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, u64 cgid)
  68 {
  69         struct blk_io_trace *t;
  70         struct ring_buffer_event *event = NULL;
  71         struct trace_buffer *buffer = NULL;
  72         int pc = 0;
  73         int cpu = smp_processor_id();
  74         bool blk_tracer = blk_tracer_enabled;
  75         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
  76
  77         if (blk_tracer) {
  78                 buffer = blk_tr->array_buffer.buffer;
  79                 pc = preempt_count();
  80                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  81                                                   sizeof(*t) + len + cgid_len,
  82                                                   0, pc);
  83                 if (!event)
  84                         return;
  85                 t = ring_buffer_event_data(event);
  86                 goto record_it;
  87         }
  88
  89         if (!bt->rchan)
  90                 return;
  91
  92         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  93         if (t) {
  94                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  95                 t->time = ktime_to_ns(ktime_get());
  96 record_it:
  97                 t->device = bt->dev;
  98                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
  99                 t->pid = pid;
 100                 t->cpu = cpu;
 101                 t->pdu_len = len + cgid_len;
 102                 if (cgid_len)
 103                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 104                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 105
 106                 if (blk_tracer)
 107                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 108         }
 109 }
 110
 111 /*
 112  * Send out a notify for this process, if we haven't done so since a trace
 113  * started
 114  */
 115 static void trace_note_tsk(struct task_struct *tsk)
 116 {
 117         unsigned long flags;
 118         struct blk_trace *bt;
 119
 120         tsk->btrace_seq = blktrace_seq;
 121         spin_lock_irqsave(&running_trace_lock, flags);
 122         list_for_each_entry(bt, &running_trace_list, running_list) {
 123                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 124                            sizeof(tsk->comm), 0);
 125         }
 126         spin_unlock_irqrestore(&running_trace_lock, flags);
 127 }
 128
 129 static void trace_note_time(struct blk_trace *bt)
 130 {
 131         struct timespec64 now;
 132         unsigned long flags;
 133         u32 words[2];
 134
 135         /* need to check user space to see if this breaks in y2038 or y2106 */
 136         ktime_get_real_ts64(&now);
 137         words[0] = (u32)now.tv_sec;
 138         words[1] = now.tv_nsec;
 139
 140         local_irq_save(flags);
 141         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
 142         local_irq_restore(flags);
 143 }
 144
 145 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 146         const char *fmt, ...)
 147 {
 148         int n;
 149         va_list args;
 150         unsigned long flags;
 151         char *buf;
 152
 153         if (unlikely(bt->trace_state != Blktrace_running &&
 154                      !blk_tracer_enabled))
 155                 return;
 156
 157         /*
 158          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 159          * message to the trace.
 160          */
 161         if (!(bt->act_mask & BLK_TC_NOTIFY))
 162                 return;
 163
 164         local_irq_save(flags);
 165         buf = this_cpu_ptr(bt->msg_data);
 166         va_start(args, fmt);
 167         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 168         va_end(args);
 169
 170         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 171                 blkcg = NULL;
 172 #ifdef CONFIG_BLK_CGROUP
 173         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 174                    blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
 175 #else
 176         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, 0);
 177 #endif
 178         local_irq_restore(flags);
 179 }
 180 EXPORT_SYMBOL_GPL(__trace_note_message);
 181
 182 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 183                          pid_t pid)
 184 {
 185         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 186                 return 1;
 187         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 188                 return 1;
 189         if (bt->pid && pid != bt->pid)
 190                 return 1;
 191
 192         return 0;
 193 }
 194
 195 /*
 196  * Data direction bit lookup
 197  */
 198 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 199                                  BLK_TC_ACT(BLK_TC_WRITE) };
 200
 201 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 202 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 203
 204 /* The ilog2() calls fall out because they're constant */
 205 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 206           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 207
 208 /*
 209  * The worker for the various blk_add_trace*() types. Fills out a
 210  * blk_io_trace structure and places it in a per-cpu subbuffer.
 211  */
 212 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 213                      int op, int op_flags, u32 what, int error, int pdu_len,
 214                      void *pdu_data, u64 cgid)
 215 {
 216         struct task_struct *tsk = current;
 217         struct ring_buffer_event *event = NULL;
 218         struct trace_buffer *buffer = NULL;
 219         struct blk_io_trace *t;
 220         unsigned long flags = 0;
 221         unsigned long *sequence;
 222         pid_t pid;
 223         int cpu, pc = 0;
 224         bool blk_tracer = blk_tracer_enabled;
 225         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
 226
 227         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 228                 return;
 229
 230         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 231         what |= MASK_TC_BIT(op_flags, SYNC);
 232         what |= MASK_TC_BIT(op_flags, RAHEAD);
 233         what |= MASK_TC_BIT(op_flags, META);
 234         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 235         what |= MASK_TC_BIT(op_flags, FUA);
 236         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 237                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 238         if (op == REQ_OP_FLUSH)
 239                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 240         if (cgid)
 241                 what |= __BLK_TA_CGROUP;
 242
 243         pid = tsk->pid;
 244         if (act_log_check(bt, what, sector, pid))
 245                 return;
 246         cpu = raw_smp_processor_id();
 247
 248         if (blk_tracer) {
 249                 tracing_record_cmdline(current);
 250
 251                 buffer = blk_tr->array_buffer.buffer;
 252                 pc = preempt_count();
 253                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 254                                                   sizeof(*t) + pdu_len + cgid_len,
 255                                                   0, pc);
 256                 if (!event)
 257                         return;
 258                 t = ring_buffer_event_data(event);
 259                 goto record_it;
 260         }
 261
 262         if (unlikely(tsk->btrace_seq != blktrace_seq))
 263                 trace_note_tsk(tsk);
 264
 265         /*
 266          * A word about the locking here - we disable interrupts to reserve
 267          * some space in the relay per-cpu buffer, to prevent an irq
 268          * from coming in and stepping on our toes.
 269          */
 270         local_irq_save(flags);
 271         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 272         if (t) {
 273                 sequence = per_cpu_ptr(bt->sequence, cpu);
 274
 275                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 276                 t->sequence = ++(*sequence);
 277                 t->time = ktime_to_ns(ktime_get());
 278 record_it:
 279                 /*
 280                  * These two are not needed in ftrace as they are in the
 281                  * generic trace_entry, filled by tracing_generic_entry_update,
 282                  * but for the trace_event->bin() synthesizer benefit we do it
 283                  * here too.
