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

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

Version: ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ 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-3.16.85 ] ~ [ 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-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 
 30 #include <trace/events/block.h>
 31 
 32 #include "trace_output.h"
 33 
 34 #ifdef CONFIG_BLK_DEV_IO_TRACE
 35 
 36 static unsigned int blktrace_seq __read_mostly = 1;
 37 
 38 static struct trace_array *blk_tr;
 39 static bool blk_tracer_enabled __read_mostly;
 40 
 41 /* Select an alternative, minimalistic output than the original one */
 42 #define TRACE_BLK_OPT_CLASSIC   0x1
 43 
 44 static struct tracer_opt blk_tracer_opts[] = {
 45         /* Default disable the minimalistic output */
 46         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
 47         { }
 48 };
 49 
 50 static struct tracer_flags blk_tracer_flags = {
 51         .val  = 0,
 52         .opts = blk_tracer_opts,
 53 };
 54 
 55 /* Global reference count of probes */
 56 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
 57 
 58 static void blk_register_tracepoints(void);
 59 static void blk_unregister_tracepoints(void);
 60 
 61 /*
 62  * Send out a notify message.
 63  */
 64 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
 65                        const void *data, size_t len)
 66 {
 67         struct blk_io_trace *t;
 68         struct ring_buffer_event *event = NULL;
 69         struct ring_buffer *buffer = NULL;
 70         int pc = 0;
 71         int cpu = smp_processor_id();
 72         bool blk_tracer = blk_tracer_enabled;
 73 
 74         if (blk_tracer) {
 75                 buffer = blk_tr->trace_buffer.buffer;
 76                 pc = preempt_count();
 77                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
 78                                                   sizeof(*t) + len,
 79                                                   0, pc);
 80                 if (!event)
 81                         return;
 82                 t = ring_buffer_event_data(event);
 83                 goto record_it;
 84         }
 85 
 86         if (!bt->rchan)
 87                 return;
 88 
 89         t = relay_reserve(bt->rchan, sizeof(*t) + len);
 90         if (t) {
 91                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
 92                 t->time = ktime_to_ns(ktime_get());
 93 record_it:
 94                 t->device = bt->dev;
 95                 t->action = action;
 96                 t->pid = pid;
 97                 t->cpu = cpu;
 98                 t->pdu_len = len;
 99                 memcpy((void *) t + sizeof(*t), data, len);
100 
101                 if (blk_tracer)
102                         trace_buffer_unlock_commit(buffer, event, 0, pc);
103         }
104 }
105 
106 /*
107  * Send out a notify for this process, if we haven't done so since a trace
108  * started
109  */
110 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
111 {
112         tsk->btrace_seq = blktrace_seq;
113         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
114 }
115 
116 static void trace_note_time(struct blk_trace *bt)
117 {
118         struct timespec now;
119         unsigned long flags;
120         u32 words[2];
121 
122         getnstimeofday(&now);
123         words[0] = now.tv_sec;
124         words[1] = now.tv_nsec;
125 
126         local_irq_save(flags);
127         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
128         local_irq_restore(flags);
129 }
130 
131 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
132 {
133         int n;
134         va_list args;
135         unsigned long flags;
136         char *buf;
137 
138         if (unlikely(bt->trace_state != Blktrace_running &&
139                      !blk_tracer_enabled))
140                 return;
141 
142         /*
143          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
144          * message to the trace.
145          */
146         if (!(bt->act_mask & BLK_TC_NOTIFY))
147                 return;
148 
149         local_irq_save(flags);
150         buf = this_cpu_ptr(bt->msg_data);
151         va_start(args, fmt);
152         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
153         va_end(args);
154 
155         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
156         local_irq_restore(flags);
157 }
158 EXPORT_SYMBOL_GPL(__trace_note_message);
159 
160 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
161                          pid_t pid)
162 {
163         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
164                 return 1;
165         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
166                 return 1;
167         if (bt->pid && pid != bt->pid)
168                 return 1;
169 
170         return 0;
171 }
172 
173 /*
174  * Data direction bit lookup
175  */
176 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
177                                  BLK_TC_ACT(BLK_TC_WRITE) };
178 
179 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
180 
181 /* The ilog2() calls fall out because they're constant */
182 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
183           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
184 
185 /*
186  * The worker for the various blk_add_trace*() types. Fills out a
187  * blk_io_trace structure and places it in a per-cpu subbuffer.
188  */
189 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
190                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
191 {
192         struct task_struct *tsk = current;
193         struct ring_buffer_event *event = NULL;
194         struct ring_buffer *buffer = NULL;
195         struct blk_io_trace *t;
196         unsigned long flags = 0;
197         unsigned long *sequence;
198         pid_t pid;
199         int cpu, pc = 0;
200         bool blk_tracer = blk_tracer_enabled;
201 
202         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
203                 return;
204 
205         what |= ddir_act[rw & WRITE];
206         what |= MASK_TC_BIT(rw, SYNC);
207         what |= MASK_TC_BIT(rw, RAHEAD);
208         what |= MASK_TC_BIT(rw, META);
209         what |= MASK_TC_BIT(rw, DISCARD);
210         what |= MASK_TC_BIT(rw, FLUSH);
211         what |= MASK_TC_BIT(rw, FUA);
212 
213         pid = tsk->pid;
214         if (act_log_check(bt, what, sector, pid))
215                 return;
216         cpu = raw_smp_processor_id();
217 
218         if (blk_tracer) {
219                 tracing_record_cmdline(current);
220 
221                 buffer = blk_tr->trace_buffer.buffer;
222                 pc = preempt_count();
223                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
224                                                   sizeof(*t) + pdu_len,
225                                                   0, pc);
226                 if (!event)
227                         return;
228                 t = ring_buffer_event_data(event);
229                 goto record_it;
230         }
231 
232         /*
233          * A word about the locking here - we disable interrupts to reserve
234          * some space in the relay per-cpu buffer, to prevent an irq
235          * from coming in and stepping on our toes.
236          */
237         local_irq_save(flags);
238 
239         if (unlikely(tsk->btrace_seq != blktrace_seq))
240                 trace_note_tsk(bt, tsk);
241 
242         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
243         if (t) {
244                 sequence = per_cpu_ptr(bt->sequence, cpu);
245 
246                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
247                 t->sequence = ++(*sequence);
248                 t->time = ktime_to_ns(ktime_get());
249 record_it:
250                 /*
251                  * These two are not needed in ftrace as they are in the
252                  * generic trace_entry, filled by tracing_generic_entry_update,
253                  * but for the trace_event->bin() synthesizer benefit we do it
254                  * here too.
