761
 762         if (likely(!bt))
 763                 return;
 764
 765         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 766                         bio_op(bio), bio->bi_opf, what, error, 0, NULL);
 767 }
 768
 769 static void blk_add_trace_bio_bounce(void *ignore,
 770                                      struct request_queue *q, struct bio *bio)
 771 {
 772         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 773 }
 774
 775 static void blk_add_trace_bio_complete(void *ignore,
 776                                        struct request_queue *q, struct bio *bio,
 777                                        int error)
 778 {
 779         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 780 }
 781
 782 static void blk_add_trace_bio_backmerge(void *ignore,
 783                                         struct request_queue *q,
 784                                         struct request *rq,
 785                                         struct bio *bio)
 786 {
 787         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 788 }
 789
 790 static void blk_add_trace_bio_frontmerge(void *ignore,
 791                                          struct request_queue *q,
 792                                          struct request *rq,
 793                                          struct bio *bio)
 794 {
 795         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 796 }
 797
 798 static void blk_add_trace_bio_queue(void *ignore,
 799                                     struct request_queue *q, struct bio *bio)
 800 {
 801         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 802 }
 803
 804 static void blk_add_trace_getrq(void *ignore,
 805                                 struct request_queue *q,
 806                                 struct bio *bio, int rw)
 807 {
 808         if (bio)
 809                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 810         else {
 811                 struct blk_trace *bt = q->blk_trace;
 812
 813                 if (bt)
 814                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 815                                         NULL);
 816         }
 817 }
 818
 819
 820 static void blk_add_trace_sleeprq(void *ignore,
 821                                   struct request_queue *q,
 822                                   struct bio *bio, int rw)
 823 {
 824         if (bio)
 825                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 826         else {
 827                 struct blk_trace *bt = q->blk_trace;
 828
 829                 if (bt)
 830                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 831                                         0, 0, NULL);
 832         }
 833 }
 834
 835 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 836 {
 837         struct blk_trace *bt = q->blk_trace;
 838
 839         if (bt)
 840                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 841 }
 842
 843 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 844                                     unsigned int depth, bool explicit)
 845 {
 846         struct blk_trace *bt = q->blk_trace;
 847
 848         if (bt) {
 849                 __be64 rpdu = cpu_to_be64(depth);
 850                 u32 what;
 851
 852                 if (explicit)
 853                         what = BLK_TA_UNPLUG_IO;
 854                 else
 855                         what = BLK_TA_UNPLUG_TIMER;
 856
 857                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 858         }
 859 }
 860
 861 static void blk_add_trace_split(void *ignore,
 862                                 struct request_queue *q, struct bio *bio,
 863                                 unsigned int pdu)
 864 {
 865         struct blk_trace *bt = q->blk_trace;
 866
 867         if (bt) {
 868                 __be64 rpdu = cpu_to_be64(pdu);
 869
 870                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 871                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 872                                 BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
 873                                 &rpdu);
 874         }
 875 }
 876
 877 /**
 878  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 879  * @ignore:     trace callback data parameter (not used)
 880  * @q:          queue the io is for
 881  * @bio:        the source bio
 882  * @dev:        target device
 883  * @from:       source sector
 884  *
 885  * Description:
 886  *     Device mapper or raid target sometimes need to split a bio because
 887  *     it spans a stripe (or similar). Add a trace for that action.
 888  *
 889  **/
 890 static void blk_add_trace_bio_remap(void *ignore,
 891                                     struct request_queue *q, struct bio *bio,
 892                                     dev_t dev, sector_t from)
 893 {
 894         struct blk_trace *bt = q->blk_trace;
 895         struct blk_io_trace_remap r;
 896
 897         if (likely(!bt))
 898                 return;
 899
 900         r.device_from = cpu_to_be32(dev);
 901         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 902         r.sector_from = cpu_to_be64(from);
 903
 904         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 905                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_error,
 906                         sizeof(r), &r);
 907 }
 908
 909 /**
 910  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 911  * @ignore:     trace callback data parameter (not used)
 912  * @q:          queue the io is for
 913  * @rq:         the source request
 914  * @dev:        target device
 915  * @from:       source sector
 916  *
 917  * Description:
 918  *     Device mapper remaps request to other devices.
