kernel/trace/blktrace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4  *
   5  */
   6 #include <linux/kernel.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/blktrace_api.h>
   9 #include <linux/percpu.h>
  10 #include <linux/init.h>
  11 #include <linux/mutex.h>
  12 #include <linux/slab.h>
  13 #include <linux/debugfs.h>
  14 #include <linux/export.h>
  15 #include <linux/time.h>
  16 #include <linux/uaccess.h>
  17 #include <linux/list.h>
  18 #include <linux/blk-cgroup.h>
  19
  20 #include "../../block/blk.h"
  21
  22 #include <trace/events/block.h>
  23
  24 #include "trace_output.h"
  25
  26 #ifdef CONFIG_BLK_DEV_IO_TRACE
  27
  28 static unsigned int blktrace_seq __read_mostly = 1;
  29
  30 static struct trace_array *blk_tr;
  31 static bool blk_tracer_enabled __read_mostly;
  32
  33 static LIST_HEAD(running_trace_list);
  34 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  35
  36 /* Select an alternative, minimalistic output than the original one */
  37 #define TRACE_BLK_OPT_CLASSIC   0x1
  38 #define TRACE_BLK_OPT_CGROUP    0x2
  39 #define TRACE_BLK_OPT_CGNAME    0x4
  40
  41 static struct tracer_opt blk_tracer_opts[] = {
  42         /* Default disable the minimalistic output */
  43         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  44 #ifdef CONFIG_BLK_CGROUP
  45         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  46         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  47 #endif
  48         { }
  49 };
  50
  51 static struct tracer_flags blk_tracer_flags = {
  52         .val  = 0,
  53         .opts = blk_tracer_opts,
  54 };
  55
  56 /* Global reference count of probes */
  57 static DEFINE_MUTEX(blk_probe_mutex);
  58 static int blk_probes_ref;
  59
  60 static void blk_register_tracepoints(void);
  61 static void blk_unregister_tracepoints(void);
  62
  63 /*
  64  * Send out a notify message.
  65  */
  66 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  67                        const void *data, size_t len,
  68                        union kernfs_node_id *cgid)
  69 {
  70         struct blk_io_trace *t;
  71         struct ring_buffer_event *event = NULL;
  72         struct ring_buffer *buffer = NULL;
  73         int pc = 0;
  74         int cpu = smp_processor_id();
  75         bool blk_tracer = blk_tracer_enabled;
  76         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  77
  78         if (blk_tracer) {
  79                 buffer = blk_tr->trace_buffer.buffer;
  80                 pc = preempt_count();
  81                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  82                                                   sizeof(*t) + len + cgid_len,
  83                                                   0, pc);
  84                 if (!event)
  85                         return;
  86                 t = ring_buffer_event_data(event);
  87                 goto record_it;
  88         }
  89
  90         if (!bt->rchan)
  91                 return;
  92
  93         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  94         if (t) {
  95                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  96                 t->time = ktime_to_ns(ktime_get());
  97 record_it:
  98                 t->device = bt->dev;
  99                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 100                 t->pid = pid;
 101                 t->cpu = cpu;
 102                 t->pdu_len = len + cgid_len;
 103                 if (cgid)
 104                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 105                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 106
 107                 if (blk_tracer)
 108                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 109         }
 110 }
 111
 112 /*
 113  * Send out a notify for this process, if we haven't done so since a trace
 114  * started
 115  */
 116 static void trace_note_tsk(struct task_struct *tsk)
 117 {
 118         unsigned long flags;
 119         struct blk_trace *bt;
 120
 121         tsk->btrace_seq = blktrace_seq;
 122         spin_lock_irqsave(&running_trace_lock, flags);
 123         list_for_each_entry(bt, &running_trace_list, running_list) {
 124                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 125                            sizeof(tsk->comm), NULL);
 126         }
 127         spin_unlock_irqrestore(&running_trace_lock, flags);
 128 }
 129
 130 static void trace_note_time(struct blk_trace *bt)
 131 {
 132         struct timespec64 now;
 133         unsigned long flags;
 134         u32 words[2];
 135
 136         /* need to check user space to see if this breaks in y2038 or y2106 */
 137         ktime_get_real_ts64(&now);
 138         words[0] = (u32)now.tv_sec;
 139         words[1] = now.tv_nsec;
 140
 141         local_irq_save(flags);
 142         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 143         local_irq_restore(flags);
 144 }
 145
 146 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 147         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         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 172                 blkcg = NULL;
 173 #ifdef CONFIG_BLK_CGROUP
 174         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 175                 blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 176 #else
 177         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 178 #endif
 179         local_irq_restore(flags);
 180 }
 181 EXPORT_SYMBOL_GPL(__trace_note_message);
 182
 183 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 184                          pid_t pid)
 185 {
 186         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 187                 return 1;
 188         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 189                 return 1;
 190         if (bt->pid && pid != bt->pid)
 191                 return 1;
 192
 193         return 0;
 194 }
 195
 196 /*
 197  * Data direction bit lookup
 198  */
 199 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 200                                  BLK_TC_ACT(BLK_TC_WRITE) };
 201
 202 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 203 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 204
 205 /* The ilog2() calls fall out because they're constant */
 206 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 207           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 208
 209 /*
 210  * The worker for the various blk_add_trace*() types. Fills out a
 211  * blk_io_trace structure and places it in a per-cpu subbuffer.
 212  */
 213 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 214                      int op, int op_flags, u32 what, int error, int pdu_len,
 215                      void *pdu_data, union kernfs_node_id *cgid)
 216 {
 217         struct task_struct *tsk = current;
 218         struct ring_buffer_event *event = NULL;
 219         struct ring_buffer *buffer = NULL;
 220         struct blk_io_trace *t;
 221         unsigned long flags = 0;
 222         unsigned long *sequence;
 223         pid_t pid;
 224         int cpu, pc = 0;
 225         bool blk_tracer = blk_tracer_enabled;
 226         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 227
 228         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 229                 return;
 230
 231         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 232         what |= MASK_TC_BIT(op_flags, SYNC);
 233         what |= MASK_TC_BIT(op_flags, RAHEAD);
 234         what |= MASK_TC_BIT(op_flags, META);
 235         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 236         what |= MASK_TC_BIT(op_flags, FUA);
 237         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 238                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 239         if (op == REQ_OP_FLUSH)
 240                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 241         if (cgid)
 242                 what |= __BLK_TA_CGROUP;
 243
 244         pid = tsk->pid;
 245         if (act_log_check(bt, what, sector, pid))
 246                 return;
 247         cpu = raw_smp_processor_id();
 248
 249         if (blk_tracer) {
 250                 tracing_record_cmdline(current);
 251
 252                 buffer = blk_tr->trace_buffer.buffer;
 253                 pc = preempt_count();
 254                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 255                                                   sizeof(*t) + pdu_len + cgid_len,
 256                                                   0, pc);
 257                 if (!event)
 258                         return;
 259                 t = ring_buffer_event_data(event);
 260                 goto record_it;
 261         }
 262
 263         if (unlikely(tsk->btrace_seq != blktrace_seq))
 264                 trace_note_tsk(tsk);
 265
 266         /*
 267          * A word about the locking here - we disable interrupts to reserve
 268          * some space in the relay per-cpu buffer, to prevent an irq
 269          * from coming in and stepping on our toes.
