kernel/trace/blktrace.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   4  *
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/kernel.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/blktrace_api.h>
  12 #include <linux/percpu.h>
  13 #include <linux/init.h>
  14 #include <linux/mutex.h>
  15 #include <linux/slab.h>
  16 #include <linux/debugfs.h>
  17 #include <linux/export.h>
  18 #include <linux/time.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/list.h>
  21 #include <linux/blk-cgroup.h>
  22
  23 #include "../../block/blk.h"
  24
  25 #include <trace/events/block.h>
  26
  27 #include "trace_output.h"
  28
  29 #ifdef CONFIG_BLK_DEV_IO_TRACE
  30
  31 static unsigned int blktrace_seq __read_mostly = 1;
  32
  33 static struct trace_array *blk_tr;
  34 static bool blk_tracer_enabled __read_mostly;
  35
  36 static LIST_HEAD(running_trace_list);
  37 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  38
  39 /* Select an alternative, minimalistic output than the original one */
  40 #define TRACE_BLK_OPT_CLASSIC   0x1
  41 #define TRACE_BLK_OPT_CGROUP    0x2
  42 #define TRACE_BLK_OPT_CGNAME    0x4
  43
  44 static struct tracer_opt blk_tracer_opts[] = {
  45         /* Default disable the minimalistic output */
  46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  47 #ifdef CONFIG_BLK_CGROUP
  48         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  49         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  50 #endif
  51         { }
  52 };
  53
  54 static struct tracer_flags blk_tracer_flags = {
  55         .val  = 0,
  56         .opts = blk_tracer_opts,
  57 };
  58
  59 /* Global reference count of probes */
  60 static DEFINE_MUTEX(blk_probe_mutex);
  61 static int blk_probes_ref;
  62
  63 static void blk_register_tracepoints(void);
  64 static void blk_unregister_tracepoints(void);
  65
  66 /*
  67  * Send out a notify message.
  68  */
  69 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  70                        const void *data, size_t len, u64 cgid)
  71 {
  72         struct blk_io_trace *t;
  73         struct ring_buffer_event *event = NULL;
  74         struct trace_buffer *buffer = NULL;
  75         int pc = 0;
  76         int cpu = smp_processor_id();
  77         bool blk_tracer = blk_tracer_enabled;
  78         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
  79
  80         if (blk_tracer) {
  81                 buffer = blk_tr->array_buffer.buffer;
  82                 pc = preempt_count();
  83                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  84                                                   sizeof(*t) + len + cgid_len,
  85                                                   0, pc);
  86                 if (!event)
  87                         return;
  88                 t = ring_buffer_event_data(event);
  89                 goto record_it;
  90         }
  91
  92         if (!bt->rchan)
  93                 return;
  94
  95         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
  96         if (t) {
  97                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
  98                 t->time = ktime_to_ns(ktime_get());
  99 record_it:
 100                 t->device = bt->dev;
 101                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 102                 t->pid = pid;
 103                 t->cpu = cpu;
 104                 t->pdu_len = len + cgid_len;
 105                 if (cgid_len)
 106                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 107                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 108
 109                 if (blk_tracer)
 110                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 111         }
 112 }
 113
 114 /*
 115  * Send out a notify for this process, if we haven't done so since a trace
 116  * started
 117  */
 118 static void trace_note_tsk(struct task_struct *tsk)
 119 {
 120         unsigned long flags;
 121         struct blk_trace *bt;
 122
 123         tsk->btrace_seq = blktrace_seq;
 124         spin_lock_irqsave(&running_trace_lock, flags);
 125         list_for_each_entry(bt, &running_trace_list, running_list) {
 126                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 127                            sizeof(tsk->comm), 0);
 128         }
 129         spin_unlock_irqrestore(&running_trace_lock, flags);
 130 }
 131
 132 static void trace_note_time(struct blk_trace *bt)
 133 {
 134         struct timespec64 now;
 135         unsigned long flags;
 136         u32 words[2];
 137
 138         /* need to check user space to see if this breaks in y2038 or y2106 */
 139         ktime_get_real_ts64(&now);
 140         words[0] = (u32)now.tv_sec;
 141         words[1] = now.tv_nsec;
 142
 143         local_irq_save(flags);
 144         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0);
 145         local_irq_restore(flags);
 146 }
 147
 148 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 149         const char *fmt, ...)
 150 {
 151         int n;
 152         va_list args;
 153         unsigned long flags;
 154         char *buf;
 155
 156         if (unlikely(bt->trace_state != Blktrace_running &&
 157                      !blk_tracer_enabled))
 158                 return;
 159
 160         /*
 161          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 162          * message to the trace.
 163          */
 164         if (!(bt->act_mask & BLK_TC_NOTIFY))
 165                 return;
 166
 167         local_irq_save(flags);
 168         buf = this_cpu_ptr(bt->msg_data);
 169         va_start(args, fmt);
 170         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 171         va_end(args);
 172
 173         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 174                 blkcg = NULL;
 175 #ifdef CONFIG_BLK_CGROUP
 176         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n,
 177                    blkcg ? cgroup_id(blkcg->css.cgroup) : 1);
 178 #else
 179         trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, 0);
 180 #endif
 181         local_irq_restore(flags);
 182 }
 183 EXPORT_SYMBOL_GPL(__trace_note_message);
 184
 185 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 186                          pid_t pid)
 187 {
 188         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 189                 return 1;
 190         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 191                 return 1;
 192         if (bt->pid && pid != bt->pid)
 193                 return 1;
 194
 195         return 0;
 196 }
 197
 198 /*
 199  * Data direction bit lookup
 200  */
 201 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 202                                  BLK_TC_ACT(BLK_TC_WRITE) };
 203
 204 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 205 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 206
 207 /* The ilog2() calls fall out because they're constant */
 208 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 209           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 210
 211 /*
 212  * The worker for the various blk_add_trace*() types. Fills out a
 213  * blk_io_trace structure and places it in a per-cpu subbuffer.
 214  */
 215 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 216                      int op, int op_flags, u32 what, int error, int pdu_len,
 217                      void *pdu_data, u64 cgid)
 218 {
 219         struct task_struct *tsk = current;
 220         struct ring_buffer_event *event = NULL;
 221         struct trace_buffer *buffer = NULL;
 222         struct blk_io_trace *t;
 223         unsigned long flags = 0;
 224         unsigned long *sequence;
 225         pid_t pid;
 226         int cpu, pc = 0;
 227         bool blk_tracer = blk_tracer_enabled;
 228         ssize_t cgid_len = cgid ? sizeof(cgid) : 0;
 229
 230         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 231                 return;
 232
 233         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 234         what |= MASK_TC_BIT(op_flags, SYNC);
 235         what |= MASK_TC_BIT(op_flags, RAHEAD);
 236         what |= MASK_TC_BIT(op_flags, META);
 237         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 238         what |= MASK_TC_BIT(op_flags, FUA);
 239         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 240                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 241         if (op == REQ_OP_FLUSH)
 242                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 243         if (cgid)
 244                 what |= __BLK_TA_CGROUP;
 245
 246         pid = tsk->pid;
 247         if (act_log_check(bt, what, sector, pid))
 248                 return;
 249         cpu = raw_smp_processor_id();
 250
 251         if (blk_tracer) {
 252                 tracing_record_cmdline(current);
 253
 254                 buffer = blk_tr->array_buffer.buffer;
 255                 pc = preempt_count();
 256                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 257                                                   sizeof(*t) + pdu_len + cgid_len,
 258                                                   0, pc);
 259                 if (!event)
 260                         return;
 261                 t = ring_buffer_event_data(event);
 262                 goto record_it;
 263         }
 264
 265         if (unlikely(tsk->btrace_seq != blktrace_seq))
 266                 trace_note_tsk(tsk);
 267
 268         /*
 269          * A word about the locking here - we disable interrupts to reserve
 270          * some space in the relay per-cpu buffer, to prevent an irq
 271          * from coming in and stepping on our toes.
