kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/export.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/list.h>
  30
  31 #include <trace/events/block.h>
  32
  33 #include "trace_output.h"
  34
  35 #ifdef CONFIG_BLK_DEV_IO_TRACE
  36
  37 static unsigned int blktrace_seq __read_mostly = 1;
  38
  39 static struct trace_array *blk_tr;
  40 static bool blk_tracer_enabled __read_mostly;
  41
  42 static LIST_HEAD(running_trace_list);
  43 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  44
  45 /* Select an alternative, minimalistic output than the original one */
  46 #define TRACE_BLK_OPT_CLASSIC   0x1
  47
  48 static struct tracer_opt blk_tracer_opts[] = {
  49         /* Default disable the minimalistic output */
  50         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  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 atomic_t blk_probes_ref = ATOMIC_INIT(0);
  61
  62 static void blk_register_tracepoints(void);
  63 static void blk_unregister_tracepoints(void);
  64
  65 /*
  66  * Send out a notify message.
  67  */
  68 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  69                        const void *data, size_t len)
  70 {
  71         struct blk_io_trace *t;
  72         struct ring_buffer_event *event = NULL;
  73         struct ring_buffer *buffer = NULL;
  74         int pc = 0;
  75         int cpu = smp_processor_id();
  76         bool blk_tracer = blk_tracer_enabled;
  77
  78         if (blk_tracer) {
  79                 buffer = blk_tr->trace_buffer.buffer;
  80                 pc = preempt_count();
  81                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  82                                                   sizeof(*t) + len,
  83                                                   0, pc);
  84                 if (!event)
  85                         return;
  86                 t = ring_buffer_event_data(event);
  87                 goto record_it;
  88         }
  89
  90         if (!bt->rchan)
  91                 return;
  92
  93         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  94         if (t) {
  95                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  96                 t->time = ktime_to_ns(ktime_get());
  97 record_it:
  98                 t->device = bt->dev;
  99                 t->action = action;
 100                 t->pid = pid;
 101                 t->cpu = cpu;
 102                 t->pdu_len = len;
 103                 memcpy((void *) t + sizeof(*t), data, len);
 104
 105                 if (blk_tracer)
 106                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 107         }
 108 }
 109
 110 /*
 111  * Send out a notify for this process, if we haven't done so since a trace
 112  * started
 113  */
 114 static void trace_note_tsk(struct task_struct *tsk)
 115 {
 116         unsigned long flags;
 117         struct blk_trace *bt;
 118
 119         tsk->btrace_seq = blktrace_seq;
 120         spin_lock_irqsave(&running_trace_lock, flags);
 121         list_for_each_entry(bt, &running_trace_list, running_list) {
 122                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 123                            sizeof(tsk->comm));
 124         }
 125         spin_unlock_irqrestore(&running_trace_lock, flags);
 126 }
 127
 128 static void trace_note_time(struct blk_trace *bt)
 129 {
 130         struct timespec now;
 131         unsigned long flags;
 132         u32 words[2];
 133
 134         getnstimeofday(&now);
 135         words[0] = now.tv_sec;
 136         words[1] = now.tv_nsec;
 137
 138         local_irq_save(flags);
 139         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 140         local_irq_restore(flags);
 141 }
 142
 143 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 144 {
 145         int n;
 146         va_list args;
 147         unsigned long flags;
 148         char *buf;
 149
 150         if (unlikely(bt->trace_state != Blktrace_running &&
 151                      !blk_tracer_enabled))
 152                 return;
 153
 154         /*
 155          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 156          * message to the trace.
 157          */
 158         if (!(bt->act_mask & BLK_TC_NOTIFY))
 159                 return;
 160
 161         local_irq_save(flags);
 162         buf = this_cpu_ptr(bt->msg_data);
 163         va_start(args, fmt);
 164         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 165         va_end(args);
 166
 167         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 168         local_irq_restore(flags);
 169 }
 170 EXPORT_SYMBOL_GPL(__trace_note_message);
 171
 172 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 173                          pid_t pid)
 174 {
 175         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 176                 return 1;
 177         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 178                 return 1;
 179         if (bt->pid && pid != bt->pid)
 180                 return 1;
 181
 182         return 0;
 183 }
 184
 185 /*
 186  * Data direction bit lookup
 187  */
 188 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 189                                  BLK_TC_ACT(BLK_TC_WRITE) };
 190
 191 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 192 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 193
 194 /* The ilog2() calls fall out because they're constant */
 195 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 196           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 197
 198 /*
 199  * The worker for the various blk_add_trace*() types. Fills out a
 200  * blk_io_trace structure and places it in a per-cpu subbuffer.
 201  */
 202 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 203                      int op, int op_flags, u32 what, int error, int pdu_len,
 204                      void *pdu_data)
 205 {
 206         struct task_struct *tsk = current;
 207         struct ring_buffer_event *event = NULL;
 208         struct ring_buffer *buffer = NULL;
 209         struct blk_io_trace *t;
 210         unsigned long flags = 0;
 211         unsigned long *sequence;
 212         pid_t pid;
 213         int cpu, pc = 0;
 214         bool blk_tracer = blk_tracer_enabled;
 215
 216         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 217                 return;
 218
 219         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 220         what |= MASK_TC_BIT(op_flags, SYNC);
 221         what |= MASK_TC_BIT(op_flags, RAHEAD);
 222         what |= MASK_TC_BIT(op_flags, META);
 223         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 224         what |= MASK_TC_BIT(op_flags, FUA);
 225         if (op == REQ_OP_DISCARD)
 226                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 227         if (op == REQ_OP_FLUSH)
 228                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 229
 230         pid = tsk->pid;
 231         if (act_log_check(bt, what, sector, pid))
 232                 return;
 233         cpu = raw_smp_processor_id();
 234
 235         if (blk_tracer) {
 236                 tracing_record_cmdline(current);
 237
 238                 buffer = blk_tr->trace_buffer.buffer;
 239                 pc = preempt_count();
 240                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 241                                                   sizeof(*t) + pdu_len,
 242                                                   0, pc);
 243                 if (!event)
 244                         return;
 245                 t = ring_buffer_event_data(event);
 246                 goto record_it;
 247         }
 248
 249         if (unlikely(tsk->btrace_seq != blktrace_seq))
 250                 trace_note_tsk(tsk);
 251
 252         /*
 253          * A word about the locking here - we disable interrupts to reserve
 254          * some space in the relay per-cpu buffer, to prevent an irq
 255          * from coming in and stepping on our toes.