 284                  */
 285                 t->cpu = cpu;
 286                 t->pid = pid;
 287
 288                 t->sector = sector;
 289                 t->bytes = bytes;
 290                 t->action = what;
 291                 t->device = bt->dev;
 292                 t->error = error;
 293                 t->pdu_len = pdu_len + cgid_len;
 294
 295                 if (cgid_len)
 296                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 297                 if (pdu_len)
 298                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 299
 300                 if (blk_tracer) {
 301                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 302                         return;
 303                 }
 304         }
 305
 306         local_irq_restore(flags);
 307 }
 308
 309 static void blk_trace_free(struct blk_trace *bt)
 310 {
 311         debugfs_remove(bt->msg_file);
 312         debugfs_remove(bt->dropped_file);
 313         relay_close(bt->rchan);
 314         debugfs_remove(bt->dir);
 315         free_percpu(bt->sequence);
 316         free_percpu(bt->msg_data);
 317         kfree(bt);
 318 }
 319
 320 static void get_probe_ref(void)
 321 {
 322         mutex_lock(&blk_probe_mutex);
 323         if (++blk_probes_ref == 1)
 324                 blk_register_tracepoints();
 325         mutex_unlock(&blk_probe_mutex);
 326 }
 327
 328 static void put_probe_ref(void)
 329 {
 330         mutex_lock(&blk_probe_mutex);
 331         if (!--blk_probes_ref)
 332                 blk_unregister_tracepoints();
 333         mutex_unlock(&blk_probe_mutex);
 334 }
 335
 336 static void blk_trace_cleanup(struct blk_trace *bt)
 337 {
 338         blk_trace_free(bt);
 339         put_probe_ref();
 340 }
 341
 342 static int __blk_trace_remove(struct request_queue *q)
 343 {
 344         struct blk_trace *bt;
 345
 346         bt = xchg(&q->blk_trace, NULL);
 347         if (!bt)
 348                 return -EINVAL;
 349
 350         if (bt->trace_state != Blktrace_running)
 351                 blk_trace_cleanup(bt);
 352
 353         return 0;
 354 }
 355
 356 int blk_trace_remove(struct request_queue *q)
 357 {
 358         int ret;
 359
 360         mutex_lock(&q->blk_trace_mutex);
 361         ret = __blk_trace_remove(q);
 362         mutex_unlock(&q->blk_trace_mutex);
 363
 364         return ret;
 365 }
 366 EXPORT_SYMBOL_GPL(blk_trace_remove);
 367
 368 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 369                                 size_t count, loff_t *ppos)
 370 {
 371         struct blk_trace *bt = filp->private_data;
 372         char buf[16];
 373
 374         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 375
 376         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 377 }
 378
 379 static const struct file_operations blk_dropped_fops = {
 380         .owner =        THIS_MODULE,
 381         .open =         simple_open,
 382         .read =         blk_dropped_read,
 383         .llseek =       default_llseek,
 384 };
 385
 386 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 387                                 size_t count, loff_t *ppos)
 388 {
 389         char *msg;
 390         struct blk_trace *bt;
 391
 392         if (count >= BLK_TN_MAX_MSG)
 393                 return -EINVAL;
 394
 395         msg = memdup_user_nul(buffer, count);
 396         if (IS_ERR(msg))
 397                 return PTR_ERR(msg);
 398
 399         bt = filp->private_data;
 400         __trace_note_message(bt, NULL, "%s", msg);
 401         kfree(msg);
 402
 403         return count;
 404 }
 405
 406 static const struct file_operations blk_msg_fops = {
 407         .owner =        THIS_MODULE,
 408         .open =         simple_open,
 409         .write =        blk_msg_write,
 410         .llseek =       noop_llseek,
 411 };
 412
 413 /*
 414  * Keep track of how many times we encountered a full subbuffer, to aid
 415  * the user space app in telling how many lost events there were.
 416  */
 417 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 418                                      void *prev_subbuf, size_t prev_padding)
 419 {
 420         struct blk_trace *bt;
 421
 422         if (!relay_buf_full(buf))
 423                 return 1;
 424
 425         bt = buf->chan->private_data;
 426         atomic_inc(&bt->dropped);
 427         return 0;
 428 }
 429
 430 static int blk_remove_buf_file_callback(struct dentry *dentry)
 431 {
 432         debugfs_remove(dentry);
 433
 434         return 0;
 435 }
 436
 437 static struct dentry *blk_create_buf_file_callback(const char *filename,
 438                                                    struct dentry *parent,
 439                                                    umode_t mode,
 440                                                    struct rchan_buf *buf,
 441                                                    int *is_global)
 442 {
 443         return debugfs_create_file(filename, mode, parent, buf,
 444                                         &relay_file_operations);
 445 }
 446
 447 static struct rchan_callbacks blk_relay_callbacks = {
 448         .subbuf_start           = blk_subbuf_start_callback,
 449         .create_buf_file        = blk_create_buf_file_callback,
 450         .remove_buf_file        = blk_remove_buf_file_callback,
 451 };
 452
 453 static void blk_trace_setup_lba(struct blk_trace *bt,
 454                                 struct block_device *bdev)
 455 {
 456         struct hd_struct *part = NULL;
 457
 458         if (bdev)
 459                 part = bdev->bd_part;
 460
 461         if (part) {
 462                 bt->start_lba = part->start_sect;
 463                 bt->end_lba = part->start_sect + part->nr_sects;
 464         } else {
 465                 bt->start_lba = 0;
 466                 bt->end_lba = -1ULL;
 467         }
 468 }
 469
 470 /*
 471  * Setup everything required to start tracing
 472  */
 473 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 474                               struct block_device *bdev,
 475                               struct blk_user_trace_setup *buts)
 476 {
 477         struct blk_trace *bt = NULL;
 478         struct dentry *dir = NULL;
 479         int ret;
 480
 481         if (!buts->buf_size || !buts->buf_nr)
 482                 return -EINVAL;
 483
 484         if (!blk_debugfs_root)
 485                 return -ENOENT;
 486
 487         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 488         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 489
 490         /*
 491          * some device names have larger paths - convert the slashes
 492          * to underscores for this to work as expected
 493          */
 494         strreplace(buts->name, '/', '_');
 495
 496         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 497         if (!bt)
 498                 return -ENOMEM;
 499
 500         ret = -ENOMEM;
 501         bt->sequence = alloc_percpu(unsigned long);
 502         if (!bt->sequence)
 503                 goto err;
 504
 505         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 506         if (!bt->msg_data)
 507                 goto err;
 508
 509         ret = -ENOENT;
 510
 511         dir = debugfs_lookup(buts->name, blk_debugfs_root);
 512         if (!dir)
 513                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 514
 515         bt->dev = dev;
 516         atomic_set(&bt->dropped, 0);
 517         INIT_LIST_HEAD(&bt->running_list);
 518
 519         ret = -EIO;
 520         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 521                                                &blk_dropped_fops);
 522
 523         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 524
 525         bt->rchan = relay_open("trace", dir, buts->buf_size,
 526                                 buts->buf_nr, &blk_relay_callbacks, bt);
 527         if (!bt->rchan)
 528                 goto err;
 529
 530         bt->act_mask = buts->act_mask;
 531         if (!bt->act_mask)
 532                 bt->act_mask = (u16) -1;
 533
 534         blk_trace_setup_lba(bt, bdev);
 535
 536         /* overwrite with user settings */
 537         if (buts->start_lba)
