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