 256          */
 257         local_irq_save(flags);
 258         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 259         if (t) {
 260                 sequence = per_cpu_ptr(bt->sequence, cpu);
 261
 262                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 263                 t->sequence = ++(*sequence);
 264                 t->time = ktime_to_ns(ktime_get());
 265 record_it:
 266                 /*
 267                  * These two are not needed in ftrace as they are in the
 268                  * generic trace_entry, filled by tracing_generic_entry_update,
 269                  * but for the trace_event->bin() synthesizer benefit we do it
 270                  * here too.
 271                  */
 272                 t->cpu = cpu;
 273                 t->pid = pid;
 274
 275                 t->sector = sector;
 276                 t->bytes = bytes;
 277                 t->action = what;
 278                 t->device = bt->dev;
 279                 t->error = error;
 280                 t->pdu_len = pdu_len;
 281
 282                 if (pdu_len)
 283                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 284
 285                 if (blk_tracer) {
 286                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 287                         return;
 288                 }
 289         }
 290
 291         local_irq_restore(flags);
 292 }
 293
 294 static struct dentry *blk_tree_root;
 295 static DEFINE_MUTEX(blk_tree_mutex);
 296
 297 static void blk_trace_free(struct blk_trace *bt)
 298 {
 299         debugfs_remove(bt->msg_file);
 300         debugfs_remove(bt->dropped_file);
 301         relay_close(bt->rchan);
 302         debugfs_remove(bt->dir);
 303         free_percpu(bt->sequence);
 304         free_percpu(bt->msg_data);
 305         kfree(bt);
 306 }
 307
 308 static void blk_trace_cleanup(struct blk_trace *bt)
 309 {
 310         blk_trace_free(bt);
 311         if (atomic_dec_and_test(&blk_probes_ref))
 312                 blk_unregister_tracepoints();
 313 }
 314
 315 int blk_trace_remove(struct request_queue *q)
 316 {
 317         struct blk_trace *bt;
 318
 319         bt = xchg(&q->blk_trace, NULL);
 320         if (!bt)
 321                 return -EINVAL;
 322
 323         if (bt->trace_state != Blktrace_running)
 324                 blk_trace_cleanup(bt);
 325
 326         return 0;
 327 }
 328 EXPORT_SYMBOL_GPL(blk_trace_remove);
 329
 330 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 331                                 size_t count, loff_t *ppos)
 332 {
 333         struct blk_trace *bt = filp->private_data;
 334         char buf[16];
 335
 336         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 337
 338         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 339 }
 340
 341 static const struct file_operations blk_dropped_fops = {
 342         .owner =        THIS_MODULE,
 343         .open =         simple_open,
 344         .read =         blk_dropped_read,
 345         .llseek =       default_llseek,
 346 };
 347
 348 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 349                                 size_t count, loff_t *ppos)
 350 {
 351         char *msg;
 352         struct blk_trace *bt;
 353
 354         if (count >= BLK_TN_MAX_MSG)
 355                 return -EINVAL;
 356
 357         msg = memdup_user_nul(buffer, count);
 358         if (IS_ERR(msg))
 359                 return PTR_ERR(msg);
 360
 361         bt = filp->private_data;
 362         __trace_note_message(bt, "%s", msg);
 363         kfree(msg);
 364
 365         return count;
 366 }
 367
 368 static const struct file_operations blk_msg_fops = {
 369         .owner =        THIS_MODULE,
 370         .open =         simple_open,
 371         .write =        blk_msg_write,
 372         .llseek =       noop_llseek,
 373 };
 374
 375 /*
 376  * Keep track of how many times we encountered a full subbuffer, to aid
 377  * the user space app in telling how many lost events there were.
 378  */
 379 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 380                                      void *prev_subbuf, size_t prev_padding)
 381 {
 382         struct blk_trace *bt;
 383
 384         if (!relay_buf_full(buf))
 385                 return 1;
 386
 387         bt = buf->chan->private_data;
 388         atomic_inc(&bt->dropped);
 389         return 0;
 390 }
 391
 392 static int blk_remove_buf_file_callback(struct dentry *dentry)
 393 {
 394         debugfs_remove(dentry);
 395
 396         return 0;
 397 }
 398
 399 static struct dentry *blk_create_buf_file_callback(const char *filename,
 400                                                    struct dentry *parent,
 401                                                    umode_t mode,
 402                                                    struct rchan_buf *buf,
 403                                                    int *is_global)
 404 {
 405         return debugfs_create_file(filename, mode, parent, buf,
 406                                         &relay_file_operations);
 407 }
 408
 409 static struct rchan_callbacks blk_relay_callbacks = {
 410         .subbuf_start           = blk_subbuf_start_callback,
 411         .create_buf_file        = blk_create_buf_file_callback,
 412         .remove_buf_file        = blk_remove_buf_file_callback,
 413 };
 414
 415 static void blk_trace_setup_lba(struct blk_trace *bt,
 416                                 struct block_device *bdev)
 417 {
 418         struct hd_struct *part = NULL;
 419
 420         if (bdev)
 421                 part = bdev->bd_part;
 422
 423         if (part) {
 424                 bt->start_lba = part->start_sect;
 425                 bt->end_lba = part->start_sect + part->nr_sects;
 426         } else {
 427                 bt->start_lba = 0;
 428                 bt->end_lba = -1ULL;
 429         }
 430 }
 431
 432 /*
 433  * Setup everything required to start tracing
 434  */
 435 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 436                        struct block_device *bdev,
 437                        struct blk_user_trace_setup *buts)
 438 {
 439         struct blk_trace *bt = NULL;
 440         struct dentry *dir = NULL;
 441         int ret;
 442
 443         if (!buts->buf_size || !buts->buf_nr)
 444                 return -EINVAL;
 445
 446         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 447         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 448
 449         /*
 450          * some device names have larger paths - convert the slashes
 451          * to underscores for this to work as expected
 452          */
 453         strreplace(buts->name, '/', '_');
 454
 455         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 456         if (!bt)
 457                 return -ENOMEM;
 458
 459         ret = -ENOMEM;
 460         bt->sequence = alloc_percpu(unsigned long);
 461         if (!bt->sequence)
 462                 goto err;
 463
 464         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 465         if (!bt->msg_data)
 466                 goto err;
 467
 468         ret = -ENOENT;
 469
 470         mutex_lock(&blk_tree_mutex);
 471         if (!blk_tree_root) {
 472                 blk_tree_root = debugfs_create_dir("block", NULL);
 473                 if (!blk_tree_root) {
 474                         mutex_unlock(&blk_tree_mutex);
 475                         goto err;
 476                 }
 477         }
