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
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/kernel.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/blktrace_api.h>
  24 #include <linux/percpu.h>
  25 #include <linux/init.h>
  26 #include <linux/mutex.h>
  27 #include <linux/slab.h>
  28 #include <linux/debugfs.h>
  29 #include <linux/export.h>
  30 #include <linux/time.h>
  31 #include <linux/uaccess.h>
  32 #include <linux/list.h>
  33
  34 #include <trace/events/block.h>
  35
  36 #include "trace_output.h"
  37
  38 #ifdef CONFIG_BLK_DEV_IO_TRACE
  39
  40 static unsigned int blktrace_seq __read_mostly = 1;
  41
  42 static struct trace_array *blk_tr;
  43 static bool blk_tracer_enabled __read_mostly;
  44
  45 static LIST_HEAD(running_trace_list);
  46 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  47
  48 /* Select an alternative, minimalistic output than the original one */
  49 #define TRACE_BLK_OPT_CLASSIC   0x1
  50
  51 static struct tracer_opt blk_tracer_opts[] = {
  52         /* Default disable the minimalistic output */
  53         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  54         { }
  55 };
  56
  57 static struct tracer_flags blk_tracer_flags = {
  58         .val  = 0,
  59         .opts = blk_tracer_opts,
  60 };
  61
  62 /* Global reference count of probes */
  63 static DEFINE_MUTEX(blk_probe_mutex);
  64 static int blk_probes_ref;
  65
  66 static void blk_register_tracepoints(void);
  67 static void blk_unregister_tracepoints(void);
  68
  69 /*
  70  * Send out a notify message.
  71  */
  72 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  73                        const void *data, size_t len)
  74 {
  75         struct blk_io_trace *t;
  76         struct ring_buffer_event *event = NULL;
  77         struct ring_buffer *buffer = NULL;
  78         int pc = 0;
  79         int cpu = smp_processor_id();
  80         bool blk_tracer = blk_tracer_enabled;
  81
  82         if (blk_tracer) {
  83                 buffer = blk_tr->trace_buffer.buffer;
  84                 pc = preempt_count();
  85                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  86                                                   sizeof(*t) + len,
  87                                                   0, pc);
  88                 if (!event)
  89                         return;
  90                 t = ring_buffer_event_data(event);
  91                 goto record_it;
  92         }
  93
  94         if (!bt->rchan)
  95                 return;
  96
  97         t = relay_reserve(bt->rchan, sizeof(*t) + len);
  98         if (t) {
  99                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 100                 t->time = ktime_to_ns(ktime_get());
 101 record_it:
 102                 t->device = bt->dev;
 103                 t->action = action;
 104                 t->pid = pid;
 105                 t->cpu = cpu;
 106                 t->pdu_len = len;
 107                 memcpy((void *) t + sizeof(*t), data, len);
 108
 109                 if (blk_tracer)
 110                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 111         }
 112 }
 113
 114 /*
 115  * Send out a notify for this process, if we haven't done so since a trace
 116  * started
 117  */
 118 static void trace_note_tsk(struct task_struct *tsk)
 119 {
 120         unsigned long flags;
 121         struct blk_trace *bt;
 122
 123         tsk->btrace_seq = blktrace_seq;
 124         spin_lock_irqsave(&running_trace_lock, flags);
 125         list_for_each_entry(bt, &running_trace_list, running_list) {
 126                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 127                            sizeof(tsk->comm));
 128         }
 129         spin_unlock_irqrestore(&running_trace_lock, flags);
 130 }
 131
 132 static void trace_note_time(struct blk_trace *bt)
 133 {
 134         struct timespec64 now;
 135         unsigned long flags;
 136         u32 words[2];
 137
 138         /* need to check user space to see if this breaks in y2038 or y2106 */
 139         ktime_get_real_ts64(&now);
 140         words[0] = (u32)now.tv_sec;
 141         words[1] = now.tv_nsec;
 142
 143         local_irq_save(flags);
 144         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 145         local_irq_restore(flags);
 146 }
 147
 148 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 149 {
 150         int n;
 151         va_list args;
 152         unsigned long flags;
 153         char *buf;
 154
 155         if (unlikely(bt->trace_state != Blktrace_running &&
 156                      !blk_tracer_enabled))
 157                 return;
 158
 159         /*
 160          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 161          * message to the trace.
 162          */
 163         if (!(bt->act_mask & BLK_TC_NOTIFY))
 164                 return;
 165
 166         local_irq_save(flags);
 167         buf = this_cpu_ptr(bt->msg_data);
 168         va_start(args, fmt);
 169         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 170         va_end(args);
 171
 172         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 173         local_irq_restore(flags);
 174 }
 175 EXPORT_SYMBOL_GPL(__trace_note_message);
 176
 177 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 178                          pid_t pid)
 179 {
 180         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 181                 return 1;
 182         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 183                 return 1;
 184         if (bt->pid && pid != bt->pid)
 185                 return 1;
 186
 187         return 0;
 188 }
 189
 190 /*
 191  * Data direction bit lookup
 192  */
 193 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 194                                  BLK_TC_ACT(BLK_TC_WRITE) };
 195
 196 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 197 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 198
 199 /* The ilog2() calls fall out because they're constant */
 200 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 201           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 202
 203 /*
 204  * The worker for the various blk_add_trace*() types. Fills out a
 205  * blk_io_trace structure and places it in a per-cpu subbuffer.
 206  */
 207 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 208                      int op, int op_flags, u32 what, int error, int pdu_len,
 209                      void *pdu_data)
 210 {
 211         struct task_struct *tsk = current;
 212         struct ring_buffer_event *event = NULL;
 213         struct ring_buffer *buffer = NULL;
 214         struct blk_io_trace *t;
 215         unsigned long flags = 0;
 216         unsigned long *sequence;
 217         pid_t pid;
 218         int cpu, pc = 0;
 219         bool blk_tracer = blk_tracer_enabled;
 220
 221         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 222                 return;
 223
 224         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 225         what |= MASK_TC_BIT(op_flags, SYNC);
 226         what |= MASK_TC_BIT(op_flags, RAHEAD);
 227         what |= MASK_TC_BIT(op_flags, META);
 228         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 229         what |= MASK_TC_BIT(op_flags, FUA);
 230         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 231                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 232         if (op == REQ_OP_FLUSH)
 233                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 234
 235         pid = tsk->pid;
 236         if (act_log_check(bt, what, sector, pid))
 237                 return;
 238         cpu = raw_smp_processor_id();
 239
 240         if (blk_tracer) {
 241                 tracing_record_cmdline(current);
 242
 243                 buffer = blk_tr->trace_buffer.buffer;
 244                 pc = preempt_count();
 245                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 246                                                   sizeof(*t) + pdu_len,
 247                                                   0, pc);
 248                 if (!event)
 249                         return;
 250                 t = ring_buffer_event_data(event);
 251                 goto record_it;
 252         }
 253
 254         if (unlikely(tsk->btrace_seq != blktrace_seq))
 255                 trace_note_tsk(tsk);
 256
 257         /*
 258          * A word about the locking here - we disable interrupts to reserve
 259          * some space in the relay per-cpu buffer, to prevent an irq
 260          * from coming in and stepping on our toes.