 272          */
 273         local_irq_save(flags);
 274         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 275         if (t) {
 276                 sequence = per_cpu_ptr(bt->sequence, cpu);
 277
 278                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 279                 t->sequence = ++(*sequence);
 280                 t->time = ktime_to_ns(ktime_get());
 281 record_it:
 282                 /*
 283                  * These two are not needed in ftrace as they are in the
 284                  * generic trace_entry, filled by tracing_generic_entry_update,
 285                  * but for the trace_event->bin() synthesizer benefit we do it
 286                  * here too.
 287                  */
 288                 t->cpu = cpu;
 289                 t->pid = pid;
 290
 291                 t->sector = sector;
 292                 t->bytes = bytes;
 293                 t->action = what;
 294                 t->device = bt->dev;
 295                 t->error = error;
 296                 t->pdu_len = pdu_len + cgid_len;
 297
 298                 if (cgid_len)
 299                         memcpy((void *)t + sizeof(*t), &cgid, cgid_len);
 300                 if (pdu_len)
 301                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 302
 303                 if (blk_tracer) {
 304                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 305                         return;
 306                 }
 307         }
 308
 309         local_irq_restore(flags);
 310 }
 311
 312 static void blk_trace_free(struct blk_trace *bt)
 313 {
 314         debugfs_remove(bt->msg_file);
 315         debugfs_remove(bt->dropped_file);
 316         relay_close(bt->rchan);
 317         debugfs_remove(bt->dir);
 318         free_percpu(bt->sequence);
 319         free_percpu(bt->msg_data);
 320         kfree(bt);
 321 }
 322
 323 static void get_probe_ref(void)
 324 {
 325         mutex_lock(&blk_probe_mutex);
 326         if (++blk_probes_ref == 1)
 327                 blk_register_tracepoints();
 328         mutex_unlock(&blk_probe_mutex);
 329 }
 330
 331 static void put_probe_ref(void)
 332 {
 333         mutex_lock(&blk_probe_mutex);
 334         if (!--blk_probes_ref)
 335                 blk_unregister_tracepoints();
 336         mutex_unlock(&blk_probe_mutex);
 337 }
 338
 339 static void blk_trace_cleanup(struct blk_trace *bt)
 340 {
 341         synchronize_rcu();
 342         blk_trace_free(bt);
 343         put_probe_ref();
 344 }
 345
 346 static int __blk_trace_remove(struct request_queue *q)
 347 {
 348         struct blk_trace *bt;
 349
 350         bt = rcu_replace_pointer(q->blk_trace, NULL,
 351                                  lockdep_is_held(&q->debugfs_mutex));
 352         if (!bt)
 353                 return -EINVAL;
 354
 355         if (bt->trace_state != Blktrace_running)
 356                 blk_trace_cleanup(bt);
 357
 358         return 0;
 359 }
 360
 361 int blk_trace_remove(struct request_queue *q)
 362 {
 363         int ret;
 364
 365         mutex_lock(&q->debugfs_mutex);
 366         ret = __blk_trace_remove(q);
 367         mutex_unlock(&q->debugfs_mutex);
 368
 369         return ret;
 370 }
 371 EXPORT_SYMBOL_GPL(blk_trace_remove);
 372
 373 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 374                                 size_t count, loff_t *ppos)
 375 {
 376         struct blk_trace *bt = filp->private_data;
 377         char buf[16];
 378
 379         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 380
 381         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 382 }
 383
 384 static const struct file_operations blk_dropped_fops = {
 385         .owner =        THIS_MODULE,
 386         .open =         simple_open,
 387         .read =         blk_dropped_read,
 388         .llseek =       default_llseek,
 389 };
 390
 391 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 392                                 size_t count, loff_t *ppos)
 393 {
 394         char *msg;
 395         struct blk_trace *bt;
 396
 397         if (count >= BLK_TN_MAX_MSG)
 398                 return -EINVAL;
 399
 400         msg = memdup_user_nul(buffer, count);
 401         if (IS_ERR(msg))
 402                 return PTR_ERR(msg);
 403
 404         bt = filp->private_data;
 405         __trace_note_message(bt, NULL, "%s", msg);
 406         kfree(msg);
 407
 408         return count;
 409 }
 410
 411 static const struct file_operations blk_msg_fops = {
 412         .owner =        THIS_MODULE,
 413         .open =         simple_open,
 414         .write =        blk_msg_write,
 415         .llseek =       noop_llseek,
 416 };
 417
 418 /*
 419  * Keep track of how many times we encountered a full subbuffer, to aid
 420  * the user space app in telling how many lost events there were.
 421  */
 422 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 423                                      void *prev_subbuf, size_t prev_padding)
 424 {
 425         struct blk_trace *bt;
 426
 427         if (!relay_buf_full(buf))
 428                 return 1;
 429
 430         bt = buf->chan->private_data;
 431         atomic_inc(&bt->dropped);
 432         return 0;
 433 }
 434
 435 static int blk_remove_buf_file_callback(struct dentry *dentry)
 436 {
 437         debugfs_remove(dentry);
 438
 439         return 0;
 440 }
 441
 442 static struct dentry *blk_create_buf_file_callback(const char *filename,
 443                                                    struct dentry *parent,
 444                                                    umode_t mode,
 445                                                    struct rchan_buf *buf,
 446                                                    int *is_global)
 447 {
 448         return debugfs_create_file(filename, mode, parent, buf,
 449                                         &relay_file_operations);
 450 }
 451
 452 static const struct rchan_callbacks blk_relay_callbacks = {
 453         .subbuf_start           = blk_subbuf_start_callback,
 454         .create_buf_file        = blk_create_buf_file_callback,
 455         .remove_buf_file        = blk_remove_buf_file_callback,
 456 };
 457
 458 static void blk_trace_setup_lba(struct blk_trace *bt,
 459                                 struct block_device *bdev)
 460 {
 461         if (bdev) {
 462                 bt->start_lba = bdev->bd_start_sect;
 463                 bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
 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         lockdep_assert_held(&q->debugfs_mutex);
 482
 483         if (!buts->buf_size || !buts->buf_nr)
 484                 return -EINVAL;
 485
 486         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 487         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 488
 489         /*
 490          * some device names have larger paths - convert the slashes
 491          * to underscores for this to work as expected
 492          */
 493         strreplace(buts->name, '/', '_');
 494
 495         /*
 496          * bdev can be NULL, as with scsi-generic, this is a helpful as
 497          * we can be.
