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, u64 cgid)
  71 {
  72         struct blk_io_trace *t;
  73         struct ring_buffer_event *event = NULL;
  74         struct trace_buffer *buffer = NULL;
  75         int pc = 0;
  76         int cpu = smp_processor_id();
  77         bool blk_tracer = blk_tracer_enabled;
  78         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
  79
  80         if (blk_tracer) {
  81                 buffer = blk_tr->array_buffer.buffer;
  82                 pc = preempt_count();
  83                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84                                                   sizeof(*t) + len + cgid_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 + cgid_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 | (cgid ? __BLK_TN_CGROUP : 0);
 102                 t->pid = pid;
 103                 t->cpu = cpu;
 104                 t->pdu_len = len + cgid_len;
 105                 if (cgid_len)
 106                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 107                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 108
 109                 if (blk_tracer)
 110                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 111         }
 112 }
 113
 114 /*
 115  * Send out a notify for this process, if we haven't done so since a trace
 116  * started
 117  */
 118 static void trace_note_tsk(struct task_struct *tsk)
 119 {
 120         unsigned long flags;
 121         struct blk_trace *bt;
 122
 123         tsk->btrace_seq = blktrace_seq;
 124         spin_lock_irqsave(&running_trace_lock, flags);
 125         list_for_each_entry(bt, &running_trace_list, running_list) {
 126                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 127                            sizeof(tsk->comm), 0);
 128         }
 129         spin_unlock_irqrestore(&running_trace_lock, flags);
 130 }
 131
 132 static void trace_note_time(struct blk_trace *bt)
 133 {
 134         struct timespec64 now;
 135         unsigned long flags;
 136         u32 words[2];
 137
 138         /* need to check user space to see if this breaks in y2038 or y2106 */
 139         ktime_get_real_ts64(&now);
 140         words[0] = (u32)now.tv_sec;
 141         words[1] = now.tv_nsec;
 142
 143         local_irq_save(flags);
 144         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
 145         local_irq_restore(flags);
 146 }
 147
 148 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 149         const char *fmt, ...)
 150 {
 151         int n;
 152         va_list args;
 153         unsigned long flags;
 154         char *buf;
 155
 156         if (unlikely(bt->trace_state != Blktrace_running &&
 157                      !blk_tracer_enabled))
 158                 return;
 159
 160         /*
 161          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 162          * message to the trace.
 163          */
 164         if (!(bt->act_mask & BLK_TC_NOTIFY))
 165                 return;
 166
 167         local_irq_save(flags);
 168         buf = this_cpu_ptr(bt->msg_data);
 169         va_start(args, fmt);
 170         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 171         va_end(args);
 172
 173         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 174                 blkcg = NULL;
 175 #ifdef CONFIG_BLK_CGROUP
 176         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n,
 177                    blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
 178 #else
 179         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, 0);
 180 #endif
 181         local_irq_restore(flags);
 182 }
 183 EXPORT_SYMBOL_GPL(__trace_note_message);
 184
 185 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 186                          pid_t pid)
 187 {
 188         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 189                 return 1;
 190         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 191                 return 1;
 192         if (bt->pid && pid != bt->pid)
 193                 return 1;
 194
 195         return 0;
 196 }
 197
 198 /*
 199  * Data direction bit lookup
 200  */
 201 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 202                                  BLK_TC_ACT(BLK_TC_WRITE) };
 203
 204 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 205 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 206
 207 /* The ilog2() calls fall out because they're constant */
 208 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 209           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 210
 211 /*
 212  * The worker for the various blk_add_trace*() types. Fills out a
 213  * blk_io_trace structure and places it in a per-cpu subbuffer.
 214  */
 215 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 216                      int op, int op_flags, u32 what, int error, int pdu_len,
 217                      void *pdu_data, u64 cgid)
 218 {
 219         struct task_struct *tsk = current;
 220         struct ring_buffer_event *event = NULL;
 221         struct trace_buffer *buffer = NULL;
 222         struct blk_io_trace *t;
 223         unsigned long flags = 0;
 224         unsigned long *sequence;
 225         pid_t pid;
 226         int cpu, pc = 0;
 227         bool blk_tracer = blk_tracer_enabled;
 228         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
 229
 230         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 231                 return;
 232
 233         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 234         what |= MASK_TC_BIT(op_flags, SYNC);
 235         what |= MASK_TC_BIT(op_flags, RAHEAD);
 236         what |= MASK_TC_BIT(op_flags, META);
 237         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 238         what |= MASK_TC_BIT(op_flags, FUA);
 239         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 240                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 241         if (op == REQ_OP_FLUSH)
 242                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 243         if (cgid)
 244                 what |= __BLK_TA_CGROUP;
 245
 246         pid = tsk->pid;
 247         if (act_log_check(bt, what, sector, pid))
 248                 return;
 249         cpu = raw_smp_processor_id();
 250
 251         if (blk_tracer) {
 252                 tracing_record_cmdline(current);
 253
 254                 buffer = blk_tr->array_buffer.buffer;
 255                 pc = preempt_count();
 256                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 257                                                   sizeof(*t) + pdu_len + cgid_len,
 258                                                   0, pc);
 259                 if (!event)
 260                         return;
 261                 t = ring_buffer_event_data(event);
 262                 goto record_it;
 263         }
 264
 265         if (unlikely(tsk->btrace_seq != blktrace_seq))
 266                 trace_note_tsk(tsk);
 267
 268         /*
 269          * A word about the locking here - we disable interrupts to reserve
 270          * some space in the relay per-cpu buffer, to prevent an irq
 271          * from coming in and stepping on our toes.
 272          */
 273         local_irq_save(flags);
 274         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 275         if (t) {
 276                 sequence = per_cpu_ptr(bt->sequence, cpu);
 277
 278                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 279                 t->sequence = ++(*sequence);
 280                 t->time = ktime_to_ns(ktime_get());
 281 record_it:
 282                 /*
 283                  * These two are not needed in ftrace as they are in the
 284                  * generic trace_entry, filled by tracing_generic_entry_update,
 285                  * but for the trace_event->bin() synthesizer benefit we do it
 286                  * here too.
