block: remove block_unplug_timer() trace point
[linux-3.10.git] / kernel / trace / blktrace.c
1 /*
2  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11  * GNU General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
16  *
17  */
18 #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/time.h>
27 #include <linux/uaccess.h>
28
29 #include <trace/events/block.h>
30
31 #include "trace_output.h"
32
33 #ifdef CONFIG_BLK_DEV_IO_TRACE
34
35 static unsigned int blktrace_seq __read_mostly = 1;
36
37 static struct trace_array *blk_tr;
38 static bool blk_tracer_enabled __read_mostly;
39
40 /* Select an alternative, minimalistic output than the original one */
41 #define TRACE_BLK_OPT_CLASSIC   0x1
42
43 static struct tracer_opt blk_tracer_opts[] = {
44         /* Default disable the minimalistic output */
45         { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
46         { }
47 };
48
49 static struct tracer_flags blk_tracer_flags = {
50         .val  = 0,
51         .opts = blk_tracer_opts,
52 };
53
54 /* Global reference count of probes */
55 static atomic_t blk_probes_ref = ATOMIC_INIT(0);
56
57 static void blk_register_tracepoints(void);
58 static void blk_unregister_tracepoints(void);
59
60 /*
61  * Send out a notify message.
62  */
63 static void trace_note(struct blk_trace *bt, pid_t pid, int action,
64                        const void *data, size_t len)
65 {
66         struct blk_io_trace *t;
67         struct ring_buffer_event *event = NULL;
68         struct ring_buffer *buffer = NULL;
69         int pc = 0;
70         int cpu = smp_processor_id();
71         bool blk_tracer = blk_tracer_enabled;
72
73         if (blk_tracer) {
74                 buffer = blk_tr->buffer;
75                 pc = preempt_count();
76                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
77                                                   sizeof(*t) + len,
78                                                   0, pc);
79                 if (!event)
80                         return;
81                 t = ring_buffer_event_data(event);
82                 goto record_it;
83         }
84
85         if (!bt->rchan)
86                 return;
87
88         t = relay_reserve(bt->rchan, sizeof(*t) + len);
89         if (t) {
90                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
91                 t->time = ktime_to_ns(ktime_get());
92 record_it:
93                 t->device = bt->dev;
94                 t->action = action;
95                 t->pid = pid;
96                 t->cpu = cpu;
97                 t->pdu_len = len;
98                 memcpy((void *) t + sizeof(*t), data, len);
99
100                 if (blk_tracer)
101                         trace_buffer_unlock_commit(buffer, event, 0, pc);
102         }
103 }
104
105 /*
106  * Send out a notify for this process, if we haven't done so since a trace
107  * started
108  */
109 static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
110 {
111         tsk->btrace_seq = blktrace_seq;
112         trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
113 }
114
115 static void trace_note_time(struct blk_trace *bt)
116 {
117         struct timespec now;
118         unsigned long flags;
119         u32 words[2];
120
121         getnstimeofday(&now);
122         words[0] = now.tv_sec;
123         words[1] = now.tv_nsec;
124
125         local_irq_save(flags);
126         trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
127         local_irq_restore(flags);
128 }
129
130 void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
131 {
132         int n;
133         va_list args;
134         unsigned long flags;
135         char *buf;
136
137         if (unlikely(bt->trace_state != Blktrace_running &&
138                      !blk_tracer_enabled))
139                 return;
140
141         /*
142          * If the BLK_TC_NOTIFY action mask isn't set, don't send any note
143          * message to the trace.
144          */
145         if (!(bt->act_mask & BLK_TC_NOTIFY))
146                 return;
147
148         local_irq_save(flags);
149         buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
150         va_start(args, fmt);
151         n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
152         va_end(args);
153
154         trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
155         local_irq_restore(flags);
156 }
157 EXPORT_SYMBOL_GPL(__trace_note_message);
158
159 static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
160                          pid_t pid)
161 {
162         if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
163                 return 1;
164         if (sector && (sector < bt->start_lba || sector > bt->end_lba))
165                 return 1;
166         if (bt->pid && pid != bt->pid)
167                 return 1;
168
169         return 0;
170 }
171
172 /*
173  * Data direction bit lookup
174  */
175 static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
176                                  BLK_TC_ACT(BLK_TC_WRITE) };
177
178 #define BLK_TC_RAHEAD           BLK_TC_AHEAD
179
180 /* The ilog2() calls fall out because they're constant */
181 #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \
182           (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name))
183
184 /*
185  * The worker for the various blk_add_trace*() types. Fills out a
186  * blk_io_trace structure and places it in a per-cpu subbuffer.
187  */
188 static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
189                      int rw, u32 what, int error, int pdu_len, void *pdu_data)
190 {
191         struct task_struct *tsk = current;
192         struct ring_buffer_event *event = NULL;
193         struct ring_buffer *buffer = NULL;
194         struct blk_io_trace *t;
195         unsigned long flags = 0;
196         unsigned long *sequence;
197         pid_t pid;
198         int cpu, pc = 0;
199         bool blk_tracer = blk_tracer_enabled;
200
201         if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
202                 return;
203
204         what |= ddir_act[rw & WRITE];
205         what |= MASK_TC_BIT(rw, SYNC);
206         what |= MASK_TC_BIT(rw, RAHEAD);
207         what |= MASK_TC_BIT(rw, META);
208         what |= MASK_TC_BIT(rw, DISCARD);
209
210         pid = tsk->pid;
211         if (act_log_check(bt, what, sector, pid))
212                 return;
213         cpu = raw_smp_processor_id();
214
215         if (blk_tracer) {
216                 tracing_record_cmdline(current);
217
218                 buffer = blk_tr->buffer;
219                 pc = preempt_count();
220                 event = trace_buffer_lock_reserve(buffer, TRACE_BLK,
221                                                   sizeof(*t) + pdu_len,
222                                                   0, pc);
223                 if (!event)
224                         return;
225                 t = ring_buffer_event_data(event);
226                 goto record_it;
227         }
228
229         /*
230          * A word about the locking here - we disable interrupts to reserve
231          * some space in the relay per-cpu buffer, to prevent an irq
232          * from coming in and stepping on our toes.
233          */
234         local_irq_save(flags);
235
236         if (unlikely(tsk->btrace_seq != blktrace_seq))
237                 trace_note_tsk(bt, tsk);
238
239         t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
240         if (t) {
241                 sequence = per_cpu_ptr(bt->sequence, cpu);
242
243                 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
244                 t->sequence = ++(*sequence);
245                 t->time = ktime_to_ns(ktime_get());
246 record_it:
247                 /*
248                  * These two are not needed in ftrace as they are in the
249                  * generic trace_entry, filled by tracing_generic_entry_update,
250                  * but for the trace_event->bin() synthesizer benefit we do it
251                  * here too.
