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_RAW_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         unsigned int trace_ctx = 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                 trace_ctx = tracing_gen_ctx_flags(0);
  83                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84                                                   sizeof(*t) + len + cgid_len,
  85                                                   trace_ctx);
  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, trace_ctx);
 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         raw_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         raw_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 __blk_trace_note_message(struct blk_trace *bt,
 149                 struct cgroup_subsys_state *css, const char *fmt, ...)
 150 {
 151         int n;
 152         va_list args;
 153         unsigned long flags;
 154         char *buf;
 155         u64 cgid = 0;
 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 #ifdef CONFIG_BLK_CGROUP
 175         if (css && (blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 176                 cgid = cgroup_id(css->cgroup);
 177         else
 178                 cgid = 1;
 179 #endif
 180         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, cgid);
 181         local_irq_restore(flags);
 182 }
 183 EXPORT_SYMBOL_GPL(__blk_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         unsigned int trace_ctx = 0;
 226         pid_t pid;
 227         int cpu;
 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->array_buffer.buffer;
 256                 trace_ctx = tracing_gen_ctx_flags(0);
 257                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 258                                                   sizeof(*t) + pdu_len + cgid_len,
 259                                                   trace_ctx);
 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, trace_ctx);
 306                         return;
 307                 }
 308         }
 309
 310         local_irq_restore(flags);
 311 }
 312
 313 static void blk_trace_free(struct request_queue *q, struct blk_trace *bt)
 314 {
 315         relay_close(bt->rchan);
 316
 317         /*
 318          * If 'bt->dir' is not set, then both 'dropped' and 'msg' are created
 319          * under 'q->debugfs_dir', thus lookup and remove them.
 320          */
 321         if (!bt->dir) {
 322                 debugfs_remove(debugfs_lookup("dropped", q->debugfs_dir));
 323                 debugfs_remove(debugfs_lookup("msg", q->debugfs_dir));
 324         } else {
 325                 debugfs_remove(bt->dir);
 326         }
 327         free_percpu(bt->sequence);
 328         free_percpu(bt->msg_data);
 329         kfree(bt);
 330 }
 331
 332 static void get_probe_ref(void)
 333 {
 334         mutex_lock(&blk_probe_mutex);
 335         if (++blk_probes_ref == 1)
 336                 blk_register_tracepoints();
 337         mutex_unlock(&blk_probe_mutex);
 338 }
 339
 340 static void put_probe_ref(void)
 341 {
 342         mutex_lock(&blk_probe_mutex);
 343         if (!--blk_probes_ref)
 344                 blk_unregister_tracepoints();
 345         mutex_unlock(&blk_probe_mutex);
 346 }
 347
 348 static void blk_trace_cleanup(struct request_queue *q, struct blk_trace *bt)
 349 {
 350         synchronize_rcu();
 351         blk_trace_free(q, bt);
 352         put_probe_ref();
 353 }
 354
 355 static int __blk_trace_remove(struct request_queue *q)
 356 {
 357         struct blk_trace *bt;
 358
 359         bt = rcu_replace_pointer(q->blk_trace, NULL,
 360                                  lockdep_is_held(&q->debugfs_mutex));
 361         if (!bt)
 362                 return -EINVAL;
 363
 364         if (bt->trace_state != Blktrace_running)
 365                 blk_trace_cleanup(q, bt);
 366
 367         return 0;
 368 }
 369
 370 int blk_trace_remove(struct request_queue *q)
 371 {
 372         int ret;
 373
 374         mutex_lock(&q->debugfs_mutex);
 375         ret = __blk_trace_remove(q);
 376         mutex_unlock(&q->debugfs_mutex);
 377
 378         return ret;
 379 }
 380 EXPORT_SYMBOL_GPL(blk_trace_remove);
 381
 382 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 383                                 size_t count, loff_t *ppos)
 384 {
 385         struct blk_trace *bt = filp->private_data;
 386         char buf[16];
 387
 388         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 389
 390         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 391 }
 392
 393 static const struct file_operations blk_dropped_fops = {
 394         .owner =        THIS_MODULE,
 395         .open =         simple_open,
 396         .read =         blk_dropped_read,
 397         .llseek =       default_llseek,
 398 };
 399
 400 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 401                                 size_t count, loff_t *ppos)
 402 {
 403         char *msg;
 404         struct blk_trace *bt;
 405
 406         if (count >= BLK_TN_MAX_MSG)
 407                 return -EINVAL;
 408
 409         msg = memdup_user_nul(buffer, count);
 410         if (IS_ERR(msg))
 411                 return PTR_ERR(msg);
 412
 413         bt = filp->private_data;
 414         __blk_trace_note_message(bt, NULL, "%s", msg);
 415         kfree(msg);
 416
 417         return count;
 418 }
 419
 420 static const struct file_operations blk_msg_fops = {
 421         .owner =        THIS_MODULE,
 422         .open =         simple_open,
 423         .write =        blk_msg_write,
 424         .llseek =       noop_llseek,
 425 };
 426
 427 /*
 428  * Keep track of how many times we encountered a full subbuffer, to aid
 429  * the user space app in telling how many lost events there were.
