kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/export.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/list.h>
  30 #include <linux/blk-cgroup.h>
  31
  32 #include "../../block/blk.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 #define TRACE_BLK_OPT_CGROUP    0x2
  51 #define TRACE_BLK_OPT_CGNAME    0x4
  52
  53 static struct tracer_opt blk_tracer_opts[] = {
  54         /* Default disable the minimalistic output */
  55         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  56 #ifdef CONFIG_BLK_CGROUP
  57         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  58         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  59 #endif
  60         { }
  61 };
  62
  63 static struct tracer_flags blk_tracer_flags = {
  64         .val  = 0,
  65         .opts = blk_tracer_opts,
  66 };
  67
  68 /* Global reference count of probes */
  69 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
  70
  71 static void blk_register_tracepoints(void);
  72 static void blk_unregister_tracepoints(void);
  73
  74 /*
  75  * Send out a notify message.
  76  */
  77 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  78                        const void *data, size_t len,
  79                        union kernfs_node_id *cgid)
  80 {
  81         struct blk_io_trace *t;
  82         struct ring_buffer_event *event = NULL;
  83         struct ring_buffer *buffer = NULL;
  84         int pc = 0;
  85         int cpu = smp_processor_id();
  86         bool blk_tracer = blk_tracer_enabled;
  87         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  88
  89         if (blk_tracer) {
  90                 buffer = blk_tr->trace_buffer.buffer;
  91                 pc = preempt_count();
  92                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  93                                                   sizeof(*t) + len + cgid_len,
  94                                                   0, pc);
  95                 if (!event)
  96                         return;
  97                 t = ring_buffer_event_data(event);
  98                 goto record_it;
  99         }
 100
 101         if (!bt->rchan)
 102                 return;
 103
 104         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
 105         if (t) {
 106                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 107                 t->time = ktime_to_ns(ktime_get());
 108 record_it:
 109                 t->device = bt->dev;
 110                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 111                 t->pid = pid;
 112                 t->cpu = cpu;
 113                 t->pdu_len = len + cgid_len;
 114                 if (cgid)
 115                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 116                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 117
 118                 if (blk_tracer)
 119                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 120         }
 121 }
 122
 123 /*
 124  * Send out a notify for this process, if we haven't done so since a trace
 125  * started
 126  */
 127 static void trace_note_tsk(struct task_struct *tsk)
 128 {
 129         unsigned long flags;
 130         struct blk_trace *bt;
 131
 132         tsk->btrace_seq = blktrace_seq;
 133         spin_lock_irqsave(&running_trace_lock, flags);
 134         list_for_each_entry(bt, &running_trace_list, running_list) {
 135                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 136                            sizeof(tsk->comm), NULL);
 137         }
 138         spin_unlock_irqrestore(&running_trace_lock, flags);
 139 }
 140
 141 static void trace_note_time(struct blk_trace *bt)
 142 {
 143         struct timespec64 now;
 144         unsigned long flags;
 145         u32 words[2];
 146
 147         /* need to check user space to see if this breaks in y2038 or y2106 */
 148         ktime_get_real_ts64(&now);
 149         words[0] = (u32)now.tv_sec;
 150         words[1] = now.tv_nsec;
 151
 152         local_irq_save(flags);
 153         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 154         local_irq_restore(flags);
 155 }
 156
 157 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 158         const char *fmt, ...)
 159 {
 160         int n;
 161         va_list args;
 162         unsigned long flags;
 163         char *buf;
 164
 165         if (unlikely(bt->trace_state != Blktrace_running &&
 166                      !blk_tracer_enabled))
 167                 return;
 168
 169         /*
 170          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 171          * message to the trace.
 172          */
 173         if (!(bt->act_mask & BLK_TC_NOTIFY))
 174                 return;
 175
 176         local_irq_save(flags);
 177         buf = this_cpu_ptr(bt->msg_data);
 178         va_start(args, fmt);
 179         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 180         va_end(args);
 181
 182         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 183                 blkcg = NULL;
 184 #ifdef CONFIG_BLK_CGROUP
 185         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 186                 blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 187 #else
 188         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 189 #endif
 190         local_irq_restore(flags);
 191 }
 192 EXPORT_SYMBOL_GPL(__trace_note_message);
 193
 194 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 195                          pid_t pid)
 196 {
 197         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 198                 return 1;
 199         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 200                 return 1;
 201         if (bt->pid && pid != bt->pid)
 202                 return 1;
 203
 204         return 0;
 205 }
 206
 207 /*
 208  * Data direction bit lookup
 209  */
 210 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 211                                  BLK_TC_ACT(BLK_TC_WRITE) };
 212
 213 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 214 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 215
 216 /* The ilog2() calls fall out because they're constant */
 217 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 218           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 219
 220 /*
 221  * The worker for the various blk_add_trace*() types. Fills out a
 222  * blk_io_trace structure and places it in a per-cpu subbuffer.
 223  */
 224 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 225                      int op, int op_flags, u32 what, int error, int pdu_len,
 226                      void *pdu_data, union kernfs_node_id *cgid)
 227 {
 228         struct task_struct *tsk = current;
 229         struct ring_buffer_event *event = NULL;
 230         struct ring_buffer *buffer = NULL;
 231         struct blk_io_trace *t;
 232         unsigned long flags = 0;
 233         unsigned long *sequence;
 234         pid_t pid;
 235         int cpu, pc = 0;
 236         bool blk_tracer = blk_tracer_enabled;
 237         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 238
 239         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 240                 return;
 241
 242         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 243         what |= MASK_TC_BIT(op_flags, SYNC);
 244         what |= MASK_TC_BIT(op_flags, RAHEAD);
 245         what |= MASK_TC_BIT(op_flags, META);
 246         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 247         what |= MASK_TC_BIT(op_flags, FUA);
 248         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 249                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 250         if (op == REQ_OP_FLUSH)
 251                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 252         if (cgid)
 253                 what |= __BLK_TA_CGROUP;
 254
 255         pid = tsk->pid;
 256         if (act_log_check(bt, what, sector, pid))
 257                 return;
 258         cpu = raw_smp_processor_id();
 259
 260         if (blk_tracer) {
 261                 tracing_record_cmdline(current);
 262
 263                 buffer = blk_tr->trace_buffer.buffer;
 264                 pc = preempt_count();
 265                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 266                                                   sizeof(*t) + pdu_len + cgid_len,
 267                                                   0, pc);
 268                 if (!event)
 269                         return;
 270                 t = ring_buffer_event_data(event);
 271                 goto record_it;
 272         }
 273
 274         if (unlikely(tsk->btrace_seq != blktrace_seq))
 275                 trace_note_tsk(tsk);
 276
 277         /*
 278          * A word about the locking here - we disable interrupts to reserve
 279          * some space in the relay per-cpu buffer, to prevent an irq
 280          * from coming in and stepping on our toes.