 287                  */
 288                 t->cpu = cpu;
 289                 t->pid = pid;
 290
 291                 t->sector = sector;
 292                 t->bytes = bytes;
 293                 t->action = what;
 294                 t->device = bt->dev;
 295                 t->error = error;
 296                 t->pdu_len = pdu_len + cgid_len;
 297
 298                 if (cgid_len)
 299                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 300                 if (pdu_len)
 301                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 302
 303                 if (blk_tracer) {
 304                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 305                         return;
 306                 }
 307         }
 308
 309         local_irq_restore(flags);
 310 }
 311
 312 static void blk_trace_free(struct blk_trace *bt)
 313 {
 314         debugfs_remove(bt->msg_file);
 315         debugfs_remove(bt->dropped_file);
 316         relay_close(bt->rchan);
 317         debugfs_remove(bt->dir);
 318         free_percpu(bt->sequence);
 319         free_percpu(bt->msg_data);
 320         kfree(bt);
 321 }
 322
 323 static void get_probe_ref(void)
 324 {
 325         mutex_lock(&blk_probe_mutex);
 326         if (++blk_probes_ref == 1)
 327                 blk_register_tracepoints();
 328         mutex_unlock(&blk_probe_mutex);
 329 }
 330
 331 static void put_probe_ref(void)
 332 {
 333         mutex_lock(&blk_probe_mutex);
 334         if (!--blk_probes_ref)
 335                 blk_unregister_tracepoints();
 336         mutex_unlock(&blk_probe_mutex);
 337 }
 338
 339 static void blk_trace_cleanup(struct blk_trace *bt)
 340 {
 341         synchronize_rcu();
 342         blk_trace_free(bt);
 343         put_probe_ref();
 344 }
 345
 346 static int __blk_trace_remove(struct request_queue *q)
 347 {
 348         struct blk_trace *bt;
 349
 350         bt = rcu_replace_pointer(q->blk_trace, NULL,
 351                                  lockdep_is_held(&q->debugfs_mutex));
 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->debugfs_mutex);
 366         ret = __blk_trace_remove(q);
 367         mutex_unlock(&q->debugfs_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         lockdep_assert_held(&q->debugfs_mutex);
 487
 488         if (!buts->buf_size || !buts->buf_nr)
 489                 return -EINVAL;
 490
 491         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 492         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 493
 494         /*
 495          * some device names have larger paths - convert the slashes
 496          * to underscores for this to work as expected
 497          */
 498         strreplace(buts->name, '/', '_');
 499
 500         /*
 501          * bdev can be NULL, as with scsi-generic, this is a helpful as
 502          * we can be.
 503          */
 504         if (rcu_dereference_protected(q->blk_trace,
 505                                       lockdep_is_held(&q->debugfs_mutex))) {
 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         /*
 525          * When tracing the whole disk reuse the existing debugfs directory
 526          * created by the block layer on init. For partitions block devices,
 527          * and scsi-generic block devices we create a temporary new debugfs
 528          * directory that will be removed once the trace ends.
 529          */
 530         if (bdev && !bdev_is_partition(bdev))
 531                 dir = q->debugfs_dir;
 532         else
 533                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 534
 535         /*
 536          * As blktrace relies on debugfs for its interface the debugfs directory
 537          * is required, contrary to the usual mantra of not checking for debugfs
 538          * files or directories.
 539          */
 540         if (IS_ERR_OR_NULL(dir)) {
 541                 pr_warn("debugfs_dir not present for %s so skipping\n",
 542                         buts->name);
 543                 ret = -ENOENT;
 544                 goto err;
 545         }
 546
 547         bt->dev = dev;
 548         atomic_set(&bt->dropped, 0);
 549         INIT_LIST_HEAD(&bt->running_list);
 550
 551         ret = -EIO;
 552         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 553                                                &blk_dropped_fops);
 554
 555         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 556
 557         bt->rchan = relay_open("trace", dir, buts->buf_size,
 558                                 buts->buf_nr, &blk_relay_callbacks, bt);
 559         if (!bt->rchan)
 560                 goto err;
 561
 562         bt->act_mask = buts->act_mask;
 563         if (!bt->act_mask)
 564                 bt->act_mask = (u16) -1;
 565
 566         blk_trace_setup_lba(bt, bdev);
 567
 568         /* overwrite with user settings */
 569         if (buts->start_lba)
 570                 bt->start_lba = buts->start_lba;
 571         if (buts->end_lba)
 572                 bt->end_lba = buts->end_lba;
 573
 574         bt->pid = buts->pid;
 575         bt->trace_state = Blktrace_setup;
 576
 577         rcu_assign_pointer(q->blk_trace, bt);
 578         get_probe_ref();
 579
 580         ret = 0;
 581 err:
 582         if (ret)
 583                 blk_trace_free(bt);
 584         return ret;
 585 }
 586
 587 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 588                              struct block_device *bdev, char __user *arg)
 589 {
 590         struct blk_user_trace_setup buts;
 591         int ret;
 592
 593         ret = copy_from_user(&buts, arg, sizeof(buts));
 594         if (ret)
 595                 return -EFAULT;
 596
 597         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 598         if (ret)
 599                 return ret;
 600
 601         if (copy_to_user(arg, &buts, sizeof(buts))) {
 602                 __blk_trace_remove(q);
 603                 return -EFAULT;
 604         }
 605         return 0;
 606 }
 607
 608 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 609                     struct block_device *bdev,
 610                     char __user *arg)
 611 {
 612         int ret;
 613
 614         mutex_lock(&q->debugfs_mutex);
 615         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 616         mutex_unlock(&q->debugfs_mutex);
 617
 618         return ret;
 619 }
 620 EXPORT_SYMBOL_GPL(blk_trace_setup);
 621
 622 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 623 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 624                                   dev_t dev, struct block_device *bdev,
 625                                   char __user *arg)
 626 {
 627         struct blk_user_trace_setup buts;
 628         struct compat_blk_user_trace_setup cbuts;
 629         int ret;
 630
 631         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 632                 return -EFAULT;
 633
 634         buts = (struct blk_user_trace_setup) {
 635                 .act_mask = cbuts.act_mask,
 636                 .buf_size = cbuts.buf_size,
 637                 .buf_nr = cbuts.buf_nr,
 638                 .start_lba = cbuts.start_lba,
 639                 .end_lba = cbuts.end_lba,
 640                 .pid = cbuts.pid,
 641         };
 642
 643         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 644         if (ret)
 645                 return ret;
 646
 647         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 648                 __blk_trace_remove(q);
 649                 return -EFAULT;
 650         }
 651
 652         return 0;
 653 }
 654 #endif
 655
 656 static int __blk_trace_startstop(struct request_queue *q, int start)
 657 {
 658         int ret;
 659         struct blk_trace *bt;
 660
 661         bt = rcu_dereference_protected(q->blk_trace,
 662                                        lockdep_is_held(&q->debugfs_mutex));
 663         if (bt == NULL)
 664                 return -EINVAL;
 665
 666         /*
 667          * For starting a trace, we can transition from a setup or stopped
 668          * trace. For stopping a trace, the state must be running
 669          */
 670         ret = -EINVAL;
 671         if (start) {
 672                 if (bt->trace_state == Blktrace_setup ||
 673                     bt->trace_state == Blktrace_stopped) {
 674                         blktrace_seq++;
 675                         smp_mb();
 676                         bt->trace_state = Blktrace_running;
 677                         spin_lock_irq(&running_trace_lock);
 678                         list_add(&bt->running_list, &running_trace_list);
 679                         spin_unlock_irq(&running_trace_lock);
 680
 681                         trace_note_time(bt);
 682                         ret = 0;
 683                 }
 684         } else {
 685                 if (bt->trace_state == Blktrace_running) {
 686                         bt->trace_state = Blktrace_stopped;
 687                         spin_lock_irq(&running_trace_lock);
 688                         list_del_init(&bt->running_list);
 689                         spin_unlock_irq(&running_trace_lock);
 690                         relay_flush(bt->rchan);
 691                         ret = 0;
 692                 }
 693         }
 694
 695         return ret;
 696 }
 697
 698 int blk_trace_startstop(struct request_queue *q, int start)
 699 {
 700         int ret;
 701
 702         mutex_lock(&q->debugfs_mutex);
 703         ret = __blk_trace_startstop(q, start);
 704         mutex_unlock(&q->debugfs_mutex);
 705
 706         return ret;
 707 }
 708 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 709
 710 /*
 711  * When reading or writing the blktrace sysfs files, the references to the
 712  * opened sysfs or device files should prevent the underlying block device
 713  * from being removed. So no further delete protection is really needed.