 478         mutex_unlock(&blk_tree_mutex);
 479
 480         dir = debugfs_create_dir(buts->name, blk_tree_root);
 481
 482         if (!dir)
 483                 goto err;
 484
 485         bt->dir = dir;
 486         bt->dev = dev;
 487         atomic_set(&bt->dropped, 0);
 488         INIT_LIST_HEAD(&bt->running_list);
 489
 490         ret = -EIO;
 491         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 492                                                &blk_dropped_fops);
 493         if (!bt->dropped_file)
 494                 goto err;
 495
 496         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 497         if (!bt->msg_file)
 498                 goto err;
 499
 500         bt->rchan = relay_open("trace", dir, buts->buf_size,
 501                                 buts->buf_nr, &blk_relay_callbacks, bt);
 502         if (!bt->rchan)
 503                 goto err;
 504
 505         bt->act_mask = buts->act_mask;
 506         if (!bt->act_mask)
 507                 bt->act_mask = (u16) -1;
 508
 509         blk_trace_setup_lba(bt, bdev);
 510
 511         /* overwrite with user settings */
 512         if (buts->start_lba)
 513                 bt->start_lba = buts->start_lba;
 514         if (buts->end_lba)
 515                 bt->end_lba = buts->end_lba;
 516
 517         bt->pid = buts->pid;
 518         bt->trace_state = Blktrace_setup;
 519
 520         ret = -EBUSY;
 521         if (cmpxchg(&q->blk_trace, NULL, bt))
 522                 goto err;
 523
 524         if (atomic_inc_return(&blk_probes_ref) == 1)
 525                 blk_register_tracepoints();
 526
 527         return 0;
 528 err:
 529         blk_trace_free(bt);
 530         return ret;
 531 }
 532
 533 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 534                     struct block_device *bdev,
 535                     char __user *arg)
 536 {
 537         struct blk_user_trace_setup buts;
 538         int ret;
 539
 540         ret = copy_from_user(&buts, arg, sizeof(buts));
 541         if (ret)
 542                 return -EFAULT;
 543
 544         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 545         if (ret)
 546                 return ret;
 547
 548         if (copy_to_user(arg, &buts, sizeof(buts))) {
 549                 blk_trace_remove(q);
 550                 return -EFAULT;
 551         }
 552         return 0;
 553 }
 554 EXPORT_SYMBOL_GPL(blk_trace_setup);
 555
 556 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 557 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 558                                   dev_t dev, struct block_device *bdev,
 559                                   char __user *arg)
 560 {
 561         struct blk_user_trace_setup buts;
 562         struct compat_blk_user_trace_setup cbuts;
 563         int ret;
 564
 565         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 566                 return -EFAULT;
 567
 568         buts = (struct blk_user_trace_setup) {
 569                 .act_mask = cbuts.act_mask,
 570                 .buf_size = cbuts.buf_size,
 571                 .buf_nr = cbuts.buf_nr,
 572                 .start_lba = cbuts.start_lba,
 573                 .end_lba = cbuts.end_lba,
 574                 .pid = cbuts.pid,
 575         };
 576
 577         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 578         if (ret)
 579                 return ret;
 580
 581         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 582                 blk_trace_remove(q);
 583                 return -EFAULT;
 584         }
 585
 586         return 0;
 587 }
 588 #endif
 589
 590 int blk_trace_startstop(struct request_queue *q, int start)
 591 {
 592         int ret;
 593         struct blk_trace *bt = q->blk_trace;
 594
 595         if (bt == NULL)
 596                 return -EINVAL;
 597
 598         /*
 599          * For starting a trace, we can transition from a setup or stopped
 600          * trace. For stopping a trace, the state must be running
 601          */
 602         ret = -EINVAL;
 603         if (start) {
 604                 if (bt->trace_state == Blktrace_setup ||
 605                     bt->trace_state == Blktrace_stopped) {
 606                         blktrace_seq++;
 607                         smp_mb();
 608                         bt->trace_state = Blktrace_running;
 609                         spin_lock_irq(&running_trace_lock);
 610                         list_add(&bt->running_list, &running_trace_list);
 611                         spin_unlock_irq(&running_trace_lock);
 612
 613                         trace_note_time(bt);
 614                         ret = 0;
 615                 }
 616         } else {
 617                 if (bt->trace_state == Blktrace_running) {
 618                         bt->trace_state = Blktrace_stopped;
 619                         spin_lock_irq(&running_trace_lock);
 620                         list_del_init(&bt->running_list);
 621                         spin_unlock_irq(&running_trace_lock);
 622                         relay_flush(bt->rchan);
 623                         ret = 0;
 624                 }
 625         }
 626
 627         return ret;
 628 }
 629 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 630
 631 /**
 632  * blk_trace_ioctl: - handle the ioctls associated with tracing
 633  * @bdev:       the block device
 634  * @cmd:        the ioctl cmd
 635  * @arg:        the argument data, if any
 636  *
 637  **/
 638 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 639 {
 640         struct request_queue *q;
 641         int ret, start = 0;
 642         char b[BDEVNAME_SIZE];
 643
 644         q = bdev_get_queue(bdev);
 645         if (!q)
 646                 return -ENXIO;
 647
 648         mutex_lock(&bdev->bd_mutex);
 649
 650         switch (cmd) {
 651         case BLKTRACESETUP:
 652                 bdevname(bdev, b);
 653                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 654                 break;
 655 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 656         case BLKTRACESETUP32:
 657                 bdevname(bdev, b);
 658                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 659                 break;
 660 #endif
 661         case BLKTRACESTART:
 662                 start = 1;
 663         case BLKTRACESTOP:
 664                 ret = blk_trace_startstop(q, start);
 665                 break;
 666         case BLKTRACETEARDOWN:
 667                 ret = blk_trace_remove(q);
 668                 break;
 669         default:
 670                 ret = -ENOTTY;
 671                 break;
 672         }
 673
 674         mutex_unlock(&bdev->bd_mutex);
 675         return ret;
 676 }
 677
 678 /**
 679  * blk_trace_shutdown: - stop and cleanup trace structures
 680  * @q:    the request queue associated with the device
 681  *
 682  **/
 683 void blk_trace_shutdown(struct request_queue *q)
 684 {
 685         if (q->blk_trace) {
 686                 blk_trace_startstop(q, 0);
 687                 blk_trace_remove(q);
 688         }
 689 }
 690
 691 /*
 692  * blktrace probes
 693  */
 694
 695 /**
 696  * blk_add_trace_rq - Add a trace for a request oriented action
 697  * @q:          queue the io is for
 698  * @rq:         the source request
 699  * @nr_bytes:   number of completed bytes
 700  * @what:       the action
 701  *
 702  * Description:
 703  *     Records an action against a request. Will log the bio offset + size.