 430  */
 431 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 432                                      void *prev_subbuf, size_t prev_padding)
 433 {
 434         struct blk_trace *bt;
 435
 436         if (!relay_buf_full(buf))
 437                 return 1;
 438
 439         bt = buf->chan->private_data;
 440         atomic_inc(&bt->dropped);
 441         return 0;
 442 }
 443
 444 static int blk_remove_buf_file_callback(struct dentry *dentry)
 445 {
 446         debugfs_remove(dentry);
 447
 448         return 0;
 449 }
 450
 451 static struct dentry *blk_create_buf_file_callback(const char *filename,
 452                                                    struct dentry *parent,
 453                                                    umode_t mode,
 454                                                    struct rchan_buf *buf,
 455                                                    int *is_global)
 456 {
 457         return debugfs_create_file(filename, mode, parent, buf,
 458                                         &relay_file_operations);
 459 }
 460
 461 static const struct rchan_callbacks blk_relay_callbacks = {
 462         .subbuf_start           = blk_subbuf_start_callback,
 463         .create_buf_file        = blk_create_buf_file_callback,
 464         .remove_buf_file        = blk_remove_buf_file_callback,
 465 };
 466
 467 static void blk_trace_setup_lba(struct blk_trace *bt,
 468                                 struct block_device *bdev)
 469 {
 470         if (bdev) {
 471                 bt->start_lba = bdev->bd_start_sect;
 472                 bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
 473         } else {
 474                 bt->start_lba = 0;
 475                 bt->end_lba = -1ULL;
 476         }
 477 }
 478
 479 /*
 480  * Setup everything required to start tracing
 481  */
 482 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 483                               struct block_device *bdev,
 484                               struct blk_user_trace_setup *buts)
 485 {
 486         struct blk_trace *bt = NULL;
 487         struct dentry *dir = NULL;
 488         int ret;
 489
 490         lockdep_assert_held(&q->debugfs_mutex);
 491
 492         if (!buts->buf_size || !buts->buf_nr)
 493                 return -EINVAL;
 494
 495         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 496         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 497
 498         /*
 499          * some device names have larger paths - convert the slashes
 500          * to underscores for this to work as expected
 501          */
 502         strreplace(buts->name, '/', '_');
 503
 504         /*
 505          * bdev can be NULL, as with scsi-generic, this is a helpful as
 506          * we can be.
 507          */
 508         if (rcu_dereference_protected(q->blk_trace,
 509                                       lockdep_is_held(&q->debugfs_mutex))) {
 510                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 511                         buts->name);
 512                 return -EBUSY;
 513         }
 514
 515         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 516         if (!bt)
 517                 return -ENOMEM;
 518
 519         ret = -ENOMEM;
 520         bt->sequence = alloc_percpu(unsigned long);
 521         if (!bt->sequence)
 522                 goto err;
 523
 524         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 525         if (!bt->msg_data)
 526                 goto err;
 527
 528         /*
 529          * When tracing the whole disk reuse the existing debugfs directory
 530          * created by the block layer on init. For partitions block devices,
 531          * and scsi-generic block devices we create a temporary new debugfs
 532          * directory that will be removed once the trace ends.
 533          */
 534         if (bdev && !bdev_is_partition(bdev))
 535                 dir = q->debugfs_dir;
 536         else
 537                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 538
 539         /*
 540          * As blktrace relies on debugfs for its interface the debugfs directory
 541          * is required, contrary to the usual mantra of not checking for debugfs
 542          * files or directories.
 543          */
 544         if (IS_ERR_OR_NULL(dir)) {
 545                 pr_warn("debugfs_dir not present for %s so skipping\n",
 546                         buts->name);
 547                 ret = -ENOENT;
 548                 goto err;
 549         }
 550
 551         bt->dev = dev;
 552         atomic_set(&bt->dropped, 0);
 553         INIT_LIST_HEAD(&bt->running_list);
 554
 555         ret = -EIO;
 556         debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
 557         debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 558
 559         bt->rchan = relay_open("trace", dir, buts->buf_size,
 560                                 buts->buf_nr, &blk_relay_callbacks, bt);
 561         if (!bt->rchan)
 562                 goto err;
 563
 564         bt->act_mask = buts->act_mask;
 565         if (!bt->act_mask)
 566                 bt->act_mask = (u16) -1;
 567
 568         blk_trace_setup_lba(bt, bdev);
 569
 570         /* overwrite with user settings */
 571         if (buts->start_lba)
 572                 bt->start_lba = buts->start_lba;
 573         if (buts->end_lba)
 574                 bt->end_lba = buts->end_lba;
 575
 576         bt->pid = buts->pid;
 577         bt->trace_state = Blktrace_setup;
 578
 579         rcu_assign_pointer(q->blk_trace, bt);
 580         get_probe_ref();
 581
 582         ret = 0;
 583 err:
 584         if (ret)
 585                 blk_trace_free(q, bt);
 586         return ret;
 587 }
 588
 589 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 590                              struct block_device *bdev, char __user *arg)
 591 {
 592         struct blk_user_trace_setup buts;
 593         int ret;
 594
 595         ret = copy_from_user(&buts, arg, sizeof(buts));
 596         if (ret)
 597                 return -EFAULT;
 598
 599         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 600         if (ret)
 601                 return ret;
 602
 603         if (copy_to_user(arg, &buts, sizeof(buts))) {
 604                 __blk_trace_remove(q);
 605                 return -EFAULT;
 606         }
 607         return 0;
 608 }
 609
 610 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 611                     struct block_device *bdev,
 612                     char __user *arg)
 613 {
 614         int ret;
 615
 616         mutex_lock(&q->debugfs_mutex);
 617         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 618         mutex_unlock(&q->debugfs_mutex);
 619
 620         return ret;
 621 }
 622 EXPORT_SYMBOL_GPL(blk_trace_setup);
 623
 624 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 625 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 626                                   dev_t dev, struct block_device *bdev,
 627                                   char __user *arg)
 628 {
 629         struct blk_user_trace_setup buts;
 630         struct compat_blk_user_trace_setup cbuts;
 631         int ret;
 632
 633         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 634                 return -EFAULT;
 635
 636         buts = (struct blk_user_trace_setup) {
 637                 .act_mask = cbuts.act_mask,
 638                 .buf_size = cbuts.buf_size,
 639                 .buf_nr = cbuts.buf_nr,
 640                 .start_lba = cbuts.start_lba,
 641                 .end_lba = cbuts.end_lba,
 642                 .pid = cbuts.pid,
 643         };
 644
 645         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 646         if (ret)
 647                 return ret;
 648
 649         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 650                 __blk_trace_remove(q);
 651                 return -EFAULT;
 652         }
 653
 654         return 0;
 655 }
 656 #endif
 657
 658 static int __blk_trace_startstop(struct request_queue *q, int start)
 659 {
 660         int ret;
 661         struct blk_trace *bt;
 662
 663         bt = rcu_dereference_protected(q->blk_trace,
 664                                        lockdep_is_held(&q->debugfs_mutex));
 665         if (bt == NULL)
 666                 return -EINVAL;
 667
 668         /*
 669          * For starting a trace, we can transition from a setup or stopped
 670          * trace. For stopping a trace, the state must be running
 671          */
 672         ret = -EINVAL;
 673         if (start) {
 674                 if (bt->trace_state == Blktrace_setup ||
 675                     bt->trace_state == Blktrace_stopped) {
 676                         blktrace_seq++;
 677                         smp_mb();
 678                         bt->trace_state = Blktrace_running;
 679                         raw_spin_lock_irq(&running_trace_lock);
 680                         list_add(&bt->running_list, &running_trace_list);
 681                         raw_spin_unlock_irq(&running_trace_lock);
 682
 683                         trace_note_time(bt);
 684                         ret = 0;
 685                 }
 686         } else {
 687                 if (bt->trace_state == Blktrace_running) {
 688                         bt->trace_state = Blktrace_stopped;
 689                         raw_spin_lock_irq(&running_trace_lock);
 690                         list_del_init(&bt->running_list);
 691                         raw_spin_unlock_irq(&running_trace_lock);
 692                         relay_flush(bt->rchan);
 693                         ret = 0;
 694                 }
 695         }
 696
 697         return ret;
 698 }
 699
 700 int blk_trace_startstop(struct request_queue *q, int start)
 701 {
 702         int ret;
 703
 704         mutex_lock(&q->debugfs_mutex);
 705         ret = __blk_trace_startstop(q, start);
 706         mutex_unlock(&q->debugfs_mutex);
 707
 708         return ret;
 709 }
 710 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 711
 712 /*
 713  * When reading or writing the blktrace sysfs files, the references to the
 714  * opened sysfs or device files should prevent the underlying block device
 715  * from being removed. So no further delete protection is really needed.