 270          */
 271         local_irq_save(flags);
 272         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 273         if (t) {
 274                 sequence = per_cpu_ptr(bt->sequence, cpu);
 275
 276                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 277                 t->sequence = ++(*sequence);
 278                 t->time = ktime_to_ns(ktime_get());
 279 record_it:
 280                 /*
 281                  * These two are not needed in ftrace as they are in the
 282                  * generic trace_entry, filled by tracing_generic_entry_update,
 283                  * but for the trace_event->bin() synthesizer benefit we do it
 284                  * here too.
 285                  */
 286                 t->cpu = cpu;
 287                 t->pid = pid;
 288
 289                 t->sector = sector;
 290                 t->bytes = bytes;
 291                 t->action = what;
 292                 t->device = bt->dev;
 293                 t->error = error;
 294                 t->pdu_len = pdu_len + cgid_len;
 295
 296                 if (cgid_len)
 297                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 298                 if (pdu_len)
 299                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 300
 301                 if (blk_tracer) {
 302                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 303                         return;
 304                 }
 305         }
 306
 307         local_irq_restore(flags);
 308 }
 309
 310 static void blk_trace_free(struct blk_trace *bt)
 311 {
 312         debugfs_remove(bt->msg_file);
 313         debugfs_remove(bt->dropped_file);
 314         relay_close(bt->rchan);
 315         debugfs_remove(bt->dir);
 316         free_percpu(bt->sequence);
 317         free_percpu(bt->msg_data);
 318         kfree(bt);
 319 }
 320
 321 static void get_probe_ref(void)
 322 {
 323         mutex_lock(&blk_probe_mutex);
 324         if (++blk_probes_ref == 1)
 325                 blk_register_tracepoints();
 326         mutex_unlock(&blk_probe_mutex);
 327 }
 328
 329 static void put_probe_ref(void)
 330 {
 331         mutex_lock(&blk_probe_mutex);
 332         if (!--blk_probes_ref)
 333                 blk_unregister_tracepoints();
 334         mutex_unlock(&blk_probe_mutex);
 335 }
 336
 337 static void blk_trace_cleanup(struct blk_trace *bt)
 338 {
 339         blk_trace_free(bt);
 340         put_probe_ref();
 341 }
 342
 343 static int __blk_trace_remove(struct request_queue *q)
 344 {
 345         struct blk_trace *bt;
 346
 347         bt = xchg(&q->blk_trace, NULL);
 348         if (!bt)
 349                 return -EINVAL;
 350
 351         if (bt->trace_state != Blktrace_running)
 352                 blk_trace_cleanup(bt);
 353
 354         return 0;
 355 }
 356
 357 int blk_trace_remove(struct request_queue *q)
 358 {
 359         int ret;
 360
 361         mutex_lock(&q->blk_trace_mutex);
 362         ret = __blk_trace_remove(q);
 363         mutex_unlock(&q->blk_trace_mutex);
 364
 365         return ret;
 366 }
 367 EXPORT_SYMBOL_GPL(blk_trace_remove);
 368
 369 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 370                                 size_t count, loff_t *ppos)
 371 {
 372         struct blk_trace *bt = filp->private_data;
 373         char buf[16];
 374
 375         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 376
 377         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 378 }
 379
 380 static const struct file_operations blk_dropped_fops = {
 381         .owner =        THIS_MODULE,
 382         .open =         simple_open,
 383         .read =         blk_dropped_read,
 384         .llseek =       default_llseek,
 385 };
 386
 387 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 388                                 size_t count, loff_t *ppos)
 389 {
 390         char *msg;
 391         struct blk_trace *bt;
 392
 393         if (count >= BLK_TN_MAX_MSG)
 394                 return -EINVAL;
 395
 396         msg = memdup_user_nul(buffer, count);
 397         if (IS_ERR(msg))
 398                 return PTR_ERR(msg);
 399
 400         bt = filp->private_data;
 401         __trace_note_message(bt, NULL, "%s", msg);
 402         kfree(msg);
 403
 404         return count;
 405 }
 406
 407 static const struct file_operations blk_msg_fops = {
 408         .owner =        THIS_MODULE,
 409         .open =         simple_open,
 410         .write =        blk_msg_write,
 411         .llseek =       noop_llseek,
 412 };
 413
 414 /*
 415  * Keep track of how many times we encountered a full subbuffer, to aid
 416  * the user space app in telling how many lost events there were.
