kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <[email protected]>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License version 2 as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  16  *
  17  */
  18 #include <linux/kernel.h>
  19 #include <linux/blkdev.h>
  20 #include <linux/blktrace_api.h>
  21 #include <linux/percpu.h>
  22 #include <linux/init.h>
  23 #include <linux/mutex.h>
  24 #include <linux/slab.h>
  25 #include <linux/debugfs.h>
  26 #include <linux/export.h>
  27 #include <linux/time.h>
  28 #include <linux/uaccess.h>
  29 #include <linux/list.h>
  30 #include <linux/blk-cgroup.h>
  31
  32 #include "../../block/blk.h"
  33
  34 #include <trace/events/block.h>
  35
  36 #include "trace_output.h"
  37
  38 #ifdef CONFIG_BLK_DEV_IO_TRACE
  39
  40 static unsigned int blktrace_seq __read_mostly = 1;
  41
  42 static struct trace_array *blk_tr;
  43 static bool blk_tracer_enabled __read_mostly;
  44
  45 static LIST_HEAD(running_trace_list);
  46 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  47
  48 /* Select an alternative, minimalistic output than the original one */
  49 #define TRACE_BLK_OPT_CLASSIC   0x1
  50 #define TRACE_BLK_OPT_CGROUP    0x2
  51 #define TRACE_BLK_OPT_CGNAME    0x4
  52
  53 static struct tracer_opt blk_tracer_opts[] = {
  54         /* Default disable the minimalistic output */
  55         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  56 #ifdef CONFIG_BLK_CGROUP
  57         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  58         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  59 #endif
  60         { }
  61 };
  62
  63 static struct tracer_flags blk_tracer_flags = {
  64         .val  = 0,
  65         .opts = blk_tracer_opts,
  66 };
  67
  68 /* Global reference count of probes */
  69 static DEFINE_MUTEX(blk_probe_mutex);
  70 static int blk_probes_ref;
  71
  72 static void blk_register_tracepoints(void);
  73 static void blk_unregister_tracepoints(void);
  74
  75 /*
  76  * Send out a notify message.
  77  */
  78 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  79                        const void *data, size_t len,
  80                        union kernfs_node_id *cgid)
  81 {
  82         struct blk_io_trace *t;
  83         struct ring_buffer_event *event = NULL;
  84         struct ring_buffer *buffer = NULL;
  85         int pc = 0;
  86         int cpu = smp_processor_id();
  87         bool blk_tracer = blk_tracer_enabled;
  88         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  89
  90         if (blk_tracer) {
  91                 buffer = blk_tr->trace_buffer.buffer;
  92                 pc = preempt_count();
  93                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  94                                                   sizeof(*t) + len + cgid_len,
  95                                                   0, pc);
  96                 if (!event)
  97                         return;
  98                 t = ring_buffer_event_data(event);
  99                 goto record_it;
 100         }
 101
 102         if (!bt->rchan)
 103                 return;
 104
 105         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
 106         if (t) {
 107                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 108                 t->time = ktime_to_ns(ktime_get());
 109 record_it:
 110                 t->device = bt->dev;
 111                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 112                 t->pid = pid;
 113                 t->cpu = cpu;
 114                 t->pdu_len = len + cgid_len;
 115                 if (cgid)
 116                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 117                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 118
 119                 if (blk_tracer)
 120                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 121         }
 122 }
 123
 124 /*
 125  * Send out a notify for this process, if we haven't done so since a trace
 126  * started
 127  */
 128 static void trace_note_tsk(struct task_struct *tsk)
 129 {
 130         unsigned long flags;
 131         struct blk_trace *bt;
 132
 133         tsk->btrace_seq = blktrace_seq;
 134         spin_lock_irqsave(&running_trace_lock, flags);
 135         list_for_each_entry(bt, &running_trace_list, running_list) {
 136                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 137                            sizeof(tsk->comm), NULL);
 138         }
 139         spin_unlock_irqrestore(&running_trace_lock, flags);
 140 }
 141
 142 static void trace_note_time(struct blk_trace *bt)
 143 {
 144         struct timespec64 now;
 145         unsigned long flags;
 146         u32 words[2];
 147
 148         /* need to check user space to see if this breaks in y2038 or y2106 */
 149         ktime_get_real_ts64(&now);
 150         words[0] = (u32)now.tv_sec;
 151         words[1] = now.tv_nsec;
 152
 153         local_irq_save(flags);
 154         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 155         local_irq_restore(flags);
 156 }
 157
 158 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 159         const char *fmt, ...)
 160 {
 161         int n;
 162         va_list args;
 163         unsigned long flags;
 164         char *buf;
 165
 166         if (unlikely(bt->trace_state != Blktrace_running &&
 167                      !blk_tracer_enabled))
 168                 return;
 169
 170         /*
 171          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 172          * message to the trace.
 173          */
 174         if (!(bt->act_mask & BLK_TC_NOTIFY))
 175                 return;
 176
 177         local_irq_save(flags);
 178         buf = this_cpu_ptr(bt->msg_data);
 179         va_start(args, fmt);
 180         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 181         va_end(args);
 182
 183         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 184                 blkcg = NULL;
 185 #ifdef CONFIG_BLK_CGROUP
 186         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 187                 blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 188 #else
 189         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 190 #endif
 191         local_irq_restore(flags);
 192 }
 193 EXPORT_SYMBOL_GPL(__trace_note_message);
 194
 195 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 196                          pid_t pid)
 197 {
 198         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 199                 return 1;
 200         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 201                 return 1;
 202         if (bt->pid && pid != bt->pid)
 203                 return 1;
 204
 205         return 0;
 206 }
 207
 208 /*
 209  * Data direction bit lookup
 210  */
 211 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 212                                  BLK_TC_ACT(BLK_TC_WRITE) };
 213
 214 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 215 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 216
 217 /* The ilog2() calls fall out because they're constant */
 218 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 219           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 220
 221 /*
 222  * The worker for the various blk_add_trace*() types. Fills out a
 223  * blk_io_trace structure and places it in a per-cpu subbuffer.
 224  */
 225 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 226                      int op, int op_flags, u32 what, int error, int pdu_len,
 227                      void *pdu_data, union kernfs_node_id *cgid)
 228 {
 229         struct task_struct *tsk = current;
 230         struct ring_buffer_event *event = NULL;
 231         struct ring_buffer *buffer = NULL;
 232         struct blk_io_trace *t;
 233         unsigned long flags = 0;
 234         unsigned long *sequence;
 235         pid_t pid;
 236         int cpu, pc = 0;
 237         bool blk_tracer = blk_tracer_enabled;
 238         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 239
 240         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 241                 return;
 242
 243         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 244         what |= MASK_TC_BIT(op_flags, SYNC);
 245         what |= MASK_TC_BIT(op_flags, RAHEAD);
 246         what |= MASK_TC_BIT(op_flags, META);
 247         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 248         what |= MASK_TC_BIT(op_flags, FUA);
 249         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 250                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 251         if (op == REQ_OP_FLUSH)
 252                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 253         if (cgid)
 254                 what |= __BLK_TA_CGROUP;
 255
 256         pid = tsk->pid;
 257         if (act_log_check(bt, what, sector, pid))
 258                 return;
 259         cpu = raw_smp_processor_id();
 260
 261         if (blk_tracer) {
 262                 tracing_record_cmdline(current);
 263
 264                 buffer = blk_tr->trace_buffer.buffer;
 265                 pc = preempt_count();
 266                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 267                                                   sizeof(*t) + pdu_len + cgid_len,
 268                                                   0, pc);
 269                 if (!event)
 270                         return;
 271                 t = ring_buffer_event_data(event);
 272                 goto record_it;
 273         }
 274
 275         if (unlikely(tsk->btrace_seq != blktrace_seq))
 276                 trace_note_tsk(tsk);
 277
 278         /*
 279          * A word about the locking here - we disable interrupts to reserve
 280          * some space in the relay per-cpu buffer, to prevent an irq
 281          * from coming in and stepping on our toes.