255                  */
256                 t->cpu = cpu;
257                 t->pid = pid;
258 
259                 t->sector = sector;
260                 t->bytes = bytes;
261                 t->action = what;
262                 t->device = bt->dev;
263                 t->error = error;
264                 t->pdu_len = pdu_len;
265 
266                 if (pdu_len)
267                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
268 
269                 if (blk_tracer) {
270                         trace_buffer_unlock_commit(buffer, event, 0, pc);
271                         return;
272                 }
273         }
274 
275         local_irq_restore(flags);
276 }
277 
278 static struct dentry *blk_tree_root;
279 static DEFINE_MUTEX(blk_tree_mutex);
280 
281 static void blk_trace_free(struct blk_trace *bt)
282 {
283         debugfs_remove(bt->msg_file);
284         debugfs_remove(bt->dropped_file);
285         relay_close(bt->rchan);
286         debugfs_remove(bt->dir);
287         free_percpu(bt->sequence);
288         free_percpu(bt->msg_data);
289         kfree(bt);
290 }
291 
292 static void blk_trace_cleanup(struct blk_trace *bt)
293 {
294         blk_trace_free(bt);
295         if (atomic_dec_and_test(&blk_probes_ref))
296                 blk_unregister_tracepoints();
297 }
298 
299 int blk_trace_remove(struct request_queue *q)
300 {
301         struct blk_trace *bt;
302 
303         bt = xchg(&q->blk_trace, NULL);
304         if (!bt)
305                 return -EINVAL;
306 
307         if (bt->trace_state != Blktrace_running)
308                 blk_trace_cleanup(bt);
309 
310         return 0;
311 }
312 EXPORT_SYMBOL_GPL(blk_trace_remove);
313 
314 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
315                                 size_t count, loff_t *ppos)
316 {
317         struct blk_trace *bt = filp->private_data;
318         char buf[16];
319 
320         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
321 
322         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
323 }
324 
325 static const struct file_operations blk_dropped_fops = {
326         .owner =        THIS_MODULE,
327         .open =         simple_open,
328         .read =         blk_dropped_read,
329         .llseek =       default_llseek,
330 };
331 
332 static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
333                                 size_t count, loff_t *ppos)
334 {
335         char *msg;
336         struct blk_trace *bt;
337 
338         if (count >= BLK_TN_MAX_MSG)
339                 return -EINVAL;
340 
341         msg = kmalloc(count + 1, GFP_KERNEL);
342         if (msg == NULL)
343                 return -ENOMEM;
344 
345         if (copy_from_user(msg, buffer, count)) {
346                 kfree(msg);
347                 return -EFAULT;
348         }
349 
350         msg[count] = '\0';
351         bt = filp->private_data;
352         __trace_note_message(bt, "%s", msg);
353         kfree(msg);
354 
355         return count;
356 }
357 
358 static const struct file_operations blk_msg_fops = {
359         .owner =        THIS_MODULE,
360         .open =         simple_open,
361         .write =        blk_msg_write,
362         .llseek =       noop_llseek,
363 };
364 
365 /*
366  * Keep track of how many times we encountered a full subbuffer, to aid
367  * the user space app in telling how many lost events there were.
368  */
369 static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
370                                      void *prev_subbuf, size_t prev_padding)
371 {
372         struct blk_trace *bt;
373 
374         if (!relay_buf_full(buf))
375                 return 1;
376 
377         bt = buf->chan->private_data;
378         atomic_inc(&bt->dropped);
379         return 0;
380 }
381 
382 static int blk_remove_buf_file_callback(struct dentry *dentry)
383 {
384         debugfs_remove(dentry);
385 
386         return 0;
387 }
388 
389 static struct dentry *blk_create_buf_file_callback(const char *filename,
390                                                    struct dentry *parent,
391                                                    umode_t mode,
392                                                    struct rchan_buf *buf,
393                                                    int *is_global)
394 {
395         return debugfs_create_file(filename, mode, parent, buf,
396                                         &relay_file_operations);
397 }
398 
399 static struct rchan_callbacks blk_relay_callbacks = {
400         .subbuf_start           = blk_subbuf_start_callback,
401         .create_buf_file        = blk_create_buf_file_callback,
402         .remove_buf_file        = blk_remove_buf_file_callback,
403 };
404 
405 static void blk_trace_setup_lba(struct blk_trace *bt,
406                                 struct block_device *bdev)
407 {
408         struct hd_struct *part = NULL;
409 
410         if (bdev)
411                 part = bdev->bd_part;
412 
413         if (part) {
414                 bt->start_lba = part->start_sect;
415                 bt->end_lba = part->start_sect + part->nr_sects;
416         } else {
417                 bt->start_lba = 0;
418                 bt->end_lba = -1ULL;
419         }
420 }
421 
422 /*
423  * Setup everything required to start tracing
424  */
425 int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
426                        struct block_device *bdev,
427                        struct blk_user_trace_setup *buts)
428 {
429         struct blk_trace *old_bt, *bt = NULL;
430         struct dentry *dir = NULL;
431         int ret, i;
432 
433         if (!buts->buf_size || !buts->buf_nr)
434                 return -EINVAL;
435 
436         strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
437         buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
438 
439         /*
440          * some device names have larger paths - convert the slashes
441          * to underscores for this to work as expected
442          */
443         for (i = 0; i < strlen(buts->name); i++)
444                 if (buts->name[i] == '/')
445                         buts->name[i] = '_';
446 
447         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
448         if (!bt)
449                 return -ENOMEM;
450 
451         ret = -ENOMEM;
452         bt->sequence = alloc_percpu(unsigned long);
453         if (!bt->sequence)
454                 goto err;
455 
456         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
457         if (!bt->msg_data)
458                 goto err;
459 
460         ret = -ENOENT;
461 
462         mutex_lock(&blk_tree_mutex);
463         if (!blk_tree_root) {
464                 blk_tree_root = debugfs_create_dir("block", NULL);
465                 if (!blk_tree_root) {
466                         mutex_unlock(&blk_tree_mutex);
467                         goto err;
468                 }
469         }
470         mutex_unlock(&blk_tree_mutex);