 498          */
 499         if (rcu_dereference_protected(q->blk_trace,
 500                                       lockdep_is_held(&q->debugfs_mutex))) {
 501                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 502                         buts->name);
 503                 return -EBUSY;
 504         }
 505
 506         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 507         if (!bt)
 508                 return -ENOMEM;
 509
 510         ret = -ENOMEM;
 511         bt->sequence = alloc_percpu(unsigned long);
 512         if (!bt->sequence)
 513                 goto err;
 514
 515         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 516         if (!bt->msg_data)
 517                 goto err;
 518
 519         /*
 520          * When tracing the whole disk reuse the existing debugfs directory
 521          * created by the block layer on init. For partitions block devices,
 522          * and scsi-generic block devices we create a temporary new debugfs
 523          * directory that will be removed once the trace ends.
 524          */
 525         if (bdev && !bdev_is_partition(bdev))
 526                 dir = q->debugfs_dir;
 527         else
 528                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 529
 530         /*
 531          * As blktrace relies on debugfs for its interface the debugfs directory
 532          * is required, contrary to the usual mantra of not checking for debugfs
 533          * files or directories.
 534          */
 535         if (IS_ERR_OR_NULL(dir)) {
 536                 pr_warn("debugfs_dir not present for %s so skipping\n",
 537                         buts->name);
 538                 ret = -ENOENT;
 539                 goto err;
 540         }
 541
 542         bt->dev = dev;
 543         atomic_set(&bt->dropped, 0);
 544         INIT_LIST_HEAD(&bt->running_list);
 545
 546         ret = -EIO;
 547         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 548                                                &blk_dropped_fops);
 549
 550         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 551
 552         bt->rchan = relay_open("trace", dir, buts->buf_size,
 553                                 buts->buf_nr, &blk_relay_callbacks, bt);
 554         if (!bt->rchan)
 555                 goto err;
 556
 557         bt->act_mask = buts->act_mask;
 558         if (!bt->act_mask)
 559                 bt->act_mask = (u16) -1;
 560
 561         blk_trace_setup_lba(bt, bdev);
 562
 563         /* overwrite with user settings */
 564         if (buts->start_lba)
 565                 bt->start_lba = buts->start_lba;
 566         if (buts->end_lba)
 567                 bt->end_lba = buts->end_lba;
 568
 569         bt->pid = buts->pid;
 570         bt->trace_state = Blktrace_setup;
 571
 572         rcu_assign_pointer(q->blk_trace, bt);
 573         get_probe_ref();
 574
 575         ret = 0;
 576 err:
 577         if (ret)
 578                 blk_trace_free(bt);
 579         return ret;
 580 }
 581
 582 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 583                              struct block_device *bdev, char __user *arg)
 584 {
 585         struct blk_user_trace_setup buts;
 586         int ret;
 587
 588         ret = copy_from_user(&buts, arg, sizeof(buts));
 589         if (ret)
 590                 return -EFAULT;
 591
 592         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 593         if (ret)
 594                 return ret;
 595
 596         if (copy_to_user(arg, &buts, sizeof(buts))) {
 597                 __blk_trace_remove(q);
 598                 return -EFAULT;
 599         }
 600         return 0;
 601 }
 602
 603 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 604                     struct block_device *bdev,
 605                     char __user *arg)
 606 {
 607         int ret;
 608
 609         mutex_lock(&q->debugfs_mutex);
 610         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 611         mutex_unlock(&q->debugfs_mutex);
 612
 613         return ret;
 614 }
 615 EXPORT_SYMBOL_GPL(blk_trace_setup);
 616
 617 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 618 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 619                                   dev_t dev, struct block_device *bdev,
 620                                   char __user *arg)
 621 {
 622         struct blk_user_trace_setup buts;
 623         struct compat_blk_user_trace_setup cbuts;
 624         int ret;
 625
 626         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 627                 return -EFAULT;
 628
 629         buts = (struct blk_user_trace_setup) {
 630                 .act_mask = cbuts.act_mask,
 631                 .buf_size = cbuts.buf_size,
 632                 .buf_nr = cbuts.buf_nr,
 633                 .start_lba = cbuts.start_lba,
 634                 .end_lba = cbuts.end_lba,
 635                 .pid = cbuts.pid,
 636         };
 637
 638         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 639         if (ret)
 640                 return ret;
 641
 642         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 643                 __blk_trace_remove(q);
 644                 return -EFAULT;
 645         }
 646
 647         return 0;
 648 }
 649 #endif
 650
 651 static int __blk_trace_startstop(struct request_queue *q, int start)
 652 {
 653         int ret;
 654         struct blk_trace *bt;
 655
 656         bt = rcu_dereference_protected(q->blk_trace,
 657                                        lockdep_is_held(&q->debugfs_mutex));
 658         if (bt == NULL)
 659                 return -EINVAL;
 660
 661         /*
 662          * For starting a trace, we can transition from a setup or stopped
 663          * trace. For stopping a trace, the state must be running
 664          */
 665         ret = -EINVAL;
 666         if (start) {
 667                 if (bt->trace_state == Blktrace_setup ||
 668                     bt->trace_state == Blktrace_stopped) {
 669                         blktrace_seq++;
 670                         smp_mb();
 671                         bt->trace_state = Blktrace_running;
 672                         spin_lock_irq(&running_trace_lock);
 673                         list_add(&bt->running_list, &running_trace_list);
 674                         spin_unlock_irq(&running_trace_lock);
 675
 676                         trace_note_time(bt);
 677                         ret = 0;
 678                 }
 679         } else {
 680                 if (bt->trace_state == Blktrace_running) {
 681                         bt->trace_state = Blktrace_stopped;
 682                         spin_lock_irq(&running_trace_lock);
 683                         list_del_init(&bt->running_list);
 684                         spin_unlock_irq(&running_trace_lock);
 685                         relay_flush(bt->rchan);
 686                         ret = 0;
 687                 }
 688         }
 689
 690         return ret;
 691 }
 692
 693 int blk_trace_startstop(struct request_queue *q, int start)
 694 {
 695         int ret;
 696
 697         mutex_lock(&q->debugfs_mutex);
 698         ret = __blk_trace_startstop(q, start);
 699         mutex_unlock(&q->debugfs_mutex);
 700
 701         return ret;
 702 }
 703 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 704
 705 /*
 706  * When reading or writing the blktrace sysfs files, the references to the
 707  * opened sysfs or device files should prevent the underlying block device
 708  * from being removed. So no further delete protection is really needed.