 281          */
 282         local_irq_save(flags);
 283         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 284         if (t) {
 285                 sequence = per_cpu_ptr(bt->sequence, cpu);
 286
 287                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 288                 t->sequence = ++(*sequence);
 289                 t->time = ktime_to_ns(ktime_get());
 290 record_it:
 291                 /*
 292                  * These two are not needed in ftrace as they are in the
 293                  * generic trace_entry, filled by tracing_generic_entry_update,
 294                  * but for the trace_event->bin() synthesizer benefit we do it
 295                  * here too.
 296                  */
 297                 t->cpu = cpu;
 298                 t->pid = pid;
 299
 300                 t->sector = sector;
 301                 t->bytes = bytes;
 302                 t->action = what;
 303                 t->device = bt->dev;
 304                 t->error = error;
 305                 t->pdu_len = pdu_len + cgid_len;
 306
 307                 if (cgid_len)
 308                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 309                 if (pdu_len)
 310                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 311
 312                 if (blk_tracer) {
 313                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 314                         return;
 315                 }
 316         }
 317
 318         local_irq_restore(flags);
 319 }
 320
 321 static void blk_trace_free(struct blk_trace *bt)
 322 {
 323         debugfs_remove(bt->msg_file);
 324         debugfs_remove(bt->dropped_file);
 325         relay_close(bt->rchan);
 326         debugfs_remove(bt->dir);
 327         free_percpu(bt->sequence);
 328         free_percpu(bt->msg_data);
 329         kfree(bt);
 330 }
 331
 332 static void blk_trace_cleanup(struct blk_trace *bt)
 333 {
 334         blk_trace_free(bt);
 335         if (atomic_dec_and_test(&blk_probes_ref))
 336                 blk_unregister_tracepoints();
 337 }
 338
 339 int blk_trace_remove(struct request_queue *q)
 340 {
 341         struct blk_trace *bt;
 342
 343         bt = xchg(&q->blk_trace, NULL);
 344         if (!bt)
 345                 return -EINVAL;
 346
 347         if (bt->trace_state != Blktrace_running)
 348                 blk_trace_cleanup(bt);
 349
 350         return 0;
 351 }
 352 EXPORT_SYMBOL_GPL(blk_trace_remove);
 353
 354 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 355                                 size_t count, loff_t *ppos)
 356 {
 357         struct blk_trace *bt = filp->private_data;
 358         char buf[16];
 359
 360         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 361
 362         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 363 }
 364
 365 static const struct file_operations blk_dropped_fops = {
 366         .owner =        THIS_MODULE,
 367         .open =         simple_open,
 368         .read =         blk_dropped_read,
 369         .llseek =       default_llseek,
 370 };
 371
 372 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 373                                 size_t count, loff_t *ppos)
 374 {
 375         char *msg;
 376         struct blk_trace *bt;
 377
 378         if (count >= BLK_TN_MAX_MSG)
 379                 return -EINVAL;
 380
 381         msg = memdup_user_nul(buffer, count);
 382         if (IS_ERR(msg))
 383                 return PTR_ERR(msg);
 384
 385         bt = filp->private_data;
 386         __trace_note_message(bt, NULL, "%s", msg);
 387         kfree(msg);
 388
 389         return count;
 390 }
 391
 392 static const struct file_operations blk_msg_fops = {
 393         .owner =        THIS_MODULE,
 394         .open =         simple_open,
 395         .write =        blk_msg_write,
 396         .llseek =       noop_llseek,
 397 };
 398
 399 /*
 400  * Keep track of how many times we encountered a full subbuffer, to aid
 401  * the user space app in telling how many lost events there were.
 402  */
 403 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 404                                      void *prev_subbuf, size_t prev_padding)
 405 {
 406         struct blk_trace *bt;
 407
 408         if (!relay_buf_full(buf))
 409                 return 1;
 410
 411         bt = buf->chan->private_data;
 412         atomic_inc(&bt->dropped);
 413         return 0;
 414 }
 415
 416 static int blk_remove_buf_file_callback(struct dentry *dentry)
 417 {
 418         debugfs_remove(dentry);
 419
 420         return 0;
 421 }
 422
 423 static struct dentry *blk_create_buf_file_callback(const char *filename,
 424                                                    struct dentry *parent,
 425                                                    umode_t mode,
 426                                                    struct rchan_buf *buf,
 427                                                    int *is_global)
 428 {
 429         return debugfs_create_file(filename, mode, parent, buf,
 430                                         &relay_file_operations);
 431 }
 432
 433 static struct rchan_callbacks blk_relay_callbacks = {
 434         .subbuf_start           = blk_subbuf_start_callback,
 435         .create_buf_file        = blk_create_buf_file_callback,
 436         .remove_buf_file        = blk_remove_buf_file_callback,
 437 };
 438
 439 static void blk_trace_setup_lba(struct blk_trace *bt,
 440                                 struct block_device *bdev)
 441 {
 442         struct hd_struct *part = NULL;
 443
 444         if (bdev)
 445                 part = bdev->bd_part;
 446
 447         if (part) {
 448                 bt->start_lba = part->start_sect;
 449                 bt->end_lba = part->start_sect + part->nr_sects;
 450         } else {
 451                 bt->start_lba = 0;
 452                 bt->end_lba = -1ULL;
 453         }
 454 }
 455
 456 /*
 457  * Setup everything required to start tracing
 458  */
 459 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 460                               struct block_device *bdev,
 461                               struct blk_user_trace_setup *buts)
 462 {
 463         struct blk_trace *bt = NULL;
 464         struct dentry *dir = NULL;
 465         int ret;
 466
 467         if (!buts->buf_size || !buts->buf_nr)
 468                 return -EINVAL;
 469
 470         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 471         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 472
 473         /*
 474          * some device names have larger paths - convert the slashes
 475          * to underscores for this to work as expected
 476          */
 477         strreplace(buts->name, '/', '_');
 478
 479         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 480         if (!bt)
 481                 return -ENOMEM;
 482
 483         ret = -ENOMEM;
 484         bt->sequence = alloc_percpu(unsigned long);
 485         if (!bt->sequence)
 486                 goto err;
 487
 488         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 489         if (!bt->msg_data)
 490                 goto err;
 491
 492         ret = -ENOENT;
 493
 494         if (!blk_debugfs_root)
 495                 goto err;
 496
 497         dir = debugfs_lookup(buts->name, blk_debugfs_root);
 498         if (!dir)
 499                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 500         if (!dir)
 501                 goto err;
 502
 503         bt->dev = dev;
 504         atomic_set(&bt->dropped, 0);
 505         INIT_LIST_HEAD(&bt->running_list);
 506
 507         ret = -EIO;
 508         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 509                                                &blk_dropped_fops);
 510         if (!bt->dropped_file)
 511                 goto err;
 512
 513         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 514         if (!bt->msg_file)
 515                 goto err;
 516