 282          */
 283         local_irq_save(flags);
 284         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 285         if (t) {
 286                 sequence = per_cpu_ptr(bt->sequence, cpu);
 287
 288                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 289                 t->sequence = ++(*sequence);
 290                 t->time = ktime_to_ns(ktime_get());
 291 record_it:
 292                 /*
 293                  * These two are not needed in ftrace as they are in the
 294                  * generic trace_entry, filled by tracing_generic_entry_update,
 295                  * but for the trace_event->bin() synthesizer benefit we do it
 296                  * here too.
 297                  */
 298                 t->cpu = cpu;
 299                 t->pid = pid;
 300
 301                 t->sector = sector;
 302                 t->bytes = bytes;
 303                 t->action = what;
 304                 t->device = bt->dev;
 305                 t->error = error;
 306                 t->pdu_len = pdu_len + cgid_len;
 307
 308                 if (cgid_len)
 309                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 310                 if (pdu_len)
 311                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 312
 313                 if (blk_tracer) {
 314                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 315                         return;
 316                 }
 317         }
 318
 319         local_irq_restore(flags);
 320 }
 321
 322 static void blk_trace_free(struct blk_trace *bt)
 323 {
 324         debugfs_remove(bt->msg_file);
 325         debugfs_remove(bt->dropped_file);
 326         relay_close(bt->rchan);
 327         debugfs_remove(bt->dir);
 328         free_percpu(bt->sequence);
 329         free_percpu(bt->msg_data);
 330         kfree(bt);
 331 }
 332
 333 static void get_probe_ref(void)
 334 {
 335         mutex_lock(&blk_probe_mutex);
 336         if (++blk_probes_ref == 1)
 337                 blk_register_tracepoints();
 338         mutex_unlock(&blk_probe_mutex);
 339 }
 340
 341 static void put_probe_ref(void)
 342 {
 343         mutex_lock(&blk_probe_mutex);
 344         if (!--blk_probes_ref)
 345                 blk_unregister_tracepoints();
 346         mutex_unlock(&blk_probe_mutex);
 347 }
 348
 349 static void blk_trace_cleanup(struct blk_trace *bt)
 350 {
 351         blk_trace_free(bt);
 352         put_probe_ref();
 353 }
 354
 355 static int __blk_trace_remove(struct request_queue *q)
 356 {
 357         struct blk_trace *bt;
 358
 359         bt = xchg(&q->blk_trace, NULL);
 360         if (!bt)
 361                 return -EINVAL;
 362
 363         if (bt->trace_state != Blktrace_running)
 364                 blk_trace_cleanup(bt);
 365
 366         return 0;
 367 }
 368
 369 int blk_trace_remove(struct request_queue *q)
 370 {
 371         int ret;
 372
 373         mutex_lock(&q->blk_trace_mutex);
 374         ret = __blk_trace_remove(q);
 375         mutex_unlock(&q->blk_trace_mutex);
 376
 377         return ret;
 378 }
 379 EXPORT_SYMBOL_GPL(blk_trace_remove);
 380
 381 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 382                                 size_t count, loff_t *ppos)
 383 {
 384         struct blk_trace *bt = filp->private_data;
 385         char buf[16];
 386
 387         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 388
 389         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 390 }
 391
 392 static const struct file_operations blk_dropped_fops = {
 393         .owner =        THIS_MODULE,
 394         .open =         simple_open,
 395         .read =         blk_dropped_read,
 396         .llseek =       default_llseek,
 397 };
 398
 399 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 400                                 size_t count, loff_t *ppos)
 401 {
 402         char *msg;
 403         struct blk_trace *bt;
 404
 405         if (count >= BLK_TN_MAX_MSG)
 406                 return -EINVAL;
 407
 408         msg = memdup_user_nul(buffer, count);
 409         if (IS_ERR(msg))
 410                 return PTR_ERR(msg);
 411
 412         bt = filp->private_data;
 413         __trace_note_message(bt, NULL, "%s", msg);
 414         kfree(msg);
 415
 416         return count;
 417 }
 418
 419 static const struct file_operations blk_msg_fops = {
 420         .owner =        THIS_MODULE,
 421         .open =         simple_open,
 422         .write =        blk_msg_write,
 423         .llseek =       noop_llseek,
 424 };
 425
 426 /*
 427  * Keep track of how many times we encountered a full subbuffer, to aid
 428  * the user space app in telling how many lost events there were.