 704  *
 705  **/
 706 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 707                              unsigned int nr_bytes, u32 what)
 708 {
 709         struct blk_trace *bt = q->blk_trace;
 710
 711         if (likely(!bt))
 712                 return;
 713
 714         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 715                 what |= BLK_TC_ACT(BLK_TC_PC);
 716                 __blk_add_trace(bt, 0, nr_bytes, req_op(rq), rq->cmd_flags,
 717                                 what, rq->errors, rq->cmd_len, rq->cmd);
 718         } else  {
 719                 what |= BLK_TC_ACT(BLK_TC_FS);
 720                 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes, req_op(rq),
 721                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 722         }
 723 }
 724
 725 static void blk_add_trace_rq_abort(void *ignore,
 726                                    struct request_queue *q, struct request *rq)
 727 {
 728         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
 729 }
 730
 731 static void blk_add_trace_rq_insert(void *ignore,
 732                                     struct request_queue *q, struct request *rq)
 733 {
 734         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
 735 }
 736
 737 static void blk_add_trace_rq_issue(void *ignore,
 738                                    struct request_queue *q, struct request *rq)
 739 {
 740         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
 741 }
 742
 743 static void blk_add_trace_rq_requeue(void *ignore,
 744                                      struct request_queue *q,
 745                                      struct request *rq)
 746 {
 747         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 748 }
 749
 750 static void blk_add_trace_rq_complete(void *ignore,
 751                                       struct request_queue *q,
 752                                       struct request *rq,
 753                                       unsigned int nr_bytes)
 754 {
 755         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 756 }
 757
 758 /**
 759  * blk_add_trace_bio - Add a trace for a bio oriented action
 760  * @q:          queue the io is for
 761  * @bio:        the source bio
 762  * @what:       the action
 763  * @error:      error, if any
 764  *
 765  * Description:
 766  *     Records an action against a bio. Will log the bio offset + size.
 767  *
 768  **/
 769 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 770                               u32 what, int error)
 771 {
 772         struct blk_trace *bt = q->blk_trace;
 773
 774         if (likely(!bt))
 775                 return;
 776
 777         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 778                         bio_op(bio), bio->bi_rw, what, error, 0, NULL);
 779 }
 780
 781 static void blk_add_trace_bio_bounce(void *ignore,
 782                                      struct request_queue *q, struct bio *bio)
 783 {
 784         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 785 }
 786
 787 static void blk_add_trace_bio_complete(void *ignore,
 788                                        struct request_queue *q, struct bio *bio,
 789                                        int error)
 790 {
 791         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 792 }
 793
 794 static void blk_add_trace_bio_backmerge(void *ignore,
 795                                         struct request_queue *q,
 796                                         struct request *rq,
 797                                         struct bio *bio)
 798 {
 799         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 800 }
 801
 802 static void blk_add_trace_bio_frontmerge(void *ignore,
 803                                          struct request_queue *q,
 804                                          struct request *rq,
 805                                          struct bio *bio)
 806 {
 807         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 808 }
 809
 810 static void blk_add_trace_bio_queue(void *ignore,
 811                                     struct request_queue *q, struct bio *bio)
 812 {
 813         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 814 }
 815
 816 static void blk_add_trace_getrq(void *ignore,
 817                                 struct request_queue *q,
 818                                 struct bio *bio, int rw)
 819 {
 820         if (bio)
 821                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 822         else {
 823                 struct blk_trace *bt = q->blk_trace;
 824
 825                 if (bt)
 826                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 827                                         NULL);
 828         }
 829 }
 830
 831
 832 static void blk_add_trace_sleeprq(void *ignore,
 833                                   struct request_queue *q,
 834                                   struct bio *bio, int rw)
 835 {
 836         if (bio)
 837                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 838         else {
 839                 struct blk_trace *bt = q->blk_trace;
 840
 841                 if (bt)
 842                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 843                                         0, 0, NULL);
 844         }
 845 }
 846
 847 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 848 {
 849         struct blk_trace *bt = q->blk_trace;
 850
 851         if (bt)
 852                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 853 }
 854
 855 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 856                                     unsigned int depth, bool explicit)
 857 {
 858         struct blk_trace *bt = q->blk_trace;
 859
 860         if (bt) {
 861                 __be64 rpdu = cpu_to_be64(depth);
 862                 u32 what;
 863
 864                 if (explicit)
 865                         what = BLK_TA_UNPLUG_IO;
 866                 else
 867                         what = BLK_TA_UNPLUG_TIMER;
 868
 869                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 870         }
 871 }
 872
 873 static void blk_add_trace_split(void *ignore,
 874                                 struct request_queue *q, struct bio *bio,
 875                                 unsigned int pdu)
 876 {
 877         struct blk_trace *bt = q->blk_trace;
 878
 879         if (bt) {
 880                 __be64 rpdu = cpu_to_be64(pdu);
 881
 882                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 883                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_rw,
 884                                 BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
 885                                 &rpdu);
 886         }
 887 }
 888
 889 /**
 890  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 891  * @ignore:     trace callback data parameter (not used)
 892  * @q:          queue the io is for
 893  * @bio:        the source bio
 894  * @dev:        target device
 895  * @from:       source sector
 896  *
 897  * Description:
 898  *     Device mapper or raid target sometimes need to split a bio because
 899  *     it spans a stripe (or similar). Add a trace for that action.
 900  *
 901  **/
 902 static void blk_add_trace_bio_remap(void *ignore,
 903                                     struct request_queue *q, struct bio *bio,
 904                                     dev_t dev, sector_t from)
 905 {
 906         struct blk_trace *bt = q->blk_trace;
 907         struct blk_io_trace_remap r;
 908
 909         if (likely(!bt))
 910                 return;
 911
 912         r.device_from = cpu_to_be32(dev);
 913         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
 914         r.sector_from = cpu_to_be64(from);
 915
 916         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 917                         bio_op(bio), bio->bi_rw, BLK_TA_REMAP, bio->bi_error,
 918                         sizeof(r), &r);
 919 }
 920
 921 /**
 922  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 923  * @ignore:     trace callback data parameter (not used)
 924  * @q:          queue the io is for
 925  * @rq:         the source request
 926  * @dev:        target device
 927  * @from:       source sector
 928  *
 929  * Description:
 930  *     Device mapper remaps request to other devices.