 517         bt->rchan = relay_open("trace", dir, buts->buf_size,
 518                                 buts->buf_nr, &blk_relay_callbacks, bt);
 519         if (!bt->rchan)
 520                 goto err;
 521
 522         bt->act_mask = buts->act_mask;
 523         if (!bt->act_mask)
 524                 bt->act_mask = (u16) -1;
 525
 526         blk_trace_setup_lba(bt, bdev);
 527
 528         /* overwrite with user settings */
 529         if (buts->start_lba)
 530                 bt->start_lba = buts->start_lba;
 531         if (buts->end_lba)
 532                 bt->end_lba = buts->end_lba;
 533
 534         bt->pid = buts->pid;
 535         bt->trace_state = Blktrace_setup;
 536
 537         ret = -EBUSY;
 538         if (cmpxchg(&q->blk_trace, NULL, bt))
 539                 goto err;
 540
 541         if (atomic_inc_return(&blk_probes_ref) == 1)
 542                 blk_register_tracepoints();
 543
 544         ret = 0;
 545 err:
 546         if (dir && !bt->dir)
 547                 dput(dir);
 548         if (ret)
 549                 blk_trace_free(bt);
 550         return ret;
 551 }
 552
 553 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 554                     struct block_device *bdev,
 555                     char __user *arg)
 556 {
 557         struct blk_user_trace_setup buts;
 558         int ret;
 559
 560         ret = copy_from_user(&buts, arg, sizeof(buts));
 561         if (ret)
 562                 return -EFAULT;
 563
 564         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 565         if (ret)
 566                 return ret;
 567
 568         if (copy_to_user(arg, &buts, sizeof(buts))) {
 569                 blk_trace_remove(q);
 570                 return -EFAULT;
 571         }
 572         return 0;
 573 }
 574 EXPORT_SYMBOL_GPL(blk_trace_setup);
 575
 576 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 577 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 578                                   dev_t dev, struct block_device *bdev,
 579                                   char __user *arg)
 580 {
 581         struct blk_user_trace_setup buts;
 582         struct compat_blk_user_trace_setup cbuts;
 583         int ret;
 584
 585         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 586                 return -EFAULT;
 587
 588         buts = (struct blk_user_trace_setup) {
 589                 .act_mask = cbuts.act_mask,
 590                 .buf_size = cbuts.buf_size,
 591                 .buf_nr = cbuts.buf_nr,
 592                 .start_lba = cbuts.start_lba,
 593                 .end_lba = cbuts.end_lba,
 594                 .pid = cbuts.pid,
 595         };
 596
 597         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 598         if (ret)
 599                 return ret;
 600
 601         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 602                 blk_trace_remove(q);
 603                 return -EFAULT;
 604         }
 605
 606         return 0;
 607 }
 608 #endif
 609
 610 int blk_trace_startstop(struct request_queue *q, int start)
 611 {
 612         int ret;
 613         struct blk_trace *bt = q->blk_trace;
 614
 615         if (bt == NULL)
 616                 return -EINVAL;
 617
 618         /*
 619          * For starting a trace, we can transition from a setup or stopped
 620          * trace. For stopping a trace, the state must be running
 621          */
 622         ret = -EINVAL;
 623         if (start) {
 624                 if (bt->trace_state == Blktrace_setup ||
 625                     bt->trace_state == Blktrace_stopped) {
 626                         blktrace_seq++;
 627                         smp_mb();
 628                         bt->trace_state = Blktrace_running;
 629                         spin_lock_irq(&running_trace_lock);
 630                         list_add(&bt->running_list, &running_trace_list);
 631                         spin_unlock_irq(&running_trace_lock);
 632
 633                         trace_note_time(bt);
 634                         ret = 0;
 635                 }
 636         } else {
 637                 if (bt->trace_state == Blktrace_running) {
 638                         bt->trace_state = Blktrace_stopped;
 639                         spin_lock_irq(&running_trace_lock);
 640                         list_del_init(&bt->running_list);
 641                         spin_unlock_irq(&running_trace_lock);
 642                         relay_flush(bt->rchan);
 643                         ret = 0;
 644                 }
 645         }
 646
 647         return ret;
 648 }
 649 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 650
 651 /*
 652  * When reading or writing the blktrace sysfs files, the references to the
 653  * opened sysfs or device files should prevent the underlying block device
 654  * from being removed. So no further delete protection is really needed.
 655  */
 656
 657 /**
 658  * blk_trace_ioctl: - handle the ioctls associated with tracing
 659  * @bdev:       the block device
 660  * @cmd:        the ioctl cmd
 661  * @arg:        the argument data, if any
 662  *
 663  **/
 664 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 665 {
 666         struct request_queue *q;
 667         int ret, start = 0;
 668         char b[BDEVNAME_SIZE];
 669
 670         q = bdev_get_queue(bdev);
 671         if (!q)
 672                 return -ENXIO;
 673
 674         mutex_lock(&q->blk_trace_mutex);
 675
 676         switch (cmd) {
 677         case BLKTRACESETUP:
 678                 bdevname(bdev, b);
 679                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 680                 break;
 681 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 682         case BLKTRACESETUP32:
 683                 bdevname(bdev, b);
 684                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 685                 break;
 686 #endif
 687         case BLKTRACESTART:
 688                 start = 1;
 689         case BLKTRACESTOP:
 690                 ret = blk_trace_startstop(q, start);
 691                 break;
 692         case BLKTRACETEARDOWN:
 693                 ret = blk_trace_remove(q);
 694                 break;
 695         default:
 696                 ret = -ENOTTY;
 697                 break;
 698         }
 699
 700         mutex_unlock(&q->blk_trace_mutex);
 701         return ret;
 702 }
 703
 704 /**
 705  * blk_trace_shutdown: - stop and cleanup trace structures
 706  * @q:    the request queue associated with the device
 707  *
 708  **/
 709 void blk_trace_shutdown(struct request_queue *q)
 710 {
 711         if (q->blk_trace) {
 712                 blk_trace_startstop(q, 0);
 713                 blk_trace_remove(q);
 714         }
 715 }
 716
 717 #ifdef CONFIG_BLK_CGROUP
 718 static union kernfs_node_id *
 719 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 720 {
 721         struct blk_trace *bt = q->blk_trace;
 722
 723         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 724                 return NULL;
 725
 726         if (!bio->bi_css)
 727                 return NULL;
 728         return cgroup_get_kernfs_id(bio->bi_css->cgroup);
 729 }
 730 #else
 731 static union kernfs_node_id *
 732 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 733 {
 734         return NULL;
 735 }
 736 #endif
 737
 738 static union kernfs_node_id *
 739 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 740 {
 741         if (!rq->bio)
 742                 return NULL;
 743         /* Use the first bio */
 744         return blk_trace_bio_get_cgid(q, rq->bio);
 745 }
 746
 747 /*
 748  * blktrace probes
 749  */
 750
 751 /**
 752  * blk_add_trace_rq - Add a trace for a request oriented action
 753  * @rq:         the source request
 754  * @error:      return status to log
 755  * @nr_bytes:   number of completed bytes
 756  * @what:       the action
 757  * @cgid:       the cgroup info
 758  *
 759  * Description:
 760  *     Records an action against a request. Will log the bio offset + size.
