kernel/trace/blktrace.c

   1 /*
   2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   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
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/kernel.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/blktrace_api.h>
  24 #include <linux/percpu.h>
  25 #include <linux/init.h>
  26 #include <linux/mutex.h>
  27 #include <linux/slab.h>
  28 #include <linux/debugfs.h>
  29 #include <linux/export.h>
  30 #include <linux/time.h>
  31 #include <linux/uaccess.h>
  32 #include <linux/list.h>
  33 #include <linux/blk-cgroup.h>
  34
  35 #include "../../block/blk.h"
  36
  37 #include <trace/events/block.h>
  38
  39 #include "trace_output.h"
  40
  41 #ifdef CONFIG_BLK_DEV_IO_TRACE
  42
  43 static unsigned int blktrace_seq __read_mostly = 1;
  44
  45 static struct trace_array *blk_tr;
  46 static bool blk_tracer_enabled __read_mostly;
  47
  48 static LIST_HEAD(running_trace_list);
  49 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock);
  50
  51 /* Select an alternative, minimalistic output than the original one */
  52 #define TRACE_BLK_OPT_CLASSIC   0x1
  53 #define TRACE_BLK_OPT_CGROUP    0x2
  54 #define TRACE_BLK_OPT_CGNAME    0x4
  55
  56 static struct tracer_opt blk_tracer_opts[] = {
  57         /* Default disable the minimalistic output */
  58         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
  59 #ifdef CONFIG_BLK_CGROUP
  60         { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) },
  61         { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) },
  62 #endif
  63         { }
  64 };
  65
  66 static struct tracer_flags blk_tracer_flags = {
  67         .val  = 0,
  68         .opts = blk_tracer_opts,
  69 };
  70
  71 /* Global reference count of probes */
  72 static DEFINE_MUTEX(blk_probe_mutex);
  73 static int blk_probes_ref;
  74
  75 static void blk_register_tracepoints(void);
  76 static void blk_unregister_tracepoints(void);
  77
  78 /*
  79  * Send out a notify message.
  80  */
  81 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
  82                        const void *data, size_t len,
  83                        union kernfs_node_id *cgid)
  84 {
  85         struct blk_io_trace *t;
  86         struct ring_buffer_event *event = NULL;
  87         struct ring_buffer *buffer = NULL;
  88         int pc = 0;
  89         int cpu = smp_processor_id();
  90         bool blk_tracer = blk_tracer_enabled;
  91         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
  92
  93         if (blk_tracer) {
  94                 buffer = blk_tr->trace_buffer.buffer;
  95                 pc = preempt_count();
  96                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
  97                                                   sizeof(*t) + len + cgid_len,
  98                                                   0, pc);
  99                 if (!event)
 100                         return;
 101                 t = ring_buffer_event_data(event);
 102                 goto record_it;
 103         }
 104
 105         if (!bt->rchan)
 106                 return;
 107
 108         t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len);
 109         if (t) {
 110                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 111                 t->time = ktime_to_ns(ktime_get());
 112 record_it:
 113                 t->device = bt->dev;
 114                 t->action = action | (cgid ? __BLK_TN_CGROUP : 0);
 115                 t->pid = pid;
 116                 t->cpu = cpu;
 117                 t->pdu_len = len + cgid_len;
 118                 if (cgid)
 119                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 120                 memcpy((void *) t + sizeof(*t) + cgid_len, data, len);
 121
 122                 if (blk_tracer)
 123                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 124         }
 125 }
 126
 127 /*
 128  * Send out a notify for this process, if we haven't done so since a trace
 129  * started
 130  */
 131 static void trace_note_tsk(struct task_struct *tsk)
 132 {
 133         unsigned long flags;
 134         struct blk_trace *bt;
 135
 136         tsk->btrace_seq = blktrace_seq;
 137         spin_lock_irqsave(&running_trace_lock, flags);
 138         list_for_each_entry(bt, &running_trace_list, running_list) {
 139                 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm,
 140                            sizeof(tsk->comm), NULL);
 141         }
 142         spin_unlock_irqrestore(&running_trace_lock, flags);
 143 }
 144
 145 static void trace_note_time(struct blk_trace *bt)
 146 {
 147         struct timespec64 now;
 148         unsigned long flags;
 149         u32 words[2];
 150
 151         /* need to check user space to see if this breaks in y2038 or y2106 */
 152         ktime_get_real_ts64(&now);
 153         words[0] = (u32)now.tv_sec;
 154         words[1] = now.tv_nsec;
 155
 156         local_irq_save(flags);
 157         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), NULL);
 158         local_irq_restore(flags);
 159 }
 160
 161 void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg,
 162         const char *fmt, ...)
 163 {
 164         int n;
 165         va_list args;
 166         unsigned long flags;
 167         char *buf;
 168
 169         if (unlikely(bt->trace_state != Blktrace_running &&
 170                      !blk_tracer_enabled))
 171                 return;
 172
 173         /*
 174          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
 175          * message to the trace.
 176          */
 177         if (!(bt->act_mask & BLK_TC_NOTIFY))
 178                 return;
 179
 180         local_irq_save(flags);
 181         buf = this_cpu_ptr(bt->msg_data);
 182         va_start(args, fmt);
 183         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
 184         va_end(args);
 185
 186         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 187                 blkcg = NULL;
 188 #ifdef CONFIG_BLK_CGROUP
 189         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n,
 190                 blkcg ? cgroup_get_kernfs_id(blkcg->css.cgroup) : NULL);
 191 #else
 192         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n, NULL);
 193 #endif
 194         local_irq_restore(flags);
 195 }
 196 EXPORT_SYMBOL_GPL(__trace_note_message);
 197
 198 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
 199                          pid_t pid)
 200 {
 201         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
 202                 return 1;
 203         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
 204                 return 1;
 205         if (bt->pid && pid != bt->pid)
 206                 return 1;
 207
 208         return 0;
 209 }
 210
 211 /*
 212  * Data direction bit lookup
 213  */
 214 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
 215                                  BLK_TC_ACT(BLK_TC_WRITE) };
 216
 217 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
 218 #define BLK_TC_PREFLUSH         BLK_TC_FLUSH
 219
 220 /* The ilog2() calls fall out because they're constant */
 221 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
 222           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
 223
 224 /*
 225  * The worker for the various blk_add_trace*() types. Fills out a
 226  * blk_io_trace structure and places it in a per-cpu subbuffer.
 227  */
 228 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
 229                      int op, int op_flags, u32 what, int error, int pdu_len,
 230                      void *pdu_data, union kernfs_node_id *cgid)
 231 {
 232         struct task_struct *tsk = current;
 233         struct ring_buffer_event *event = NULL;
 234         struct ring_buffer *buffer = NULL;
 235         struct blk_io_trace *t;
 236         unsigned long flags = 0;
 237         unsigned long *sequence;
 238         pid_t pid;
 239         int cpu, pc = 0;
 240         bool blk_tracer = blk_tracer_enabled;
 241         ssize_t cgid_len = cgid ? sizeof(*cgid) : 0;
 242
 243         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
 244                 return;
 245
 246         what |= ddir_act[op_is_write(op) ? WRITE : READ];
 247         what |= MASK_TC_BIT(op_flags, SYNC);
 248         what |= MASK_TC_BIT(op_flags, RAHEAD);
 249         what |= MASK_TC_BIT(op_flags, META);
 250         what |= MASK_TC_BIT(op_flags, PREFLUSH);
 251         what |= MASK_TC_BIT(op_flags, FUA);
 252         if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)
 253                 what |= BLK_TC_ACT(BLK_TC_DISCARD);
 254         if (op == REQ_OP_FLUSH)
 255                 what |= BLK_TC_ACT(BLK_TC_FLUSH);
 256         if (cgid)
 257                 what |= __BLK_TA_CGROUP;
 258
 259         pid = tsk->pid;
 260         if (act_log_check(bt, what, sector, pid))
 261                 return;
 262         cpu = raw_smp_processor_id();
 263
 264         if (blk_tracer) {
 265                 tracing_record_cmdline(current);
 266
 267                 buffer = blk_tr->trace_buffer.buffer;
 268                 pc = preempt_count();
 269                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 270                                                   sizeof(*t) + pdu_len + cgid_len,
 271                                                   0, pc);
 272                 if (!event)
 273                         return;
 274                 t = ring_buffer_event_data(event);
 275                 goto record_it;
 276         }
 277
 278         if (unlikely(tsk->btrace_seq != blktrace_seq))
 279                 trace_note_tsk(tsk);
 280
 281         /*
 282          * A word about the locking here - we disable interrupts to reserve
 283          * some space in the relay per-cpu buffer, to prevent an irq
 284          * from coming in and stepping on our toes.