 538                 bt->start_lba = buts->start_lba;
 539         if (buts->end_lba)
 540                 bt->end_lba = buts->end_lba;
 541
 542         bt->pid = buts->pid;
 543         bt->trace_state = Blktrace_setup;
 544
 545         ret = -EBUSY;
 546         if (cmpxchg(&q->blk_trace, NULL, bt))
 547                 goto err;
 548
 549         get_probe_ref();
 550
 551         ret = 0;
 552 err:
 553         if (dir && !bt->dir)
 554                 dput(dir);
 555         if (ret)
 556                 blk_trace_free(bt);
 557         return ret;
 558 }
 559
 560 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 561                              struct block_device *bdev, char __user *arg)
 562 {
 563         struct blk_user_trace_setup buts;
 564         int ret;
 565
 566         ret = copy_from_user(&buts, arg, sizeof(buts));
 567         if (ret)
 568                 return -EFAULT;
 569
 570         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 571         if (ret)
 572                 return ret;
 573
 574         if (copy_to_user(arg, &buts, sizeof(buts))) {
 575                 __blk_trace_remove(q);
 576                 return -EFAULT;
 577         }
 578         return 0;
 579 }
 580
 581 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 582                     struct block_device *bdev,
 583                     char __user *arg)
 584 {
 585         int ret;
 586
 587         mutex_lock(&q->blk_trace_mutex);
 588         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 589         mutex_unlock(&q->blk_trace_mutex);
 590
 591         return ret;
 592 }
 593 EXPORT_SYMBOL_GPL(blk_trace_setup);
 594
 595 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 596 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 597                                   dev_t dev, struct block_device *bdev,
 598                                   char __user *arg)
 599 {
 600         struct blk_user_trace_setup buts;
 601         struct compat_blk_user_trace_setup cbuts;
 602         int ret;
 603
 604         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 605                 return -EFAULT;
 606
 607         buts = (struct blk_user_trace_setup) {
 608                 .act_mask = cbuts.act_mask,
 609                 .buf_size = cbuts.buf_size,
 610                 .buf_nr = cbuts.buf_nr,
 611                 .start_lba = cbuts.start_lba,
 612                 .end_lba = cbuts.end_lba,
 613                 .pid = cbuts.pid,
 614         };
 615
 616         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 617         if (ret)
 618                 return ret;
 619
 620         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 621                 __blk_trace_remove(q);
 622                 return -EFAULT;
 623         }
 624
 625         return 0;
 626 }
 627 #endif
 628
 629 static int __blk_trace_startstop(struct request_queue *q, int start)
 630 {
 631         int ret;
 632         struct blk_trace *bt = q->blk_trace;
 633
 634         if (bt == NULL)
 635                 return -EINVAL;
 636
 637         /*
 638          * For starting a trace, we can transition from a setup or stopped
 639          * trace. For stopping a trace, the state must be running
 640          */
 641         ret = -EINVAL;
 642         if (start) {
 643                 if (bt->trace_state == Blktrace_setup ||
 644                     bt->trace_state == Blktrace_stopped) {
 645                         blktrace_seq++;
 646                         smp_mb();
 647                         bt->trace_state = Blktrace_running;
 648                         spin_lock_irq(&running_trace_lock);
 649                         list_add(&bt->running_list, &running_trace_list);
 650                         spin_unlock_irq(&running_trace_lock);
 651
 652                         trace_note_time(bt);
 653                         ret = 0;
 654                 }
 655         } else {
 656                 if (bt->trace_state == Blktrace_running) {
 657                         bt->trace_state = Blktrace_stopped;
 658                         spin_lock_irq(&running_trace_lock);
 659                         list_del_init(&bt->running_list);
 660                         spin_unlock_irq(&running_trace_lock);
 661                         relay_flush(bt->rchan);
 662                         ret = 0;
 663                 }
 664         }
 665
 666         return ret;
 667 }
 668
 669 int blk_trace_startstop(struct request_queue *q, int start)
 670 {
 671         int ret;
 672
 673         mutex_lock(&q->blk_trace_mutex);
 674         ret = __blk_trace_startstop(q, start);
 675         mutex_unlock(&q->blk_trace_mutex);
 676
 677         return ret;
 678 }
 679 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 680
 681 /*
 682  * When reading or writing the blktrace sysfs files, the references to the
 683  * opened sysfs or device files should prevent the underlying block device
 684  * from being removed. So no further delete protection is really needed.
 685  */
 686
 687 /**
 688  * blk_trace_ioctl: - handle the ioctls associated with tracing
 689  * @bdev:       the block device
 690  * @cmd:        the ioctl cmd
 691  * @arg:        the argument data, if any
 692  *
 693  **/
 694 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 695 {
 696         struct request_queue *q;
 697         int ret, start = 0;
 698         char b[BDEVNAME_SIZE];
 699
 700         q = bdev_get_queue(bdev);
 701         if (!q)
 702                 return -ENXIO;
 703
 704         mutex_lock(&q->blk_trace_mutex);
 705
 706         switch (cmd) {
 707         case BLKTRACESETUP:
 708                 bdevname(bdev, b);
 709                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 710                 break;
 711 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 712         case BLKTRACESETUP32:
 713                 bdevname(bdev, b);
 714                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 715                 break;
 716 #endif
 717         case BLKTRACESTART:
 718                 start = 1;
 719                 /* fall through */
 720         case BLKTRACESTOP:
 721                 ret = __blk_trace_startstop(q, start);
 722                 break;
 723         case BLKTRACETEARDOWN:
 724                 ret = __blk_trace_remove(q);
 725                 break;
 726         default:
 727                 ret = -ENOTTY;
 728                 break;
 729         }
 730
 731         mutex_unlock(&q->blk_trace_mutex);
 732         return ret;
 733 }
 734
 735 /**
 736  * blk_trace_shutdown: - stop and cleanup trace structures
 737  * @q:    the request queue associated with the device
 738  *
 739  **/
 740 void blk_trace_shutdown(struct request_queue *q)
 741 {
 742         mutex_lock(&q->blk_trace_mutex);
 743
 744         if (q->blk_trace) {
 745                 __blk_trace_startstop(q, 0);
 746                 __blk_trace_remove(q);
 747         }
 748
 749         mutex_unlock(&q->blk_trace_mutex);
 750 }
 751
 752 #ifdef CONFIG_BLK_CGROUP
 753 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 754 {
 755         struct blk_trace *bt = q->blk_trace;
 756
 757         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 758                 return 0;
 759
 760         if (!bio->bi_blkg)
 761                 return 0;
 762         return cgroup_id(bio_blkcg(bio)->css.cgroup);
 763 }
 764 #else
 765 u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 766 {
 767         return 0;
 768 }
 769 #endif
 770
 771 static u64
 772 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 773 {
 774         if (!rq->bio)
 775                 return 0;
 776         /* Use the first bio */
 777         return blk_trace_bio_get_cgid(q, rq->bio);
 778 }
 779
 780 /*
 781  * blktrace probes
 782  */
 783
 784 /**
 785  * blk_add_trace_rq - Add a trace for a request oriented action
 786  * @rq:         the source request
 787  * @error:      return status to log
 788  * @nr_bytes:   number of completed bytes
 789  * @what:       the action
 790  * @cgid:       the cgroup info
 791  *
 792  * Description:
 793  *     Records an action against a request. Will log the bio offset + size.