 761  *
 762  **/
 763 static void blk_add_trace_rq(struct request *rq, int error,
 764                              unsigned int nr_bytes, u32 what,
 765                              union kernfs_node_id *cgid)
 766 {
 767         struct blk_trace *bt = rq->q->blk_trace;
 768
 769         if (likely(!bt))
 770                 return;
 771
 772         if (blk_rq_is_passthrough(rq))
 773                 what |= BLK_TC_ACT(BLK_TC_PC);
 774         else
 775                 what |= BLK_TC_ACT(BLK_TC_FS);
 776
 777         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 778                         rq->cmd_flags, what, error, 0, NULL, cgid);
 779 }
 780
 781 static void blk_add_trace_rq_insert(void *ignore,
 782                                     struct request_queue *q, struct request *rq)
 783 {
 784         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 785                          blk_trace_request_get_cgid(q, rq));
 786 }
 787
 788 static void blk_add_trace_rq_issue(void *ignore,
 789                                    struct request_queue *q, struct request *rq)
 790 {
 791         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 792                          blk_trace_request_get_cgid(q, rq));
 793 }
 794
 795 static void blk_add_trace_rq_requeue(void *ignore,
 796                                      struct request_queue *q,
 797                                      struct request *rq)
 798 {
 799         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 800                          blk_trace_request_get_cgid(q, rq));
 801 }
 802
 803 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 804                         int error, unsigned int nr_bytes)
 805 {
 806         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 807                          blk_trace_request_get_cgid(rq->q, rq));
 808 }
 809
 810 /**
 811  * blk_add_trace_bio - Add a trace for a bio oriented action
 812  * @q:          queue the io is for
 813  * @bio:        the source bio
 814  * @what:       the action
 815  * @error:      error, if any
 816  *
 817  * Description:
 818  *     Records an action against a bio. Will log the bio offset + size.
 819  *
 820  **/
 821 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 822                               u32 what, int error, union kernfs_node_id *cgid)
 823 {
 824         struct blk_trace *bt = q->blk_trace;
 825
 826         if (likely(!bt))
 827                 return;
 828
 829         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 830                         bio_op(bio), bio->bi_opf, what, error, 0, NULL, cgid);
 831 }
 832
 833 static void blk_add_trace_bio_bounce(void *ignore,
 834                                      struct request_queue *q, struct bio *bio)
 835 {
 836         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0,
 837                           blk_trace_bio_get_cgid(q, bio));
 838 }
 839
 840 static void blk_add_trace_bio_complete(void *ignore,
 841                                        struct request_queue *q, struct bio *bio,
 842                                        int error)
 843 {
 844         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error,
 845                           blk_trace_bio_get_cgid(q, bio));
 846 }
 847
 848 static void blk_add_trace_bio_backmerge(void *ignore,
 849                                         struct request_queue *q,
 850                                         struct request *rq,
 851                                         struct bio *bio)
 852 {
 853         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0,
 854                          blk_trace_bio_get_cgid(q, bio));
 855 }
 856
 857 static void blk_add_trace_bio_frontmerge(void *ignore,
 858                                          struct request_queue *q,
 859                                          struct request *rq,
 860                                          struct bio *bio)
 861 {
 862         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0,
 863                           blk_trace_bio_get_cgid(q, bio));
 864 }
 865
 866 static void blk_add_trace_bio_queue(void *ignore,
 867                                     struct request_queue *q, struct bio *bio)
 868 {
 869         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0,
 870                           blk_trace_bio_get_cgid(q, bio));
 871 }
 872
 873 static void blk_add_trace_getrq(void *ignore,
 874                                 struct request_queue *q,
 875                                 struct bio *bio, int rw)
 876 {
 877         if (bio)
 878                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0,
 879                                   blk_trace_bio_get_cgid(q, bio));
 880         else {
 881                 struct blk_trace *bt = q->blk_trace;
 882
 883                 if (bt)
 884                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 885                                         NULL, NULL);
 886         }
 887 }
 888
 889
 890 static void blk_add_trace_sleeprq(void *ignore,
 891                                   struct request_queue *q,
 892                                   struct bio *bio, int rw)
 893 {
 894         if (bio)
 895                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0,
 896                                   blk_trace_bio_get_cgid(q, bio));
 897         else {
 898                 struct blk_trace *bt = q->blk_trace;
 899
 900                 if (bt)
 901                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 902                                         0, 0, NULL, NULL);
 903         }
 904 }
 905
 906 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 907 {
 908         struct blk_trace *bt = q->blk_trace;
 909
 910         if (bt)
 911                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
 912 }
 913
 914 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 915                                     unsigned int depth, bool explicit)
 916 {
 917         struct blk_trace *bt = q->blk_trace;
 918
 919         if (bt) {
 920                 __be64 rpdu = cpu_to_be64(depth);
 921                 u32 what;
 922
 923                 if (explicit)
 924                         what = BLK_TA_UNPLUG_IO;
 925                 else
 926                         what = BLK_TA_UNPLUG_TIMER;
 927
 928                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
 929         }
 930 }
 931
 932 static void blk_add_trace_split(void *ignore,
 933                                 struct request_queue *q, struct bio *bio,
 934                                 unsigned int pdu)
 935 {
 936         struct blk_trace *bt = q->blk_trace;
 937
 938         if (bt) {
 939                 __be64 rpdu = cpu_to_be64(pdu);
 940
 941                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 942                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 943                                 BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
 944                                 &rpdu, blk_trace_bio_get_cgid(q, bio));
 945         }
 946 }
 947
 948 /**
 949  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 950  * @ignore:     trace callback data parameter (not used)
 951  * @q:          queue the io is for
 952  * @bio:        the source bio
 953  * @dev:        target device
 954  * @from:       source sector
 955  *
 956  * Description:
 957  *     Device mapper or raid target sometimes need to split a bio because
 958  *     it spans a stripe (or similar). Add a trace for that action.