 261          */
 262         local_irq_save(flags);
 263         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 264         if (t) {
 265                 sequence = per_cpu_ptr(bt->sequence, cpu);
 266
 267                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 268                 t->sequence = ++(*sequence);
 269                 t->time = ktime_to_ns(ktime_get());
 270 record_it:
 271                 /*
 272                  * These two are not needed in ftrace as they are in the
 273                  * generic trace_entry, filled by tracing_generic_entry_update,
 274                  * but for the trace_event->bin() synthesizer benefit we do it
 275                  * here too.
 276                  */
 277                 t->cpu = cpu;
 278                 t->pid = pid;
 279
 280                 t->sector = sector;
 281                 t->bytes = bytes;
 282                 t->action = what;
 283                 t->device = bt->dev;
 284                 t->error = error;
 285                 t->pdu_len = pdu_len;
 286
 287                 if (pdu_len)
 288                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 289
 290                 if (blk_tracer) {
 291                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 292                         return;
 293                 }
 294         }
 295
 296         local_irq_restore(flags);
 297 }
 298
 299 static struct dentry *blk_tree_root;
 300 static DEFINE_MUTEX(blk_tree_mutex);
 301
 302 static void blk_trace_free(struct blk_trace *bt)
 303 {
 304         debugfs_remove(bt->msg_file);
 305         debugfs_remove(bt->dropped_file);
 306         relay_close(bt->rchan);
 307         debugfs_remove(bt->dir);
 308         free_percpu(bt->sequence);
 309         free_percpu(bt->msg_data);
 310         kfree(bt);
 311 }
 312
 313 static void get_probe_ref(void)
 314 {
 315         mutex_lock(&blk_probe_mutex);
 316         if (++blk_probes_ref == 1)
 317                 blk_register_tracepoints();
 318         mutex_unlock(&blk_probe_mutex);
 319 }
 320
 321 static void put_probe_ref(void)
 322 {
 323         mutex_lock(&blk_probe_mutex);
 324         if (!--blk_probes_ref)
 325                 blk_unregister_tracepoints();
 326         mutex_unlock(&blk_probe_mutex);
 327 }
 328
 329 static void blk_trace_cleanup(struct blk_trace *bt)
 330 {
 331         synchronize_rcu();
 332         blk_trace_free(bt);
 333         put_probe_ref();
 334 }
 335
 336 static int __blk_trace_remove(struct request_queue *q)
 337 {
 338         struct blk_trace *bt;
 339
 340         bt = xchg(&q->blk_trace, NULL);
 341         if (!bt)
 342                 return -EINVAL;
 343
 344         if (bt->trace_state != Blktrace_running)
 345                 blk_trace_cleanup(bt);
 346
 347         return 0;
 348 }
 349
 350 int blk_trace_remove(struct request_queue *q)
 351 {
 352         int ret;
 353
 354         mutex_lock(&q->blk_trace_mutex);
 355         ret = __blk_trace_remove(q);
 356         mutex_unlock(&q->blk_trace_mutex);
 357
 358         return ret;
 359 }
 360 EXPORT_SYMBOL_GPL(blk_trace_remove);
 361
 362 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 363                                 size_t count, loff_t *ppos)
 364 {
 365         struct blk_trace *bt = filp->private_data;
 366         char buf[16];
 367
 368         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 369
 370         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 371 }
 372
 373 static const struct file_operations blk_dropped_fops = {
 374         .owner =        THIS_MODULE,
 375         .open =         simple_open,
 376         .read =         blk_dropped_read,
 377         .llseek =       default_llseek,
 378 };
 379
 380 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 381                                 size_t count, loff_t *ppos)
 382 {
 383         char *msg;
 384         struct blk_trace *bt;
 385
 386         if (count >= BLK_TN_MAX_MSG)
 387                 return -EINVAL;
 388
 389         msg = memdup_user_nul(buffer, count);
 390         if (IS_ERR(msg))
 391                 return PTR_ERR(msg);
 392
 393         bt = filp->private_data;
 394         __trace_note_message(bt, "%s", msg);
 395         kfree(msg);
 396
 397         return count;
 398 }
 399
 400 static const struct file_operations blk_msg_fops = {
 401         .owner =        THIS_MODULE,
 402         .open =         simple_open,
 403         .write =        blk_msg_write,
 404         .llseek =       noop_llseek,
 405 };
 406
 407 /*
 408  * Keep track of how many times we encountered a full subbuffer, to aid
 409  * the user space app in telling how many lost events there were.
 410  */
 411 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 412                                      void *prev_subbuf, size_t prev_padding)
 413 {
 414         struct blk_trace *bt;
 415
 416         if (!relay_buf_full(buf))
 417                 return 1;
 418
 419         bt = buf->chan->private_data;
 420         atomic_inc(&bt->dropped);
 421         return 0;
 422 }
 423
 424 static int blk_remove_buf_file_callback(struct dentry *dentry)
 425 {
 426         debugfs_remove(dentry);
 427
 428         return 0;
 429 }
 430
 431 static struct dentry *blk_create_buf_file_callback(const char *filename,
 432                                                    struct dentry *parent,
 433                                                    umode_t mode,
 434                                                    struct rchan_buf *buf,
 435                                                    int *is_global)
 436 {
 437         return debugfs_create_file(filename, mode, parent, buf,
 438                                         &relay_file_operations);
 439 }
 440
 441 static struct rchan_callbacks blk_relay_callbacks = {
 442         .subbuf_start           = blk_subbuf_start_callback,
 443         .create_buf_file        = blk_create_buf_file_callback,
 444         .remove_buf_file        = blk_remove_buf_file_callback,
 445 };
 446
 447 static void blk_trace_setup_lba(struct blk_trace *bt,
 448                                 struct block_device *bdev)
 449 {
 450         struct hd_struct *part = NULL;
 451
 452         if (bdev)
 453                 part = bdev->bd_part;
 454
 455         if (part) {
 456                 bt->start_lba = part->start_sect;
 457                 bt->end_lba = part->start_sect + part->nr_sects;
 458         } else {
 459                 bt->start_lba = 0;
 460                 bt->end_lba = -1ULL;
 461         }
 462 }
 463
 464 /*
 465  * Setup everything required to start tracing
 466  */
 467 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 468                        struct block_device *bdev,
 469                        struct blk_user_trace_setup *buts)
 470 {
 471         struct blk_trace *bt = NULL;
 472         struct dentry *dir = NULL;
 473         int ret;
 474
 475         if (!buts->buf_size || !buts->buf_nr)
 476                 return -EINVAL;
 477
 478         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 479         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 480
 481         /*
 482          * some device names have larger paths - convert the slashes
 483          * to underscores for this to work as expected
 484          */
 485         strreplace(buts->name, '/', '_');
 486
 487         /*
 488          * bdev can be NULL, as with scsi-generic, this is a helpful as
 489          * we can be.