 919  *     Add a trace for that action.
 920  *
 921  **/
 922 static void blk_add_trace_rq_remap(void *ignore,
 923                                    struct request_queue *q,
 924                                    struct request *rq, dev_t dev,
 925                                    sector_t from)
 926 {
 927         struct blk_trace *bt = q->blk_trace;
 928         struct blk_io_trace_remap r;
 929
 930         if (likely(!bt))
 931                 return;
 932
 933         r.device_from = cpu_to_be32(dev);
 934         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 935         r.sector_from = cpu_to_be64(from);
 936
 937         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 938                         rq_data_dir(rq), 0, BLK_TA_REMAP, !!rq->errors,
 939                         sizeof(r), &r);
 940 }
 941
 942 /**
 943  * blk_add_driver_data - Add binary message with driver-specific data
 944  * @q:          queue the io is for
 945  * @rq:         io request
 946  * @data:       driver-specific data
 947  * @len:        length of driver-specific data
 948  *
 949  * Description:
 950  *     Some drivers might want to write driver-specific data per request.
 951  *
 952  **/
 953 void blk_add_driver_data(struct request_queue *q,
 954                          struct request *rq,
 955                          void *data, size_t len)
 956 {
 957         struct blk_trace *bt = q->blk_trace;
 958
 959         if (likely(!bt))
 960                 return;
 961
 962         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
 963                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 964 }
 965 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 966
 967 static void blk_register_tracepoints(void)
 968 {
 969         int ret;
 970
 971         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 972         WARN_ON(ret);
 973         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 974         WARN_ON(ret);
 975         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 976         WARN_ON(ret);
 977         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 978         WARN_ON(ret);
 979         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 980         WARN_ON(ret);
 981         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
 982         WARN_ON(ret);
 983         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
 984         WARN_ON(ret);
 985         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
 986         WARN_ON(ret);
 987         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
 988         WARN_ON(ret);
 989         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
 990         WARN_ON(ret);
 991         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
 992         WARN_ON(ret);
 993         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
 994         WARN_ON(ret);
 995         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
 996         WARN_ON(ret);
 997         ret = register_trace_block_split(blk_add_trace_split, NULL);
 998         WARN_ON(ret);
 999         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1000         WARN_ON(ret);
1001         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1002         WARN_ON(ret);
1003 }
1004
1005 static void blk_unregister_tracepoints(void)
1006 {
1007         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1008         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1009         unregister_trace_block_split(blk_add_trace_split, NULL);
1010         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1011         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1012         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1013         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1014         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1015         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1016         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1017         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1018         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1019         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1020         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1021         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1022         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1023
1024         tracepoint_synchronize_unregister();
1025 }
1026
1027 /*
1028  * struct blk_io_tracer formatting routines
1029  */
1030
1031 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1032 {
1033         int i = 0;
1034         int tc = t->action >> BLK_TC_SHIFT;
1035
1036         if (t->action == BLK_TN_MESSAGE) {
1037                 rwbs[i++] = 'N';
1038                 goto out;
1039         }
1040
1041         if (tc & BLK_TC_FLUSH)
1042                 rwbs[i++] = 'F';
1043
1044         if (tc & BLK_TC_DISCARD)
1045                 rwbs[i++] = 'D';