252                  */
253                 t->cpu = cpu;
254                 t->pid = pid;
255
256                 t->sector = sector;
257                 t->bytes = bytes;
258                 t->action = what;
259                 t->device = bt->dev;
260                 t->error = error;
261                 t->pdu_len = pdu_len;
262
263                 if (pdu_len)
264                         memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
265
266                 if (blk_tracer) {
267                         trace_buffer_unlock_commit(buffer, event, 0, pc);
268                         return;
269                 }
270         }
271
272         local_irq_restore(flags);
273 }
274
275 static struct dentry *blk_tree_root;
276 static DEFINE_MUTEX(blk_tree_mutex);
277
278 static void blk_trace_free(struct blk_trace *bt)
279 {
280         debugfs_remove(bt->msg_file);
281         debugfs_remove(bt->dropped_file);
282         relay_close(bt->rchan);
283         debugfs_remove(bt->dir);
284         free_percpu(bt->sequence);
285         free_percpu(bt->msg_data);
286         kfree(bt);
287 }
288
289 static void blk_trace_cleanup(struct blk_trace *bt)
290 {
291         blk_trace_free(bt);
292         if (atomic_dec_and_test(&blk_probes_ref))
293                 blk_unregister_tracepoints();
294 }
295
296 int blk_trace_remove(struct request_queue *q)
297 {
298         struct blk_trace *bt;
299
300         bt = xchg(&q->blk_trace, NULL);
301         if (!bt)
302                 return -EINVAL;
303
304         if (bt->trace_state != Blktrace_running)
305                 blk_trace_cleanup(bt);
306
307         return 0;
308 }
309 EXPORT_SYMBOL_GPL(blk_trace_remove);
310
311 static int blk_dropped_open(struct inode *inode, struct file *filp)
312 {
313         filp->private_data = inode->i_private;
314
315         return 0;
316 }
317
318 static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
319                                 size_t count, loff_t *ppos)
320 {
321         struct blk_trace *bt = filp->private_data;
322         char buf[16];
323
324         snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
325
326         return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
327 }
328
329 static const struct file_operations blk_dropped_fops = {
330         .owner =        THIS_MODULE,
331         .open =         blk_dropped_open,
332         .read =         blk_dropped_read,
333         .llseek =       default_llseek,
334 };
335
336 static int blk_msg_open(struct inode *inode, struct file *filp)
337 {
338         filp->private_data = inode->i_private;
339
340         return 0;
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 =         blk_msg_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                                                    int 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, i;
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         for (i = 0; i < strlen(buts->name); i++)
455                 if (buts->name[i] == '/')
456                         buts->name[i] = '_';
457
458         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
459         if (!bt)
460                 return -ENOMEM;
461
462         ret = -ENOMEM;
463         bt->sequence = alloc_percpu(unsigned long);
464         if (!bt->sequence)
465                 goto err;
466
467         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
468         if (!bt->msg_data)
469                 goto err;
470
471         ret = -ENOENT;
472
473         mutex_lock(&blk_tree_mutex);
474         if (!blk_tree_root) {
475                 blk_tree_root = debugfs_create_dir("block", NULL);
476                 if (!blk_tree_root) {
477                         mutex_unlock(&blk_tree_mutex);
478                         goto err;
479                 }
480         }
481         mutex_unlock(&blk_tree_mutex);
482
483         dir = debugfs_create_dir(buts->name, blk_tree_root);
484
485         if (!dir)
486                 goto err;
487
488         bt->dir = dir;
489         bt->dev = dev;
490         atomic_set(&bt->dropped, 0);
491
492         ret = -EIO;
493         bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
494                                                &blk_dropped_fops);
495         if (!bt->dropped_file)
496                 goto err;
497
498         bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
499         if (!bt->msg_file)
500                 goto err;
501
502         bt->rchan = relay_open("trace", dir, buts->buf_size,
503                                 buts->buf_nr, &blk_relay_callbacks, bt);
504         if (!bt->rchan)
505                 goto err;
506
507         bt->act_mask = buts->act_mask;
508         if (!bt->act_mask)
509                 bt->act_mask = (u16) -1;
510
511         blk_trace_setup_lba(bt, bdev);
512
513         /* overwrite with user settings */
514         if (buts->start_lba)
515                 bt->start_lba = buts->start_lba;
516         if (buts->end_lba)
517                 bt->end_lba = buts->end_lba;
518
519         bt->pid = buts->pid;
520         bt->trace_state = Blktrace_setup;
521
522         ret = -EBUSY;
523         old_bt = xchg(&q->blk_trace, bt);
524         if (old_bt) {
525                 (void) xchg(&q->blk_trace, old_bt);
526                 goto err;
527         }
528
529         if (atomic_inc_return(&blk_probes_ref) == 1)
530                 blk_register_tracepoints();
531
532         return 0;
533 err:
534         blk_trace_free(bt);
535         return ret;
536 }
537
538 int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
539                     struct block_device *bdev,
540                     char __user *arg)
541 {
542         struct blk_user_trace_setup buts;
543         int ret;
544
545         ret = copy_from_user(&buts, arg, sizeof(buts));
546         if (ret)
547                 return -EFAULT;
548
549         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
550         if (ret)
551                 return ret;
552
553         if (copy_to_user(arg, &buts, sizeof(buts))) {
554                 blk_trace_remove(q);
555                 return -EFAULT;
556         }
557         return 0;
558 }
559 EXPORT_SYMBOL_GPL(blk_trace_setup);
560
561 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
562 static int compat_blk_trace_setup(struct request_queue *q, char *name,
563                                   dev_t dev, struct block_device *bdev,
564                                   char __user *arg)