 959  *
 960  **/
 961 static void blk_add_trace_bio_remap(void *ignore,
 962                                     struct request_queue *q, struct bio *bio,
 963                                     dev_t dev, sector_t from)
 964 {
 965         struct blk_trace *bt = q->blk_trace;
 966         struct blk_io_trace_remap r;
 967
 968         if (likely(!bt))
 969                 return;
 970
 971         r.device_from = cpu_to_be32(dev);
 972         r.device_to   = cpu_to_be32(bio_dev(bio));
 973         r.sector_from = cpu_to_be64(from);
 974
 975         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 976                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
 977                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
 978 }
 979
 980 /**
 981  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
 982  * @ignore:     trace callback data parameter (not used)
 983  * @q:          queue the io is for
 984  * @rq:         the source request
 985  * @dev:        target device
 986  * @from:       source sector
 987  *
 988  * Description:
 989  *     Device mapper remaps request to other devices.
 990  *     Add a trace for that action.
 991  *
 992  **/
 993 static void blk_add_trace_rq_remap(void *ignore,
 994                                    struct request_queue *q,
 995                                    struct request *rq, dev_t dev,
 996                                    sector_t from)
 997 {
 998         struct blk_trace *bt = q->blk_trace;
 999         struct blk_io_trace_remap r;
1000
1001         if (likely(!bt))
1002                 return;
1003
1004         r.device_from = cpu_to_be32(dev);
1005         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1006         r.sector_from = cpu_to_be64(from);
1007
1008         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1009                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1010                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1011 }
1012
1013 /**
1014  * blk_add_driver_data - Add binary message with driver-specific data
1015  * @q:          queue the io is for
1016  * @rq:         io request
1017  * @data:       driver-specific data
1018  * @len:        length of driver-specific data
1019  *
1020  * Description:
1021  *     Some drivers might want to write driver-specific data per request.
1022  *
1023  **/
1024 void blk_add_driver_data(struct request_queue *q,
1025                          struct request *rq,
1026                          void *data, size_t len)
1027 {
1028         struct blk_trace *bt = q->blk_trace;
1029
1030         if (likely(!bt))
1031                 return;
1032
1033         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1034                                 BLK_TA_DRV_DATA, 0, len, data,
1035                                 blk_trace_request_get_cgid(q, rq));
1036 }
1037 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1038
1039 static void blk_register_tracepoints(void)
1040 {
1041         int ret;
1042
1043         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1044         WARN_ON(ret);
1045         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1046         WARN_ON(ret);
1047         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1048         WARN_ON(ret);
1049         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1050         WARN_ON(ret);
1051         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1052         WARN_ON(ret);
1053         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1054         WARN_ON(ret);
1055         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1056         WARN_ON(ret);
1057         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1058         WARN_ON(ret);
1059         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1060         WARN_ON(ret);
1061         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1062         WARN_ON(ret);
1063         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1064         WARN_ON(ret);
1065         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1066         WARN_ON(ret);
1067         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1068         WARN_ON(ret);
1069         ret = register_trace_block_split(blk_add_trace_split, NULL);
1070         WARN_ON(ret);
1071         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1072         WARN_ON(ret);
1073         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1074         WARN_ON(ret);
1075 }
1076
1077 static void blk_unregister_tracepoints(void)
1078 {
1079         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1080         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1081         unregister_trace_block_split(blk_add_trace_split, NULL);
1082         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1083         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1084         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1085         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1086         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1087         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1088         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1089         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1090         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1091         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1092         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1093         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1094         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1095
1096         tracepoint_synchronize_unregister();
1097 }
1098
1099 /*
1100  * struct blk_io_tracer formatting routines
1101  */
1102
1103 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1104 {
1105         int i = 0;
1106         int tc = t->action >> BLK_TC_SHIFT;
1107
1108         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1109                 rwbs[i++] = 'N';
1110                 goto out;
1111         }
1112
1113         if (tc & BLK_TC_FLUSH)
1114                 rwbs[i++] = 'F';
1115
1116         if (tc & BLK_TC_DISCARD)
1117                 rwbs[i++] = 'D';
1118         else if (tc & BLK_TC_WRITE)
1119                 rwbs[i++] = 'W';
1120         else if (t->bytes)
1121                 rwbs[i++] = 'R';
1122         else
1123                 rwbs[i++] = 'N';
1124
1125         if (tc & BLK_TC_FUA)
1126                 rwbs[i++] = 'F';
1127         if (tc & BLK_TC_AHEAD)
1128                 rwbs[i++] = 'A';
1129         if (tc & BLK_TC_SYNC)
1130                 rwbs[i++] = 'S';
1131         if (tc & BLK_TC_META)
1132                 rwbs[i++] = 'M';
1133 out:
1134         rwbs[i] = '\0';
1135 }
1136
1137 static inline
1138 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1139 {
1140         return (const struct blk_io_trace *)ent;
1141 }
1142
1143 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1144 {
1145         return (void *)(te_blk_io_trace(ent) + 1) +
1146                 (has_cg ? sizeof(union kernfs_node_id) : 0);
1147 }
1148
1149 static inline const void *cgid_start(const struct trace_entry *ent)
1150 {
1151         return (void *)(te_blk_io_trace(ent) + 1);
1152 }
1153
1154 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1155 {
1156         return te_blk_io_trace(ent)->pdu_len -
1157                         (has_cg ? sizeof(union kernfs_node_id) : 0);
1158 }
1159
1160 static inline u32 t_action(const struct trace_entry *ent)
1161 {
1162         return te_blk_io_trace(ent)->action;
1163 }
1164
1165 static inline u32 t_bytes(const struct trace_entry *ent)
1166 {
1167         return te_blk_io_trace(ent)->bytes;
1168 }
1169
1170 static inline u32 t_sec(const struct trace_entry *ent)
1171 {
1172         return te_blk_io_trace(ent)->bytes >> 9;
1173 }
1174
1175 static inline unsigned long long t_sector(const struct trace_entry *ent)
1176 {
1177         return te_blk_io_trace(ent)->sector;
1178 }
1179
1180 static inline __u16 t_error(const struct trace_entry *ent)
1181 {
1182         return te_blk_io_trace(ent)->error;
1183 }
1184
1185 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1186 {