 716  */
 717
 718 /**
 719  * blk_trace_ioctl: - handle the ioctls associated with tracing
 720  * @bdev:       the block device
 721  * @cmd:        the ioctl cmd
 722  * @arg:        the argument data, if any
 723  *
 724  **/
 725 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 726 {
 727         struct request_queue *q;
 728         int ret, start = 0;
 729         char b[BDEVNAME_SIZE];
 730
 731         q = bdev_get_queue(bdev);
 732         if (!q)
 733                 return -ENXIO;
 734
 735         mutex_lock(&q->debugfs_mutex);
 736
 737         switch (cmd) {
 738         case BLKTRACESETUP:
 739                 bdevname(bdev, b);
 740                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 741                 break;
 742 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 743         case BLKTRACESETUP32:
 744                 bdevname(bdev, b);
 745                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 746                 break;
 747 #endif
 748         case BLKTRACESTART:
 749                 start = 1;
 750                 fallthrough;
 751         case BLKTRACESTOP:
 752                 ret = __blk_trace_startstop(q, start);
 753                 break;
 754         case BLKTRACETEARDOWN:
 755                 ret = __blk_trace_remove(q);
 756                 break;
 757         default:
 758                 ret = -ENOTTY;
 759                 break;
 760         }
 761
 762         mutex_unlock(&q->debugfs_mutex);
 763         return ret;
 764 }
 765
 766 /**
 767  * blk_trace_shutdown: - stop and cleanup trace structures
 768  * @q:    the request queue associated with the device
 769  *
 770  **/
 771 void blk_trace_shutdown(struct request_queue *q)
 772 {
 773         if (rcu_dereference_protected(q->blk_trace,
 774                                       lockdep_is_held(&q->debugfs_mutex))) {
 775                 __blk_trace_startstop(q, 0);
 776                 __blk_trace_remove(q);
 777         }
 778 }
 779
 780 #ifdef CONFIG_BLK_CGROUP
 781 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 782 {
 783         struct cgroup_subsys_state *blkcg_css;
 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         blkcg_css = bio_blkcg_css(bio);
 792         if (!blkcg_css)
 793                 return 0;
 794         return cgroup_id(blkcg_css->cgroup);
 795 }
 796 #else
 797 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 798 {
 799         return 0;
 800 }
 801 #endif
 802
 803 static u64
 804 blk_trace_request_get_cgid(struct request *rq)
 805 {
 806         if (!rq->bio)
 807                 return 0;
 808         /* Use the first bio */
 809         return blk_trace_bio_get_cgid(rq->q, rq->bio);
 810 }
 811
 812 /*
 813  * blktrace probes
 814  */
 815
 816 /**
 817  * blk_add_trace_rq - Add a trace for a request oriented action
 818  * @rq:         the source request
 819  * @error:      return status to log
 820  * @nr_bytes:   number of completed bytes
 821  * @what:       the action
 822  * @cgid:       the cgroup info
 823  *
 824  * Description:
 825  *     Records an action against a request. Will log the bio offset + size.
 826  *
 827  **/
 828 static void blk_add_trace_rq(struct request *rq, blk_status_t error,
 829                              unsigned int nr_bytes, u32 what, u64 cgid)
 830 {
 831         struct blk_trace *bt;
 832
 833         rcu_read_lock();
 834         bt = rcu_dereference(rq->q->blk_trace);
 835         if (likely(!bt)) {
 836                 rcu_read_unlock();
 837                 return;
 838         }
 839
 840         if (blk_rq_is_passthrough(rq))
 841                 what |= BLK_TC_ACT(BLK_TC_PC);
 842         else
 843                 what |= BLK_TC_ACT(BLK_TC_FS);
 844
 845         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 846                         rq->cmd_flags, what, blk_status_to_errno(error), 0,
 847                         NULL, cgid);
 848         rcu_read_unlock();
 849 }
 850
 851 static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
 852 {
 853         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 854                          blk_trace_request_get_cgid(rq));
 855 }
 856
 857 static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
 858 {
 859         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 860                          blk_trace_request_get_cgid(rq));
 861 }
 862
 863 static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
 864 {
 865         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
 866                          blk_trace_request_get_cgid(rq));
 867 }
 868
 869 static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
 870 {
 871         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 872                          blk_trace_request_get_cgid(rq));
 873 }
 874
 875 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 876                         blk_status_t error, unsigned int nr_bytes)
 877 {
 878         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 879                          blk_trace_request_get_cgid(rq));
 880 }
 881
 882 /**
 883  * blk_add_trace_bio - Add a trace for a bio oriented action
 884  * @q:          queue the io is for
 885  * @bio:        the source bio
 886  * @what:       the action
 887  * @error:      error, if any
 888  *
 889  * Description:
 890  *     Records an action against a bio. Will log the bio offset + size.