 931  *     Add a trace for that action.
 932  *
 933  **/
 934 static void blk_add_trace_rq_remap(void *ignore,
 935                                    struct request_queue *q,
 936                                    struct request *rq, dev_t dev,
 937                                    sector_t from)
 938 {
 939         struct blk_trace *bt = q->blk_trace;
 940         struct blk_io_trace_remap r;
 941
 942         if (likely(!bt))
 943                 return;
 944
 945         r.device_from = cpu_to_be32(dev);
 946         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
 947         r.sector_from = cpu_to_be64(from);
 948
 949         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
 950                         rq_data_dir(rq), 0, BLK_TA_REMAP, !!rq->errors,
 951                         sizeof(r), &r);
 952 }
 953
 954 /**
 955  * blk_add_driver_data - Add binary message with driver-specific data
 956  * @q:          queue the io is for
 957  * @rq:         io request
 958  * @data:       driver-specific data
 959  * @len:        length of driver-specific data
 960  *
 961  * Description:
 962  *     Some drivers might want to write driver-specific data per request.
 963  *
 964  **/
 965 void blk_add_driver_data(struct request_queue *q,
 966                          struct request *rq,
 967                          void *data, size_t len)
 968 {
 969         struct blk_trace *bt = q->blk_trace;
 970
 971         if (likely(!bt))
 972                 return;
 973
 974         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
 975                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 0,
 976                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 977         else
 978                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 0,
 979                                 BLK_TA_DRV_DATA, rq->errors, len, data);
 980 }
 981 EXPORT_SYMBOL_GPL(blk_add_driver_data);
 982
 983 static void blk_register_tracepoints(void)
 984 {
 985         int ret;
 986
 987         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
 988         WARN_ON(ret);
 989         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
 990         WARN_ON(ret);
 991         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
 992         WARN_ON(ret);
 993         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
 994         WARN_ON(ret);
 995         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
 996         WARN_ON(ret);
 997         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
 998         WARN_ON(ret);
 999         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1000         WARN_ON(ret);
1001         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1002         WARN_ON(ret);
1003         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1004         WARN_ON(ret);
1005         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1006         WARN_ON(ret);
1007         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1008         WARN_ON(ret);
1009         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1010         WARN_ON(ret);
1011         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1012         WARN_ON(ret);
1013         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1014         WARN_ON(ret);
1015         ret = register_trace_block_split(blk_add_trace_split, NULL);
1016         WARN_ON(ret);
1017         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1018         WARN_ON(ret);
1019         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1020         WARN_ON(ret);
1021 }
1022
1023 static void blk_unregister_tracepoints(void)
1024 {
1025         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1026         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1027         unregister_trace_block_split(blk_add_trace_split, NULL);
1028         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1029         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1030         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1031         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1032         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1033         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1034         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1035         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1036         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1037         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1038         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1039         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1040         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1041         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1042
1043         tracepoint_synchronize_unregister();
1044 }
1045
1046 /*
1047  * struct blk_io_tracer formatting routines
1048  */
1049
1050 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1051 {
1052         int i = 0;
1053         int tc = t->action >> BLK_TC_SHIFT;
1054
1055         if (t->action == BLK_TN_MESSAGE) {
1056                 rwbs[i++] = 'N';
1057                 goto out;
1058         }
1059
1060         if (tc & BLK_TC_FLUSH)
1061                 rwbs[i++] = 'F';
1062
1063         if (tc & BLK_TC_DISCARD)
1064                 rwbs[i++] = 'D';
1065         else if (tc & BLK_TC_WRITE)
1066                 rwbs[i++] = 'W';
1067         else if (t->bytes)
1068                 rwbs[i++] = 'R';
1069         else
1070                 rwbs[i++] = 'N';
1071
1072         if (tc & BLK_TC_FUA)
1073                 rwbs[i++] = 'F';
1074         if (tc & BLK_TC_AHEAD)
1075                 rwbs[i++] = 'A';
1076         if (tc & BLK_TC_SYNC)
1077                 rwbs[i++] = 'S';
1078         if (tc & BLK_TC_META)
1079                 rwbs[i++] = 'M';
1080 out:
1081         rwbs[i] = '\0';
1082 }
1083
1084 static inline
1085 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1086 {
1087         return (const struct blk_io_trace *)ent;
1088 }
1089
1090 static inline const void *pdu_start(const struct trace_entry *ent)
1091 {
1092         return te_blk_io_trace(ent) + 1;
1093 }
1094
1095 static inline u32 t_action(const struct trace_entry *ent)
1096 {
1097         return te_blk_io_trace(ent)->action;
1098 }
1099
1100 static inline u32 t_bytes(const struct trace_entry *ent)
1101 {
1102         return te_blk_io_trace(ent)->bytes;
1103 }
1104
1105 static inline u32 t_sec(const struct trace_entry *ent)
1106 {
1107         return te_blk_io_trace(ent)->bytes >> 9;
1108 }
1109
1110 static inline unsigned long long t_sector(const struct trace_entry *ent)
1111 {
1112         return te_blk_io_trace(ent)->sector;
1113 }
1114
1115 static inline __u16 t_error(const struct trace_entry *ent)
1116 {
1117         return te_blk_io_trace(ent)->error;
1118 }
1119
1120 static __u64 get_pdu_int(const struct trace_entry *ent)
1121 {
1122         const __u64 *val = pdu_start(ent);
1123         return be64_to_cpu(*val);
1124 }
1125
1126 static void get_pdu_remap(const struct trace_entry *ent,
1127                           struct blk_io_trace_remap *r)
1128 {
1129         const struct blk_io_trace_remap *__r = pdu_start(ent);
1130         __u64 sector_from = __r->sector_from;
1131
1132         r->device_from = be32_to_cpu(__r->device_from);
1133         r->device_to   = be32_to_cpu(__r->device_to);
1134         r->sector_from = be64_to_cpu(sector_from);
1135 }
1136
1137 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1138
1139 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1140 {
1141         char rwbs[RWBS_LEN];
1142         unsigned long long ts  = iter->ts;