 794  *
 795  **/
 796 static void blk_add_trace_rq(struct request *rq, int error,
 797                              unsigned int nr_bytes, u32 what, u64 cgid)
 798 {
 799         struct blk_trace *bt = rq->q->blk_trace;
 800
 801         if (likely(!bt))
 802                 return;
 803
 804         if (blk_rq_is_passthrough(rq))
 805                 what |= BLK_TC_ACT(BLK_TC_PC);
 806         else
 807                 what |= BLK_TC_ACT(BLK_TC_FS);
 808
 809         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 810                         rq->cmd_flags, what, error, 0, NULL, cgid);
 811 }
 812
 813 static void blk_add_trace_rq_insert(void *ignore,
 814                                     struct request_queue *q, struct request *rq)
 815 {
 816         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 817                          blk_trace_request_get_cgid(q, rq));
 818 }
 819
 820 static void blk_add_trace_rq_issue(void *ignore,
 821                                    struct request_queue *q, struct request *rq)
 822 {
 823         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 824                          blk_trace_request_get_cgid(q, rq));
 825 }
 826
 827 static void blk_add_trace_rq_requeue(void *ignore,
 828                                      struct request_queue *q,
 829                                      struct request *rq)
 830 {
 831         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 832                          blk_trace_request_get_cgid(q, rq));
 833 }
 834
 835 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 836                         int error, unsigned int nr_bytes)
 837 {
 838         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 839                          blk_trace_request_get_cgid(rq->q, rq));
 840 }
 841
 842 /**
 843  * blk_add_trace_bio - Add a trace for a bio oriented action
 844  * @q:          queue the io is for
 845  * @bio:        the source bio
 846  * @what:       the action
 847  * @error:      error, if any
 848  *
 849  * Description:
 850  *     Records an action against a bio. Will log the bio offset + size.
 851  *
 852  **/
 853 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 854                               u32 what, int error)
 855 {
 856         struct blk_trace *bt = q->blk_trace;
 857
 858         if (likely(!bt))
 859                 return;
 860
 861         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 862                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 863                         blk_trace_bio_get_cgid(q, bio));
 864 }
 865
 866 static void blk_add_trace_bio_bounce(void *ignore,
 867                                      struct request_queue *q, struct bio *bio)
 868 {
 869         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
 870 }
 871
 872 static void blk_add_trace_bio_complete(void *ignore,
 873                                        struct request_queue *q, struct bio *bio,
 874                                        int error)
 875 {
 876         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
 877 }
 878
 879 static void blk_add_trace_bio_backmerge(void *ignore,
 880                                         struct request_queue *q,
 881                                         struct request *rq,
 882                                         struct bio *bio)
 883 {
 884         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
 885 }
 886
 887 static void blk_add_trace_bio_frontmerge(void *ignore,
 888                                          struct request_queue *q,
 889                                          struct request *rq,
 890                                          struct bio *bio)
 891 {
 892         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
 893 }
 894
 895 static void blk_add_trace_bio_queue(void *ignore,
 896                                     struct request_queue *q, struct bio *bio)
 897 {
 898         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
 899 }
 900
 901 static void blk_add_trace_getrq(void *ignore,
 902                                 struct request_queue *q,
 903                                 struct bio *bio, int rw)
 904 {
 905         if (bio)
 906                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
 907         else {
 908                 struct blk_trace *bt = q->blk_trace;
 909
 910                 if (bt)
 911                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 912                                         NULL, 0);
 913         }
 914 }
 915
 916
 917 static void blk_add_trace_sleeprq(void *ignore,
 918                                   struct request_queue *q,
 919                                   struct bio *bio, int rw)
 920 {
 921         if (bio)
 922                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
 923         else {
 924                 struct blk_trace *bt = q->blk_trace;
 925
 926                 if (bt)
 927                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 928                                         0, 0, NULL, 0);
 929         }
 930 }
 931
 932 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 933 {
 934         struct blk_trace *bt = q->blk_trace;
 935
 936         if (bt)
 937                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 938 }
 939
 940 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 941                                     unsigned int depth, bool explicit)
 942 {
 943         struct blk_trace *bt = q->blk_trace;
 944
 945         if (bt) {
 946                 __be64 rpdu = cpu_to_be64(depth);
 947                 u32 what;
 948
 949                 if (explicit)
 950                         what = BLK_TA_UNPLUG_IO;
 951                 else
 952                         what = BLK_TA_UNPLUG_TIMER;
 953
 954                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
 955         }
 956 }
 957
 958 static void blk_add_trace_split(void *ignore,
 959                                 struct request_queue *q, struct bio *bio,
 960                                 unsigned int pdu)
 961 {
 962         struct blk_trace *bt = q->blk_trace;
 963
 964         if (bt) {
 965                 __be64 rpdu = cpu_to_be64(pdu);
 966
 967                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 968                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 969                                 BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
 970                                 &rpdu, blk_trace_bio_get_cgid(q, bio));
 971         }
 972 }
 973
 974 /**
 975  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 976  * @ignore:     trace callback data parameter (not used)
 977  * @q:          queue the io is for
 978  * @bio:        the source bio
 979  * @dev:        target device
 980  * @from:       source sector
 981  *
 982  * Description:
 983  *     Device mapper or raid target sometimes need to split a bio because
 984  *     it spans a stripe (or similar). Add a trace for that action.
 985  *
 986  **/
 987 static void blk_add_trace_bio_remap(void *ignore,
 988                                     struct request_queue *q, struct bio *bio,
 989                                     dev_t dev, sector_t from)
 990 {
 991         struct blk_trace *bt = q->blk_trace;
 992         struct blk_io_trace_remap r;
 993
 994         if (likely(!bt))
 995                 return;
 996
 997         r.device_from = cpu_to_be32(dev);
 998         r.device_to   = cpu_to_be32(bio_dev(bio));
 999         r.sector_from = cpu_to_be64(from);
1000
1001         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1002                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1003                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1004 }
1005
1006 /**
1007  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1008  * @ignore:     trace callback data parameter (not used)
1009  * @q:          queue the io is for
1010  * @rq:         the source request
1011  * @dev:        target device
1012  * @from:       source sector
1013  *
1014  * Description:
1015  *     Device mapper remaps request to other devices.