 891  *
 892  **/
 893 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 894                               u32 what, int error)
 895 {
 896         struct blk_trace *bt;
 897
 898         rcu_read_lock();
 899         bt = rcu_dereference(q->blk_trace);
 900         if (likely(!bt)) {
 901                 rcu_read_unlock();
 902                 return;
 903         }
 904
 905         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 906                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 907                         blk_trace_bio_get_cgid(q, bio));
 908         rcu_read_unlock();
 909 }
 910
 911 static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio)
 912 {
 913         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0);
 914 }
 915
 916 static void blk_add_trace_bio_complete(void *ignore,
 917                                        struct request_queue *q, struct bio *bio)
 918 {
 919         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
 920                           blk_status_to_errno(bio->bi_status));
 921 }
 922
 923 static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio)
 924 {
 925         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE,
 926                         0);
 927 }
 928
 929 static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio)
 930 {
 931         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE,
 932                         0);
 933 }
 934
 935 static void blk_add_trace_bio_queue(void *ignore, struct bio *bio)
 936 {
 937         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0);
 938 }
 939
 940 static void blk_add_trace_getrq(void *ignore, struct bio *bio)
 941 {
 942         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0);
 943 }
 944
 945 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 946 {
 947         struct blk_trace *bt;
 948
 949         rcu_read_lock();
 950         bt = rcu_dereference(q->blk_trace);
 951         if (bt)
 952                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 953         rcu_read_unlock();
 954 }
 955
 956 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 957                                     unsigned int depth, bool explicit)
 958 {
 959         struct blk_trace *bt;
 960
 961         rcu_read_lock();
 962         bt = rcu_dereference(q->blk_trace);
 963         if (bt) {
 964                 __be64 rpdu = cpu_to_be64(depth);
 965                 u32 what;
 966
 967                 if (explicit)
 968                         what = BLK_TA_UNPLUG_IO;
 969                 else
 970                         what = BLK_TA_UNPLUG_TIMER;
 971
 972                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
 973         }
 974         rcu_read_unlock();
 975 }
 976
 977 static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu)
 978 {
 979         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 980         struct blk_trace *bt;
 981
 982         rcu_read_lock();
 983         bt = rcu_dereference(q->blk_trace);
 984         if (bt) {
 985                 __be64 rpdu = cpu_to_be64(pdu);
 986
 987                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 988                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 989                                 BLK_TA_SPLIT,
 990                                 blk_status_to_errno(bio->bi_status),
 991                                 sizeof(rpdu), &rpdu,
 992                                 blk_trace_bio_get_cgid(q, bio));
 993         }
 994         rcu_read_unlock();
 995 }
 996
 997 /**
 998  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 999  * @ignore:     trace callback data parameter (not used)
1000  * @bio:        the source bio
1001  * @dev:        source device
1002  * @from:       source sector
1003  *
1004  * Called after a bio is remapped to a different device and/or sector.
1005  **/
1006 static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev,
1007                                     sector_t from)
1008 {
1009         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1010         struct blk_trace *bt;
1011         struct blk_io_trace_remap r;
1012
1013         rcu_read_lock();
1014         bt = rcu_dereference(q->blk_trace);
1015         if (likely(!bt)) {
1016                 rcu_read_unlock();
1017                 return;
1018         }
1019
1020         r.device_from = cpu_to_be32(dev);
1021         r.device_to   = cpu_to_be32(bio_dev(bio));
1022         r.sector_from = cpu_to_be64(from);
1023
1024         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1025                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1026                         blk_status_to_errno(bio->bi_status),
1027                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1028         rcu_read_unlock();
1029 }
1030
1031 /**
1032  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1033  * @ignore:     trace callback data parameter (not used)
1034  * @rq:         the source request
1035  * @dev:        target device
1036  * @from:       source sector
1037  *
1038  * Description:
1039  *     Device mapper remaps request to other devices.
1040  *     Add a trace for that action.
1041  *
1042  **/
1043 static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
1044                                    sector_t from)
1045 {
1046         struct blk_trace *bt;
1047         struct blk_io_trace_remap r;
1048
1049         rcu_read_lock();
1050         bt = rcu_dereference(rq->q->blk_trace);
1051         if (likely(!bt)) {
1052                 rcu_read_unlock();
1053                 return;
1054         }
1055
1056         r.device_from = cpu_to_be32(dev);
1057         r.device_to   = cpu_to_be32(disk_devt(rq->q->disk));
1058         r.sector_from = cpu_to_be64(from);
1059
1060         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1061                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1062                         sizeof(r), &r, blk_trace_request_get_cgid(rq));
1063         rcu_read_unlock();
1064 }
1065
1066 /**
1067  * blk_add_driver_data - Add binary message with driver-specific data
1068  * @rq:         io request
1069  * @data:       driver-specific data
1070  * @len:        length of driver-specific data
1071  *
1072  * Description:
1073  *     Some drivers might want to write driver-specific data per request.