1187         const __u64 *val = pdu_start(ent, has_cg);
1188         return be64_to_cpu(*val);
1189 }
1190
1191 static void get_pdu_remap(const struct trace_entry *ent,
1192                           struct blk_io_trace_remap *r, bool has_cg)
1193 {
1194         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1195         __u64 sector_from = __r->sector_from;
1196
1197         r->device_from = be32_to_cpu(__r->device_from);
1198         r->device_to   = be32_to_cpu(__r->device_to);
1199         r->sector_from = be64_to_cpu(sector_from);
1200 }
1201
1202 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1203         bool has_cg);
1204
1205 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1206         bool has_cg)
1207 {
1208         char rwbs[RWBS_LEN];
1209         unsigned long long ts  = iter->ts;
1210         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1211         unsigned secs          = (unsigned long)ts;
1212         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1213
1214         fill_rwbs(rwbs, t);
1215
1216         trace_seq_printf(&iter->seq,
1217                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1218                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1219                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1220 }
1221
1222 static void blk_log_action(struct trace_iterator *iter, const char *act,
1223         bool has_cg)
1224 {
1225         char rwbs[RWBS_LEN];
1226         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1227
1228         fill_rwbs(rwbs, t);
1229         if (has_cg) {
1230                 const union kernfs_node_id *id = cgid_start(iter->ent);
1231
1232                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1233                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1234
1235                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1236                                 sizeof(blkcg_name_buf));
1237                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1238                                  MAJOR(t->device), MINOR(t->device),
1239                                  blkcg_name_buf, act, rwbs);
1240                 } else
1241                         trace_seq_printf(&iter->seq,
1242                                  "%3d,%-3d %x,%-x %2s %3s ",
1243                                  MAJOR(t->device), MINOR(t->device),
1244                                  id->ino, id->generation, act, rwbs);
1245         } else
1246                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1247                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1248 }
1249
1250 static void blk_log_dump_pdu(struct trace_seq *s,
1251         const struct trace_entry *ent, bool has_cg)
1252 {
1253         const unsigned char *pdu_buf;
1254         int pdu_len;
1255         int i, end;
1256
1257         pdu_buf = pdu_start(ent, has_cg);
1258         pdu_len = pdu_real_len(ent, has_cg);
1259
1260         if (!pdu_len)
1261                 return;
1262
1263         /* find the last zero that needs to be printed */
1264         for (end = pdu_len - 1; end >= 0; end--)
1265                 if (pdu_buf[end])
1266                         break;
1267         end++;
1268
1269         trace_seq_putc(s, '(');
1270
1271         for (i = 0; i < pdu_len; i++) {
1272
1273                 trace_seq_printf(s, "%s%02x",
1274                                  i == 0 ? "" : " ", pdu_buf[i]);
1275
1276                 /*
1277                  * stop when the rest is just zeroes and indicate so
1278                  * with a ".." appended
1279                  */
1280                 if (i == end && end != pdu_len - 1) {
1281                         trace_seq_puts(s, " ..) ");
1282                         return;
1283                 }
1284         }
1285
1286         trace_seq_puts(s, ") ");
1287 }
1288
1289 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1290 {
1291         char cmd[TASK_COMM_LEN];
1292
1293         trace_find_cmdline(ent->pid, cmd);
1294
1295         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1296                 trace_seq_printf(s, "%u ", t_bytes(ent));
1297                 blk_log_dump_pdu(s, ent, has_cg);
1298                 trace_seq_printf(s, "[%s]\n", cmd);
1299         } else {
1300                 if (t_sec(ent))
1301                         trace_seq_printf(s, "%llu + %u [%s]\n",
1302                                                 t_sector(ent), t_sec(ent), cmd);
1303                 else
1304                         trace_seq_printf(s, "[%s]\n", cmd);
1305         }
1306 }
1307
1308 static void blk_log_with_error(struct trace_seq *s,
1309                               const struct trace_entry *ent, bool has_cg)
1310 {
1311         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1312                 blk_log_dump_pdu(s, ent, has_cg);
1313                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1314         } else {
1315                 if (t_sec(ent))
1316                         trace_seq_printf(s, "%llu + %u [%d]\n",
1317                                          t_sector(ent),
1318                                          t_sec(ent), t_error(ent));
1319                 else
1320                         trace_seq_printf(s, "%llu [%d]\n",
1321                                          t_sector(ent), t_error(ent));
1322         }
1323 }
1324
1325 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1326 {
1327         struct blk_io_trace_remap r = { .device_from = 0, };
1328
1329         get_pdu_remap(ent, &r, has_cg);
1330         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1331                          t_sector(ent), t_sec(ent),
1332                          MAJOR(r.device_from), MINOR(r.device_from),
1333                          (unsigned long long)r.sector_from);
1334 }
1335
1336 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1337 {
1338         char cmd[TASK_COMM_LEN];
1339
1340         trace_find_cmdline(ent->pid, cmd);
1341
1342         trace_seq_printf(s, "[%s]\n", cmd);
1343 }
1344
1345 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1346 {
1347         char cmd[TASK_COMM_LEN];
1348
1349         trace_find_cmdline(ent->pid, cmd);
1350
1351         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1352 }
1353
1354 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1355 {
1356         char cmd[TASK_COMM_LEN];
1357
1358         trace_find_cmdline(ent->pid, cmd);
1359
1360         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1361                          get_pdu_int(ent, has_cg), cmd);
1362 }
1363
1364 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1365                         bool has_cg)
1366 {
1367
1368         trace_seq_putmem(s, pdu_start(ent, has_cg),
1369                 pdu_real_len(ent, has_cg));
1370         trace_seq_putc(s, '\n');
1371 }
1372
1373 /*
1374  * struct tracer operations
1375  */
1376
1377 static void blk_tracer_print_header(struct seq_file *m)
1378 {
1379         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1380                 return;
1381         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1382                     "#  |     |     |           |   |   |\n");
1383 }
1384
1385 static void blk_tracer_start(struct trace_array *tr)
1386 {
1387         blk_tracer_enabled = true;
1388 }
1389
1390 static int blk_tracer_init(struct trace_array *tr)
1391 {
1392         blk_tr = tr;
1393         blk_tracer_start(tr);
1394         return 0;
1395 }
1396
1397 static void blk_tracer_stop(struct trace_array *tr)
1398 {
1399         blk_tracer_enabled = false;
1400 }
1401
1402 static void blk_tracer_reset(struct trace_array *tr)
1403 {
1404         blk_tracer_stop(tr);
1405 }
1406
1407 static const struct {
1408         const char *act[2];
1409         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1410                             bool has_cg);
1411 } what2act[] = {