471 
472         dir = debugfs_create_dir(buts->name, blk_tree_root);
473 
474         if (!dir)
475                 goto err;
476 
477         bt->dir = dir;
478         bt->dev = dev;
479         atomic_set(&bt->dropped, 0);
480 
481         ret = -EIO;
482         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
483                                                &blk_dropped_fops);
484         if (!bt->dropped_file)
485                 goto err;
486 
487         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
488         if (!bt->msg_file)
489                 goto err;
490 
491         bt->rchan = relay_open("trace", dir, buts->buf_size,
492                                 buts->buf_nr, &blk_relay_callbacks, bt);
493         if (!bt->rchan)
494                 goto err;
495 
496         bt->act_mask = buts->act_mask;
497         if (!bt->act_mask)
498                 bt->act_mask = (u16) -1;
499 
500         blk_trace_setup_lba(bt, bdev);
501 
502         /* overwrite with user settings */
503         if (buts->start_lba)
504                 bt->start_lba = buts->start_lba;
505         if (buts->end_lba)
506                 bt->end_lba = buts->end_lba;
507 
508         bt->pid = buts->pid;
509         bt->trace_state = Blktrace_setup;
510 
511         ret = -EBUSY;
512         old_bt = xchg(&q->blk_trace, bt);
513         if (old_bt) {
514                 (void) xchg(&q->blk_trace, old_bt);
515                 goto err;
516         }
517 
518         if (atomic_inc_return(&blk_probes_ref) == 1)
519                 blk_register_tracepoints();
520 
521         return 0;
522 err:
523         blk_trace_free(bt);
524         return ret;
525 }
526 
527 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
528                     struct block_device *bdev,
529                     char __user *arg)
530 {
531         struct blk_user_trace_setup buts;
532         int ret;
533 
534         ret = copy_from_user(&buts, arg, sizeof(buts));
535         if (ret)
536                 return -EFAULT;
537 
538         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
539         if (ret)
540                 return ret;
541 
542         if (copy_to_user(arg, &buts, sizeof(buts))) {
543                 blk_trace_remove(q);
544                 return -EFAULT;
545         }
546         return 0;
547 }
548 EXPORT_SYMBOL_GPL(blk_trace_setup);
549 
550 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
551 static int compat_blk_trace_setup(struct request_queue *q, char *name,
552                                   dev_t dev, struct block_device *bdev,
553                                   char __user *arg)
554 {
555         struct blk_user_trace_setup buts;
556         struct compat_blk_user_trace_setup cbuts;
557         int ret;
558 
559         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
560                 return -EFAULT;
561 
562         buts = (struct blk_user_trace_setup) {
563                 .act_mask = cbuts.act_mask,
564                 .buf_size = cbuts.buf_size,
565                 .buf_nr = cbuts.buf_nr,
566                 .start_lba = cbuts.start_lba,
567                 .end_lba = cbuts.end_lba,
568                 .pid = cbuts.pid,
569         };
570         memcpy(&buts.name, &cbuts.name, 32);
571 
572         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
573         if (ret)
574                 return ret;
575 
576         if (copy_to_user(arg, &buts.name, 32)) {
577                 blk_trace_remove(q);
578                 return -EFAULT;
579         }
580 
581         return 0;
582 }
583 #endif
584 
585 int blk_trace_startstop(struct request_queue *q, int start)
586 {
587         int ret;
588         struct blk_trace *bt = q->blk_trace;
589 
590         if (bt == NULL)
591                 return -EINVAL;
592 
593         /*
594          * For starting a trace, we can transition from a setup or stopped
595          * trace. For stopping a trace, the state must be running
596          */
597         ret = -EINVAL;
598         if (start) {
599                 if (bt->trace_state == Blktrace_setup ||
600                     bt->trace_state == Blktrace_stopped) {
601                         blktrace_seq++;
602                         smp_mb();
603                         bt->trace_state = Blktrace_running;
604 
605                         trace_note_time(bt);
606                         ret = 0;
607                 }
608         } else {
609                 if (bt->trace_state == Blktrace_running) {
610                         bt->trace_state = Blktrace_stopped;
611                         relay_flush(bt->rchan);
612                         ret = 0;
613                 }
614         }
615 
616         return ret;
617 }
618 EXPORT_SYMBOL_GPL(blk_trace_startstop);
619 
620 /**
621  * blk_trace_ioctl: - handle the ioctls associated with tracing
622  * @bdev:       the block device
623  * @cmd:        the ioctl cmd
624  * @arg:        the argument data, if any
625  *
626  **/
627 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
628 {
629         struct request_queue *q;
630         int ret, start = 0;
631         char b[BDEVNAME_SIZE];
632 
633         q = bdev_get_queue(bdev);
634         if (!q)
635                 return -ENXIO;
636 
637         mutex_lock(&bdev->bd_mutex);
638 
639         switch (cmd) {
640         case BLKTRACESETUP:
641                 bdevname(bdev, b);
642                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
643                 break;
644 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
645         case BLKTRACESETUP32:
646                 bdevname(bdev, b);
647                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
648                 break;
649 #endif
650         case BLKTRACESTART:
651                 start = 1;
652         case BLKTRACESTOP:
653                 ret = blk_trace_startstop(q, start);
654                 break;
655         case BLKTRACETEARDOWN:
656                 ret = blk_trace_remove(q);
657                 break;
658         default:
659                 ret = -ENOTTY;
660                 break;
661         }
662 
663         mutex_unlock(&bdev->bd_mutex);
664         return ret;
665 }
666 
667 /**
668  * blk_trace_shutdown: - stop and cleanup trace structures
669  * @q:    the request queue associated with the device
670  *
671  **/
672 void blk_trace_shutdown(struct request_queue *q)
673 {
674         if (q->blk_trace) {
675                 blk_trace_startstop(q, 0);
676                 blk_trace_remove(q);
677         }
678 }
679 
680 /*
681  * blktrace probes
682  */
683 
684 /**
685  * blk_add_trace_rq - Add a trace for a request oriented action
686  * @q:          queue the io is for
687  * @rq:         the source request
688  * @nr_bytes:   number of completed bytes
689  * @what:       the action
690  *
691  * Description:
692  *     Records an action against a request. Will log the bio offset + size.
