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 timespec now;
 135         unsigned long flags;
 136         u32 words[2];
 137
 138         getnstimeofday(&now);
 139         words[0] = now.tv_sec;
 140         words[1] = now.tv_nsec;
 141
 142         local_irq_save(flags);
 143         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
 144         local_irq_restore(flags);
 145 }
 146
 147 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
 148 {
 149         int n;
 150         va_list args;
 151         unsigned long flags;
 152         char *buf;
 153
 154         if (unlikely(bt->trace_state != Blktrace_running &&
 155                      !blk_tracer_enabled))
 156                 return;
 157
 158         /*
 159          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 160          * message to the trace.
 161          */
 162         if (!(bt->act_mask & BLK_TC_NOTIFY))
 163                 return;
 164
 165         local_irq_save(flags);
 166         buf = this_cpu_ptr(bt->msg_data);
 167         va_start(args, fmt);
 168         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 169         va_end(args);
 170
 171         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
 172         local_irq_restore(flags);
 173 }
 174 EXPORT_SYMBOL_GPL(__trace_note_message);
 175
 176 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 177                          pid_t pid)
 178 {
 179         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 180                 return 1;
 181         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 182                 return 1;
 183         if (bt->pid && pid != bt->pid)
 184                 return 1;
 185
 186         return 0;
 187 }
 188
 189 /*
 190  * Data direction bit lookup
 191  */
 192 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 193                                  BLK_TC_ACT(BLK_TC_WRITE) };
 194
 195 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 196
 197 /* The ilog2() calls fall out because they're constant */
 198 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 199           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 200
 201 /*
 202  * The worker for the various blk_add_trace*() types. Fills out a
 203  * blk_io_trace structure and places it in a per-cpu subbuffer.
 204  */
 205 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 206                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
 207 {
 208         struct task_struct *tsk = current;
 209         struct ring_buffer_event *event = NULL;
 210         struct ring_buffer *buffer = NULL;
 211         struct blk_io_trace *t;
 212         unsigned long flags = 0;
 213         unsigned long *sequence;
 214         pid_t pid;
 215         int cpu, pc = 0;
 216         bool blk_tracer = blk_tracer_enabled;
 217
 218         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 219                 return;
 220
 221         what |= ddir_act[rw & WRITE];
 222         what |= MASK_TC_BIT(rw, SYNC);
 223         what |= MASK_TC_BIT(rw, RAHEAD);
 224         what |= MASK_TC_BIT(rw, META);
 225         what |= MASK_TC_BIT(rw, DISCARD);
 226         what |= MASK_TC_BIT(rw, FLUSH);
 227         what |= MASK_TC_BIT(rw, FUA);
 228
 229         pid = tsk->pid;
 230         if (act_log_check(bt, what, sector, pid))
 231                 return;
 232         cpu = raw_smp_processor_id();
 233
 234         if (blk_tracer) {
 235                 tracing_record_cmdline(current);
 236
 237                 buffer = blk_tr->trace_buffer.buffer;
 238                 pc = preempt_count();
 239                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 240                                                   sizeof(*t) + pdu_len,
 241                                                   0, pc);
 242                 if (!event)
 243                         return;
 244                 t = ring_buffer_event_data(event);
 245                 goto record_it;
 246         }
 247
 248         if (unlikely(tsk->btrace_seq != blktrace_seq))
 249                 trace_note_tsk(tsk);
 250
 251         /*
 252          * A word about the locking here - we disable interrupts to reserve
 253          * some space in the relay per-cpu buffer, to prevent an irq
 254          * from coming in and stepping on our toes.
 255          */
 256         local_irq_save(flags);
 257         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
 258         if (t) {
 259                 sequence = per_cpu_ptr(bt->sequence, cpu);
 260
 261                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 262                 t->sequence = ++(*sequence);
 263                 t->time = ktime_to_ns(ktime_get());
 264 record_it:
 265                 /*
 266                  * These two are not needed in ftrace as they are in the
 267                  * generic trace_entry, filled by tracing_generic_entry_update,
 268                  * but for the trace_event->bin() synthesizer benefit we do it
 269                  * here too.