1016  *     Add a trace for that action.
1017  *
1018  **/
1019 static void blk_add_trace_rq_remap(void *ignore,
1020                                    struct request_queue *q,
1021                                    struct request *rq, dev_t dev,
1022                                    sector_t from)
1023 {
1024         struct blk_trace *bt = q->blk_trace;
1025         struct blk_io_trace_remap r;
1026
1027         if (likely(!bt))
1028                 return;
1029
1030         r.device_from = cpu_to_be32(dev);
1031         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1032         r.sector_from = cpu_to_be64(from);
1033
1034         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1035                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1036                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1037 }
1038
1039 /**
1040  * blk_add_driver_data - Add binary message with driver-specific data
1041  * @q:          queue the io is for
1042  * @rq:         io request
1043  * @data:       driver-specific data
1044  * @len:        length of driver-specific data
1045  *
1046  * Description:
1047  *     Some drivers might want to write driver-specific data per request.
1048  *
1049  **/
1050 void blk_add_driver_data(struct request_queue *q,
1051                          struct request *rq,
1052                          void *data, size_t len)
1053 {
1054         struct blk_trace *bt = q->blk_trace;
1055
1056         if (likely(!bt))
1057                 return;
1058
1059         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1060                                 BLK_TA_DRV_DATA, 0, len, data,
1061                                 blk_trace_request_get_cgid(q, rq));
1062 }
1063 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1064
1065 static void blk_register_tracepoints(void)
1066 {
1067         int ret;
1068
1069         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1070         WARN_ON(ret);
1071         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1072         WARN_ON(ret);
1073         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1074         WARN_ON(ret);
1075         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1076         WARN_ON(ret);
1077         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1078         WARN_ON(ret);
1079         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1080         WARN_ON(ret);
1081         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1082         WARN_ON(ret);
1083         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1084         WARN_ON(ret);
1085         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1086         WARN_ON(ret);
1087         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1088         WARN_ON(ret);
1089         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1090         WARN_ON(ret);
1091         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1092         WARN_ON(ret);
1093         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1094         WARN_ON(ret);
1095         ret = register_trace_block_split(blk_add_trace_split, NULL);
1096         WARN_ON(ret);
1097         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1098         WARN_ON(ret);
1099         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1100         WARN_ON(ret);
1101 }
1102
1103 static void blk_unregister_tracepoints(void)
1104 {
1105         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1106         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1107         unregister_trace_block_split(blk_add_trace_split, NULL);
1108         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1109         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1110         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1111         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1112         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1113         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1114         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1115         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1116         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1117         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1118         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1119         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1120         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1121
1122         tracepoint_synchronize_unregister();
1123 }
1124
1125 /*
1126  * struct blk_io_tracer formatting routines
1127  */
1128
1129 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1130 {
1131         int i = 0;
1132         int tc = t->action >> BLK_TC_SHIFT;
1133
1134         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1135                 rwbs[i++] = 'N';
1136                 goto out;
1137         }
1138
1139         if (tc & BLK_TC_FLUSH)
1140                 rwbs[i++] = 'F';
1141
1142         if (tc & BLK_TC_DISCARD)
1143                 rwbs[i++] = 'D';
1144         else if (tc & BLK_TC_WRITE)
1145                 rwbs[i++] = 'W';
1146         else if (t->bytes)
1147                 rwbs[i++] = 'R';
1148         else
1149                 rwbs[i++] = 'N';
1150
1151         if (tc & BLK_TC_FUA)
1152                 rwbs[i++] = 'F';
1153         if (tc & BLK_TC_AHEAD)
1154                 rwbs[i++] = 'A';
1155         if (tc & BLK_TC_SYNC)
1156                 rwbs[i++] = 'S';
1157         if (tc & BLK_TC_META)
1158                 rwbs[i++] = 'M';
1159 out:
1160         rwbs[i] = '\0';
1161 }
1162
1163 static inline
1164 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1165 {
1166         return (const struct blk_io_trace *)ent;
1167 }
1168
1169 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1170 {
1171         return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1172 }
1173
1174 static inline u64 t_cgid(const struct trace_entry *ent)
1175 {
1176         return *(u64 *)(te_blk_io_trace(ent) + 1);
1177 }
1178
1179 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1180 {
1181         return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1182 }
1183
1184 static inline u32 t_action(const struct trace_entry *ent)
1185 {
1186         return te_blk_io_trace(ent)->action;
1187 }
1188
1189 static inline u32 t_bytes(const struct trace_entry *ent)
1190 {
1191         return te_blk_io_trace(ent)->bytes;
1192 }
1193
1194 static inline u32 t_sec(const struct trace_entry *ent)
1195 {
1196         return te_blk_io_trace(ent)->bytes >> 9;
1197 }
1198
1199 static inline unsigned long long t_sector(const struct trace_entry *ent)
1200 {
1201         return te_blk_io_trace(ent)->sector;
1202 }
1203
1204 static inline __u16 t_error(const struct trace_entry *ent)
1205 {
1206         return te_blk_io_trace(ent)->error;
1207 }
1208
1209 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1210 {
1211         const __u64 *val = pdu_start(ent, has_cg);
1212         return be64_to_cpu(*val);
1213 }
1214
1215 static void get_pdu_remap(const struct trace_entry *ent,
1216                           struct blk_io_trace_remap *r, bool has_cg)
1217 {
1218         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1219         __u64 sector_from = __r->sector_from;
1220
1221         r->device_from = be32_to_cpu(__r->device_from);
1222         r->device_to   = be32_to_cpu(__r->device_to);
1223         r->sector_from = be64_to_cpu(sector_from);
1224 }
1225
1226 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1227         bool has_cg);
1228
1229 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1230         bool has_cg)
1231 {
1232         char rwbs[RWBS_LEN];
1233         unsigned long long ts  = iter->ts;
1234         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1235         unsigned secs          = (unsigned long)ts;
1236         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1237
1238         fill_rwbs(rwbs, t);
1239
1240         trace_seq_printf(&iter->seq,
1241                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1242                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1243                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1244 }
1245
1246 static void blk_log_action(struct trace_iterator *iter, const char *act,
1247         bool has_cg)
1248 {
1249         char rwbs[RWBS_LEN];
1250         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1251
1252         fill_rwbs(rwbs, t);
1253         if (has_cg) {
1254                 u64 id = t_cgid(iter->ent);
1255
1256                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1257                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1258
1259                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1260                                 sizeof(blkcg_name_buf));
1261                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1262                                  MAJOR(t->device), MINOR(t->device),
1263                                  blkcg_name_buf, act, rwbs);
1264                 } else {
1265                         /*
1266                          * The cgid portion used to be "INO,GEN".  Userland
1267                          * builds a FILEID_INO32_GEN fid out of them and
1268                          * opens the cgroup using open_by_handle_at(2).