693  *
694  **/
695 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
696                              unsigned int nr_bytes, u32 what)
697 {
698         struct blk_trace *bt = q->blk_trace;
699 
700         if (likely(!bt))
701                 return;
702 
703         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
704                 what |= BLK_TC_ACT(BLK_TC_PC);
705                 __blk_add_trace(bt, 0, nr_bytes, rq->cmd_flags,
706                                 what, rq->errors, rq->cmd_len, rq->cmd);
707         } else  {
708                 what |= BLK_TC_ACT(BLK_TC_FS);
709                 __blk_add_trace(bt, blk_rq_pos(rq), nr_bytes,
710                                 rq->cmd_flags, what, rq->errors, 0, NULL);
711         }
712 }
713 
714 static void blk_add_trace_rq_abort(void *ignore,
715                                    struct request_queue *q, struct request *rq)
716 {
717         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ABORT);
718 }
719 
720 static void blk_add_trace_rq_insert(void *ignore,
721                                     struct request_queue *q, struct request *rq)
722 {
723         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_INSERT);
724 }
725 
726 static void blk_add_trace_rq_issue(void *ignore,
727                                    struct request_queue *q, struct request *rq)
728 {
729         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_ISSUE);
730 }
731 
732 static void blk_add_trace_rq_requeue(void *ignore,
733                                      struct request_queue *q,
734                                      struct request *rq)
735 {
736         blk_add_trace_rq(q, rq, blk_rq_bytes(rq), BLK_TA_REQUEUE);
737 }
738 
739 static void blk_add_trace_rq_complete(void *ignore,
740                                       struct request_queue *q,
741                                       struct request *rq,
742                                       unsigned int nr_bytes)
743 {
744         blk_add_trace_rq(q, rq, nr_bytes, BLK_TA_COMPLETE);
745 }
746 
747 /**
748  * blk_add_trace_bio - Add a trace for a bio oriented action
749  * @q:          queue the io is for
750  * @bio:        the source bio
751  * @what:       the action
752  * @error:      error, if any
753  *
754  * Description:
755  *     Records an action against a bio. Will log the bio offset + size.
756  *
757  **/
758 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
759                               u32 what, int error)
760 {
761         struct blk_trace *bt = q->blk_trace;
762 
763         if (likely(!bt))
764                 return;
765 
766         if (!error && !bio_flagged(bio, BIO_UPTODATE))
767                 error = EIO;
768 
769         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
770                         error, 0, NULL);
771 }
772 
773 static void blk_add_trace_bio_bounce(void *ignore,
774                                      struct request_queue *q, struct bio *bio)
775 {
776         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
777 }
778 
779 static void blk_add_trace_bio_complete(void *ignore,
780                                        struct request_queue *q, struct bio *bio,
781                                        int error)
782 {
783         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
784 }
785 
786 static void blk_add_trace_bio_backmerge(void *ignore,
787                                         struct request_queue *q,
788                                         struct request *rq,
789                                         struct bio *bio)
790 {
791         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
792 }
793 
794 static void blk_add_trace_bio_frontmerge(void *ignore,
795                                          struct request_queue *q,
796                                          struct request *rq,
797                                          struct bio *bio)
798 {
799         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
800 }
801 
802 static void blk_add_trace_bio_queue(void *ignore,
803                                     struct request_queue *q, struct bio *bio)
804 {
805         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
806 }
807 
808 static void blk_add_trace_getrq(void *ignore,
809                                 struct request_queue *q,
810                                 struct bio *bio, int rw)
811 {
812         if (bio)
813                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
814         else {
815                 struct blk_trace *bt = q->blk_trace;
816 
817                 if (bt)
818                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
819         }
820 }
821 
822 
823 static void blk_add_trace_sleeprq(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_SLEEPRQ, 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_SLEEPRQ,
834                                         0, 0, NULL);
835         }
836 }
837 
838 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
839 {
840         struct blk_trace *bt = q->blk_trace;
841 
842         if (bt)
843                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
844 }
845 
846 static void blk_add_trace_unplug(void *ignore, struct request_queue *q,
847                                     unsigned int depth, bool explicit)
848 {
849         struct blk_trace *bt = q->blk_trace;
850 
851         if (bt) {
852                 __be64 rpdu = cpu_to_be64(depth);
853                 u32 what;
854 
855                 if (explicit)
856                         what = BLK_TA_UNPLUG_IO;
857                 else
858                         what = BLK_TA_UNPLUG_TIMER;
859 
860                 __blk_add_trace(bt, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu);
861         }
862 }
863 
864 static void blk_add_trace_split(void *ignore,
865                                 struct request_queue *q, struct bio *bio,
866                                 unsigned int pdu)
867 {
868         struct blk_trace *bt = q->blk_trace;
869 
870         if (bt) {
871                 __be64 rpdu = cpu_to_be64(pdu);
872 
873                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
874                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
875                                 sizeof(rpdu), &rpdu);
876         }
877 }
878 
879 /**
880  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
881  * @ignore:     trace callback data parameter (not used)
882  * @q:          queue the io is for
883  * @bio:        the source bio
884  * @dev:        target device
885  * @from:       source sector
886  *
887  * Description:
888  *     Device mapper or raid target sometimes need to split a bio because
889  *     it spans a stripe (or similar). Add a trace for that action.
890  *
891  **/
892 static void blk_add_trace_bio_remap(void *ignore,
893                                     struct request_queue *q, struct bio *bio,
894                                     dev_t dev, sector_t from)
895 {
896         struct blk_trace *bt = q->blk_trace;
897         struct blk_io_trace_remap r;
898 
899         if (likely(!bt))
900                 return;
901 
902         r.device_from = cpu_to_be32(dev);
903         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
904         r.sector_from = cpu_to_be64(from);
905 
906         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
907                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
908                         sizeof(r), &r);
909 }
910 
911 /**
912  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
913  * @ignore:     trace callback data parameter (not used)
914  * @q:          queue the io is for
915  * @rq:         the source request
916  * @dev:        target device
917  * @from:       source sector
918  *
919  * Description:
920  *     Device mapper remaps request to other devices.
921  *     Add a trace for that action.