 417  */
 418 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 419                                      void *prev_subbuf, size_t prev_padding)
 420 {
 421         struct blk_trace *bt;
 422
 423         if (!relay_buf_full(buf))
 424                 return 1;
 425
 426         bt = buf->chan->private_data;
 427         atomic_inc(&bt->dropped);
 428         return 0;
 429 }
 430
 431 static int blk_remove_buf_file_callback(struct dentry *dentry)
 432 {
 433         debugfs_remove(dentry);
 434
 435         return 0;
 436 }
 437
 438 static struct dentry *blk_create_buf_file_callback(const char *filename,
 439                                                    struct dentry *parent,
 440                                                    umode_t mode,
 441                                                    struct rchan_buf *buf,
 442                                                    int *is_global)
 443 {
 444         return debugfs_create_file(filename, mode, parent, buf,
 445                                         &relay_file_operations);
 446 }
 447
 448 static struct rchan_callbacks blk_relay_callbacks = {
 449         .subbuf_start           = blk_subbuf_start_callback,
 450         .create_buf_file        = blk_create_buf_file_callback,
 451         .remove_buf_file        = blk_remove_buf_file_callback,
 452 };
 453
 454 static void blk_trace_setup_lba(struct blk_trace *bt,
 455                                 struct block_device *bdev)
 456 {
 457         struct hd_struct *part = NULL;
 458
 459         if (bdev)
 460                 part = bdev->bd_part;
 461
 462         if (part) {
 463                 bt->start_lba = part->start_sect;
 464                 bt->end_lba = part->start_sect + part->nr_sects;
 465         } else {
 466                 bt->start_lba = 0;
 467                 bt->end_lba = -1ULL;
 468         }
 469 }
 470
 471 /*
 472  * Setup everything required to start tracing
 473  */
 474 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 475                               struct block_device *bdev,
 476                               struct blk_user_trace_setup *buts)
 477 {
 478         struct blk_trace *bt = NULL;
 479         struct dentry *dir = NULL;
 480         int ret;
 481
 482         if (!buts->buf_size || !buts->buf_nr)
 483                 return -EINVAL;
 484
 485         if (!blk_debugfs_root)
 486                 return -ENOENT;
 487
 488         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 489         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 490
 491         /*
 492          * some device names have larger paths - convert the slashes
 493          * to underscores for this to work as expected
 494          */
 495         strreplace(buts->name, '/', '_');
 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         dir = debugfs_lookup(buts->name, blk_debugfs_root);
 513         if (!dir)
 514                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 515
 516         bt->dev = dev;
 517         atomic_set(&bt->dropped, 0);
 518         INIT_LIST_HEAD(&bt->running_list);
 519
 520         ret = -EIO;
 521         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 522                                                &blk_dropped_fops);
 523
 524         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 525
 526         bt->rchan = relay_open("trace", dir, buts->buf_size,
 527                                 buts->buf_nr, &blk_relay_callbacks, bt);
 528         if (!bt->rchan)
 529                 goto err;
 530
 531         bt->act_mask = buts->act_mask;
 532         if (!bt->act_mask)
 533                 bt->act_mask = (u16) -1;
 534
 535         blk_trace_setup_lba(bt, bdev);
 536
 537         /* overwrite with user settings */
 538         if (buts->start_lba)
 539                 bt->start_lba = buts->start_lba;
 540         if (buts->end_lba)
 541                 bt->end_lba = buts->end_lba;
 542
 543         bt->pid = buts->pid;
 544         bt->trace_state = Blktrace_setup;
 545
 546         ret = -EBUSY;
 547         if (cmpxchg(&q->blk_trace, NULL, bt))
 548                 goto err;
 549
 550         get_probe_ref();
 551
 552         ret = 0;
 553 err:
 554         if (dir && !bt->dir)
 555                 dput(dir);
 556         if (ret)
 557                 blk_trace_free(bt);
 558         return ret;
 559 }
 560
 561 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 562                              struct block_device *bdev, char __user *arg)
 563 {
 564         struct blk_user_trace_setup buts;
 565         int ret;
 566
 567         ret = copy_from_user(&buts, arg, sizeof(buts));
 568         if (ret)
 569                 return -EFAULT;
 570
 571         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 572         if (ret)
 573                 return ret;
 574
 575         if (copy_to_user(arg, &buts, sizeof(buts))) {
 576                 __blk_trace_remove(q);
 577                 return -EFAULT;
 578         }
 579         return 0;
 580 }
 581
 582 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 583                     struct block_device *bdev,
 584                     char __user *arg)
 585 {
 586         int ret;
 587
 588         mutex_lock(&q->blk_trace_mutex);
 589         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 590         mutex_unlock(&q->blk_trace_mutex);
 591
 592         return ret;
 593 }
 594 EXPORT_SYMBOL_GPL(blk_trace_setup);
 595
 596 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 597 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 598                                   dev_t dev, struct block_device *bdev,
 599                                   char __user *arg)
 600 {
 601         struct blk_user_trace_setup buts;
 602         struct compat_blk_user_trace_setup cbuts;
 603         int ret;
 604
 605         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 606                 return -EFAULT;
 607
 608         buts = (struct blk_user_trace_setup) {
 609                 .act_mask = cbuts.act_mask,
 610                 .buf_size = cbuts.buf_size,
 611                 .buf_nr = cbuts.buf_nr,
 612                 .start_lba = cbuts.start_lba,
 613                 .end_lba = cbuts.end_lba,
 614                 .pid = cbuts.pid,
 615         };
 616
 617         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 618         if (ret)
 619                 return ret;
 620
 621         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 622                 __blk_trace_remove(q);
 623                 return -EFAULT;
 624         }
 625
 626         return 0;
 627 }
 628 #endif
 629
 630 static int __blk_trace_startstop(struct request_queue *q, int start)
 631 {
 632         int ret;
 633         struct blk_trace *bt = q->blk_trace;
 634
 635         if (bt == NULL)
 636                 return -EINVAL;
 637
 638         /*
 639          * For starting a trace, we can transition from a setup or stopped
 640          * trace. For stopping a trace, the state must be running
 641          */
 642         ret = -EINVAL;
 643         if (start) {
 644                 if (bt->trace_state == Blktrace_setup ||
 645                     bt->trace_state == Blktrace_stopped) {
 646                         blktrace_seq++;
 647                         smp_mb();
 648                         bt->trace_state = Blktrace_running;
 649                         spin_lock_irq(&running_trace_lock);
 650                         list_add(&bt->running_list, &running_trace_list);
 651                         spin_unlock_irq(&running_trace_lock);
 652
 653                         trace_note_time(bt);
 654                         ret = 0;
 655                 }
 656         } else {
 657                 if (bt->trace_state == Blktrace_running) {
 658                         bt->trace_state = Blktrace_stopped;
 659                         spin_lock_irq(&running_trace_lock);
 660                         list_del_init(&bt->running_list);
 661                         spin_unlock_irq(&running_trace_lock);
 662                         relay_flush(bt->rchan);
 663                         ret = 0;
 664                 }
 665         }
 666
 667         return ret;
 668 }
 669
 670 int blk_trace_startstop(struct request_queue *q, int start)
 671 {
 672         int ret;
 673
 674         mutex_lock(&q->blk_trace_mutex);
 675         ret = __blk_trace_startstop(q, start);
 676         mutex_unlock(&q->blk_trace_mutex);
 677
 678         return ret;
 679 }
 680 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 681
 682 /*
 683  * When reading or writing the blktrace sysfs files, the references to the
 684  * opened sysfs or device files should prevent the underlying block device
 685  * from being removed. So no further delete protection is really needed.