565 {
566         struct blk_user_trace_setup buts;
567         struct compat_blk_user_trace_setup cbuts;
568         int ret;
569
570         if (copy_from_user(&cbuts, arg, sizeof(cbuts)))
571                 return -EFAULT;
572
573         buts = (struct blk_user_trace_setup) {
574                 .act_mask = cbuts.act_mask,
575                 .buf_size = cbuts.buf_size,
576                 .buf_nr = cbuts.buf_nr,
577                 .start_lba = cbuts.start_lba,
578                 .end_lba = cbuts.end_lba,
579                 .pid = cbuts.pid,
580         };
581         memcpy(&buts.name, &cbuts.name, 32);
582
583         ret = do_blk_trace_setup(q, name, dev, bdev, &buts);
584         if (ret)
585                 return ret;
586
587         if (copy_to_user(arg, &buts.name, 32)) {
588                 blk_trace_remove(q);
589                 return -EFAULT;
590         }
591
592         return 0;
593 }
594 #endif
595
596 int blk_trace_startstop(struct request_queue *q, int start)
597 {
598         int ret;
599         struct blk_trace *bt = q->blk_trace;
600
601         if (bt == NULL)
602                 return -EINVAL;
603
604         /*
605          * For starting a trace, we can transition from a setup or stopped
606          * trace. For stopping a trace, the state must be running
607          */
608         ret = -EINVAL;
609         if (start) {
610                 if (bt->trace_state == Blktrace_setup ||
611                     bt->trace_state == Blktrace_stopped) {
612                         blktrace_seq++;
613                         smp_mb();
614                         bt->trace_state = Blktrace_running;
615
616                         trace_note_time(bt);
617                         ret = 0;
618                 }
619         } else {
620                 if (bt->trace_state == Blktrace_running) {
621                         bt->trace_state = Blktrace_stopped;
622                         relay_flush(bt->rchan);
623                         ret = 0;
624                 }
625         }
626
627         return ret;
628 }
629 EXPORT_SYMBOL_GPL(blk_trace_startstop);
630
631 /**
632  * blk_trace_ioctl: - handle the ioctls associated with tracing
633  * @bdev:       the block device
634  * @cmd:        the ioctl cmd
635  * @arg:        the argument data, if any
636  *
637  **/
638 int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
639 {
640         struct request_queue *q;
641         int ret, start = 0;
642         char b[BDEVNAME_SIZE];
643
644         q = bdev_get_queue(bdev);
645         if (!q)
646                 return -ENXIO;
647
648         mutex_lock(&bdev->bd_mutex);
649
650         switch (cmd) {
651         case BLKTRACESETUP:
652                 bdevname(bdev, b);
653                 ret = blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
654                 break;
655 #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64)
656         case BLKTRACESETUP32:
657                 bdevname(bdev, b);
658                 ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg);
659                 break;
660 #endif
661         case BLKTRACESTART:
662                 start = 1;
663         case BLKTRACESTOP:
664                 ret = blk_trace_startstop(q, start);
665                 break;
666         case BLKTRACETEARDOWN:
667                 ret = blk_trace_remove(q);
668                 break;
669         default:
670                 ret = -ENOTTY;
671                 break;
672         }
673
674         mutex_unlock(&bdev->bd_mutex);
675         return ret;
676 }
677
678 /**
679  * blk_trace_shutdown: - stop and cleanup trace structures
680  * @q:    the request queue associated with the device
681  *
682  **/
683 void blk_trace_shutdown(struct request_queue *q)
684 {
685         if (q->blk_trace) {
686                 blk_trace_startstop(q, 0);
687                 blk_trace_remove(q);
688         }
689 }
690
691 /*
692  * blktrace probes
693  */
694
695 /**
696  * blk_add_trace_rq - Add a trace for a request oriented action
697  * @q:          queue the io is for
698  * @rq:         the source request
699  * @what:       the action
700  *
701  * Description:
702  *     Records an action against a request. Will log the bio offset + size.
703  *
704  **/
705 static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
706                              u32 what)
707 {
708         struct blk_trace *bt = q->blk_trace;
709
710         if (likely(!bt))
711                 return;
712
713         if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
714                 what |= BLK_TC_ACT(BLK_TC_PC);
715                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), rq->cmd_flags,
716                                 what, rq->errors, rq->cmd_len, rq->cmd);
717         } else  {
718                 what |= BLK_TC_ACT(BLK_TC_FS);
719                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
720                                 rq->cmd_flags, what, rq->errors, 0, NULL);
721         }
722 }
723
724 static void blk_add_trace_rq_abort(void *ignore,
725                                    struct request_queue *q, struct request *rq)
726 {
727         blk_add_trace_rq(q, rq, BLK_TA_ABORT);
728 }
729
730 static void blk_add_trace_rq_insert(void *ignore,
731                                     struct request_queue *q, struct request *rq)
732 {
733         blk_add_trace_rq(q, rq, BLK_TA_INSERT);
734 }
735
736 static void blk_add_trace_rq_issue(void *ignore,
737                                    struct request_queue *q, struct request *rq)
738 {
739         blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
740 }
741
742 static void blk_add_trace_rq_requeue(void *ignore,
743                                      struct request_queue *q,
744                                      struct request *rq)
745 {
746         blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
747 }
748
749 static void blk_add_trace_rq_complete(void *ignore,
750                                       struct request_queue *q,
751                                       struct request *rq)
752 {
753         blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
754 }
755
756 /**
757  * blk_add_trace_bio - Add a trace for a bio oriented action
758  * @q:          queue the io is for
759  * @bio:        the source bio
760  * @what:       the action
761  * @error:      error, if any
762  *
763  * Description:
764  *     Records an action against a bio. Will log the bio offset + size.