1143         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1144         unsigned secs          = (unsigned long)ts;
1145         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1146
1147         fill_rwbs(rwbs, t);
1148
1149         trace_seq_printf(&iter->seq,
1150                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1151                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1152                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1153 }
1154
1155 static void blk_log_action(struct trace_iterator *iter, const char *act)
1156 {
1157         char rwbs[RWBS_LEN];
1158         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1159
1160         fill_rwbs(rwbs, t);
1161         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1162                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1163 }
1164
1165 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1166 {
1167         const unsigned char *pdu_buf;
1168         int pdu_len;
1169         int i, end;
1170
1171         pdu_buf = pdu_start(ent);
1172         pdu_len = te_blk_io_trace(ent)->pdu_len;
1173
1174         if (!pdu_len)
1175                 return;
1176
1177         /* find the last zero that needs to be printed */
1178         for (end = pdu_len - 1; end >= 0; end--)
1179                 if (pdu_buf[end])
1180                         break;
1181         end++;
1182
1183         trace_seq_putc(s, '(');
1184
1185         for (i = 0; i < pdu_len; i++) {
1186
1187                 trace_seq_printf(s, "%s%02x",
1188                                  i == 0 ? "" : " ", pdu_buf[i]);
1189
1190                 /*
1191                  * stop when the rest is just zeroes and indicate so
1192                  * with a ".." appended
1193                  */
1194                 if (i == end && end != pdu_len - 1) {
1195                         trace_seq_puts(s, " ..) ");
1196                         return;
1197                 }
1198         }
1199
1200         trace_seq_puts(s, ") ");
1201 }
1202
1203 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1204 {
1205         char cmd[TASK_COMM_LEN];
1206
1207         trace_find_cmdline(ent->pid, cmd);
1208
1209         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1210                 trace_seq_printf(s, "%u ", t_bytes(ent));
1211                 blk_log_dump_pdu(s, ent);
1212                 trace_seq_printf(s, "[%s]\n", cmd);
1213         } else {
1214                 if (t_sec(ent))
1215                         trace_seq_printf(s, "%llu + %u [%s]\n",
1216                                                 t_sector(ent), t_sec(ent), cmd);
1217                 else
1218                         trace_seq_printf(s, "[%s]\n", cmd);
1219         }
1220 }
1221
1222 static void blk_log_with_error(struct trace_seq *s,
1223                               const struct trace_entry *ent)
1224 {
1225         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1226                 blk_log_dump_pdu(s, ent);
1227                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1228         } else {
1229                 if (t_sec(ent))
1230                         trace_seq_printf(s, "%llu + %u [%d]\n",
1231                                          t_sector(ent),
1232                                          t_sec(ent), t_error(ent));
1233                 else
1234                         trace_seq_printf(s, "%llu [%d]\n",
1235                                          t_sector(ent), t_error(ent));
1236         }
1237 }
1238
1239 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1240 {
1241         struct blk_io_trace_remap r = { .device_from = 0, };
1242
1243         get_pdu_remap(ent, &r);
1244         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1245                          t_sector(ent), t_sec(ent),
1246                          MAJOR(r.device_from), MINOR(r.device_from),
1247                          (unsigned long long)r.sector_from);
1248 }
1249
1250 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1251 {
1252         char cmd[TASK_COMM_LEN];
1253
1254         trace_find_cmdline(ent->pid, cmd);
1255
1256         trace_seq_printf(s, "[%s]\n", cmd);
1257 }
1258
1259 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1260 {
1261         char cmd[TASK_COMM_LEN];
1262
1263         trace_find_cmdline(ent->pid, cmd);
1264
1265         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1266 }
1267
1268 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1269 {
1270         char cmd[TASK_COMM_LEN];
1271
1272         trace_find_cmdline(ent->pid, cmd);
1273
1274         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1275                          get_pdu_int(ent), cmd);
1276 }
1277
1278 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1279 {
1280         const struct blk_io_trace *t = te_blk_io_trace(ent);
1281
1282         trace_seq_putmem(s, t + 1, t->pdu_len);
1283         trace_seq_putc(s, '\n');
1284 }
1285
1286 /*
1287  * struct tracer operations
1288  */
1289
1290 static void blk_tracer_print_header(struct seq_file *m)
1291 {
1292         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1293                 return;
1294         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1295                     "#  |     |     |           |   |   |\n");
1296 }
1297
1298 static void blk_tracer_start(struct trace_array *tr)
1299 {
1300         blk_tracer_enabled = true;
1301 }
1302
1303 static int blk_tracer_init(struct trace_array *tr)
1304 {
1305         blk_tr = tr;
1306         blk_tracer_start(tr);
1307         return 0;
1308 }
1309
1310 static void blk_tracer_stop(struct trace_array *tr)
1311 {
1312         blk_tracer_enabled = false;
1313 }
1314
1315 static void blk_tracer_reset(struct trace_array *tr)
1316 {
1317         blk_tracer_stop(tr);
1318 }
1319
1320 static const struct {
1321         const char *act[2];
1322         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1323 } what2act[] = {
1324         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1325         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1326         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1327         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1328         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1329         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1330         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1331         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1332         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1333         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1334         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1335         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1336         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1337         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1338         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1339 };
1340
1341 static enum print_line_t print_one_line(struct trace_iterator *iter,
1342                                         bool classic)
1343 {
1344         struct trace_array *tr = iter->tr;
1345         struct trace_seq *s = &iter->seq;
1346         const struct blk_io_trace *t;
1347         u16 what;
1348         bool long_act;
1349         blk_log_action_t *log_action;
1350
1351         t          = te_blk_io_trace(iter->ent);
1352         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1353         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1354         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1355
1356         if (t->action == BLK_TN_MESSAGE) {
1357                 log_action(iter, long_act ? "message" : "m");
1358                 blk_log_msg(s, iter->ent);