1269                          * While 32bit ino setups are still the same, 64bit
1270                          * ones now use the 64bit ino as the whole ID and
1271                          * no longer use generation.
1272                          *
1273                          * Regarldess of the content, always output
1274                          * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1275                          * be mapped back to @id on both 64 and 32bit ino
1276                          * setups.  See __kernfs_fh_to_dentry().
1277                          */
1278                         trace_seq_printf(&iter->seq,
1279                                  "%3d,%-3d %llx,%-llx %2s %3s ",
1280                                  MAJOR(t->device), MINOR(t->device),
1281                                  id & U32_MAX, id >> 32, act, rwbs);
1282                 }
1283         } else
1284                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1285                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1286 }
1287
1288 static void blk_log_dump_pdu(struct trace_seq *s,
1289         const struct trace_entry *ent, bool has_cg)
1290 {
1291         const unsigned char *pdu_buf;
1292         int pdu_len;
1293         int i, end;
1294
1295         pdu_buf = pdu_start(ent, has_cg);
1296         pdu_len = pdu_real_len(ent, has_cg);
1297
1298         if (!pdu_len)
1299                 return;
1300
1301         /* find the last zero that needs to be printed */
1302         for (end = pdu_len - 1; end >= 0; end--)
1303                 if (pdu_buf[end])
1304                         break;
1305         end++;
1306
1307         trace_seq_putc(s, '(');
1308
1309         for (i = 0; i < pdu_len; i++) {
1310
1311                 trace_seq_printf(s, "%s%02x",
1312                                  i == 0 ? "" : " ", pdu_buf[i]);
1313
1314                 /*
1315                  * stop when the rest is just zeroes and indicate so
1316                  * with a ".." appended
1317                  */
1318                 if (i == end && end != pdu_len - 1) {
1319                         trace_seq_puts(s, " ..) ");
1320                         return;
1321                 }
1322         }
1323
1324         trace_seq_puts(s, ") ");
1325 }
1326
1327 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1328 {
1329         char cmd[TASK_COMM_LEN];
1330
1331         trace_find_cmdline(ent->pid, cmd);
1332
1333         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1334                 trace_seq_printf(s, "%u ", t_bytes(ent));
1335                 blk_log_dump_pdu(s, ent, has_cg);
1336                 trace_seq_printf(s, "[%s]\n", cmd);
1337         } else {
1338                 if (t_sec(ent))
1339                         trace_seq_printf(s, "%llu + %u [%s]\n",
1340                                                 t_sector(ent), t_sec(ent), cmd);
1341                 else
1342                         trace_seq_printf(s, "[%s]\n", cmd);
1343         }
1344 }
1345
1346 static void blk_log_with_error(struct trace_seq *s,
1347                               const struct trace_entry *ent, bool has_cg)
1348 {
1349         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1350                 blk_log_dump_pdu(s, ent, has_cg);
1351                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1352         } else {
1353                 if (t_sec(ent))
1354                         trace_seq_printf(s, "%llu + %u [%d]\n",
1355                                          t_sector(ent),
1356                                          t_sec(ent), t_error(ent));
1357                 else
1358                         trace_seq_printf(s, "%llu [%d]\n",
1359                                          t_sector(ent), t_error(ent));
1360         }
1361 }
1362
1363 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1364 {
1365         struct blk_io_trace_remap r = { .device_from = 0, };
1366
1367         get_pdu_remap(ent, &r, has_cg);
1368         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1369                          t_sector(ent), t_sec(ent),
1370                          MAJOR(r.device_from), MINOR(r.device_from),
1371                          (unsigned long long)r.sector_from);
1372 }
1373
1374 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1375 {
1376         char cmd[TASK_COMM_LEN];
1377
1378         trace_find_cmdline(ent->pid, cmd);
1379
1380         trace_seq_printf(s, "[%s]\n", cmd);
1381 }
1382
1383 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1384 {
1385         char cmd[TASK_COMM_LEN];
1386
1387         trace_find_cmdline(ent->pid, cmd);
1388
1389         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1390 }
1391
1392 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1393 {
1394         char cmd[TASK_COMM_LEN];
1395
1396         trace_find_cmdline(ent->pid, cmd);
1397
1398         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1399                          get_pdu_int(ent, has_cg), cmd);
1400 }
1401
1402 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1403                         bool has_cg)
1404 {
1405
1406         trace_seq_putmem(s, pdu_start(ent, has_cg),
1407                 pdu_real_len(ent, has_cg));
1408         trace_seq_putc(s, '\n');
1409 }
1410
1411 /*
1412  * struct tracer operations
1413  */
1414
1415 static void blk_tracer_print_header(struct seq_file *m)
1416 {
1417         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1418                 return;
1419         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1420                     "#  |     |     |           |   |   |\n");
1421 }
1422
1423 static void blk_tracer_start(struct trace_array *tr)
1424 {
1425         blk_tracer_enabled = true;
1426 }
1427
1428 static int blk_tracer_init(struct trace_array *tr)
1429 {
1430         blk_tr = tr;
1431         blk_tracer_start(tr);
1432         return 0;
1433 }
1434
1435 static void blk_tracer_stop(struct trace_array *tr)
1436 {
1437         blk_tracer_enabled = false;
1438 }
1439
1440 static void blk_tracer_reset(struct trace_array *tr)
1441 {
1442         blk_tracer_stop(tr);
1443 }
1444
1445 static const struct {
1446         const char *act[2];
1447         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1448                             bool has_cg);
1449 } what2act[] = {
1450         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1451         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1452         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1453         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1454         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1455         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1456         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1457         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1458         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1459         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1460         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1461         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1462         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1463         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1464         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1465 };
1466
1467 static enum print_line_t print_one_line(struct trace_iterator *iter,
1468                                         bool classic)
1469 {
1470         struct trace_array *tr = iter->tr;
1471         struct trace_seq *s = &iter->seq;
1472         const struct blk_io_trace *t;
1473         u16 what;
1474         bool long_act;