1074  *
1075  **/
1076 void blk_add_driver_data(struct request *rq, void *data, size_t len)
1077 {
1078         struct blk_trace *bt;
1079
1080         rcu_read_lock();
1081         bt = rcu_dereference(rq->q->blk_trace);
1082         if (likely(!bt)) {
1083                 rcu_read_unlock();
1084                 return;
1085         }
1086
1087         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1088                                 BLK_TA_DRV_DATA, 0, len, data,
1089                                 blk_trace_request_get_cgid(rq));
1090         rcu_read_unlock();
1091 }
1092 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1093
1094 static void blk_register_tracepoints(void)
1095 {
1096         int ret;
1097
1098         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1099         WARN_ON(ret);
1100         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1101         WARN_ON(ret);
1102         ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1103         WARN_ON(ret);
1104         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1105         WARN_ON(ret);
1106         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1107         WARN_ON(ret);
1108         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1109         WARN_ON(ret);
1110         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1111         WARN_ON(ret);
1112         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1113         WARN_ON(ret);
1114         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1115         WARN_ON(ret);
1116         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1117         WARN_ON(ret);
1118         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1119         WARN_ON(ret);
1120         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1121         WARN_ON(ret);
1122         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1123         WARN_ON(ret);
1124         ret = register_trace_block_split(blk_add_trace_split, NULL);
1125         WARN_ON(ret);
1126         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1127         WARN_ON(ret);
1128         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1129         WARN_ON(ret);
1130 }
1131
1132 static void blk_unregister_tracepoints(void)
1133 {
1134         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1135         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1136         unregister_trace_block_split(blk_add_trace_split, NULL);
1137         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1138         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1139         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1140         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1141         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1142         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1143         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1144         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1145         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1146         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1147         unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1148         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1149         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1150
1151         tracepoint_synchronize_unregister();
1152 }
1153
1154 /*
1155  * struct blk_io_tracer formatting routines
1156  */
1157
1158 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1159 {
1160         int i = 0;
1161         int tc = t->action >> BLK_TC_SHIFT;
1162
1163         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1164                 rwbs[i++] = 'N';
1165                 goto out;
1166         }
1167
1168         if (tc & BLK_TC_FLUSH)
1169                 rwbs[i++] = 'F';
1170
1171         if (tc & BLK_TC_DISCARD)
1172                 rwbs[i++] = 'D';
1173         else if (tc & BLK_TC_WRITE)
1174                 rwbs[i++] = 'W';
1175         else if (t->bytes)
1176                 rwbs[i++] = 'R';
1177         else
1178                 rwbs[i++] = 'N';
1179
1180         if (tc & BLK_TC_FUA)
1181                 rwbs[i++] = 'F';
1182         if (tc & BLK_TC_AHEAD)
1183                 rwbs[i++] = 'A';
1184         if (tc & BLK_TC_SYNC)
1185                 rwbs[i++] = 'S';
1186         if (tc & BLK_TC_META)
1187                 rwbs[i++] = 'M';
1188 out:
1189         rwbs[i] = '\0';
1190 }
1191
1192 static inline
1193 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1194 {
1195         return (const struct blk_io_trace *)ent;
1196 }
1197
1198 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1199 {
1200         return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1201 }
1202
1203 static inline u64 t_cgid(const struct trace_entry *ent)
1204 {
1205         return *(u64 *)(te_blk_io_trace(ent) + 1);
1206 }
1207
1208 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1209 {
1210         return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1211 }
1212
1213 static inline u32 t_action(const struct trace_entry *ent)
1214 {
1215         return te_blk_io_trace(ent)->action;
1216 }
1217
1218 static inline u32 t_bytes(const struct trace_entry *ent)
1219 {
1220         return te_blk_io_trace(ent)->bytes;
1221 }
1222
1223 static inline u32 t_sec(const struct trace_entry *ent)
1224 {
1225         return te_blk_io_trace(ent)->bytes >> 9;
1226 }
1227
1228 static inline unsigned long long t_sector(const struct trace_entry *ent)
1229 {
1230         return te_blk_io_trace(ent)->sector;
1231 }
1232
1233 static inline __u16 t_error(const struct trace_entry *ent)
1234 {
1235         return te_blk_io_trace(ent)->error;
1236 }
1237
1238 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1239 {
1240         const __be64 *val = pdu_start(ent, has_cg);
1241         return be64_to_cpu(*val);
1242 }
1243
1244 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1245         bool has_cg);
1246
1247 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1248         bool has_cg)
1249 {
1250         char rwbs[RWBS_LEN];
1251         unsigned long long ts  = iter->ts;
1252         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1253         unsigned secs          = (unsigned long)ts;
1254         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1255
1256         fill_rwbs(rwbs, t);
1257
1258         trace_seq_printf(&iter->seq,
1259                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1260                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1261                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1262 }
1263
1264 static void blk_log_action(struct trace_iterator *iter, const char *act,
1265         bool has_cg)
1266 {
1267         char rwbs[RWBS_LEN];
1268         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1269
1270         fill_rwbs(rwbs, t);
1271         if (has_cg) {
1272                 u64 id = t_cgid(iter->ent);
1273
1274                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1275                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1276
1277                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1278                                 sizeof(blkcg_name_buf));
1279                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1280                                  MAJOR(t->device), MINOR(t->device),
1281                                  blkcg_name_buf, act, rwbs);
1282                 } else {
1283                         /*
1284                          * The cgid portion used to be "INO,GEN".  Userland
1285                          * builds a FILEID_INO32_GEN fid out of them and
1286                          * opens the cgroup using open_by_handle_at(2).
1287                          * While 32bit ino setups are still the same, 64bit
1288                          * ones now use the 64bit ino as the whole ID and
1289                          * no longer use generation.
1290                          *
1291                          * Regardless of the content, always output
1292                          * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1293                          * be mapped back to @id on both 64 and 32bit ino
1294                          * setups.  See __kernfs_fh_to_dentry().