 686  */
 687
 688 /**
 689  * blk_trace_ioctl: - handle the ioctls associated with tracing
 690  * @bdev:       the block device
 691  * @cmd:        the ioctl cmd
 692  * @arg:        the argument data, if any
 693  *
 694  **/
 695 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 696 {
 697         struct request_queue *q;
 698         int ret, start = 0;
 699         char b[BDEVNAME_SIZE];
 700
 701         q = bdev_get_queue(bdev);
 702         if (!q)
 703                 return -ENXIO;
 704
 705         mutex_lock(&q->blk_trace_mutex);
 706
 707         switch (cmd) {
 708         case BLKTRACESETUP:
 709                 bdevname(bdev, b);
 710                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 711                 break;
 712 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 713         case BLKTRACESETUP32:
 714                 bdevname(bdev, b);
 715                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 716                 break;
 717 #endif
 718         case BLKTRACESTART:
 719                 start = 1;
 720                 /* fall through */
 721         case BLKTRACESTOP:
 722                 ret = __blk_trace_startstop(q, start);
 723                 break;
 724         case BLKTRACETEARDOWN:
 725                 ret = __blk_trace_remove(q);
 726                 break;
 727         default:
 728                 ret = -ENOTTY;
 729                 break;
 730         }
 731
 732         mutex_unlock(&q->blk_trace_mutex);
 733         return ret;
 734 }
 735
 736 /**
 737  * blk_trace_shutdown: - stop and cleanup trace structures
 738  * @q:    the request queue associated with the device
 739  *
 740  **/
 741 void blk_trace_shutdown(struct request_queue *q)
 742 {
 743         mutex_lock(&q->blk_trace_mutex);
 744
 745         if (q->blk_trace) {
 746                 __blk_trace_startstop(q, 0);
 747                 __blk_trace_remove(q);
 748         }
 749
 750         mutex_unlock(&q->blk_trace_mutex);
 751 }
 752
 753 #ifdef CONFIG_BLK_CGROUP
 754 static union kernfs_node_id *
 755 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 756 {
 757         struct blk_trace *bt = q->blk_trace;
 758
 759         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 760                 return NULL;
 761
 762         if (!bio->bi_blkg)
 763                 return NULL;
 764         return cgroup_get_kernfs_id(bio_blkcg(bio)->css.cgroup);
 765 }
 766 #else
 767 static union kernfs_node_id *
 768 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 769 {
 770         return NULL;
 771 }
 772 #endif
 773
 774 static union kernfs_node_id *
 775 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 776 {
 777         if (!rq->bio)
 778                 return NULL;
 779         /* Use the first bio */
 780         return blk_trace_bio_get_cgid(q, rq->bio);
 781 }
 782
 783 /*
 784  * blktrace probes
 785  */
 786
 787 /**
 788  * blk_add_trace_rq - Add a trace for a request oriented action
 789  * @rq:         the source request
 790  * @error:      return status to log
 791  * @nr_bytes:   number of completed bytes
 792  * @what:       the action
 793  * @cgid:       the cgroup info
 794  *
 795  * Description:
 796  *     Records an action against a request. Will log the bio offset + size.
 797  *
 798  **/
 799 static void blk_add_trace_rq(struct request *rq, int error,
 800                              unsigned int nr_bytes, u32 what,
 801                              union kernfs_node_id *cgid)
 802 {
 803         struct blk_trace *bt = rq->q->blk_trace;
 804
 805         if (likely(!bt))
 806                 return;
 807
 808         if (blk_rq_is_passthrough(rq))
 809                 what |= BLK_TC_ACT(BLK_TC_PC);
 810         else
 811                 what |= BLK_TC_ACT(BLK_TC_FS);
 812
 813         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 814                         rq->cmd_flags, what, error, 0, NULL, cgid);
 815 }
 816
 817 static void blk_add_trace_rq_insert(void *ignore,
 818                                     struct request_queue *q, struct request *rq)
 819 {
 820         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 821                          blk_trace_request_get_cgid(q, rq));
 822 }
 823
 824 static void blk_add_trace_rq_issue(void *ignore,
 825                                    struct request_queue *q, struct request *rq)
 826 {
 827         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 828                          blk_trace_request_get_cgid(q, rq));
 829 }
 830
 831 static void blk_add_trace_rq_requeue(void *ignore,
 832                                      struct request_queue *q,
 833                                      struct request *rq)
 834 {
 835         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 836                          blk_trace_request_get_cgid(q, rq));
 837 }
 838
 839 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 840                         int error, unsigned int nr_bytes)
 841 {
 842         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 843                          blk_trace_request_get_cgid(rq->q, rq));
 844 }
 845
 846 /**
 847  * blk_add_trace_bio - Add a trace for a bio oriented action
 848  * @q:          queue the io is for
 849  * @bio:        the source bio
 850  * @what:       the action
 851  * @error:      error, if any
 852  *
 853  * Description:
 854  *     Records an action against a bio. Will log the bio offset + size.
 855  *
 856  **/
 857 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 858                               u32 what, int error)
 859 {
 860         struct blk_trace *bt = q->blk_trace;
 861
 862         if (likely(!bt))
 863                 return;
 864
 865         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 866                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 867                         blk_trace_bio_get_cgid(q, bio));
 868 }
 869
 870 static void blk_add_trace_bio_bounce(void *ignore,
 871                                      struct request_queue *q, struct bio *bio)
 872 {
 873         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 874 }
 875
 876 static void blk_add_trace_bio_complete(void *ignore,
 877                                        struct request_queue *q, struct bio *bio,
 878                                        int error)
 879 {
 880         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 881 }
 882
 883 static void blk_add_trace_bio_backmerge(void *ignore,
 884                                         struct request_queue *q,
 885                                         struct request *rq,
 886                                         struct bio *bio)
 887 {
 888         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 889 }
 890
 891 static void blk_add_trace_bio_frontmerge(void *ignore,
 892                                          struct request_queue *q,
 893                                          struct request *rq,
 894                                          struct bio *bio)
 895 {
 896         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 897 }
 898
 899 static void blk_add_trace_bio_queue(void *ignore,
 900                                     struct request_queue *q, struct bio *bio)
 901 {
 902         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 903 }
 904
 905 static void blk_add_trace_getrq(void *ignore,
 906                                 struct request_queue *q,
 907                                 struct bio *bio, int rw)
 908 {
 909         if (bio)
 910                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 911         else {
 912                 struct blk_trace *bt = q->blk_trace;
 913
 914                 if (bt)
 915                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 916                                         NULL, NULL);
 917         }
 918 }
 919
 920
 921 static void blk_add_trace_sleeprq(void *ignore,
 922                                   struct request_queue *q,
 923                                   struct bio *bio, int rw)
 924 {
 925         if (bio)
 926                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 927         else {
 928                 struct blk_trace *bt = q->blk_trace;
 929
 930                 if (bt)
 931                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 932                                         0, 0, NULL, NULL);
 933         }
 934 }
 935
 936 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 937 {
 938         struct blk_trace *bt = q->blk_trace;
 939
 940         if (bt)
 941                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
 942 }
 943
 944 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 945                                     unsigned int depth, bool explicit)
 946 {
 947         struct blk_trace *bt = q->blk_trace;
 948
 949         if (bt) {
 950                 __be64 rpdu = cpu_to_be64(depth);
 951                 u32 what;
 952
 953                 if (explicit)
 954                         what = BLK_TA_UNPLUG_IO;
 955                 else
 956                         what = BLK_TA_UNPLUG_TIMER;
 957
 958                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
 959         }
 960 }
 961
 962 static void blk_add_trace_split(void *ignore,
 963                                 struct request_queue *q, struct bio *bio,
 964                                 unsigned int pdu)
 965 {
 966         struct blk_trace *bt = q->blk_trace;
 967
 968         if (bt) {
 969                 __be64 rpdu = cpu_to_be64(pdu);
 970
 971                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 972                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 973                                 BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
 974                                 &rpdu, blk_trace_bio_get_cgid(q, bio));
 975         }
 976 }
 977
 978 /**
 979  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 980  * @ignore:     trace callback data parameter (not used)
 981  * @q:          queue the io is for
 982  * @bio:        the source bio
 983  * @dev:        target device
 984  * @from:       source sector
 985  *
 986  * Description:
 987  *     Device mapper or raid target sometimes need to split a bio because
 988  *     it spans a stripe (or similar). Add a trace for that action.