765  *
766  **/
767 static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
768                               u32 what, int error)
769 {
770         struct blk_trace *bt = q->blk_trace;
771
772         if (likely(!bt))
773                 return;
774
775         if (!error && !bio_flagged(bio, BIO_UPTODATE))
776                 error = EIO;
777
778         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
779                         error, 0, NULL);
780 }
781
782 static void blk_add_trace_bio_bounce(void *ignore,
783                                      struct request_queue *q, struct bio *bio)
784 {
785         blk_add_trace_bio(q, bio, BLK_TA_BOUNCE, 0);
786 }
787
788 static void blk_add_trace_bio_complete(void *ignore,
789                                        struct request_queue *q, struct bio *bio,
790                                        int error)
791 {
792         blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, error);
793 }
794
795 static void blk_add_trace_bio_backmerge(void *ignore,
796                                         struct request_queue *q,
797                                         struct bio *bio)
798 {
799         blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE, 0);
800 }
801
802 static void blk_add_trace_bio_frontmerge(void *ignore,
803                                          struct request_queue *q,
804                                          struct bio *bio)
805 {
806         blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE, 0);
807 }
808
809 static void blk_add_trace_bio_queue(void *ignore,
810                                     struct request_queue *q, struct bio *bio)
811 {
812         blk_add_trace_bio(q, bio, BLK_TA_QUEUE, 0);
813 }
814
815 static void blk_add_trace_getrq(void *ignore,
816                                 struct request_queue *q,
817                                 struct bio *bio, int rw)
818 {
819         if (bio)
820                 blk_add_trace_bio(q, bio, BLK_TA_GETRQ, 0);
821         else {
822                 struct blk_trace *bt = q->blk_trace;
823
824                 if (bt)
825                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
826         }
827 }
828
829
830 static void blk_add_trace_sleeprq(void *ignore,
831                                   struct request_queue *q,
832                                   struct bio *bio, int rw)
833 {
834         if (bio)
835                 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ, 0);
836         else {
837                 struct blk_trace *bt = q->blk_trace;
838
839                 if (bt)
840                         __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
841                                         0, 0, NULL);
842         }
843 }
844
845 static void blk_add_trace_plug(void *ignore, struct request_queue *q)
846 {
847         struct blk_trace *bt = q->blk_trace;
848
849         if (bt)
850                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
851 }
852
853 static void blk_add_trace_unplug_io(void *ignore, struct request_queue *q)
854 {
855         struct blk_trace *bt = q->blk_trace;
856
857         if (bt) {
858                 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
859                 __be64 rpdu = cpu_to_be64(pdu);
860
861                 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
862                                 sizeof(rpdu), &rpdu);
863         }
864 }
865
866 static void blk_add_trace_split(void *ignore,
867                                 struct request_queue *q, struct bio *bio,
868                                 unsigned int pdu)
869 {
870         struct blk_trace *bt = q->blk_trace;
871
872         if (bt) {
873                 __be64 rpdu = cpu_to_be64(pdu);
874
875                 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
876                                 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
877                                 sizeof(rpdu), &rpdu);
878         }
879 }
880
881 /**
882  * blk_add_trace_bio_remap - Add a trace for a bio-remap operation
883  * @ignore:     trace callback data parameter (not used)
884  * @q:          queue the io is for
885  * @bio:        the source bio
886  * @dev:        target device
887  * @from:       source sector
888  *
889  * Description:
890  *     Device mapper or raid target sometimes need to split a bio because
891  *     it spans a stripe (or similar). Add a trace for that action.
892  *
893  **/
894 static void blk_add_trace_bio_remap(void *ignore,
895                                     struct request_queue *q, struct bio *bio,
896                                     dev_t dev, sector_t from)
897 {
898         struct blk_trace *bt = q->blk_trace;
899         struct blk_io_trace_remap r;
900
901         if (likely(!bt))
902                 return;
903
904         r.device_from = cpu_to_be32(dev);
905         r.device_to   = cpu_to_be32(bio->bi_bdev->bd_dev);
906         r.sector_from = cpu_to_be64(from);
907
908         __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
909                         BLK_TA_REMAP, !bio_flagged(bio, BIO_UPTODATE),
910                         sizeof(r), &r);
911 }
912
913 /**
914  * blk_add_trace_rq_remap - Add a trace for a request-remap operation
915  * @ignore:     trace callback data parameter (not used)
916  * @q:          queue the io is for
917  * @rq:         the source request
918  * @dev:        target device
919  * @from:       source sector
920  *
921  * Description:
922  *     Device mapper remaps request to other devices.
923  *     Add a trace for that action.
924  *
925  **/
926 static void blk_add_trace_rq_remap(void *ignore,
927                                    struct request_queue *q,
928                                    struct request *rq, dev_t dev,
929                                    sector_t from)
930 {
931         struct blk_trace *bt = q->blk_trace;
932         struct blk_io_trace_remap r;
933
934         if (likely(!bt))
935                 return;
936
937         r.device_from = cpu_to_be32(dev);
938         r.device_to   = cpu_to_be32(disk_devt(rq->rq_disk));
939         r.sector_from = cpu_to_be64(from);
940
941         __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq),
942                         rq_data_dir(rq), BLK_TA_REMAP, !!rq->errors,
943                         sizeof(r), &r);
944 }
945
946 /**
947  * blk_add_driver_data - Add binary message with driver-specific data
948  * @q:          queue the io is for
949  * @rq:         io request
950  * @data:       driver-specific data
951  * @len:        length of driver-specific data
952  *
953  * Description:
954  *     Some drivers might want to write driver-specific data per request.
955  *
956  **/
957 void blk_add_driver_data(struct request_queue *q,
958                          struct request *rq,
959                          void *data, size_t len)
960 {
961         struct blk_trace *bt = q->blk_trace;
962
963         if (likely(!bt))
964                 return;
965
966         if (rq->cmd_type == REQ_TYPE_BLOCK_PC)
967                 __blk_add_trace(bt, 0, blk_rq_bytes(rq), 0,
968                                 BLK_TA_DRV_DATA, rq->errors, len, data);
969         else
970                 __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), 0,
971                                 BLK_TA_DRV_DATA, rq->errors, len, data);
972 }
973 EXPORT_SYMBOL_GPL(blk_add_driver_data);
974
975 static void blk_register_tracepoints(void)
976 {
977         int ret;
978
979         ret = register_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
980         WARN_ON(ret);
981         ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
982         WARN_ON(ret);
983         ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
984         WARN_ON(ret);
985         ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