922  *
923  **/
924 static void blk_add_trace_rq_remap(void *ignore,
925                                    struct request_queue *q,
926                                    struct request *rq, dev_t dev,
927                                    sector_t from)
928 {
929         struct blk_trace *bt = q->blk_trace;
930         struct blk_io_trace_remap r;
931 
932         if (likely(!bt))
933                 return;
934 
935         r.device_from = cpu_to_be32(dev);
936         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
937         r.sector_from = cpu_to_be64(from);
938 
939         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
940                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
941                         sizeof(r), &r);
942 }
943 
944 /**
945  * blk_add_driver_data - Add binary message with driver-specific data
946  * @q:          queue the io is for
947  * @rq:         io request
948  * @data:       driver-specific data
949  * @len:        length of driver-specific data
950  *
951  * Description:
952  *     Some drivers might want to write driver-specific data per request.
953  *
954  **/
955 void blk_add_driver_data(struct request_queue *q,
956                          struct request *rq,
957                          void *data, size_t len)
958 {
959         struct blk_trace *bt = q->blk_trace;
960 
961         if (likely(!bt))
962                 return;
963 
964         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
965                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
966                                 BLK_TA_DRV_DATA, rq->errors, len, data);
967         else
968                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
969                                 BLK_TA_DRV_DATA, rq->errors, len, data);
970 }
971 EXPORT_SYMBOL_GPL(blk_add_driver_data);
972 
973 static void blk_register_tracepoints(void)
974 {
975         int ret;
976 
977         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
978         WARN_ON(ret);
979         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
980         WARN_ON(ret);
981         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
982         WARN_ON(ret);
983         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
984         WARN_ON(ret);
985         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
986         WARN_ON(ret);
987         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
988         WARN_ON(ret);
989         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
990         WARN_ON(ret);
991         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
992         WARN_ON(ret);
993         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
994         WARN_ON(ret);
995         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
996         WARN_ON(ret);
997         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
998         WARN_ON(ret);
999         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1000         WARN_ON(ret);
1001         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1002         WARN_ON(ret);
1003         ret = register_trace_block_unplug(blk_add_trace_unplug, NULL);
1004         WARN_ON(ret);
1005         ret = register_trace_block_split(blk_add_trace_split, NULL);
1006         WARN_ON(ret);
1007         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1008         WARN_ON(ret);
1009         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1010         WARN_ON(ret);
1011 }
1012 
1013 static void blk_unregister_tracepoints(void)
1014 {
1015         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1016         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1017         unregister_trace_block_split(blk_add_trace_split, NULL);
1018         unregister_trace_block_unplug(blk_add_trace_unplug, NULL);
1019         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1020         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1021         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1022         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1023         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1024         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1025         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1026         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1027         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1028         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1029         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1030         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1031         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1032 
1033         tracepoint_synchronize_unregister();
1034 }
1035 
1036 /*
1037  * struct blk_io_tracer formatting routines
1038  */
1039 
1040 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1041 {
1042         int i = 0;
1043         int tc = t->action >> BLK_TC_SHIFT;
1044 
1045         if (t->action == BLK_TN_MESSAGE) {
1046                 rwbs[i++] = 'N';
1047                 goto out;
1048         }
1049 
1050         if (tc & BLK_TC_FLUSH)
1051                 rwbs[i++] = 'F';
1052 
1053         if (tc & BLK_TC_DISCARD)
1054                 rwbs[i++] = 'D';
1055         else if (tc & BLK_TC_WRITE)
1056                 rwbs[i++] = 'W';
1057         else if (t->bytes)
1058                 rwbs[i++] = 'R';
1059         else
1060                 rwbs[i++] = 'N';
1061 
1062         if (tc & BLK_TC_FUA)
1063                 rwbs[i++] = 'F';
1064         if (tc & BLK_TC_AHEAD)
1065                 rwbs[i++] = 'A';
1066         if (tc & BLK_TC_SYNC)
1067                 rwbs[i++] = 'S';
1068         if (tc & BLK_TC_META)
1069                 rwbs[i++] = 'M';
1070 out:
1071         rwbs[i] = '\0';
1072 }
1073 
1074 static inline
1075 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1076 {
1077         return (const struct blk_io_trace *)ent;
1078 }
1079 
1080 static inline const void *pdu_start(const struct trace_entry *ent)
1081 {
1082         return te_blk_io_trace(ent) + 1;
1083 }
1084 
1085 static inline u32 t_action(const struct trace_entry *ent)
1086 {
1087         return te_blk_io_trace(ent)->action;
1088 }
1089 
1090 static inline u32 t_bytes(const struct trace_entry *ent)
1091 {
1092         return te_blk_io_trace(ent)->bytes;
1093 }
1094 
1095 static inline u32 t_sec(const struct trace_entry *ent)
1096 {
1097         return te_blk_io_trace(ent)->bytes >> 9;
1098 }
1099 
1100 static inline unsigned long long t_sector(const struct trace_entry *ent)
1101 {
1102         return te_blk_io_trace(ent)->sector;
1103 }
1104 
1105 static inline __u16 t_error(const struct trace_entry *ent)
1106 {
1107         return te_blk_io_trace(ent)->error;
1108 }
1109 
1110 static __u64 get_pdu_int(const struct trace_entry *ent)
1111 {
1112         const __u64 *val = pdu_start(ent);
1113         return be64_to_cpu(*val);
1114 }
1115 
1116 static void get_pdu_remap(const struct trace_entry *ent,
1117                           struct blk_io_trace_remap *r)
1118 {
1119         const struct blk_io_trace_remap *__r = pdu_start(ent);
1120         __u64 sector_from = __r->sector_from;
1121 
1122         r->device_from = be32_to_cpu(__r->device_from);
1123         r->device_to   = be32_to_cpu(__r->device_to);
1124         r->sector_from = be64_to_cpu(sector_from);
1125 }
1126 
1127 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1128 
1129 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1130 {
1131         char rwbs[RWBS_LEN];
1132         unsigned long long ts  = iter->ts;
1133         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1134         unsigned secs          = (unsigned long)ts;
1135         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1136 
1137         fill_rwbs(rwbs, t);
1138 
1139         return trace_seq_printf(&iter->seq,
1140                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1141                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1142                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1143 }
1144 
1145 static int blk_log_action(struct trace_iterator *iter, const char *act)
1146 {
1147         char rwbs[RWBS_LEN];
1148         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1149 
1150         fill_rwbs(rwbs, t);
1151         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1152                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1153 }