 490          */
 491         if (q->blk_trace) {
 492                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 493                         buts->name);
 494                 return -EBUSY;
 495         }
 496
 497         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 498         if (!bt)
 499                 return -ENOMEM;
 500
 501         ret = -ENOMEM;
 502         bt->sequence = alloc_percpu(unsigned long);
 503         if (!bt->sequence)
 504                 goto err;
 505
 506         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 507         if (!bt->msg_data)
 508                 goto err;
 509
 510         ret = -ENOENT;
 511
 512         mutex_lock(&blk_tree_mutex);
 513         if (!blk_tree_root) {
 514                 blk_tree_root = debugfs_create_dir("block", NULL);
 515                 if (!blk_tree_root) {
 516                         mutex_unlock(&blk_tree_mutex);
 517                         goto err;
 518                 }
 519         }
 520         mutex_unlock(&blk_tree_mutex);
 521
 522         dir = debugfs_create_dir(buts->name, blk_tree_root);
 523
 524         if (!dir)
 525                 goto err;
 526
 527         bt->dir = dir;
 528         bt->dev = dev;
 529         atomic_set(&bt->dropped, 0);
 530         INIT_LIST_HEAD(&bt->running_list);
 531
 532         ret = -EIO;
 533         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 534                                                &blk_dropped_fops);
 535         if (!bt->dropped_file)
 536                 goto err;
 537
 538         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 539         if (!bt->msg_file)
 540                 goto err;
 541
 542         bt->rchan = relay_open("trace", dir, buts->buf_size,
 543                                 buts->buf_nr, &blk_relay_callbacks, bt);
 544         if (!bt->rchan)
 545                 goto err;
 546
 547         bt->act_mask = buts->act_mask;
 548         if (!bt->act_mask)
 549                 bt->act_mask = (u16) -1;
 550
 551         blk_trace_setup_lba(bt, bdev);
 552
 553         /* overwrite with user settings */
 554         if (buts->start_lba)
 555                 bt->start_lba = buts->start_lba;
 556         if (buts->end_lba)
 557                 bt->end_lba = buts->end_lba;
 558
 559         bt->pid = buts->pid;
 560         bt->trace_state = Blktrace_setup;
 561
 562         ret = -EBUSY;
 563         if (cmpxchg(&q->blk_trace, NULL, bt))
 564                 goto err;
 565
 566         get_probe_ref();
 567
 568         return 0;
 569 err:
 570         blk_trace_free(bt);
 571         return ret;
 572 }
 573
 574 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 575                              struct block_device *bdev, char __user *arg)
 576 {
 577         struct blk_user_trace_setup buts;
 578         int ret;
 579
 580         ret = copy_from_user(&buts, arg, sizeof(buts));
 581         if (ret)
 582                 return -EFAULT;
 583
 584         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 585         if (ret)
 586                 return ret;
 587
 588         if (copy_to_user(arg, &buts, sizeof(buts))) {
 589                 __blk_trace_remove(q);
 590                 return -EFAULT;
 591         }
 592         return 0;
 593 }
 594
 595 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 596                     struct block_device *bdev,
 597                     char __user *arg)
 598 {
 599         int ret;
 600
 601         mutex_lock(&q->blk_trace_mutex);
 602         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 603         mutex_unlock(&q->blk_trace_mutex);
 604
 605         return ret;
 606 }
 607 EXPORT_SYMBOL_GPL(blk_trace_setup);
 608
 609 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 610 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 611                                   dev_t dev, struct block_device *bdev,
 612                                   char __user *arg)
 613 {
 614         struct blk_user_trace_setup buts;
 615         struct compat_blk_user_trace_setup cbuts;
 616         int ret;
 617
 618         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 619                 return -EFAULT;
 620
 621         buts = (struct blk_user_trace_setup) {
 622                 .act_mask = cbuts.act_mask,
 623                 .buf_size = cbuts.buf_size,
 624                 .buf_nr = cbuts.buf_nr,
 625                 .start_lba = cbuts.start_lba,
 626                 .end_lba = cbuts.end_lba,
 627                 .pid = cbuts.pid,
 628         };
 629
 630         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 631         if (ret)
 632                 return ret;
 633
 634         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 635                 __blk_trace_remove(q);
 636                 return -EFAULT;
 637         }
 638
 639         return 0;
 640 }
 641 #endif
 642
 643 static int __blk_trace_startstop(struct request_queue *q, int start)
 644 {
 645         int ret;
 646         struct blk_trace *bt;
 647
 648         bt = rcu_dereference_protected(q->blk_trace,
 649                                        lockdep_is_held(&q->blk_trace_mutex));
 650         if (bt == NULL)
 651                 return -EINVAL;
 652
 653         /*
 654          * For starting a trace, we can transition from a setup or stopped
 655          * trace. For stopping a trace, the state must be running
 656          */
 657         ret = -EINVAL;
 658         if (start) {
 659                 if (bt->trace_state == Blktrace_setup ||
 660                     bt->trace_state == Blktrace_stopped) {
 661                         blktrace_seq++;
 662                         smp_mb();
 663                         bt->trace_state = Blktrace_running;
 664                         spin_lock_irq(&running_trace_lock);
 665                         list_add(&bt->running_list, &running_trace_list);
 666                         spin_unlock_irq(&running_trace_lock);
 667
 668                         trace_note_time(bt);
 669                         ret = 0;
 670                 }
 671         } else {
 672                 if (bt->trace_state == Blktrace_running) {
 673                         bt->trace_state = Blktrace_stopped;
 674                         spin_lock_irq(&running_trace_lock);
 675                         list_del_init(&bt->running_list);
 676                         spin_unlock_irq(&running_trace_lock);
 677                         relay_flush(bt->rchan);
 678                         ret = 0;
 679                 }
 680         }
 681
 682         return ret;
 683 }
 684
 685 int blk_trace_startstop(struct request_queue *q, int start)
 686 {
 687         int ret;
 688
 689         mutex_lock(&q->blk_trace_mutex);
 690         ret = __blk_trace_startstop(q, start);
 691         mutex_unlock(&q->blk_trace_mutex);
 692
 693         return ret;
 694 }
 695 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 696
 697 /*
 698  * When reading or writing the blktrace sysfs files, the references to the
 699  * opened sysfs or device files should prevent the underlying block device
 700  * from being removed. So no further delete protection is really needed.