 285          */
 286         local_irq_save(flags);
 287         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len);
 288         if (t) {
 289                 sequence = per_cpu_ptr(bt->sequence, cpu);
 290
 291                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 292                 t->sequence = ++(*sequence);
 293                 t->time = ktime_to_ns(ktime_get());
 294 record_it:
 295                 /*
 296                  * These two are not needed in ftrace as they are in the
 297                  * generic trace_entry, filled by tracing_generic_entry_update,
 298                  * but for the trace_event->bin() synthesizer benefit we do it
 299                  * here too.
 300                  */
 301                 t->cpu = cpu;
 302                 t->pid = pid;
 303
 304                 t->sector = sector;
 305                 t->bytes = bytes;
 306                 t->action = what;
 307                 t->device = bt->dev;
 308                 t->error = error;
 309                 t->pdu_len = pdu_len + cgid_len;
 310
 311                 if (cgid_len)
 312                         memcpy((void *)t + sizeof(*t), cgid, cgid_len);
 313                 if (pdu_len)
 314                         memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len);
 315
 316                 if (blk_tracer) {
 317                         trace_buffer_unlock_commit(blk_tr, buffer, event, 0, pc);
 318                         return;
 319                 }
 320         }
 321
 322         local_irq_restore(flags);
 323 }
 324
 325 static void blk_trace_free(struct blk_trace *bt)
 326 {
 327         debugfs_remove(bt->msg_file);
 328         debugfs_remove(bt->dropped_file);
 329         relay_close(bt->rchan);
 330         debugfs_remove(bt->dir);
 331         free_percpu(bt->sequence);
 332         free_percpu(bt->msg_data);
 333         kfree(bt);
 334 }
 335
 336 static void get_probe_ref(void)
 337 {
 338         mutex_lock(&blk_probe_mutex);
 339         if (++blk_probes_ref == 1)
 340                 blk_register_tracepoints();
 341         mutex_unlock(&blk_probe_mutex);
 342 }
 343
 344 static void put_probe_ref(void)
 345 {
 346         mutex_lock(&blk_probe_mutex);
 347         if (!--blk_probes_ref)
 348                 blk_unregister_tracepoints();
 349         mutex_unlock(&blk_probe_mutex);
 350 }
 351
 352 static void blk_trace_cleanup(struct blk_trace *bt)
 353 {
 354         synchronize_rcu();
 355         blk_trace_free(bt);
 356         put_probe_ref();
 357 }
 358
 359 static int __blk_trace_remove(struct request_queue *q)
 360 {
 361         struct blk_trace *bt;
 362
 363         bt = xchg(&q->blk_trace, NULL);
 364         if (!bt)
 365                 return -EINVAL;
 366
 367         if (bt->trace_state != Blktrace_running)
 368                 blk_trace_cleanup(bt);
 369
 370         return 0;
 371 }
 372
 373 int blk_trace_remove(struct request_queue *q)
 374 {
 375         int ret;
 376
 377         mutex_lock(&q->blk_trace_mutex);
 378         ret = __blk_trace_remove(q);
 379         mutex_unlock(&q->blk_trace_mutex);
 380
 381         return ret;
 382 }
 383 EXPORT_SYMBOL_GPL(blk_trace_remove);
 384
 385 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
 386                                 size_t count, loff_t *ppos)
 387 {
 388         struct blk_trace *bt = filp->private_data;
 389         char buf[16];
 390
 391         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
 392
 393         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
 394 }
 395
 396 static const struct file_operations blk_dropped_fops = {
 397         .owner =        THIS_MODULE,
 398         .open =         simple_open,
 399         .read =         blk_dropped_read,
 400         .llseek =       default_llseek,
 401 };
 402
 403 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
 404                                 size_t count, loff_t *ppos)
 405 {
 406         char *msg;
 407         struct blk_trace *bt;
 408
 409         if (count >= BLK_TN_MAX_MSG)
 410                 return -EINVAL;
 411
 412         msg = memdup_user_nul(buffer, count);
 413         if (IS_ERR(msg))
 414                 return PTR_ERR(msg);
 415
 416         bt = filp->private_data;
 417         __trace_note_message(bt, NULL, "%s", msg);
 418         kfree(msg);
 419
 420         return count;
 421 }
 422
 423 static const struct file_operations blk_msg_fops = {
 424         .owner =        THIS_MODULE,
 425         .open =         simple_open,
 426         .write =        blk_msg_write,
 427         .llseek =       noop_llseek,
 428 };
 429
 430 /*
 431  * Keep track of how many times we encountered a full subbuffer, to aid
 432  * the user space app in telling how many lost events there were.
 433  */
 434 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
 435                                      void *prev_subbuf, size_t prev_padding)
 436 {
 437         struct blk_trace *bt;
 438
 439         if (!relay_buf_full(buf))
 440                 return 1;
 441
 442         bt = buf->chan->private_data;
 443         atomic_inc(&bt->dropped);
 444         return 0;
 445 }
 446
 447 static int blk_remove_buf_file_callback(struct dentry *dentry)
 448 {
 449         debugfs_remove(dentry);
 450
 451         return 0;
 452 }
 453
 454 static struct dentry *blk_create_buf_file_callback(const char *filename,
 455                                                    struct dentry *parent,
 456                                                    umode_t mode,
 457                                                    struct rchan_buf *buf,
 458                                                    int *is_global)
 459 {
 460         return debugfs_create_file(filename, mode, parent, buf,
 461                                         &relay_file_operations);
 462 }
 463
 464 static struct rchan_callbacks blk_relay_callbacks = {
 465         .subbuf_start           = blk_subbuf_start_callback,
 466         .create_buf_file        = blk_create_buf_file_callback,
 467         .remove_buf_file        = blk_remove_buf_file_callback,
 468 };
 469
 470 static void blk_trace_setup_lba(struct blk_trace *bt,
 471                                 struct block_device *bdev)
 472 {
 473         struct hd_struct *part = NULL;
 474
 475         if (bdev)
 476                 part = bdev->bd_part;
 477
 478         if (part) {
 479                 bt->start_lba = part->start_sect;
 480                 bt->end_lba = part->start_sect + part->nr_sects;
 481         } else {
 482                 bt->start_lba = 0;
 483                 bt->end_lba = -1ULL;
 484         }
 485 }
 486
 487 /*
 488  * Setup everything required to start tracing
 489  */
 490 static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 491                               struct block_device *bdev,
 492                               struct blk_user_trace_setup *buts)
 493 {
 494         struct blk_trace *bt = NULL;
 495         struct dentry *dir = NULL;
 496         int ret;
 497
 498         if (!buts->buf_size || !buts->buf_nr)
 499                 return -EINVAL;
 500
 501         if (!blk_debugfs_root)
 502                 return -ENOENT;
 503
 504         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
 505         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
 506
 507         /*
 508          * some device names have larger paths - convert the slashes
 509          * to underscores for this to work as expected
 510          */
 511         strreplace(buts->name, '/', '_');
 512
 513         /*
 514          * bdev can be NULL, as with scsi-generic, this is a helpful as
 515          * we can be.