 709  */
 710
 711 /**
 712  * blk_trace_ioctl: - handle the ioctls associated with tracing
 713  * @bdev:       the block device
 714  * @cmd:        the ioctl cmd
 715  * @arg:        the argument data, if any
 716  *
 717  **/
 718 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 719 {
 720         struct request_queue *q;
 721         int ret, start = 0;
 722         char b[BDEVNAME_SIZE];
 723
 724         q = bdev_get_queue(bdev);
 725         if (!q)
 726                 return -ENXIO;
 727
 728         mutex_lock(&q->debugfs_mutex);
 729
 730         switch (cmd) {
 731         case BLKTRACESETUP:
 732                 bdevname(bdev, b);
 733                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 734                 break;
 735 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 736         case BLKTRACESETUP32:
 737                 bdevname(bdev, b);
 738                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 739                 break;
 740 #endif
 741         case BLKTRACESTART:
 742                 start = 1;
 743                 fallthrough;
 744         case BLKTRACESTOP:
 745                 ret = __blk_trace_startstop(q, start);
 746                 break;
 747         case BLKTRACETEARDOWN:
 748                 ret = __blk_trace_remove(q);
 749                 break;
 750         default:
 751                 ret = -ENOTTY;
 752                 break;
 753         }
 754
 755         mutex_unlock(&q->debugfs_mutex);
 756         return ret;
 757 }
 758
 759 /**
 760  * blk_trace_shutdown: - stop and cleanup trace structures
 761  * @q:    the request queue associated with the device
 762  *
 763  **/
 764 void blk_trace_shutdown(struct request_queue *q)
 765 {
 766         mutex_lock(&q->debugfs_mutex);
 767         if (rcu_dereference_protected(q->blk_trace,
 768                                       lockdep_is_held(&q->debugfs_mutex))) {
 769                 __blk_trace_startstop(q, 0);
 770                 __blk_trace_remove(q);
 771         }
 772
 773         mutex_unlock(&q->debugfs_mutex);
 774 }
 775
 776 #ifdef CONFIG_BLK_CGROUP
 777 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 778 {
 779         struct blk_trace *bt;
 780
 781         /* We don't use the 'bt' value here except as an optimization... */
 782         bt = rcu_dereference_protected(q->blk_trace, 1);
 783         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 784                 return 0;
 785
 786         if (!bio->bi_blkg)
 787                 return 0;
 788         return cgroup_id(bio_blkcg(bio)->css.cgroup);
 789 }
 790 #else
 791 static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 792 {
 793         return 0;
 794 }
 795 #endif
 796
 797 static u64
 798 blk_trace_request_get_cgid(struct request *rq)
 799 {
 800         if (!rq->bio)
 801                 return 0;
 802         /* Use the first bio */
 803         return blk_trace_bio_get_cgid(rq->q, rq->bio);
 804 }
 805
 806 /*
 807  * blktrace probes
 808  */
 809
 810 /**
 811  * blk_add_trace_rq - Add a trace for a request oriented action
 812  * @rq:         the source request
 813  * @error:      return status to log
 814  * @nr_bytes:   number of completed bytes
 815  * @what:       the action
 816  * @cgid:       the cgroup info
 817  *
 818  * Description:
 819  *     Records an action against a request. Will log the bio offset + size.
 820  *
 821  **/
 822 static void blk_add_trace_rq(struct request *rq, int error,
 823                              unsigned int nr_bytes, u32 what, u64 cgid)
 824 {
 825         struct blk_trace *bt;
 826
 827         rcu_read_lock();
 828         bt = rcu_dereference(rq->q->blk_trace);
 829         if (likely(!bt)) {
 830                 rcu_read_unlock();
 831                 return;
 832         }
 833
 834         if (blk_rq_is_passthrough(rq))
 835                 what |= BLK_TC_ACT(BLK_TC_PC);
 836         else
 837                 what |= BLK_TC_ACT(BLK_TC_FS);
 838
 839         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 840                         rq->cmd_flags, what, error, 0, NULL, cgid);
 841         rcu_read_unlock();
 842 }
 843
 844 static void blk_add_trace_rq_insert(void *ignore, struct request *rq)
 845 {
 846         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 847                          blk_trace_request_get_cgid(rq));
 848 }
 849
 850 static void blk_add_trace_rq_issue(void *ignore, struct request *rq)
 851 {
 852         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 853                          blk_trace_request_get_cgid(rq));
 854 }
 855
 856 static void blk_add_trace_rq_merge(void *ignore, struct request *rq)
 857 {
 858         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE,
 859                          blk_trace_request_get_cgid(rq));
 860 }
 861
 862 static void blk_add_trace_rq_requeue(void *ignore, struct request *rq)
 863 {
 864         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 865                          blk_trace_request_get_cgid(rq));
 866 }
 867
 868 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 869                         int error, unsigned int nr_bytes)
 870 {
 871         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 872                          blk_trace_request_get_cgid(rq));
 873 }
 874
 875 /**
 876  * blk_add_trace_bio - Add a trace for a bio oriented action
 877  * @q:          queue the io is for
 878  * @bio:        the source bio
 879  * @what:       the action
 880  * @error:      error, if any
 881  *
 882  * Description:
 883  *     Records an action against a bio. Will log the bio offset + size.
 884  *
 885  **/
 886 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 887                               u32 what, int error)
 888 {
 889         struct blk_trace *bt;
 890
 891         rcu_read_lock();
 892         bt = rcu_dereference(q->blk_trace);
 893         if (likely(!bt)) {
 894                 rcu_read_unlock();
 895                 return;
 896         }
 897
 898         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 899                         bio_op(bio), bio->bi_opf, what, error, 0, NULL,
 900                         blk_trace_bio_get_cgid(q, bio));
 901         rcu_read_unlock();
 902 }
 903
 904 static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio)
 905 {
 906         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0);
 907 }
 908
 909 static void blk_add_trace_bio_complete(void *ignore,
 910                                        struct request_queue *q, struct bio *bio)
 911 {
 912         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE,
 913                           blk_status_to_errno(bio->bi_status));
 914 }
 915
 916 static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio)
 917 {
 918         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE,
 919                         0);
 920 }
 921
 922 static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio)
 923 {
 924         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE,
 925                         0);
 926 }
 927
 928 static void blk_add_trace_bio_queue(void *ignore, struct bio *bio)
 929 {
 930         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0);
 931 }
 932
 933 static void blk_add_trace_getrq(void *ignore, struct bio *bio)
 934 {
 935         blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0);
 936 }
 937
 938 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 939 {
 940         struct blk_trace *bt;
 941
 942         rcu_read_lock();
 943         bt = rcu_dereference(q->blk_trace);
 944         if (bt)
 945                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0);
 946         rcu_read_unlock();
 947 }
 948
 949 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 950                                     unsigned int depth, bool explicit)
 951 {
 952         struct blk_trace *bt;
 953
 954         rcu_read_lock();
 955         bt = rcu_dereference(q->blk_trace);
 956         if (bt) {
 957                 __be64 rpdu = cpu_to_be64(depth);
 958                 u32 what;
 959
 960                 if (explicit)
 961                         what = BLK_TA_UNPLUG_IO;
 962                 else
 963                         what = BLK_TA_UNPLUG_TIMER;
 964
 965                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0);
 966         }
 967         rcu_read_unlock();
 968 }
 969
 970 static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu)
 971 {
 972         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
 973         struct blk_trace *bt;
 974
 975         rcu_read_lock();
 976         bt = rcu_dereference(q->blk_trace);
 977         if (bt) {
 978                 __be64 rpdu = cpu_to_be64(pdu);
 979
 980                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 981                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 982                                 BLK_TA_SPLIT,
 983                                 blk_status_to_errno(bio->bi_status),
 984                                 sizeof(rpdu), &rpdu,
 985                                 blk_trace_bio_get_cgid(q, bio));
 986         }
 987         rcu_read_unlock();
 988 }
 989
 990 /**
 991  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
 992  * @ignore:     trace callback data parameter (not used)
 993  * @bio:        the source bio
 994  * @dev:        source device
 995  * @from:       source sector
 996  *
 997  * Called after a bio is remapped to a different device and/or sector.