1046         else if (tc & BLK_TC_WRITE)
1047                 rwbs[i++] = 'W';
1048         else if (t->bytes)
1049                 rwbs[i++] = 'R';
1050         else
1051                 rwbs[i++] = 'N';
1052
1053         if (tc & BLK_TC_FUA)
1054                 rwbs[i++] = 'F';
1055         if (tc & BLK_TC_AHEAD)
1056                 rwbs[i++] = 'A';
1057         if (tc & BLK_TC_SYNC)
1058                 rwbs[i++] = 'S';
1059         if (tc & BLK_TC_META)
1060                 rwbs[i++] = 'M';
1061 out:
1062         rwbs[i] = '\0';
1063 }
1064
1065 static inline
1066 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1067 {
1068         return (const struct blk_io_trace *)ent;
1069 }
1070
1071 static inline const void *pdu_start(const struct trace_entry *ent)
1072 {
1073         return te_blk_io_trace(ent) + 1;
1074 }
1075
1076 static inline u32 t_action(const struct trace_entry *ent)
1077 {
1078         return te_blk_io_trace(ent)->action;
1079 }
1080
1081 static inline u32 t_bytes(const struct trace_entry *ent)
1082 {
1083         return te_blk_io_trace(ent)->bytes;
1084 }
1085
1086 static inline u32 t_sec(const struct trace_entry *ent)
1087 {
1088         return te_blk_io_trace(ent)->bytes >> 9;
1089 }
1090
1091 static inline unsigned long long t_sector(const struct trace_entry *ent)
1092 {
1093         return te_blk_io_trace(ent)->sector;
1094 }
1095
1096 static inline __u16 t_error(const struct trace_entry *ent)
1097 {
1098         return te_blk_io_trace(ent)->error;
1099 }
1100
1101 static __u64 get_pdu_int(const struct trace_entry *ent)
1102 {
1103         const __u64 *val = pdu_start(ent);
1104         return be64_to_cpu(*val);
1105 }
1106
1107 static void get_pdu_remap(const struct trace_entry *ent,
1108                           struct blk_io_trace_remap *r)
1109 {
1110         const struct blk_io_trace_remap *__r = pdu_start(ent);
1111         __u64 sector_from = __r->sector_from;
1112
1113         r->device_from = be32_to_cpu(__r->device_from);
1114         r->device_to   = be32_to_cpu(__r->device_to);
1115         r->sector_from = be64_to_cpu(sector_from);
1116 }
1117
1118 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1119
1120 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1121 {
1122         char rwbs[RWBS_LEN];
1123         unsigned long long ts  = iter->ts;
1124         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1125         unsigned secs          = (unsigned long)ts;
1126         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1127
1128         fill_rwbs(rwbs, t);
1129
1130         trace_seq_printf(&iter->seq,
1131                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1132                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1133                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1134 }
1135
1136 static void blk_log_action(struct trace_iterator *iter, const char *act)
1137 {
1138         char rwbs[RWBS_LEN];
1139         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1140
1141         fill_rwbs(rwbs, t);
1142         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1143                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1144 }
1145
1146 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1147 {
1148         const unsigned char *pdu_buf;
1149         int pdu_len;
1150         int i, end;
1151
1152         pdu_buf = pdu_start(ent);
1153         pdu_len = te_blk_io_trace(ent)->pdu_len;
1154
1155         if (!pdu_len)
1156                 return;
1157
1158         /* find the last zero that needs to be printed */
1159         for (end = pdu_len - 1; end >= 0; end--)
1160                 if (pdu_buf[end])
1161                         break;
1162         end++;
1163
1164         trace_seq_putc(s, '(');
1165
1166         for (i = 0; i < pdu_len; i++) {
1167
1168                 trace_seq_printf(s, "%s%02x",
1169                                  i == 0 ? "" : " ", pdu_buf[i]);
1170
1171                 /*
1172                  * stop when the rest is just zeroes and indicate so
1173                  * with a ".." appended
1174                  */
1175                 if (i == end && end != pdu_len - 1) {
1176                         trace_seq_puts(s, " ..) ");
1177                         return;
1178                 }
1179         }
1180
1181         trace_seq_puts(s, ") ");
1182 }
1183
1184 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1185 {
1186         char cmd[TASK_COMM_LEN];
1187
1188         trace_find_cmdline(ent->pid, cmd);
1189
1190         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1191                 trace_seq_printf(s, "%u ", t_bytes(ent));
1192                 blk_log_dump_pdu(s, ent);
1193                 trace_seq_printf(s, "[%s]\n", cmd);
1194         } else {
1195                 if (t_sec(ent))
1196                         trace_seq_printf(s, "%llu + %u [%s]\n",
1197                                                 t_sector(ent), t_sec(ent), cmd);