 516          */
 517         if (q->blk_trace) {
 518                 pr_warn("Concurrent blktraces are not allowed on %s\n",
 519                         buts->name);
 520                 return -EBUSY;
 521         }
 522
 523         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
 524         if (!bt)
 525                 return -ENOMEM;
 526
 527         ret = -ENOMEM;
 528         bt->sequence = alloc_percpu(unsigned long);
 529         if (!bt->sequence)
 530                 goto err;
 531
 532         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
 533         if (!bt->msg_data)
 534                 goto err;
 535
 536 #ifdef CONFIG_BLK_DEBUG_FS
 537         /*
 538          * When tracing whole make_request drivers (multiqueue) block devices,
 539          * reuse the existing debugfs directory created by the block layer on
 540          * init. For request-based block devices, all partitions block devices,
 541          * and scsi-generic block devices we create a temporary new debugfs
 542          * directory that will be removed once the trace ends.
 543          */
 544         if (q->mq_ops && bdev && bdev == bdev->bd_contains)
 545                 dir = q->debugfs_dir;
 546         else
 547 #endif
 548                 bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
 549         if (!dir)
 550                 goto err;
 551
 552         /*
 553          * As blktrace relies on debugfs for its interface the debugfs directory
 554          * is required, contrary to the usual mantra of not checking for debugfs
 555          * files or directories.
 556          */
 557         if (IS_ERR_OR_NULL(dir)) {
 558                 pr_warn("debugfs_dir not present for %s so skipping\n",
 559                         buts->name);
 560                 ret = -ENOENT;
 561                 goto err;
 562         }
 563
 564         bt->dev = dev;
 565         atomic_set(&bt->dropped, 0);
 566         INIT_LIST_HEAD(&bt->running_list);
 567
 568         ret = -EIO;
 569         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
 570                                                &blk_dropped_fops);
 571         if (!bt->dropped_file)
 572                 goto err;
 573
 574         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
 575         if (!bt->msg_file)
 576                 goto err;
 577
 578         bt->rchan = relay_open("trace", dir, buts->buf_size,
 579                                 buts->buf_nr, &blk_relay_callbacks, bt);
 580         if (!bt->rchan)
 581                 goto err;
 582
 583         bt->act_mask = buts->act_mask;
 584         if (!bt->act_mask)
 585                 bt->act_mask = (u16) -1;
 586
 587         blk_trace_setup_lba(bt, bdev);
 588
 589         /* overwrite with user settings */
 590         if (buts->start_lba)
 591                 bt->start_lba = buts->start_lba;
 592         if (buts->end_lba)
 593                 bt->end_lba = buts->end_lba;
 594
 595         bt->pid = buts->pid;
 596         bt->trace_state = Blktrace_setup;
 597
 598         ret = -EBUSY;
 599         if (cmpxchg(&q->blk_trace, NULL, bt))
 600                 goto err;
 601
 602         get_probe_ref();
 603
 604         ret = 0;
 605 err:
 606         if (ret)
 607                 blk_trace_free(bt);
 608         return ret;
 609 }
 610
 611 static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 612                              struct block_device *bdev, char __user *arg)
 613 {
 614         struct blk_user_trace_setup buts;
 615         int ret;
 616
 617         ret = copy_from_user(&buts, arg, sizeof(buts));
 618         if (ret)
 619                 return -EFAULT;
 620
 621         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 622         if (ret)
 623                 return ret;
 624
 625         if (copy_to_user(arg, &buts, sizeof(buts))) {
 626                 __blk_trace_remove(q);
 627                 return -EFAULT;
 628         }
 629         return 0;
 630 }
 631
 632 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
 633                     struct block_device *bdev,
 634                     char __user *arg)
 635 {
 636         int ret;
 637
 638         mutex_lock(&q->blk_trace_mutex);
 639         ret = __blk_trace_setup(q, name, dev, bdev, arg);
 640         mutex_unlock(&q->blk_trace_mutex);
 641
 642         return ret;
 643 }
 644 EXPORT_SYMBOL_GPL(blk_trace_setup);
 645
 646 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 647 static int compat_blk_trace_setup(struct request_queue *q, char *name,
 648                                   dev_t dev, struct block_device *bdev,
 649                                   char __user *arg)
 650 {
 651         struct blk_user_trace_setup buts;
 652         struct compat_blk_user_trace_setup cbuts;
 653         int ret;
 654
 655         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
 656                 return -EFAULT;
 657
 658         buts = (struct blk_user_trace_setup) {
 659                 .act_mask = cbuts.act_mask,
 660                 .buf_size = cbuts.buf_size,
 661                 .buf_nr = cbuts.buf_nr,
 662                 .start_lba = cbuts.start_lba,
 663                 .end_lba = cbuts.end_lba,
 664                 .pid = cbuts.pid,
 665         };
 666
 667         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
 668         if (ret)
 669                 return ret;
 670
 671         if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) {
 672                 __blk_trace_remove(q);
 673                 return -EFAULT;
 674         }
 675
 676         return 0;
 677 }
 678 #endif
 679
 680 static int __blk_trace_startstop(struct request_queue *q, int start)
 681 {
 682         int ret;
 683         struct blk_trace *bt;
 684
 685         bt = rcu_dereference_protected(q->blk_trace,
 686                                        lockdep_is_held(&q->blk_trace_mutex));
 687         if (bt == NULL)
 688                 return -EINVAL;
 689
 690         /*
 691          * For starting a trace, we can transition from a setup or stopped
 692          * trace. For stopping a trace, the state must be running
 693          */
 694         ret = -EINVAL;
 695         if (start) {
 696                 if (bt->trace_state == Blktrace_setup ||
 697                     bt->trace_state == Blktrace_stopped) {
 698                         blktrace_seq++;
 699                         smp_mb();
 700                         bt->trace_state = Blktrace_running;
 701                         spin_lock_irq(&running_trace_lock);
 702                         list_add(&bt->running_list, &running_trace_list);
 703                         spin_unlock_irq(&running_trace_lock);
 704
 705                         trace_note_time(bt);
 706                         ret = 0;
 707                 }
 708         } else {
 709                 if (bt->trace_state == Blktrace_running) {
 710                         bt->trace_state = Blktrace_stopped;
 711                         spin_lock_irq(&running_trace_lock);
 712                         list_del_init(&bt->running_list);
 713                         spin_unlock_irq(&running_trace_lock);
 714                         relay_flush(bt->rchan);
 715                         ret = 0;
 716                 }
 717         }
 718
 719         return ret;
 720 }
 721
 722 int blk_trace_startstop(struct request_queue *q, int start)
 723 {
 724         int ret;
 725
 726         mutex_lock(&q->blk_trace_mutex);
 727         ret = __blk_trace_startstop(q, start);
 728         mutex_unlock(&q->blk_trace_mutex);
 729
 730         return ret;
 731 }
 732 EXPORT_SYMBOL_GPL(blk_trace_startstop);
 733
 734 /*
 735  * When reading or writing the blktrace sysfs files, the references to the
 736  * opened sysfs or device files should prevent the underlying block device
 737  * from being removed. So no further delete protection is really needed.