 998  **/
 999 static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev,
1000                                     sector_t from)
1001 {
1002         struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1003         struct blk_trace *bt;
1004         struct blk_io_trace_remap r;
1005
1006         rcu_read_lock();
1007         bt = rcu_dereference(q->blk_trace);
1008         if (likely(!bt)) {
1009                 rcu_read_unlock();
1010                 return;
1011         }
1012
1013         r.device_from = cpu_to_be32(dev);
1014         r.device_to   = cpu_to_be32(bio_dev(bio));
1015         r.sector_from = cpu_to_be64(from);
1016
1017         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1018                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1019                         blk_status_to_errno(bio->bi_status),
1020                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1021         rcu_read_unlock();
1022 }
1023
1024 /**
1025  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1026  * @ignore:     trace callback data parameter (not used)
1027  * @rq:         the source request
1028  * @dev:        target device
1029  * @from:       source sector
1030  *
1031  * Description:
1032  *     Device mapper remaps request to other devices.
1033  *     Add a trace for that action.
1034  *
1035  **/
1036 static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev,
1037                                    sector_t from)
1038 {
1039         struct blk_trace *bt;
1040         struct blk_io_trace_remap r;
1041
1042         rcu_read_lock();
1043         bt = rcu_dereference(rq->q->blk_trace);
1044         if (likely(!bt)) {
1045                 rcu_read_unlock();
1046                 return;
1047         }
1048
1049         r.device_from = cpu_to_be32(dev);
1050         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1051         r.sector_from = cpu_to_be64(from);
1052
1053         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1054                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1055                         sizeof(r), &r, blk_trace_request_get_cgid(rq));
1056         rcu_read_unlock();
1057 }
1058
1059 /**
1060  * blk_add_driver_data - Add binary message with driver-specific data
1061  * @rq:         io request
1062  * @data:       driver-specific data
1063  * @len:        length of driver-specific data
1064  *
1065  * Description:
1066  *     Some drivers might want to write driver-specific data per request.
1067  *
1068  **/
1069 void blk_add_driver_data(struct request *rq, void *data, size_t len)
1070 {
1071         struct blk_trace *bt;
1072
1073         rcu_read_lock();
1074         bt = rcu_dereference(rq->q->blk_trace);
1075         if (likely(!bt)) {
1076                 rcu_read_unlock();
1077                 return;
1078         }
1079
1080         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1081                                 BLK_TA_DRV_DATA, 0, len, data,
1082                                 blk_trace_request_get_cgid(rq));
1083         rcu_read_unlock();
1084 }
1085 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1086
1087 static void blk_register_tracepoints(void)
1088 {
1089         int ret;
1090
1091         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1092         WARN_ON(ret);
1093         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1094         WARN_ON(ret);
1095         ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1096         WARN_ON(ret);
1097         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1098         WARN_ON(ret);
1099         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1100         WARN_ON(ret);
1101         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1102         WARN_ON(ret);
1103         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1104         WARN_ON(ret);
1105         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1106         WARN_ON(ret);
1107         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1108         WARN_ON(ret);
1109         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1110         WARN_ON(ret);
1111         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1112         WARN_ON(ret);
1113         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1114         WARN_ON(ret);
1115         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1116         WARN_ON(ret);
1117         ret = register_trace_block_split(blk_add_trace_split, NULL);
1118         WARN_ON(ret);
1119         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1120         WARN_ON(ret);
1121         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1122         WARN_ON(ret);
1123 }
1124
1125 static void blk_unregister_tracepoints(void)
1126 {
1127         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1128         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1129         unregister_trace_block_split(blk_add_trace_split, NULL);
1130         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1131         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1132         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1133         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1134         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1135         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1136         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1137         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1138         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1139         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1140         unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL);
1141         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1142         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1143
1144         tracepoint_synchronize_unregister();
1145 }
1146
1147 /*
1148  * struct blk_io_tracer formatting routines
1149  */
1150
1151 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1152 {
1153         int i = 0;
1154         int tc = t->action >> BLK_TC_SHIFT;
1155
1156         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1157                 rwbs[i++] = 'N';
1158                 goto out;
1159         }
1160
1161         if (tc & BLK_TC_FLUSH)
1162                 rwbs[i++] = 'F';
1163
1164         if (tc & BLK_TC_DISCARD)
1165                 rwbs[i++] = 'D';
1166         else if (tc & BLK_TC_WRITE)
1167                 rwbs[i++] = 'W';
1168         else if (t->bytes)
1169                 rwbs[i++] = 'R';
1170         else
1171                 rwbs[i++] = 'N';
1172
1173         if (tc & BLK_TC_FUA)
1174                 rwbs[i++] = 'F';
1175         if (tc & BLK_TC_AHEAD)
1176                 rwbs[i++] = 'A';
1177         if (tc & BLK_TC_SYNC)
1178                 rwbs[i++] = 'S';
1179         if (tc & BLK_TC_META)
1180                 rwbs[i++] = 'M';
1181 out:
1182         rwbs[i] = '\0';
1183 }
1184
1185 static inline
1186 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1187 {
1188         return (const struct blk_io_trace *)ent;
1189 }
1190
1191 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1192 {
1193         return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0);
1194 }
1195
1196 static inline u64 t_cgid(const struct trace_entry *ent)
1197 {
1198         return *(u64 *)(te_blk_io_trace(ent) + 1);
1199 }
1200
1201 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1202 {
1203         return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0);
1204 }
1205
1206 static inline u32 t_action(const struct trace_entry *ent)
1207 {
1208         return te_blk_io_trace(ent)->action;
1209 }
1210
1211 static inline u32 t_bytes(const struct trace_entry *ent)
1212 {
1213         return te_blk_io_trace(ent)->bytes;
1214 }
1215
1216 static inline u32 t_sec(const struct trace_entry *ent)
1217 {
1218         return te_blk_io_trace(ent)->bytes >> 9;
1219 }
1220
1221 static inline unsigned long long t_sector(const struct trace_entry *ent)
1222 {
1223         return te_blk_io_trace(ent)->sector;
1224 }
1225
1226 static inline __u16 t_error(const struct trace_entry *ent)
1227 {
1228         return te_blk_io_trace(ent)->error;
1229 }
1230
1231 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1232 {
1233         const __be64 *val = pdu_start(ent, has_cg);
1234         return be64_to_cpu(*val);
1235 }
1236
1237 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1238         bool has_cg);
1239
1240 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1241         bool has_cg)
1242 {
1243         char rwbs[RWBS_LEN];
1244         unsigned long long ts  = iter->ts;
1245         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1246         unsigned secs          = (unsigned long)ts;
1247         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1248
1249         fill_rwbs(rwbs, t);
1250
1251         trace_seq_printf(&iter->seq,
1252                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1253                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1254                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1255 }
1256
1257 static void blk_log_action(struct trace_iterator *iter, const char *act,
1258         bool has_cg)
1259 {
1260         char rwbs[RWBS_LEN];
1261         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1262
1263         fill_rwbs(rwbs, t);
1264         if (has_cg) {
1265                 u64 id = t_cgid(iter->ent);
1266
1267                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1268                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1269
1270                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1271                                 sizeof(blkcg_name_buf));
1272                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1273                                  MAJOR(t->device), MINOR(t->device),
1274                                  blkcg_name_buf, act, rwbs);
1275                 } else {
1276                         /*
1277                          * The cgid portion used to be "INO,GEN".  Userland
1278                          * builds a FILEID_INO32_GEN fid out of them and
1279                          * opens the cgroup using open_by_handle_at(2).