986         WARN_ON(ret);
987         ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
988         WARN_ON(ret);
989         ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
990         WARN_ON(ret);
991         ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
992         WARN_ON(ret);
993         ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
994         WARN_ON(ret);
995         ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
996         WARN_ON(ret);
997         ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
998         WARN_ON(ret);
999         ret = register_trace_block_getrq(blk_add_trace_getrq, NULL);
1000         WARN_ON(ret);
1001         ret = register_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1002         WARN_ON(ret);
1003         ret = register_trace_block_plug(blk_add_trace_plug, NULL);
1004         WARN_ON(ret);
1005         ret = register_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1006         WARN_ON(ret);
1007         ret = register_trace_block_split(blk_add_trace_split, NULL);
1008         WARN_ON(ret);
1009         ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1010         WARN_ON(ret);
1011         ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1012         WARN_ON(ret);
1013 }
1014
1015 static void blk_unregister_tracepoints(void)
1016 {
1017         unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL);
1018         unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL);
1019         unregister_trace_block_split(blk_add_trace_split, NULL);
1020         unregister_trace_block_unplug_io(blk_add_trace_unplug_io, NULL);
1021         unregister_trace_block_plug(blk_add_trace_plug, NULL);
1022         unregister_trace_block_sleeprq(blk_add_trace_sleeprq, NULL);
1023         unregister_trace_block_getrq(blk_add_trace_getrq, NULL);
1024         unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL);
1025         unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL);
1026         unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL);
1027         unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL);
1028         unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL);
1029         unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL);
1030         unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL);
1031         unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL);
1032         unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL);
1033         unregister_trace_block_rq_abort(blk_add_trace_rq_abort, NULL);
1034
1035         tracepoint_synchronize_unregister();
1036 }
1037
1038 /*
1039  * struct blk_io_tracer formatting routines
1040  */
1041
1042 static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
1043 {
1044         int i = 0;
1045         int tc = t->action >> BLK_TC_SHIFT;
1046
1047         if (t->action == BLK_TN_MESSAGE) {
1048                 rwbs[i++] = 'N';
1049                 goto out;
1050         }
1051
1052         if (tc & BLK_TC_DISCARD)
1053                 rwbs[i++] = 'D';
1054         else if (tc & BLK_TC_WRITE)
1055                 rwbs[i++] = 'W';
1056         else if (t->bytes)
1057                 rwbs[i++] = 'R';
1058         else
1059                 rwbs[i++] = 'N';
1060
1061         if (tc & BLK_TC_AHEAD)
1062                 rwbs[i++] = 'A';
1063         if (tc & BLK_TC_BARRIER)
1064                 rwbs[i++] = 'B';
1065         if (tc & BLK_TC_SYNC)
1066                 rwbs[i++] = 'S';
1067         if (tc & BLK_TC_META)
1068                 rwbs[i++] = 'M';
1069 out:
1070         rwbs[i] = '\0';
1071 }
1072
1073 static inline
1074 const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
1075 {
1076         return (const struct blk_io_trace *)ent;
1077 }
1078
1079 static inline const void *pdu_start(const struct trace_entry *ent)
1080 {
1081         return te_blk_io_trace(ent) + 1;
1082 }
1083
1084 static inline u32 t_action(const struct trace_entry *ent)
1085 {
1086         return te_blk_io_trace(ent)->action;
1087 }
1088
1089 static inline u32 t_bytes(const struct trace_entry *ent)
1090 {
1091         return te_blk_io_trace(ent)->bytes;
1092 }
1093
1094 static inline u32 t_sec(const struct trace_entry *ent)
1095 {
1096         return te_blk_io_trace(ent)->bytes >> 9;
1097 }
1098
1099 static inline unsigned long long t_sector(const struct trace_entry *ent)
1100 {
1101         return te_blk_io_trace(ent)->sector;
1102 }
1103
1104 static inline __u16 t_error(const struct trace_entry *ent)
1105 {
1106         return te_blk_io_trace(ent)->error;
1107 }
1108
1109 static __u64 get_pdu_int(const struct trace_entry *ent)
1110 {
1111         const __u64 *val = pdu_start(ent);
1112         return be64_to_cpu(*val);
1113 }
1114
1115 static void get_pdu_remap(const struct trace_entry *ent,
1116                           struct blk_io_trace_remap *r)
1117 {
1118         const struct blk_io_trace_remap *__r = pdu_start(ent);
1119         __u64 sector_from = __r->sector_from;
1120
1121         r->device_from = be32_to_cpu(__r->device_from);
1122         r->device_to   = be32_to_cpu(__r->device_to);
1123         r->sector_from = be64_to_cpu(sector_from);
1124 }
1125
1126 typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1127
1128 static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1129 {
1130         char rwbs[6];
1131         unsigned long long ts  = iter->ts;
1132         unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1133         unsigned secs          = (unsigned long)ts;
1134         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1135
1136         fill_rwbs(rwbs, t);
1137
1138         return trace_seq_printf(&iter->seq,
1139                                 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1140                                 MAJOR(t->device), MINOR(t->device), iter->cpu,
1141                                 secs, nsec_rem, iter->ent->pid, act, rwbs);
1142 }
1143
1144 static int blk_log_action(struct trace_iterator *iter, const char *act)
1145 {
1146         char rwbs[6];
1147         const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1148
1149         fill_rwbs(rwbs, t);
1150         return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1151                                 MAJOR(t->device), MINOR(t->device), act, rwbs);
1152 }
1153
1154 static int blk_log_dump_pdu(struct trace_seq *s, const struct trace_entry *ent)
1155 {
1156         const unsigned char *pdu_buf;
1157         int pdu_len;
1158         int i, end, ret;
1159
1160         pdu_buf = pdu_start(ent);
1161         pdu_len = te_blk_io_trace(ent)->pdu_len;
1162
1163         if (!pdu_len)
1164                 return 1;
1165
1166         /* find the last zero that needs to be printed */
1167         for (end = pdu_len - 1; end >= 0; end--)
1168                 if (pdu_buf[end])
1169                         break;
1170         end++;
1171
1172         if (!trace_seq_putc(s, '('))
1173                 return 0;
1174
1175         for (i = 0; i < pdu_len; i++) {
1176
1177                 ret = trace_seq_printf(s, "%s%02x",
1178                                        i == 0 ? "" : " ", pdu_buf[i]);
1179                 if (!ret)
1180                         return ret;
1181
1182                 /*
1183                  * stop when the rest is just zeroes and indicate so
1184                  * with a ".." appended
1185                  */
1186                 if (i == end && end != pdu_len - 1)
1187                         return trace_seq_puts(s, " ..) ");
1188         }
1189
1190         return trace_seq_puts(s, ") ");
1191 }
1192
1193 static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1194 {
1195         char cmd[TASK_COMM_LEN];
1196
1197         trace_find_cmdline(ent->pid, cmd);
1198