 714  */
 715
 716 /**
 717  * blk_trace_ioctl: - handle the ioctls associated with tracing
 718  * @bdev:       the block device
 719  * @cmd:        the ioctl cmd
 720  * @arg:        the argument data, if any
 721  *
 722  **/
 723 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 724 {
 725         struct request_queue *q;
 726         int ret, start = 0;
 727         char b[BDEVNAME_SIZE];
 728
 729         q = bdev_get_queue(bdev);
 730         if (!q)
 731                 return -ENXIO;
 732
 733         mutex_lock(&q->debugfs_mutex);
 734
 735         switch (cmd) {
 736         case BLKTRACESETUP:
 737                 bdevname(bdev, b);
 738                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 739                 break;
 740 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 741         case BLKTRACESETUP32:
 742                 bdevname(bdev, b);
 743                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 744                 break;
 745 #endif
 746         case BLKTRACESTART:
 747                 start = 1;
 748                 fallthrough;
 749         case BLKTRACESTOP:
 750                 ret = __blk_trace_startstop(q, start);
 751                 break;
 752         case BLKTRACETEARDOWN:
 753                 ret = __blk_trace_remove(q);
 754                 break;
 755         default:
 756                 ret = -ENOTTY;
 757                 break;
 758         }
 759
 760         mutex_unlock(&q->debugfs_mutex);
 761         return ret;
 762 }
 763
 764 /**
 765  * blk_trace_shutdown: - stop and cleanup trace structures
 766  * @q:    the request queue associated with the device
 767  *
 768  **/
 769 void blk_trace_shutdown(struct request_queue *q)
 770 {
 771         mutex_lock(&q->debugfs_mutex);
 772         if (rcu_dereference_protected(q->blk_trace,
 773                                       lockdep_is_held(&q->debugfs_mutex))) {
 774                 __blk_trace_startstop(q, 0);
 775                 __blk_trace_remove(q);
 776         }
 777
 778         mutex_unlock(&q->debugfs_mutex);
 779 }
 780
 781 #ifdef CONFIG_BLK_CGROUP
 782 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 783 {
 784         struct blk_trace *bt;
 785
 786         /* We don't use the 'bt' value here except as an optimization... */
 787         bt = rcu_dereference_protected(q->blk_trace, 1);
 788         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 789                 return 0;
 790
 791         if (!bio->bi_blkg)
 792                 return 0;
 793         return cgroup_id(bio_blkcg(bio)->css.cgroup);
 794 }
 795 #else
 796 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 797 {
 798         return 0;
 799 }
 800 #endif
 801
 802 static u64
 803 blk_trace_request_get_cgid(struct request *rq)
 804 {
 805         if (!rq->bio)
 806                 return 0;
 807         /* Use the first bio */
 808         return blk_trace_bio_get_cgid(rq->q, rq->bio);
 809 }
 810
 811 /*
 812  * blktrace probes
 813  */
 814
 815 /**
 816  * blk_add_trace_rq - Add a trace for a request oriented action
 817  * @rq:         the source request
 818  * @error:      return status to log
 819  * @nr_bytes:   number of completed bytes
 820  * @what:       the action
 821  * @cgid:       the cgroup info
 822  *
 823  * Description:
 824  *     Records an action against a request. Will log the bio offset + size.
 825  *
 826  **/
 827 static void blk_add_trace_rq(struct request *rq, int error,
 828                              unsigned int nr_bytes, u32 what, u64 cgid)
 829 {
 830         struct blk_trace *bt;
 831
 832         rcu_read_lock();
 833         bt = rcu_dereference(rq->q->blk_trace);
 834         if (likely(!bt)) {
 835                 rcu_read_unlock();
 836                 return;
 837         }
 838
 839         if (blk_rq_is_passthrough(rq))
 840                 what |= BLK_TC_ACT(BLK_TC_PC);
 841         else
 842                 what |= BLK_TC_ACT(BLK_TC_FS);
 843
 844         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 845                         rq->cmd_flags, what, error, 0, NULL, cgid);
 846         rcu_read_unlock();
 847 }
 848
 849 static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
 850 {
 851         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 852                          blk_trace_request_get_cgid(rq));
 853 }
 854
 855 static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
 856 {
 857         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 858                          blk_trace_request_get_cgid(rq));
 859 }
 860
 861 static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
 862 {
 863         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
 864                          blk_trace_request_get_cgid(rq));
 865 }
 866
 867 static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
 868 {
 869         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 870                          blk_trace_request_get_cgid(rq));
 871 }
 872
 873 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 874                         int error, unsigned int nr_bytes)
 875 {
 876         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 877                          blk_trace_request_get_cgid(rq));
 878 }
 879
 880 /**
 881  * blk_add_trace_bio - Add a trace for a bio oriented action
 882  * @q:          queue the io is for
 883  * @bio:        the source bio
 884  * @what:       the action
 885  * @error:      error, if any
 886  *
 887  * Description:
 888  *     Records an action against a bio. Will log the bio offset + size.