 738  */
 739
 740 /**
 741  * blk_trace_ioctl: - handle the ioctls associated with tracing
 742  * @bdev:       the block device
 743  * @cmd:        the ioctl cmd
 744  * @arg:        the argument data, if any
 745  *
 746  **/
 747 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
 748 {
 749         struct request_queue *q;
 750         int ret, start = 0;
 751         char b[BDEVNAME_SIZE];
 752
 753         q = bdev_get_queue(bdev);
 754         if (!q)
 755                 return -ENXIO;
 756
 757         mutex_lock(&q->blk_trace_mutex);
 758
 759         switch (cmd) {
 760         case BLKTRACESETUP:
 761                 bdevname(bdev, b);
 762                 ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 763                 break;
 764 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
 765         case BLKTRACESETUP32:
 766                 bdevname(bdev, b);
 767                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
 768                 break;
 769 #endif
 770         case BLKTRACESTART:
 771                 start = 1;
 772         case BLKTRACESTOP:
 773                 ret = __blk_trace_startstop(q, start);
 774                 break;
 775         case BLKTRACETEARDOWN:
 776                 ret = __blk_trace_remove(q);
 777                 break;
 778         default:
 779                 ret = -ENOTTY;
 780                 break;
 781         }
 782
 783         mutex_unlock(&q->blk_trace_mutex);
 784         return ret;
 785 }
 786
 787 /**
 788  * blk_trace_shutdown: - stop and cleanup trace structures
 789  * @q:    the request queue associated with the device
 790  *
 791  **/
 792 void blk_trace_shutdown(struct request_queue *q)
 793 {
 794         mutex_lock(&q->blk_trace_mutex);
 795         if (rcu_dereference_protected(q->blk_trace,
 796                                       lockdep_is_held(&q->blk_trace_mutex))) {
 797                 __blk_trace_startstop(q, 0);
 798                 __blk_trace_remove(q);
 799         }
 800
 801         mutex_unlock(&q->blk_trace_mutex);
 802 }
 803
 804 #ifdef CONFIG_BLK_CGROUP
 805 static union kernfs_node_id *
 806 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 807 {
 808         struct blk_trace *bt;
 809
 810         /* We don't use the 'bt' value here except as an optimization... */
 811         bt = rcu_dereference_protected(q->blk_trace, 1);
 812         if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP))
 813                 return NULL;
 814
 815         if (!bio->bi_css)
 816                 return NULL;
 817         return cgroup_get_kernfs_id(bio->bi_css->cgroup);
 818 }
 819 #else
 820 static union kernfs_node_id *
 821 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio)
 822 {
 823         return NULL;
 824 }
 825 #endif
 826
 827 static union kernfs_node_id *
 828 blk_trace_request_get_cgid(struct request_queue *q, struct request *rq)
 829 {
 830         if (!rq->bio)
 831                 return NULL;
 832         /* Use the first bio */
 833         return blk_trace_bio_get_cgid(q, rq->bio);
 834 }
 835
 836 /*
 837  * blktrace probes
 838  */
 839
 840 /**
 841  * blk_add_trace_rq - Add a trace for a request oriented action
 842  * @rq:         the source request
 843  * @error:      return status to log
 844  * @nr_bytes:   number of completed bytes
 845  * @what:       the action
 846  * @cgid:       the cgroup info
 847  *
 848  * Description:
 849  *     Records an action against a request. Will log the bio offset + size.
 850  *
 851  **/
 852 static void blk_add_trace_rq(struct request *rq, int error,
 853                              unsigned int nr_bytes, u32 what,
 854                              union kernfs_node_id *cgid)
 855 {
 856         struct blk_trace *bt;
 857
 858         rcu_read_lock();
 859         bt = rcu_dereference(rq->q->blk_trace);
 860         if (likely(!bt)) {
 861                 rcu_read_unlock();
 862                 return;
 863         }
 864
 865         if (blk_rq_is_passthrough(rq))
 866                 what |= BLK_TC_ACT(BLK_TC_PC);
 867         else
 868                 what |= BLK_TC_ACT(BLK_TC_FS);
 869
 870         __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq),
 871                         rq->cmd_flags, what, error, 0, NULL, cgid);
 872         rcu_read_unlock();
 873 }
 874
 875 static void blk_add_trace_rq_insert(void *ignore,
 876                                     struct request_queue *q, struct request *rq)
 877 {
 878         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT,
 879                          blk_trace_request_get_cgid(q, rq));
 880 }
 881
 882 static void blk_add_trace_rq_issue(void *ignore,
 883                                    struct request_queue *q, struct request *rq)
 884 {
 885         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE,
 886                          blk_trace_request_get_cgid(q, rq));
 887 }
 888
 889 static void blk_add_trace_rq_requeue(void *ignore,
 890                                      struct request_queue *q,
 891                                      struct request *rq)
 892 {
 893         blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE,
 894                          blk_trace_request_get_cgid(q, rq));
 895 }
 896
 897 static void blk_add_trace_rq_complete(void *ignore, struct request *rq,
 898                         int error, unsigned int nr_bytes)
 899 {
 900         blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE,
 901                          blk_trace_request_get_cgid(rq->q, rq));
 902 }
 903
 904 /**
 905  * blk_add_trace_bio - Add a trace for a bio oriented action
 906  * @q:          queue the io is for
 907  * @bio:        the source bio
 908  * @what:       the action
 909  * @error:      error, if any
 910  *
 911  * Description:
 912  *     Records an action against a bio. Will log the bio offset + size.
 913  *
 914  **/
 915 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
 916                               u32 what, int error, union kernfs_node_id *cgid)
 917 {
 918         struct blk_trace *bt;
 919
 920         rcu_read_lock();
 921         bt = rcu_dereference(q->blk_trace);
 922         if (likely(!bt)) {
 923                 rcu_read_unlock();
 924                 return;
 925         }
 926
 927         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
 928                         bio_op(bio), bio->bi_opf, what, error, 0, NULL, cgid);
 929         rcu_read_unlock();
 930 }
 931
 932 static void blk_add_trace_bio_bounce(void *ignore,
 933                                      struct request_queue *q, struct bio *bio)
 934 {
 935         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0,
 936                           blk_trace_bio_get_cgid(q, bio));
 937 }
 938
 939 static void blk_add_trace_bio_complete(void *ignore,
 940                                        struct request_queue *q, struct bio *bio,
 941                                        int error)
 942 {
 943         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error,
 944                           blk_trace_bio_get_cgid(q, bio));
 945 }
 946
 947 static void blk_add_trace_bio_backmerge(void *ignore,
 948                                         struct request_queue *q,
 949                                         struct request *rq,
 950                                         struct bio *bio)
 951 {
 952         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0,
 953                          blk_trace_bio_get_cgid(q, bio));
 954 }
 955
 956 static void blk_add_trace_bio_frontmerge(void *ignore,
 957                                          struct request_queue *q,
 958                                          struct request *rq,
 959                                          struct bio *bio)
 960 {
 961         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0,
 962                           blk_trace_bio_get_cgid(q, bio));
 963 }
 964
 965 static void blk_add_trace_bio_queue(void *ignore,
 966                                     struct request_queue *q, struct bio *bio)
 967 {
 968         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0,
 969                           blk_trace_bio_get_cgid(q, bio));
 970 }
 971
 972 static void blk_add_trace_getrq(void *ignore,
 973                                 struct request_queue *q,
 974                                 struct bio *bio, int rw)
 975 {
 976         if (bio)
 977                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0,
 978                                   blk_trace_bio_get_cgid(q, bio));
 979         else {
 980                 struct blk_trace *bt;
 981
 982                 rcu_read_lock();
 983                 bt = rcu_dereference(q->blk_trace);
 984                 if (bt)
 985                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_GETRQ, 0, 0,
 986                                         NULL, NULL);
 987                 rcu_read_unlock();
 988         }
 989 }
 990
 991
 992 static void blk_add_trace_sleeprq(void *ignore,
 993                                   struct request_queue *q,
 994                                   struct bio *bio, int rw)
 995 {
 996         if (bio)
 997                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0,
 998                                   blk_trace_bio_get_cgid(q, bio));
 999         else {
1000                 struct blk_trace *bt;
1001
1002                 rcu_read_lock();
1003                 bt = rcu_dereference(q->blk_trace);
1004                 if (bt)
1005                         __blk_add_trace(bt, 0, 0, rw, 0, BLK_TA_SLEEPRQ,
1006                                         0, 0, NULL, NULL);
1007                 rcu_read_unlock();
1008         }
1009 }
1010
1011 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
1012 {
1013         struct blk_trace *bt;
1014
1015         rcu_read_lock();
1016         bt = rcu_dereference(q->blk_trace);
1017         if (bt)
1018                 __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, NULL);
1019         rcu_read_unlock();
1020 }
1021
1022 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
1023                                     unsigned int depth, bool explicit)
1024 {
1025         struct blk_trace *bt;
1026
1027         rcu_read_lock();
1028         bt = rcu_dereference(q->blk_trace);
1029         if (bt) {
1030                 __be64 rpdu = cpu_to_be64(depth);
1031                 u32 what;
1032
1033                 if (explicit)
1034                         what = BLK_TA_UNPLUG_IO;
1035                 else
1036                         what = BLK_TA_UNPLUG_TIMER;
1037
1038                 __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, NULL);
1039         }
1040         rcu_read_unlock();
1041 }
1042
1043 static void blk_add_trace_split(void *ignore,
1044                                 struct request_queue *q, struct bio *bio,
1045                                 unsigned int pdu)
1046 {
1047         struct blk_trace *bt;
1048
1049         rcu_read_lock();
1050         bt = rcu_dereference(q->blk_trace);
1051         if (bt) {
1052                 __be64 rpdu = cpu_to_be64(pdu);
1053
1054                 __blk_add_trace(bt, bio->bi_iter.bi_sector,
1055                                 bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf,
1056                                 BLK_TA_SPLIT,
1057                                 blk_status_to_errno(bio->bi_status),
1058                                 sizeof(rpdu), &rpdu,
1059                                 blk_trace_bio_get_cgid(q, bio));
1060         }
1061         rcu_read_unlock();
1062 }
1063
1064 /**
1065  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
1066  * @ignore:     trace callback data parameter (not used)
1067  * @q:          queue the io is for
1068  * @bio:        the source bio
1069  * @dev:        target device
1070  * @from:       source sector
1071  *
1072  * Description:
1073  *     Device mapper or raid target sometimes need to split a bio because
1074  *     it spans a stripe (or similar). Add a trace for that action.