 270                  */
 271                 t->cpu = cpu;
 272                 t->pid = pid;
 273
 274                 t->sector = sector;
 275                 t->bytes = bytes;
 276                 t->action = what;
 277                 t->device = bt->dev;
 278                 t->error = error;
 279                 t->pdu_len = pdu_len;
 280
 281                 if (pdu_len)
 282                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
 283
 284                 if (blk_tracer) {
 285                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 286                         return;
 287                 }
 288         }
 289
 290         local_irq_restore(flags);
 291 }
 292
 293 static struct dentry *blk_tree_root;
 294 static DEFINE_MUTEX(blk_tree_mutex);
 295
 296 static void blk_trace_free(struct blk_trace *bt)
 297 {
 298         debugfs_remove(bt->msg_file);
 299         debugfs_remove(bt->dropped_file);
 300         relay_close(bt->rchan);
 301         debugfs_remove(bt->dir);
 302         free_percpu(bt->sequence);
 303         free_percpu(bt->msg_data);
 304         kfree(bt);
 305 }
 306
 307 static void get_probe_ref(void)
 308 {
 309         mutex_lock(&blk_probe_mutex);
 310         if (++blk_probes_ref == 1)
 311                 blk_register_tracepoints();
 312         mutex_unlock(&blk_probe_mutex);
 313 }
 314
 315 static void put_probe_ref(void)
 316 {
 317         mutex_lock(&blk_probe_mutex);
 318         if (!--blk_probes_ref)
 319                 blk_unregister_tracepoints();
 320         mutex_unlock(&blk_probe_mutex);
 321 }
 322
 323 static void blk_trace_cleanup(struct blk_trace *bt)
 324 {
 325         synchronize_rcu();
 326         blk_trace_free(bt);
 327         put_probe_ref();
 328 }
 329
 330 static int __blk_trace_remove(struct request_queue *q)
 331 {
 332         struct blk_trace *bt;
 333
 334         bt = xchg(&q->blk_trace, NULL);
 335         if (!bt)
 336                 return -EINVAL;
 337
 338         if (bt->trace_state != Blktrace_running)
 339                 blk_trace_cleanup(bt);
 340
 341         return 0;
 342 }
 343
 344 int blk_trace_remove(struct request_queue *q)
 345 {
 346         int ret;
 347
 348         mutex_lock(&q->blk_trace_mutex);
 349         ret = __blk_trace_remove(q);
 350         mutex_unlock(&q->blk_trace_mutex);
 351
 352         return ret;
 353 }
 354 EXPORT_SYMBOL_GPL(blk_trace_remove);
 355
 356 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 357                                 size_t count, loff_t *ppos)
 358 {
 359         struct blk_trace *bt = filp->private_data;
 360         char buf[16];
 361
 362         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 363
 364         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 365 }
 366
 367 static const struct file_operations blk_dropped_fops = {
 368         .owner =        THIS_MODULE,
 369         .open =         simple_open,
 370         .read =         blk_dropped_read,
 371         .llseek =       default_llseek,
 372 };
 373
 374 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 375                                 size_t count, loff_t *ppos)
 376 {
 377         char *msg;
 378         struct blk_trace *bt;
 379
 380         if (count >= BLK_TN_MAX_MSG)
 381                 return -EINVAL;
 382
 383         msg = kmalloc(count + 1, GFP_KERNEL);
 384         if (msg == NULL)
 385                 return -ENOMEM;
 386
 387         if (copy_from_user(msg, buffer, count)) {
 388                 kfree(msg);
 389                 return -EFAULT;
 390         }
 391
 392         msg[count] = '\0';
 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, 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,
 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->bi_rw, 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, BLK_TA_GETRQ, 0, 0, NULL);
 915                 rcu_read_unlock();
 916         }
 917 }
 918
 919
 920 static void blk_add_trace_sleeprq(void *ignore,
 921                                   struct request_queue *q,
 922                                   struct bio *bio, int rw)
 923 {
 924         if (bio)
 925                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 926         else {
 927                 struct blk_trace *bt;
 928
 929                 rcu_read_lock();
 930                 bt = rcu_dereference(q->blk_trace);
 931                 if (bt)
 932                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
 933                                         0, 0, NULL);
 934                 rcu_read_unlock();
 935         }
 936 }
 937
 938 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 939 {
 940         struct blk_trace *bt;
 941
 942         rcu_read_lock();
 943         bt = rcu_dereference(q->blk_trace);
 944         if (bt)
 945                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
 946         rcu_read_unlock();
 947 }
 948
 949 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 950                                     unsigned int depth, bool explicit)
 951 {
 952         struct blk_trace *bt;
 953
 954         rcu_read_lock();
 955         bt = rcu_dereference(q->blk_trace);
 956         if (bt) {
 957                 __be64 rpdu = cpu_to_be64(depth);
 958                 u32 what;
 959
 960                 if (explicit)
 961                         what = BLK_TA_UNPLUG_IO;
 962                 else
 963                         what = BLK_TA_UNPLUG_TIMER;
 964
 965                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
 966         }
 967         rcu_read_unlock();
 968 }
 969
 970 static void blk_add_trace_split(void *ignore,
 971                                 struct request_queue *q, struct bio *bio,
 972                                 unsigned int pdu)
 973 {
 974         struct blk_trace *bt;
 975
 976         rcu_read_lock();
 977         bt = rcu_dereference(q->blk_trace);
 978         if (bt) {
 979                 __be64 rpdu = cpu_to_be64(pdu);
 980
 981                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 982                                 bio->bi_iter.bi_size, bio->bi_rw, BLK_TA_SPLIT,
 983                                 bio->bi_error, sizeof(rpdu), &rpdu);
 984         }
 985         rcu_read_unlock();
 986 }
 987
 988 /**
 989  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 990  * @ignore:     trace callback data parameter (not used)
 991  * @q:          queue the io is for
 992  * @bio:        the source bio
 993  * @dev:        target device
 994  * @from:       source sector
 995  *
 996  * Description:
 997  *     Device mapper or raid target sometimes need to split a bio because
 998  *     it spans a stripe (or similar). Add a trace for that action.