1359                 return trace_handle_return(s);
1360         }
1361
1362         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1363                 trace_seq_printf(s, "Unknown action %x\n", what);
1364         else {
1365                 log_action(iter, what2act[what].act[long_act]);
1366                 what2act[what].print(s, iter->ent);
1367         }
1368
1369         return trace_handle_return(s);
1370 }
1371
1372 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1373                                                int flags, struct trace_event *event)
1374 {
1375         return print_one_line(iter, false);
1376 }
1377
1378 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1379 {
1380         struct trace_seq *s = &iter->seq;
1381         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1382         const int offset = offsetof(struct blk_io_trace, sector);
1383         struct blk_io_trace old = {
1384                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1385                 .time     = iter->ts,
1386         };
1387
1388         trace_seq_putmem(s, &old, offset);
1389         trace_seq_putmem(s, &t->sector,
1390                          sizeof(old) - offset + t->pdu_len);
1391 }
1392
1393 static enum print_line_t
1394 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1395                              struct trace_event *event)
1396 {
1397         blk_trace_synthesize_old_trace(iter);
1398
1399         return trace_handle_return(&iter->seq);
1400 }
1401
1402 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1403 {
1404         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1405                 return TRACE_TYPE_UNHANDLED;
1406
1407         return print_one_line(iter, true);
1408 }
1409
1410 static int
1411 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1412 {
1413         /* don't output context-info for blk_classic output */
1414         if (bit == TRACE_BLK_OPT_CLASSIC) {
1415                 if (set)
1416                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1417                 else
1418                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1419         }
1420         return 0;
1421 }
1422
1423 static struct tracer blk_tracer __read_mostly = {
1424         .name           = "blk",
1425         .init           = blk_tracer_init,
1426         .reset          = blk_tracer_reset,
1427         .start          = blk_tracer_start,
1428         .stop           = blk_tracer_stop,
1429         .print_header   = blk_tracer_print_header,
1430         .print_line     = blk_tracer_print_line,
1431         .flags          = &blk_tracer_flags,
1432         .set_flag       = blk_tracer_set_flag,
1433 };
1434
1435 static struct trace_event_functions trace_blk_event_funcs = {
1436         .trace          = blk_trace_event_print,
1437         .binary         = blk_trace_event_print_binary,
1438 };
1439
1440 static struct trace_event trace_blk_event = {
1441         .type           = TRACE_BLK,
1442         .funcs          = &trace_blk_event_funcs,
1443 };
1444
1445 static int __init init_blk_tracer(void)
1446 {
1447         if (!register_trace_event(&trace_blk_event)) {
1448                 pr_warn("Warning: could not register block events\n");
1449                 return 1;
1450         }
1451
1452         if (register_tracer(&blk_tracer) != 0) {
1453                 pr_warn("Warning: could not register the block tracer\n");
1454                 unregister_trace_event(&trace_blk_event);
1455                 return 1;
1456         }
1457
1458         return 0;
1459 }
1460
1461 device_initcall(init_blk_tracer);
1462
1463 static int blk_trace_remove_queue(struct request_queue *q)
1464 {
1465         struct blk_trace *bt;
1466
1467         bt = xchg(&q->blk_trace, NULL);
1468         if (bt == NULL)
1469                 return -EINVAL;
1470
1471         if (atomic_dec_and_test(&blk_probes_ref))
1472                 blk_unregister_tracepoints();
1473
1474         blk_trace_free(bt);
1475         return 0;
1476 }
1477
1478 /*
1479  * Setup everything required to start tracing
1480  */
1481 static int blk_trace_setup_queue(struct request_queue *q,
1482                                  struct block_device *bdev)
1483 {
1484         struct blk_trace *bt = NULL;
1485         int ret = -ENOMEM;
1486
1487         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1488         if (!bt)
1489                 return -ENOMEM;
1490
1491         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1492         if (!bt->msg_data)
1493                 goto free_bt;
1494
1495         bt->dev = bdev->bd_dev;
1496         bt->act_mask = (u16)-1;
1497
1498         blk_trace_setup_lba(bt, bdev);
1499
1500         ret = -EBUSY;
1501         if (cmpxchg(&q->blk_trace, NULL, bt))
1502                 goto free_bt;
1503
1504         if (atomic_inc_return(&blk_probes_ref) == 1)
1505                 blk_register_tracepoints();
1506         return 0;
1507
1508 free_bt:
1509         blk_trace_free(bt);
1510         return ret;
1511 }
1512
1513 /*
1514  * sysfs interface to enable and configure tracing
1515  */
1516
1517 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1518                                          struct device_attribute *attr,
1519                                          char *buf);
1520 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1521                                           struct device_attribute *attr,
1522                                           const char *buf, size_t count);
1523 #define BLK_TRACE_DEVICE_ATTR(_name) \
1524         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1525                     sysfs_blk_trace_attr_show, \
1526                     sysfs_blk_trace_attr_store)
1527
1528 static BLK_TRACE_DEVICE_ATTR(enable);
1529 static BLK_TRACE_DEVICE_ATTR(act_mask);
1530 static BLK_TRACE_DEVICE_ATTR(pid);
1531 static BLK_TRACE_DEVICE_ATTR(start_lba);
1532 static BLK_TRACE_DEVICE_ATTR(end_lba);
1533
1534 static struct attribute *blk_trace_attrs[] = {
1535         &dev_attr_enable.attr,
1536         &dev_attr_act_mask.attr,
1537         &dev_attr_pid.attr,
1538         &dev_attr_start_lba.attr,
1539         &dev_attr_end_lba.attr,
1540         NULL
1541 };
1542
1543 struct attribute_group blk_trace_attr_group = {
1544         .name  = "trace",
1545         .attrs = blk_trace_attrs,
1546 };
1547
1548 static const struct {
1549         int mask;
1550         const char *str;
1551 } mask_maps[] = {
1552         { BLK_TC_READ,          "read"          },
1553         { BLK_TC_WRITE,         "write"         },
1554         { BLK_TC_FLUSH,         "flush"         },
1555         { BLK_TC_SYNC,          "sync"          },
1556         { BLK_TC_QUEUE,         "queue"         },
1557         { BLK_TC_REQUEUE,       "requeue"       },
1558         { BLK_TC_ISSUE,         "issue"         },
1559         { BLK_TC_COMPLETE,      "complete"      },
1560         { BLK_TC_FS,            "fs"            },
1561         { BLK_TC_PC,            "pc"            },
1562         { BLK_TC_NOTIFY,        "notify"        },
1563         { BLK_TC_AHEAD,         "ahead"         },
1564         { BLK_TC_META,          "meta"          },
1565         { BLK_TC_DISCARD,       "discard"       },
1566         { BLK_TC_DRV_DATA,      "drv_data"      },
1567         { BLK_TC_FUA,           "fua"           },
1568 };
1569
1570 static int blk_trace_str2mask(const char *str)
1571 {
1572         int i;
1573         int mask = 0;
1574         char *buf, *s, *token;
1575
1576         buf = kstrdup(str, GFP_KERNEL);
1577         if (buf == NULL)
1578                 return -ENOMEM;
1579         s = strstrip(buf);
1580
1581         while (1) {
1582                 token = strsep(&s, ",");