1475         blk_log_action_t *log_action;
1476         bool has_cg;
1477
1478         t          = te_blk_io_trace(iter->ent);
1479         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1480         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1481         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1482         has_cg     = t->action & __BLK_TA_CGROUP;
1483
1484         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1485                 log_action(iter, long_act ? "message" : "m", has_cg);
1486                 blk_log_msg(s, iter->ent, has_cg);
1487                 return trace_handle_return(s);
1488         }
1489
1490         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1491                 trace_seq_printf(s, "Unknown action %x\n", what);
1492         else {
1493                 log_action(iter, what2act[what].act[long_act], has_cg);
1494                 what2act[what].print(s, iter->ent, has_cg);
1495         }
1496
1497         return trace_handle_return(s);
1498 }
1499
1500 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1501                                                int flags, struct trace_event *event)
1502 {
1503         return print_one_line(iter, false);
1504 }
1505
1506 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1507 {
1508         struct trace_seq *s = &iter->seq;
1509         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1510         const int offset = offsetof(struct blk_io_trace, sector);
1511         struct blk_io_trace old = {
1512                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1513                 .time     = iter->ts,
1514         };
1515
1516         trace_seq_putmem(s, &old, offset);
1517         trace_seq_putmem(s, &t->sector,
1518                          sizeof(old) - offset + t->pdu_len);
1519 }
1520
1521 static enum print_line_t
1522 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1523                              struct trace_event *event)
1524 {
1525         blk_trace_synthesize_old_trace(iter);
1526
1527         return trace_handle_return(&iter->seq);
1528 }
1529
1530 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1531 {
1532         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1533                 return TRACE_TYPE_UNHANDLED;
1534
1535         return print_one_line(iter, true);
1536 }
1537
1538 static int
1539 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1540 {
1541         /* don't output context-info for blk_classic output */
1542         if (bit == TRACE_BLK_OPT_CLASSIC) {
1543                 if (set)
1544                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1545                 else
1546                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1547         }
1548         return 0;
1549 }
1550
1551 static struct tracer blk_tracer __read_mostly = {
1552         .name           = "blk",
1553         .init           = blk_tracer_init,
1554         .reset          = blk_tracer_reset,
1555         .start          = blk_tracer_start,
1556         .stop           = blk_tracer_stop,
1557         .print_header   = blk_tracer_print_header,
1558         .print_line     = blk_tracer_print_line,
1559         .flags          = &blk_tracer_flags,
1560         .set_flag       = blk_tracer_set_flag,
1561 };
1562
1563 static struct trace_event_functions trace_blk_event_funcs = {
1564         .trace          = blk_trace_event_print,
1565         .binary         = blk_trace_event_print_binary,
1566 };
1567
1568 static struct trace_event trace_blk_event = {
1569         .type           = TRACE_BLK,
1570         .funcs          = &trace_blk_event_funcs,
1571 };
1572
1573 static int __init init_blk_tracer(void)
1574 {
1575         if (!register_trace_event(&trace_blk_event)) {
1576                 pr_warn("Warning: could not register block events\n");
1577                 return 1;
1578         }
1579
1580         if (register_tracer(&blk_tracer) != 0) {
1581                 pr_warn("Warning: could not register the block tracer\n");
1582                 unregister_trace_event(&trace_blk_event);
1583                 return 1;
1584         }
1585
1586         return 0;
1587 }
1588
1589 device_initcall(init_blk_tracer);
1590
1591 static int blk_trace_remove_queue(struct request_queue *q)
1592 {
1593         struct blk_trace *bt;
1594
1595         bt = xchg(&q->blk_trace, NULL);
1596         if (bt == NULL)
1597                 return -EINVAL;
1598
1599         put_probe_ref();
1600         blk_trace_free(bt);
1601         return 0;
1602 }
1603
1604 /*
1605  * Setup everything required to start tracing
1606  */
1607 static int blk_trace_setup_queue(struct request_queue *q,
1608                                  struct block_device *bdev)
1609 {
1610         struct blk_trace *bt = NULL;
1611         int ret = -ENOMEM;
1612
1613         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1614         if (!bt)
1615                 return -ENOMEM;
1616
1617         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1618         if (!bt->msg_data)
1619                 goto free_bt;
1620
1621         bt->dev = bdev->bd_dev;
1622         bt->act_mask = (u16)-1;
1623
1624         blk_trace_setup_lba(bt, bdev);
1625
1626         ret = -EBUSY;
1627         if (cmpxchg(&q->blk_trace, NULL, bt))
1628                 goto free_bt;
1629
1630         get_probe_ref();
1631         return 0;
1632
1633 free_bt:
1634         blk_trace_free(bt);
1635         return ret;
1636 }
1637
1638 /*
1639  * sysfs interface to enable and configure tracing
1640  */
1641
1642 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1643                                          struct device_attribute *attr,
1644                                          char *buf);
1645 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1646                                           struct device_attribute *attr,
1647                                           const char *buf, size_t count);
1648 #define BLK_TRACE_DEVICE_ATTR(_name) \
1649         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1650                     sysfs_blk_trace_attr_show, \
1651                     sysfs_blk_trace_attr_store)
1652
1653 static BLK_TRACE_DEVICE_ATTR(enable);
1654 static BLK_TRACE_DEVICE_ATTR(act_mask);
1655 static BLK_TRACE_DEVICE_ATTR(pid);
1656 static BLK_TRACE_DEVICE_ATTR(start_lba);
1657 static BLK_TRACE_DEVICE_ATTR(end_lba);
1658
1659 static struct attribute *blk_trace_attrs[] = {
1660         &dev_attr_enable.attr,
1661         &dev_attr_act_mask.attr,
1662         &dev_attr_pid.attr,
1663         &dev_attr_start_lba.attr,
1664         &dev_attr_end_lba.attr,
1665         NULL
1666 };
1667
1668 struct attribute_group blk_trace_attr_group = {
1669         .name  = "trace",
1670         .attrs = blk_trace_attrs,
1671 };
1672
1673 static const struct {
1674         int mask;
1675         const char *str;
1676 } mask_maps[] = {
1677         { BLK_TC_READ,          "read"          },
1678         { BLK_TC_WRITE,         "write"         },
1679         { BLK_TC_FLUSH,         "flush"         },
1680         { BLK_TC_SYNC,          "sync"          },
1681         { BLK_TC_QUEUE,         "queue"         },
1682         { BLK_TC_REQUEUE,       "requeue"       },
1683         { BLK_TC_ISSUE,         "issue"         },
1684         { BLK_TC_COMPLETE,      "complete"      },
1685         { BLK_TC_FS,            "fs"            },
1686         { BLK_TC_PC,            "pc"            },
1687         { BLK_TC_NOTIFY,        "notify"        },
1688         { BLK_TC_AHEAD,         "ahead"         },
1689         { BLK_TC_META,          "meta"          },
1690         { BLK_TC_DISCARD,       "discard"       },