1075  *
1076  **/
1077 static void blk_add_trace_bio_remap(void *ignore,
1078                                     struct request_queue *q, struct bio *bio,
1079                                     dev_t dev, sector_t from)
1080 {
1081         struct blk_trace *bt;
1082         struct blk_io_trace_remap r;
1083
1084         rcu_read_lock();
1085         bt = rcu_dereference(q->blk_trace);
1086         if (likely(!bt)) {
1087                 rcu_read_unlock();
1088                 return;
1089         }
1090
1091         r.device_from = cpu_to_be32(dev);
1092         r.device_to   = cpu_to_be32(bio_dev(bio));
1093         r.sector_from = cpu_to_be64(from);
1094
1095         __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size,
1096                         bio_op(bio), bio->bi_opf, BLK_TA_REMAP,
1097                         blk_status_to_errno(bio->bi_status),
1098                         sizeof(r), &r, blk_trace_bio_get_cgid(q, bio));
1099         rcu_read_unlock();
1100 }
1101
1102 /**
1103  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
1104  * @ignore:     trace callback data parameter (not used)
1105  * @q:          queue the io is for
1106  * @rq:         the source request
1107  * @dev:        target device
1108  * @from:       source sector
1109  *
1110  * Description:
1111  *     Device mapper remaps request to other devices.
1112  *     Add a trace for that action.
1113  *
1114  **/
1115 static void blk_add_trace_rq_remap(void *ignore,
1116                                    struct request_queue *q,
1117                                    struct request *rq, dev_t dev,
1118                                    sector_t from)
1119 {
1120         struct blk_trace *bt;
1121         struct blk_io_trace_remap r;
1122
1123         rcu_read_lock();
1124         bt = rcu_dereference(q->blk_trace);
1125         if (likely(!bt)) {
1126                 rcu_read_unlock();
1127                 return;
1128         }
1129
1130         r.device_from = cpu_to_be32(dev);
1131         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
1132         r.sector_from = cpu_to_be64(from);
1133
1134         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
1135                         rq_data_dir(rq), 0, BLK_TA_REMAP, 0,
1136                         sizeof(r), &r, blk_trace_request_get_cgid(q, rq));
1137         rcu_read_unlock();
1138 }
1139
1140 /**
1141  * blk_add_driver_data - Add binary message with driver-specific data
1142  * @q:          queue the io is for
1143  * @rq:         io request
1144  * @data:       driver-specific data
1145  * @len:        length of driver-specific data
1146  *
1147  * Description:
1148  *     Some drivers might want to write driver-specific data per request.
1149  *
1150  **/
1151 void blk_add_driver_data(struct request_queue *q,
1152                          struct request *rq,
1153                          void *data, size_t len)
1154 {
1155         struct blk_trace *bt;
1156
1157         rcu_read_lock();
1158         bt = rcu_dereference(q->blk_trace);
1159         if (likely(!bt)) {
1160                 rcu_read_unlock();
1161                 return;
1162         }
1163
1164         __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0,
1165                                 BLK_TA_DRV_DATA, 0, len, data,
1166                                 blk_trace_request_get_cgid(q, rq));
1167         rcu_read_unlock();
1168 }
1169 EXPORT_SYMBOL_GPL(blk_add_driver_data);
1170
1171 static void blk_register_tracepoints(void)
1172 {
1173         int ret;
1174
1175         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1176         WARN_ON(ret);
1177         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1178         WARN_ON(ret);
1179         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1180         WARN_ON(ret);
1181         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1182         WARN_ON(ret);
1183         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1184         WARN_ON(ret);
1185         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1186         WARN_ON(ret);
1187         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1188         WARN_ON(ret);
1189         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1190         WARN_ON(ret);
1191         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1192         WARN_ON(ret);
1193         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1194         WARN_ON(ret);
1195         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1196         WARN_ON(ret);
1197         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1198         WARN_ON(ret);
1199         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1200         WARN_ON(ret);
1201         ret = register_trace_block_split(blk_add_trace_split, NULL);
1202         WARN_ON(ret);
1203         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1204         WARN_ON(ret);
1205         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1206         WARN_ON(ret);
1207 }
1208
1209 static void blk_unregister_tracepoints(void)
1210 {
1211         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1212         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1213         unregister_trace_block_split(blk_add_trace_split, NULL);
1214         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1215         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1216         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1217         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1218         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1219         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1220         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1221         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1222         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1223         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1224         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1225         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1226         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1227
1228         tracepoint_synchronize_unregister();
1229 }
1230
1231 /*
1232  * struct blk_io_tracer formatting routines
1233  */
1234
1235 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1236 {
1237         int i = 0;
1238         int tc = t->action >> BLK_TC_SHIFT;
1239
1240         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1241                 rwbs[i++] = 'N';
1242                 goto out;
1243         }
1244
1245         if (tc & BLK_TC_FLUSH)
1246                 rwbs[i++] = 'F';
1247
1248         if (tc & BLK_TC_DISCARD)
1249                 rwbs[i++] = 'D';
1250         else if (tc & BLK_TC_WRITE)
1251                 rwbs[i++] = 'W';
1252         else if (t->bytes)
1253                 rwbs[i++] = 'R';
1254         else
1255                 rwbs[i++] = 'N';
1256
1257         if (tc & BLK_TC_FUA)
1258                 rwbs[i++] = 'F';
1259         if (tc & BLK_TC_AHEAD)
1260                 rwbs[i++] = 'A';
1261         if (tc & BLK_TC_SYNC)
1262                 rwbs[i++] = 'S';
1263         if (tc & BLK_TC_META)
1264                 rwbs[i++] = 'M';
1265 out:
1266         rwbs[i] = '\0';
1267 }
1268
1269 static inline
1270 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1271 {
1272         return (const struct blk_io_trace *)ent;
1273 }
1274
1275 static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg)
1276 {
1277         return (void *)(te_blk_io_trace(ent) + 1) +
1278                 (has_cg ? sizeof(union kernfs_node_id) : 0);
1279 }
1280
1281 static inline const void *cgid_start(const struct trace_entry *ent)
1282 {
1283         return (void *)(te_blk_io_trace(ent) + 1);
1284 }
1285
1286 static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg)
1287 {
1288         return te_blk_io_trace(ent)->pdu_len -
1289                         (has_cg ? sizeof(union kernfs_node_id) : 0);
1290 }
1291
1292 static inline u32 t_action(const struct trace_entry *ent)
1293 {
1294         return te_blk_io_trace(ent)->action;
1295 }
1296
1297 static inline u32 t_bytes(const struct trace_entry *ent)
1298 {
1299         return te_blk_io_trace(ent)->bytes;
1300 }
1301
1302 static inline u32 t_sec(const struct trace_entry *ent)
1303 {
1304         return te_blk_io_trace(ent)->bytes >> 9;
1305 }
1306
1307 static inline unsigned long long t_sector(const struct trace_entry *ent)