 889  *
 890  **/
 891 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 892                               u32 what, int error)
 893 {
 894         struct blk_trace *bt;
 895
 896         rcu_read_lock();
 897         bt = rcu_dereference(q->blk_trace);
 898         if (likely(!bt)) {
 899                 rcu_read_unlock();
 900                 return;
 901         }
 902
 903         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 904                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 905                         blk_trace_bio_get_cgid(q, bio));
 906         rcu_read_unlock();
 907 }
 908
 909 static void blk_add_trace_bio_bounce(void *ignore,
 910                                      struct request_queue *q, struct bio *bio)
 911 {
 912         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 913 }
 914
 915 static void blk_add_trace_bio_complete(void *ignore,
 916                                        struct request_queue *q, struct bio *bio)
 917 {
 918         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
 919                           blk_status_to_errno(bio->bi_status));
 920 }
 921
 922 static void blk_add_trace_bio_backmerge(void *ignore,
 923                                         struct request_queue *q,
 924                                         struct request *rq,
 925                                         struct bio *bio)
 926 {
 927         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 928 }
 929
 930 static void blk_add_trace_bio_frontmerge(void *ignore,
 931                                          struct request_queue *q,
 932                                          struct request *rq,
 933                                          struct bio *bio)
 934 {
 935         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 936 }
 937
 938 static void blk_add_trace_bio_queue(void *ignore,
 939                                     struct request_queue *q, struct bio *bio)
 940 {
 941         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 942 }
 943
 944 static void blk_add_trace_getrq(void *ignore,
 945                                 struct request_queue *q,
 946                                 struct bio *bio, int rw)
 947 {
 948         if (bio)
 949                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 950         else {
 951                 struct blk_trace *bt;
 952
 953                 rcu_read_lock();
 954                 bt = rcu_dereference(q->blk_trace);
 955                 if (bt)
 956                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 957                                         NULL, 0);
 958                 rcu_read_unlock();
 959         }
 960 }
 961
 962
 963 static void blk_add_trace_sleeprq(void *ignore,
 964                                   struct request_queue *q,
 965                                   struct bio *bio, int rw)
 966 {
 967         if (bio)
 968                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 969         else {
 970                 struct blk_trace *bt;
 971
 972                 rcu_read_lock();
 973                 bt = rcu_dereference(q->blk_trace);
 974                 if (bt)
 975                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 976                                         0, 0, NULL, 0);
 977                 rcu_read_unlock();
 978         }
 979 }
 980
 981 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 982 {
 983         struct blk_trace *bt;
 984
 985         rcu_read_lock();
 986         bt = rcu_dereference(q->blk_trace);
 987         if (bt)
 988                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 989         rcu_read_unlock();
 990 }
 991
 992 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 993                                     unsigned int depth, bool explicit)
 994 {
 995         struct blk_trace *bt;
 996
 997         rcu_read_lock();
 998         bt = rcu_dereference(q->blk_trace);
 999         if (bt) {
1000                 __be64 rpdu = cpu_to_be64(depth);
1001                 u32 what;
1002
1003                 if (explicit)
1004                         what = BLK_TA_UNPLUG_IO;
1005                 else
1006                         what = BLK_TA_UNPLUG_TIMER;
1007
1008                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
1009         }
1010         rcu_read_unlock();
1011 }
1012
1013 static void blk_add_trace_split(void *ignore,
1014                                 struct request_queue *q, struct bio *bio,
1015                                 unsigned int pdu)
1016 {
1017         struct blk_trace *bt;
1018
1019         rcu_read_lock();
1020         bt = rcu_dereference(q->blk_trace);
1021         if (bt) {
1022                 __be64 rpdu = cpu_to_be64(pdu);
1023
1024                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
1025                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
1026                                 BLK_TA_SPLIT,
1027                                 blk_status_to_errno(bio->bi_status),
1028                                 sizeof(rpdu), &rpdu,
1029                                 blk_trace_bio_get_cgid(q, bio));
1030         }
1031         rcu_read_unlock();
1032 }
1033
1034 /**
1035  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1036  * @ignore:     trace callback data parameter (not used)
1037  * @q:          queue the io is for
1038  * @bio:        the source bio
1039  * @dev:        target device
1040  * @from:       source sector
1041  *
1042  * Description:
1043  *     Device mapper or raid target sometimes need to split a bio because
1044  *     it spans a stripe (or similar). Add a trace for that action.
1045  *
1046  **/
1047 static void blk_add_trace_bio_remap(void *ignore,
1048                                     struct request_queue *q, struct bio *bio,
1049                                     dev_t dev, sector_t from)
1050 {
1051         struct blk_trace *bt;
1052         struct blk_io_trace_remap r;
1053
1054         rcu_read_lock();
1055         bt = rcu_dereference(q->blk_trace);
1056         if (likely(!bt)) {
1057                 rcu_read_unlock();
1058                 return;
1059         }
1060
1061         r.device_from = cpu_to_be32(dev);
1062         r.device_to   = cpu_to_be32(bio_dev(bio));
1063         r.sector_from = cpu_to_be64(from);
1064
1065         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1066                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1067                         blk_status_to_errno(bio->bi_status),
1068                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1069         rcu_read_unlock();
1070 }
1071
1072 /**
1073  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1074  * @ignore:     trace callback data parameter (not used)
1075  * @q:          queue the io is for
1076  * @rq:         the source request
1077  * @dev:        target device
1078  * @from:       source sector
1079  *
1080  * Description:
1081  *     Device mapper remaps request to other devices.
1082  *     Add a trace for that action.
1083  *
1084  **/
1085 static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
1086                                    sector_t from)
1087 {
1088         struct blk_trace *bt;
1089         struct blk_io_trace_remap r;
1090
1091         rcu_read_lock();
1092         bt = rcu_dereference(rq->q->blk_trace);
1093         if (likely(!bt)) {
1094                 rcu_read_unlock();
1095                 return;
1096         }
1097
1098         r.device_from = cpu_to_be32(dev);
1099         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1100         r.sector_from = cpu_to_be64(from);
1101
1102         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1103                         req_op(rq), rq->cmd_flags, BLK_TA_REMAP, 0,
1104                         sizeof(r), &r, blk_trace_request_get_cgid(rq));
1105         rcu_read_unlock();
1106 }
1107
1108 /**
1109  * blk_add_driver_data - Add binary message with driver-specific data
1110  * @rq:         io request
1111  * @data:       driver-specific data
1112  * @len:        length of driver-specific data
1113  *
1114  * Description:
1115  *     Some drivers might want to write driver-specific data per request.