1154 
1155 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1156 {
1157         const unsigned char *pdu_buf;
1158         int pdu_len;
1159         int i, end, ret;
1160 
1161         pdu_buf = pdu_start(ent);
1162         pdu_len = te_blk_io_trace(ent)->pdu_len;
1163 
1164         if (!pdu_len)
1165                 return 1;
1166 
1167         /* find the last zero that needs to be printed */
1168         for (end = pdu_len - 1; end >= 0; end--)
1169                 if (pdu_buf[end])
1170                         break;
1171         end++;
1172 
1173         if (!trace_seq_putc(s, '('))
1174                 return 0;
1175 
1176         for (i = 0; i < pdu_len; i++) {
1177 
1178                 ret = trace_seq_printf(s, "%s%02x",
1179                                        i == 0 ? "" : " ", pdu_buf[i]);
1180                 if (!ret)
1181                         return ret;
1182 
1183                 /*
1184                  * stop when the rest is just zeroes and indicate so
1185                  * with a ".." appended
1186                  */
1187                 if (i == end && end != pdu_len - 1)
1188                         return trace_seq_puts(s, " ..) ");
1189         }
1190 
1191         return trace_seq_puts(s, ") ");
1192 }
1193 
1194 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1195 {
1196         char cmd[TASK_COMM_LEN];
1197 
1198         trace_find_cmdline(ent->pid, cmd);
1199 
1200         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1201                 int ret;
1202 
1203                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1204                 if (!ret)
1205                         return 0;
1206                 ret = blk_log_dump_pdu(s, ent);
1207                 if (!ret)
1208                         return 0;
1209                 return trace_seq_printf(s, "[%s]\n", cmd);
1210         } else {
1211                 if (t_sec(ent))
1212                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1213                                                 t_sector(ent), t_sec(ent), cmd);
1214                 return trace_seq_printf(s, "[%s]\n", cmd);
1215         }
1216 }
1217 
1218 static int blk_log_with_error(struct trace_seq *s,
1219                               const struct trace_entry *ent)
1220 {
1221         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1222                 int ret;
1223 
1224                 ret = blk_log_dump_pdu(s, ent);
1225                 if (ret)
1226                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1227                 return 0;
1228         } else {
1229                 if (t_sec(ent))
1230                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1231                                                 t_sector(ent),
1232                                                 t_sec(ent), t_error(ent));
1233                 return trace_seq_printf(s, "%llu [%d]\n",
1234                                         t_sector(ent), t_error(ent));
1235         }
1236 }
1237 
1238 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1239 {
1240         struct blk_io_trace_remap r = { .device_from = 0, };
1241 
1242         get_pdu_remap(ent, &r);
1243         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1244                                 t_sector(ent), t_sec(ent),
1245                                 MAJOR(r.device_from), MINOR(r.device_from),
1246                                 (unsigned long long)r.sector_from);
1247 }
1248 
1249 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1250 {
1251         char cmd[TASK_COMM_LEN];
1252 
1253         trace_find_cmdline(ent->pid, cmd);
1254 
1255         return trace_seq_printf(s, "[%s]\n", cmd);
1256 }
1257 
1258 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1259 {
1260         char cmd[TASK_COMM_LEN];
1261 
1262         trace_find_cmdline(ent->pid, cmd);
1263 
1264         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1265 }
1266 
1267 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1268 {
1269         char cmd[TASK_COMM_LEN];
1270 
1271         trace_find_cmdline(ent->pid, cmd);
1272 
1273         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1274                                 get_pdu_int(ent), cmd);
1275 }
1276 
1277 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1278 {
1279         int ret;
1280         const struct blk_io_trace *t = te_blk_io_trace(ent);
1281 
1282         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1283         if (ret)
1284                 return trace_seq_putc(s, '\n');
1285         return ret;
1286 }
1287 
1288 /*
1289  * struct tracer operations
1290  */
1291 
1292 static void blk_tracer_print_header(struct seq_file *m)
1293 {
1294         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1295                 return;
1296         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1297                     "#  |     |     |           |   |   |\n");
1298 }
1299 
1300 static void blk_tracer_start(struct trace_array *tr)
1301 {
1302         blk_tracer_enabled = true;
1303 }
1304 
1305 static int blk_tracer_init(struct trace_array *tr)
1306 {
1307         blk_tr = tr;
1308         blk_tracer_start(tr);
1309         return 0;
1310 }
1311 
1312 static void blk_tracer_stop(struct trace_array *tr)
1313 {
1314         blk_tracer_enabled = false;
1315 }
1316 
1317 static void blk_tracer_reset(struct trace_array *tr)
1318 {
1319         blk_tracer_stop(tr);
1320 }
1321 
1322 static const struct {
1323         const char *act[2];
1324         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1325 } what2act[] = {
1326         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1327         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1328         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1329         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1330         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1331         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1332         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1333         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1334         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1335         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1336         [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1337         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1338         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1339         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1340         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1341 };
1342 
1343 static enum print_line_t print_one_line(struct trace_iterator *iter,
1344                                         bool classic)
1345 {
1346         struct trace_seq *s = &iter->seq;
1347         const struct blk_io_trace *t;
1348         u16 what;
1349         int ret;
1350         bool long_act;
1351         blk_log_action_t *log_action;
1352 
1353         t          = te_blk_io_trace(iter->ent);
1354         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1355         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1356         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1357 
1358         if (t->action == BLK_TN_MESSAGE) {
1359                 ret = log_action(iter, long_act ? "message" : "m");
1360                 if (ret)
1361                         ret = blk_log_msg(s, iter->ent);
1362                 goto out;
1363         }
1364 
1365         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1366                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1367         else {
1368                 ret = log_action(iter, what2act[what].act[long_act]);
1369                 if (ret)
1370                         ret = what2act[what].print(s, iter->ent);
1371         }
1372 out:
1373         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
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 int 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         if (!trace_seq_putmem(s, &old, offset))
1393                 return 0;
1394         return trace_seq_putmem(s, &t->sector,
1395                                 sizeof(old) - offset + t->pdu_len);
1396 }
1397 
1398 static enum print_line_t
1399 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1400                              struct trace_event *event)
1401 {
1402         return blk_trace_synthesize_old_trace(iter) ?