1198                 else
1199                         trace_seq_printf(s, "[%s]\n", cmd);
1200         }
1201 }
1202
1203 static void blk_log_with_error(struct trace_seq *s,
1204                               const struct trace_entry *ent)
1205 {
1206         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1207                 blk_log_dump_pdu(s, ent);
1208                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1209         } else {
1210                 if (t_sec(ent))
1211                         trace_seq_printf(s, "%llu + %u [%d]\n",
1212                                          t_sector(ent),
1213                                          t_sec(ent), t_error(ent));
1214                 else
1215                         trace_seq_printf(s, "%llu [%d]\n",
1216                                          t_sector(ent), t_error(ent));
1217         }
1218 }
1219
1220 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1221 {
1222         struct blk_io_trace_remap r = { .device_from = 0, };
1223
1224         get_pdu_remap(ent, &r);
1225         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1226                          t_sector(ent), t_sec(ent),
1227                          MAJOR(r.device_from), MINOR(r.device_from),
1228                          (unsigned long long)r.sector_from);
1229 }
1230
1231 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1232 {
1233         char cmd[TASK_COMM_LEN];
1234
1235         trace_find_cmdline(ent->pid, cmd);
1236
1237         trace_seq_printf(s, "[%s]\n", cmd);
1238 }
1239
1240 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1241 {
1242         char cmd[TASK_COMM_LEN];
1243
1244         trace_find_cmdline(ent->pid, cmd);
1245
1246         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1247 }
1248
1249 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1250 {
1251         char cmd[TASK_COMM_LEN];
1252
1253         trace_find_cmdline(ent->pid, cmd);
1254
1255         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1256                          get_pdu_int(ent), cmd);
1257 }
1258
1259 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1260 {
1261         const struct blk_io_trace *t = te_blk_io_trace(ent);
1262
1263         trace_seq_putmem(s, t + 1, t->pdu_len);
1264         trace_seq_putc(s, '\n');
1265 }
1266
1267 /*
1268  * struct tracer operations
1269  */
1270
1271 static void blk_tracer_print_header(struct seq_file *m)
1272 {
1273         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1274                 return;
1275         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1276                     "#  |     |     |           |   |   |\n");
1277 }
1278
1279 static void blk_tracer_start(struct trace_array *tr)
1280 {
1281         blk_tracer_enabled = true;
1282 }
1283
1284 static int blk_tracer_init(struct trace_array *tr)
1285 {
1286         blk_tr = tr;
1287         blk_tracer_start(tr);
1288         return 0;
1289 }
1290
1291 static void blk_tracer_stop(struct trace_array *tr)
1292 {
1293         blk_tracer_enabled = false;
1294 }
1295
1296 static void blk_tracer_reset(struct trace_array *tr)
1297 {
1298         blk_tracer_stop(tr);
1299 }
1300
1301 static const struct {
1302         const char *act[2];
1303         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1304 } what2act[] = {
1305         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1306         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1307         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1308         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1309         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1310         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1311         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1312         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1313         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1314         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1315         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1316         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1317         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1318         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1319         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1320 };
1321
1322 static enum print_line_t print_one_line(struct trace_iterator *iter,
1323                                         bool classic)
1324 {
1325         struct trace_array *tr = iter->tr;
1326         struct trace_seq *s = &iter->seq;
1327         const struct blk_io_trace *t;
1328         u16 what;
1329         bool long_act;
1330         blk_log_action_t *log_action;
1331
1332         t          = te_blk_io_trace(iter->ent);
1333         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1334         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1335         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1336
1337         if (t->action == BLK_TN_MESSAGE) {
1338                 log_action(iter, long_act ? "message" : "m");
1339                 blk_log_msg(s, iter->ent);
1340                 return trace_handle_return(s);
1341         }
1342
1343         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1344                 trace_seq_printf(s, "Unknown action %x\n", what);
1345         else {
1346                 log_action(iter, what2act[what].act[long_act]);
1347                 what2act[what].print(s, iter->ent);
1348         }
1349
1350         return trace_handle_return(s);
1351 }
1352
1353 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1354                                                int flags, struct trace_event *event)
1355 {
1356         return print_one_line(iter, false);
1357 }
1358
1359 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1360 {
1361         struct trace_seq *s = &iter->seq;
1362         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1363         const int offset = offsetof(struct blk_io_trace, sector);
1364         struct blk_io_trace old = {
1365                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1366                 .time     = iter->ts,
1367         };
1368
1369         trace_seq_putmem(s, &old, offset);
1370         trace_seq_putmem(s, &t->sector,
1371                          sizeof(old) - offset + t->pdu_len);
1372 }
1373
1374 static enum print_line_t
1375 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1376                              struct trace_event *event)
1377 {
1378         blk_trace_synthesize_old_trace(iter);
1379
1380         return trace_handle_return(&iter->seq);
1381 }
1382
1383 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1384 {
1385         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1386                 return TRACE_TYPE_UNHANDLED;
1387
1388         return print_one_line(iter, true);
1389 }
1390
1391 static int
1392 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1393 {
1394         /* don't output context-info for blk_classic output */
1395         if (bit == TRACE_BLK_OPT_CLASSIC) {
1396                 if (set)
1397                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1398                 else
1399                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1400         }
1401         return 0;
1402 }
1403
1404 static struct tracer blk_tracer __read_mostly = {
1405         .name           = "blk",
1406         .init           = blk_tracer_init,
1407         .reset          = blk_tracer_reset,
1408         .start          = blk_tracer_start,
1409         .stop           = blk_tracer_stop,
1410         .print_header   = blk_tracer_print_header,
1411         .print_line     = blk_tracer_print_line,
1412         .flags          = &blk_tracer_flags,
1413         .set_flag       = blk_tracer_set_flag,
1414 };
1415
1416 static struct trace_event_functions trace_blk_event_funcs = {
1417         .trace          = blk_trace_event_print,
1418         .binary         = blk_trace_event_print_binary,
1419 };
1420
1421 static struct trace_event trace_blk_event = {
1422         .type           = TRACE_BLK,
1423         .funcs          = &trace_blk_event_funcs,
1424 };
1425
1426 static int __init init_blk_tracer(void)
1427 {
1428         if (!register_trace_event(&trace_blk_event)) {
1429                 pr_warn("Warning: could not register block events\n");
1430                 return 1;
1431         }
1432
1433         if (register_tracer(&blk_tracer) != 0) {
1434                 pr_warn("Warning: could not register the block tracer\n");
1435                 unregister_trace_event(&trace_blk_event);
1436                 return 1;
1437         }
1438
1439         return 0;
1440 }
1441
1442 device_initcall(init_blk_tracer);
1443
1444 static int blk_trace_remove_queue(struct request_queue *q)
1445 {
1446         struct blk_trace *bt;
1447
1448         bt = xchg(&q->blk_trace, NULL);
1449         if (bt == NULL)
1450                 return -EINVAL;
1451
1452         if (atomic_dec_and_test(&blk_probes_ref))
1453                 blk_unregister_tracepoints();
1454
1455         blk_trace_free(bt);
1456         return 0;
1457 }
1458
1459 /*
1460  * Setup everything required to start tracing
1461  */
1462 static int blk_trace_setup_queue(struct request_queue *q,
1463                                  struct block_device *bdev)
1464 {
1465         struct blk_trace *bt = NULL;
1466         int ret = -ENOMEM;
1467
1468         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1469         if (!bt)
1470                 return -ENOMEM;
1471
1472         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1473         if (!bt->msg_data)
1474                 goto free_bt;
1475
1476         bt->dev = bdev->bd_dev;
1477         bt->act_mask = (u16)-1;
1478
1479         blk_trace_setup_lba(bt, bdev);
1480
1481         ret = -EBUSY;
1482         if (cmpxchg(&q->blk_trace, NULL, bt))