 701  */
 702
 703 /**
 704  * blk_trace_ioctl: - handle the ioctls associated with tracing
 705  * @bdev:       the block device
 706  * @cmd:        the ioctl cmd
 707  * @arg:        the argument data, if any
 708  *
 709  **/
 710 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 711 {
 712         struct request_queue *q;
 713         int ret, start = 0;
 714         char b[BDEVNAME_SIZE];
 715
 716         q = bdev_get_queue(bdev);
 717         if (!q)
 718                 return -ENXIO;
 719
 720         mutex_lock(&q->blk_trace_mutex);
 721
 722         switch (cmd) {
 723         case BLKTRACESETUP:
 724                 bdevname(bdev, b);
 725                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 726                 break;
 727 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 728         case BLKTRACESETUP32:
 729                 bdevname(bdev, b);
 730                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 731                 break;
 732 #endif
 733         case BLKTRACESTART:
 734                 start = 1;
 735         case BLKTRACESTOP:
 736                 ret = __blk_trace_startstop(q, start);
 737                 break;
 738         case BLKTRACETEARDOWN:
 739                 ret = __blk_trace_remove(q);
 740                 break;
 741         default:
 742                 ret = -ENOTTY;
 743                 break;
 744         }
 745
 746         mutex_unlock(&q->blk_trace_mutex);
 747         return ret;
 748 }
 749
 750 /**
 751  * blk_trace_shutdown: - stop and cleanup trace structures
 752  * @q:    the request queue associated with the device
 753  *
 754  **/
 755 void blk_trace_shutdown(struct request_queue *q)
 756 {
 757         mutex_lock(&q->blk_trace_mutex);
 758         if (rcu_dereference_protected(q->blk_trace,
 759                                       lockdep_is_held(&q->blk_trace_mutex))) {
 760                 __blk_trace_startstop(q, 0);
 761                 __blk_trace_remove(q);
 762         }
 763
 764         mutex_unlock(&q->blk_trace_mutex);
 765 }
 766
 767 /*
 768  * blktrace probes
 769  */
 770
 771 /**
 772  * blk_add_trace_rq - Add a trace for a request oriented action
 773  * @q:          queue the io is for
 774  * @rq:         the source request
 775  * @nr_bytes:   number of completed bytes
 776  * @what:       the action
 777  *
 778  * Description:
 779  *     Records an action against a request. Will log the bio offset + size.
 780  *
 781  **/
 782 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
 783                              unsigned int nr_bytes, u32 what)
 784 {
 785         struct blk_trace *bt;
 786
 787         rcu_read_lock();
 788         bt = rcu_dereference(q->blk_trace);
 789         if (likely(!bt)) {
 790                 rcu_read_unlock();
 791                 return;
 792         }
 793
 794         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
 795                 what |= BLK_TC_ACT(BLK_TC_PC);
 796                 __blk_add_trace(bt, 0, nr_bytes, req_op(rq), rq->cmd_flags,
 797                                 what, rq->errors, rq->cmd_len, rq->cmd);
 798         } else  {
 799                 what |= BLK_TC_ACT(BLK_TC_FS);
 800                 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes, req_op(rq),
 801                                 rq->cmd_flags, what, rq->errors, 0, NULL);
 802         }
 803         rcu_read_unlock();
 804 }
 805
 806 static void blk_add_trace_rq_abort(void *ignore,
 807                                    struct request_queue *q, struct request *rq)
 808 {
 809         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
 810 }
 811
 812 static void blk_add_trace_rq_insert(void *ignore,
 813                                     struct request_queue *q, struct request *rq)
 814 {
 815         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
 816 }
 817
 818 static void blk_add_trace_rq_issue(void *ignore,
 819                                    struct request_queue *q, struct request *rq)
 820 {
 821         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
 822 }
 823
 824 static void blk_add_trace_rq_requeue(void *ignore,
 825                                      struct request_queue *q,
 826                                      struct request *rq)
 827 {
 828         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
 829 }
 830
 831 static void blk_add_trace_rq_complete(void *ignore,
 832                                       struct request_queue *q,
 833                                       struct request *rq,
 834                                       unsigned int nr_bytes)
 835 {
 836         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
 837 }
 838
 839 /**
 840  * blk_add_trace_bio - Add a trace for a bio oriented action
 841  * @q:          queue the io is for
 842  * @bio:        the source bio
 843  * @what:       the action
 844  * @error:      error, if any
 845  *
 846  * Description:
 847  *     Records an action against a bio. Will log the bio offset + size.
 848  *
 849  **/
 850 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 851                               u32 what, int error)
 852 {
 853         struct blk_trace *bt;
 854
 855         rcu_read_lock();
 856         bt = rcu_dereference(q->blk_trace);
 857         if (likely(!bt)) {
 858                 rcu_read_unlock();
 859                 return;
 860         }
 861
 862         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 863                         bio_op(bio), bio->bi_opf, what, error, 0, NULL);
 864         rcu_read_unlock();
 865 }
 866
 867 static void blk_add_trace_bio_bounce(void *ignore,
 868                                      struct request_queue *q, struct bio *bio)
 869 {
 870         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 871 }
 872
 873 static void blk_add_trace_bio_complete(void *ignore,
 874                                        struct request_queue *q, struct bio *bio,
 875                                        int error)
 876 {
 877         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 878 }
 879
 880 static void blk_add_trace_bio_backmerge(void *ignore,
 881                                         struct request_queue *q,
 882                                         struct request *rq,
 883                                         struct bio *bio)
 884 {
 885         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 886 }
 887
 888 static void blk_add_trace_bio_frontmerge(void *ignore,
 889                                          struct request_queue *q,
 890                                          struct request *rq,
 891                                          struct bio *bio)
 892 {
 893         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 894 }
 895
 896 static void blk_add_trace_bio_queue(void *ignore,
 897                                     struct request_queue *q, struct bio *bio)
 898 {
 899         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 900 }
 901
 902 static void blk_add_trace_getrq(void *ignore,
 903                                 struct request_queue *q,
 904                                 struct bio *bio, int rw)
 905 {
 906         if (bio)
 907                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 908         else {
 909                 struct blk_trace *bt;
 910
 911                 rcu_read_lock();
 912                 bt = rcu_dereference(q->blk_trace);
 913                 if (bt)
 914                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 915                                         NULL);
 916                 rcu_read_unlock();
 917         }
 918 }
 919
 920
 921 static void blk_add_trace_sleeprq(void *ignore,
 922                                   struct request_queue *q,
 923                                   struct bio *bio, int rw)
 924 {
 925         if (bio)
 926                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 927         else {
 928                 struct blk_trace *bt;
 929
 930                 rcu_read_lock();
 931                 bt = rcu_dereference(q->blk_trace);
 932                 if (bt)
 933                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 934                                         0, 0, NULL);
 935                 rcu_read_unlock();
 936         }
 937 }
 938
 939 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 940 {
 941         struct blk_trace *bt;
 942
 943         rcu_read_lock();
 944         bt = rcu_dereference(q->blk_trace);
 945         if (bt)
 946                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 947         rcu_read_unlock();
 948 }
 949
 950 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 951                                     unsigned int depth, bool explicit)
 952 {
 953         struct blk_trace *bt;
 954
 955         rcu_read_lock();
 956         bt = rcu_dereference(q->blk_trace);
 957         if (bt) {
 958                 __be64 rpdu = cpu_to_be64(depth);
 959                 u32 what;
 960
 961                 if (explicit)
 962                         what = BLK_TA_UNPLUG_IO;
 963                 else
 964                         what = BLK_TA_UNPLUG_TIMER;
 965
 966                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 967         }
 968         rcu_read_unlock();
 969 }
 970
 971 static void blk_add_trace_split(void *ignore,
 972                                 struct request_queue *q, struct bio *bio,
 973                                 unsigned int pdu)
 974 {
 975         struct blk_trace *bt;
 976
 977         rcu_read_lock();
 978         bt = rcu_dereference(q->blk_trace);
 979         if (bt) {
 980                 __be64 rpdu = cpu_to_be64(pdu);
 981
 982                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 983                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 984                                 BLK_TA_SPLIT, bio->bi_error, sizeof(rpdu),
 985                                 &rpdu);
 986         }
 987         rcu_read_unlock();
 988 }
 989
 990 /**
 991  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 992  * @ignore:     trace callback data parameter (not used)
 993  * @q:          queue the io is for
 994  * @bio:        the source bio
 995  * @dev:        target device
 996  * @from:       source sector
 997  *
 998  * Description:
 999  *     Device mapper or raid target sometimes need to split a bio because
1000  *     it spans a stripe (or similar). Add a trace for that action.