1403                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
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 blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1415 {
1416         /* don't output context-info for blk_classic output */
1417         if (bit == TRACE_BLK_OPT_CLASSIC) {
1418                 if (set)
1419                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1420                 else
1421                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1422         }
1423         return 0;
1424 }
1425 
1426 static struct tracer blk_tracer __read_mostly = {
1427         .name           = "blk",
1428         .init           = blk_tracer_init,
1429         .reset          = blk_tracer_reset,
1430         .start          = blk_tracer_start,
1431         .stop           = blk_tracer_stop,
1432         .print_header   = blk_tracer_print_header,
1433         .print_line     = blk_tracer_print_line,
1434         .flags          = &blk_tracer_flags,
1435         .set_flag       = blk_tracer_set_flag,
1436 };
1437 
1438 static struct trace_event_functions trace_blk_event_funcs = {
1439         .trace          = blk_trace_event_print,
1440         .binary         = blk_trace_event_print_binary,
1441 };
1442 
1443 static struct trace_event trace_blk_event = {
1444         .type           = TRACE_BLK,
1445         .funcs          = &trace_blk_event_funcs,
1446 };
1447 
1448 static int __init init_blk_tracer(void)
1449 {
1450         if (!register_ftrace_event(&trace_blk_event)) {
1451                 pr_warning("Warning: could not register block events\n");
1452                 return 1;
1453         }
1454 
1455         if (register_tracer(&blk_tracer) != 0) {
1456                 pr_warning("Warning: could not register the block tracer\n");
1457                 unregister_ftrace_event(&trace_blk_event);
1458                 return 1;
1459         }
1460 
1461         return 0;
1462 }
1463 
1464 device_initcall(init_blk_tracer);
1465 
1466 static int blk_trace_remove_queue(struct request_queue *q)
1467 {
1468         struct blk_trace *bt;
1469 
1470         bt = xchg(&q->blk_trace, NULL);
1471         if (bt == NULL)
1472                 return -EINVAL;
1473 
1474         if (atomic_dec_and_test(&blk_probes_ref))
1475                 blk_unregister_tracepoints();
1476 
1477         blk_trace_free(bt);
1478         return 0;
1479 }
1480 
1481 /*
1482  * Setup everything required to start tracing
1483  */
1484 static int blk_trace_setup_queue(struct request_queue *q,
1485                                  struct block_device *bdev)
1486 {
1487         struct blk_trace *old_bt, *bt = NULL;
1488         int ret = -ENOMEM;
1489 
1490         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1491         if (!bt)
1492                 return -ENOMEM;
1493 
1494         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1495         if (!bt->msg_data)
1496                 goto free_bt;
1497 
1498         bt->dev = bdev->bd_dev;
1499         bt->act_mask = (u16)-1;
1500 
1501         blk_trace_setup_lba(bt, bdev);
1502 
1503         old_bt = xchg(&q->blk_trace, bt);
1504         if (old_bt != NULL) {
1505                 (void)xchg(&q->blk_trace, old_bt);
1506                 ret = -EBUSY;
1507                 goto free_bt;
1508         }
1509 
1510         if (atomic_inc_return(&blk_probes_ref) == 1)
1511                 blk_register_tracepoints();
1512         return 0;
1513 
1514 free_bt:
1515         blk_trace_free(bt);
1516         return ret;
1517 }
1518 
1519 /*
1520  * sysfs interface to enable and configure tracing
1521  */
1522 
1523 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1524                                          struct device_attribute *attr,
1525                                          char *buf);
1526 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1527                                           struct device_attribute *attr,
1528                                           const char *buf, size_t count);
1529 #define BLK_TRACE_DEVICE_ATTR(_name) \
1530         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1531                     sysfs_blk_trace_attr_show, \
1532                     sysfs_blk_trace_attr_store)
1533 
1534 static BLK_TRACE_DEVICE_ATTR(enable);
1535 static BLK_TRACE_DEVICE_ATTR(act_mask);
1536 static BLK_TRACE_DEVICE_ATTR(pid);
1537 static BLK_TRACE_DEVICE_ATTR(start_lba);
1538 static BLK_TRACE_DEVICE_ATTR(end_lba);
1539 
1540 static struct attribute *blk_trace_attrs[] = {
1541         &dev_attr_enable.attr,
1542         &dev_attr_act_mask.attr,
1543         &dev_attr_pid.attr,
1544         &dev_attr_start_lba.attr,
1545         &dev_attr_end_lba.attr,
1546         NULL
1547 };
1548 
1549 struct attribute_group blk_trace_attr_group = {
1550         .name  = "trace",
1551         .attrs = blk_trace_attrs,
1552 };
1553 
1554 static const struct {
1555         int mask;
1556         const char *str;
1557 } mask_maps[] = {
1558         { BLK_TC_READ,          "read"          },
1559         { BLK_TC_WRITE,         "write"         },
1560         { BLK_TC_FLUSH,         "flush"         },
1561         { BLK_TC_SYNC,          "sync"          },
1562         { BLK_TC_QUEUE,         "queue"         },
1563         { BLK_TC_REQUEUE,       "requeue"       },
1564         { BLK_TC_ISSUE,         "issue"         },
1565         { BLK_TC_COMPLETE,      "complete"      },
1566         { BLK_TC_FS,            "fs"            },
1567         { BLK_TC_PC,            "pc"            },
1568         { BLK_TC_AHEAD,         "ahead"         },
1569         { BLK_TC_META,          "meta"          },
1570         { BLK_TC_DISCARD,       "discard"       },
1571         { BLK_TC_DRV_DATA,      "drv_data"      },
1572         { BLK_TC_FUA,           "fua"           },
1573 };
1574 
1575 static int blk_trace_str2mask(const char *str)
1576 {
1577         int i;
1578         int mask = 0;
1579         char *buf, *s, *token;
1580 
1581         buf = kstrdup(str, GFP_KERNEL);
1582         if (buf == NULL)
1583                 return -ENOMEM;
1584         s = strstrip(buf);
1585 
1586         while (1) {
1587                 token = strsep(&s, ",");
1588                 if (token == NULL)
1589                         break;
1590 
1591                 if (*token == '\0')
1592                         continue;
1593 
1594                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1595                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1596                                 mask |= mask_maps[i].mask;
1597                                 break;
1598                         }
1599                 }
1600                 if (i == ARRAY_SIZE(mask_maps)) {
1601                         mask = -EINVAL;
1602                         break;
1603                 }
1604         }