1295                          */
1296                         trace_seq_printf(&iter->seq,
1297                                  "%3d,%-3d %llx,%-llx %2s %3s ",
1298                                  MAJOR(t->device), MINOR(t->device),
1299                                  id & U32_MAX, id >> 32, act, rwbs);
1300                 }
1301         } else
1302                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1303                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1304 }
1305
1306 static void blk_log_dump_pdu(struct trace_seq *s,
1307         const struct trace_entry *ent, bool has_cg)
1308 {
1309         const unsigned char *pdu_buf;
1310         int pdu_len;
1311         int i, end;
1312
1313         pdu_buf = pdu_start(ent, has_cg);
1314         pdu_len = pdu_real_len(ent, has_cg);
1315
1316         if (!pdu_len)
1317                 return;
1318
1319         /* find the last zero that needs to be printed */
1320         for (end = pdu_len - 1; end >= 0; end--)
1321                 if (pdu_buf[end])
1322                         break;
1323         end++;
1324
1325         trace_seq_putc(s, '(');
1326
1327         for (i = 0; i < pdu_len; i++) {
1328
1329                 trace_seq_printf(s, "%s%02x",
1330                                  i == 0 ? "" : " ", pdu_buf[i]);
1331
1332                 /*
1333                  * stop when the rest is just zeros and indicate so
1334                  * with a ".." appended
1335                  */
1336                 if (i == end && end != pdu_len - 1) {
1337                         trace_seq_puts(s, " ..) ");
1338                         return;
1339                 }
1340         }
1341
1342         trace_seq_puts(s, ") ");
1343 }
1344
1345 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1346 {
1347         char cmd[TASK_COMM_LEN];
1348
1349         trace_find_cmdline(ent->pid, cmd);
1350
1351         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1352                 trace_seq_printf(s, "%u ", t_bytes(ent));
1353                 blk_log_dump_pdu(s, ent, has_cg);
1354                 trace_seq_printf(s, "[%s]\n", cmd);
1355         } else {
1356                 if (t_sec(ent))
1357                         trace_seq_printf(s, "%llu + %u [%s]\n",
1358                                                 t_sector(ent), t_sec(ent), cmd);
1359                 else
1360                         trace_seq_printf(s, "[%s]\n", cmd);
1361         }
1362 }
1363
1364 static void blk_log_with_error(struct trace_seq *s,
1365                               const struct trace_entry *ent, bool has_cg)
1366 {
1367         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1368                 blk_log_dump_pdu(s, ent, has_cg);
1369                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1370         } else {
1371                 if (t_sec(ent))
1372                         trace_seq_printf(s, "%llu + %u [%d]\n",
1373                                          t_sector(ent),
1374                                          t_sec(ent), t_error(ent));
1375                 else
1376                         trace_seq_printf(s, "%llu [%d]\n",
1377                                          t_sector(ent), t_error(ent));
1378         }
1379 }
1380
1381 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1382 {
1383         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1384
1385         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1386                          t_sector(ent), t_sec(ent),
1387                          MAJOR(be32_to_cpu(__r->device_from)),
1388                          MINOR(be32_to_cpu(__r->device_from)),
1389                          be64_to_cpu(__r->sector_from));
1390 }
1391
1392 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1393 {
1394         char cmd[TASK_COMM_LEN];
1395
1396         trace_find_cmdline(ent->pid, cmd);
1397
1398         trace_seq_printf(s, "[%s]\n", cmd);
1399 }
1400
1401 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1402 {
1403         char cmd[TASK_COMM_LEN];
1404
1405         trace_find_cmdline(ent->pid, cmd);
1406
1407         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1408 }
1409
1410 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1411 {
1412         char cmd[TASK_COMM_LEN];
1413
1414         trace_find_cmdline(ent->pid, cmd);
1415
1416         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1417                          get_pdu_int(ent, has_cg), cmd);
1418 }
1419
1420 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1421                         bool has_cg)
1422 {
1423
1424         trace_seq_putmem(s, pdu_start(ent, has_cg),
1425                 pdu_real_len(ent, has_cg));
1426         trace_seq_putc(s, '\n');
1427 }
1428
1429 /*
1430  * struct tracer operations
1431  */
1432
1433 static void blk_tracer_print_header(struct seq_file *m)
1434 {
1435         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1436                 return;
1437         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1438                     "#  |     |     |           |   |   |\n");
1439 }
1440
1441 static void blk_tracer_start(struct trace_array *tr)
1442 {
1443         blk_tracer_enabled = true;
1444 }
1445
1446 static int blk_tracer_init(struct trace_array *tr)
1447 {
1448         blk_tr = tr;
1449         blk_tracer_start(tr);
1450         return 0;
1451 }
1452
1453 static void blk_tracer_stop(struct trace_array *tr)
1454 {
1455         blk_tracer_enabled = false;
1456 }
1457
1458 static void blk_tracer_reset(struct trace_array *tr)
1459 {
1460         blk_tracer_stop(tr);
1461 }
1462
1463 static const struct {
1464         const char *act[2];
1465         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1466                             bool has_cg);
1467 } what2act[] = {
1468         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1469         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1470         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1471         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1472         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1473         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1474         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1475         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1476         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1477         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1478         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1479         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1480         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1481         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1482         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1483 };
1484
1485 static enum print_line_t print_one_line(struct trace_iterator *iter,
1486                                         bool classic)
1487 {
1488         struct trace_array *tr = iter->tr;
1489         struct trace_seq *s = &iter->seq;
1490         const struct blk_io_trace *t;
1491         u16 what;
1492         bool long_act;
1493         blk_log_action_t *log_action;
1494         bool has_cg;
1495
1496         t          = te_blk_io_trace(iter->ent);
1497         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1498         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1499         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1500         has_cg     = t->action & __BLK_TA_CGROUP;
1501
