kernel/trace/blktrace.c

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