1116  *
1117  **/
1118 void blk_add_driver_data(struct request *rq, void *data, size_t len)
1119 {
1120         struct blk_trace *bt;
1121
1122         rcu_read_lock();
1123         bt = rcu_dereference(rq->q->blk_trace);
1124         if (likely(!bt)) {
1125                 rcu_read_unlock();
1126                 return;
1127         }
1128
1129         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1130                                 BLK_TA_DRV_DATA, 0, len, data,
1131                                 blk_trace_request_get_cgid(rq));
1132         rcu_read_unlock();
1133 }
1134 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1135
1136 static void blk_register_tracepoints(void)
1137 {
1138         int ret;
1139
1140         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1141         WARN_ON(ret);
1142         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1143         WARN_ON(ret);
1144         ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1145         WARN_ON(ret);
1146         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1147         WARN_ON(ret);
1148         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1149         WARN_ON(ret);
1150         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1151         WARN_ON(ret);
1152         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1153         WARN_ON(ret);
1154         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1155         WARN_ON(ret);
1156         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1157         WARN_ON(ret);
1158         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1159         WARN_ON(ret);
1160         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1161         WARN_ON(ret);
1162         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1163         WARN_ON(ret);
1164         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1165         WARN_ON(ret);
1166         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1167         WARN_ON(ret);
1168         ret = register_trace_block_split(blk_add_trace_split, NULL);
1169         WARN_ON(ret);
1170         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1171         WARN_ON(ret);
1172         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1173         WARN_ON(ret);
1174 }
1175
1176 static void blk_unregister_tracepoints(void)
1177 {
1178         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1179         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1180         unregister_trace_block_split(blk_add_trace_split, NULL);
1181         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1182         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1183         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1184         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1185         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1186         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1187         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1188         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1189         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1190         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1191         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1192         unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1193         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1194         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1195
1196         tracepoint_synchronize_unregister();
1197 }
1198
1199 /*
1200  * struct blk_io_tracer formatting routines
1201  */
1202
1203 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1204 {
1205         int i = 0;
1206         int tc = t->action >> BLK_TC_SHIFT;
1207
1208         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1209                 rwbs[i++] = 'N';
1210                 goto out;
1211         }
1212
1213         if (tc & BLK_TC_FLUSH)
1214                 rwbs[i++] = 'F';
1215
1216         if (tc & BLK_TC_DISCARD)
1217                 rwbs[i++] = 'D';
1218         else if (tc & BLK_TC_WRITE)
1219                 rwbs[i++] = 'W';
1220         else if (t->bytes)
1221                 rwbs[i++] = 'R';
1222         else
1223                 rwbs[i++] = 'N';
1224
1225         if (tc & BLK_TC_FUA)
1226                 rwbs[i++] = 'F';
1227         if (tc & BLK_TC_AHEAD)
1228                 rwbs[i++] = 'A';
1229         if (tc & BLK_TC_SYNC)
1230                 rwbs[i++] = 'S';
1231         if (tc & BLK_TC_META)
1232                 rwbs[i++] = 'M';
1233 out:
1234         rwbs[i] = '\0';
1235 }
1236
1237 static inline
1238 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1239 {
1240         return (const struct blk_io_trace *)ent;
1241 }
1242
1243 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1244 {
1245         return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1246 }
1247
1248 static inline u64 t_cgid(const struct trace_entry *ent)
1249 {
1250         return *(u64 *)(te_blk_io_trace(ent) + 1);
1251 }
1252
1253 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1254 {
1255         return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1256 }
1257
1258 static inline u32 t_action(const struct trace_entry *ent)
1259 {
1260         return te_blk_io_trace(ent)->action;
1261 }
1262
1263 static inline u32 t_bytes(const struct trace_entry *ent)
1264 {
1265         return te_blk_io_trace(ent)->bytes;
1266 }
1267
1268 static inline u32 t_sec(const struct trace_entry *ent)
1269 {
1270         return te_blk_io_trace(ent)->bytes >> 9;
1271 }
1272
1273 static inline unsigned long long t_sector(const struct trace_entry *ent)
1274 {
1275         return te_blk_io_trace(ent)->sector;
1276 }
1277
1278 static inline __u16 t_error(const struct trace_entry *ent)
1279 {
1280         return te_blk_io_trace(ent)->error;
1281 }
1282
1283 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1284 {
1285         const __be64 *val = pdu_start(ent, has_cg);
1286         return be64_to_cpu(*val);
1287 }
1288
1289 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1290         bool has_cg);
1291
1292 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1293         bool has_cg)
1294 {
1295         char rwbs[RWBS_LEN];
1296         unsigned long long ts  = iter->ts;
1297         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1298         unsigned secs          = (unsigned long)ts;
1299         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1300
1301         fill_rwbs(rwbs, t);
1302
1303         trace_seq_printf(&iter->seq,
1304                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1305                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1306                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1307 }
1308
1309 static void blk_log_action(struct trace_iterator *iter, const char *act,
1310         bool has_cg)
1311 {
1312         char rwbs[RWBS_LEN];
1313         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1314
1315         fill_rwbs(rwbs, t);
1316         if (has_cg) {
1317                 u64 id = t_cgid(iter->ent);
1318
1319                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1320                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1321
1322                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1323                                 sizeof(blkcg_name_buf));
1324                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1325                                  MAJOR(t->device), MINOR(t->device),
1326                                  blkcg_name_buf, act, rwbs);
1327                 } else {
1328                         /*
1329                          * The cgid portion used to be "INO,GEN".  Userland
1330                          * builds a FILEID_INO32_GEN fid out of them and
1331                          * opens the cgroup using open_by_handle_at(2).
1332                          * While 32bit ino setups are still the same, 64bit
1333                          * ones now use the 64bit ino as the whole ID and
1334                          * no longer use generation.
1335                          *
1336                          * Regarldess of the content, always output
1337                          * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1338                          * be mapped back to @id on both 64 and 32bit ino
1339                          * setups.  See __kernfs_fh_to_dentry().