 429  */
 430 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 431                                      void *prev_subbuf, size_t prev_padding)
 432 {
 433         struct blk_trace *bt;
 434
 435         if (!relay_buf_full(buf))
 436                 return 1;
 437
 438         bt = buf->chan->private_data;
 439         atomic_inc(&bt->dropped);
 440         return 0;
 441 }
 442
 443 static int blk_remove_buf_file_callback(struct dentry *dentry)
 444 {
 445         debugfs_remove(dentry);
 446
 447         return 0;
 448 }
 449
 450 static struct dentry *blk_create_buf_file_callback(const char *filename,
 451                                                    struct dentry *parent,
 452                                                    umode_t mode,
 453                                                    struct rchan_buf *buf,
 454                                                    int *is_global)
 455 {
 456         return debugfs_create_file(filename, mode, parent, buf,
 457                                         &relay_file_operations);
 458 }
 459
 460 static struct rchan_callbacks blk_relay_callbacks = {
 461         .subbuf_start           = blk_subbuf_start_callback,
 462         .create_buf_file        = blk_create_buf_file_callback,
 463         .remove_buf_file        = blk_remove_buf_file_callback,
 464 };
 465
 466 static void blk_trace_setup_lba(struct blk_trace *bt,
 467                                 struct block_device *bdev)
 468 {
 469         struct hd_struct *part = NULL;
 470
 471         if (bdev)
 472                 part = bdev->bd_part;
 473
 474         if (part) {
 475                 bt->start_lba = part->start_sect;
 476                 bt->end_lba = part->start_sect + part->nr_sects;
 477         } else {
 478                 bt->start_lba = 0;
 479                 bt->end_lba = -1ULL;
 480         }
 481 }
 482
 483 /*
 484  * Setup everything required to start tracing
 485  */
 486 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 487                               struct block_device *bdev,
 488                               struct blk_user_trace_setup *buts)
 489 {
 490         struct blk_trace *bt = NULL;
 491         struct dentry *dir = NULL;
 492         int ret;
 493
 494         if (!buts->buf_size || !buts->buf_nr)
 495                 return -EINVAL;
 496
 497         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 498         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 499
 500         /*
 501          * some device names have larger paths - convert the slashes
 502          * to underscores for this to work as expected
 503          */
 504         strreplace(buts->name, '/', '_');
 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         ret = -ENOENT;
 520
 521         if (!blk_debugfs_root)
 522                 goto err;
 523
 524         dir = debugfs_lookup(buts->name, blk_debugfs_root);
 525         if (!dir)
 526                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 527         if (!dir)
 528                 goto err;
 529
 530         bt->dev = dev;
 531         atomic_set(&bt->dropped, 0);
 532         INIT_LIST_HEAD(&bt->running_list);
 533
 534         ret = -EIO;
 535         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 536                                                &blk_dropped_fops);
 537         if (!bt->dropped_file)
 538                 goto err;
 539
 540         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 541         if (!bt->msg_file)
 542                 goto err;
 543
 544         bt->rchan = relay_open("trace", dir, buts->buf_size,
 545                                 buts->buf_nr, &blk_relay_callbacks, bt);
 546         if (!bt->rchan)
 547                 goto err;
 548
 549         bt->act_mask = buts->act_mask;
 550         if (!bt->act_mask)
 551                 bt->act_mask = (u16) -1;
 552
 553         blk_trace_setup_lba(bt, bdev);
 554
 555         /* overwrite with user settings */
 556         if (buts->start_lba)
 557                 bt->start_lba = buts->start_lba;
 558         if (buts->end_lba)
 559                 bt->end_lba = buts->end_lba;
 560
 561         bt->pid = buts->pid;
 562         bt->trace_state = Blktrace_setup;
 563
 564         ret = -EBUSY;
 565         if (cmpxchg(&q->blk_trace, NULL, bt))
 566                 goto err;
 567
 568         get_probe_ref();
 569
 570         ret = 0;
 571 err:
 572         if (dir && !bt->dir)
 573                 dput(dir);
 574         if (ret)
 575                 blk_trace_free(bt);
 576         return ret;
 577 }
 578
 579 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 580                              struct block_device *bdev, char __user *arg)
 581 {
 582         struct blk_user_trace_setup buts;
 583         int ret;
 584
 585         ret = copy_from_user(&buts, arg, sizeof(buts));
 586         if (ret)
 587                 return -EFAULT;
 588
 589         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 590         if (ret)
 591                 return ret;
 592
 593         if (copy_to_user(arg, &buts, sizeof(buts))) {
 594                 blk_trace_remove(q);
 595                 return -EFAULT;
 596         }
 597         return 0;
 598 }
 599
 600 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 601                     struct block_device *bdev,
 602                     char __user *arg)
 603 {
 604         int ret;
 605
 606         mutex_lock(&q->blk_trace_mutex);
 607         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 608         mutex_unlock(&q->blk_trace_mutex);
 609
 610         return ret;
 611 }
 612 EXPORT_SYMBOL_GPL(blk_trace_setup);
 613
 614 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 615 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 616                                   dev_t dev, struct block_device *bdev,
 617                                   char __user *arg)
 618 {
 619         struct blk_user_trace_setup buts;
 620         struct compat_blk_user_trace_setup cbuts;
 621         int ret;
 622
 623         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 624                 return -EFAULT;
 625
 626         buts = (struct blk_user_trace_setup) {
 627                 .act_mask = cbuts.act_mask,
 628                 .buf_size = cbuts.buf_size,
 629                 .buf_nr = cbuts.buf_nr,
 630                 .start_lba = cbuts.start_lba,
 631                 .end_lba = cbuts.end_lba,
 632                 .pid = cbuts.pid,
 633         };
 634
 635         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 636         if (ret)
 637                 return ret;
 638
 639         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 640                 blk_trace_remove(q);
 641                 return -EFAULT;
 642         }
 643
 644         return 0;
 645 }
 646 #endif
 647
 648 static int __blk_trace_startstop(struct request_queue *q, int start)
 649 {
 650         int ret;
 651         struct blk_trace *bt = q->blk_trace;
 652
 653         if (bt == NULL)
 654                 return -EINVAL;
 655
 656         /*
 657          * For starting a trace, we can transition from a setup or stopped
 658          * trace. For stopping a trace, the state must be running
 659          */
 660         ret = -EINVAL;
 661         if (start) {
 662                 if (bt->trace_state == Blktrace_setup ||
 663                     bt->trace_state == Blktrace_stopped) {
 664                         blktrace_seq++;
 665                         smp_mb();
 666                         bt->trace_state = Blktrace_running;
 667                         spin_lock_irq(&running_trace_lock);
 668                         list_add(&bt->running_list, &running_trace_list);
 669                         spin_unlock_irq(&running_trace_lock);
 670
 671                         trace_note_time(bt);
 672                         ret = 0;
 673                 }
 674         } else {
 675                 if (bt->trace_state == Blktrace_running) {
 676                         bt->trace_state = Blktrace_stopped;
 677                         spin_lock_irq(&running_trace_lock);
 678                         list_del_init(&bt->running_list);
 679                         spin_unlock_irq(&running_trace_lock);
 680                         relay_flush(bt->rchan);
 681                         ret = 0;
 682                 }
 683         }
 684
 685         return ret;
 686 }
 687
 688 int blk_trace_startstop(struct request_queue *q, int start)
 689 {
 690         int ret;
 691
 692         mutex_lock(&q->blk_trace_mutex);
 693         ret = __blk_trace_startstop(q, start);
 694         mutex_unlock(&q->blk_trace_mutex);
 695
 696         return ret;
 697 }
 698 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 699
 700 /*
 701  * When reading or writing the blktrace sysfs files, the references to the
 702  * opened sysfs or device files should prevent the underlying block device
 703  * from being removed. So no further delete protection is really needed.