1001  *
1002  **/
1003 static void blk_add_trace_bio_remap(void *ignore,
1004                                     struct request_queue *q, struct bio *bio,
1005                                     dev_t dev, sector_t from)
1006 {
1007         struct blk_trace *bt;
1008         struct blk_io_trace_remap r;
1009
1010         rcu_read_lock();
1011         bt = rcu_dereference(q->blk_trace);
1012         if (likely(!bt)) {
1013                 rcu_read_unlock();
1014                 return;
1015         }
1016
1017         r.device_from = cpu_to_be32(dev);
1018         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
1019         r.sector_from = cpu_to_be64(from);
1020
1021         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1022                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_error,
1023                         sizeof(r), &r);
1024         rcu_read_unlock();
1025 }
1026
1027 /**
1028  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1029  * @ignore:     trace callback data parameter (not used)
1030  * @q:          queue the io is for
1031  * @rq:         the source request
1032  * @dev:        target device
1033  * @from:       source sector
1034  *
1035  * Description:
1036  *     Device mapper remaps request to other devices.
1037  *     Add a trace for that action.
1038  *
1039  **/
1040 static void blk_add_trace_rq_remap(void *ignore,
1041                                    struct request_queue *q,
1042                                    struct request *rq, dev_t dev,
1043                                    sector_t from)
1044 {
1045         struct blk_trace *bt;
1046         struct blk_io_trace_remap r;
1047
1048         rcu_read_lock();
1049         bt = rcu_dereference(q->blk_trace);
1050         if (likely(!bt)) {
1051                 rcu_read_unlock();
1052                 return;
1053         }
1054
1055         r.device_from = cpu_to_be32(dev);
1056         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1057         r.sector_from = cpu_to_be64(from);
1058
1059         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1060                         rq_data_dir(rq), 0, BLK_TA_REMAP, !!rq->errors,
1061                         sizeof(r), &r);
1062         rcu_read_unlock();
1063 }
1064
1065 /**
1066  * blk_add_driver_data - Add binary message with driver-specific data
1067  * @q:          queue the io is for
1068  * @rq:         io request
1069  * @data:       driver-specific data
1070  * @len:        length of driver-specific data
1071  *
1072  * Description:
1073  *     Some drivers might want to write driver-specific data per request.
1074  *
1075  **/
1076 void blk_add_driver_data(struct request_queue *q,
1077                          struct request *rq,
1078                          void *data, size_t len)
1079 {
1080         struct blk_trace *bt;
1081
1082         rcu_read_lock();
1083         bt = rcu_dereference(q->blk_trace);
1084         if (likely(!bt)) {
1085                 rcu_read_unlock();
1086                 return;
1087         }
1088
1089         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1090                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0, 0,
1091                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1092         else
1093                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0, 0,
1094                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1095         rcu_read_unlock();
1096 }
1097 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1098
1099 static void blk_register_tracepoints(void)
1100 {
1101         int ret;
1102
1103         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1104         WARN_ON(ret);
1105         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1106         WARN_ON(ret);
1107         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1108         WARN_ON(ret);
1109         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1110         WARN_ON(ret);
1111         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1112         WARN_ON(ret);
1113         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1114         WARN_ON(ret);
1115         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1116         WARN_ON(ret);
1117         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1118         WARN_ON(ret);
1119         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1120         WARN_ON(ret);
1121         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1122         WARN_ON(ret);
1123         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1124         WARN_ON(ret);
1125         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1126         WARN_ON(ret);
1127         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1128         WARN_ON(ret);
1129         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1130         WARN_ON(ret);
1131         ret = register_trace_block_split(blk_add_trace_split, NULL);
1132         WARN_ON(ret);
1133         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1134         WARN_ON(ret);
1135         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1136         WARN_ON(ret);
1137 }
1138
1139 static void blk_unregister_tracepoints(void)
1140 {
1141         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1142         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1143         unregister_trace_block_split(blk_add_trace_split, NULL);
1144         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1145         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1146         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1147         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1148         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1149         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1150         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1151         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1152         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1153         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1154         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1155         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1156         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1157         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1158
1159         tracepoint_synchronize_unregister();
1160 }
1161
1162 /*
1163  * struct blk_io_tracer formatting routines
1164  */
1165
1166 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1167 {
1168         int i = 0;
1169         int tc = t->action >> BLK_TC_SHIFT;
1170
1171         if (t->action == BLK_TN_MESSAGE) {
1172                 rwbs[i++] = 'N';
1173                 goto out;
1174         }
1175
1176         if (tc & BLK_TC_FLUSH)
1177                 rwbs[i++] = 'F';
1178
1179         if (tc & BLK_TC_DISCARD)
1180                 rwbs[i++] = 'D';
1181         else if (tc & BLK_TC_WRITE)
1182                 rwbs[i++] = 'W';
1183         else if (t->bytes)
1184                 rwbs[i++] = 'R';
1185         else
1186                 rwbs[i++] = 'N';
1187
1188         if (tc & BLK_TC_FUA)
1189                 rwbs[i++] = 'F';
1190         if (tc & BLK_TC_AHEAD)
1191                 rwbs[i++] = 'A';
1192         if (tc & BLK_TC_SYNC)
1193                 rwbs[i++] = 'S';
1194         if (tc & BLK_TC_META)
1195                 rwbs[i++] = 'M';
1196 out:
1197         rwbs[i] = '\0';
1198 }
1199
1200 static inline
1201 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1202 {
1203         return (const struct blk_io_trace *)ent;
1204 }
1205
1206 static inline const void *pdu_start(const struct trace_entry *ent)
1207 {
1208         return te_blk_io_trace(ent) + 1;
1209 }
1210
1211 static inline u32 t_action(const struct trace_entry *ent)
1212 {
1213         return te_blk_io_trace(ent)->action;
1214 }
1215
1216 static inline u32 t_bytes(const struct trace_entry *ent)
1217 {
1218         return te_blk_io_trace(ent)->bytes;
1219 }
1220
1221 static inline u32 t_sec(const struct trace_entry *ent)
1222 {
1223         return te_blk_io_trace(ent)->bytes >> 9;
1224 }
1225
1226 static inline unsigned long long t_sector(const struct trace_entry *ent)
1227 {
1228         return te_blk_io_trace(ent)->sector;
1229 }
1230