1502         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1503                 log_action(iter, long_act ? "message" : "m", has_cg);
1504                 blk_log_msg(s, iter->ent, has_cg);
1505                 return trace_handle_return(s);
1506         }
1507
1508         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1509                 trace_seq_printf(s, "Unknown action %x\n", what);
1510         else {
1511                 log_action(iter, what2act[what].act[long_act], has_cg);
1512                 what2act[what].print(s, iter->ent, has_cg);
1513         }
1514
1515         return trace_handle_return(s);
1516 }
1517
1518 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1519                                                int flags, struct trace_event *event)
1520 {
1521         return print_one_line(iter, false);
1522 }
1523
1524 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1525 {
1526         struct trace_seq *s = &iter->seq;
1527         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1528         const int offset = offsetof(struct blk_io_trace, sector);
1529         struct blk_io_trace old = {
1530                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1531                 .time     = iter->ts,
1532         };
1533
1534         trace_seq_putmem(s, &old, offset);
1535         trace_seq_putmem(s, &t->sector,
1536                          sizeof(old) - offset + t->pdu_len);
1537 }
1538
1539 static enum print_line_t
1540 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1541                              struct trace_event *event)
1542 {
1543         blk_trace_synthesize_old_trace(iter);
1544
1545         return trace_handle_return(&iter->seq);
1546 }
1547
1548 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1549 {
1550         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1551                 return TRACE_TYPE_UNHANDLED;
1552
1553         return print_one_line(iter, true);
1554 }
1555
1556 static int
1557 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1558 {
1559         /* don't output context-info for blk_classic output */
1560         if (bit == TRACE_BLK_OPT_CLASSIC) {
1561                 if (set)
1562                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1563                 else
1564                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1565         }
1566         return 0;
1567 }
1568
1569 static struct tracer blk_tracer __read_mostly = {
1570         .name           = "blk",
1571         .init           = blk_tracer_init,
1572         .reset          = blk_tracer_reset,
1573         .start          = blk_tracer_start,
1574         .stop           = blk_tracer_stop,
1575         .print_header   = blk_tracer_print_header,
1576         .print_line     = blk_tracer_print_line,
1577         .flags          = &blk_tracer_flags,
1578         .set_flag       = blk_tracer_set_flag,
1579 };
1580
1581 static struct trace_event_functions trace_blk_event_funcs = {
1582         .trace          = blk_trace_event_print,
1583         .binary         = blk_trace_event_print_binary,
1584 };
1585
1586 static struct trace_event trace_blk_event = {
1587         .type           = TRACE_BLK,
1588         .funcs          = &trace_blk_event_funcs,
1589 };
1590
1591 static int __init init_blk_tracer(void)
1592 {
1593         if (!register_trace_event(&trace_blk_event)) {
1594                 pr_warn("Warning: could not register block events\n");
1595                 return 1;
1596         }
1597
1598         if (register_tracer(&blk_tracer) != 0) {
1599                 pr_warn("Warning: could not register the block tracer\n");
1600                 unregister_trace_event(&trace_blk_event);
1601                 return 1;
1602         }
1603
1604         return 0;
1605 }
1606
1607 device_initcall(init_blk_tracer);
1608
1609 static int blk_trace_remove_queue(struct request_queue *q)
1610 {
1611         struct blk_trace *bt;
1612
1613         bt = rcu_replace_pointer(q->blk_trace, NULL,
1614                                  lockdep_is_held(&q->debugfs_mutex));
1615         if (bt == NULL)
1616                 return -EINVAL;
1617
1618         if (bt->trace_state == Blktrace_running) {
1619                 bt->trace_state = Blktrace_stopped;
1620                 raw_spin_lock_irq(&running_trace_lock);
1621                 list_del_init(&bt->running_list);
1622                 raw_spin_unlock_irq(&running_trace_lock);
1623                 relay_flush(bt->rchan);
1624         }
1625
1626         put_probe_ref();
1627         synchronize_rcu();
1628         blk_trace_free(q, bt);
1629         return 0;
1630 }
1631
1632 /*
1633  * Setup everything required to start tracing
1634  */
1635 static int blk_trace_setup_queue(struct request_queue *q,
1636                                  struct block_device *bdev)
1637 {
1638         struct blk_trace *bt = NULL;
1639         int ret = -ENOMEM;
1640
1641         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1642         if (!bt)
1643                 return -ENOMEM;
1644
1645         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1646         if (!bt->msg_data)
1647                 goto free_bt;
1648
1649         bt->dev = bdev->bd_dev;
1650         bt->act_mask = (u16)-1;
1651
1652         blk_trace_setup_lba(bt, bdev);
1653
1654         rcu_assign_pointer(q->blk_trace, bt);
1655         get_probe_ref();
1656         return 0;
1657
1658 free_bt:
1659         blk_trace_free(q, bt);
1660         return ret;
1661 }
1662
1663 /*
1664  * sysfs interface to enable and configure tracing
1665  */
1666
1667 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1668                                          struct device_attribute *attr,
1669                                          char *buf);
1670 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1671                                           struct device_attribute *attr,
1672                                           const char *buf, size_t count);
1673 #define BLK_TRACE_DEVICE_ATTR(_name) \
1674         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1675                     sysfs_blk_trace_attr_show, \
1676                     sysfs_blk_trace_attr_store)
1677
1678 static BLK_TRACE_DEVICE_ATTR(enable);
1679 static BLK_TRACE_DEVICE_ATTR(act_mask);
1680 static BLK_TRACE_DEVICE_ATTR(pid);
1681 static BLK_TRACE_DEVICE_ATTR(start_lba);
1682 static BLK_TRACE_DEVICE_ATTR(end_lba);
1683
1684 static struct attribute *blk_trace_attrs[] = {
1685         &dev_attr_enable.attr,
1686         &dev_attr_act_mask.attr,
1687         &dev_attr_pid.attr,
1688         &dev_attr_start_lba.attr,
1689         &dev_attr_end_lba.attr,
1690         NULL
1691 };
1692
1693 struct attribute_group blk_trace_attr_group = {
1694         .name  = "trace",
1695         .attrs = blk_trace_attrs,
1696 };
1697
1698 static const struct {
1699         int mask;
1700         const char *str;
1701 } mask_maps[] = {
1702         { BLK_TC_READ,          "read"          },
1703         { BLK_TC_WRITE,         "write"         },
1704         { BLK_TC_FLUSH,         "flush"         },
1705         { BLK_TC_SYNC,          "sync"          },
1706         { BLK_TC_QUEUE,         "queue"         },
1707         { BLK_TC_REQUEUE,       "requeue"       },
1708         { BLK_TC_ISSUE,         "issue"         },
1709         { BLK_TC_COMPLETE,      "complete"      },
1710         { BLK_TC_FS,            "fs"            },
1711         { BLK_TC_PC,            "pc"            },