1280                          * While 32bit ino setups are still the same, 64bit
1281                          * ones now use the 64bit ino as the whole ID and
1282                          * no longer use generation.
1283                          *
1284                          * Regardless of the content, always output
1285                          * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can
1286                          * be mapped back to @id on both 64 and 32bit ino
1287                          * setups.  See __kernfs_fh_to_dentry().
1288                          */
1289                         trace_seq_printf(&iter->seq,
1290                                  "%3d,%-3d %llx,%-llx %2s %3s ",
1291                                  MAJOR(t->device), MINOR(t->device),
1292                                  id & U32_MAX, id >> 32, act, rwbs);
1293                 }
1294         } else
1295                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1296                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1297 }
1298
1299 static void blk_log_dump_pdu(struct trace_seq *s,
1300         const struct trace_entry *ent, bool has_cg)
1301 {
1302         const unsigned char *pdu_buf;
1303         int pdu_len;
1304         int i, end;
1305
1306         pdu_buf = pdu_start(ent, has_cg);
1307         pdu_len = pdu_real_len(ent, has_cg);
1308
1309         if (!pdu_len)
1310                 return;
1311
1312         /* find the last zero that needs to be printed */
1313         for (end = pdu_len - 1; end >= 0; end--)
1314                 if (pdu_buf[end])
1315                         break;
1316         end++;
1317
1318         trace_seq_putc(s, '(');
1319
1320         for (i = 0; i < pdu_len; i++) {
1321
1322                 trace_seq_printf(s, "%s%02x",
1323                                  i == 0 ? "" : " ", pdu_buf[i]);
1324
1325                 /*
1326                  * stop when the rest is just zeros and indicate so
1327                  * with a ".." appended
1328                  */
1329                 if (i == end && end != pdu_len - 1) {
1330                         trace_seq_puts(s, " ..) ");
1331                         return;
1332                 }
1333         }
1334
1335         trace_seq_puts(s, ") ");
1336 }
1337
1338 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1339 {
1340         char cmd[TASK_COMM_LEN];
1341
1342         trace_find_cmdline(ent->pid, cmd);
1343
1344         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1345                 trace_seq_printf(s, "%u ", t_bytes(ent));
1346                 blk_log_dump_pdu(s, ent, has_cg);
1347                 trace_seq_printf(s, "[%s]\n", cmd);
1348         } else {
1349                 if (t_sec(ent))
1350                         trace_seq_printf(s, "%llu + %u [%s]\n",
1351                                                 t_sector(ent), t_sec(ent), cmd);
1352                 else
1353                         trace_seq_printf(s, "[%s]\n", cmd);
1354         }
1355 }
1356
1357 static void blk_log_with_error(struct trace_seq *s,
1358                               const struct trace_entry *ent, bool has_cg)
1359 {
1360         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1361                 blk_log_dump_pdu(s, ent, has_cg);
1362                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1363         } else {
1364                 if (t_sec(ent))
1365                         trace_seq_printf(s, "%llu + %u [%d]\n",
1366                                          t_sector(ent),
1367                                          t_sec(ent), t_error(ent));
1368                 else
1369                         trace_seq_printf(s, "%llu [%d]\n",
1370                                          t_sector(ent), t_error(ent));
1371         }
1372 }
1373
1374 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1375 {
1376         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1377
1378         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1379                          t_sector(ent), t_sec(ent),
1380                          MAJOR(be32_to_cpu(__r->device_from)),
1381                          MINOR(be32_to_cpu(__r->device_from)),
1382                          be64_to_cpu(__r->sector_from));
1383 }
1384
1385 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1386 {
1387         char cmd[TASK_COMM_LEN];
1388
1389         trace_find_cmdline(ent->pid, cmd);
1390
1391         trace_seq_printf(s, "[%s]\n", cmd);
1392 }
1393
1394 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1395 {
1396         char cmd[TASK_COMM_LEN];
1397
1398         trace_find_cmdline(ent->pid, cmd);
1399
1400         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1401 }
1402
1403 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1404 {
1405         char cmd[TASK_COMM_LEN];
1406
1407         trace_find_cmdline(ent->pid, cmd);
1408
1409         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1410                          get_pdu_int(ent, has_cg), cmd);
1411 }
1412
1413 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1414                         bool has_cg)
1415 {
1416
1417         trace_seq_putmem(s, pdu_start(ent, has_cg),
1418                 pdu_real_len(ent, has_cg));
1419         trace_seq_putc(s, '\n');
1420 }
1421
1422 /*
1423  * struct tracer operations
1424  */
1425
1426 static void blk_tracer_print_header(struct seq_file *m)
1427 {
1428         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1429                 return;
1430         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1431                     "#  |     |     |           |   |   |\n");
1432 }
1433
1434 static void blk_tracer_start(struct trace_array *tr)
1435 {
1436         blk_tracer_enabled = true;
1437 }
1438
1439 static int blk_tracer_init(struct trace_array *tr)
1440 {
1441         blk_tr = tr;
1442         blk_tracer_start(tr);
1443         return 0;
1444 }
1445
1446 static void blk_tracer_stop(struct trace_array *tr)
1447 {
1448         blk_tracer_enabled = false;
1449 }
1450
1451 static void blk_tracer_reset(struct trace_array *tr)
1452 {
1453         blk_tracer_stop(tr);
1454 }
1455
1456 static const struct {
1457         const char *act[2];
1458         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1459                             bool has_cg);
1460 } what2act[] = {
1461         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1462         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1463         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1464         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1465         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1466         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1467         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1468         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1469         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1470         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1471         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1472         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1473         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1474         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1475         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1476 };
1477
1478 static enum print_line_t print_one_line(struct trace_iterator *iter,
1479                                         bool classic)
1480 {
1481         struct trace_array *tr = iter->tr;
1482         struct trace_seq *s = &iter->seq;