1199         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1200                 int ret;
1201
1202                 ret = trace_seq_printf(s, "%u ", t_bytes(ent));
1203                 if (!ret)
1204                         return 0;
1205                 ret = blk_log_dump_pdu(s, ent);
1206                 if (!ret)
1207                         return 0;
1208                 return trace_seq_printf(s, "[%s]\n", cmd);
1209         } else {
1210                 if (t_sec(ent))
1211                         return trace_seq_printf(s, "%llu + %u [%s]\n",
1212                                                 t_sector(ent), t_sec(ent), cmd);
1213                 return trace_seq_printf(s, "[%s]\n", cmd);
1214         }
1215 }
1216
1217 static int blk_log_with_error(struct trace_seq *s,
1218                               const struct trace_entry *ent)
1219 {
1220         if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) {
1221                 int ret;
1222
1223                 ret = blk_log_dump_pdu(s, ent);
1224                 if (ret)
1225                         return trace_seq_printf(s, "[%d]\n", t_error(ent));
1226                 return 0;
1227         } else {
1228                 if (t_sec(ent))
1229                         return trace_seq_printf(s, "%llu + %u [%d]\n",
1230                                                 t_sector(ent),
1231                                                 t_sec(ent), t_error(ent));
1232                 return trace_seq_printf(s, "%llu [%d]\n",
1233                                         t_sector(ent), t_error(ent));
1234         }
1235 }
1236
1237 static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1238 {
1239         struct blk_io_trace_remap r = { .device_from = 0, };
1240
1241         get_pdu_remap(ent, &r);
1242         return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1243                                 t_sector(ent), t_sec(ent),
1244                                 MAJOR(r.device_from), MINOR(r.device_from),
1245                                 (unsigned long long)r.sector_from);
1246 }
1247
1248 static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1249 {
1250         char cmd[TASK_COMM_LEN];
1251
1252         trace_find_cmdline(ent->pid, cmd);
1253
1254         return trace_seq_printf(s, "[%s]\n", cmd);
1255 }
1256
1257 static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1258 {
1259         char cmd[TASK_COMM_LEN];
1260
1261         trace_find_cmdline(ent->pid, cmd);
1262
1263         return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1264 }
1265
1266 static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1267 {
1268         char cmd[TASK_COMM_LEN];
1269
1270         trace_find_cmdline(ent->pid, cmd);
1271
1272         return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1273                                 get_pdu_int(ent), cmd);
1274 }
1275
1276 static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1277 {
1278         int ret;
1279         const struct blk_io_trace *t = te_blk_io_trace(ent);
1280
1281         ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1282         if (ret)
1283                 return trace_seq_putc(s, '\n');
1284         return ret;
1285 }
1286
1287 /*
1288  * struct tracer operations
1289  */
1290
1291 static void blk_tracer_print_header(struct seq_file *m)
1292 {
1293         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1294                 return;
1295         seq_puts(m, "# DEV   CPU TIMESTAMP     PID ACT FLG\n"
1296                     "#  |     |     |           |   |   |\n");
1297 }
1298
1299 static void blk_tracer_start(struct trace_array *tr)
1300 {
1301         blk_tracer_enabled = true;
1302 }
1303
1304 static int blk_tracer_init(struct trace_array *tr)
1305 {
1306         blk_tr = tr;
1307         blk_tracer_start(tr);
1308         return 0;
1309 }
1310
1311 static void blk_tracer_stop(struct trace_array *tr)
1312 {
1313         blk_tracer_enabled = false;
1314 }
1315
1316 static void blk_tracer_reset(struct trace_array *tr)
1317 {
1318         blk_tracer_stop(tr);
1319 }
1320
1321 static const struct {
1322         const char *act[2];
1323         int        (*print)(struct trace_seq *s, const struct trace_entry *ent);
1324 } what2act[] = {
1325         [__BLK_TA_QUEUE]        = {{  "Q", "queue" },      blk_log_generic },
1326         [__BLK_TA_BACKMERGE]    = {{  "M", "backmerge" },  blk_log_generic },
1327         [__BLK_TA_FRONTMERGE]   = {{  "F", "frontmerge" }, blk_log_generic },
1328         [__BLK_TA_GETRQ]        = {{  "G", "getrq" },      blk_log_generic },
1329         [__BLK_TA_SLEEPRQ]      = {{  "S", "sleeprq" },    blk_log_generic },
1330         [__BLK_TA_REQUEUE]      = {{  "R", "requeue" },    blk_log_with_error },
1331         [__BLK_TA_ISSUE]        = {{  "D", "issue" },      blk_log_generic },
1332         [__BLK_TA_COMPLETE]     = {{  "C", "complete" },   blk_log_with_error },
1333         [__BLK_TA_PLUG]         = {{  "P", "plug" },       blk_log_plug },
1334         [__BLK_TA_UNPLUG_IO]    = {{  "U", "unplug_io" },  blk_log_unplug },
1335         [__BLK_TA_INSERT]       = {{  "I", "insert" },     blk_log_generic },
1336         [__BLK_TA_SPLIT]        = {{  "X", "split" },      blk_log_split },
1337         [__BLK_TA_BOUNCE]       = {{  "B", "bounce" },     blk_log_generic },
1338         [__BLK_TA_REMAP]        = {{  "A", "remap" },      blk_log_remap },
1339 };
1340
1341 static enum print_line_t print_one_line(struct trace_iterator *iter,
1342                                         bool classic)
1343 {
1344         struct trace_seq *s = &iter->seq;
1345         const struct blk_io_trace *t;
1346         u16 what;
1347         int ret;
1348         bool long_act;
1349         blk_log_action_t *log_action;
1350
1351         t          = te_blk_io_trace(iter->ent);
1352         what       = t->action & ((1 << BLK_TC_SHIFT) - 1);
1353         long_act   = !!(trace_flags & TRACE_ITER_VERBOSE);
1354         log_action = classic ? &blk_log_action_classic : &blk_log_action;
1355
1356         if (t->action == BLK_TN_MESSAGE) {
1357                 ret = log_action(iter, long_act ? "message" : "m");
1358                 if (ret)
1359                         ret = blk_log_msg(s, iter->ent);
1360                 goto out;
1361         }
1362
1363         if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1364                 ret = trace_seq_printf(s, "Unknown action %x\n", what);
1365         else {
1366                 ret = log_action(iter, what2act[what].act[long_act]);
1367                 if (ret)
1368                         ret = what2act[what].print(s, iter->ent);
1369         }
1370 out:
1371         return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1372 }
1373
1374 static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1375                                                int flags, struct trace_event *event)
1376 {
1377         return print_one_line(iter, false);
1378 }
1379
1380 static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1381 {
1382         struct trace_seq *s = &iter->seq;
1383         struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1384         const int offset = offsetof(struct blk_io_trace, sector);
1385         struct blk_io_trace old = {
1386                 .magic    = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1387                 .time     = iter->ts,
1388         };
1389
1390         if (!trace_seq_putmem(s, &old, offset))
1391                 return 0;
1392         return trace_seq_putmem(s, &t->sector,
1393                                 sizeof(old) - offset + t->pdu_len);
1394 }
1395
1396 static enum print_line_t
1397 blk_trace_event_print_binary(struct trace_iterator *iter, int flags,
1398                              struct trace_event *event)
1399 {
1400         return blk_trace_synthesize_old_trace(iter) ?