1231 static inline __u16 t_error(const struct trace_entry *ent)
1232 {
1233         return te_blk_io_trace(ent)->error;
1234 }
1235
1236 static __u64 get_pdu_int(const struct trace_entry *ent)
1237 {
1238         const __u64 *val = pdu_start(ent);
1239         return be64_to_cpu(*val);
1240 }
1241
1242 static void get_pdu_remap(const struct trace_entry *ent,
1243                           struct blk_io_trace_remap *r)
1244 {
1245         const struct blk_io_trace_remap *__r = pdu_start(ent);
1246         __u64 sector_from = __r->sector_from;
1247
1248         r->device_from = be32_to_cpu(__r->device_from);
1249         r->device_to   = be32_to_cpu(__r->device_to);
1250         r->sector_from = be64_to_cpu(sector_from);
1251 }
1252
1253 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1254
1255 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1256 {
1257         char rwbs[RWBS_LEN];
1258         unsigned long long ts  = iter->ts;
1259         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1260         unsigned secs          = (unsigned long)ts;
1261         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1262
1263         fill_rwbs(rwbs, t);
1264
1265         trace_seq_printf(&iter->seq,
1266                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1267                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1268                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1269 }
1270
1271 static void blk_log_action(struct trace_iterator *iter, const char *act)
1272 {
1273         char rwbs[RWBS_LEN];
1274         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1275
1276         fill_rwbs(rwbs, t);
1277         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1278                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1279 }
1280
1281 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1282 {
1283         const unsigned char *pdu_buf;
1284         int pdu_len;
1285         int i, end;
1286
1287         pdu_buf = pdu_start(ent);
1288         pdu_len = te_blk_io_trace(ent)->pdu_len;
1289
1290         if (!pdu_len)
1291                 return;
1292
1293         /* find the last zero that needs to be printed */
1294         for (end = pdu_len - 1; end >= 0; end--)
1295                 if (pdu_buf[end])
1296                         break;
1297         end++;
1298
1299         trace_seq_putc(s, '(');
1300
1301         for (i = 0; i < pdu_len; i++) {
1302
1303                 trace_seq_printf(s, "%s%02x",
1304                                  i == 0 ? "" : " ", pdu_buf[i]);
1305
1306                 /*
1307                  * stop when the rest is just zeroes and indicate so
1308                  * with a ".." appended
1309                  */
1310                 if (i == end && end != pdu_len - 1) {
1311                         trace_seq_puts(s, " ..) ");
1312                         return;
1313                 }
1314         }
1315
1316         trace_seq_puts(s, ") ");
1317 }
1318
1319 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1320 {
1321         char cmd[TASK_COMM_LEN];
1322
1323         trace_find_cmdline(ent->pid, cmd);
1324
1325         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1326                 trace_seq_printf(s, "%u ", t_bytes(ent));
1327                 blk_log_dump_pdu(s, ent);
1328                 trace_seq_printf(s, "[%s]\n", cmd);
1329         } else {
1330                 if (t_sec(ent))
1331                         trace_seq_printf(s, "%llu + %u [%s]\n",
1332                                                 t_sector(ent), t_sec(ent), cmd);
1333                 else
1334                         trace_seq_printf(s, "[%s]\n", cmd);
1335         }
1336 }
1337
1338 static void blk_log_with_error(struct trace_seq *s,
1339                               const struct trace_entry *ent)
1340 {
1341         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1342                 blk_log_dump_pdu(s, ent);
1343                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1344         } else {
1345                 if (t_sec(ent))
1346                         trace_seq_printf(s, "%llu + %u [%d]\n",
1347                                          t_sector(ent),
1348                                          t_sec(ent), t_error(ent));
1349                 else
1350                         trace_seq_printf(s, "%llu [%d]\n",
1351                                          t_sector(ent), t_error(ent));
1352         }
1353 }
1354
1355 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1356 {
1357         struct blk_io_trace_remap r = { .device_from = 0, };
1358
1359         get_pdu_remap(ent, &r);
1360         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1361                          t_sector(ent), t_sec(ent),
1362                          MAJOR(r.device_from), MINOR(r.device_from),
1363                          (unsigned long long)r.sector_from);
1364 }
1365
1366 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1367 {
1368         char cmd[TASK_COMM_LEN];
1369
1370         trace_find_cmdline(ent->pid, cmd);
1371
1372         trace_seq_printf(s, "[%s]\n", cmd);
1373 }
1374
1375 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1376 {
1377         char cmd[TASK_COMM_LEN];
1378
1379         trace_find_cmdline(ent->pid, cmd);
1380
1381         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1382 }
1383
1384 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1385 {
1386         char cmd[TASK_COMM_LEN];
1387
1388         trace_find_cmdline(ent->pid, cmd);
1389
1390         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1391                          get_pdu_int(ent), cmd);
1392 }
1393
1394 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1395 {
1396         const struct blk_io_trace *t = te_blk_io_trace(ent);
1397
1398         trace_seq_putmem(s, t + 1, t->pdu_len);
1399         trace_seq_putc(s, '\n');
1400 }
1401
1402 /*
1403  * struct tracer operations
1404  */
1405
1406 static void blk_tracer_print_header(struct seq_file *m)
1407 {
1408         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1409                 return;
1410         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1411                     "#  |     |     |           |   |   |\n");
1412 }
1413
1414 static void blk_tracer_start(struct trace_array *tr)
1415 {
1416         blk_tracer_enabled = true;
1417 }
1418
1419 static int blk_tracer_init(struct trace_array *tr)
1420 {
1421         blk_tr = tr;
1422         blk_tracer_start(tr);
1423         return 0;
1424 }
1425
1426 static void blk_tracer_stop(struct trace_array *tr)
1427 {
1428         blk_tracer_enabled = false;
1429 }
1430
1431 static void blk_tracer_reset(struct trace_array *tr)
1432 {
1433         blk_tracer_stop(tr);
1434 }
1435
1436 static const struct {
1437         const char *act[2];
1438         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1439 } what2act[] = {
1440         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1441         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1442         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1443         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1444         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1445         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1446         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1447         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1448         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1449         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1450         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1451         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1452         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1453         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1454         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1455 };
1456
1457 static enum print_line_t print_one_line(struct trace_iterator *iter,
1458                                         bool classic)
1459 {
1460         struct trace_array *tr = iter->tr;
1461         struct trace_seq *s = &iter->seq;
1462         const struct blk_io_trace *t;
1463         u16 what;
1464         bool long_act;
1465         blk_log_action_t *log_action;
1466
1467         t          = te_blk_io_trace(iter->ent);