1340                          */
1341                         trace_seq_printf(&iter->seq,
1342                                  "%3d,%-3d %llx,%-llx %2s %3s ",
1343                                  MAJOR(t->device), MINOR(t->device),
1344                                  id & U32_MAX, id >> 32, act, rwbs);
1345                 }
1346         } else
1347                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1348                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1349 }
1350
1351 static void blk_log_dump_pdu(struct trace_seq *s,
1352         const struct trace_entry *ent, bool has_cg)
1353 {
1354         const unsigned char *pdu_buf;
1355         int pdu_len;
1356         int i, end;
1357
1358         pdu_buf = pdu_start(ent, has_cg);
1359         pdu_len = pdu_real_len(ent, has_cg);
1360
1361         if (!pdu_len)
1362                 return;
1363
1364         /* find the last zero that needs to be printed */
1365         for (end = pdu_len - 1; end >= 0; end--)
1366                 if (pdu_buf[end])
1367                         break;
1368         end++;
1369
1370         trace_seq_putc(s, '(');
1371
1372         for (i = 0; i < pdu_len; i++) {
1373
1374                 trace_seq_printf(s, "%s%02x",
1375                                  i == 0 ? "" : " ", pdu_buf[i]);
1376
1377                 /*
1378                  * stop when the rest is just zeroes and indicate so
1379                  * with a ".." appended
1380                  */
1381                 if (i == end && end != pdu_len - 1) {
1382                         trace_seq_puts(s, " ..) ");
1383                         return;
1384                 }
1385         }
1386
1387         trace_seq_puts(s, ") ");
1388 }
1389
1390 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1391 {
1392         char cmd[TASK_COMM_LEN];
1393
1394         trace_find_cmdline(ent->pid, cmd);
1395
1396         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1397                 trace_seq_printf(s, "%u ", t_bytes(ent));
1398                 blk_log_dump_pdu(s, ent, has_cg);
1399                 trace_seq_printf(s, "[%s]\n", cmd);
1400         } else {
1401                 if (t_sec(ent))
1402                         trace_seq_printf(s, "%llu + %u [%s]\n",
1403                                                 t_sector(ent), t_sec(ent), cmd);
1404                 else
1405                         trace_seq_printf(s, "[%s]\n", cmd);
1406         }
1407 }
1408
1409 static void blk_log_with_error(struct trace_seq *s,
1410                               const struct trace_entry *ent, bool has_cg)
1411 {
1412         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1413                 blk_log_dump_pdu(s, ent, has_cg);
1414                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1415         } else {
1416                 if (t_sec(ent))
1417                         trace_seq_printf(s, "%llu + %u [%d]\n",
1418                                          t_sector(ent),
1419                                          t_sec(ent), t_error(ent));
1420                 else
1421                         trace_seq_printf(s, "%llu [%d]\n",
1422                                          t_sector(ent), t_error(ent));
1423         }
1424 }
1425
1426 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1427 {
1428         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1429
1430         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1431                          t_sector(ent), t_sec(ent),
1432                          MAJOR(be32_to_cpu(__r->device_from)),
1433                          MINOR(be32_to_cpu(__r->device_from)),
1434                          be64_to_cpu(__r->sector_from));
1435 }
1436
1437 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1438 {
1439         char cmd[TASK_COMM_LEN];
1440
1441         trace_find_cmdline(ent->pid, cmd);
1442
1443         trace_seq_printf(s, "[%s]\n", cmd);
1444 }
1445
1446 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1447 {
1448         char cmd[TASK_COMM_LEN];
1449
1450         trace_find_cmdline(ent->pid, cmd);
1451
1452         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1453 }
1454
1455 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1456 {
1457         char cmd[TASK_COMM_LEN];
1458
1459         trace_find_cmdline(ent->pid, cmd);
1460
1461         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1462                          get_pdu_int(ent, has_cg), cmd);
1463 }
1464
1465 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1466                         bool has_cg)
1467 {
1468
1469         trace_seq_putmem(s, pdu_start(ent, has_cg),
1470                 pdu_real_len(ent, has_cg));
1471         trace_seq_putc(s, '\n');
1472 }
1473
1474 /*
1475  * struct tracer operations
1476  */
1477
1478 static void blk_tracer_print_header(struct seq_file *m)
1479 {
1480         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1481                 return;
1482         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1483                     "#  |     |     |           |   |   |\n");
1484 }
1485
1486 static void blk_tracer_start(struct trace_array *tr)
1487 {
1488         blk_tracer_enabled = true;
1489 }
1490
1491 static int blk_tracer_init(struct trace_array *tr)
1492 {
1493         blk_tr = tr;
1494         blk_tracer_start(tr);
1495         return 0;
1496 }
1497
1498 static void blk_tracer_stop(struct trace_array *tr)
1499 {
1500         blk_tracer_enabled = false;
1501 }
1502
1503 static void blk_tracer_reset(struct trace_array *tr)
1504 {
1505         blk_tracer_stop(tr);
1506 }
1507
1508 static const struct {
1509         const char *act[2];
1510         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1511                             bool has_cg);
1512 } what2act[] = {
1513         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1514         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1515         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1516         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1517         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1518         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1519         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1520         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1521         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1522         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1523         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1524         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1525         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1526         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1527         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1528 };
1529
1530 static enum print_line_t print_one_line(struct trace_iterator *iter,
1531                                         bool classic)
1532 {
1533         struct trace_array *tr = iter->tr;
1534         struct trace_seq *s = &iter->seq;
1535         const struct blk_io_trace *t;
1536         u16 what;
1537         bool long_act;
1538         blk_log_action_t *log_action;
1539         bool has_cg;
1540
1541         t          = te_blk_io_trace(iter->ent);
1542         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1543         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1544         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1545         has_cg     = t->action & __BLK_TA_CGROUP;
1546
1547         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1548                 log_action(iter, long_act ? "message" : "m", has_cg);
1549                 blk_log_msg(s, iter->ent, has_cg);