1483         const struct blk_io_trace *t;
1484         u16 what;
1485         bool long_act;
1486         blk_log_action_t *log_action;
1487         bool has_cg;
1488
1489         t          = te_blk_io_trace(iter->ent);
1490         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1491         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1492         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1493         has_cg     = t->action & __BLK_TA_CGROUP;
1494
1495         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1496                 log_action(iter, long_act ? "message" : "m", has_cg);
1497                 blk_log_msg(s, iter->ent, has_cg);
1498                 return trace_handle_return(s);
1499         }
1500
1501         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1502                 trace_seq_printf(s, "Unknown action %x\n", what);
1503         else {
1504                 log_action(iter, what2act[what].act[long_act], has_cg);
1505                 what2act[what].print(s, iter->ent, has_cg);
1506         }
1507
1508         return trace_handle_return(s);
1509 }
1510
1511 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1512                                                int flags, struct trace_event *event)
1513 {
1514         return print_one_line(iter, false);
1515 }
1516
1517 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1518 {
1519         struct trace_seq *s = &iter->seq;
1520         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1521         const int offset = offsetof(struct blk_io_trace, sector);
1522         struct blk_io_trace old = {
1523                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1524                 .time     = iter->ts,
1525         };
1526
1527         trace_seq_putmem(s, &old, offset);
1528         trace_seq_putmem(s, &t->sector,
1529                          sizeof(old) - offset + t->pdu_len);
1530 }
1531
1532 static enum print_line_t
1533 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1534                              struct trace_event *event)
1535 {
1536         blk_trace_synthesize_old_trace(iter);
1537
1538         return trace_handle_return(&iter->seq);
1539 }
1540
1541 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1542 {
1543         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1544                 return TRACE_TYPE_UNHANDLED;
1545
1546         return print_one_line(iter, true);
1547 }
1548
1549 static int
1550 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1551 {
1552         /* don't output context-info for blk_classic output */
1553         if (bit == TRACE_BLK_OPT_CLASSIC) {
1554                 if (set)
1555                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1556                 else
1557                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1558         }
1559         return 0;
1560 }
1561
1562 static struct tracer blk_tracer __read_mostly = {
1563         .name           = "blk",
1564         .init           = blk_tracer_init,
1565         .reset          = blk_tracer_reset,
1566         .start          = blk_tracer_start,
1567         .stop           = blk_tracer_stop,
1568         .print_header   = blk_tracer_print_header,
1569         .print_line     = blk_tracer_print_line,
1570         .flags          = &blk_tracer_flags,
1571         .set_flag       = blk_tracer_set_flag,
1572 };
1573
1574 static struct trace_event_functions trace_blk_event_funcs = {
1575         .trace          = blk_trace_event_print,
1576         .binary         = blk_trace_event_print_binary,
1577 };
1578
1579 static struct trace_event trace_blk_event = {
1580         .type           = TRACE_BLK,
1581         .funcs          = &trace_blk_event_funcs,
1582 };
1583
1584 static int __init init_blk_tracer(void)
1585 {
1586         if (!register_trace_event(&trace_blk_event)) {
1587                 pr_warn("Warning: could not register block events\n");
1588                 return 1;
1589         }
1590
1591         if (register_tracer(&blk_tracer) != 0) {
1592                 pr_warn("Warning: could not register the block tracer\n");
1593                 unregister_trace_event(&trace_blk_event);
1594                 return 1;
1595         }
1596
1597         return 0;
1598 }
1599
1600 device_initcall(init_blk_tracer);
1601
1602 static int blk_trace_remove_queue(struct request_queue *q)
1603 {
1604         struct blk_trace *bt;
1605
1606         bt = rcu_replace_pointer(q->blk_trace, NULL,
1607                                  lockdep_is_held(&q->debugfs_mutex));
1608         if (bt == NULL)
1609                 return -EINVAL;
1610
1611         put_probe_ref();
1612         synchronize_rcu();
1613         blk_trace_free(bt);
1614         return 0;
1615 }
1616
1617 /*
1618  * Setup everything required to start tracing
1619  */
1620 static int blk_trace_setup_queue(struct request_queue *q,
1621                                  struct block_device *bdev)
1622 {
1623         struct blk_trace *bt = NULL;
1624         int ret = -ENOMEM;
1625
1626         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1627         if (!bt)
1628                 return -ENOMEM;
1629
1630         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1631         if (!bt->msg_data)
1632                 goto free_bt;
1633
1634         bt->dev = bdev->bd_dev;
1635         bt->act_mask = (u16)-1;
1636
1637         blk_trace_setup_lba(bt, bdev);
1638
1639         rcu_assign_pointer(q->blk_trace, bt);
1640         get_probe_ref();
1641         return 0;
1642
1643 free_bt:
1644         blk_trace_free(bt);
1645         return ret;
1646 }
1647
1648 /*
1649  * sysfs interface to enable and configure tracing
1650  */
1651
1652 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1653                                          struct device_attribute *attr,
1654                                          char *buf);
1655 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1656                                           struct device_attribute *attr,
1657                                           const char *buf, size_t count);
1658 #define BLK_TRACE_DEVICE_ATTR(_name) \
1659         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1660                     sysfs_blk_trace_attr_show, \
1661                     sysfs_blk_trace_attr_store)
1662
1663 static BLK_TRACE_DEVICE_ATTR(enable);
1664 static BLK_TRACE_DEVICE_ATTR(act_mask);
1665 static BLK_TRACE_DEVICE_ATTR(pid);
1666 static BLK_TRACE_DEVICE_ATTR(start_lba);
1667 static BLK_TRACE_DEVICE_ATTR(end_lba);
1668
1669 static struct attribute *blk_trace_attrs[] = {
1670         &dev_attr_enable.attr,
1671         &dev_attr_act_mask.attr,
1672         &dev_attr_pid.attr,
1673         &dev_attr_start_lba.attr,
1674         &dev_attr_end_lba.attr,
1675         NULL
1676 };
1677
1678 struct attribute_group blk_trace_attr_group = {
1679         .name  = "trace",
1680         .attrs = blk_trace_attrs,
1681 };
1682
1683 static const struct {
1684         int mask;
1685         const char *str;
1686 } mask_maps[] = {
1687         { BLK_TC_READ,          "read"          },
1688         { BLK_TC_WRITE,         "write"         },
1689         { BLK_TC_FLUSH,         "flush"         },
1690         { BLK_TC_SYNC,          "sync"          },
1691         { BLK_TC_QUEUE,         "queue"         },
1692         { BLK_TC_REQUEUE,       "requeue"       },
1693         { BLK_TC_ISSUE,         "issue"         },
1694         { BLK_TC_COMPLETE,      "complete"      },
1695         { BLK_TC_FS,            "fs"            },