1712         { BLK_TC_NOTIFY,        "notify"        },
1713         { BLK_TC_AHEAD,         "ahead"         },
1714         { BLK_TC_META,          "meta"          },
1715         { BLK_TC_DISCARD,       "discard"       },
1716         { BLK_TC_DRV_DATA,      "drv_data"      },
1717         { BLK_TC_FUA,           "fua"           },
1718 };
1719
1720 static int blk_trace_str2mask(const char *str)
1721 {
1722         int i;
1723         int mask = 0;
1724         char *buf, *s, *token;
1725
1726         buf = kstrdup(str, GFP_KERNEL);
1727         if (buf == NULL)
1728                 return -ENOMEM;
1729         s = strstrip(buf);
1730
1731         while (1) {
1732                 token = strsep(&s, ",");
1733                 if (token == NULL)
1734                         break;
1735
1736                 if (*token == '\0')
1737                         continue;
1738
1739                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1740                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1741                                 mask |= mask_maps[i].mask;
1742                                 break;
1743                         }
1744                 }
1745                 if (i == ARRAY_SIZE(mask_maps)) {
1746                         mask = -EINVAL;
1747                         break;
1748                 }
1749         }
1750         kfree(buf);
1751
1752         return mask;
1753 }
1754
1755 static ssize_t blk_trace_mask2str(char *buf, int mask)
1756 {
1757         int i;
1758         char *p = buf;
1759
1760         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1761                 if (mask & mask_maps[i].mask) {
1762                         p += sprintf(p, "%s%s",
1763                                     (p == buf) ? "" : ",", mask_maps[i].str);
1764                 }
1765         }
1766         *p++ = '\n';
1767
1768         return p - buf;
1769 }
1770
1771 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1772                                          struct device_attribute *attr,
1773                                          char *buf)
1774 {
1775         struct block_device *bdev = dev_to_bdev(dev);
1776         struct request_queue *q = bdev_get_queue(bdev);
1777         struct blk_trace *bt;
1778         ssize_t ret = -ENXIO;
1779
1780         mutex_lock(&q->debugfs_mutex);
1781
1782         bt = rcu_dereference_protected(q->blk_trace,
1783                                        lockdep_is_held(&q->debugfs_mutex));
1784         if (attr == &dev_attr_enable) {
1785                 ret = sprintf(buf, "%u\n", !!bt);
1786                 goto out_unlock_bdev;
1787         }
1788
1789         if (bt == NULL)
1790                 ret = sprintf(buf, "disabled\n");
1791         else if (attr == &dev_attr_act_mask)
1792                 ret = blk_trace_mask2str(buf, bt->act_mask);
1793         else if (attr == &dev_attr_pid)
1794                 ret = sprintf(buf, "%u\n", bt->pid);
1795         else if (attr == &dev_attr_start_lba)
1796                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1797         else if (attr == &dev_attr_end_lba)
1798                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1799
1800 out_unlock_bdev:
1801         mutex_unlock(&q->debugfs_mutex);
1802         return ret;
1803 }
1804
1805 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1806                                           struct device_attribute *attr,
1807                                           const char *buf, size_t count)
1808 {
1809         struct block_device *bdev = dev_to_bdev(dev);
1810         struct request_queue *q = bdev_get_queue(bdev);
1811         struct blk_trace *bt;
1812         u64 value;
1813         ssize_t ret = -EINVAL;
1814
1815         if (count == 0)
1816                 goto out;
1817
1818         if (attr == &dev_attr_act_mask) {
1819                 if (kstrtoull(buf, 0, &value)) {
1820                         /* Assume it is a list of trace category names */
1821                         ret = blk_trace_str2mask(buf);
1822                         if (ret < 0)
1823                                 goto out;
1824                         value = ret;
1825                 }
1826         } else {
1827                 if (kstrtoull(buf, 0, &value))
1828                         goto out;
1829         }
1830
1831         mutex_lock(&q->debugfs_mutex);
1832
1833         bt = rcu_dereference_protected(q->blk_trace,
1834                                        lockdep_is_held(&q->debugfs_mutex));
1835         if (attr == &dev_attr_enable) {
1836                 if (!!value == !!bt) {
1837                         ret = 0;
1838                         goto out_unlock_bdev;
1839                 }
1840                 if (value)
1841                         ret = blk_trace_setup_queue(q, bdev);
1842                 else
1843                         ret = blk_trace_remove_queue(q);
1844                 goto out_unlock_bdev;
1845         }
1846
1847         ret = 0;
1848         if (bt == NULL) {
1849                 ret = blk_trace_setup_queue(q, bdev);
1850                 bt = rcu_dereference_protected(q->blk_trace,
1851                                 lockdep_is_held(&q->debugfs_mutex));
1852         }
1853
1854         if (ret == 0) {
1855                 if (attr == &dev_attr_act_mask)
1856                         bt->act_mask = value;
1857                 else if (attr == &dev_attr_pid)
1858                         bt->pid = value;
1859                 else if (attr == &dev_attr_start_lba)
1860                         bt->start_lba = value;
1861                 else if (attr == &dev_attr_end_lba)
1862                         bt->end_lba = value;
1863         }
1864
1865 out_unlock_bdev:
1866         mutex_unlock(&q->debugfs_mutex);
1867 out:
1868         return ret ? ret : count;
1869 }
1870 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1871
1872 #ifdef CONFIG_EVENT_TRACING
1873
1874 /**
1875  * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string.
1876  * @rwbs:       buffer to be filled
1877  * @op:         REQ_OP_XXX for the tracepoint
1878  *
1879  * Description:
1880  *     Maps the REQ_OP_XXX to character and fills the buffer provided by the
1881  *     caller with resulting string.
1882  *
1883  **/
1884 void blk_fill_rwbs(char *rwbs, unsigned int op)
1885 {
1886         int i = 0;
1887
1888         if (op & REQ_PREFLUSH)
1889                 rwbs[i++] = 'F';
1890
1891         switch (op & REQ_OP_MASK) {
1892         case REQ_OP_WRITE:
1893                 rwbs[i++] = 'W';
1894                 break;
1895         case REQ_OP_DISCARD:
1896                 rwbs[i++] = 'D';
1897                 break;
1898         case REQ_OP_SECURE_ERASE:
1899                 rwbs[i++] = 'D';
1900                 rwbs[i++] = 'E';
1901                 break;
1902         case REQ_OP_FLUSH:
1903                 rwbs[i++] = 'F';
1904                 break;
1905         case REQ_OP_READ:
1906                 rwbs[i++] = 'R';
1907                 break;
1908         default:
1909                 rwbs[i++] = 'N';
1910         }
1911
1912         if (op & REQ_FUA)
1913                 rwbs[i++] = 'F';
1914         if (op & REQ_RAHEAD)
1915                 rwbs[i++] = 'A';
1916         if (op & REQ_SYNC)
1917                 rwbs[i++] = 'S';
1918         if (op & REQ_META)
1919                 rwbs[i++] = 'M';
1920
1921         rwbs[i] = '\0';
1922 }
1923 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1924
1925 #endif /* CONFIG_EVENT_TRACING */
1926