 999  *
1000  **/
1001 static void blk_add_trace_bio_remap(void *ignore,
1002                                     struct request_queue *q, struct bio *bio,
1003                                     dev_t dev, sector_t from)
1004 {
1005         struct blk_trace *bt;
1006         struct blk_io_trace_remap r;
1007
1008         rcu_read_lock();
1009         bt = rcu_dereference(q->blk_trace);
1010         if (likely(!bt)) {
1011                 rcu_read_unlock();
1012                 return;
1013         }
1014
1015         r.device_from = cpu_to_be32(dev);
1016         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
1017         r.sector_from = cpu_to_be64(from);
1018
1019         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1020                         bio->bi_rw, BLK_TA_REMAP, bio->bi_error,
1021                         sizeof(r), &r);
1022         rcu_read_unlock();
1023 }
1024
1025 /**
1026  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1027  * @ignore:     trace callback data parameter (not used)
1028  * @q:          queue the io is for
1029  * @rq:         the source request
1030  * @dev:        target device
1031  * @from:       source sector
1032  *
1033  * Description:
1034  *     Device mapper remaps request to other devices.
1035  *     Add a trace for that action.
1036  *
1037  **/
1038 static void blk_add_trace_rq_remap(void *ignore,
1039                                    struct request_queue *q,
1040                                    struct request *rq, dev_t dev,
1041                                    sector_t from)
1042 {
1043         struct blk_trace *bt;
1044         struct blk_io_trace_remap r;
1045
1046         rcu_read_lock();
1047         bt = rcu_dereference(q->blk_trace);
1048         if (likely(!bt)) {
1049                 rcu_read_unlock();
1050                 return;
1051         }
1052
1053         r.device_from = cpu_to_be32(dev);
1054         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1055         r.sector_from = cpu_to_be64(from);
1056
1057         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1058                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
1059                         sizeof(r), &r);
1060         rcu_read_unlock();
1061 }
1062
1063 /**
1064  * blk_add_driver_data - Add binary message with driver-specific data
1065  * @q:          queue the io is for
1066  * @rq:         io request
1067  * @data:       driver-specific data
1068  * @len:        length of driver-specific data
1069  *
1070  * Description:
1071  *     Some drivers might want to write driver-specific data per request.
1072  *
1073  **/
1074 void blk_add_driver_data(struct request_queue *q,
1075                          struct request *rq,
1076                          void *data, size_t len)
1077 {
1078         struct blk_trace *bt;
1079
1080         rcu_read_lock();
1081         bt = rcu_dereference(q->blk_trace);
1082         if (likely(!bt)) {
1083                 rcu_read_unlock();
1084                 return;
1085         }
1086
1087         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
1088                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
1089                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1090         else
1091                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
1092                                 BLK_TA_DRV_DATA, rq->errors, len, data);
1093         rcu_read_unlock();
1094 }
1095 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1096
1097 static void blk_register_tracepoints(void)
1098 {
1099         int ret;
1100
1101         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1102         WARN_ON(ret);
1103         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1104         WARN_ON(ret);
1105         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1106         WARN_ON(ret);
1107         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1108         WARN_ON(ret);
1109         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1110         WARN_ON(ret);
1111         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1112         WARN_ON(ret);
1113         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1114         WARN_ON(ret);
1115         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1116         WARN_ON(ret);
1117         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1118         WARN_ON(ret);
1119         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1120         WARN_ON(ret);
1121         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1122         WARN_ON(ret);
1123         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1124         WARN_ON(ret);
1125         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1126         WARN_ON(ret);
1127         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1128         WARN_ON(ret);
1129         ret = register_trace_block_split(blk_add_trace_split, NULL);
1130         WARN_ON(ret);
1131         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1132         WARN_ON(ret);
1133         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1134         WARN_ON(ret);
1135 }
1136
1137 static void blk_unregister_tracepoints(void)
1138 {
1139         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1140         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1141         unregister_trace_block_split(blk_add_trace_split, NULL);