1468         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1469         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1470         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1471
1472         if (t->action == BLK_TN_MESSAGE) {
1473                 log_action(iter, long_act ? "message" : "m");
1474                 blk_log_msg(s, iter->ent);
1475                 return trace_handle_return(s);
1476         }
1477
1478         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1479                 trace_seq_printf(s, "Unknown action %x\n", what);
1480         else {
1481                 log_action(iter, what2act[what].act[long_act]);
1482                 what2act[what].print(s, iter->ent);
1483         }
1484
1485         return trace_handle_return(s);
1486 }
1487
1488 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1489                                                int flags, struct trace_event *event)
1490 {
1491         return print_one_line(iter, false);
1492 }
1493
1494 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1495 {
1496         struct trace_seq *s = &iter->seq;
1497         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1498         const int offset = offsetof(struct blk_io_trace, sector);
1499         struct blk_io_trace old = {
1500                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1501                 .time     = iter->ts,
1502         };
1503
1504         trace_seq_putmem(s, &old, offset);
1505         trace_seq_putmem(s, &t->sector,
1506                          sizeof(old) - offset + t->pdu_len);
1507 }
1508
1509 static enum print_line_t
1510 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1511                              struct trace_event *event)
1512 {
1513         blk_trace_synthesize_old_trace(iter);
1514
1515         return trace_handle_return(&iter->seq);
1516 }
1517
1518 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1519 {
1520         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1521                 return TRACE_TYPE_UNHANDLED;
1522
1523         return print_one_line(iter, true);
1524 }
1525
1526 static int
1527 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1528 {
1529         /* don't output context-info for blk_classic output */
1530         if (bit == TRACE_BLK_OPT_CLASSIC) {
1531                 if (set)
1532                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1533                 else
1534                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1535         }
1536         return 0;
1537 }
1538
1539 static struct tracer blk_tracer __read_mostly = {
1540         .name           = "blk",
1541         .init           = blk_tracer_init,
1542         .reset          = blk_tracer_reset,
1543         .start          = blk_tracer_start,
1544         .stop           = blk_tracer_stop,
1545         .print_header   = blk_tracer_print_header,
1546         .print_line     = blk_tracer_print_line,
1547         .flags          = &blk_tracer_flags,
1548         .set_flag       = blk_tracer_set_flag,
1549 };
1550
1551 static struct trace_event_functions trace_blk_event_funcs = {
1552         .trace          = blk_trace_event_print,
1553         .binary         = blk_trace_event_print_binary,
1554 };
1555
1556 static struct trace_event trace_blk_event = {
1557         .type           = TRACE_BLK,
1558         .funcs          = &trace_blk_event_funcs,
1559 };
1560
1561 static int __init init_blk_tracer(void)
1562 {
1563         if (!register_trace_event(&trace_blk_event)) {
1564                 pr_warn("Warning: could not register block events\n");
1565                 return 1;
1566         }
1567
1568         if (register_tracer(&blk_tracer) != 0) {
1569                 pr_warn("Warning: could not register the block tracer\n");
1570                 unregister_trace_event(&trace_blk_event);
1571                 return 1;
1572         }
1573
1574         return 0;
1575 }
1576
1577 device_initcall(init_blk_tracer);
1578
1579 static int blk_trace_remove_queue(struct request_queue *q)
1580 {
1581         struct blk_trace *bt;
1582
1583         bt = xchg(&q->blk_trace, NULL);
1584         if (bt == NULL)
1585                 return -EINVAL;
1586
1587         if (bt->trace_state == Blktrace_running) {
1588                 bt->trace_state = Blktrace_stopped;
1589                 spin_lock_irq(&running_trace_lock);
1590                 list_del_init(&bt->running_list);
1591                 spin_unlock_irq(&running_trace_lock);
1592                 relay_flush(bt->rchan);
1593         }
1594
1595         put_probe_ref();
1596         synchronize_rcu();
1597         blk_trace_free(bt);
1598         return 0;
1599 }
1600
1601 /*
1602  * Setup everything required to start tracing
1603  */
1604 static int blk_trace_setup_queue(struct request_queue *q,
1605                                  struct block_device *bdev)
1606 {
1607         struct blk_trace *bt = NULL;
1608         int ret = -ENOMEM;
1609
1610         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1611         if (!bt)
1612                 return -ENOMEM;
1613
1614         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1615         if (!bt->msg_data)
1616                 goto free_bt;
1617
1618         bt->dev = bdev->bd_dev;
1619         bt->act_mask = (u16)-1;
1620
1621         blk_trace_setup_lba(bt, bdev);
1622
1623         ret = -EBUSY;
1624         if (cmpxchg(&q->blk_trace, NULL, bt))
1625                 goto free_bt;
1626
1627         get_probe_ref();
1628         return 0;
1629
1630 free_bt:
1631         blk_trace_free(bt);
1632         return ret;
1633 }
1634
1635 /*
1636  * sysfs interface to enable and configure tracing
1637  */
1638
1639 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1640                                          struct device_attribute *attr,
1641                                          char *buf);
1642 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1643                                           struct device_attribute *attr,
1644                                           const char *buf, size_t count);
1645 #define BLK_TRACE_DEVICE_ATTR(_name) \
1646         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1647                     sysfs_blk_trace_attr_show, \
1648                     sysfs_blk_trace_attr_store)
1649
1650 static BLK_TRACE_DEVICE_ATTR(enable);
1651 static BLK_TRACE_DEVICE_ATTR(act_mask);
1652 static BLK_TRACE_DEVICE_ATTR(pid);
1653 static BLK_TRACE_DEVICE_ATTR(start_lba);
1654 static BLK_TRACE_DEVICE_ATTR(end_lba);
1655
1656 static struct attribute *blk_trace_attrs[] = {
1657         &dev_attr_enable.attr,
1658         &dev_attr_act_mask.attr,
1659         &dev_attr_pid.attr,
1660         &dev_attr_start_lba.attr,
1661         &dev_attr_end_lba.attr,
1662         NULL
1663 };
1664
1665 struct attribute_group blk_trace_attr_group = {
1666         .name  = "trace",
1667         .attrs = blk_trace_attrs,
1668 };
1669
1670 static const struct {
1671         int mask;
1672         const char *str;
1673 } mask_maps[] = {
1674         { BLK_TC_READ,          "read"          },
1675         { BLK_TC_WRITE,         "write"         },
1676         { BLK_TC_FLUSH,         "flush"         },
1677         { BLK_TC_SYNC,          "sync"          },
1678         { BLK_TC_QUEUE,         "queue"         },
1679         { BLK_TC_REQUEUE,       "requeue"       },
1680         { BLK_TC_ISSUE,         "issue"         },
1681         { BLK_TC_COMPLETE,      "complete"      },
1682         { BLK_TC_FS,            "fs"            },
1683         { BLK_TC_PC,            "pc"            },
1684         { BLK_TC_NOTIFY,        "notify"        },
1685         { BLK_TC_AHEAD,         "ahead"         },
1686         { BLK_TC_META,          "meta"          },
1687         { BLK_TC_DISCARD,       "discard"       },
1688         { BLK_TC_DRV_DATA,      "drv_data"      },
1689         { BLK_TC_FUA,           "fua"           },
1690 };
1691
1692 static int blk_trace_str2mask(const char *str)
1693 {
1694         int i;
1695         int mask = 0;
1696         char *buf, *s, *token;
1697
1698         buf = kstrdup(str, GFP_KERNEL);
1699         if (buf == NULL)
1700                 return -ENOMEM;
1701         s = strstrip(buf);
1702
1703         while (1) {
1704                 token = strsep(&s, ",");