1401                         TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1402 }
1403
1404 static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1405 {
1406         if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1407                 return TRACE_TYPE_UNHANDLED;
1408
1409         return print_one_line(iter, true);
1410 }
1411
1412 static int blk_tracer_set_flag(u32 old_flags, u32 bit, int set)
1413 {
1414         /* don't output context-info for blk_classic output */
1415         if (bit == TRACE_BLK_OPT_CLASSIC) {
1416                 if (set)
1417                         trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1418                 else
1419                         trace_flags |= TRACE_ITER_CONTEXT_INFO;
1420         }
1421         return 0;
1422 }
1423
1424 static struct tracer blk_tracer __read_mostly = {
1425         .name           = "blk",
1426         .init           = blk_tracer_init,
1427         .reset          = blk_tracer_reset,
1428         .start          = blk_tracer_start,
1429         .stop           = blk_tracer_stop,
1430         .print_header   = blk_tracer_print_header,
1431         .print_line     = blk_tracer_print_line,
1432         .flags          = &blk_tracer_flags,
1433         .set_flag       = blk_tracer_set_flag,
1434 };
1435
1436 static struct trace_event_functions trace_blk_event_funcs = {
1437         .trace          = blk_trace_event_print,
1438         .binary         = blk_trace_event_print_binary,
1439 };
1440
1441 static struct trace_event trace_blk_event = {
1442         .type           = TRACE_BLK,
1443         .funcs          = &trace_blk_event_funcs,
1444 };
1445
1446 static int __init init_blk_tracer(void)
1447 {
1448         if (!register_ftrace_event(&trace_blk_event)) {
1449                 pr_warning("Warning: could not register block events\n");
1450                 return 1;
1451         }
1452
1453         if (register_tracer(&blk_tracer) != 0) {
1454                 pr_warning("Warning: could not register the block tracer\n");
1455                 unregister_ftrace_event(&trace_blk_event);
1456                 return 1;
1457         }
1458
1459         return 0;
1460 }
1461
1462 device_initcall(init_blk_tracer);
1463
1464 static int blk_trace_remove_queue(struct request_queue *q)
1465 {
1466         struct blk_trace *bt;
1467
1468         bt = xchg(&q->blk_trace, NULL);
1469         if (bt == NULL)
1470                 return -EINVAL;
1471
1472         if (atomic_dec_and_test(&blk_probes_ref))
1473                 blk_unregister_tracepoints();
1474
1475         blk_trace_free(bt);
1476         return 0;
1477 }
1478
1479 /*
1480  * Setup everything required to start tracing
1481  */
1482 static int blk_trace_setup_queue(struct request_queue *q,
1483                                  struct block_device *bdev)
1484 {
1485         struct blk_trace *old_bt, *bt = NULL;
1486         int ret = -ENOMEM;
1487
1488         bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1489         if (!bt)
1490                 return -ENOMEM;
1491
1492         bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1493         if (!bt->msg_data)
1494                 goto free_bt;
1495
1496         bt->dev = bdev->bd_dev;
1497         bt->act_mask = (u16)-1;
1498
1499         blk_trace_setup_lba(bt, bdev);
1500
1501         old_bt = xchg(&q->blk_trace, bt);
1502         if (old_bt != NULL) {
1503                 (void)xchg(&q->blk_trace, old_bt);
1504                 ret = -EBUSY;
1505                 goto free_bt;
1506         }
1507
1508         if (atomic_inc_return(&blk_probes_ref) == 1)
1509                 blk_register_tracepoints();
1510         return 0;
1511
1512 free_bt:
1513         blk_trace_free(bt);
1514         return ret;
1515 }
1516
1517 /*
1518  * sysfs interface to enable and configure tracing
1519  */
1520
1521 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1522                                          struct device_attribute *attr,
1523                                          char *buf);
1524 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1525                                           struct device_attribute *attr,
1526                                           const char *buf, size_t count);
1527 #define BLK_TRACE_DEVICE_ATTR(_name) \
1528         DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1529                     sysfs_blk_trace_attr_show, \
1530                     sysfs_blk_trace_attr_store)
1531
1532 static BLK_TRACE_DEVICE_ATTR(enable);
1533 static BLK_TRACE_DEVICE_ATTR(act_mask);
1534 static BLK_TRACE_DEVICE_ATTR(pid);
1535 static BLK_TRACE_DEVICE_ATTR(start_lba);
1536 static BLK_TRACE_DEVICE_ATTR(end_lba);
1537
1538 static struct attribute *blk_trace_attrs[] = {
1539         &dev_attr_enable.attr,
1540         &dev_attr_act_mask.attr,
1541         &dev_attr_pid.attr,
1542         &dev_attr_start_lba.attr,
1543         &dev_attr_end_lba.attr,
1544         NULL
1545 };
1546
1547 struct attribute_group blk_trace_attr_group = {
1548         .name  = "trace",
1549         .attrs = blk_trace_attrs,
1550 };
1551
1552 static const struct {
1553         int mask;
1554         const char *str;
1555 } mask_maps[] = {
1556         { BLK_TC_READ,          "read"          },
1557         { BLK_TC_WRITE,         "write"         },
1558         { BLK_TC_BARRIER,       "barrier"       },
1559         { BLK_TC_SYNC,          "sync"          },
1560         { BLK_TC_QUEUE,         "queue"         },
1561         { BLK_TC_REQUEUE,       "requeue"       },
1562         { BLK_TC_ISSUE,         "issue"         },
1563         { BLK_TC_COMPLETE,      "complete"      },
1564         { BLK_TC_FS,            "fs"            },
1565         { BLK_TC_PC,            "pc"            },
1566         { BLK_TC_AHEAD,         "ahead"         },
1567         { BLK_TC_META,          "meta"          },
1568         { BLK_TC_DISCARD,       "discard"       },
1569         { BLK_TC_DRV_DATA,      "drv_data"      },
1570 };
1571
1572 static int blk_trace_str2mask(const char *str)
1573 {
1574         int i;
1575         int mask = 0;
1576         char *buf, *s, *token;
1577
1578         buf = kstrdup(str, GFP_KERNEL);
1579         if (buf == NULL)
1580                 return -ENOMEM;
1581         s = strstrip(buf);
1582
1583         while (1) {
1584                 token = strsep(&s, ",");
1585                 if (token == NULL)
1586                         break;
1587
1588                 if (*token == '\0')
1589                         continue;
1590
1591                 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1592                         if (strcasecmp(token, mask_maps[i].str) == 0) {
1593                                 mask |= mask_maps[i].mask;
1594                                 break;