 704  */
 705
 706 /**
 707  * blk_trace_ioctl: - handle the ioctls associated with tracing
 708  * @bdev:       the block device
 709  * @cmd:        the ioctl cmd
 710  * @arg:        the argument data, if any
 711  *
 712  **/
 713 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 714 {
 715         struct request_queue *q;
 716         int ret, start = 0;
 717         char b[BDEVNAME_SIZE];
 718
 719         q = bdev_get_queue(bdev);
 720         if (!q)
 721                 return -ENXIO;
 722
 723         mutex_lock(&q->blk_trace_mutex);
 724
 725         switch (cmd) {
 726         case BLKTRACESETUP:
 727                 bdevname(bdev, b);
 728                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 729                 break;
 730 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 731         case BLKTRACESETUP32:
 732                 bdevname(bdev, b);
 733                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 734                 break;
 735 #endif
 736         case BLKTRACESTART:
 737                 start = 1;
 738         case BLKTRACESTOP:
 739                 ret = __blk_trace_startstop(q, start);
 740                 break;
 741         case BLKTRACETEARDOWN:
 742                 ret = __blk_trace_remove(q);
 743                 break;
 744         default:
 745                 ret = -ENOTTY;
 746                 break;
 747         }
 748
 749         mutex_unlock(&q->blk_trace_mutex);
 750         return ret;
 751 }
 752
 753 /**
 754  * blk_trace_shutdown: - stop and cleanup trace structures
 755  * @q:    the request queue associated with the device
 756  *
 757  **/
 758 void blk_trace_shutdown(struct request_queue *q)
 759 {
 760         mutex_lock(&q->blk_trace_mutex);
 761
 762         if (q->blk_trace) {
 763                 __blk_trace_startstop(q, 0);
 764                 __blk_trace_remove(q);
 765         }
 766
 767         mutex_unlock(&q->blk_trace_mutex);
 768 }
 769
 770 #ifdef CONFIG_BLK_CGROUP
 771 static union kernfs_node_id *
 772 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 773 {
 774         struct blk_trace *bt = q->blk_trace;
 775
 776         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 777                 return NULL;
 778
 779         if (!bio->bi_css)
 780                 return NULL;
 781         return cgroup_get_kernfs_id(bio->bi_css->cgroup);
 782 }
 783 #else
 784 static union kernfs_node_id *
 785 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 786 {
 787         return NULL;
 788 }
 789 #endif
 790
 791 static union kernfs_node_id *
 792 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 793 {
 794         if (!rq->bio)
 795                 return NULL;
 796         /* Use the first bio */
 797         return blk_trace_bio_get_cgid(q, rq->bio);
 798 }
 799
 800 /*
 801  * blktrace probes
 802  */
 803
 804 /**
 805  * blk_add_trace_rq - Add a trace for a request oriented action
 806  * @rq:         the source request
 807  * @error:      return status to log
 808  * @nr_bytes:   number of completed bytes
 809  * @what:       the action
 810  * @cgid:       the cgroup info
 811  *
 812  * Description:
 813  *     Records an action against a request. Will log the bio offset + size.
 814  *
 815  **/
 816 static void blk_add_trace_rq(struct request *rq, int error,
 817                              unsigned int nr_bytes, u32 what,
 818                              union kernfs_node_id *cgid)
 819 {
 820         struct blk_trace *bt = rq->q->blk_trace;
 821
 822         if (likely(!bt))
 823                 return;
 824
 825         if (blk_rq_is_passthrough(rq))
 826                 what |= BLK_TC_ACT(BLK_TC_PC);
 827         else
 828                 what |= BLK_TC_ACT(BLK_TC_FS);
 829
 830         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 831                         rq->cmd_flags, what, error, 0, NULL, cgid);
 832 }
 833
 834 static void blk_add_trace_rq_insert(void *ignore,
 835                                     struct request_queue *q, struct request *rq)
 836 {
 837         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 838                          blk_trace_request_get_cgid(q, rq));
 839 }
 840
 841 static void blk_add_trace_rq_issue(void *ignore,
 842                                    struct request_queue *q, struct request *rq)
 843 {
 844         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 845                          blk_trace_request_get_cgid(q, rq));
 846 }
 847
 848 static void blk_add_trace_rq_requeue(void *ignore,
 849                                      struct request_queue *q,
 850                                      struct request *rq)
 851 {
 852         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 853                          blk_trace_request_get_cgid(q, rq));
 854 }
 855
 856 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 857                         int error, unsigned int nr_bytes)
 858 {
 859         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 860                          blk_trace_request_get_cgid(rq->q, rq));
 861 }
 862
 863 /**
 864  * blk_add_trace_bio - Add a trace for a bio oriented action
 865  * @q:          queue the io is for
 866  * @bio:        the source bio
 867  * @what:       the action
 868  * @error:      error, if any
 869  *
 870  * Description:
 871  *     Records an action against a bio. Will log the bio offset + size.
 872  *
 873  **/
 874 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 875                               u32 what, int error, union kernfs_node_id *cgid)
 876 {
 877         struct blk_trace *bt = q->blk_trace;
 878
 879         if (likely(!bt))
 880                 return;
 881
 882         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 883                         bio_op(bio), bio->bi_opf, what, error, 0, NULL, cgid);
 884 }
 885
 886 static void blk_add_trace_bio_bounce(void *ignore,
 887                                      struct request_queue *q, struct bio *bio)
 888 {
 889         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0,
 890                           blk_trace_bio_get_cgid(q, bio));
 891 }
 892
 893 static void blk_add_trace_bio_complete(void *ignore,
 894                                        struct request_queue *q, struct bio *bio,
 895                                        int error)
 896 {
 897         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error,
 898                           blk_trace_bio_get_cgid(q, bio));
 899 }
 900
 901 static void blk_add_trace_bio_backmerge(void *ignore,
 902                                         struct request_queue *q,
 903                                         struct request *rq,
 904                                         struct bio *bio)
 905 {
 906         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0,
 907                          blk_trace_bio_get_cgid(q, bio));
 908 }
 909
 910 static void blk_add_trace_bio_frontmerge(void *ignore,
 911                                          struct request_queue *q,
 912                                          struct request *rq,
 913                                          struct bio *bio)
 914 {
 915         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0,
 916                           blk_trace_bio_get_cgid(q, bio));
 917 }
 918
 919 static void blk_add_trace_bio_queue(void *ignore,
 920                                     struct request_queue *q, struct bio *bio)
 921 {
 922         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0,
 923                           blk_trace_bio_get_cgid(q, bio));
 924 }
 925
 926 static void blk_add_trace_getrq(void *ignore,
 927                                 struct request_queue *q,
 928                                 struct bio *bio, int rw)
 929 {
 930         if (bio)
 931                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0,
 932                                   blk_trace_bio_get_cgid(q, bio));
 933         else {
 934                 struct blk_trace *bt = q->blk_trace;
 935
 936                 if (bt)
 937                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 938                                         NULL, NULL);
 939         }
 940 }
 941
 942
 943 static void blk_add_trace_sleeprq(void *ignore,
 944                                   struct request_queue *q,
 945                                   struct bio *bio, int rw)
 946 {
 947         if (bio)
 948                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0,
 949                                   blk_trace_bio_get_cgid(q, bio));
 950         else {
 951                 struct blk_trace *bt = q->blk_trace;
 952
 953                 if (bt)
 954                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
 955                                         0, 0, NULL, NULL);
 956         }
 957 }
 958
 959 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
 960 {
 961         struct blk_trace *bt = q->blk_trace;
 962
 963         if (bt)
 964                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
 965 }
 966
 967 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
 968                                     unsigned int depth, bool explicit)
 969 {
 970         struct blk_trace *bt = q->blk_trace;
 971
 972         if (bt) {
 973                 __be64 rpdu = cpu_to_be64(depth);
 974                 u32 what;
 975
 976                 if (explicit)
 977                         what = BLK_TA_UNPLUG_IO;
 978                 else
 979                         what = BLK_TA_UNPLUG_TIMER;
 980
 981                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
 982         }
 983 }
 984
 985 static void blk_add_trace_split(void *ignore,
 986                                 struct request_queue *q, struct bio *bio,
 987                                 unsigned int pdu)
 988 {
 989         struct blk_trace *bt = q->blk_trace;
 990
 991         if (bt) {
 992                 __be64 rpdu = cpu_to_be64(pdu);
 993
 994                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
 995                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
 996                                 BLK_TA_SPLIT, bio->bi_status, sizeof(rpdu),
 997                                 &rpdu, blk_trace_bio_get_cgid(q, bio));
 998         }
 999 }
1000
1001 /**
1002  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1003  * @ignore:     trace callback data parameter (not used)
1004  * @q:          queue the io is for
1005  * @bio:        the source bio
1006  * @dev:        target device
1007  * @from:       source sector
1008  *
1009  * Description:
1010  *     Device mapper or raid target sometimes need to split a bio because
1011  *     it spans a stripe (or similar). Add a trace for that action.