 989  *
 990  **/
 991 static void blk_add_trace_bio_remap(void *ignore,
 992                                     struct request_queue *q, struct bio *bio,
 993                                     dev_t dev, sector_t from)
 994 {
 995         struct blk_trace *bt = q->blk_trace;
 996         struct blk_io_trace_remap r;
 997
 998         if (likely(!bt))
 999                 return;
1000
1001         r.device_from = cpu_to_be32(dev);
1002         r.device_to   = cpu_to_be32(bio_dev(bio));
1003         r.sector_from = cpu_to_be64(from);
1004
1005         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1006                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1007                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1008 }
1009
1010 /**
1011  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1012  * @ignore:     trace callback data parameter (not used)
1013  * @q:          queue the io is for
1014  * @rq:         the source request
1015  * @dev:        target device
1016  * @from:       source sector
1017  *
1018  * Description:
1019  *     Device mapper remaps request to other devices.
1020  *     Add a trace for that action.
1021  *
1022  **/
1023 static void blk_add_trace_rq_remap(void *ignore,
1024                                    struct request_queue *q,
1025                                    struct request *rq, dev_t dev,
1026                                    sector_t from)
1027 {
1028         struct blk_trace *bt = q->blk_trace;
1029         struct blk_io_trace_remap r;
1030
1031         if (likely(!bt))
1032                 return;
1033
1034         r.device_from = cpu_to_be32(dev);
1035         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1036         r.sector_from = cpu_to_be64(from);
1037
1038         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1039                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1040                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1041 }
1042
1043 /**
1044  * blk_add_driver_data - Add binary message with driver-specific data
1045  * @q:          queue the io is for
1046  * @rq:         io request
1047  * @data:       driver-specific data
1048  * @len:        length of driver-specific data
1049  *
1050  * Description:
1051  *     Some drivers might want to write driver-specific data per request.
1052  *
1053  **/
1054 void blk_add_driver_data(struct request_queue *q,
1055                          struct request *rq,
1056                          void *data, size_t len)
1057 {
1058         struct blk_trace *bt = q->blk_trace;
1059
1060         if (likely(!bt))
1061                 return;
1062
1063         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1064                                 BLK_TA_DRV_DATA, 0, len, data,
1065                                 blk_trace_request_get_cgid(q, rq));
1066 }
1067 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1068
1069 static void blk_register_tracepoints(void)
1070 {
1071         int ret;
1072
1073         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1074         WARN_ON(ret);
1075         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1076         WARN_ON(ret);
1077         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1078         WARN_ON(ret);
1079         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1080         WARN_ON(ret);
1081         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1082         WARN_ON(ret);
1083         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1084         WARN_ON(ret);
1085         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1086         WARN_ON(ret);
1087         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1088         WARN_ON(ret);
1089         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1090         WARN_ON(ret);
1091         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1092         WARN_ON(ret);
1093         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1094         WARN_ON(ret);
1095         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1096         WARN_ON(ret);
1097         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1098         WARN_ON(ret);
1099         ret = register_trace_block_split(blk_add_trace_split, NULL);
1100         WARN_ON(ret);
1101         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1102         WARN_ON(ret);
1103         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1104         WARN_ON(ret);
1105 }
1106
1107 static void blk_unregister_tracepoints(void)
1108 {
1109         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1110         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1111         unregister_trace_block_split(blk_add_trace_split, NULL);
1112         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1113         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1114         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1115         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1116         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1117         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1118         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1119         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1120         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1121         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1122         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1123         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1124         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1125
1126         tracepoint_synchronize_unregister();
1127 }
1128
1129 /*
1130  * struct blk_io_tracer formatting routines
1131  */
1132
1133 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1134 {
1135         int i = 0;
1136         int tc = t->action >> BLK_TC_SHIFT;
1137
1138         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1139                 rwbs[i++] = 'N';
1140                 goto out;
1141         }
1142
1143         if (tc & BLK_TC_FLUSH)
1144                 rwbs[i++] = 'F';
1145
1146         if (tc & BLK_TC_DISCARD)
1147                 rwbs[i++] = 'D';
1148         else if (tc & BLK_TC_WRITE)
1149                 rwbs[i++] = 'W';
1150         else if (t->bytes)
1151                 rwbs[i++] = 'R';
1152         else
1153                 rwbs[i++] = 'N';
1154
1155         if (tc & BLK_TC_FUA)
1156                 rwbs[i++] = 'F';
1157         if (tc & BLK_TC_AHEAD)
1158                 rwbs[i++] = 'A';
1159         if (tc & BLK_TC_SYNC)
1160                 rwbs[i++] = 'S';
1161         if (tc & BLK_TC_META)
1162                 rwbs[i++] = 'M';
1163 out:
1164         rwbs[i] = '\0';
1165 }
1166
1167 static inline
1168 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1169 {
1170         return (const struct blk_io_trace *)ent;
1171 }
1172
1173 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1174 {
1175         return (void *)(te_blk_io_trace(ent) + 1) +
1176                 (has_cg ? sizeof(union kernfs_node_id) : 0);
1177 }
1178
1179 static inline const void *cgid_start(const struct trace_entry *ent)
1180 {
1181         return (void *)(te_blk_io_trace(ent) + 1);
1182 }
1183
1184 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1185 {
1186         return te_blk_io_trace(ent)->pdu_len -
1187                         (has_cg ? sizeof(union kernfs_node_id) : 0);
1188 }
1189
1190 static inline u32 t_action(const struct trace_entry *ent)
1191 {
1192         return te_blk_io_trace(ent)->action;
1193 }
1194
1195 static inline u32 t_bytes(const struct trace_entry *ent)
1196 {
1197         return te_blk_io_trace(ent)->bytes;
1198 }
1199
1200 static inline u32 t_sec(const struct trace_entry *ent)
1201 {
1202         return te_blk_io_trace(ent)->bytes >> 9;
1203 }
1204
1205 static inline unsigned long long t_sector(const struct trace_entry *ent)
1206 {
1207         return te_blk_io_trace(ent)->sector;
1208 }
1209
1210 static inline __u16 t_error(const struct trace_entry *ent)
1211 {
1212         return te_blk_io_trace(ent)->error;
1213 }
1214
1215 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1216 {