1142         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1143         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1144         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1145         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1146         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1147         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1148         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1149         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1150         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1151         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1152         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1153         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1154         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1155         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1156
1157         tracepoint_synchronize_unregister();
1158 }
1159
1160 /*
1161  * struct blk_io_tracer formatting routines
1162  */
1163
1164 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1165 {
1166         int i = 0;
1167         int tc = t->action >> BLK_TC_SHIFT;
1168
1169         if (t->action == BLK_TN_MESSAGE) {
1170                 rwbs[i++] = 'N';
1171                 goto out;
1172         }
1173
1174         if (tc & BLK_TC_FLUSH)
1175                 rwbs[i++] = 'F';
1176
1177         if (tc & BLK_TC_DISCARD)
1178                 rwbs[i++] = 'D';
1179         else if (tc & BLK_TC_WRITE)
1180                 rwbs[i++] = 'W';
1181         else if (t->bytes)
1182                 rwbs[i++] = 'R';
1183         else
1184                 rwbs[i++] = 'N';
1185
1186         if (tc & BLK_TC_FUA)
1187                 rwbs[i++] = 'F';
1188         if (tc & BLK_TC_AHEAD)
1189                 rwbs[i++] = 'A';
1190         if (tc & BLK_TC_SYNC)
1191                 rwbs[i++] = 'S';
1192         if (tc & BLK_TC_META)
1193                 rwbs[i++] = 'M';
1194 out:
1195         rwbs[i] = '\0';
1196 }
1197
1198 static inline
1199 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1200 {
1201         return (const struct blk_io_trace *)ent;
1202 }
1203
1204 static inline const void *pdu_start(const struct trace_entry *ent)
1205 {
1206         return te_blk_io_trace(ent) + 1;
1207 }
1208
1209 static inline u32 t_action(const struct trace_entry *ent)
1210 {
1211         return te_blk_io_trace(ent)->action;
1212 }
1213
1214 static inline u32 t_bytes(const struct trace_entry *ent)
1215 {
1216         return te_blk_io_trace(ent)->bytes;
1217 }
1218
1219 static inline u32 t_sec(const struct trace_entry *ent)
1220 {
1221         return te_blk_io_trace(ent)->bytes >> 9;
1222 }
1223
1224 static inline unsigned long long t_sector(const struct trace_entry *ent)
1225 {
1226         return te_blk_io_trace(ent)->sector;
1227 }
1228
1229 static inline __u16 t_error(const struct trace_entry *ent)
1230 {
1231         return te_blk_io_trace(ent)->error;
1232 }
1233
1234 static __u64 get_pdu_int(const struct trace_entry *ent)
1235 {
1236         const __u64 *val = pdu_start(ent);
1237         return be64_to_cpu(*val);
1238 }
1239
1240 static void get_pdu_remap(const struct trace_entry *ent,
1241                           struct blk_io_trace_remap *r)
1242 {
1243         const struct blk_io_trace_remap *__r = pdu_start(ent);
1244         __u64 sector_from = __r->sector_from;
1245
1246         r->device_from = be32_to_cpu(__r->device_from);
1247         r->device_to   = be32_to_cpu(__r->device_to);
1248         r->sector_from = be64_to_cpu(sector_from);
1249 }
1250
1251 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1252
1253 static void blk_log_action_classic(struct trace_iterator *iter, const char *act)
1254 {
1255         char rwbs[RWBS_LEN];
1256         unsigned long long ts  = iter->ts;
1257         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1258         unsigned secs          = (unsigned long)ts;
1259         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1260
1261         fill_rwbs(rwbs, t);
1262
1263         trace_seq_printf(&iter->seq,
1264                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1265                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1266                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1267 }
1268
1269 static void blk_log_action(struct trace_iterator *iter, const char *act)
1270 {
1271         char rwbs[RWBS_LEN];
1272         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1273
1274         fill_rwbs(rwbs, t);
1275         trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1276                          MAJOR(t->device), MINOR(t->device), act, rwbs);
1277 }
1278
1279 static void blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1280 {
1281         const unsigned char *pdu_buf;
1282         int pdu_len;
1283         int i, end;
1284
1285         pdu_buf = pdu_start(ent);
1286         pdu_len = te_blk_io_trace(ent)->pdu_len;
1287
1288         if (!pdu_len)
1289                 return;
1290
1291         /* find the last zero that needs to be printed */
1292         for (end = pdu_len - 1; end >= 0; end--)
1293                 if (pdu_buf[end])
1294                         break;
1295         end++;
1296
1297         trace_seq_putc(s, '(');
1298
1299         for (i = 0; i < pdu_len; i++) {
1300
1301                 trace_seq_printf(s, "%s%02x",
1302                                  i == 0 ? "" : " ", pdu_buf[i]);
1303
1304                 /*