1412         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1413         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1414         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1415         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1416         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1417         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1418         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1419         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1420         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1421         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1422         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1423         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1424         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1425         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1426         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1427 };
1428
1429 static enum print_line_t print_one_line(struct trace_iterator *iter,
1430                                         bool classic)
1431 {
1432         struct trace_array *tr = iter->tr;
1433         struct trace_seq *s = &iter->seq;
1434         const struct blk_io_trace *t;
1435         u16 what;
1436         bool long_act;
1437         blk_log_action_t *log_action;
1438         bool has_cg;
1439
1440         t          = te_blk_io_trace(iter->ent);
1441         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1442         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1443         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1444         has_cg     = t->action & __BLK_TA_CGROUP;
1445
1446         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1447                 log_action(iter, long_act ? "message" : "m", has_cg);
1448                 blk_log_msg(s, iter->ent, has_cg);
1449                 return trace_handle_return(s);
1450         }
1451
1452         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1453                 trace_seq_printf(s, "Unknown action %x\n", what);
1454         else {
1455                 log_action(iter, what2act[what].act[long_act], has_cg);
1456                 what2act[what].print(s, iter->ent, has_cg);
1457         }
1458
1459         return trace_handle_return(s);
1460 }
1461
1462 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1463                                                int flags, struct trace_event *event)
1464 {
1465         return print_one_line(iter, false);
1466 }
1467
1468 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1469 {
1470         struct trace_seq *s = &iter->seq;
1471         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1472         const int offset = offsetof(struct blk_io_trace, sector);
1473         struct blk_io_trace old = {
1474                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1475                 .time     = iter->ts,
1476         };
1477
1478         trace_seq_putmem(s, &old, offset);
1479         trace_seq_putmem(s, &t->sector,
1480                          sizeof(old) - offset + t->pdu_len);
1481 }
1482
1483 static enum print_line_t
1484 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1485                              struct trace_event *event)
1486 {
1487         blk_trace_synthesize_old_trace(iter);
1488
1489         return trace_handle_return(&iter->seq);
1490 }
1491
1492 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1493 {
1494         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1495                 return TRACE_TYPE_UNHANDLED;
1496
1497         return print_one_line(iter, true);
1498 }
1499
1500 static int
1501 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1502 {
1503         /* don't output context-info for blk_classic output */
1504         if (bit == TRACE_BLK_OPT_CLASSIC) {
1505                 if (set)
1506                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1507                 else
1508                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1509         }
1510         return 0;
1511 }
1512
1513 static struct tracer blk_tracer __read_mostly = {
1514         .name           = "blk",
1515         .init           = blk_tracer_init,
1516         .reset          = blk_tracer_reset,
1517         .start          = blk_tracer_start,
1518         .stop           = blk_tracer_stop,
1519         .print_header   = blk_tracer_print_header,
1520         .print_line     = blk_tracer_print_line,
1521         .flags          = &blk_tracer_flags,
1522         .set_flag       = blk_tracer_set_flag,
1523 };
1524
1525 static struct trace_event_functions trace_blk_event_funcs = {
1526         .trace          = blk_trace_event_print,
1527         .binary         = blk_trace_event_print_binary,
1528 };
1529
1530 static struct trace_event trace_blk_event = {
1531         .type           = TRACE_BLK,
1532         .funcs          = &trace_blk_event_funcs,
1533 };
1534
1535 static int __init init_blk_tracer(void)
1536 {
1537         if (!register_trace_event(&trace_blk_event)) {
1538                 pr_warn("Warning: could not register block events\n");
1539                 return 1;
1540         }
1541
1542         if (register_tracer(&blk_tracer) != 0) {
1543                 pr_warn("Warning: could not register the block tracer\n");
1544                 unregister_trace_event(&trace_blk_event);
1545                 return 1;
1546         }
1547
1548         return 0;
1549 }
1550
1551 device_initcall(init_blk_tracer);
1552
1553 static int blk_trace_remove_queue(struct request_queue *q)
1554 {
1555         struct blk_trace *bt;
1556
1557         bt = xchg(&q->blk_trace, NULL);
1558         if (bt == NULL)
1559                 return -EINVAL;
1560
1561         if (atomic_dec_and_test(&blk_probes_ref))
1562                 blk_unregister_tracepoints();
1563
1564         blk_trace_free(bt);
1565         return 0;
1566 }
1567
1568 /*
1569  * Setup everything required to start tracing
1570  */
1571 static int blk_trace_setup_queue(struct request_queue *q,
1572                                  struct block_device *bdev)
1573 {
1574         struct blk_trace *bt = NULL;
1575         int ret = -ENOMEM;
1576
1577         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1578         if (!bt)
1579                 return -ENOMEM;
1580
1581         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1582         if (!bt->msg_data)
1583                 goto free_bt;
1584
1585         bt->dev = bdev->bd_dev;
1586         bt->act_mask = (u16)-1;
1587
1588         blk_trace_setup_lba(bt, bdev);
1589
1590         ret = -EBUSY;
1591         if (cmpxchg(&q->blk_trace, NULL, bt))
1592                 goto free_bt;
1593
1594         if (atomic_inc_return(&blk_probes_ref) == 1)
1595                 blk_register_tracepoints();
1596         return 0;
1597
1598 free_bt:
1599         blk_trace_free(bt);
1600         return ret;
1601 }
1602
1603 /*
1604  * sysfs interface to enable and configure tracing
1605  */
1606
1607 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1608                                          struct device_attribute *attr,
1609                                          char *buf);
1610 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1611                                           struct device_attribute *attr,
1612                                           const char *buf, size_t count);
1613 #define BLK_TRACE_DEVICE_ATTR(_name) \
1614         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1615                     sysfs_blk_trace_attr_show, \
1616                     sysfs_blk_trace_attr_store)
1617
1618 static BLK_TRACE_DEVICE_ATTR(enable);
1619 static BLK_TRACE_DEVICE_ATTR(act_mask);
1620 static BLK_TRACE_DEVICE_ATTR(pid);
1621 static BLK_TRACE_DEVICE_ATTR(start_lba);
1622 static BLK_TRACE_DEVICE_ATTR(end_lba);
1623
1624 static struct attribute *blk_trace_attrs[] = {
1625         &dev_attr_enable.attr,
1626         &dev_attr_act_mask.attr,
1627         &dev_attr_pid.attr,
1628         &dev_attr_start_lba.attr,
1629         &dev_attr_end_lba.attr,
1630         NULL
1631 };
1632
1633 struct attribute_group blk_trace_attr_group = {
1634         .name  = "trace",
1635         .attrs = blk_trace_attrs,
1636 };
1637
1638 static const struct {
1639         int mask;
1640         const char *str;
1641 } mask_maps[] = {
1642         { BLK_TC_READ,          "read"          },