1217         const __u64 *val = pdu_start(ent, has_cg);
1218         return be64_to_cpu(*val);
1219 }
1220
1221 static void get_pdu_remap(const struct trace_entry *ent,
1222                           struct blk_io_trace_remap *r, bool has_cg)
1223 {
1224         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1225         __u64 sector_from = __r->sector_from;
1226
1227         r->device_from = be32_to_cpu(__r->device_from);
1228         r->device_to   = be32_to_cpu(__r->device_to);
1229         r->sector_from = be64_to_cpu(sector_from);
1230 }
1231
1232 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1233         bool has_cg);
1234
1235 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1236         bool has_cg)
1237 {
1238         char rwbs[RWBS_LEN];
1239         unsigned long long ts  = iter->ts;
1240         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1241         unsigned secs          = (unsigned long)ts;
1242         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1243
1244         fill_rwbs(rwbs, t);
1245
1246         trace_seq_printf(&iter->seq,
1247                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1248                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1249                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1250 }
1251
1252 static void blk_log_action(struct trace_iterator *iter, const char *act,
1253         bool has_cg)
1254 {
1255         char rwbs[RWBS_LEN];
1256         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1257
1258         fill_rwbs(rwbs, t);
1259         if (has_cg) {
1260                 const union kernfs_node_id *id = cgid_start(iter->ent);
1261
1262                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1263                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1264
1265                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1266                                 sizeof(blkcg_name_buf));
1267                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1268                                  MAJOR(t->device), MINOR(t->device),
1269                                  blkcg_name_buf, act, rwbs);
1270                 } else
1271                         trace_seq_printf(&iter->seq,
1272                                  "%3d,%-3d %x,%-x %2s %3s ",
1273                                  MAJOR(t->device), MINOR(t->device),
1274                                  id->ino, id->generation, act, rwbs);
1275         } else
1276                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1277                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1278 }
1279
1280 static void blk_log_dump_pdu(struct trace_seq *s,
1281         const struct trace_entry *ent, bool has_cg)
1282 {
1283         const unsigned char *pdu_buf;
1284         int pdu_len;
1285         int i, end;
1286
1287         pdu_buf = pdu_start(ent, has_cg);
1288         pdu_len = pdu_real_len(ent, has_cg);
1289
1290         if (!pdu_len)
1291                 return;
1292
1293         /* find the last zero that needs to be printed */
1294         for (end = pdu_len - 1; end >= 0; end--)
1295                 if (pdu_buf[end])
1296                         break;
1297         end++;
1298
1299         trace_seq_putc(s, '(');
1300
1301         for (i = 0; i < pdu_len; i++) {
1302
1303                 trace_seq_printf(s, "%s%02x",
1304                                  i == 0 ? "" : " ", pdu_buf[i]);
1305
1306                 /*
1307                  * stop when the rest is just zeroes and indicate so
1308                  * with a ".." appended
1309                  */
1310                 if (i == end && end != pdu_len - 1) {
1311                         trace_seq_puts(s, " ..) ");
1312                         return;
1313                 }
1314         }
1315
1316         trace_seq_puts(s, ") ");
1317 }
1318
1319 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1320 {
1321         char cmd[TASK_COMM_LEN];
1322
1323         trace_find_cmdline(ent->pid, cmd);
1324
1325         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1326                 trace_seq_printf(s, "%u ", t_bytes(ent));
1327                 blk_log_dump_pdu(s, ent, has_cg);
1328                 trace_seq_printf(s, "[%s]\n", cmd);
1329         } else {
1330                 if (t_sec(ent))
1331                         trace_seq_printf(s, "%llu + %u [%s]\n",
1332                                                 t_sector(ent), t_sec(ent), cmd);
1333                 else
1334                         trace_seq_printf(s, "[%s]\n", cmd);
1335         }
1336 }
1337
1338 static void blk_log_with_error(struct trace_seq *s,
1339                               const struct trace_entry *ent, bool has_cg)
1340 {
1341         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1342                 blk_log_dump_pdu(s, ent, has_cg);
1343                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1344         } else {
1345                 if (t_sec(ent))
1346                         trace_seq_printf(s, "%llu + %u [%d]\n",
1347                                          t_sector(ent),
1348                                          t_sec(ent), t_error(ent));
1349                 else
1350                         trace_seq_printf(s, "%llu [%d]\n",
1351                                          t_sector(ent), t_error(ent));
1352         }
1353 }
1354
1355 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1356 {
1357         struct blk_io_trace_remap r = { .device_from = 0, };
1358
1359         get_pdu_remap(ent, &r, has_cg);
1360         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1361                          t_sector(ent), t_sec(ent),
1362                          MAJOR(r.device_from), MINOR(r.device_from),
1363                          (unsigned long long)r.sector_from);
1364 }
1365
1366 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1367 {
1368         char cmd[TASK_COMM_LEN];
1369
1370         trace_find_cmdline(ent->pid, cmd);
1371
1372         trace_seq_printf(s, "[%s]\n", cmd);
1373 }
1374
1375 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1376 {
1377         char cmd[TASK_COMM_LEN];
1378
1379         trace_find_cmdline(ent->pid, cmd);
1380
1381         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1382 }
1383
1384 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1385 {
1386         char cmd[TASK_COMM_LEN];
1387
1388         trace_find_cmdline(ent->pid, cmd);
1389
1390         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1391                          get_pdu_int(ent, has_cg), cmd);
1392 }
1393
1394 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1395                         bool has_cg)
1396 {
1397
1398         trace_seq_putmem(s, pdu_start(ent, has_cg),
1399                 pdu_real_len(ent, has_cg));
1400         trace_seq_putc(s, '\n');
1401 }
1402
1403 /*
1404  * struct tracer operations
1405  */
1406
1407 static void blk_tracer_print_header(struct seq_file *m)
1408 {
1409         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1410                 return;
1411         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1412                     "#  |     |     |           |   |   |\n");
1413 }
1414
1415 static void blk_tracer_start(struct trace_array *tr)
1416 {
1417         blk_tracer_enabled = true;
1418 }
1419
1420 static int blk_tracer_init(struct trace_array *tr)
1421 {
1422         blk_tr = tr;
1423         blk_tracer_start(tr);
1424         return 0;
1425 }
1426
1427 static void blk_tracer_stop(struct trace_array *tr)
1428 {
1429         blk_tracer_enabled = false;
1430 }
1431
1432 static void blk_tracer_reset(struct trace_array *tr)
1433 {
1434         blk_tracer_stop(tr);
1435 }
1436
1437 static const struct {
1438         const char *act[2];
1439         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1440                             bool has_cg);
1441 } what2act[] = {