1305                  * stop when the rest is just zeroes and indicate so
1306                  * with a ".." appended
1307                  */
1308                 if (i == end && end != pdu_len - 1) {
1309                         trace_seq_puts(s, " ..) ");
1310                         return;
1311                 }
1312         }
1313
1314         trace_seq_puts(s, ") ");
1315 }
1316
1317 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1318 {
1319         char cmd[TASK_COMM_LEN];
1320
1321         trace_find_cmdline(ent->pid, cmd);
1322
1323         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1324                 trace_seq_printf(s, "%u ", t_bytes(ent));
1325                 blk_log_dump_pdu(s, ent);
1326                 trace_seq_printf(s, "[%s]\n", cmd);
1327         } else {
1328                 if (t_sec(ent))
1329                         trace_seq_printf(s, "%llu + %u [%s]\n",
1330                                                 t_sector(ent), t_sec(ent), cmd);
1331                 else
1332                         trace_seq_printf(s, "[%s]\n", cmd);
1333         }
1334 }
1335
1336 static void blk_log_with_error(struct trace_seq *s,
1337                               const struct trace_entry *ent)
1338 {
1339         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1340                 blk_log_dump_pdu(s, ent);
1341                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1342         } else {
1343                 if (t_sec(ent))
1344                         trace_seq_printf(s, "%llu + %u [%d]\n",
1345                                          t_sector(ent),
1346                                          t_sec(ent), t_error(ent));
1347                 else
1348                         trace_seq_printf(s, "%llu [%d]\n",
1349                                          t_sector(ent), t_error(ent));
1350         }
1351 }
1352
1353 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1354 {
1355         struct blk_io_trace_remap r = { .device_from = 0, };
1356
1357         get_pdu_remap(ent, &r);
1358         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1359                          t_sector(ent), t_sec(ent),
1360                          MAJOR(r.device_from), MINOR(r.device_from),
1361                          (unsigned long long)r.sector_from);
1362 }
1363
1364 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1365 {
1366         char cmd[TASK_COMM_LEN];
1367
1368         trace_find_cmdline(ent->pid, cmd);
1369
1370         trace_seq_printf(s, "[%s]\n", cmd);
1371 }
1372
1373 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1374 {
1375         char cmd[TASK_COMM_LEN];
1376
1377         trace_find_cmdline(ent->pid, cmd);
1378
1379         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1380 }
1381
1382 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1383 {
1384         char cmd[TASK_COMM_LEN];
1385
1386         trace_find_cmdline(ent->pid, cmd);
1387
1388         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1389                          get_pdu_int(ent), cmd);
1390 }
1391
1392 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1393 {
1394         const struct blk_io_trace *t = te_blk_io_trace(ent);
1395
1396         trace_seq_putmem(s, t + 1, t->pdu_len);
1397         trace_seq_putc(s, '\n');
1398 }
1399
1400 /*
1401  * struct tracer operations
1402  */
1403
1404 static void blk_tracer_print_header(struct seq_file *m)
1405 {
1406         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1407                 return;
1408         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1409                     "#  |     |     |           |   |   |\n");
1410 }
1411
1412 static void blk_tracer_start(struct trace_array *tr)
1413 {
1414         blk_tracer_enabled = true;
1415 }
1416
1417 static int blk_tracer_init(struct trace_array *tr)
1418 {
1419         blk_tr = tr;
1420         blk_tracer_start(tr);
1421         return 0;
1422 }
1423
1424 static void blk_tracer_stop(struct trace_array *tr)
1425 {
1426         blk_tracer_enabled = false;
1427 }
1428
1429 static void blk_tracer_reset(struct trace_array *tr)
1430 {
1431         blk_tracer_stop(tr);
1432 }
1433
1434 static const struct {
1435         const char *act[2];
1436         void       (*print)(struct trace_seq *s, const struct trace_entry *ent);
1437 } what2act[] = {
1438         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1439         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1440         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1441         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1442         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1443         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1444         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1445         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1446         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1447         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1448         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1449         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1450         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1451         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1452         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1453 };
1454
1455 static enum print_line_t print_one_line(struct trace_iterator *iter,
1456                                         bool classic)
1457 {
1458         struct trace_array *tr = iter->tr;
1459         struct trace_seq *s = &iter->seq;
1460         const struct blk_io_trace *t;
1461         u16 what;
1462         bool long_act;
1463         blk_log_action_t *log_action;
1464
1465         t          = te_blk_io_trace(iter->ent);
1466         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1467         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1468         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1469
1470         if (t->action == BLK_TN_MESSAGE) {
1471                 log_action(iter, long_act ? "message" : "m");
1472                 blk_log_msg(s, iter->ent);
1473         }
1474
1475         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1476                 trace_seq_printf(s, "Unknown action %x\n", what);
1477         else {
1478                 log_action(iter, what2act[what].act[long_act]);
1479                 what2act[what].print(s, iter->ent);
1480         }
1481
1482         return trace_handle_return(s);
1483 }
1484
1485 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1486                                                int flags, struct trace_event *event)
1487 {
1488         return print_one_line(iter, false);
1489 }
1490
1491 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1492 {
1493         struct trace_seq *s = &iter->seq;
1494         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1495         const int offset = offsetof(struct blk_io_trace, sector);
1496         struct blk_io_trace old = {
1497                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1498                 .time     = iter->ts,
1499         };
1500
1501         trace_seq_putmem(s, &old, offset);
1502         trace_seq_putmem(s, &t->sector,
1503                          sizeof(old) - offset + t->pdu_len);
1504 }
1505
1506 static enum print_line_t
1507 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1508                              struct trace_event *event)
1509 {
1510         blk_trace_synthesize_old_trace(iter);
1511
1512         return trace_handle_return(&iter->seq);
1513 }
1514
1515 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1516 {
1517         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1518                 return TRACE_TYPE_UNHANDLED;
1519
1520         return print_one_line(iter, true);
1521 }
1522
1523 static int
1524 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1525 {
1526         /* don't output context-info for blk_classic output */
1527         if (bit == TRACE_BLK_OPT_CLASSIC) {
1528                 if (set)
1529                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1530                 else
1531                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1532         }
1533         return 0;
1534 }
1535
1536 static struct tracer blk_tracer __read_mostly = {
1537         .name           = "blk",
1538         .init           = blk_tracer_init,
1539         .reset          = blk_tracer_reset,
1540         .start          = blk_tracer_start,
1541         .stop           = blk_tracer_stop,
1542         .print_header   = blk_tracer_print_header,
1543         .print_line     = blk_tracer_print_line,
1544         .flags          = &blk_tracer_flags,
1545         .set_flag       = blk_tracer_set_flag,
1546 };
1547
1548 static struct trace_event_functions trace_blk_event_funcs = {
1549         .trace          = blk_trace_event_print,
1550         .binary         = blk_trace_event_print_binary,
1551 };
1552
1553 static struct trace_event trace_blk_event = {
1554         .type           = TRACE_BLK,
1555         .funcs          = &trace_blk_event_funcs,
1556 };
1557
1558 static int __init init_blk_tracer(void)
1559 {
1560         if (!register_trace_event(&trace_blk_event)) {
1561                 pr_warning("Warning: could not register block events\n");
1562                 return 1;
1563         }
1564
1565         if (register_tracer(&blk_tracer) != 0) {
1566                 pr_warning("Warning: could not register the block tracer\n");
1567                 unregister_trace_event(&trace_blk_event);
1568                 return 1;
1569         }
1570
1571         return 0;
1572 }
1573
1574 device_initcall(init_blk_tracer);
1575
1576 static int blk_trace_remove_queue(struct request_queue *q)
1577 {
1578         struct blk_trace *bt;
1579
1580         bt = xchg(&q->blk_trace, NULL);
1581         if (bt == NULL)
1582                 return -EINVAL;
1583
1584         if (bt->trace_state == Blktrace_running) {
1585                 bt->trace_state = Blktrace_stopped;
1586                 spin_lock_irq(&running_trace_lock);
1587                 list_del_init(&bt->running_list);
1588                 spin_unlock_irq(&running_trace_lock);
1589                 relay_flush(bt->rchan);
1590         }
1591
1592         put_probe_ref();
1593         synchronize_rcu();
1594         blk_trace_free(bt);
1595         return 0;
1596 }
1597
1598 /*
1599  * Setup everything required to start tracing
1600  */
1601 static int blk_trace_setup_queue(struct request_queue *q,
1602                                  struct block_device *bdev)
1603 {
1604         struct blk_trace *bt = NULL;
1605         int ret = -ENOMEM;
1606
1607         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1608         if (!bt)
1609                 return -ENOMEM;
1610
1611         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1612         if (!bt->msg_data)
1613                 goto free_bt;
1614
1615         bt->dev = bdev->bd_dev;
