~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/trace/blktrace.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.72 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp