kernel/trace/blktrace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4  *
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/blktrace_api.h>
  12 #include <linux/percpu.h>
  13 #include <linux/init.h>
  14 #include <linux/mutex.h>
  15 #include <linux/slab.h>
  16 #include <linux/debugfs.h>
  17 #include <linux/export.h>
  18 #include <linux/time.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/list.h>
  21 #include <linux/blk-cgroup.h>
  22
  23 #include "../../block/blk.h"
  24
  25 #include <trace/events/block.h>
  26
  27 #include "trace_output.h"
  28
  29 #ifdef CONFIG_BLK_DEV_IO_TRACE
  30
  31 static unsigned int blktrace_seq __read_mostly = 1;
  32
  33 static struct trace_array *blk_tr;
  34 static bool blk_tracer_enabled __read_mostly;
  35
  36 static LIST_HEAD(running_trace_list);
  37 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  38
  39 /* Select an alternative, minimalistic output than the original one */
  40 #define TRACE_BLK_OPT_CLASSIC   0x1
  41 #define TRACE_BLK_OPT_CGROUP    0x2
  42 #define TRACE_BLK_OPT_CGNAME    0x4
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47 #ifdef CONFIG_BLK_CGROUP
  48         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  49         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  50 #endif
  51         { }
  52 };
  53
  54 static struct tracer_flags blk_tracer_flags = {
  55         .val  = 0,
  56         .opts = blk_tracer_opts,
  57 };
  58
  59 /* Global reference count of probes */
  60 static DEFINE_MUTEX(blk_probe_mutex);
  61 static int blk_probes_ref;
  62
  63 static void blk_register_tracepoints(void);
  64 static void blk_unregister_tracepoints(void);
  65
  66 /*
  67  * Send out a notify message.
  68  */
  69 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  70                        const void *data, size_t len,
  71                        union kernfs_node_id *cgid)
  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         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  80
  81         if (blk_tracer) {
  82                 buffer = blk_tr->trace_buffer.buffer;
  83                 pc = preempt_count();
  84                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  85                                                   sizeof(*t) + len + cgid_len,
  86                                                   0, pc);
  87                 if (!event)
  88                         return;
  89                 t = ring_buffer_event_data(event);
  90                 goto record_it;
  91         }
  92
  93         if (!bt->rchan)
  94                 return;
  95
  96         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  97         if (t) {
  98                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  99                 t->time = ktime_to_ns(ktime_get());
 100 record_it:
 101                 t->device = bt->dev;
 102                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 103                 t->pid = pid;
 104                 t->cpu = cpu;
 105                 t->pdu_len = len + cgid_len;
 106                 if (cgid)
 107                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 108                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 109
 110                 if (blk_tracer)
 111                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 112         }
 113 }
 114
 115 /*
 116  * Send out a notify for this process, if we haven't done so since a trace
 117  * started
 118  */
 119 static void trace_note_tsk(struct task_struct *tsk)
 120 {
 121         unsigned long flags;
 122         struct blk_trace *bt;
 123
 124         tsk->btrace_seq = blktrace_seq;
 125         spin_lock_irqsave(&running_trace_lock, flags);
 126         list_for_each_entry(bt, &running_trace_list, running_list) {
 127                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 128                            sizeof(tsk->comm), NULL);
 129         }
 130         spin_unlock_irqrestore(&running_trace_lock, flags);
 131 }
 132
 133 static void trace_note_time(struct blk_trace *bt)
 134 {
 135         struct timespec64 now;
 136         unsigned long flags;
 137         u32 words[2];
 138
 139         /* need to check user space to see if this breaks in y2038 or y2106 */
 140         ktime_get_real_ts64(&now);
 141         words[0] = (u32)now.tv_sec;
 142         words[1] = now.tv_nsec;
 143
 144         local_irq_save(flags);
 145         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 146         local_irq_restore(flags);
 147 }
 148
 149 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 150         const char *fmt, ...)
 151 {
 152         int n;
 153         va_list args;
 154         unsigned long flags;
 155         char *buf;
 156
 157         if (unlikely(bt->trace_state != Blktrace_running &&
 158                      !blk_tracer_enabled))
 159                 return;
 160
 161         /*
 162          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 163          * message to the trace.
 164          */
 165         if (!(bt->act_mask & BLK_TC_NOTIFY))
 166                 return;
 167
 168         local_irq_save(flags);
 169         buf = this_cpu_ptr(bt->msg_data);
 170         va_start(args, fmt);
 171         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 172         va_end(args);
 173
 174         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 175                 blkcg = NULL;
 176 #ifdef CONFIG_BLK_CGROUP
 177         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 178                 blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 179 #else
 180         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 181 #endif
 182         local_irq_restore(flags);
 183 }
 184 EXPORT_SYMBOL_GPL(__trace_note_message);
 185
 186 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 187                          pid_t pid)
 188 {
 189         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 190                 return 1;
 191         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 192                 return 1;
 193         if (bt->pid && pid != bt->pid)
 194                 return 1;
 195
 196         return 0;
 197 }
 198
 199 /*
 200  * Data direction bit lookup
 201  */
 202 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 203                                  BLK_TC_ACT(BLK_TC_WRITE) };
 204
 205 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 206 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 207
 208 /* The ilog2() calls fall out because they're constant */
 209 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 210           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 211
 212 /*
 213  * The worker for the various blk_add_trace*() types. Fills out a
 214  * blk_io_trace structure and places it in a per-cpu subbuffer.
 215  */
 216 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 217                      int op, int op_flags, u32 what, int error, int pdu_len,
 218                      void *pdu_data, union kernfs_node_id *cgid)
 219 {
 220         struct task_struct *tsk = current;
 221         struct ring_buffer_event *event = NULL;
 222         struct ring_buffer *buffer = NULL;
 223         struct blk_io_trace *t;
 224         unsigned long flags = 0;
 225         unsigned long *sequence;
 226         pid_t pid;
 227         int cpu, pc = 0;
 228         bool blk_tracer = blk_tracer_enabled;
 229         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 230
 231         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 232                 return;
 233
 234         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 235         what |= MASK_TC_BIT(op_flags, SYNC);
 236         what |= MASK_TC_BIT(op_flags, RAHEAD);
 237         what |= MASK_TC_BIT(op_flags, META);
 238         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 239         what |= MASK_TC_BIT(op_flags, FUA);
 240         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 241                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 242         if (op == REQ_OP_FLUSH)
 243                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 244         if (cgid)
 245                 what |= __BLK_TA_CGROUP;
 246
 247         pid = tsk->pid;
 248         if (act_log_check(bt, what, sector, pid))
 249                 return;
 250         cpu = raw_smp_processor_id();
 251
 252         if (blk_tracer) {
 253                 tracing_record_cmdline(current);
 254
 255                 buffer = blk_tr->trace_buffer.buffer;
 256                 pc = preempt_count();
 257                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 258                                                   sizeof(*t) + pdu_len + cgid_len,
 259                                                   0, pc);
 260                 if (!event)
 261                         return;
 262                 t = ring_buffer_event_data(event);
 263                 goto record_it;
 264         }
 265
 266         if (unlikely(tsk->btrace_seq != blktrace_seq))
 267                 trace_note_tsk(tsk);
 268
 269         /*
 270          * A word about the locking here - we disable interrupts to reserve
 271          * some space in the relay per-cpu buffer, to prevent an irq
 272          * from coming in and stepping on our toes.
 273          */
 274         local_irq_save(flags);
 275         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 276         if (t) {
 277                 sequence = per_cpu_ptr(bt->sequence, cpu);
 278
 279                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 280                 t->sequence = ++(*sequence);
 281                 t->time = ktime_to_ns(ktime_get());
 282 record_it:
 283                 /*
 284                  * These two are not needed in ftrace as they are in the
 285                  * generic trace_entry, filled by tracing_generic_entry_update,
 286                  * but for the trace_event->bin() synthesizer benefit we do it
 287                  * here too.
 288                  */
 289                 t->cpu = cpu;
 290                 t->pid = pid;
 291
 292                 t->sector = sector;
 293                 t->bytes = bytes;
 294                 t->action = what;
 295                 t->device = bt->dev;
 296                 t->error = error;
 297                 t->pdu_len = pdu_len + cgid_len;
 298
 299                 if (cgid_len)
 300                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 301                 if (pdu_len)
 302                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 303
 304                 if (blk_tracer) {
 305                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 306                         return;
 307                 }
 308         }
 309
 310         local_irq_restore(flags);
 311 }
 312
 313 static void blk_trace_free(struct blk_trace *bt)
 314 {
 315         debugfs_remove(bt->msg_file);
 316         debugfs_remove(bt->dropped_file);
 317         relay_close(bt->rchan);
 318         debugfs_remove(bt->dir);
 319         free_percpu(bt->sequence);
 320         free_percpu(bt->msg_data);
 321         kfree(bt);
 322 }
 323
 324 static void get_probe_ref(void)
 325 {
 326         mutex_lock(&blk_probe_mutex);
 327         if (++blk_probes_ref == 1)
 328                 blk_register_tracepoints();
 329         mutex_unlock(&blk_probe_mutex);
 330 }
 331
 332 static void put_probe_ref(void)
 333 {
 334         mutex_lock(&blk_probe_mutex);
 335         if (!--blk_probes_ref)
 336                 blk_unregister_tracepoints();
 337         mutex_unlock(&blk_probe_mutex);
 338 }
 339
 340 static void blk_trace_cleanup(struct blk_trace *bt)
 341 {
 342         synchronize_rcu();
 343         blk_trace_free(bt);
 344         put_probe_ref();
 345 }
 346
 347 static int __blk_trace_remove(struct request_queue *q)
 348 {
 349         struct blk_trace *bt;
 350
 351         bt = xchg(&q->blk_trace, NULL);
 352         if (!bt)
 353                 return -EINVAL;
 354
 355         if (bt->trace_state != Blktrace_running)
 356                 blk_trace_cleanup(bt);
 357
 358         return 0;
 359 }
 360
 361 int blk_trace_remove(struct request_queue *q)
 362 {
 363         int ret;
 364
 365         mutex_lock(&q->blk_trace_mutex);
 366         ret = __blk_trace_remove(q);
 367         mutex_unlock(&q->blk_trace_mutex);
 368
 369         return ret;
 370 }
 371 EXPORT_SYMBOL_GPL(blk_trace_remove);
 372
 373 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 374                                 size_t count, loff_t *ppos)
 375 {
 376         struct blk_trace *bt = filp->private_data;
 377         char buf[16];
 378
 379         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 380
 381         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 382 }
 383
 384 static const struct file_operations blk_dropped_fops = {
 385         .owner =        THIS_MODULE,
 386         .open =         simple_open,
 387         .read =         blk_dropped_read,
 388         .llseek =       default_llseek,
 389 };
 390
 391 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 392                                 size_t count, loff_t *ppos)
 393 {
 394         char *msg;
 395         struct blk_trace *bt;
 396
 397         if (count >= BLK_TN_MAX_MSG)
 398                 return -EINVAL;
 399
 400         msg = memdup_user_nul(buffer, count);
 401         if (IS_ERR(msg))
 402                 return PTR_ERR(msg);
 403
 404         bt = filp->private_data;
 405         __trace_note_message(bt, NULL, "%s", msg);
 406         kfree(msg);
 407
 408         return count;
 409 }
 410
 411 static const struct file_operations blk_msg_fops = {
 412         .owner =        THIS_MODULE,
 413         .open =         simple_open,
 414         .write =        blk_msg_write,
 415         .llseek =       noop_llseek,
 416 };
 417
 418 /*
 419  * Keep track of how many times we encountered a full subbuffer, to aid
 420  * the user space app in telling how many lost events there were.
 421  */
 422 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 423                                      void *prev_subbuf, size_t prev_padding)
 424 {
 425         struct blk_trace *bt;
 426
 427         if (!relay_buf_full(buf))
 428                 return 1;
 429
 430         bt = buf->chan->private_data;
 431         atomic_inc(&bt->dropped);
 432         return 0;
 433 }
 434
 435 static int blk_remove_buf_file_callback(struct dentry *dentry)
 436 {
 437         debugfs_remove(dentry);
 438
 439         return 0;
 440 }
 441
 442 static struct dentry *blk_create_buf_file_callback(const char *filename,
 443                                                    struct dentry *parent,
 444                                                    umode_t mode,
 445                                                    struct rchan_buf *buf,
 446                                                    int *is_global)
 447 {
 448         return debugfs_create_file(filename, mode, parent, buf,
 449                                         &relay_file_operations);
 450 }
 451
 452 static struct rchan_callbacks blk_relay_callbacks = {
 453         .subbuf_start           = blk_subbuf_start_callback,
 454         .create_buf_file        = blk_create_buf_file_callback,
 455         .remove_buf_file        = blk_remove_buf_file_callback,
 456 };
 457
 458 static void blk_trace_setup_lba(struct blk_trace *bt,
 459                                 struct block_device *bdev)
 460 {
 461         struct hd_struct *part = NULL;
 462
 463         if (bdev)
 464                 part = bdev->bd_part;
 465
 466         if (part) {
 467                 bt->start_lba = part->start_sect;
 468                 bt->end_lba = part->start_sect + part->nr_sects;
 469         } else {
 470                 bt->start_lba = 0;
 471                 bt->end_lba = -1ULL;
 472         }
 473 }
 474
 475 /*
 476  * Setup everything required to start tracing
 477  */
 478 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 479                               struct block_device *bdev,
 480                               struct blk_user_trace_setup *buts)
 481 {
 482         struct blk_trace *bt = NULL;
 483         struct dentry *dir = NULL;
 484         int ret;
 485
 486         if (!buts->buf_size || !buts->buf_nr)
 487                 return -EINVAL;
 488
 489         if (!blk_debugfs_root)
 490                 return -ENOENT;
 491
 492         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 493         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 494
 495         /*
 496          * some device names have larger paths - convert the slashes
 497          * to underscores for this to work as expected
 498          */
 499         strreplace(buts->name, '/', '_');
 500
 501         /*
 502          * bdev can be NULL, as with scsi-generic, this is a helpful as
 503          * we can be.
 504          */
 505         if (q->blk_trace) {
 506                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 507                         buts->name);
 508                 return -EBUSY;
 509         }
 510
 511         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 512         if (!bt)
 513                 return -ENOMEM;
 514
 515         ret = -ENOMEM;
 516         bt->sequence = alloc_percpu(unsigned long);
 517         if (!bt->sequence)
 518                 goto err;
 519
 520         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 521         if (!bt->msg_data)
 522                 goto err;
 523
 524 #ifdef CONFIG_BLK_DEBUG_FS
 525         /*
 526          * When tracing whole make_request drivers (multiqueue) block devices,
 527          * reuse the existing debugfs directory created by the block layer on
 528          * init. For request-based block devices, all partitions block devices,
 529          * and scsi-generic block devices we create a temporary new debugfs
 530          * directory that will be removed once the trace ends.
 531          */
 532         if (queue_is_mq(q) && bdev && bdev == bdev->bd_contains)
 533                 dir = q->debugfs_dir;
 534         else
 535 #endif
 536                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 537
 538         /*
 539          * As blktrace relies on debugfs for its interface the debugfs directory
 540          * is required, contrary to the usual mantra of not checking for debugfs
 541          * files or directories.
 542          */
 543         if (IS_ERR_OR_NULL(dir)) {
 544                 pr_warn("debugfs_dir not present for %s so skipping\n",
 545                         buts->name);
 546                 ret = -ENOENT;
 547                 goto err;
 548         }
 549
 550         bt->dev = dev;
 551         atomic_set(&bt->dropped, 0);
 552         INIT_LIST_HEAD(&bt->running_list);
 553
 554         ret = -EIO;
 555         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 556                                                &blk_dropped_fops);
 557
 558         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 559
 560         bt->rchan = relay_open("trace", dir, buts->buf_size,
 561                                 buts->buf_nr, &blk_relay_callbacks, bt);
 562         if (!bt->rchan)
 563                 goto err;
 564
 565         bt->act_mask = buts->act_mask;
 566         if (!bt->act_mask)
 567                 bt->act_mask = (u16) -1;
 568
 569         blk_trace_setup_lba(bt, bdev);
 570
 571         /* overwrite with user settings */
 572         if (buts->start_lba)
 573                 bt->start_lba = buts->start_lba;
 574         if (buts->end_lba)
 575                 bt->end_lba = buts->end_lba;
 576
 577         bt->pid = buts->pid;
 578         bt->trace_state = Blktrace_setup;
 579
 580         ret = -EBUSY;
 581         if (cmpxchg(&q->blk_trace, NULL, bt))
 582                 goto err;
 583
 584         get_probe_ref();
 585
 586         ret = 0;
 587 err:
 588         if (ret)
 589                 blk_trace_free(bt);
 590         return ret;
 591 }
 592
 593 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 594                              struct block_device *bdev, char __user *arg)
 595 {
 596         struct blk_user_trace_setup buts;
 597         int ret;
 598
 599         ret = copy_from_user(&buts, arg, sizeof(buts));
 600         if (ret)
 601                 return -EFAULT;
 602
 603         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 604         if (ret)
 605                 return ret;
 606
 607         if (copy_to_user(arg, &buts, sizeof(buts))) {
 608                 __blk_trace_remove(q);
 609                 return -EFAULT;
 610         }
 611         return 0;
 612 }
 613
 614 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 615                     struct block_device *bdev,
 616                     char __user *arg)
 617 {
 618         int ret;
 619
 620         mutex_lock(&q->blk_trace_mutex);
 621         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 622         mutex_unlock(&q->blk_trace_mutex);
 623
 624         return ret;
 625 }
 626 EXPORT_SYMBOL_GPL(blk_trace_setup);
 627
 628 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 629 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 630                                   dev_t dev, struct block_device *bdev,
 631                                   char __user *arg)
 632 {
 633         struct blk_user_trace_setup buts;
 634         struct compat_blk_user_trace_setup cbuts;
 635         int ret;
 636
 637         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 638                 return -EFAULT;
 639
 640         buts = (struct blk_user_trace_setup) {
 641                 .act_mask = cbuts.act_mask,
 642                 .buf_size = cbuts.buf_size,
 643                 .buf_nr = cbuts.buf_nr,
 644                 .start_lba = cbuts.start_lba,
 645                 .end_lba = cbuts.end_lba,
 646                 .pid = cbuts.pid,
 647         };
 648
 649         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 650         if (ret)
 651                 return ret;
 652
 653         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 654                 __blk_trace_remove(q);
 655                 return -EFAULT;
 656         }
 657
 658         return 0;
 659 }
 660 #endif
 661
 662 static int __blk_trace_startstop(struct request_queue *q, int start)
 663 {
 664         int ret;
 665         struct blk_trace *bt;
 666
 667         bt = rcu_dereference_protected(q->blk_trace,
 668                                        lockdep_is_held(&q->blk_trace_mutex));
 669         if (bt == NULL)
 670                 return -EINVAL;
 671
 672         /*
 673          * For starting a trace, we can transition from a setup or stopped
 674          * trace. For stopping a trace, the state must be running
 675          */
 676         ret = -EINVAL;
 677         if (start) {
 678                 if (bt->trace_state == Blktrace_setup ||
 679                     bt->trace_state == Blktrace_stopped) {
 680                         blktrace_seq++;
 681                         smp_mb();
 682                         bt->trace_state = Blktrace_running;
 683                         spin_lock_irq(&running_trace_lock);
 684                         list_add(&bt->running_list, &running_trace_list);
 685                         spin_unlock_irq(&running_trace_lock);
 686
 687                         trace_note_time(bt);
 688                         ret = 0;
 689                 }
 690         } else {
 691                 if (bt->trace_state == Blktrace_running) {
 692                         bt->trace_state = Blktrace_stopped;
 693                         spin_lock_irq(&running_trace_lock);
 694                         list_del_init(&bt->running_list);
 695                         spin_unlock_irq(&running_trace_lock);
 696                         relay_flush(bt->rchan);
 697                         ret = 0;
 698                 }
 699         }
 700
 701         return ret;
 702 }
 703
 704 int blk_trace_startstop(struct request_queue *q, int start)
 705 {
 706         int ret;
 707
 708         mutex_lock(&q->blk_trace_mutex);
 709         ret = __blk_trace_startstop(q, start);
 710         mutex_unlock(&q->blk_trace_mutex);
 711
 712         return ret;
 713 }
 714 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 715
 716 /*
 717  * When reading or writing the blktrace sysfs files, the references to the
 718  * opened sysfs or device files should prevent the underlying block device
 719  * from being removed. So no further delete protection is really needed.
 720  */
 721
 722 /**
 723  * blk_trace_ioctl: - handle the ioctls associated with tracing
 724  * @bdev:       the block device
 725  * @cmd:        the ioctl cmd
 726  * @arg:        the argument data, if any
 727  *
 728  **/
 729 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 730 {
 731         struct request_queue *q;
 732         int ret, start = 0;
 733         char b[BDEVNAME_SIZE];
 734
 735         q = bdev_get_queue(bdev);
 736         if (!q)
 737                 return -ENXIO;
 738
 739         mutex_lock(&q->blk_trace_mutex);
 740
 741         switch (cmd) {
 742         case BLKTRACESETUP:
 743                 bdevname(bdev, b);
 744                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 745                 break;
 746 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 747         case BLKTRACESETUP32:
 748                 bdevname(bdev, b);
 749                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 750                 break;
 751 #endif
 752         case BLKTRACESTART:
 753                 start = 1;
 754                 /* fall through */
 755         case BLKTRACESTOP:
 756                 ret = __blk_trace_startstop(q, start);
 757                 break;
 758         case BLKTRACETEARDOWN:
 759                 ret = __blk_trace_remove(q);
 760                 break;
 761         default:
 762                 ret = -ENOTTY;
 763                 break;
 764         }
 765
 766         mutex_unlock(&q->blk_trace_mutex);
 767         return ret;
 768 }
 769
 770 /**
 771  * blk_trace_shutdown: - stop and cleanup trace structures
 772  * @q:    the request queue associated with the device
 773  *
 774  **/
 775 void blk_trace_shutdown(struct request_queue *q)
 776 {
 777         mutex_lock(&q->blk_trace_mutex);
 778         if (rcu_dereference_protected(q->blk_trace,
 779                                       lockdep_is_held(&q->blk_trace_mutex))) {
 780                 __blk_trace_startstop(q, 0);
 781                 __blk_trace_remove(q);
 782         }
 783
 784         mutex_unlock(&q->blk_trace_mutex);
 785 }
 786
 787 #ifdef CONFIG_BLK_CGROUP
 788 static union kernfs_node_id *
 789 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 790 {
 791         struct blk_trace *bt;
 792
 793         /* We don't use the 'bt' value here except as an optimization... */
 794         bt = rcu_dereference_protected(q->blk_trace, 1);
 795         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 796                 return NULL;
 797
 798         if (!bio->bi_blkg)
 799                 return NULL;
 800         return cgroup_get_kernfs_id(bio_blkcg(bio)->css.cgroup);
 801 }
 802 #else
 803 static union kernfs_node_id *
 804 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 805 {
 806         return NULL;
 807 }
 808 #endif
 809
 810 static union kernfs_node_id *
 811 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 812 {
 813         if (!rq->bio)
 814                 return NULL;
 815         /* Use the first bio */
 816         return blk_trace_bio_get_cgid(q, rq->bio);
 817 }
 818
 819 /*
 820  * blktrace probes
 821  */
 822
 823 /**
 824  * blk_add_trace_rq - Add a trace for a request oriented action
 825  * @rq:         the source request
 826  * @error:      return status to log
 827  * @nr_bytes:   number of completed bytes
 828  * @what:       the action
 829  * @cgid:       the cgroup info
 830  *
 831  * Description:
 832  *     Records an action against a request. Will log the bio offset + size.
 833  *
 834  **/
 835 static void blk_add_trace_rq(struct request *rq, int error,
 836                              unsigned int nr_bytes, u32 what,
 837                              union kernfs_node_id *cgid)
 838 {
 839         struct blk_trace *bt;
 840
 841         rcu_read_lock();
 842         bt = rcu_dereference(rq->q->blk_trace);
 843         if (likely(!bt)) {
 844                 rcu_read_unlock();
 845                 return;
 846         }
 847
 848         if (blk_rq_is_passthrough(rq))
 849                 what |= BLK_TC_ACT(BLK_TC_PC);
 850         else
 851                 what |= BLK_TC_ACT(BLK_TC_FS);
 852
 853         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 854                         rq->cmd_flags, what, error, 0, NULL, cgid);
 855         rcu_read_unlock();
 856 }
 857
 858 static void blk_add_trace_rq_insert(void *ignore,
 859                                     struct request_queue *q, struct request *rq)
 860 {
 861         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 862                          blk_trace_request_get_cgid(q, rq));
 863 }
 864
 865 static void blk_add_trace_rq_issue(void *ignore,
 866                                    struct request_queue *q, struct request *rq)
 867 {
 868         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 869                          blk_trace_request_get_cgid(q, rq));
 870 }
 871
 872 static void blk_add_trace_rq_requeue(void *ignore,
 873                                      struct request_queue *q,
 874                                      struct request *rq)
 875 {
 876         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 877                          blk_trace_request_get_cgid(q, rq));
 878 }
 879
 880 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 881                         int error, unsigned int nr_bytes)
 882 {
 883         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 884                          blk_trace_request_get_cgid(rq->q, rq));
 885 }
 886
 887 /**
 888  * blk_add_trace_bio - Add a trace for a bio oriented action
 889  * @q:          queue the io is for
 890  * @bio:        the source bio
 891  * @what:       the action
 892  * @error:      error, if any
 893  *
 894  * Description:
 895  *     Records an action against a bio. Will log the bio offset + size.
 896  *
 897  **/
 898 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 899                               u32 what, int error)
 900 {
 901         struct blk_trace *bt;
 902
 903         rcu_read_lock();
 904         bt = rcu_dereference(q->blk_trace);
 905         if (likely(!bt)) {
 906                 rcu_read_unlock();
 907                 return;
 908         }
 909
 910         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 911                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 912                         blk_trace_bio_get_cgid(q, bio));
 913         rcu_read_unlock();
 914 }
 915
 916 static void blk_add_trace_bio_bounce(void *ignore,
 917                                      struct request_queue *q, struct bio *bio)
 918 {
 919         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 920 }
 921
 922 static void blk_add_trace_bio_complete(void *ignore,
 923                                        struct request_queue *q, struct bio *bio,
 924                                        int error)
 925 {
 926         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 927 }
 928
 929 static void blk_add_trace_bio_backmerge(void *ignore,
 930                                         struct request_queue *q,
 931                                         struct request *rq,
 932                                         struct bio *bio)
 933 {
 934         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 935 }
 936
 937 static void blk_add_trace_bio_frontmerge(void *ignore,
 938                                          struct request_queue *q,
 939                                          struct request *rq,
 940                                          struct bio *bio)
 941 {
 942         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 943 }
 944
 945 static void blk_add_trace_bio_queue(void *ignore,
 946                                     struct request_queue *q, struct bio *bio)
 947 {
 948         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 949 }
 950
 951 static void blk_add_trace_getrq(void *ignore,
 952                                 struct request_queue *q,
 953                                 struct bio *bio, int rw)
 954 {
 955         if (bio)
 956                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 957         else {
 958                 struct blk_trace *bt;
 959
 960                 rcu_read_lock();
 961                 bt = rcu_dereference(q->blk_trace);
 962                 if (bt)
 963                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 964                                         NULL, NULL);
 965                 rcu_read_unlock();
 966         }
 967 }
 968
 969
 970 static void blk_add_trace_sleeprq(void *ignore,
 971                                   struct request_queue *q,
 972                                   struct bio *bio, int rw)
 973 {
 974         if (bio)
 975                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 976         else {
 977                 struct blk_trace *bt;
 978
 979                 rcu_read_lock();
 980                 bt = rcu_dereference(q->blk_trace);
 981                 if (bt)
 982                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 983                                         0, 0, NULL, NULL);
 984                 rcu_read_unlock();
 985         }
 986 }
 987
 988 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 989 {
 990         struct blk_trace *bt;
 991
 992         rcu_read_lock();
 993         bt = rcu_dereference(q->blk_trace);
 994         if (bt)
 995                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
 996         rcu_read_unlock();
 997 }
 998
 999 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
1000                                     unsigned int depth, bool explicit)
1001 {
1002         struct blk_trace *bt;
1003
1004         rcu_read_lock();
1005         bt = rcu_dereference(q->blk_trace);
1006         if (bt) {
1007                 __be64 rpdu = cpu_to_be64(depth);
1008                 u32 what;
1009
1010                 if (explicit)
1011                         what = BLK_TA_UNPLUG_IO;
1012                 else
1013                         what = BLK_TA_UNPLUG_TIMER;
1014
1015                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
1016         }
1017         rcu_read_unlock();
1018 }
1019
1020 static void blk_add_trace_split(void *ignore,
1021                                 struct request_queue *q, struct bio *bio,
1022                                 unsigned int pdu)
1023 {
1024         struct blk_trace *bt;
1025
1026         rcu_read_lock();
1027         bt = rcu_dereference(q->blk_trace);
1028         if (bt) {
1029                 __be64 rpdu = cpu_to_be64(pdu);
1030
1031                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
1032                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
1033                                 BLK_TA_SPLIT,
1034                                 blk_status_to_errno(bio->bi_status),
1035                                 sizeof(rpdu), &rpdu,
1036                                 blk_trace_bio_get_cgid(q, bio));
1037         }
1038         rcu_read_unlock();
1039 }
1040
1041 /**
1042  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1043  * @ignore:     trace callback data parameter (not used)
1044  * @q:          queue the io is for
1045  * @bio:        the source bio
1046  * @dev:        target device
1047  * @from:       source sector
1048  *
1049  * Description:
1050  *     Device mapper or raid target sometimes need to split a bio because
1051  *     it spans a stripe (or similar). Add a trace for that action.
1052  *
1053  **/
1054 static void blk_add_trace_bio_remap(void *ignore,
1055                                     struct request_queue *q, struct bio *bio,
1056                                     dev_t dev, sector_t from)
1057 {
1058         struct blk_trace *bt;
1059         struct blk_io_trace_remap r;
1060
1061         rcu_read_lock();
1062         bt = rcu_dereference(q->blk_trace);
1063         if (likely(!bt)) {
1064                 rcu_read_unlock();
1065                 return;
1066         }
1067
1068         r.device_from = cpu_to_be32(dev);
1069         r.device_to   = cpu_to_be32(bio_dev(bio));
1070         r.sector_from = cpu_to_be64(from);
1071
1072         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1073                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1074                         blk_status_to_errno(bio->bi_status),
1075                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1076         rcu_read_unlock();
1077 }
1078
1079 /**
1080  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1081  * @ignore:     trace callback data parameter (not used)
1082  * @q:          queue the io is for
1083  * @rq:         the source request
1084  * @dev:        target device
1085  * @from:       source sector
1086  *
1087  * Description:
1088  *     Device mapper remaps request to other devices.
1089  *     Add a trace for that action.
1090  *
1091  **/
1092 static void blk_add_trace_rq_remap(void *ignore,
1093                                    struct request_queue *q,
1094                                    struct request *rq, dev_t dev,
1095                                    sector_t from)
1096 {
1097         struct blk_trace *bt;
1098         struct blk_io_trace_remap r;
1099
1100         rcu_read_lock();
1101         bt = rcu_dereference(q->blk_trace);
1102         if (likely(!bt)) {
1103                 rcu_read_unlock();
1104                 return;
1105         }
1106
1107         r.device_from = cpu_to_be32(dev);
1108         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1109         r.sector_from = cpu_to_be64(from);
1110
1111         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1112                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1113                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1114         rcu_read_unlock();
1115 }
1116
1117 /**
1118  * blk_add_driver_data - Add binary message with driver-specific data
1119  * @q:          queue the io is for
1120  * @rq:         io request
1121  * @data:       driver-specific data
1122  * @len:        length of driver-specific data
1123  *
1124  * Description:
1125  *     Some drivers might want to write driver-specific data per request.
1126  *
1127  **/
1128 void blk_add_driver_data(struct request_queue *q,
1129                          struct request *rq,
1130                          void *data, size_t len)
1131 {
1132         struct blk_trace *bt;
1133
1134         rcu_read_lock();
1135         bt = rcu_dereference(q->blk_trace);
1136         if (likely(!bt)) {
1137                 rcu_read_unlock();
1138                 return;
1139         }
1140
1141         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1142                                 BLK_TA_DRV_DATA, 0, len, data,
1143                                 blk_trace_request_get_cgid(q, rq));
1144         rcu_read_unlock();
1145 }
1146 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1147
1148 static void blk_register_tracepoints(void)
1149 {
1150         int ret;
1151
1152         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1153         WARN_ON(ret);
1154         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1155         WARN_ON(ret);
1156         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1157         WARN_ON(ret);
1158         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1159         WARN_ON(ret);
1160         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1161         WARN_ON(ret);
1162         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1163         WARN_ON(ret);
1164         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1165         WARN_ON(ret);
1166         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1167         WARN_ON(ret);
1168         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1169         WARN_ON(ret);
1170         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1171         WARN_ON(ret);
1172         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1173         WARN_ON(ret);
1174         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1175         WARN_ON(ret);
1176         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1177         WARN_ON(ret);
1178         ret = register_trace_block_split(blk_add_trace_split, NULL);
1179         WARN_ON(ret);
1180         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1181         WARN_ON(ret);
1182         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1183         WARN_ON(ret);
1184 }
1185
1186 static void blk_unregister_tracepoints(void)
1187 {
1188         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1189         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1190         unregister_trace_block_split(blk_add_trace_split, NULL);
1191         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1192         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1193         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1194         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1195         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1196         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1197         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1198         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1199         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1200         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1201         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1202         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1203         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1204
1205         tracepoint_synchronize_unregister();
1206 }
1207
1208 /*
1209  * struct blk_io_tracer formatting routines
1210  */
1211
1212 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1213 {
1214         int i = 0;
1215         int tc = t->action >> BLK_TC_SHIFT;
1216
1217         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1218                 rwbs[i++] = 'N';
1219                 goto out;
1220         }
1221
1222         if (tc & BLK_TC_FLUSH)
1223                 rwbs[i++] = 'F';
1224
1225         if (tc & BLK_TC_DISCARD)
1226                 rwbs[i++] = 'D';
1227         else if (tc & BLK_TC_WRITE)
1228                 rwbs[i++] = 'W';
1229         else if (t->bytes)
1230                 rwbs[i++] = 'R';
1231         else
1232                 rwbs[i++] = 'N';
1233
1234         if (tc & BLK_TC_FUA)
1235                 rwbs[i++] = 'F';
1236         if (tc & BLK_TC_AHEAD)
1237                 rwbs[i++] = 'A';
1238         if (tc & BLK_TC_SYNC)
1239                 rwbs[i++] = 'S';
1240         if (tc & BLK_TC_META)
1241                 rwbs[i++] = 'M';
1242 out:
1243         rwbs[i] = '\0';
1244 }
1245
1246 static inline
1247 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1248 {
1249         return (const struct blk_io_trace *)ent;
1250 }
1251
1252 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1253 {
1254         return (void *)(te_blk_io_trace(ent) + 1) +
1255                 (has_cg ? sizeof(union kernfs_node_id) : 0);
1256 }
1257
1258 static inline const void *cgid_start(const struct trace_entry *ent)
1259 {
1260         return (void *)(te_blk_io_trace(ent) + 1);
1261 }
1262
1263 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1264 {
1265         return te_blk_io_trace(ent)->pdu_len -
1266                         (has_cg ? sizeof(union kernfs_node_id) : 0);
1267 }
1268
1269 static inline u32 t_action(const struct trace_entry *ent)
1270 {
1271         return te_blk_io_trace(ent)->action;
1272 }
1273
1274 static inline u32 t_bytes(const struct trace_entry *ent)
1275 {
1276         return te_blk_io_trace(ent)->bytes;
1277 }
1278
1279 static inline u32 t_sec(const struct trace_entry *ent)
1280 {
1281         return te_blk_io_trace(ent)->bytes >> 9;
1282 }
1283
1284 static inline unsigned long long t_sector(const struct trace_entry *ent)
1285 {
1286         return te_blk_io_trace(ent)->sector;
1287 }
1288
1289 static inline __u16 t_error(const struct trace_entry *ent)
1290 {
1291         return te_blk_io_trace(ent)->error;
1292 }
1293
1294 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1295 {
1296         const __be64 *val = pdu_start(ent, has_cg);
1297         return be64_to_cpu(*val);
1298 }
1299
1300 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1301         bool has_cg);
1302
1303 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1304         bool has_cg)
1305 {
1306         char rwbs[RWBS_LEN];
1307         unsigned long long ts  = iter->ts;
1308         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1309         unsigned secs          = (unsigned long)ts;
1310         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1311
1312         fill_rwbs(rwbs, t);
1313
1314         trace_seq_printf(&iter->seq,
1315                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1316                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1317                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1318 }
1319
1320 static void blk_log_action(struct trace_iterator *iter, const char *act,
1321         bool has_cg)
1322 {
1323         char rwbs[RWBS_LEN];
1324         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1325
1326         fill_rwbs(rwbs, t);
1327         if (has_cg) {
1328                 const union kernfs_node_id *id = cgid_start(iter->ent);
1329
1330                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1331                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1332
1333                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1334                                 sizeof(blkcg_name_buf));
1335                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1336                                  MAJOR(t->device), MINOR(t->device),
1337                                  blkcg_name_buf, act, rwbs);
1338                 } else
1339                         trace_seq_printf(&iter->seq,
1340                                  "%3d,%-3d %x,%-x %2s %3s ",
1341                                  MAJOR(t->device), MINOR(t->device),
1342                                  id->ino, id->generation, act, rwbs);
1343         } else
1344                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1345                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1346 }
1347
1348 static void blk_log_dump_pdu(struct trace_seq *s,
1349         const struct trace_entry *ent, bool has_cg)
1350 {
1351         const unsigned char *pdu_buf;
1352         int pdu_len;
1353         int i, end;
1354
1355         pdu_buf = pdu_start(ent, has_cg);
1356         pdu_len = pdu_real_len(ent, has_cg);
1357
1358         if (!pdu_len)
1359                 return;
1360
1361         /* find the last zero that needs to be printed */
1362         for (end = pdu_len - 1; end >= 0; end--)
1363                 if (pdu_buf[end])
1364                         break;
1365         end++;
1366
1367         trace_seq_putc(s, '(');
1368
1369         for (i = 0; i < pdu_len; i++) {
1370
1371                 trace_seq_printf(s, "%s%02x",
1372                                  i == 0 ? "" : " ", pdu_buf[i]);
1373
1374                 /*
1375                  * stop when the rest is just zeroes and indicate so
1376                  * with a ".." appended
1377                  */
1378                 if (i == end && end != pdu_len - 1) {
1379                         trace_seq_puts(s, " ..) ");
1380                         return;
1381                 }
1382         }
1383
1384         trace_seq_puts(s, ") ");
1385 }
1386
1387 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1388 {
1389         char cmd[TASK_COMM_LEN];
1390
1391         trace_find_cmdline(ent->pid, cmd);
1392
1393         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1394                 trace_seq_printf(s, "%u ", t_bytes(ent));
1395                 blk_log_dump_pdu(s, ent, has_cg);
1396                 trace_seq_printf(s, "[%s]\n", cmd);
1397         } else {
1398                 if (t_sec(ent))
1399                         trace_seq_printf(s, "%llu + %u [%s]\n",
1400                                                 t_sector(ent), t_sec(ent), cmd);
1401                 else
1402                         trace_seq_printf(s, "[%s]\n", cmd);
1403         }
1404 }
1405
1406 static void blk_log_with_error(struct trace_seq *s,
1407                               const struct trace_entry *ent, bool has_cg)
1408 {
1409         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1410                 blk_log_dump_pdu(s, ent, has_cg);
1411                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1412         } else {
1413                 if (t_sec(ent))
1414                         trace_seq_printf(s, "%llu + %u [%d]\n",
1415                                          t_sector(ent),
1416                                          t_sec(ent), t_error(ent));
1417                 else
1418                         trace_seq_printf(s, "%llu [%d]\n",
1419                                          t_sector(ent), t_error(ent));
1420         }
1421 }
1422
1423 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1424 {
1425         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1426
1427         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1428                          t_sector(ent), t_sec(ent),
1429                          MAJOR(be32_to_cpu(__r->device_from)),
1430                          MINOR(be32_to_cpu(__r->device_from)),
1431                          be64_to_cpu(__r->sector_from));
1432 }
1433
1434 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1435 {
1436         char cmd[TASK_COMM_LEN];
1437
1438         trace_find_cmdline(ent->pid, cmd);
1439
1440         trace_seq_printf(s, "[%s]\n", cmd);
1441 }
1442
1443 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1444 {
1445         char cmd[TASK_COMM_LEN];
1446
1447         trace_find_cmdline(ent->pid, cmd);
1448
1449         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1450 }
1451
1452 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1453 {
1454         char cmd[TASK_COMM_LEN];
1455
1456         trace_find_cmdline(ent->pid, cmd);
1457
1458         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1459                          get_pdu_int(ent, has_cg), cmd);
1460 }
1461
1462 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1463                         bool has_cg)
1464 {
1465
1466         trace_seq_putmem(s, pdu_start(ent, has_cg),
1467                 pdu_real_len(ent, has_cg));
1468         trace_seq_putc(s, '\n');
1469 }
1470
1471 /*
1472  * struct tracer operations
1473  */
1474
1475 static void blk_tracer_print_header(struct seq_file *m)
1476 {
1477         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1478                 return;
1479         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1480                     "#  |     |     |           |   |   |\n");
1481 }
1482
1483 static void blk_tracer_start(struct trace_array *tr)
1484 {
1485         blk_tracer_enabled = true;
1486 }
1487
1488 static int blk_tracer_init(struct trace_array *tr)
1489 {
1490         blk_tr = tr;
1491         blk_tracer_start(tr);
1492         return 0;
1493 }
1494
1495 static void blk_tracer_stop(struct trace_array *tr)
1496 {
1497         blk_tracer_enabled = false;
1498 }
1499
1500 static void blk_tracer_reset(struct trace_array *tr)
1501 {
1502         blk_tracer_stop(tr);
1503 }
1504
1505 static const struct {
1506         const char *act[2];
1507         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1508                             bool has_cg);
1509 } what2act[] = {
1510         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1511         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1512         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1513         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1514         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1515         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1516         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1517         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1518         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1519         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1520         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1521         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1522         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1523         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1524         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1525 };
1526
1527 static enum print_line_t print_one_line(struct trace_iterator *iter,
1528                                         bool classic)
1529 {
1530         struct trace_array *tr = iter->tr;
1531         struct trace_seq *s = &iter->seq;
1532         const struct blk_io_trace *t;
1533         u16 what;
1534         bool long_act;
1535         blk_log_action_t *log_action;
1536         bool has_cg;
1537
1538         t          = te_blk_io_trace(iter->ent);
1539         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1540         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1541         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1542         has_cg     = t->action & __BLK_TA_CGROUP;
1543
1544         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1545                 log_action(iter, long_act ? "message" : "m", has_cg);
1546                 blk_log_msg(s, iter->ent, has_cg);
1547                 return trace_handle_return(s);
1548         }
1549
1550         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1551                 trace_seq_printf(s, "Unknown action %x\n", what);
1552         else {
1553                 log_action(iter, what2act[what].act[long_act], has_cg);
1554                 what2act[what].print(s, iter->ent, has_cg);
1555         }
1556
1557         return trace_handle_return(s);
1558 }
1559
1560 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1561                                                int flags, struct trace_event *event)
1562 {
1563         return print_one_line(iter, false);
1564 }
1565
1566 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1567 {
1568         struct trace_seq *s = &iter->seq;
1569         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1570         const int offset = offsetof(struct blk_io_trace, sector);
1571         struct blk_io_trace old = {
1572                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1573                 .time     = iter->ts,
1574         };
1575
1576         trace_seq_putmem(s, &old, offset);
1577         trace_seq_putmem(s, &t->sector,
1578                          sizeof(old) - offset + t->pdu_len);
1579 }
1580
1581 static enum print_line_t
1582 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1583                              struct trace_event *event)
1584 {
1585         blk_trace_synthesize_old_trace(iter);
1586
1587         return trace_handle_return(&iter->seq);
1588 }
1589
1590 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1591 {
1592         if ((iter->ent->type != TRACE_BLK) ||
1593             !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1594                 return TRACE_TYPE_UNHANDLED;
1595
1596         return print_one_line(iter, true);
1597 }
1598
1599 static int
1600 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1601 {
1602         /* don't output context-info for blk_classic output */
1603         if (bit == TRACE_BLK_OPT_CLASSIC) {
1604                 if (set)
1605                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1606                 else
1607                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1608         }
1609         return 0;
1610 }
1611
1612 static struct tracer blk_tracer __read_mostly = {
1613         .name           = "blk",
1614         .init           = blk_tracer_init,
1615         .reset          = blk_tracer_reset,
1616         .start          = blk_tracer_start,
1617         .stop           = blk_tracer_stop,
1618         .print_header   = blk_tracer_print_header,
1619         .print_line     = blk_tracer_print_line,
1620         .flags          = &blk_tracer_flags,
1621         .set_flag       = blk_tracer_set_flag,
1622 };
1623
1624 static struct trace_event_functions trace_blk_event_funcs = {
1625         .trace          = blk_trace_event_print,
1626         .binary         = blk_trace_event_print_binary,
1627 };
1628
1629 static struct trace_event trace_blk_event = {
1630         .type           = TRACE_BLK,
1631         .funcs          = &trace_blk_event_funcs,
1632 };
1633
1634 static int __init init_blk_tracer(void)
1635 {
1636         if (!register_trace_event(&trace_blk_event)) {
1637                 pr_warn("Warning: could not register block events\n");
1638                 return 1;
1639         }
1640
1641         if (register_tracer(&blk_tracer) != 0) {
1642                 pr_warn("Warning: could not register the block tracer\n");
1643                 unregister_trace_event(&trace_blk_event);
1644                 return 1;
1645         }
1646
1647         return 0;
1648 }
1649
1650 device_initcall(init_blk_tracer);
1651
1652 static int blk_trace_remove_queue(struct request_queue *q)
1653 {
1654         struct blk_trace *bt;
1655
1656         bt = xchg(&q->blk_trace, NULL);
1657         if (bt == NULL)
1658                 return -EINVAL;
1659
1660         if (bt->trace_state == Blktrace_running) {
1661                 bt->trace_state = Blktrace_stopped;
1662                 spin_lock_irq(&running_trace_lock);
1663                 list_del_init(&bt->running_list);
1664                 spin_unlock_irq(&running_trace_lock);
1665                 relay_flush(bt->rchan);
1666         }
1667
1668         put_probe_ref();
1669         synchronize_rcu();
1670         blk_trace_free(bt);
1671         return 0;
1672 }
1673
1674 /*
1675  * Setup everything required to start tracing
1676  */
1677 static int blk_trace_setup_queue(struct request_queue *q,
1678                                  struct block_device *bdev)
1679 {
1680         struct blk_trace *bt = NULL;
1681         int ret = -ENOMEM;
1682
1683         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1684         if (!bt)
1685                 return -ENOMEM;
1686
1687         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1688         if (!bt->msg_data)
1689                 goto free_bt;
1690
1691         bt->dev = bdev->bd_dev;
1692         bt->act_mask = (u16)-1;
1693
1694         blk_trace_setup_lba(bt, bdev);
1695
1696         ret = -EBUSY;
1697         if (cmpxchg(&q->blk_trace, NULL, bt))
1698                 goto free_bt;
1699
1700         get_probe_ref();
1701         return 0;
1702
1703 free_bt:
1704         blk_trace_free(bt);
1705         return ret;
1706 }
1707
1708 /*
1709  * sysfs interface to enable and configure tracing
1710  */
1711
1712 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1713                                          struct device_attribute *attr,
1714                                          char *buf);
1715 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1716                                           struct device_attribute *attr,
1717                                           const char *buf, size_t count);
1718 #define BLK_TRACE_DEVICE_ATTR(_name) \
1719         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1720                     sysfs_blk_trace_attr_show, \
1721                     sysfs_blk_trace_attr_store)
1722
1723 static BLK_TRACE_DEVICE_ATTR(enable);
1724 static BLK_TRACE_DEVICE_ATTR(act_mask);
1725 static BLK_TRACE_DEVICE_ATTR(pid);
1726 static BLK_TRACE_DEVICE_ATTR(start_lba);
1727 static BLK_TRACE_DEVICE_ATTR(end_lba);
1728
1729 static struct attribute *blk_trace_attrs[] = {
1730         &dev_attr_enable.attr,
1731         &dev_attr_act_mask.attr,
1732         &dev_attr_pid.attr,
1733         &dev_attr_start_lba.attr,
1734         &dev_attr_end_lba.attr,
1735         NULL
1736 };
1737
1738 struct attribute_group blk_trace_attr_group = {
1739         .name  = "trace",
1740         .attrs = blk_trace_attrs,
1741 };
1742
1743 static const struct {
1744         int mask;
1745         const char *str;
1746 } mask_maps[] = {
1747         { BLK_TC_READ,          "read"          },
1748         { BLK_TC_WRITE,         "write"         },
1749         { BLK_TC_FLUSH,         "flush"         },
1750         { BLK_TC_SYNC,          "sync"          },
1751         { BLK_TC_QUEUE,         "queue"         },
1752         { BLK_TC_REQUEUE,       "requeue"       },
1753         { BLK_TC_ISSUE,         "issue"         },
1754         { BLK_TC_COMPLETE,      "complete"      },
1755         { BLK_TC_FS,            "fs"            },
1756         { BLK_TC_PC,            "pc"            },
1757         { BLK_TC_NOTIFY,        "notify"        },
1758         { BLK_TC_AHEAD,         "ahead"         },
1759         { BLK_TC_META,          "meta"          },
1760         { BLK_TC_DISCARD,       "discard"       },
1761         { BLK_TC_DRV_DATA,      "drv_data"      },
1762         { BLK_TC_FUA,           "fua"           },
1763 };
1764
1765 static int blk_trace_str2mask(const char *str)
1766 {
1767         int i;
1768         int mask = 0;
1769         char *buf, *s, *token;
1770
1771         buf = kstrdup(str, GFP_KERNEL);
1772         if (buf == NULL)
1773                 return -ENOMEM;
1774         s = strstrip(buf);
1775
1776         while (1) {
1777                 token = strsep(&s, ",");
1778                 if (token == NULL)
1779                         break;
1780
1781                 if (*token == '\0')
1782                         continue;
1783
1784                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1785                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1786                                 mask |= mask_maps[i].mask;
1787                                 break;
1788                         }
1789                 }
1790                 if (i == ARRAY_SIZE(mask_maps)) {
1791                         mask = -EINVAL;
1792                         break;
1793                 }
1794         }
1795         kfree(buf);
1796
1797         return mask;
1798 }
1799
1800 static ssize_t blk_trace_mask2str(char *buf, int mask)
1801 {
1802         int i;
1803         char *p = buf;
1804
1805         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1806                 if (mask & mask_maps[i].mask) {
1807                         p += sprintf(p, "%s%s",
1808                                     (p == buf) ? "" : ",", mask_maps[i].str);
1809                 }
1810         }
1811         *p++ = '\n';
1812
1813         return p - buf;
1814 }
1815
1816 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1817 {
1818         if (bdev->bd_disk == NULL)
1819                 return NULL;
1820
1821         return bdev_get_queue(bdev);
1822 }
1823
1824 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1825                                          struct device_attribute *attr,
1826                                          char *buf)
1827 {
1828         struct hd_struct *p = dev_to_part(dev);
1829         struct request_queue *q;
1830         struct block_device *bdev;
1831         struct blk_trace *bt;
1832         ssize_t ret = -ENXIO;
1833
1834         bdev = bdget(part_devt(p));
1835         if (bdev == NULL)
1836                 goto out;
1837
1838         q = blk_trace_get_queue(bdev);
1839         if (q == NULL)
1840                 goto out_bdput;
1841
1842         mutex_lock(&q->blk_trace_mutex);
1843
1844         bt = rcu_dereference_protected(q->blk_trace,
1845                                        lockdep_is_held(&q->blk_trace_mutex));
1846         if (attr == &dev_attr_enable) {
1847                 ret = sprintf(buf, "%u\n", !!bt);
1848                 goto out_unlock_bdev;
1849         }
1850
1851         if (bt == NULL)
1852                 ret = sprintf(buf, "disabled\n");
1853         else if (attr == &dev_attr_act_mask)
1854                 ret = blk_trace_mask2str(buf, bt->act_mask);
1855         else if (attr == &dev_attr_pid)
1856                 ret = sprintf(buf, "%u\n", bt->pid);
1857         else if (attr == &dev_attr_start_lba)
1858                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1859         else if (attr == &dev_attr_end_lba)
1860                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1861
1862 out_unlock_bdev:
1863         mutex_unlock(&q->blk_trace_mutex);
1864 out_bdput:
1865         bdput(bdev);
1866 out:
1867         return ret;
1868 }
1869
1870 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1871                                           struct device_attribute *attr,
1872                                           const char *buf, size_t count)
1873 {
1874         struct block_device *bdev;
1875         struct request_queue *q;
1876         struct hd_struct *p;
1877         struct blk_trace *bt;
1878         u64 value;
1879         ssize_t ret = -EINVAL;
1880
1881         if (count == 0)
1882                 goto out;
1883
1884         if (attr == &dev_attr_act_mask) {
1885                 if (kstrtoull(buf, 0, &value)) {
1886                         /* Assume it is a list of trace category names */
1887                         ret = blk_trace_str2mask(buf);
1888                         if (ret < 0)
1889                                 goto out;
1890                         value = ret;
1891                 }
1892         } else if (kstrtoull(buf, 0, &value))
1893                 goto out;
1894
1895         ret = -ENXIO;
1896
1897         p = dev_to_part(dev);
1898         bdev = bdget(part_devt(p));
1899         if (bdev == NULL)
1900                 goto out;
1901
1902         q = blk_trace_get_queue(bdev);
1903         if (q == NULL)
1904                 goto out_bdput;
1905
1906         mutex_lock(&q->blk_trace_mutex);
1907
1908         bt = rcu_dereference_protected(q->blk_trace,
1909                                        lockdep_is_held(&q->blk_trace_mutex));
1910         if (attr == &dev_attr_enable) {
1911                 if (!!value == !!bt) {
1912                         ret = 0;
1913                         goto out_unlock_bdev;
1914                 }
1915                 if (value)
1916                         ret = blk_trace_setup_queue(q, bdev);
1917                 else
1918                         ret = blk_trace_remove_queue(q);
1919                 goto out_unlock_bdev;
1920         }
1921
1922         ret = 0;
1923         if (bt == NULL) {
1924                 ret = blk_trace_setup_queue(q, bdev);
1925                 bt = rcu_dereference_protected(q->blk_trace,
1926                                 lockdep_is_held(&q->blk_trace_mutex));
1927         }
1928
1929         if (ret == 0) {
1930                 if (attr == &dev_attr_act_mask)
1931                         bt->act_mask = value;
1932                 else if (attr == &dev_attr_pid)
1933                         bt->pid = value;
1934                 else if (attr == &dev_attr_start_lba)
1935                         bt->start_lba = value;
1936                 else if (attr == &dev_attr_end_lba)
1937                         bt->end_lba = value;
1938         }
1939
1940 out_unlock_bdev:
1941         mutex_unlock(&q->blk_trace_mutex);
1942 out_bdput:
1943         bdput(bdev);
1944 out:
1945         return ret ? ret : count;
1946 }
1947
1948 int blk_trace_init_sysfs(struct device *dev)
1949 {
1950         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1951 }
1952
1953 void blk_trace_remove_sysfs(struct device *dev)
1954 {
1955         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1956 }
1957
1958 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1959
1960 #ifdef CONFIG_EVENT_TRACING
1961
1962 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1963 {
1964         int i = 0;
1965
1966         if (op & REQ_PREFLUSH)
1967                 rwbs[i++] = 'F';
1968
1969         switch (op & REQ_OP_MASK) {
1970         case REQ_OP_WRITE:
1971         case REQ_OP_WRITE_SAME:
1972                 rwbs[i++] = 'W';
1973                 break;
1974         case REQ_OP_DISCARD:
1975                 rwbs[i++] = 'D';
1976                 break;
1977         case REQ_OP_SECURE_ERASE:
1978                 rwbs[i++] = 'D';
1979                 rwbs[i++] = 'E';
1980                 break;
1981         case REQ_OP_FLUSH:
1982                 rwbs[i++] = 'F';
1983                 break;
1984         case REQ_OP_READ:
1985                 rwbs[i++] = 'R';
1986                 break;
1987         default:
1988                 rwbs[i++] = 'N';
1989         }
1990
1991         if (op & REQ_FUA)
1992                 rwbs[i++] = 'F';
1993         if (op & REQ_RAHEAD)
1994                 rwbs[i++] = 'A';
1995         if (op & REQ_SYNC)
1996                 rwbs[i++] = 'S';
1997         if (op & REQ_META)
1998                 rwbs[i++] = 'M';
1999
2000         rwbs[i] = '\0';
2001 }
2002 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
2003
2004 #endif /* CONFIG_EVENT_TRACING */
2005