1308 {
1309         return te_blk_io_trace(ent)->sector;
1310 }
1311
1312 static inline __u16 t_error(const struct trace_entry *ent)
1313 {
1314         return te_blk_io_trace(ent)->error;
1315 }
1316
1317 static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg)
1318 {
1319         const __be64 *val = pdu_start(ent, has_cg);
1320         return be64_to_cpu(*val);
1321 }
1322
1323 typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act,
1324         bool has_cg);
1325
1326 static void blk_log_action_classic(struct trace_iterator *iter, const char *act,
1327         bool has_cg)
1328 {
1329         char rwbs[RWBS_LEN];
1330         unsigned long long ts  = iter->ts;
1331         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1332         unsigned secs          = (unsigned long)ts;
1333         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1334
1335         fill_rwbs(rwbs, t);
1336
1337         trace_seq_printf(&iter->seq,
1338                          "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1339                          MAJOR(t->device), MINOR(t->device), iter->cpu,
1340                          secs, nsec_rem, iter->ent->pid, act, rwbs);
1341 }
1342
1343 static void blk_log_action(struct trace_iterator *iter, const char *act,
1344         bool has_cg)
1345 {
1346         char rwbs[RWBS_LEN];
1347         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1348
1349         fill_rwbs(rwbs, t);
1350         if (has_cg) {
1351                 const union kernfs_node_id *id = cgid_start(iter->ent);
1352
1353                 if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) {
1354                         char blkcg_name_buf[NAME_MAX + 1] = "<...>";
1355
1356                         cgroup_path_from_kernfs_id(id, blkcg_name_buf,
1357                                 sizeof(blkcg_name_buf));
1358                         trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ",
1359                                  MAJOR(t->device), MINOR(t->device),
1360                                  blkcg_name_buf, act, rwbs);
1361                 } else
1362                         trace_seq_printf(&iter->seq,
1363                                  "%3d,%-3d %x,%-x %2s %3s ",
1364                                  MAJOR(t->device), MINOR(t->device),
1365                                  id->ino, id->generation, act, rwbs);
1366         } else
1367                 trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1368                                  MAJOR(t->device), MINOR(t->device), act, rwbs);
1369 }
1370
1371 static void blk_log_dump_pdu(struct trace_seq *s,
1372         const struct trace_entry *ent, bool has_cg)
1373 {
1374         const unsigned char *pdu_buf;
1375         int pdu_len;
1376         int i, end;
1377
1378         pdu_buf = pdu_start(ent, has_cg);
1379         pdu_len = pdu_real_len(ent, has_cg);
1380
1381         if (!pdu_len)
1382                 return;
1383
1384         /* find the last zero that needs to be printed */
1385         for (end = pdu_len - 1; end >= 0; end--)
1386                 if (pdu_buf[end])
1387                         break;
1388         end++;
1389
1390         trace_seq_putc(s, '(');
1391
1392         for (i = 0; i < pdu_len; i++) {
1393
1394                 trace_seq_printf(s, "%s%02x",
1395                                  i == 0 ? "" : " ", pdu_buf[i]);
1396
1397                 /*
1398                  * stop when the rest is just zeroes and indicate so
1399                  * with a ".." appended
1400                  */
1401                 if (i == end && end != pdu_len - 1) {
1402                         trace_seq_puts(s, " ..) ");
1403                         return;
1404                 }
1405         }
1406
1407         trace_seq_puts(s, ") ");
1408 }
1409
1410 static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1411 {
1412         char cmd[TASK_COMM_LEN];
1413
1414         trace_find_cmdline(ent->pid, cmd);
1415
1416         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1417                 trace_seq_printf(s, "%u ", t_bytes(ent));
1418                 blk_log_dump_pdu(s, ent, has_cg);
1419                 trace_seq_printf(s, "[%s]\n", cmd);
1420         } else {
1421                 if (t_sec(ent))
1422                         trace_seq_printf(s, "%llu + %u [%s]\n",
1423                                                 t_sector(ent), t_sec(ent), cmd);
1424                 else
1425                         trace_seq_printf(s, "[%s]\n", cmd);
1426         }
1427 }
1428
1429 static void blk_log_with_error(struct trace_seq *s,
1430                               const struct trace_entry *ent, bool has_cg)
1431 {
1432         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1433                 blk_log_dump_pdu(s, ent, has_cg);
1434                 trace_seq_printf(s, "[%d]\n", t_error(ent));
1435         } else {
1436                 if (t_sec(ent))
1437                         trace_seq_printf(s, "%llu + %u [%d]\n",
1438                                          t_sector(ent),
1439                                          t_sec(ent), t_error(ent));
1440                 else
1441                         trace_seq_printf(s, "%llu [%d]\n",
1442                                          t_sector(ent), t_error(ent));
1443         }
1444 }
1445
1446 static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1447 {
1448         const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg);
1449
1450         trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1451                          t_sector(ent), t_sec(ent),
1452                          MAJOR(be32_to_cpu(__r->device_from)),
1453                          MINOR(be32_to_cpu(__r->device_from)),
1454                          be64_to_cpu(__r->sector_from));
1455 }
1456
1457 static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1458 {
1459         char cmd[TASK_COMM_LEN];
1460
1461         trace_find_cmdline(ent->pid, cmd);
1462
1463         trace_seq_printf(s, "[%s]\n", cmd);
1464 }
1465
1466 static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1467 {
1468         char cmd[TASK_COMM_LEN];
1469
1470         trace_find_cmdline(ent->pid, cmd);
1471
1472         trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg));
1473 }
1474
1475 static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg)
1476 {
1477         char cmd[TASK_COMM_LEN];
1478
1479         trace_find_cmdline(ent->pid, cmd);
1480
1481         trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1482                          get_pdu_int(ent, has_cg), cmd);
1483 }
1484
1485 static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent,
1486                         bool has_cg)
1487 {
1488
1489         trace_seq_putmem(s, pdu_start(ent, has_cg),
1490                 pdu_real_len(ent, has_cg));
1491         trace_seq_putc(s, '\n');
1492 }
1493
1494 /*
1495  * struct tracer operations
1496  */
1497
1498 static void blk_tracer_print_header(struct seq_file *m)
1499 {
1500         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1501                 return;
1502         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1503                     "#  |     |     |           |   |   |\n");
1504 }
1505
1506 static void blk_tracer_start(struct trace_array *tr)
1507 {
1508         blk_tracer_enabled = true;
1509 }
1510
1511 static int blk_tracer_init(struct trace_array *tr)
1512 {
1513         blk_tr = tr;
1514         blk_tracer_start(tr);
1515         return 0;
1516 }
1517
1518 static void blk_tracer_stop(struct trace_array *tr)
1519 {
1520         blk_tracer_enabled = false;
1521 }
1522
1523 static void blk_tracer_reset(struct trace_array *tr)
1524 {
1525         blk_tracer_stop(tr);
1526 }
1527
1528 static const struct {
1529         const char *act[2];
1530         void       (*print)(struct trace_seq *s, const struct trace_entry *ent,
1531                             bool has_cg);
1532 } what2act[] = {
1533         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1534         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1535         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1536         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1537         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1538         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1539         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1540         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1541         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1542         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1543         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1544         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1545         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1546         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1547         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1548 };