1691         { BLK_TC_DRV_DATA,      "drv_data"      },
1692         { BLK_TC_FUA,           "fua"           },
1693 };
1694
1695 static int blk_trace_str2mask(const char *str)
1696 {
1697         int i;
1698         int mask = 0;
1699         char *buf, *s, *token;
1700
1701         buf = kstrdup(str, GFP_KERNEL);
1702         if (buf == NULL)
1703                 return -ENOMEM;
1704         s = strstrip(buf);
1705
1706         while (1) {
1707                 token = strsep(&s, ",");
1708                 if (token == NULL)
1709                         break;
1710
1711                 if (*token == '\0')
1712                         continue;
1713
1714                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1715                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1716                                 mask |= mask_maps[i].mask;
1717                                 break;
1718                         }
1719                 }
1720                 if (i == ARRAY_SIZE(mask_maps)) {
1721                         mask = -EINVAL;
1722                         break;
1723                 }
1724         }
1725         kfree(buf);
1726
1727         return mask;
1728 }
1729
1730 static ssize_t blk_trace_mask2str(char *buf, int mask)
1731 {
1732         int i;
1733         char *p = buf;
1734
1735         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1736                 if (mask & mask_maps[i].mask) {
1737                         p += sprintf(p, "%s%s",
1738                                     (p == buf) ? "" : ",", mask_maps[i].str);
1739                 }
1740         }
1741         *p++ = '\n';
1742
1743         return p - buf;
1744 }
1745
1746 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1747 {
1748         if (bdev->bd_disk == NULL)
1749                 return NULL;
1750
1751         return bdev_get_queue(bdev);
1752 }
1753
1754 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1755                                          struct device_attribute *attr,
1756                                          char *buf)
1757 {
1758         struct hd_struct *p = dev_to_part(dev);
1759         struct request_queue *q;
1760         struct block_device *bdev;
1761         ssize_t ret = -ENXIO;
1762
1763         bdev = bdget(part_devt(p));
1764         if (bdev == NULL)
1765                 goto out;
1766
1767         q = blk_trace_get_queue(bdev);
1768         if (q == NULL)
1769                 goto out_bdput;
1770
1771         mutex_lock(&q->blk_trace_mutex);
1772
1773         if (attr == &dev_attr_enable) {
1774                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1775                 goto out_unlock_bdev;
1776         }
1777
1778         if (q->blk_trace == NULL)
1779                 ret = sprintf(buf, "disabled\n");
1780         else if (attr == &dev_attr_act_mask)
1781                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1782         else if (attr == &dev_attr_pid)
1783                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1784         else if (attr == &dev_attr_start_lba)
1785                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1786         else if (attr == &dev_attr_end_lba)
1787                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1788
1789 out_unlock_bdev:
1790         mutex_unlock(&q->blk_trace_mutex);
1791 out_bdput:
1792         bdput(bdev);
1793 out:
1794         return ret;
1795 }
1796
1797 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1798                                           struct device_attribute *attr,
1799                                           const char *buf, size_t count)
1800 {
1801         struct block_device *bdev;
1802         struct request_queue *q;
1803         struct hd_struct *p;
1804         u64 value;
1805         ssize_t ret = -EINVAL;
1806
1807         if (count == 0)
1808                 goto out;
1809
1810         if (attr == &dev_attr_act_mask) {
1811                 if (kstrtoull(buf, 0, &value)) {
1812                         /* Assume it is a list of trace category names */
1813                         ret = blk_trace_str2mask(buf);
1814                         if (ret < 0)
1815                                 goto out;
1816                         value = ret;
1817                 }
1818         } else if (kstrtoull(buf, 0, &value))
1819                 goto out;
1820
1821         ret = -ENXIO;
1822
1823         p = dev_to_part(dev);
1824         bdev = bdget(part_devt(p));
1825         if (bdev == NULL)
1826                 goto out;
1827
1828         q = blk_trace_get_queue(bdev);
1829         if (q == NULL)
1830                 goto out_bdput;
1831
1832         mutex_lock(&q->blk_trace_mutex);
1833
1834         if (attr == &dev_attr_enable) {
1835                 if (!!value == !!q->blk_trace) {
1836                         ret = 0;
1837                         goto out_unlock_bdev;
1838                 }
1839                 if (value)
1840                         ret = blk_trace_setup_queue(q, bdev);
1841                 else
1842                         ret = blk_trace_remove_queue(q);
1843                 goto out_unlock_bdev;
1844         }
1845
1846         ret = 0;
1847         if (q->blk_trace == NULL)
1848                 ret = blk_trace_setup_queue(q, bdev);
1849
1850         if (ret == 0) {
1851                 if (attr == &dev_attr_act_mask)
1852                         q->blk_trace->act_mask = value;
1853                 else if (attr == &dev_attr_pid)
1854                         q->blk_trace->pid = value;
1855                 else if (attr == &dev_attr_start_lba)
1856                         q->blk_trace->start_lba = value;
1857                 else if (attr == &dev_attr_end_lba)
1858                         q->blk_trace->end_lba = value;
1859         }
1860
1861 out_unlock_bdev:
1862         mutex_unlock(&q->blk_trace_mutex);
1863 out_bdput:
1864         bdput(bdev);
1865 out:
1866         return ret ? ret : count;
1867 }
1868
1869 int blk_trace_init_sysfs(struct device *dev)
1870 {
1871         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1872 }
1873
1874 void blk_trace_remove_sysfs(struct device *dev)
1875 {
1876         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1877 }
1878
1879 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1880
1881 #ifdef CONFIG_EVENT_TRACING
1882
1883 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1884 {
1885         int i = 0;
1886
1887         if (op & REQ_PREFLUSH)
1888                 rwbs[i++] = 'F';
1889
1890         switch (op & REQ_OP_MASK) {
1891         case REQ_OP_WRITE:
1892         case REQ_OP_WRITE_SAME:
1893                 rwbs[i++] = 'W';
1894                 break;
1895         case REQ_OP_DISCARD:
1896                 rwbs[i++] = 'D';
1897                 break;
1898         case REQ_OP_SECURE_ERASE:
1899                 rwbs[i++] = 'D';
1900                 rwbs[i++] = 'E';
1901                 break;
1902         case REQ_OP_FLUSH:
1903                 rwbs[i++] = 'F';
1904                 break;
1905         case REQ_OP_READ:
1906                 rwbs[i++] = 'R';
1907                 break;
1908         default:
1909                 rwbs[i++] = 'N';
1910         }
1911
1912         if (op & REQ_FUA)
1913                 rwbs[i++] = 'F';
1914         if (op & REQ_RAHEAD)
1915                 rwbs[i++] = 'A';
1916         if (op & REQ_SYNC)
1917                 rwbs[i++] = 'S';
1918         if (op & REQ_META)
1919                 rwbs[i++] = 'M';
1920
1921         rwbs[i] = '\0';
1922 }
1923 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1924
1925 #endif /* CONFIG_EVENT_TRACING */
1926