1550                 return trace_handle_return(s);
1551         }
1552
1553         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1554                 trace_seq_printf(s, "Unknown action %x\n", what);
1555         else {
1556                 log_action(iter, what2act[what].act[long_act], has_cg);
1557                 what2act[what].print(s, iter->ent, has_cg);
1558         }
1559
1560         return trace_handle_return(s);
1561 }
1562
1563 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1564                                                int flags, struct trace_event *event)
1565 {
1566         return print_one_line(iter, false);
1567 }
1568
1569 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1570 {
1571         struct trace_seq *s = &iter->seq;
1572         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1573         const int offset = offsetof(struct blk_io_trace, sector);
1574         struct blk_io_trace old = {
1575                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1576                 .time     = iter->ts,
1577         };
1578
1579         trace_seq_putmem(s, &old, offset);
1580         trace_seq_putmem(s, &t->sector,
1581                          sizeof(old) - offset + t->pdu_len);
1582 }
1583
1584 static enum print_line_t
1585 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1586                              struct trace_event *event)
1587 {
1588         blk_trace_synthesize_old_trace(iter);
1589
1590         return trace_handle_return(&iter->seq);
1591 }
1592
1593 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1594 {
1595         if ((iter->ent->type != TRACE_BLK) ||
1596             !(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1597                 return TRACE_TYPE_UNHANDLED;
1598
1599         return print_one_line(iter, true);
1600 }
1601
1602 static int
1603 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1604 {
1605         /* don't output context-info for blk_classic output */
1606         if (bit == TRACE_BLK_OPT_CLASSIC) {
1607                 if (set)
1608                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1609                 else
1610                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1611         }
1612         return 0;
1613 }
1614
1615 static struct tracer blk_tracer __read_mostly = {
1616         .name           = "blk",
1617         .init           = blk_tracer_init,
1618         .reset          = blk_tracer_reset,
1619         .start          = blk_tracer_start,
1620         .stop           = blk_tracer_stop,
1621         .print_header   = blk_tracer_print_header,
1622         .print_line     = blk_tracer_print_line,
1623         .flags          = &blk_tracer_flags,
1624         .set_flag       = blk_tracer_set_flag,
1625 };
1626
1627 static struct trace_event_functions trace_blk_event_funcs = {
1628         .trace          = blk_trace_event_print,
1629         .binary         = blk_trace_event_print_binary,
1630 };
1631
1632 static struct trace_event trace_blk_event = {
1633         .type           = TRACE_BLK,
1634         .funcs          = &trace_blk_event_funcs,
1635 };
1636
1637 static int __init init_blk_tracer(void)
1638 {
1639         if (!register_trace_event(&trace_blk_event)) {
1640                 pr_warn("Warning: could not register block events\n");
1641                 return 1;
1642         }
1643
1644         if (register_tracer(&blk_tracer) != 0) {
1645                 pr_warn("Warning: could not register the block tracer\n");
1646                 unregister_trace_event(&trace_blk_event);
1647                 return 1;
1648         }
1649
1650         return 0;
1651 }
1652
1653 device_initcall(init_blk_tracer);
1654
1655 static int blk_trace_remove_queue(struct request_queue *q)
1656 {
1657         struct blk_trace *bt;
1658
1659         bt = rcu_replace_pointer(q->blk_trace, NULL,
1660                                  lockdep_is_held(&q->debugfs_mutex));
1661         if (bt == NULL)
1662                 return -EINVAL;
1663
1664         if (bt->trace_state == Blktrace_running) {
1665                 bt->trace_state = Blktrace_stopped;
1666                 spin_lock_irq(&running_trace_lock);
1667                 list_del_init(&bt->running_list);
1668                 spin_unlock_irq(&running_trace_lock);
1669                 relay_flush(bt->rchan);
1670         }
1671
1672         put_probe_ref();
1673         synchronize_rcu();
1674         blk_trace_free(bt);
1675         return 0;
1676 }
1677
1678 /*
1679  * Setup everything required to start tracing
1680  */
1681 static int blk_trace_setup_queue(struct request_queue *q,
1682                                  struct block_device *bdev)
1683 {
1684         struct blk_trace *bt = NULL;
1685         int ret = -ENOMEM;
1686
1687         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1688         if (!bt)
1689                 return -ENOMEM;
1690
1691         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1692         if (!bt->msg_data)
1693                 goto free_bt;
1694
1695         bt->dev = bdev->bd_dev;
1696         bt->act_mask = (u16)-1;
1697
1698         blk_trace_setup_lba(bt, bdev);
1699
1700         rcu_assign_pointer(q->blk_trace, bt);
1701         get_probe_ref();
1702         return 0;
1703
1704 free_bt:
1705         blk_trace_free(bt);
1706         return ret;
1707 }
1708
1709 /*
1710  * sysfs interface to enable and configure tracing
1711  */
1712
1713 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1714                                          struct device_attribute *attr,
1715                                          char *buf);
1716 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1717                                           struct device_attribute *attr,
1718                                           const char *buf, size_t count);
1719 #define BLK_TRACE_DEVICE_ATTR(_name) \
1720         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1721                     sysfs_blk_trace_attr_show, \
1722                     sysfs_blk_trace_attr_store)
1723
1724 static BLK_TRACE_DEVICE_ATTR(enable);
1725 static BLK_TRACE_DEVICE_ATTR(act_mask);
1726 static BLK_TRACE_DEVICE_ATTR(pid);
1727 static BLK_TRACE_DEVICE_ATTR(start_lba);
1728 static BLK_TRACE_DEVICE_ATTR(end_lba);
1729
1730 static struct attribute *blk_trace_attrs[] = {
1731         &dev_attr_enable.attr,
1732         &dev_attr_act_mask.attr,
1733         &dev_attr_pid.attr,
1734         &dev_attr_start_lba.attr,
1735         &dev_attr_end_lba.attr,
1736         NULL
1737 };
1738
1739 struct attribute_group blk_trace_attr_group = {
1740         .name  = "trace",
1741         .attrs = blk_trace_attrs,
1742 };
1743
1744 static const struct {
1745         int mask;
1746         const char *str;
1747 } mask_maps[] = {
1748         { BLK_TC_READ,          "read"          },
1749         { BLK_TC_WRITE,         "write"         },
1750         { BLK_TC_FLUSH,         "flush"         },
1751         { BLK_TC_SYNC,          "sync"          },
1752         { BLK_TC_QUEUE,         "queue"         },
1753         { BLK_TC_REQUEUE,       "requeue"       },
1754         { BLK_TC_ISSUE,         "issue"         },
1755         { BLK_TC_COMPLETE,      "complete"      },
1756         { BLK_TC_FS,            "fs"            },
1757         { BLK_TC_PC,            "pc"            },
1758         { BLK_TC_NOTIFY,        "notify"        },
1759         { BLK_TC_AHEAD,         "ahead"         },
1760         { BLK_TC_META,          "meta"          },
1761         { BLK_TC_DISCARD,       "discard"       },
1762         { BLK_TC_DRV_DATA,      "drv_data"      },
1763         { BLK_TC_FUA,           "fua"           },
1764 };
1765
1766 static int blk_trace_str2mask(const char *str)