1595                         }
1596                 }
1597                 if (i == ARRAY_SIZE(mask_maps)) {
1598                         mask = -EINVAL;
1599                         break;
1600                 }
1601         }
1602         kfree(buf);
1603
1604         return mask;
1605 }
1606
1607 static ssize_t blk_trace_mask2str(char *buf, int mask)
1608 {
1609         int i;
1610         char *p = buf;
1611
1612         for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1613                 if (mask & mask_maps[i].mask) {
1614                         p += sprintf(p, "%s%s",
1615                                     (p == buf) ? "" : ",", mask_maps[i].str);
1616                 }
1617         }
1618         *p++ = '\n';
1619
1620         return p - buf;
1621 }
1622
1623 static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1624 {
1625         if (bdev->bd_disk == NULL)
1626                 return NULL;
1627
1628         return bdev_get_queue(bdev);
1629 }
1630
1631 static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1632                                          struct device_attribute *attr,
1633                                          char *buf)
1634 {
1635         struct hd_struct *p = dev_to_part(dev);
1636         struct request_queue *q;
1637         struct block_device *bdev;
1638         ssize_t ret = -ENXIO;
1639
1640         bdev = bdget(part_devt(p));
1641         if (bdev == NULL)
1642                 goto out;
1643
1644         q = blk_trace_get_queue(bdev);
1645         if (q == NULL)
1646                 goto out_bdput;
1647
1648         mutex_lock(&bdev->bd_mutex);
1649
1650         if (attr == &dev_attr_enable) {
1651                 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1652                 goto out_unlock_bdev;
1653         }
1654
1655         if (q->blk_trace == NULL)
1656                 ret = sprintf(buf, "disabled\n");
1657         else if (attr == &dev_attr_act_mask)
1658                 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1659         else if (attr == &dev_attr_pid)
1660                 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1661         else if (attr == &dev_attr_start_lba)
1662                 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1663         else if (attr == &dev_attr_end_lba)
1664                 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1665
1666 out_unlock_bdev:
1667         mutex_unlock(&bdev->bd_mutex);
1668 out_bdput:
1669         bdput(bdev);
1670 out:
1671         return ret;
1672 }
1673
1674 static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1675                                           struct device_attribute *attr,
1676                                           const char *buf, size_t count)
1677 {
1678         struct block_device *bdev;
1679         struct request_queue *q;
1680         struct hd_struct *p;
1681         u64 value;
1682         ssize_t ret = -EINVAL;
1683
1684         if (count == 0)
1685                 goto out;
1686
1687         if (attr == &dev_attr_act_mask) {
1688                 if (sscanf(buf, "%llx", &value) != 1) {
1689                         /* Assume it is a list of trace category names */
1690                         ret = blk_trace_str2mask(buf);
1691                         if (ret < 0)
1692                                 goto out;
1693                         value = ret;
1694                 }
1695         } else if (sscanf(buf, "%llu", &value) != 1)
1696                 goto out;
1697
1698         ret = -ENXIO;
1699
1700         p = dev_to_part(dev);
1701         bdev = bdget(part_devt(p));
1702         if (bdev == NULL)
1703                 goto out;
1704
1705         q = blk_trace_get_queue(bdev);
1706         if (q == NULL)
1707                 goto out_bdput;
1708
1709         mutex_lock(&bdev->bd_mutex);
1710
1711         if (attr == &dev_attr_enable) {
1712                 if (value)
1713                         ret = blk_trace_setup_queue(q, bdev);
1714                 else
1715                         ret = blk_trace_remove_queue(q);
1716                 goto out_unlock_bdev;
1717         }
1718
1719         ret = 0;
1720         if (q->blk_trace == NULL)
1721                 ret = blk_trace_setup_queue(q, bdev);
1722
1723         if (ret == 0) {
1724                 if (attr == &dev_attr_act_mask)
1725                         q->blk_trace->act_mask = value;
1726                 else if (attr == &dev_attr_pid)
1727                         q->blk_trace->pid = value;
1728                 else if (attr == &dev_attr_start_lba)
1729                         q->blk_trace->start_lba = value;
1730                 else if (attr == &dev_attr_end_lba)
1731                         q->blk_trace->end_lba = value;
1732         }
1733
1734 out_unlock_bdev:
1735         mutex_unlock(&bdev->bd_mutex);
1736 out_bdput:
1737         bdput(bdev);
1738 out:
1739         return ret ? ret : count;
1740 }
1741
1742 int blk_trace_init_sysfs(struct device *dev)
1743 {
1744         return sysfs_create_group(&dev->kobj, &blk_trace_attr_group);
1745 }
1746
1747 void blk_trace_remove_sysfs(struct device *dev)
1748 {
1749         sysfs_remove_group(&dev->kobj, &blk_trace_attr_group);
1750 }
1751
1752 #endif /* CONFIG_BLK_DEV_IO_TRACE */
1753
1754 #ifdef CONFIG_EVENT_TRACING
1755
1756 void blk_dump_cmd(char *buf, struct request *rq)
1757 {
1758         int i, end;
1759         int len = rq->cmd_len;
1760         unsigned char *cmd = rq->cmd;
1761
1762         if (rq->cmd_type != REQ_TYPE_BLOCK_PC) {
1763                 buf[0] = '\0';
1764                 return;
1765         }
1766
1767         for (end = len - 1; end >= 0; end--)
1768                 if (cmd[end])
1769                         break;
1770         end++;
1771
1772         for (i = 0; i < len; i++) {
1773                 buf += sprintf(buf, "%s%02x", i == 0 ? "" : " ", cmd[i]);
1774                 if (i == end && end != len - 1) {
1775                         sprintf(buf, " ..");
1776                         break;
1777                 }
1778         }
1779 }
1780
1781 void blk_fill_rwbs(char *rwbs, u32 rw, int bytes)
1782 {
1783         int i = 0;
1784
1785         if (rw & WRITE)
1786                 rwbs[i++] = 'W';
1787         else if (rw & REQ_DISCARD)
1788                 rwbs[i++] = 'D';
1789         else if (bytes)
1790                 rwbs[i++] = 'R';
1791         else
1792                 rwbs[i++] = 'N';
1793
1794         if (rw & REQ_RAHEAD)
1795                 rwbs[i++] = 'A';
1796         if (rw & REQ_SYNC)
1797                 rwbs[i++] = 'S';
1798         if (rw & REQ_META)
1799                 rwbs[i++] = 'M';
1800         if (rw & REQ_SECURE)
1801                 rwbs[i++] = 'E';
1802
1803         rwbs[i] = '\0';
1804 }
1805
1806 #endif /* CONFIG_EVENT_TRACING */
1807