1643         { BLK_TC_WRITE,         "write"         },
1644         { BLK_TC_FLUSH,         "flush"         },
1645         { BLK_TC_SYNC,          "sync"          },
1646         { BLK_TC_QUEUE,         "queue"         },
1647         { BLK_TC_REQUEUE,       "requeue"       },
1648         { BLK_TC_ISSUE,         "issue"         },
1649         { BLK_TC_COMPLETE,      "complete"      },
1650         { BLK_TC_FS,            "fs"            },
1651         { BLK_TC_PC,            "pc"            },
1652         { BLK_TC_NOTIFY,        "notify"        },
1653         { BLK_TC_AHEAD,         "ahead"         },
1654         { BLK_TC_META,          "meta"          },
1655         { BLK_TC_DISCARD,       "discard"       },
1656         { BLK_TC_DRV_DATA,      "drv_data"      },
1657         { BLK_TC_FUA,           "fua"           },
1658 };
1659
1660 static int blk_trace_str2mask(const char *str)
1661 {
1662         int i;
1663         int mask = 0;
1664         char *buf, *s, *token;
1665
1666         buf = kstrdup(str, GFP_KERNEL);
1667         if (buf == NULL)
1668                 return -ENOMEM;
1669         s = strstrip(buf);
1670
1671         while (1) {
1672                 token = strsep(&s, ",");
1673                 if (token == NULL)
1674                         break;
1675
1676                 if (*token == '\0')
1677                         continue;
1678
1679                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1680                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1681                                 mask |= mask_maps[i].mask;
1682                                 break;
1683                         }
1684                 }
1685                 if (i == ARRAY_SIZE(mask_maps)) {
1686                         mask = -EINVAL;
1687                         break;
1688                 }
1689         }
1690         kfree(buf);
1691
1692         return mask;
1693 }
1694
1695 static ssize_t blk_trace_mask2str(char *buf, int mask)
1696 {
1697         int i;
1698         char *p = buf;
1699
1700         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1701                 if (mask & mask_maps[i].mask) {
1702                         p += sprintf(p, "%s%s",
1703                                     (p == buf) ? "" : ",", mask_maps[i].str);
1704                 }
1705         }
1706         *p++ = '\n';
1707
1708         return p - buf;
1709 }
1710
1711 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1712 {
1713         if (bdev->bd_disk == NULL)
1714                 return NULL;
1715
1716         return bdev_get_queue(bdev);
1717 }
1718
1719 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1720                                          struct device_attribute *attr,
1721                                          char *buf)
1722 {
1723         struct hd_struct *p = dev_to_part(dev);
1724         struct request_queue *q;
1725         struct block_device *bdev;
1726         ssize_t ret = -ENXIO;
1727
1728         bdev = bdget(part_devt(p));
1729         if (bdev == NULL)
1730                 goto out;
1731
1732         q = blk_trace_get_queue(bdev);
1733         if (q == NULL)
1734                 goto out_bdput;
1735
1736         mutex_lock(&q->blk_trace_mutex);
1737
1738         if (attr == &dev_attr_enable) {
1739                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1740                 goto out_unlock_bdev;
1741         }
1742
1743         if (q->blk_trace == NULL)
1744                 ret = sprintf(buf, "disabled\n");
1745         else if (attr == &dev_attr_act_mask)
1746                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1747         else if (attr == &dev_attr_pid)
1748                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1749         else if (attr == &dev_attr_start_lba)
1750                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1751         else if (attr == &dev_attr_end_lba)
1752                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1753
1754 out_unlock_bdev:
1755         mutex_unlock(&q->blk_trace_mutex);
1756 out_bdput:
1757         bdput(bdev);
1758 out:
1759         return ret;
1760 }
1761
1762 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1763                                           struct device_attribute *attr,
1764                                           const char *buf, size_t count)
1765 {
1766         struct block_device *bdev;
1767         struct request_queue *q;
1768         struct hd_struct *p;
1769         u64 value;
1770         ssize_t ret = -EINVAL;
1771
1772         if (count == 0)
1773                 goto out;
1774
1775         if (attr == &dev_attr_act_mask) {
1776                 if (kstrtoull(buf, 0, &value)) {
1777                         /* Assume it is a list of trace category names */
1778                         ret = blk_trace_str2mask(buf);
1779                         if (ret < 0)
1780                                 goto out;
1781                         value = ret;
1782                 }
1783         } else if (kstrtoull(buf, 0, &value))
1784                 goto out;
1785
1786         ret = -ENXIO;
1787
1788         p = dev_to_part(dev);
1789         bdev = bdget(part_devt(p));
1790         if (bdev == NULL)
1791                 goto out;
1792
1793         q = blk_trace_get_queue(bdev);
1794         if (q == NULL)
1795                 goto out_bdput;
1796
1797         mutex_lock(&q->blk_trace_mutex);
1798
1799         if (attr == &dev_attr_enable) {
1800                 if (value)
1801                         ret = blk_trace_setup_queue(q, bdev);
1802                 else
1803                         ret = blk_trace_remove_queue(q);
1804                 goto out_unlock_bdev;
1805         }
1806
1807         ret = 0;
1808         if (q->blk_trace == NULL)
1809                 ret = blk_trace_setup_queue(q, bdev);
1810
1811         if (ret == 0) {
1812                 if (attr == &dev_attr_act_mask)
1813                         q->blk_trace->act_mask = value;
1814                 else if (attr == &dev_attr_pid)
1815                         q->blk_trace->pid = value;
1816                 else if (attr == &dev_attr_start_lba)
1817                         q->blk_trace->start_lba = value;
1818                 else if (attr == &dev_attr_end_lba)
1819                         q->blk_trace->end_lba = value;
1820         }
1821
1822 out_unlock_bdev:
1823         mutex_unlock(&q->blk_trace_mutex);
1824 out_bdput:
1825         bdput(bdev);
1826 out:
1827         return ret ? ret : count;
1828 }
1829
1830 int blk_trace_init_sysfs(struct device *dev)
1831 {
1832         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1833 }
1834
1835 void blk_trace_remove_sysfs(struct device *dev)
1836 {
1837         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1838 }
1839
1840 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1841
1842 #ifdef CONFIG_EVENT_TRACING
1843
1844 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1845 {
1846         int i = 0;
1847
1848         if (op & REQ_PREFLUSH)
1849                 rwbs[i++] = 'F';
1850
1851         switch (op & REQ_OP_MASK) {
1852         case REQ_OP_WRITE:
1853         case REQ_OP_WRITE_SAME:
1854                 rwbs[i++] = 'W';
1855                 break;
1856         case REQ_OP_DISCARD:
1857                 rwbs[i++] = 'D';
1858                 break;
1859         case REQ_OP_SECURE_ERASE:
1860                 rwbs[i++] = 'D';
1861                 rwbs[i++] = 'E';
1862                 break;
1863         case REQ_OP_FLUSH:
1864                 rwbs[i++] = 'F';
1865                 break;
1866         case REQ_OP_READ:
1867                 rwbs[i++] = 'R';
1868                 break;
1869         default:
1870                 rwbs[i++] = 'N';
1871         }
1872
1873         if (op & REQ_FUA)
1874                 rwbs[i++] = 'F';
1875         if (op & REQ_RAHEAD)
1876                 rwbs[i++] = 'A';
1877         if (op & REQ_SYNC)
1878                 rwbs[i++] = 'S';
1879         if (op & REQ_META)
1880                 rwbs[i++] = 'M';
1881
1882         rwbs[i] = '\0';
1883 }
1884 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1885
1886 #endif /* CONFIG_EVENT_TRACING */
1887