1442         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1443         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1444         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1445         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1446         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1447         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1448         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1449         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1450         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1451         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1452         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1453         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1454         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1455         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1456         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1457 };
1458
1459 static enum print_line_t print_one_line(struct trace_iterator *iter,
1460                                         bool classic)
1461 {
1462         struct trace_array *tr = iter->tr;
1463         struct trace_seq *s = &iter->seq;
1464         const struct blk_io_trace *t;
1465         u16 what;
1466         bool long_act;
1467         blk_log_action_t *log_action;
1468         bool has_cg;
1469
1470         t          = te_blk_io_trace(iter->ent);
1471         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1472         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1473         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1474         has_cg     = t->action & __BLK_TA_CGROUP;
1475
1476         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1477                 log_action(iter, long_act ? "message" : "m", has_cg);
1478                 blk_log_msg(s, iter->ent, has_cg);
1479                 return trace_handle_return(s);
1480         }
1481
1482         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1483                 trace_seq_printf(s, "Unknown action %x\n", what);
1484         else {
1485                 log_action(iter, what2act[what].act[long_act], has_cg);
1486                 what2act[what].print(s, iter->ent, has_cg);
1487         }
1488
1489         return trace_handle_return(s);
1490 }
1491
1492 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1493                                                int flags, struct trace_event *event)
1494 {
1495         return print_one_line(iter, false);
1496 }
1497
1498 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1499 {
1500         struct trace_seq *s = &iter->seq;
1501         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1502         const int offset = offsetof(struct blk_io_trace, sector);
1503         struct blk_io_trace old = {
1504                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1505                 .time     = iter->ts,
1506         };
1507
1508         trace_seq_putmem(s, &old, offset);
1509         trace_seq_putmem(s, &t->sector,
1510                          sizeof(old) - offset + t->pdu_len);
1511 }
1512
1513 static enum print_line_t
1514 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1515                              struct trace_event *event)
1516 {
1517         blk_trace_synthesize_old_trace(iter);
1518
1519         return trace_handle_return(&iter->seq);
1520 }
1521
1522 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1523 {
1524         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1525                 return TRACE_TYPE_UNHANDLED;
1526
1527         return print_one_line(iter, true);
1528 }
1529
1530 static int
1531 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1532 {
1533         /* don't output context-info for blk_classic output */
1534         if (bit == TRACE_BLK_OPT_CLASSIC) {
1535                 if (set)
1536                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1537                 else
1538                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1539         }
1540         return 0;
1541 }
1542
1543 static struct tracer blk_tracer __read_mostly = {
1544         .name           = "blk",
1545         .init           = blk_tracer_init,
1546         .reset          = blk_tracer_reset,
1547         .start          = blk_tracer_start,
1548         .stop           = blk_tracer_stop,
1549         .print_header   = blk_tracer_print_header,
1550         .print_line     = blk_tracer_print_line,
1551         .flags          = &blk_tracer_flags,
1552         .set_flag       = blk_tracer_set_flag,
1553 };
1554
1555 static struct trace_event_functions trace_blk_event_funcs = {
1556         .trace          = blk_trace_event_print,
1557         .binary         = blk_trace_event_print_binary,
1558 };
1559
1560 static struct trace_event trace_blk_event = {
1561         .type           = TRACE_BLK,
1562         .funcs          = &trace_blk_event_funcs,
1563 };
1564
1565 static int __init init_blk_tracer(void)
1566 {
1567         if (!register_trace_event(&trace_blk_event)) {
1568                 pr_warn("Warning: could not register block events\n");
1569                 return 1;
1570         }
1571
1572         if (register_tracer(&blk_tracer) != 0) {
1573                 pr_warn("Warning: could not register the block tracer\n");
1574                 unregister_trace_event(&trace_blk_event);
1575                 return 1;
1576         }
1577
1578         return 0;
1579 }
1580
1581 device_initcall(init_blk_tracer);
1582
1583 static int blk_trace_remove_queue(struct request_queue *q)
1584 {
1585         struct blk_trace *bt;
1586
1587         bt = xchg(&q->blk_trace, NULL);
1588         if (bt == NULL)
1589                 return -EINVAL;
1590
1591         put_probe_ref();
1592         blk_trace_free(bt);
1593         return 0;
1594 }
1595
1596 /*
1597  * Setup everything required to start tracing
1598  */
1599 static int blk_trace_setup_queue(struct request_queue *q,
1600                                  struct block_device *bdev)
1601 {
1602         struct blk_trace *bt = NULL;
1603         int ret = -ENOMEM;
1604
1605         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1606         if (!bt)
1607                 return -ENOMEM;
1608
1609         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1610         if (!bt->msg_data)
1611                 goto free_bt;
1612
1613         bt->dev = bdev->bd_dev;
1614         bt->act_mask = (u16)-1;
1615
1616         blk_trace_setup_lba(bt, bdev);
1617
1618         ret = -EBUSY;
1619         if (cmpxchg(&q->blk_trace, NULL, bt))
1620                 goto free_bt;
1621
1622         get_probe_ref();
1623         return 0;
1624
1625 free_bt:
1626         blk_trace_free(bt);
1627         return ret;
1628 }
1629
1630 /*
1631  * sysfs interface to enable and configure tracing
1632  */
1633
1634 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1635                                          struct device_attribute *attr,
1636                                          char *buf);
1637 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1638                                           struct device_attribute *attr,
1639                                           const char *buf, size_t count);
1640 #define BLK_TRACE_DEVICE_ATTR(_name) \
1641         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1642                     sysfs_blk_trace_attr_show, \
1643                     sysfs_blk_trace_attr_store)
1644
1645 static BLK_TRACE_DEVICE_ATTR(enable);
1646 static BLK_TRACE_DEVICE_ATTR(act_mask);
1647 static BLK_TRACE_DEVICE_ATTR(pid);
1648 static BLK_TRACE_DEVICE_ATTR(start_lba);
1649 static BLK_TRACE_DEVICE_ATTR(end_lba);
1650
1651 static struct attribute *blk_trace_attrs[] = {
1652         &dev_attr_enable.attr,
1653         &dev_attr_act_mask.attr,
1654         &dev_attr_pid.attr,
1655         &dev_attr_start_lba.attr,
1656         &dev_attr_end_lba.attr,
1657         NULL
1658 };
1659
1660 struct attribute_group blk_trace_attr_group = {
1661         .name  = "trace",
1662         .attrs = blk_trace_attrs,
1663 };
1664
1665 static const struct {
1666         int mask;
1667         const char *str;
1668 } mask_maps[] = {
1669         { BLK_TC_READ,          "read"          },
1670         { BLK_TC_WRITE,         "write"         },
1671         { BLK_TC_FLUSH,         "flush"         },
1672         { BLK_TC_SYNC,          "sync"          },