1549
1550 static enum print_line_t print_one_line(struct trace_iterator *iter,
1551                                         bool classic)
1552 {
1553         struct trace_array *tr = iter->tr;
1554         struct trace_seq *s = &iter->seq;
1555         const struct blk_io_trace *t;
1556         u16 what;
1557         bool long_act;
1558         blk_log_action_t *log_action;
1559         bool has_cg;
1560
1561         t          = te_blk_io_trace(iter->ent);
1562         what       = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP;
1563         long_act   = !!(tr->trace_flags & TRACE_ITER_VERBOSE);
1564         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1565         has_cg     = t->action & __BLK_TA_CGROUP;
1566
1567         if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) {
1568                 log_action(iter, long_act ? "message" : "m", has_cg);
1569                 blk_log_msg(s, iter->ent, has_cg);
1570                 return trace_handle_return(s);
1571         }
1572
1573         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1574                 trace_seq_printf(s, "Unknown action %x\n", what);
1575         else {
1576                 log_action(iter, what2act[what].act[long_act], has_cg);
1577                 what2act[what].print(s, iter->ent, has_cg);
1578         }
1579
1580         return trace_handle_return(s);
1581 }
1582
1583 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1584                                                int flags, struct trace_event *event)
1585 {
1586         return print_one_line(iter, false);
1587 }
1588
1589 static void blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1590 {
1591         struct trace_seq *s = &iter->seq;
1592         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1593         const int offset = offsetof(struct blk_io_trace, sector);
1594         struct blk_io_trace old = {
1595                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1596                 .time     = iter->ts,
1597         };
1598
1599         trace_seq_putmem(s, &old, offset);
1600         trace_seq_putmem(s, &t->sector,
1601                          sizeof(old) - offset + t->pdu_len);
1602 }
1603
1604 static enum print_line_t
1605 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1606                              struct trace_event *event)
1607 {
1608         blk_trace_synthesize_old_trace(iter);
1609
1610         return trace_handle_return(&iter->seq);
1611 }
1612
1613 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1614 {
1615         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1616                 return TRACE_TYPE_UNHANDLED;
1617
1618         return print_one_line(iter, true);
1619 }
1620
1621 static int
1622 blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1623 {
1624         /* don't output context-info for blk_classic output */
1625         if (bit == TRACE_BLK_OPT_CLASSIC) {
1626                 if (set)
1627                         tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1628                 else
1629                         tr->trace_flags |= TRACE_ITER_CONTEXT_INFO;
1630         }
1631         return 0;
1632 }
1633
1634 static struct tracer blk_tracer __read_mostly = {
1635         .name           = "blk",
1636         .init           = blk_tracer_init,
1637         .reset          = blk_tracer_reset,
1638         .start          = blk_tracer_start,
1639         .stop           = blk_tracer_stop,
1640         .print_header   = blk_tracer_print_header,
1641         .print_line     = blk_tracer_print_line,
1642         .flags          = &blk_tracer_flags,
1643         .set_flag       = blk_tracer_set_flag,
1644 };
1645
1646 static struct trace_event_functions trace_blk_event_funcs = {
1647         .trace          = blk_trace_event_print,
1648         .binary         = blk_trace_event_print_binary,
1649 };
1650
1651 static struct trace_event trace_blk_event = {
1652         .type           = TRACE_BLK,
1653         .funcs          = &trace_blk_event_funcs,
1654 };
1655
1656 static int __init init_blk_tracer(void)
1657 {
1658         if (!register_trace_event(&trace_blk_event)) {
1659                 pr_warn("Warning: could not register block events\n");
1660                 return 1;
1661         }
1662
1663         if (register_tracer(&blk_tracer) != 0) {
1664                 pr_warn("Warning: could not register the block tracer\n");
1665                 unregister_trace_event(&trace_blk_event);
1666                 return 1;
1667         }
1668
1669         return 0;
1670 }
1671
1672 device_initcall(init_blk_tracer);
1673
1674 static int blk_trace_remove_queue(struct request_queue *q)
1675 {
1676         struct blk_trace *bt;
1677
1678         bt = xchg(&q->blk_trace, NULL);
1679         if (bt == NULL)
1680                 return -EINVAL;
1681
1682         if (bt->trace_state == Blktrace_running) {
1683                 bt->trace_state = Blktrace_stopped;
1684                 spin_lock_irq(&running_trace_lock);
1685                 list_del_init(&bt->running_list);
1686                 spin_unlock_irq(&running_trace_lock);
1687                 relay_flush(bt->rchan);
1688         }
1689
1690         put_probe_ref();
1691         synchronize_rcu();
1692         blk_trace_free(bt);
1693         return 0;
1694 }
1695
1696 /*
1697  * Setup everything required to start tracing
1698  */
1699 static int blk_trace_setup_queue(struct request_queue *q,
1700                                  struct block_device *bdev)
1701 {
1702         struct blk_trace *bt = NULL;
1703         int ret = -ENOMEM;
1704
1705         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1706         if (!bt)
1707                 return -ENOMEM;
1708
1709         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1710         if (!bt->msg_data)
1711                 goto free_bt;
1712
1713         bt->dev = bdev->bd_dev;
1714         bt->act_mask = (u16)-1;
1715
1716         blk_trace_setup_lba(bt, bdev);
1717
1718         ret = -EBUSY;
1719         if (cmpxchg(&q->blk_trace, NULL, bt))
1720                 goto free_bt;
1721
1722         get_probe_ref();
1723         return 0;
1724
1725 free_bt:
1726         blk_trace_free(bt);
1727         return ret;
1728 }
1729
1730 /*
1731  * sysfs interface to enable and configure tracing
1732  */
1733
1734 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1735                                          struct device_attribute *attr,
1736                                          char *buf);
1737 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1738                                           struct device_attribute *attr,
1739                                           const char *buf, size_t count);
1740 #define BLK_TRACE_DEVICE_ATTR(_name) \
1741         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1742                     sysfs_blk_trace_attr_show, \
1743                     sysfs_blk_trace_attr_store)
1744
1745 static BLK_TRACE_DEVICE_ATTR(enable);
1746 static BLK_TRACE_DEVICE_ATTR(act_mask);
1747 static BLK_TRACE_DEVICE_ATTR(pid);
1748 static BLK_TRACE_DEVICE_ATTR(start_lba);
1749 static BLK_TRACE_DEVICE_ATTR(end_lba);
1750
1751 static struct attribute *blk_trace_attrs[] = {
1752         &dev_attr_enable.attr,
1753         &dev_attr_act_mask.attr,
1754         &dev_attr_pid.attr,
1755         &dev_attr_start_lba.attr,
1756         &dev_attr_end_lba.attr,
1757         NULL
1758 };
1759
1760 struct attribute_group blk_trace_attr_group = {
1761         .name  = "trace",
1762         .attrs = blk_trace_attrs,
1763 };
1764
1765 static const struct {
1766         int mask;
1767         const char *str;
1768 } mask_maps[] = {
1769         { BLK_TC_READ,          "read"          },
1770         { BLK_TC_WRITE,         "write"         },
1771         { BLK_TC_FLUSH,         "flush"         },
1772         { BLK_TC_SYNC,          "sync"          },
1773         { BLK_TC_QUEUE,         "queue"         },
1774         { BLK_TC_REQUEUE,       "requeue"       },