1696         { BLK_TC_PC,            "pc"            },
1697         { BLK_TC_NOTIFY,        "notify"        },
1698         { BLK_TC_AHEAD,         "ahead"         },
1699         { BLK_TC_META,          "meta"          },
1700         { BLK_TC_DISCARD,       "discard"       },
1701         { BLK_TC_DRV_DATA,      "drv_data"      },
1702         { BLK_TC_FUA,           "fua"           },
1703 };
1704
1705 static int blk_trace_str2mask(const char *str)
1706 {
1707         int i;
1708         int mask = 0;
1709         char *buf, *s, *token;
1710
1711         buf = kstrdup(str, GFP_KERNEL);
1712         if (buf == NULL)
1713                 return -ENOMEM;
1714         s = strstrip(buf);
1715
1716         while (1) {
1717                 token = strsep(&s, ",");
1718                 if (token == NULL)
1719                         break;
1720
1721                 if (*token == '\0')
1722                         continue;
1723
1724                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1725                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1726                                 mask |= mask_maps[i].mask;
1727                                 break;
1728                         }
1729                 }
1730                 if (i == ARRAY_SIZE(mask_maps)) {
1731                         mask = -EINVAL;
1732                         break;
1733                 }
1734         }
1735         kfree(buf);
1736
1737         return mask;
1738 }
1739
1740 static ssize_t blk_trace_mask2str(char *buf, int mask)
1741 {
1742         int i;
1743         char *p = buf;
1744
1745         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1746                 if (mask & mask_maps[i].mask) {
1747                         p += sprintf(p, "%s%s",
1748                                     (p == buf) ? "" : ",", mask_maps[i].str);
1749                 }
1750         }
1751         *p++ = '\n';
1752
1753         return p - buf;
1754 }
1755
1756 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1757                                          struct device_attribute *attr,
1758                                          char *buf)
1759 {
1760         struct block_device *bdev = dev_to_bdev(dev);
1761         struct request_queue *q = bdev_get_queue(bdev);
1762         struct blk_trace *bt;
1763         ssize_t ret = -ENXIO;
1764
1765         mutex_lock(&q->debugfs_mutex);
1766
1767         bt = rcu_dereference_protected(q->blk_trace,
1768                                        lockdep_is_held(&q->debugfs_mutex));
1769         if (attr == &dev_attr_enable) {
1770                 ret = sprintf(buf, "%u\n", !!bt);
1771                 goto out_unlock_bdev;
1772         }
1773
1774         if (bt == NULL)
1775                 ret = sprintf(buf, "disabled\n");
1776         else if (attr == &dev_attr_act_mask)
1777                 ret = blk_trace_mask2str(buf, bt->act_mask);
1778         else if (attr == &dev_attr_pid)
1779                 ret = sprintf(buf, "%u\n", bt->pid);
1780         else if (attr == &dev_attr_start_lba)
1781                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1782         else if (attr == &dev_attr_end_lba)
1783                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1784
1785 out_unlock_bdev:
1786         mutex_unlock(&q->debugfs_mutex);
1787         return ret;
1788 }
1789
1790 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1791                                           struct device_attribute *attr,
1792                                           const char *buf, size_t count)
1793 {
1794         struct block_device *bdev = dev_to_bdev(dev);
1795         struct request_queue *q = bdev_get_queue(bdev);
1796         struct blk_trace *bt;
1797         u64 value;
1798         ssize_t ret = -EINVAL;
1799
1800         if (count == 0)
1801                 goto out;
1802
1803         if (attr == &dev_attr_act_mask) {
1804                 if (kstrtoull(buf, 0, &value)) {
1805                         /* Assume it is a list of trace category names */
1806                         ret = blk_trace_str2mask(buf);
1807                         if (ret < 0)
1808                                 goto out;
1809                         value = ret;
1810                 }
1811         } else {
1812                 if (kstrtoull(buf, 0, &value))
1813                         goto out;
1814         }
1815
1816         mutex_lock(&q->debugfs_mutex);
1817
1818         bt = rcu_dereference_protected(q->blk_trace,
1819                                        lockdep_is_held(&q->debugfs_mutex));
1820         if (attr == &dev_attr_enable) {
1821                 if (!!value == !!bt) {
1822                         ret = 0;
1823                         goto out_unlock_bdev;
1824                 }
1825                 if (value)
1826                         ret = blk_trace_setup_queue(q, bdev);
1827                 else
1828                         ret = blk_trace_remove_queue(q);
1829                 goto out_unlock_bdev;
1830         }
1831
1832         ret = 0;
1833         if (bt == NULL) {
1834                 ret = blk_trace_setup_queue(q, bdev);
1835                 bt = rcu_dereference_protected(q->blk_trace,
1836                                 lockdep_is_held(&q->debugfs_mutex));
1837         }
1838
1839         if (ret == 0) {
1840                 if (attr == &dev_attr_act_mask)
1841                         bt->act_mask = value;
1842                 else if (attr == &dev_attr_pid)
1843                         bt->pid = value;
1844                 else if (attr == &dev_attr_start_lba)
1845                         bt->start_lba = value;
1846                 else if (attr == &dev_attr_end_lba)
1847                         bt->end_lba = value;
1848         }
1849
1850 out_unlock_bdev:
1851         mutex_unlock(&q->debugfs_mutex);
1852 out:
1853         return ret ? ret : count;
1854 }
1855
1856 int blk_trace_init_sysfs(struct device *dev)
1857 {
1858         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1859 }
1860
1861 void blk_trace_remove_sysfs(struct device *dev)
1862 {
1863         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1864 }
1865
1866 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1867
1868 #ifdef CONFIG_EVENT_TRACING
1869
1870 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1871 {
1872         int i = 0;
1873
1874         if (op & REQ_PREFLUSH)
1875                 rwbs[i++] = 'F';
1876
1877         switch (op & REQ_OP_MASK) {
1878         case REQ_OP_WRITE:
1879         case REQ_OP_WRITE_SAME:
1880                 rwbs[i++] = 'W';
1881                 break;
1882         case REQ_OP_DISCARD:
1883                 rwbs[i++] = 'D';
1884                 break;
1885         case REQ_OP_SECURE_ERASE:
1886                 rwbs[i++] = 'D';
1887                 rwbs[i++] = 'E';
1888                 break;
1889         case REQ_OP_FLUSH:
1890                 rwbs[i++] = 'F';
1891                 break;
1892         case REQ_OP_READ:
1893                 rwbs[i++] = 'R';
1894                 break;
1895         default:
1896                 rwbs[i++] = 'N';
1897         }
1898
1899         if (op & REQ_FUA)
1900                 rwbs[i++] = 'F';
1901         if (op & REQ_RAHEAD)
1902                 rwbs[i++] = 'A';
1903         if (op & REQ_SYNC)
1904                 rwbs[i++] = 'S';
1905         if (op & REQ_META)
1906                 rwbs[i++] = 'M';
1907
1908         rwbs[i] = '\0';
1909 }
1910 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1911
1912 #endif /* CONFIG_EVENT_TRACING */
1913