1012  *
1013  **/
1014 static void blk_add_trace_bio_remap(void *ignore,
1015                                     struct request_queue *q, struct bio *bio,
1016                                     dev_t dev, sector_t from)
1017 {
1018         struct blk_trace *bt = q->blk_trace;
1019         struct blk_io_trace_remap r;
1020
1021         if (likely(!bt))
1022                 return;
1023
1024         r.device_from = cpu_to_be32(dev);
1025         r.device_to   = cpu_to_be32(bio_dev(bio));
1026         r.sector_from = cpu_to_be64(from);
1027
1028         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1029                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP, bio->bi_status,
1030                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1031 }
1032
1033 /**
1034  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1035  * @ignore:     trace callback data parameter (not used)
1036  * @q:          queue the io is for
1037  * @rq:         the source request
1038  * @dev:        target device
1039  * @from:       source sector
1040  *
1041  * Description:
1042  *     Device mapper remaps request to other devices.
1043  *     Add a trace for that action.
1044  *
1045  **/
1046 static void blk_add_trace_rq_remap(void *ignore,
1047                                    struct request_queue *q,
1048                                    struct request *rq, dev_t dev,
1049                                    sector_t from)
1050 {
1051         struct blk_trace *bt = q->blk_trace;
1052         struct blk_io_trace_remap r;
1053
1054         if (likely(!bt))
1055                 return;
1056
1057         r.device_from = cpu_to_be32(dev);
1058         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1059         r.sector_from = cpu_to_be64(from);
1060
1061         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1062                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1063                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1064 }
1065
1066 /**
1067  * blk_add_driver_data - Add binary message with driver-specific data
1068  * @q:          queue the io is for
1069  * @rq:         io request
1070  * @data:       driver-specific data
1071  * @len:        length of driver-specific data
1072  *
1073  * Description:
1074  *     Some drivers might want to write driver-specific data per request.
1075  *
1076  **/
1077 void blk_add_driver_data(struct request_queue *q,
1078                          struct request *rq,
1079                          void *data, size_t len)
1080 {
1081         struct blk_trace *bt = q->blk_trace;
1082
1083         if (likely(!bt))
1084                 return;
1085
1086         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1087                                 BLK_TA_DRV_DATA, 0, len, data,
1088                                 blk_trace_request_get_cgid(q, rq));
1089 }
1090 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1091
1092 static void blk_register_tracepoints(void)
1093 {
1094         int ret;
1095
1096         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1097         WARN_ON(ret);
1098         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1099         WARN_ON(ret);
1100         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1101         WARN_ON(ret);
1102         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1103         WARN_ON(ret);
1104         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1105         WARN_ON(ret);
1106         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1107         WARN_ON(ret);
1108         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1109         WARN_ON(ret);
1110         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1111         WARN_ON(ret);
1112         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1113         WARN_ON(ret);
1114         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1115         WARN_ON(ret);
1116         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1117         WARN_ON(ret);
1118         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1119         WARN_ON(ret);
1120         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1121         WARN_ON(ret);
1122         ret = register_trace_block_split(blk_add_trace_split, NULL);
1123         WARN_ON(ret);
1124         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1125         WARN_ON(ret);
1126         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1127         WARN_ON(ret);
1128 }
1129
1130 static void blk_unregister_tracepoints(void)
1131 {
1132         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1133         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1134         unregister_trace_block_split(blk_add_trace_split, NULL);
1135         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1136         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1137         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1138         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1139         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1140         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1141         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1142         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1143         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1144         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1145         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1146         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1147         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1148
1149         tracepoint_synchronize_unregister();
1150 }
1151
1152 /*
1153  * struct blk_io_tracer formatting routines
1154  */
1155
1156 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1157 {
1158         int i = 0;
1159         int tc = t->action >> BLK_TC_SHIFT;
1160
1161         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1162                 rwbs[i++] = 'N';
1163                 goto out;
1164         }
1165
1166         if (tc & BLK_TC_FLUSH)
1167                 rwbs[i++] = 'F';
1168
1169         if (tc & BLK_TC_DISCARD)
1170                 rwbs[i++] = 'D';
1171         else if (tc & BLK_TC_WRITE)
1172                 rwbs[i++] = 'W';
1173         else if (t->bytes)
1174                 rwbs[i++] = 'R';
1175         else
1176                 rwbs[i++] = 'N';
1177
1178         if (tc & BLK_TC_FUA)
1179                 rwbs[i++] = 'F';
1180         if (tc & BLK_TC_AHEAD)
1181                 rwbs[i++] = 'A';
1182         if (tc & BLK_TC_SYNC)
1183                 rwbs[i++] = 'S';
1184         if (tc & BLK_TC_META)
1185                 rwbs[i++] = 'M';
1186 out:
1187         rwbs[i] = '\0';
1188 }
1189
1190 static inline
1191 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1192 {
1193         return (const struct blk_io_trace *)ent;
1194 }
1195
1196 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1197 {
1198         return (void *)(te_blk_io_trace(ent) + 1) +
1199                 (has_cg ? sizeof(union kernfs_node_id) : 0);
1200 }
1201
1202 static inline const void *cgid_start(const struct trace_entry *ent)
1203 {
1204         return (void *)(te_blk_io_trace(ent) + 1);
1205 }
1206
1207 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1208 {
1209         return te_blk_io_trace(ent)->pdu_len -
1210                         (has_cg ? sizeof(union kernfs_node_id) : 0);
1211 }
1212
1213 static inline u32 t_action(const struct trace_entry *ent)
1214 {
1215         return te_blk_io_trace(ent)->action;
1216 }
1217
1218 static inline u32 t_bytes(const struct trace_entry *ent)
1219 {
1220         return te_blk_io_trace(ent)->bytes;
1221 }
1222
1223 static inline u32 t_sec(const struct trace_entry *ent)
1224 {
1225         return te_blk_io_trace(ent)->bytes >> 9;
1226 }
1227
1228 static inline unsigned long long t_sector(const struct trace_entry *ent)
1229 {
1230         return te_blk_io_trace(ent)->sector;
1231 }
1232
1233 static inline __u16 t_error(const struct trace_entry *ent)
1234 {
1235         return te_blk_io_trace(ent)->error;
1236 }
1237