1605         kfree(buf);
1606 
1607         return mask;
1608 }
1609 
1610 static ssize_t blk_trace_mask2str(char *buf, int mask)
1611 {
1612         int i;
1613         char *p = buf;
1614 
1615         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1616                 if (mask & mask_maps[i].mask) {
1617                         p += sprintf(p, "%s%s",
1618                                     (p == buf) ? "" : ",", mask_maps[i].str);
1619                 }
1620         }
1621         *p++ = '\n';
1622 
1623         return p - buf;
1624 }
1625 
1626 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1627 {
1628         if (bdev->bd_disk == NULL)
1629                 return NULL;
1630 
1631         return bdev_get_queue(bdev);
1632 }
1633 
1634 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1635                                          struct device_attribute *attr,
1636                                          char *buf)
1637 {
1638         struct hd_struct *p = dev_to_part(dev);
1639         struct request_queue *q;
1640         struct block_device *bdev;
1641         ssize_t ret = -ENXIO;
1642 
1643         bdev = bdget(part_devt(p));
1644         if (bdev == NULL)
1645                 goto out;
1646 
1647         q = blk_trace_get_queue(bdev);
1648         if (q == NULL)
1649                 goto out_bdput;
1650 
1651         mutex_lock(&bdev->bd_mutex);
1652 
1653         if (attr == &dev_attr_enable) {
1654                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1655                 goto out_unlock_bdev;
1656         }
1657 
1658         if (q->blk_trace == NULL)
1659                 ret = sprintf(buf, "disabled\n");
1660         else if (attr == &dev_attr_act_mask)
1661                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1662         else if (attr == &dev_attr_pid)
1663                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1664         else if (attr == &dev_attr_start_lba)
1665                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1666         else if (attr == &dev_attr_end_lba)
1667                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1668 
1669 out_unlock_bdev:
1670         mutex_unlock(&bdev->bd_mutex);
1671 out_bdput:
1672         bdput(bdev);
1673 out:
1674         return ret;
1675 }
1676 
1677 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1678                                           struct device_attribute *attr,
1679                                           const char *buf, size_t count)
1680 {
1681         struct block_device *bdev;
1682         struct request_queue *q;
1683         struct hd_struct *p;
1684         u64 value;
1685         ssize_t ret = -EINVAL;
1686 
1687         if (count == 0)
1688                 goto out;
1689 
1690         if (attr == &dev_attr_act_mask) {
1691                 if (sscanf(buf, "%llx", &value) != 1) {
1692                         /* Assume it is a list of trace category names */
1693                         ret = blk_trace_str2mask(buf);
1694                         if (ret < 0)
1695                                 goto out;
1696                         value = ret;
1697                 }
1698         } else if (sscanf(buf, "%llu", &value) != 1)
1699                 goto out;
1700 
1701         ret = -ENXIO;
1702 
1703         p = dev_to_part(dev);
1704         bdev = bdget(part_devt(p));
1705         if (bdev == NULL)
1706                 goto out;
1707 
1708         q = blk_trace_get_queue(bdev);
1709         if (q == NULL)
1710                 goto out_bdput;
1711 
1712         mutex_lock(&bdev->bd_mutex);
1713 
1714         if (attr == &dev_attr_enable) {
1715                 if (value)
1716                         ret = blk_trace_setup_queue(q, bdev);
1717                 else
1718                         ret = blk_trace_remove_queue(q);
1719                 goto out_unlock_bdev;
1720         }
1721 
1722         ret = 0;
1723         if (q->blk_trace == NULL)
1724                 ret = blk_trace_setup_queue(q, bdev);
1725 
1726         if (ret == 0) {
1727                 if (attr == &dev_attr_act_mask)
1728                         q->blk_trace->act_mask = value;
1729                 else if (attr == &dev_attr_pid)
1730                         q->blk_trace->pid = value;
1731                 else if (attr == &dev_attr_start_lba)
1732                         q->blk_trace->start_lba = value;
1733                 else if (attr == &dev_attr_end_lba)
1734                         q->blk_trace->end_lba = value;
1735         }
1736 
1737 out_unlock_bdev:
1738         mutex_unlock(&bdev->bd_mutex);
1739 out_bdput:
1740         bdput(bdev);
1741 out:
1742         return ret ? ret : count;
1743 }
1744 
1745 int blk_trace_init_sysfs(struct device *dev)
1746 {
1747         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1748 }
1749 
1750 void blk_trace_remove_sysfs(struct device *dev)
1751 {
1752         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1753 }
1754 
1755 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1756 
1757 #ifdef CONFIG_EVENT_TRACING
1758 
1759 void blk_dump_cmd(char *buf, struct request *rq)
1760 {
1761         int i, end;
1762         int len = rq->cmd_len;
1763         unsigned char *cmd = rq->cmd;
1764 
1765         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1766                 buf[0] = '\0';
1767                 return;
1768         }
1769 
1770         for (end = len - 1; end >= 0; end--)
1771                 if (cmd[end])
1772                         break;
1773         end++;
1774 
1775         for (i = 0; i < len; i++) {
1776                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1777                 if (i == end && end != len - 1) {
1778                         sprintf(buf, " ..");
1779                         break;
1780                 }
1781         }
1782 }
1783 
1784 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1785 {
1786         int i = 0;
1787 
1788         if (rw & REQ_FLUSH)
1789                 rwbs[i++] = 'F';
1790 
1791         if (rw & WRITE)
1792                 rwbs[i++] = 'W';
1793         else if (rw & REQ_DISCARD)
1794                 rwbs[i++] = 'D';
1795         else if (bytes)
1796                 rwbs[i++] = 'R';
1797         else
1798                 rwbs[i++] = 'N';
1799 
1800         if (rw & REQ_FUA)
1801                 rwbs[i++] = 'F';
1802         if (rw & REQ_RAHEAD)
1803                 rwbs[i++] = 'A';
1804         if (rw & REQ_SYNC)
1805                 rwbs[i++] = 'S';
1806         if (rw & REQ_META)
1807                 rwbs[i++] = 'M';
1808         if (rw & REQ_SECURE)
1809                 rwbs[i++] = 'E';
1810 
1811         rwbs[i] = '\0';
1812 }
1813 EXPORT_SYMBOL_GPL(blk_fill_rwbs);
1814 
1815 #endif /* CONFIG_EVENT_TRACING */
1816 
1817 

~ [ 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