1673         { BLK_TC_QUEUE,         "queue"         },
1674         { BLK_TC_REQUEUE,       "requeue"       },
1675         { BLK_TC_ISSUE,         "issue"         },
1676         { BLK_TC_COMPLETE,      "complete"      },
1677         { BLK_TC_FS,            "fs"            },
1678         { BLK_TC_PC,            "pc"            },
1679         { BLK_TC_NOTIFY,        "notify"        },
1680         { BLK_TC_AHEAD,         "ahead"         },
1681         { BLK_TC_META,          "meta"          },
1682         { BLK_TC_DISCARD,       "discard"       },
1683         { BLK_TC_DRV_DATA,      "drv_data"      },
1684         { BLK_TC_FUA,           "fua"           },
1685 };
1686
1687 static int blk_trace_str2mask(const char *str)
1688 {
1689         int i;
1690         int mask = 0;
1691         char *buf, *s, *token;
1692
1693         buf = kstrdup(str, GFP_KERNEL);
1694         if (buf == NULL)
1695                 return -ENOMEM;
1696         s = strstrip(buf);
1697
1698         while (1) {
1699                 token = strsep(&s, ",");
1700                 if (token == NULL)
1701                         break;
1702
1703                 if (*token == '\0')
1704                         continue;
1705
1706                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1707                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1708                                 mask |= mask_maps[i].mask;
1709                                 break;
1710                         }
1711                 }
1712                 if (i == ARRAY_SIZE(mask_maps)) {
1713                         mask = -EINVAL;
1714                         break;
1715                 }
1716         }
1717         kfree(buf);
1718
1719         return mask;
1720 }
1721
1722 static ssize_t blk_trace_mask2str(char *buf, int mask)
1723 {
1724         int i;
1725         char *p = buf;
1726
1727         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1728                 if (mask & mask_maps[i].mask) {
1729                         p += sprintf(p, "%s%s",
1730                                     (p == buf) ? "" : ",", mask_maps[i].str);
1731                 }
1732         }
1733         *p++ = '\n';
1734
1735         return p - buf;
1736 }
1737
1738 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1739 {
1740         if (bdev->bd_disk == NULL)
1741                 return NULL;
1742
1743         return bdev_get_queue(bdev);
1744 }
1745
1746 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1747                                          struct device_attribute *attr,
1748                                          char *buf)
1749 {
1750         struct hd_struct *p = dev_to_part(dev);
1751         struct request_queue *q;
1752         struct block_device *bdev;
1753         ssize_t ret = -ENXIO;
1754
1755         bdev = bdget(part_devt(p));
1756         if (bdev == NULL)
1757                 goto out;
1758
1759         q = blk_trace_get_queue(bdev);
1760         if (q == NULL)
1761                 goto out_bdput;
1762
1763         mutex_lock(&q->blk_trace_mutex);
1764
1765         if (attr == &dev_attr_enable) {
1766                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1767                 goto out_unlock_bdev;
1768         }
1769
1770         if (q->blk_trace == NULL)
1771                 ret = sprintf(buf, "disabled\n");
1772         else if (attr == &dev_attr_act_mask)
1773                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1774         else if (attr == &dev_attr_pid)
1775                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1776         else if (attr == &dev_attr_start_lba)
1777                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1778         else if (attr == &dev_attr_end_lba)
1779                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1780
1781 out_unlock_bdev:
1782         mutex_unlock(&q->blk_trace_mutex);
1783 out_bdput:
1784         bdput(bdev);
1785 out:
1786         return ret;
1787 }
1788
1789 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1790                                           struct device_attribute *attr,
1791                                           const char *buf, size_t count)
1792 {
1793         struct block_device *bdev;
1794         struct request_queue *q;
1795         struct hd_struct *p;
1796         u64 value;
1797         ssize_t ret = -EINVAL;
1798
1799         if (count == 0)
1800                 goto out;
1801
1802         if (attr == &dev_attr_act_mask) {
1803                 if (kstrtoull(buf, 0, &value)) {
1804                         /* Assume it is a list of trace category names */
1805                         ret = blk_trace_str2mask(buf);
1806                         if (ret < 0)
1807                                 goto out;
1808                         value = ret;
1809                 }
1810         } else if (kstrtoull(buf, 0, &value))
1811                 goto out;
1812
1813         ret = -ENXIO;
1814
1815         p = dev_to_part(dev);
1816         bdev = bdget(part_devt(p));
1817         if (bdev == NULL)
1818                 goto out;
1819
1820         q = blk_trace_get_queue(bdev);
1821         if (q == NULL)
1822                 goto out_bdput;
1823
1824         mutex_lock(&q->blk_trace_mutex);
1825
1826         if (attr == &dev_attr_enable) {
1827                 if (!!value == !!q->blk_trace) {
1828                         ret = 0;
1829                         goto out_unlock_bdev;
1830                 }
1831                 if (value)
1832                         ret = blk_trace_setup_queue(q, bdev);
1833                 else
1834                         ret = blk_trace_remove_queue(q);
1835                 goto out_unlock_bdev;
1836         }
1837
1838         ret = 0;
1839         if (q->blk_trace == NULL)
1840                 ret = blk_trace_setup_queue(q, bdev);
1841
1842         if (ret == 0) {
1843                 if (attr == &dev_attr_act_mask)
1844                         q->blk_trace->act_mask = value;
1845                 else if (attr == &dev_attr_pid)
1846                         q->blk_trace->pid = value;
1847                 else if (attr == &dev_attr_start_lba)
1848                         q->blk_trace->start_lba = value;
1849                 else if (attr == &dev_attr_end_lba)
1850                         q->blk_trace->end_lba = value;
1851         }
1852
1853 out_unlock_bdev:
1854         mutex_unlock(&q->blk_trace_mutex);
1855 out_bdput:
1856         bdput(bdev);
1857 out:
1858         return ret ? ret : count;
1859 }
1860
1861 int blk_trace_init_sysfs(struct device *dev)
1862 {
1863         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1864 }
1865
1866 void blk_trace_remove_sysfs(struct device *dev)
1867 {
1868         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1869 }
1870
1871 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1872
1873 #ifdef CONFIG_EVENT_TRACING
1874
1875 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1876 {
1877         int i = 0;
1878
1879         if (op & REQ_PREFLUSH)
1880                 rwbs[i++] = 'F';
1881
1882         switch (op & REQ_OP_MASK) {
1883         case REQ_OP_WRITE:
1884         case REQ_OP_WRITE_SAME:
1885                 rwbs[i++] = 'W';
1886                 break;
1887         case REQ_OP_DISCARD:
1888                 rwbs[i++] = 'D';
1889                 break;
1890         case REQ_OP_SECURE_ERASE:
1891                 rwbs[i++] = 'D';
1892                 rwbs[i++] = 'E';
1893                 break;
1894         case REQ_OP_FLUSH:
1895                 rwbs[i++] = 'F';
1896                 break;
1897         case REQ_OP_READ:
1898                 rwbs[i++] = 'R';
1899                 break;
1900         default:
1901                 rwbs[i++] = 'N';
1902         }
1903
1904         if (op & REQ_FUA)
1905                 rwbs[i++] = 'F';
1906         if (op & REQ_RAHEAD)
1907                 rwbs[i++] = 'A';
1908         if (op & REQ_SYNC)
1909                 rwbs[i++] = 'S';
1910         if (op & REQ_META)
1911                 rwbs[i++] = 'M';
1912
1913         rwbs[i] = '\0';
1914 }
1915 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1916
1917 #endif /* CONFIG_EVENT_TRACING */
1918