1238 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1239 {
1240         const __u64 *val = pdu_start(ent, has_cg);
1241         return be64_to_cpu(*val);
1242 }
1243
1244 static void get_pdu_remap(const struct trace_entry *ent,
1245                           struct blk_io_trace_remap *r, bool has_cg)
1246 {
1247         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1248         __u64 sector_from = __r->sector_from;
1249
1250         r->device_from = be32_to_cpu(__r->device_from);
1251         r->device_to   = be32_to_cpu(__r->device_to);
1252         r->sector_from = be64_to_cpu(sector_from);
1253 }
1254
1255 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1256         bool has_cg);
1257
1258 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1259         bool has_cg)
1260 {
1261         char rwbs[RWBS_LEN];
1262         unsigned long long ts  = iter->ts;
1263         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1264         unsigned secs          = (unsigned long)ts;
1265         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1266
1267         fill_rwbs(rwbs, t);
1268
1269         trace_seq_printf(&iter->seq,
1270                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1271                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1272                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1273 }
1274
1275 static void blk_log_action(struct trace_iterator *iter, const char *act,
1276         bool has_cg)
1277 {
1278         char rwbs[RWBS_LEN];
1279         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1280
1281         fill_rwbs(rwbs, t);
1282         if (has_cg) {
1283                 const union kernfs_node_id *id = cgid_start(iter->ent);
1284
1285                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1286                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1287
1288                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1289                                 sizeof(blkcg_name_buf));
1290                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1291                                  MAJOR(t->device), MINOR(t->device),
1292                                  blkcg_name_buf, act, rwbs);
1293                 } else
1294                         trace_seq_printf(&iter->seq,
1295                                  "%3d,%-3d %x,%-x %2s %3s ",
1296                                  MAJOR(t->device), MINOR(t->device),
1297                                  id->ino, id->generation, act, rwbs);
1298         } else
1299                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1300                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1301 }
1302
1303 static void blk_log_dump_pdu(struct trace_seq *s,
1304         const struct trace_entry *ent, bool has_cg)
1305 {
1306         const unsigned char *pdu_buf;
1307         int pdu_len;
1308         int i, end;
1309
1310         pdu_buf = pdu_start(ent, has_cg);
1311         pdu_len = pdu_real_len(ent, has_cg);
1312
1313         if (!pdu_len)
1314                 return;
1315
1316         /* find the last zero that needs to be printed */
1317         for (end = pdu_len - 1; end >= 0; end--)
1318                 if (pdu_buf[end])
1319                         break;
1320         end++;
1321
1322         trace_seq_putc(s, '(');
1323
1324         for (i = 0; i < pdu_len; i++) {
1325
1326                 trace_seq_printf(s, "%s%02x",
1327                                  i == 0 ? "" : " ", pdu_buf[i]);
1328
1329                 /*
1330                  * stop when the rest is just zeroes and indicate so
1331                  * with a ".." appended
1332                  */
1333                 if (i == end && end != pdu_len - 1) {
1334                         trace_seq_puts(s, " ..) ");
1335                         return;
1336                 }
1337         }
1338
1339         trace_seq_puts(s, ") ");
1340 }
1341
1342 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1343 {
1344         char cmd[TASK_COMM_LEN];
1345
1346         trace_find_cmdline(ent->pid, cmd);
1347
1348         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1349                 trace_seq_printf(s, "%u ", t_bytes(ent));
1350                 blk_log_dump_pdu(s, ent, has_cg);
1351                 trace_seq_printf(s, "[%s]\n", cmd);
1352         } else {
1353                 if (t_sec(ent))
1354                         trace_seq_printf(s, "%llu + %u [%s]\n",
1355                                                 t_sector(ent), t_sec(ent), cmd);
1356                 else
1357                         trace_seq_printf(s, "[%s]\n", cmd);
1358         }
1359 }
1360
1361 static void blk_log_with_error(struct trace_seq *s,
1362                               const struct trace_entry *ent, bool has_cg)
1363 {
1364         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1365                 blk_log_dump_pdu(s, ent, has_cg);
1366                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1367         } else {
1368                 if (t_sec(ent))
1369                         trace_seq_printf(s, "%llu + %u [%d]\n",
1370                                          t_sector(ent),
1371                                          t_sec(ent), t_error(ent));
1372                 else
1373                         trace_seq_printf(s, "%llu [%d]\n",
1374                                          t_sector(ent), t_error(ent));
1375         }
1376 }
1377
1378 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1379 {
1380         struct blk_io_trace_remap r = { .device_from = 0, };
1381
1382         get_pdu_remap(ent, &r, has_cg);
1383         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1384                          t_sector(ent), t_sec(ent),
1385                          MAJOR(r.device_from), MINOR(r.device_from),
1386                          (unsigned long long)r.sector_from);
1387 }
1388
1389 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1390 {
1391         char cmd[TASK_COMM_LEN];
1392
1393         trace_find_cmdline(ent->pid, cmd);
1394
1395         trace_seq_printf(s, "[%s]\n", cmd);
1396 }
1397
1398 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1399 {
1400         char cmd[TASK_COMM_LEN];
1401
1402         trace_find_cmdline(ent->pid, cmd);
1403
1404         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1405 }
1406
1407 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1408 {
1409         char cmd[TASK_COMM_LEN];
1410
1411         trace_find_cmdline(ent->pid, cmd);
1412
1413         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1414                          get_pdu_int(ent, has_cg), cmd);
1415 }
1416
1417 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1418                         bool has_cg)
1419 {
1420
1421         trace_seq_putmem(s, pdu_start(ent, has_cg),
1422                 pdu_real_len(ent, has_cg));
1423         trace_seq_putc(s, '\n');
1424 }
1425
1426 /*
1427  * struct tracer operations
1428  */
1429
1430 static void blk_tracer_print_header(struct seq_file *m)
1431 {
1432         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1433                 return;
1434         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1435                     "#  |     |     |           |   |   |\n");
1436 }
1437
1438 static void blk_tracer_start(struct trace_array *tr)
1439 {
1440         blk_tracer_enabled = true;
1441 }
1442
1443 static int blk_tracer_init(struct trace_array *tr)
1444 {
1445         blk_tr = tr;
1446         blk_tracer_start(tr);
1447         return 0;
1448 }
1449
1450 static void blk_tracer_stop(struct trace_array *tr)
1451 {
1452         blk_tracer_enabled = false;
1453 }
1454
1455 static void blk_tracer_reset(struct trace_array *tr)
1456 {
1457         blk_tracer_stop(tr);
1458 }
1459
1460 static const struct {
1461         const char *act[2];
1462         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1463                             bool has_cg);
1464 } what2act[] = {
1465         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1466         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1467         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1468         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1469         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1470         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1471         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1472         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1473         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1474         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1475         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1476         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1477         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1478         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1479         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1480 };
1481
1482 static enum print_line_t print_one_line(struct trace_iterator *iter,
1483                                         bool classic)
1484 {
1485         struct trace_array *tr = iter->tr;
1486         struct trace_seq *s = &iter->seq;
1487         const struct blk_io_trace *t;
1488         u16 what;
1489         bool long_act;
1490         blk_log_action_t *log_action;
1491         bool has_cg;
1492
1493         t          = te_blk_io_trace(iter->ent);
1494         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1495         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1496         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1497         has_cg     = t->action & __BLK_TA_CGROUP;
1498
1499         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1500                 log_action(iter, long_act ? "message" : "m", has_cg);
1501                 blk_log_msg(s, iter->ent, has_cg);
1502                 return trace_handle_return(s);
1503         }
1504
1505         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1506                 trace_seq_printf(s, "Unknown action %x\n", what);
1507         else {
1508                 log_action(iter, what2act[what].act[long_act], has_cg);
1509                 what2act[what].print(s, iter->ent, has_cg);
1510         }
1511
1512         return trace_handle_return(s);
1513 }
1514
1515 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1516                                                int flags, struct trace_event *event)
1517 {
1518         return print_one_line(iter, false);
1519 }
1520
1521 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1522 {
1523         struct trace_seq *s = &iter->seq;
1524         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1525         const int offset = offsetof(struct blk_io_trace, sector);
1526         struct blk_io_trace old = {
1527                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1528                 .time     = iter->ts,
1529         };
1530
1531         trace_seq_putmem(s, &old, offset);
1532         trace_seq_putmem(s, &t->sector,
1533                          sizeof(old) - offset + t->pdu_len);
1534 }
1535
1536 static enum print_line_t
1537 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1538                              struct trace_event *event)
1539 {
1540         blk_trace_synthesize_old_trace(iter);
1541
1542         return trace_handle_return(&iter->seq);
1543 }
1544
1545 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1546 {
1547         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1548                 return TRACE_TYPE_UNHANDLED;
1549
1550         return print_one_line(iter, true);
1551 }
1552
1553 static int
1554 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1555 {
1556         /* don't output context-info for blk_classic output */
1557         if (bit == TRACE_BLK_OPT_CLASSIC) {
1558                 if (set)
1559                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1560                 else
1561                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1562         }
1563         return 0;
1564 }
1565
1566 static struct tracer blk_tracer __read_mostly = {
1567         .name           = "blk",
1568         .init           = blk_tracer_init,
1569         .reset          = blk_tracer_reset,
1570         .start          = blk_tracer_start,
1571         .stop           = blk_tracer_stop,
1572         .print_header   = blk_tracer_print_header,
1573         .print_line     = blk_tracer_print_line,
1574         .flags          = &blk_tracer_flags,
1575         .set_flag       = blk_tracer_set_flag,
1576 };
1577
1578 static struct trace_event_functions trace_blk_event_funcs = {
1579         .trace          = blk_trace_event_print,
1580         .binary         = blk_trace_event_print_binary,
1581 };
1582
1583 static struct trace_event trace_blk_event = {
1584         .type           = TRACE_BLK,
1585         .funcs          = &trace_blk_event_funcs,
1586 };
1587
1588 static int __init init_blk_tracer(void)
1589 {
1590         if (!register_trace_event(&trace_blk_event)) {
1591                 pr_warn("Warning: could not register block events\n");
1592                 return 1;
1593         }
1594
1595         if (register_tracer(&blk_tracer) != 0) {
1596                 pr_warn("Warning: could not register the block tracer\n");
1597                 unregister_trace_event(&trace_blk_event);
1598                 return 1;
1599         }
1600
1601         return 0;
1602 }
1603
1604 device_initcall(init_blk_tracer);
1605
1606 static int blk_trace_remove_queue(struct request_queue *q)
1607 {
1608         struct blk_trace *bt;
1609
1610         bt = xchg(&q->blk_trace, NULL);
1611         if (bt == NULL)
1612                 return -EINVAL;
1613
1614         put_probe_ref();
1615         blk_trace_free(bt);
1616         return 0;
1617 }
1618
1619 /*
1620  * Setup everything required to start tracing
1621  */
1622 static int blk_trace_setup_queue(struct request_queue *q,
1623                                  struct block_device *bdev)
1624 {
1625         struct blk_trace *bt = NULL;
1626         int ret = -ENOMEM;
1627
1628         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1629         if (!bt)
1630                 return -ENOMEM;
1631
1632         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1633         if (!bt->msg_data)
1634                 goto free_bt;
1635
1636         bt->dev = bdev->bd_dev;
1637         bt->act_mask = (u16)-1;
1638
1639         blk_trace_setup_lba(bt, bdev);
1640
1641         ret = -EBUSY;
1642         if (cmpxchg(&q->blk_trace, NULL, bt))
1643                 goto free_bt;
1644
1645         get_probe_ref();
1646         return 0;
1647
1648 free_bt:
1649         blk_trace_free(bt);
1650         return ret;
1651 }
1652
1653 /*
1654  * sysfs interface to enable and configure tracing
1655  */
1656
1657 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1658                                          struct device_attribute *attr,
1659                                          char *buf);
1660 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1661                                           struct device_attribute *attr,
1662                                           const char *buf, size_t count);
1663 #define BLK_TRACE_DEVICE_ATTR(_name) \
1664         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1665                     sysfs_blk_trace_attr_show, \
1666                     sysfs_blk_trace_attr_store)
1667
1668 static BLK_TRACE_DEVICE_ATTR(enable);
1669 static BLK_TRACE_DEVICE_ATTR(act_mask);
1670 static BLK_TRACE_DEVICE_ATTR(pid);
1671 static BLK_TRACE_DEVICE_ATTR(start_lba);
1672 static BLK_TRACE_DEVICE_ATTR(end_lba);
1673
1674 static struct attribute *blk_trace_attrs[] = {
1675         &dev_attr_enable.attr,
1676         &dev_attr_act_mask.attr,
1677         &dev_attr_pid.attr,
1678         &dev_attr_start_lba.attr,
1679         &dev_attr_end_lba.attr,
1680         NULL
1681 };
1682
1683 struct attribute_group blk_trace_attr_group = {
1684         .name  = "trace",
1685         .attrs = blk_trace_attrs,
1686 };
1687
1688 static const struct {
1689         int mask;
1690         const char *str;
1691 } mask_maps[] = {
1692         { BLK_TC_READ,          "read"          },
1693         { BLK_TC_WRITE,         "write"         },