1705                 if (token == NULL)
1706                         break;
1707
1708                 if (*token == '\0')
1709                         continue;
1710
1711                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1712                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1713                                 mask |= mask_maps[i].mask;
1714                                 break;
1715                         }
1716                 }
1717                 if (i == ARRAY_SIZE(mask_maps)) {
1718                         mask = -EINVAL;
1719                         break;
1720                 }
1721         }
1722         kfree(buf);
1723
1724         return mask;
1725 }
1726
1727 static ssize_t blk_trace_mask2str(char *buf, int mask)
1728 {
1729         int i;
1730         char *p = buf;
1731
1732         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1733                 if (mask & mask_maps[i].mask) {
1734                         p += sprintf(p, "%s%s",
1735                                     (p == buf) ? "" : ",", mask_maps[i].str);
1736                 }
1737         }
1738         *p++ = '\n';
1739
1740         return p - buf;
1741 }
1742
1743 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1744 {
1745         if (bdev->bd_disk == NULL)
1746                 return NULL;
1747
1748         return bdev_get_queue(bdev);
1749 }
1750
1751 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1752                                          struct device_attribute *attr,
1753                                          char *buf)
1754 {
1755         struct hd_struct *p = dev_to_part(dev);
1756         struct request_queue *q;
1757         struct block_device *bdev;
1758         struct blk_trace *bt;
1759         ssize_t ret = -ENXIO;
1760
1761         bdev = bdget(part_devt(p));
1762         if (bdev == NULL)
1763                 goto out;
1764
1765         q = blk_trace_get_queue(bdev);
1766         if (q == NULL)
1767                 goto out_bdput;
1768
1769         mutex_lock(&q->blk_trace_mutex);
1770
1771         bt = rcu_dereference_protected(q->blk_trace,
1772                                        lockdep_is_held(&q->blk_trace_mutex));
1773         if (attr == &dev_attr_enable) {
1774                 ret = sprintf(buf, "%u\n", !!bt);
1775                 goto out_unlock_bdev;
1776         }
1777
1778         if (bt == NULL)
1779                 ret = sprintf(buf, "disabled\n");
1780         else if (attr == &dev_attr_act_mask)
1781                 ret = blk_trace_mask2str(buf, bt->act_mask);
1782         else if (attr == &dev_attr_pid)
1783                 ret = sprintf(buf, "%u\n", bt->pid);
1784         else if (attr == &dev_attr_start_lba)
1785                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1786         else if (attr == &dev_attr_end_lba)
1787                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1788
1789 out_unlock_bdev:
1790         mutex_unlock(&q->blk_trace_mutex);
1791 out_bdput:
1792         bdput(bdev);
1793 out:
1794         return ret;
1795 }
1796
1797 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1798                                           struct device_attribute *attr,
1799                                           const char *buf, size_t count)
1800 {
1801         struct block_device *bdev;
1802         struct request_queue *q;
1803         struct hd_struct *p;
1804         struct blk_trace *bt;
1805         u64 value;
1806         ssize_t ret = -EINVAL;
1807
1808         if (count == 0)
1809                 goto out;
1810
1811         if (attr == &dev_attr_act_mask) {
1812                 if (sscanf(buf, "%llx", &value) != 1) {
1813                         /* Assume it is a list of trace category names */
1814                         ret = blk_trace_str2mask(buf);
1815                         if (ret < 0)
1816                                 goto out;
1817                         value = ret;
1818                 }
1819         } else if (sscanf(buf, "%llu", &value) != 1)
1820                 goto out;
1821
1822         ret = -ENXIO;
1823
1824         p = dev_to_part(dev);
1825         bdev = bdget(part_devt(p));
1826         if (bdev == NULL)
1827                 goto out;
1828
1829         q = blk_trace_get_queue(bdev);
1830         if (q == NULL)
1831                 goto out_bdput;
1832
1833         mutex_lock(&q->blk_trace_mutex);
1834
1835         bt = rcu_dereference_protected(q->blk_trace,
1836                                        lockdep_is_held(&q->blk_trace_mutex));
1837         if (attr == &dev_attr_enable) {
1838                 if (!!value == !!bt) {
1839                         ret = 0;
1840                         goto out_unlock_bdev;
1841                 }
1842                 if (value)
1843                         ret = blk_trace_setup_queue(q, bdev);
1844                 else
1845                         ret = blk_trace_remove_queue(q);
1846                 goto out_unlock_bdev;
1847         }
1848
1849         ret = 0;
1850         if (bt == NULL) {
1851                 ret = blk_trace_setup_queue(q, bdev);
1852                 bt = rcu_dereference_protected(q->blk_trace,
1853                                 lockdep_is_held(&q->blk_trace_mutex));
1854         }
1855
1856         if (ret == 0) {
1857                 if (attr == &dev_attr_act_mask)
1858                         bt->act_mask = value;
1859                 else if (attr == &dev_attr_pid)
1860                         bt->pid = value;
1861                 else if (attr == &dev_attr_start_lba)
1862                         bt->start_lba = value;
1863                 else if (attr == &dev_attr_end_lba)
1864                         bt->end_lba = value;
1865         }
1866
1867 out_unlock_bdev:
1868         mutex_unlock(&q->blk_trace_mutex);
1869 out_bdput:
1870         bdput(bdev);
1871 out:
1872         return ret ? ret : count;
1873 }
1874
1875 int blk_trace_init_sysfs(struct device *dev)
1876 {
1877         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1878 }
1879
1880 void blk_trace_remove_sysfs(struct device *dev)
1881 {
1882         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1883 }
1884
1885 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1886
1887 #ifdef CONFIG_EVENT_TRACING
1888
1889 void blk_dump_cmd(char *buf, struct request *rq)
1890 {
1891         int i, end;
1892         int len = rq->cmd_len;
1893         unsigned char *cmd = rq->cmd;
1894
1895         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1896                 buf[0] = '\0';
1897                 return;
1898         }
1899
1900         for (end = len - 1; end >= 0; end--)
1901                 if (cmd[end])
1902                         break;
1903         end++;
1904
1905         for (i = 0; i < len; i++) {
1906                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1907                 if (i == end && end != len - 1) {
1908                         sprintf(buf, " ..");
1909                         break;
1910                 }
1911         }
1912 }
1913
1914 void blk_fill_rwbs(char *rwbs, int op, u32 rw, int bytes)
1915 {
1916         int i = 0;
1917
1918         if (rw & REQ_PREFLUSH)
1919                 rwbs[i++] = 'F';
1920
1921         switch (op) {
1922         case REQ_OP_WRITE:
1923         case REQ_OP_WRITE_SAME:
1924                 rwbs[i++] = 'W';
1925                 break;
1926         case REQ_OP_DISCARD:
1927                 rwbs[i++] = 'D';
1928                 break;
1929         case REQ_OP_SECURE_ERASE:
1930                 rwbs[i++] = 'D';
1931                 rwbs[i++] = 'E';
1932                 break;
1933         case REQ_OP_FLUSH:
1934                 rwbs[i++] = 'F';
1935                 break;
1936         case REQ_OP_READ:
1937                 rwbs[i++] = 'R';
1938                 break;
1939         default:
1940                 rwbs[i++] = 'N';
1941         }
1942
1943         if (rw & REQ_FUA)
1944                 rwbs[i++] = 'F';
1945         if (rw & REQ_RAHEAD)
1946                 rwbs[i++] = 'A';
1947         if (rw & REQ_SYNC)
1948                 rwbs[i++] = 'S';
1949         if (rw & REQ_META)
1950                 rwbs[i++] = 'M';
1951
1952         rwbs[i] = '\0';
1953 }
1954 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1955
1956 #endif /* CONFIG_EVENT_TRACING */
1957