1616         bt->act_mask = (u16)-1;
1617
1618         blk_trace_setup_lba(bt, bdev);
1619
1620         ret = -EBUSY;
1621         if (cmpxchg(&q->blk_trace, NULL, bt))
1622                 goto free_bt;
1623
1624         get_probe_ref();
1625         return 0;
1626
1627 free_bt:
1628         blk_trace_free(bt);
1629         return ret;
1630 }
1631
1632 /*
1633  * sysfs interface to enable and configure tracing
1634  */
1635
1636 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1637                                          struct device_attribute *attr,
1638                                          char *buf);
1639 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1640                                           struct device_attribute *attr,
1641                                           const char *buf, size_t count);
1642 #define BLK_TRACE_DEVICE_ATTR(_name) \
1643         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1644                     sysfs_blk_trace_attr_show, \
1645                     sysfs_blk_trace_attr_store)
1646
1647 static BLK_TRACE_DEVICE_ATTR(enable);
1648 static BLK_TRACE_DEVICE_ATTR(act_mask);
1649 static BLK_TRACE_DEVICE_ATTR(pid);
1650 static BLK_TRACE_DEVICE_ATTR(start_lba);
1651 static BLK_TRACE_DEVICE_ATTR(end_lba);
1652
1653 static struct attribute *blk_trace_attrs[] = {
1654         &dev_attr_enable.attr,
1655         &dev_attr_act_mask.attr,
1656         &dev_attr_pid.attr,
1657         &dev_attr_start_lba.attr,
1658         &dev_attr_end_lba.attr,
1659         NULL
1660 };
1661
1662 struct attribute_group blk_trace_attr_group = {
1663         .name  = "trace",
1664         .attrs = blk_trace_attrs,
1665 };
1666
1667 static const struct {
1668         int mask;
1669         const char *str;
1670 } mask_maps[] = {
1671         { BLK_TC_READ,          "read"          },
1672         { BLK_TC_WRITE,         "write"         },
1673         { BLK_TC_FLUSH,         "flush"         },
1674         { BLK_TC_SYNC,          "sync"          },
1675         { BLK_TC_QUEUE,         "queue"         },
1676         { BLK_TC_REQUEUE,       "requeue"       },
1677         { BLK_TC_ISSUE,         "issue"         },
1678         { BLK_TC_COMPLETE,      "complete"      },
1679         { BLK_TC_FS,            "fs"            },
1680         { BLK_TC_PC,            "pc"            },
1681         { BLK_TC_AHEAD,         "ahead"         },
1682         { BLK_TC_META,          "meta"          },
1683         { BLK_TC_DISCARD,       "discard"       },
1684         { BLK_TC_DRV_DATA,      "drv_data"      },
1685         { BLK_TC_FUA,           "fua"           },
1686 };
1687
1688 static int blk_trace_str2mask(const char *str)
1689 {
1690         int i;
1691         int mask = 0;
1692         char *buf, *s, *token;
1693
1694         buf = kstrdup(str, GFP_KERNEL);
1695         if (buf == NULL)
1696                 return -ENOMEM;
1697         s = strstrip(buf);
1698
1699         while (1) {
1700                 token = strsep(&s, ",");
1701                 if (token == NULL)
1702                         break;
1703
1704                 if (*token == '\0')
1705                         continue;
1706
1707                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1708                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1709                                 mask |= mask_maps[i].mask;
1710                                 break;
1711                         }
1712                 }
1713                 if (i == ARRAY_SIZE(mask_maps)) {
1714                         mask = -EINVAL;
1715                         break;
1716                 }
1717         }
1718         kfree(buf);
1719
1720         return mask;
1721 }
1722
1723 static ssize_t blk_trace_mask2str(char *buf, int mask)
1724 {
1725         int i;
1726         char *p = buf;
1727
1728         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1729                 if (mask & mask_maps[i].mask) {
1730                         p += sprintf(p, "%s%s",
1731                                     (p == buf) ? "" : ",", mask_maps[i].str);
1732                 }
1733         }
1734         *p++ = '\n';
1735
1736         return p - buf;
1737 }
1738
1739 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1740 {
1741         if (bdev->bd_disk == NULL)
1742                 return NULL;
1743
1744         return bdev_get_queue(bdev);
1745 }
1746
1747 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1748                                          struct device_attribute *attr,
1749                                          char *buf)
1750 {
1751         struct hd_struct *p = dev_to_part(dev);
1752         struct request_queue *q;
1753         struct block_device *bdev;
1754         struct blk_trace *bt;
1755         ssize_t ret = -ENXIO;
1756
1757         bdev = bdget(part_devt(p));
1758         if (bdev == NULL)
1759                 goto out;
1760
1761         q = blk_trace_get_queue(bdev);
1762         if (q == NULL)
1763                 goto out_bdput;
1764
1765         mutex_lock(&q->blk_trace_mutex);
1766
1767         bt = rcu_dereference_protected(q->blk_trace,
1768                                        lockdep_is_held(&q->blk_trace_mutex));
1769         if (attr == &dev_attr_enable) {
1770                 ret = sprintf(buf, "%u\n", !!bt);
1771                 goto out_unlock_bdev;
1772         }
1773
1774         if (bt == NULL)
1775                 ret = sprintf(buf, "disabled\n");
1776         else if (attr == &dev_attr_act_mask)