1483                 goto free_bt;
1484
1485         if (atomic_inc_return(&blk_probes_ref) == 1)
1486                 blk_register_tracepoints();
1487         return 0;
1488
1489 free_bt:
1490         blk_trace_free(bt);
1491         return ret;
1492 }
1493
1494 /*
1495  * sysfs interface to enable and configure tracing
1496  */
1497
1498 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1499                                          struct device_attribute *attr,
1500                                          char *buf);
1501 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1502                                           struct device_attribute *attr,
1503                                           const char *buf, size_t count);
1504 #define BLK_TRACE_DEVICE_ATTR(_name) \
1505         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1506                     sysfs_blk_trace_attr_show, \
1507                     sysfs_blk_trace_attr_store)
1508
1509 static BLK_TRACE_DEVICE_ATTR(enable);
1510 static BLK_TRACE_DEVICE_ATTR(act_mask);
1511 static BLK_TRACE_DEVICE_ATTR(pid);
1512 static BLK_TRACE_DEVICE_ATTR(start_lba);
1513 static BLK_TRACE_DEVICE_ATTR(end_lba);
1514
1515 static struct attribute *blk_trace_attrs[] = {
1516         &dev_attr_enable.attr,
1517         &dev_attr_act_mask.attr,
1518         &dev_attr_pid.attr,
1519         &dev_attr_start_lba.attr,
1520         &dev_attr_end_lba.attr,
1521         NULL
1522 };
1523
1524 struct attribute_group blk_trace_attr_group = {
1525         .name  = "trace",
1526         .attrs = blk_trace_attrs,
1527 };
1528
1529 static const struct {
1530         int mask;
1531         const char *str;
1532 } mask_maps[] = {
1533         { BLK_TC_READ,          "read"          },
1534         { BLK_TC_WRITE,         "write"         },
1535         { BLK_TC_FLUSH,         "flush"         },
1536         { BLK_TC_SYNC,          "sync"          },
1537         { BLK_TC_QUEUE,         "queue"         },
1538         { BLK_TC_REQUEUE,       "requeue"       },
1539         { BLK_TC_ISSUE,         "issue"         },
1540         { BLK_TC_COMPLETE,      "complete"      },
1541         { BLK_TC_FS,            "fs"            },
1542         { BLK_TC_PC,            "pc"            },
1543         { BLK_TC_NOTIFY,        "notify"        },
1544         { BLK_TC_AHEAD,         "ahead"         },
1545         { BLK_TC_META,          "meta"          },
1546         { BLK_TC_DISCARD,       "discard"       },
1547         { BLK_TC_DRV_DATA,      "drv_data"      },
1548         { BLK_TC_FUA,           "fua"           },
1549 };
1550
1551 static int blk_trace_str2mask(const char *str)
1552 {
1553         int i;
1554         int mask = 0;
1555         char *buf, *s, *token;
1556
1557         buf = kstrdup(str, GFP_KERNEL);
1558         if (buf == NULL)
1559                 return -ENOMEM;
1560         s = strstrip(buf);
1561
1562         while (1) {
1563                 token = strsep(&s, ",");
1564                 if (token == NULL)
1565                         break;
1566
1567                 if (*token == '\0')
1568                         continue;
1569
1570                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1571                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1572                                 mask |= mask_maps[i].mask;
1573                                 break;
1574                         }
1575                 }
1576                 if (i == ARRAY_SIZE(mask_maps)) {
1577                         mask = -EINVAL;
1578                         break;
1579                 }
1580         }
1581         kfree(buf);
1582
1583         return mask;
1584 }
1585
1586 static ssize_t blk_trace_mask2str(char *buf, int mask)
1587 {
1588         int i;
1589         char *p = buf;
1590
1591         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1592                 if (mask & mask_maps[i].mask) {
1593                         p += sprintf(p, "%s%s",
1594                                     (p == buf) ? "" : ",", mask_maps[i].str);
1595                 }
1596         }
1597         *p++ = '\n';
1598
1599         return p - buf;
1600 }
1601
1602 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1603 {
1604         if (bdev->bd_disk == NULL)
1605                 return NULL;
1606
1607         return bdev_get_queue(bdev);
1608 }
1609
1610 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1611                                          struct device_attribute *attr,
1612                                          char *buf)
1613 {
1614         struct hd_struct *p = dev_to_part(dev);
1615         struct request_queue *q;
1616         struct block_device *bdev;
1617         ssize_t ret = -ENXIO;
1618
1619         bdev = bdget(part_devt(p));
1620         if (bdev == NULL)
1621                 goto out;
1622
1623         q = blk_trace_get_queue(bdev);
1624         if (q == NULL)
1625                 goto out_bdput;
1626
1627         mutex_lock(&bdev->bd_mutex);