1775         { BLK_TC_ISSUE,         "issue"         },
1776         { BLK_TC_COMPLETE,      "complete"      },
1777         { BLK_TC_FS,            "fs"            },
1778         { BLK_TC_PC,            "pc"            },
1779         { BLK_TC_NOTIFY,        "notify"        },
1780         { BLK_TC_AHEAD,         "ahead"         },
1781         { BLK_TC_META,          "meta"          },
1782         { BLK_TC_DISCARD,       "discard"       },
1783         { BLK_TC_DRV_DATA,      "drv_data"      },
1784         { BLK_TC_FUA,           "fua"           },
1785 };
1786
1787 static int blk_trace_str2mask(const char *str)
1788 {
1789         int i;
1790         int mask = 0;
1791         char *buf, *s, *token;
1792
1793         buf = kstrdup(str, GFP_KERNEL);
1794         if (buf == NULL)
1795                 return -ENOMEM;
1796         s = strstrip(buf);
1797
1798         while (1) {
1799                 token = strsep(&s, ",");
1800                 if (token == NULL)
1801                         break;
1802
1803                 if (*token == '\0')
1804                         continue;
1805
1806                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1807                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1808                                 mask |= mask_maps[i].mask;
1809                                 break;
1810                         }
1811                 }
1812                 if (i == ARRAY_SIZE(mask_maps)) {
1813                         mask = -EINVAL;
1814                         break;
1815                 }
1816         }
1817         kfree(buf);
1818
1819         return mask;
1820 }
1821
1822 static ssize_t blk_trace_mask2str(char *buf, int mask)
1823 {
1824         int i;
1825         char *p = buf;
1826
1827         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1828                 if (mask & mask_maps[i].mask) {
1829                         p += sprintf(p, "%s%s",
1830                                     (p == buf) ? "" : ",", mask_maps[i].str);
1831                 }
1832         }
1833         *p++ = '\n';
1834
1835         return p - buf;
1836 }
1837
1838 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1839 {
1840         if (bdev->bd_disk == NULL)
1841                 return NULL;
1842
1843         return bdev_get_queue(bdev);
1844 }
1845
1846 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1847                                          struct device_attribute *attr,
1848                                          char *buf)
1849 {
1850         struct hd_struct *p = dev_to_part(dev);
1851         struct request_queue *q;
1852         struct block_device *bdev;
1853         struct blk_trace *bt;
1854         ssize_t ret = -ENXIO;
1855
1856         bdev = bdget(part_devt(p));
1857         if (bdev == NULL)
1858                 goto out;
1859
1860         q = blk_trace_get_queue(bdev);
1861         if (q == NULL)
1862                 goto out_bdput;
1863
1864         mutex_lock(&q->blk_trace_mutex);
1865
1866         bt = rcu_dereference_protected(q->blk_trace,
1867                                        lockdep_is_held(&q->blk_trace_mutex));
1868         if (attr == &dev_attr_enable) {
1869                 ret = sprintf(buf, "%u\n", !!bt);
1870                 goto out_unlock_bdev;
1871         }
1872
1873         if (bt == NULL)
1874                 ret = sprintf(buf, "disabled\n");
1875         else if (attr == &dev_attr_act_mask)
1876                 ret = blk_trace_mask2str(buf, bt->act_mask);
1877         else if (attr == &dev_attr_pid)
1878                 ret = sprintf(buf, "%u\n", bt->pid);
1879         else if (attr == &dev_attr_start_lba)
1880                 ret = sprintf(buf, "%llu\n", bt->start_lba);
1881         else if (attr == &dev_attr_end_lba)
1882                 ret = sprintf(buf, "%llu\n", bt->end_lba);
1883
1884 out_unlock_bdev:
1885         mutex_unlock(&q->blk_trace_mutex);
1886 out_bdput:
1887         bdput(bdev);
1888 out:
1889         return ret;
1890 }
1891
1892 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1893                                           struct device_attribute *attr,
1894                                           const char *buf, size_t count)
1895 {
1896         struct block_device *bdev;
1897         struct request_queue *q;
1898         struct hd_struct *p;
1899         struct blk_trace *bt;
1900         u64 value;
1901         ssize_t ret = -EINVAL;
1902
1903         if (count == 0)
1904                 goto out;
1905
1906         if (attr == &dev_attr_act_mask) {
1907                 if (kstrtoull(buf, 0, &value)) {
1908                         /* Assume it is a list of trace category names */
1909                         ret = blk_trace_str2mask(buf);
1910                         if (ret < 0)
1911                                 goto out;
1912                         value = ret;
1913                 }
1914         } else if (kstrtoull(buf, 0, &value))
1915                 goto out;
1916
1917         ret = -ENXIO;
1918
1919         p = dev_to_part(dev);
1920         bdev = bdget(part_devt(p));
1921         if (bdev == NULL)
1922                 goto out;
1923
1924         q = blk_trace_get_queue(bdev);
1925         if (q == NULL)
1926                 goto out_bdput;
1927
1928         mutex_lock(&q->blk_trace_mutex);
1929
1930         bt = rcu_dereference_protected(q->blk_trace,
1931                                        lockdep_is_held(&q->blk_trace_mutex));
1932         if (attr == &dev_attr_enable) {
1933                 if (!!value == !!bt) {
1934                         ret = 0;
1935                         goto out_unlock_bdev;
1936                 }
1937                 if (value)
1938                         ret = blk_trace_setup_queue(q, bdev);
1939                 else
1940                         ret = blk_trace_remove_queue(q);
1941                 goto out_unlock_bdev;
1942         }
1943
1944         ret = 0;
1945         if (bt == NULL) {
1946                 ret = blk_trace_setup_queue(q, bdev);
1947                 bt = rcu_dereference_protected(q->blk_trace,
1948                                 lockdep_is_held(&q->blk_trace_mutex));
1949         }
1950
1951         if (ret == 0) {
1952                 if (attr == &dev_attr_act_mask)
1953                         bt->act_mask = value;
1954                 else if (attr == &dev_attr_pid)
1955                         bt->pid = value;
1956                 else if (attr == &dev_attr_start_lba)
1957                         bt->start_lba = value;
1958                 else if (attr == &dev_attr_end_lba)
1959                         bt->end_lba = value;
1960         }
1961
1962 out_unlock_bdev:
1963         mutex_unlock(&q->blk_trace_mutex);
1964 out_bdput:
1965         bdput(bdev);
1966 out:
1967         return ret ? ret : count;
1968 }
1969
1970 int blk_trace_init_sysfs(struct device *dev)
1971 {
1972         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1973 }
1974
1975 void blk_trace_remove_sysfs(struct device *dev)
1976 {
1977         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1978 }
1979
1980 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1981
1982 #ifdef CONFIG_EVENT_TRACING
1983
1984 void blk_fill_rwbs(char *rwbs, unsigned int op, int bytes)
1985 {
1986         int i = 0;
1987
1988         if (op & REQ_PREFLUSH)
1989                 rwbs[i++] = 'F';
1990
1991         switch (op & REQ_OP_MASK) {
1992         case REQ_OP_WRITE:
1993         case REQ_OP_WRITE_SAME:
1994                 rwbs[i++] = 'W';
1995                 break;
1996         case REQ_OP_DISCARD:
1997                 rwbs[i++] = 'D';
1998                 break;
1999         case REQ_OP_SECURE_ERASE:
2000                 rwbs[i++] = 'D';
2001                 rwbs[i++] = 'E';
2002                 break;
2003         case REQ_OP_FLUSH:
2004                 rwbs[i++] = 'F';
2005                 break;
2006         case REQ_OP_READ:
2007                 rwbs[i++] = 'R';
2008                 break;
2009         default:
2010                 rwbs[i++] = 'N';
2011         }
2012
2013         if (op & REQ_FUA)
2014                 rwbs[i++] = 'F';
2015         if (op & REQ_RAHEAD)
2016                 rwbs[i++] = 'A';
2017         if (op & REQ_SYNC)
2018                 rwbs[i++] = 'S';
2019         if (op & REQ_META)
2020                 rwbs[i++] = 'M';
2021
2022         rwbs[i] = '\0';
2023 }
2024 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
2025
2026 #endif /* CONFIG_EVENT_TRACING */
2027