1767 {
1768         int i;
1769         int mask = 0;
1770         char *buf, *s, *token;
1771
1772         buf = kstrdup(str, GFP_KERNEL);
1773         if (buf == NULL)
1774                 return -ENOMEM;
1775         s = strstrip(buf);
1776
1777         while (1) {
1778                 token = strsep(&s, ",");
1779                 if (token == NULL)
1780                         break;
1781
1782                 if (*token == '\0')
1783                         continue;
1784
1785                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1786                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1787                                 mask |= mask_maps[i].mask;
1788                                 break;
1789                         }
1790                 }
1791                 if (i == ARRAY_SIZE(mask_maps)) {
1792                         mask = -EINVAL;
1793                         break;
1794                 }
1795         }
1796         kfree(buf);
1797
1798         return mask;
1799 }
1800
1801 static ssize_t blk_trace_mask2str(char *buf, int mask)
1802 {
1803         int i;
1804         char *p = buf;
1805
1806         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1807                 if (mask & mask_maps[i].mask) {
1808                         p += sprintf(p, "%s%s",
1809                                     (p == buf) ? "" : ",", mask_maps[i].str);
1810                 }
1811         }
1812         *p++ = '\n';
1813
1814         return p - buf;
1815 }
1816
1817 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1818 {
1819         if (bdev->bd_disk == NULL)
1820                 return NULL;
1821
1822         return bdev_get_queue(bdev);
1823 }
1824
1825 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1826                                          struct device_attribute *attr,
1827                                          char *buf)
1828 {
1829         struct block_device *bdev = bdget_part(dev_to_part(dev));
1830         struct request_queue *q;
1831         struct blk_trace *bt;
1832         ssize_t ret = -ENXIO;
1833
1834         if (bdev == NULL)
1835                 goto out;
1836
1837         q = blk_trace_get_queue(bdev);
1838         if (q == NULL)
1839                 goto out_bdput;
1840
1841         mutex_lock(&q->debugfs_mutex);
1842
1843         bt = rcu_dereference_protected(q->blk_trace,
1844                                        lockdep_is_held(&q->debugfs_mutex));
1845         if (attr == &dev_attr_enable) {
1846                 ret = sprintf(buf, "%u\n", !!bt);
1847                 goto out_unlock_bdev;
1848         }
1849
1850         if (bt == NULL)
1851                 ret = sprintf(buf, "disabled\n");
1852         else if (attr == &dev_attr_act_mask)
1853                 ret = blk_trace_mask2str(buf, bt->act_mask);
1854         else if (attr == &dev_attr_pid)
1855                 ret = sprintf(buf, "%u\n", bt->pid);
1856         else if (attr == &dev_attr_start_lba)
1857                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1858         else if (attr == &dev_attr_end_lba)
1859                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1860
1861 out_unlock_bdev:
1862         mutex_unlock(&q->debugfs_mutex);
1863 out_bdput:
1864         bdput(bdev);
1865 out:
1866         return ret;
1867 }
1868
1869 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1870                                           struct device_attribute *attr,
1871                                           const char *buf, size_t count)
1872 {
1873         struct block_device *bdev;
1874         struct request_queue *q;
1875         struct blk_trace *bt;
1876         u64 value;
1877         ssize_t ret = -EINVAL;
1878
1879         if (count == 0)
1880                 goto out;
1881
1882         if (attr == &dev_attr_act_mask) {
1883                 if (kstrtoull(buf, 0, &value)) {
1884                         /* Assume it is a list of trace category names */
1885                         ret = blk_trace_str2mask(buf);
1886                         if (ret < 0)
1887                                 goto out;
1888                         value = ret;
1889                 }
1890         } else if (kstrtoull(buf, 0, &value))
1891                 goto out;
1892
1893         ret = -ENXIO;
1894         bdev = bdget_part(dev_to_part(dev));
1895         if (bdev == NULL)
1896                 goto out;
1897
1898         q = blk_trace_get_queue(bdev);
1899         if (q == NULL)
1900                 goto out_bdput;
1901
1902         mutex_lock(&q->debugfs_mutex);
1903
1904         bt = rcu_dereference_protected(q->blk_trace,
1905                                        lockdep_is_held(&q->debugfs_mutex));
1906         if (attr == &dev_attr_enable) {
1907                 if (!!value == !!bt) {
1908                         ret = 0;
1909                         goto out_unlock_bdev;
1910                 }
1911                 if (value)
1912                         ret = blk_trace_setup_queue(q, bdev);
1913                 else
1914                         ret = blk_trace_remove_queue(q);
1915                 goto out_unlock_bdev;
1916         }
1917
1918         ret = 0;
1919         if (bt == NULL) {
1920                 ret = blk_trace_setup_queue(q, bdev);
1921                 bt = rcu_dereference_protected(q->blk_trace,
1922                                 lockdep_is_held(&q->debugfs_mutex));
1923         }
1924
1925         if (ret == 0) {
1926                 if (attr == &dev_attr_act_mask)
1927                         bt->act_mask = value;
1928                 else if (attr == &dev_attr_pid)
1929                         bt->pid = value;
1930                 else if (attr == &dev_attr_start_lba)
1931                         bt->start_lba = value;
1932                 else if (attr == &dev_attr_end_lba)
1933                         bt->end_lba = value;
1934         }
1935
1936 out_unlock_bdev:
1937         mutex_unlock(&q->debugfs_mutex);
1938 out_bdput:
1939         bdput(bdev);
1940 out:
1941         return ret ? ret : count;
1942 }
1943
1944 int blk_trace_init_sysfs(struct device *dev)
1945 {
1946         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1947 }
1948
1949 void blk_trace_remove_sysfs(struct device *dev)
1950 {
1951         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1952 }
1953
1954 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1955
1956 #ifdef CONFIG_EVENT_TRACING
1957
1958 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1959 {
1960         int i = 0;
1961
1962         if (op & REQ_PREFLUSH)
1963                 rwbs[i++] = 'F';
1964
1965         switch (op & REQ_OP_MASK) {
1966         case REQ_OP_WRITE:
1967         case REQ_OP_WRITE_SAME:
1968                 rwbs[i++] = 'W';
1969                 break;
1970         case REQ_OP_DISCARD:
1971                 rwbs[i++] = 'D';
1972                 break;
1973         case REQ_OP_SECURE_ERASE:
1974                 rwbs[i++] = 'D';
1975                 rwbs[i++] = 'E';
1976                 break;
1977         case REQ_OP_FLUSH:
1978                 rwbs[i++] = 'F';
1979                 break;
1980         case REQ_OP_READ:
1981                 rwbs[i++] = 'R';
1982                 break;
1983         default:
1984                 rwbs[i++] = 'N';
1985         }
1986
1987         if (op & REQ_FUA)
1988                 rwbs[i++] = 'F';
1989         if (op & REQ_RAHEAD)
1990                 rwbs[i++] = 'A';
1991         if (op & REQ_SYNC)
1992                 rwbs[i++] = 'S';
1993         if (op & REQ_META)
1994                 rwbs[i++] = 'M';
1995
1996         rwbs[i] = '\0';
1997 }
1998 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1999
2000 #endif /* CONFIG_EVENT_TRACING */
2001