1628
1629         if (attr == &dev_attr_enable) {
1630                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1631                 goto out_unlock_bdev;
1632         }
1633
1634         if (q->blk_trace == NULL)
1635                 ret = sprintf(buf, "disabled\n");
1636         else if (attr == &dev_attr_act_mask)
1637                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1638         else if (attr == &dev_attr_pid)
1639                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1640         else if (attr == &dev_attr_start_lba)
1641                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1642         else if (attr == &dev_attr_end_lba)
1643                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1644
1645 out_unlock_bdev:
1646         mutex_unlock(&bdev->bd_mutex);
1647 out_bdput:
1648         bdput(bdev);
1649 out:
1650         return ret;
1651 }
1652
1653 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1654                                           struct device_attribute *attr,
1655                                           const char *buf, size_t count)
1656 {
1657         struct block_device *bdev;
1658         struct request_queue *q;
1659         struct hd_struct *p;
1660         u64 value;
1661         ssize_t ret = -EINVAL;
1662
1663         if (count == 0)
1664                 goto out;
1665
1666         if (attr == &dev_attr_act_mask) {
1667                 if (sscanf(buf, "%llx", &value) != 1) {
1668                         /* Assume it is a list of trace category names */
1669                         ret = blk_trace_str2mask(buf);
1670                         if (ret < 0)
1671                                 goto out;
1672                         value = ret;
1673                 }
1674         } else if (sscanf(buf, "%llu", &value) != 1)
1675                 goto out;
1676
1677         ret = -ENXIO;
1678
1679         p = dev_to_part(dev);
1680         bdev = bdget(part_devt(p));
1681         if (bdev == NULL)
1682                 goto out;
1683
1684         q = blk_trace_get_queue(bdev);
1685         if (q == NULL)
1686                 goto out_bdput;
1687
1688         mutex_lock(&bdev->bd_mutex);
1689
1690         if (attr == &dev_attr_enable) {
1691                 if (value)
1692                         ret = blk_trace_setup_queue(q, bdev);
1693                 else
1694                         ret = blk_trace_remove_queue(q);
1695                 goto out_unlock_bdev;
1696         }
1697
1698         ret = 0;
1699         if (q->blk_trace == NULL)
1700                 ret = blk_trace_setup_queue(q, bdev);
1701
1702         if (ret == 0) {
1703                 if (attr == &dev_attr_act_mask)
1704                         q->blk_trace->act_mask = value;
1705                 else if (attr == &dev_attr_pid)
1706                         q->blk_trace->pid = value;
1707                 else if (attr == &dev_attr_start_lba)
1708                         q->blk_trace->start_lba = value;
1709                 else if (attr == &dev_attr_end_lba)
1710                         q->blk_trace->end_lba = value;
1711         }
1712
1713 out_unlock_bdev:
1714         mutex_unlock(&bdev->bd_mutex);
1715 out_bdput:
1716         bdput(bdev);
1717 out:
1718         return ret ? ret : count;
1719 }
1720
1721 int blk_trace_init_sysfs(struct device *dev)
1722 {
1723         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1724 }
1725
1726 void blk_trace_remove_sysfs(struct device *dev)
1727 {
1728         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1729 }
1730
1731 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1732
1733 #ifdef CONFIG_EVENT_TRACING
1734
1735 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1736 {
1737         int i = 0;
1738
1739         if (op & REQ_PREFLUSH)
1740                 rwbs[i++] = 'F';
1741
1742         switch (op & REQ_OP_MASK) {
1743         case REQ_OP_WRITE:
1744         case REQ_OP_WRITE_SAME:
1745                 rwbs[i++] = 'W';
1746                 break;
1747         case REQ_OP_DISCARD:
1748                 rwbs[i++] = 'D';
1749                 break;
1750         case REQ_OP_SECURE_ERASE:
1751                 rwbs[i++] = 'D';
1752                 rwbs[i++] = 'E';
1753                 break;
1754         case REQ_OP_FLUSH:
1755                 rwbs[i++] = 'F';
1756                 break;
1757         case REQ_OP_READ:
1758                 rwbs[i++] = 'R';
1759                 break;
1760         default:
1761                 rwbs[i++] = 'N';
1762         }
1763
1764         if (op & REQ_FUA)
1765                 rwbs[i++] = 'F';
1766         if (op & REQ_RAHEAD)
1767                 rwbs[i++] = 'A';
1768         if (op & REQ_SYNC)
1769                 rwbs[i++] = 'S';
1770         if (op & REQ_META)
1771                 rwbs[i++] = 'M';
1772
1773         rwbs[i] = '\0';
1774 }
1775 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1776
1777 #endif /* CONFIG_EVENT_TRACING */
1778