1777                 ret = blk_trace_mask2str(buf, bt->act_mask);
1778         else if (attr == &dev_attr_pid)
1779                 ret = sprintf(buf, "%u\n", bt->pid);
1780         else if (attr == &dev_attr_start_lba)
1781                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1782         else if (attr == &dev_attr_end_lba)
1783                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1784
1785 out_unlock_bdev:
1786         mutex_unlock(&q->blk_trace_mutex);
1787 out_bdput:
1788         bdput(bdev);
1789 out:
1790         return ret;
1791 }
1792
1793 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1794                                           struct device_attribute *attr,
1795                                           const char *buf, size_t count)
1796 {
1797         struct block_device *bdev;
1798         struct request_queue *q;
1799         struct hd_struct *p;
1800         struct blk_trace *bt;
1801         u64 value;
1802         ssize_t ret = -EINVAL;
1803
1804         if (count == 0)
1805                 goto out;
1806
1807         if (attr == &dev_attr_act_mask) {
1808                 if (sscanf(buf, "%llx", &value) != 1) {
1809                         /* Assume it is a list of trace category names */
1810                         ret = blk_trace_str2mask(buf);
1811                         if (ret < 0)
1812                                 goto out;
1813                         value = ret;
1814                 }
1815         } else if (sscanf(buf, "%llu", &value) != 1)
1816                 goto out;
1817
1818         ret = -ENXIO;
1819
1820         p = dev_to_part(dev);
1821         bdev = bdget(part_devt(p));
1822         if (bdev == NULL)
1823                 goto out;
1824
1825         q = blk_trace_get_queue(bdev);
1826         if (q == NULL)
1827                 goto out_bdput;
1828
1829         mutex_lock(&q->blk_trace_mutex);
1830
1831         bt = rcu_dereference_protected(q->blk_trace,
1832                                        lockdep_is_held(&q->blk_trace_mutex));
1833         if (attr == &dev_attr_enable) {
1834                 if (!!value == !!bt) {
1835                         ret = 0;
1836                         goto out_unlock_bdev;
1837                 }
1838                 if (value)
1839                         ret = blk_trace_setup_queue(q, bdev);
1840                 else
1841                         ret = blk_trace_remove_queue(q);
1842                 goto out_unlock_bdev;
1843         }
1844
1845         ret = 0;
1846         if (bt == NULL) {
1847                 ret = blk_trace_setup_queue(q, bdev);
1848                 bt = rcu_dereference_protected(q->blk_trace,
1849                                 lockdep_is_held(&q->blk_trace_mutex));
1850         }
1851
1852         if (ret == 0) {
1853                 if (attr == &dev_attr_act_mask)
1854                         bt->act_mask = value;
1855                 else if (attr == &dev_attr_pid)
1856                         bt->pid = value;
1857                 else if (attr == &dev_attr_start_lba)
1858                         bt->start_lba = value;
1859                 else if (attr == &dev_attr_end_lba)
1860                         bt->end_lba = value;
1861         }
1862
1863 out_unlock_bdev:
1864         mutex_unlock(&q->blk_trace_mutex);
1865 out_bdput:
1866         bdput(bdev);
1867 out:
1868         return ret ? ret : count;
1869 }
1870
1871 int blk_trace_init_sysfs(struct device *dev)
1872 {
1873         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1874 }
1875
1876 void blk_trace_remove_sysfs(struct device *dev)
1877 {
1878         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1879 }
1880
1881 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1882
1883 #ifdef CONFIG_EVENT_TRACING
1884
1885 void blk_dump_cmd(char *buf, struct request *rq)
1886 {
1887         int i, end;
1888         int len = rq->cmd_len;
1889         unsigned char *cmd = rq->cmd;
1890
1891         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1892                 buf[0] = '\0';
1893                 return;
1894         }
1895
1896         for (end = len - 1; end >= 0; end--)
1897                 if (cmd[end])
1898                         break;
1899         end++;
1900
1901         for (i = 0; i < len; i++) {
1902                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1903                 if (i == end && end != len - 1) {
1904                         sprintf(buf, " ..");
1905                         break;
1906                 }
1907         }
1908 }
1909
1910 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1911 {
1912         int i = 0;
1913
1914         if (rw & REQ_FLUSH)
1915                 rwbs[i++] = 'F';
1916
1917         if (rw & WRITE)
1918                 rwbs[i++] = 'W';
1919         else if (rw & REQ_DISCARD)
1920                 rwbs[i++] = 'D';
1921         else if (bytes)
1922                 rwbs[i++] = 'R';
1923         else
1924                 rwbs[i++] = 'N';
1925
1926         if (rw & REQ_FUA)
1927                 rwbs[i++] = 'F';
1928         if (rw & REQ_RAHEAD)
1929                 rwbs[i++] = 'A';
1930         if (rw & REQ_SYNC)
1931                 rwbs[i++] = 'S';
1932         if (rw & REQ_META)
1933                 rwbs[i++] = 'M';
1934         if (rw & REQ_SECURE)
1935                 rwbs[i++] = 'E';
1936
1937         rwbs[i] = '\0';
1938 }
1939 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1940
1941 #endif /* CONFIG_EVENT_TRACING */
1942