1583                 if (token == NULL)
1584                         break;
1585
1586                 if (*token == '\0')
1587                         continue;
1588
1589                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1590                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1591                                 mask |= mask_maps[i].mask;
1592                                 break;
1593                         }
1594                 }
1595                 if (i == ARRAY_SIZE(mask_maps)) {
1596                         mask = -EINVAL;
1597                         break;
1598                 }
1599         }
1600         kfree(buf);
1601
1602         return mask;
1603 }
1604
1605 static ssize_t blk_trace_mask2str(char *buf, int mask)
1606 {
1607         int i;
1608         char *p = buf;
1609
1610         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1611                 if (mask & mask_maps[i].mask) {
1612                         p += sprintf(p, "%s%s",
1613                                     (p == buf) ? "" : ",", mask_maps[i].str);
1614                 }
1615         }
1616         *p++ = '\n';
1617
1618         return p - buf;
1619 }
1620
1621 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1622 {
1623         if (bdev->bd_disk == NULL)
1624                 return NULL;
1625
1626         return bdev_get_queue(bdev);
1627 }
1628
1629 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1630                                          struct device_attribute *attr,
1631                                          char *buf)
1632 {
1633         struct hd_struct *p = dev_to_part(dev);
1634         struct request_queue *q;
1635         struct block_device *bdev;
1636         ssize_t ret = -ENXIO;
1637
1638         bdev = bdget(part_devt(p));
1639         if (bdev == NULL)
1640                 goto out;
1641
1642         q = blk_trace_get_queue(bdev);
1643         if (q == NULL)
1644                 goto out_bdput;
1645
1646         mutex_lock(&bdev->bd_mutex);
1647
1648         if (attr == &dev_attr_enable) {
1649                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1650                 goto out_unlock_bdev;
1651         }
1652
1653         if (q->blk_trace == NULL)
1654                 ret = sprintf(buf, "disabled\n");
1655         else if (attr == &dev_attr_act_mask)
1656                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1657         else if (attr == &dev_attr_pid)
1658                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1659         else if (attr == &dev_attr_start_lba)
1660                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1661         else if (attr == &dev_attr_end_lba)
1662                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1663
1664 out_unlock_bdev:
1665         mutex_unlock(&bdev->bd_mutex);
1666 out_bdput:
1667         bdput(bdev);
1668 out:
1669         return ret;
1670 }
1671
1672 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1673                                           struct device_attribute *attr,
1674                                           const char *buf, size_t count)
1675 {
1676         struct block_device *bdev;
1677         struct request_queue *q;
1678         struct hd_struct *p;
1679         u64 value;
1680         ssize_t ret = -EINVAL;
1681
1682         if (count == 0)
1683                 goto out;
1684
1685         if (attr == &dev_attr_act_mask) {
1686                 if (sscanf(buf, "%llx", &value) != 1) {
1687                         /* Assume it is a list of trace category names */
1688                         ret = blk_trace_str2mask(buf);
1689                         if (ret < 0)
1690                                 goto out;
1691                         value = ret;
1692                 }
1693         } else if (sscanf(buf, "%llu", &value) != 1)
1694                 goto out;
1695
1696         ret = -ENXIO;
1697
1698         p = dev_to_part(dev);
1699         bdev = bdget(part_devt(p));
1700         if (bdev == NULL)
1701                 goto out;
1702
1703         q = blk_trace_get_queue(bdev);
1704         if (q == NULL)
1705                 goto out_bdput;
1706
1707         mutex_lock(&bdev->bd_mutex);
1708
1709         if (attr == &dev_attr_enable) {
1710                 if (value)
1711                         ret = blk_trace_setup_queue(q, bdev);
1712                 else
1713                         ret = blk_trace_remove_queue(q);
1714                 goto out_unlock_bdev;
1715         }
1716
1717         ret = 0;
1718         if (q->blk_trace == NULL)
1719                 ret = blk_trace_setup_queue(q, bdev);
1720
1721         if (ret == 0) {
1722                 if (attr == &dev_attr_act_mask)
1723                         q->blk_trace->act_mask = value;
1724                 else if (attr == &dev_attr_pid)
1725                         q->blk_trace->pid = value;
1726                 else if (attr == &dev_attr_start_lba)
1727                         q->blk_trace->start_lba = value;
1728                 else if (attr == &dev_attr_end_lba)
1729                         q->blk_trace->end_lba = value;
1730         }
1731
1732 out_unlock_bdev:
1733         mutex_unlock(&bdev->bd_mutex);
1734 out_bdput:
1735         bdput(bdev);
1736 out:
1737         return ret ? ret : count;
1738 }
1739
1740 int blk_trace_init_sysfs(struct device *dev)
1741 {
1742         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1743 }
1744
1745 void blk_trace_remove_sysfs(struct device *dev)
1746 {
1747         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1748 }
1749
1750 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1751
1752 #ifdef CONFIG_EVENT_TRACING
1753
1754 void blk_dump_cmd(char *buf, struct request *rq)
1755 {
1756         int i, end;
1757         int len = rq->cmd_len;
1758         unsigned char *cmd = rq->cmd;
1759
1760         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1761                 buf[0] = '\0';
1762                 return;
1763         }
1764
1765         for (end = len - 1; end >= 0; end--)
1766                 if (cmd[end])
1767                         break;
1768         end++;
1769
1770         for (i = 0; i < len; i++) {
1771                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1772                 if (i == end && end != len - 1) {
1773                         sprintf(buf, " ..");
1774                         break;
1775                 }
1776         }
1777 }
1778
1779 void blk_fill_rwbs(char *rwbs, int op, u32 rw, int bytes)
1780 {
1781         int i = 0;
1782
1783         if (rw & REQ_PREFLUSH)
1784                 rwbs[i++] = 'F';
1785
1786         switch (op) {
1787         case REQ_OP_WRITE:
1788         case REQ_OP_WRITE_SAME:
1789                 rwbs[i++] = 'W';
1790                 break;
1791         case REQ_OP_DISCARD:
1792                 rwbs[i++] = 'D';
1793                 break;
1794         case REQ_OP_SECURE_ERASE:
1795                 rwbs[i++] = 'D';
1796                 rwbs[i++] = 'E';
1797                 break;
1798         case REQ_OP_FLUSH:
1799                 rwbs[i++] = 'F';
1800                 break;
1801         case REQ_OP_READ:
1802                 rwbs[i++] = 'R';
1803                 break;
1804         default:
1805                 rwbs[i++] = 'N';
1806         }
1807
1808         if (rw & REQ_FUA)
1809                 rwbs[i++] = 'F';
1810         if (rw & REQ_RAHEAD)
1811                 rwbs[i++] = 'A';
1812         if (rw & REQ_SYNC)
1813                 rwbs[i++] = 'S';
1814         if (rw & REQ_META)
1815                 rwbs[i++] = 'M';
1816
1817         rwbs[i] = '\0';
1818 }
1819 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1820
1821 #endif /* CONFIG_EVENT_TRACING */
1822