1694         { BLK_TC_FLUSH,         "flush"         },
1695         { BLK_TC_SYNC,          "sync"          },
1696         { BLK_TC_QUEUE,         "queue"         },
1697         { BLK_TC_REQUEUE,       "requeue"       },
1698         { BLK_TC_ISSUE,         "issue"         },
1699         { BLK_TC_COMPLETE,      "complete"      },
1700         { BLK_TC_FS,            "fs"            },
1701         { BLK_TC_PC,            "pc"            },
1702         { BLK_TC_NOTIFY,        "notify"        },
1703         { BLK_TC_AHEAD,         "ahead"         },
1704         { BLK_TC_META,          "meta"          },
1705         { BLK_TC_DISCARD,       "discard"       },
1706         { BLK_TC_DRV_DATA,      "drv_data"      },
1707         { BLK_TC_FUA,           "fua"           },
1708 };
1709
1710 static int blk_trace_str2mask(const char *str)
1711 {
1712         int i;
1713         int mask = 0;
1714         char *buf, *s, *token;
1715
1716         buf = kstrdup(str, GFP_KERNEL);
1717         if (buf == NULL)
1718                 return -ENOMEM;
1719         s = strstrip(buf);
1720
1721         while (1) {
1722                 token = strsep(&s, ",");
1723                 if (token == NULL)
1724                         break;
1725
1726                 if (*token == '\0')
1727                         continue;
1728
1729                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1730                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1731                                 mask |= mask_maps[i].mask;
1732                                 break;
1733                         }
1734                 }
1735                 if (i == ARRAY_SIZE(mask_maps)) {
1736                         mask = -EINVAL;
1737                         break;
1738                 }
1739         }
1740         kfree(buf);
1741
1742         return mask;
1743 }
1744
1745 static ssize_t blk_trace_mask2str(char *buf, int mask)
1746 {
1747         int i;
1748         char *p = buf;
1749
1750         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1751                 if (mask & mask_maps[i].mask) {
1752                         p += sprintf(p, "%s%s",
1753                                     (p == buf) ? "" : ",", mask_maps[i].str);
1754                 }
1755         }
1756         *p++ = '\n';
1757
1758         return p - buf;
1759 }
1760
1761 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1762 {
1763         if (bdev->bd_disk == NULL)
1764                 return NULL;
1765
1766         return bdev_get_queue(bdev);
1767 }
1768
1769 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1770                                          struct device_attribute *attr,
1771                                          char *buf)
1772 {
1773         struct hd_struct *p = dev_to_part(dev);
1774         struct request_queue *q;
1775         struct block_device *bdev;
1776         ssize_t ret = -ENXIO;
1777
1778         bdev = bdget(part_devt(p));
1779         if (bdev == NULL)
1780                 goto out;
1781
1782         q = blk_trace_get_queue(bdev);
1783         if (q == NULL)
1784                 goto out_bdput;
1785
1786         mutex_lock(&q->blk_trace_mutex);
1787
1788         if (attr == &dev_attr_enable) {
1789                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1790                 goto out_unlock_bdev;
1791         }
1792
1793         if (q->blk_trace == NULL)
1794                 ret = sprintf(buf, "disabled\n");
1795         else if (attr == &dev_attr_act_mask)
1796                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1797         else if (attr == &dev_attr_pid)
1798                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1799         else if (attr == &dev_attr_start_lba)
1800                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1801         else if (attr == &dev_attr_end_lba)
1802                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1803
1804 out_unlock_bdev:
1805         mutex_unlock(&q->blk_trace_mutex);
1806 out_bdput:
1807         bdput(bdev);
1808 out:
1809         return ret;
1810 }
1811
1812 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1813                                           struct device_attribute *attr,
1814                                           const char *buf, size_t count)
1815 {
1816         struct block_device *bdev;
1817         struct request_queue *q;
1818         struct hd_struct *p;
1819         u64 value;
1820         ssize_t ret = -EINVAL;
1821
1822         if (count == 0)
1823                 goto out;
1824
1825         if (attr == &dev_attr_act_mask) {
1826                 if (kstrtoull(buf, 0, &value)) {
1827                         /* Assume it is a list of trace category names */
1828                         ret = blk_trace_str2mask(buf);
1829                         if (ret < 0)
1830                                 goto out;
1831                         value = ret;
1832                 }
1833         } else if (kstrtoull(buf, 0, &value))
1834                 goto out;
1835
1836         ret = -ENXIO;
1837
1838         p = dev_to_part(dev);
1839         bdev = bdget(part_devt(p));
1840         if (bdev == NULL)
1841                 goto out;
1842
1843         q = blk_trace_get_queue(bdev);
1844         if (q == NULL)
1845                 goto out_bdput;
1846
1847         mutex_lock(&q->blk_trace_mutex);
1848
1849         if (attr == &dev_attr_enable) {
1850                 if (value)
1851                         ret = blk_trace_setup_queue(q, bdev);
1852                 else
1853                         ret = blk_trace_remove_queue(q);
1854                 goto out_unlock_bdev;
1855         }
1856
1857         ret = 0;
1858         if (q->blk_trace == NULL)
1859                 ret = blk_trace_setup_queue(q, bdev);
1860
1861         if (ret == 0) {
1862                 if (attr == &dev_attr_act_mask)
1863                         q->blk_trace->act_mask = value;
1864                 else if (attr == &dev_attr_pid)
1865                         q->blk_trace->pid = value;
1866                 else if (attr == &dev_attr_start_lba)
1867                         q->blk_trace->start_lba = value;
1868                 else if (attr == &dev_attr_end_lba)
1869                         q->blk_trace->end_lba = value;
1870         }
1871
1872 out_unlock_bdev:
1873         mutex_unlock(&q->blk_trace_mutex);
1874 out_bdput:
1875         bdput(bdev);
1876 out:
1877         return ret ? ret : count;
1878 }
1879
1880 int blk_trace_init_sysfs(struct device *dev)
1881 {
1882         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1883 }
1884
1885 void blk_trace_remove_sysfs(struct device *dev)
1886 {
1887         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1888 }
1889
1890 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1891
1892 #ifdef CONFIG_EVENT_TRACING
1893
1894 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1895 {
1896         int i = 0;
1897
1898         if (op & REQ_PREFLUSH)
1899                 rwbs[i++] = 'F';
1900
1901         switch (op & REQ_OP_MASK) {
1902         case REQ_OP_WRITE:
1903         case REQ_OP_WRITE_SAME:
1904                 rwbs[i++] = 'W';
1905                 break;
1906         case REQ_OP_DISCARD:
1907                 rwbs[i++] = 'D';
1908                 break;
1909         case REQ_OP_SECURE_ERASE:
1910                 rwbs[i++] = 'D';
1911                 rwbs[i++] = 'E';
1912                 break;
1913         case REQ_OP_FLUSH:
1914                 rwbs[i++] = 'F';
1915                 break;
1916         case REQ_OP_READ:
1917                 rwbs[i++] = 'R';
1918                 break;
1919         default:
1920                 rwbs[i++] = 'N';
1921         }
1922
1923         if (op & REQ_FUA)
1924                 rwbs[i++] = 'F';
1925         if (op & REQ_RAHEAD)
1926                 rwbs[i++] = 'A';
1927         if (op & REQ_SYNC)
1928                 rwbs[i++] = 'S';
1929         if (op & REQ_META)
1930                 rwbs[i++] = 'M';
1931
1932         rwbs[i] = '\0';
1933 }
1934 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1935
1936 #endif /* CONFIG_EVENT_TRACING */
1937