blob: b2cbb4e5d7673c7a49ea557d962022f7ba3594f9 [file] [log] [blame]
Jens Axboe2056a782006-03-23 20:00:26 +01001/*
Jens Axboe0fe23472006-09-04 15:41:16 +02002 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
Jens Axboe2056a782006-03-23 20:00:26 +01003 *
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 */
Jens Axboe2056a782006-03-23 20:00:26 +010018#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/debugfs.h>
Olaf Kirchbe1c6342006-12-01 10:39:12 +010025#include <linux/time.h>
Jens Axboe2056a782006-03-23 20:00:26 +010026#include <asm/uaccess.h>
27
Jens Axboe2056a782006-03-23 20:00:26 +010028static unsigned int blktrace_seq __read_mostly = 1;
29
30/*
Olaf Kirchbe1c6342006-12-01 10:39:12 +010031 * Send out a notify message.
32 */
Jens Axboea8630552006-12-04 09:30:58 +010033static void trace_note(struct blk_trace *bt, pid_t pid, int action,
34 const void *data, size_t len)
Olaf Kirchbe1c6342006-12-01 10:39:12 +010035{
36 struct blk_io_trace *t;
Olaf Kirchbe1c6342006-12-01 10:39:12 +010037
38 t = relay_reserve(bt->rchan, sizeof(*t) + len);
Jens Axboed3d9d2a2006-12-04 09:27:41 +010039 if (t) {
40 const int cpu = smp_processor_id();
Olaf Kirchbe1c6342006-12-01 10:39:12 +010041
Jens Axboed3d9d2a2006-12-04 09:27:41 +010042 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
Ingo Molnar2997c8c2008-01-11 13:35:54 +010043 t->time = ktime_to_ns(ktime_get());
Jens Axboed3d9d2a2006-12-04 09:27:41 +010044 t->device = bt->dev;
45 t->action = action;
46 t->pid = pid;
47 t->cpu = cpu;
48 t->pdu_len = len;
49 memcpy((void *) t + sizeof(*t), data, len);
50 }
Olaf Kirchbe1c6342006-12-01 10:39:12 +010051}
52
53/*
Jens Axboe2056a782006-03-23 20:00:26 +010054 * Send out a notify for this process, if we haven't done so since a trace
55 * started
56 */
57static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
58{
Jens Axboea8630552006-12-04 09:30:58 +010059 tsk->btrace_seq = blktrace_seq;
60 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
Olaf Kirchbe1c6342006-12-01 10:39:12 +010061}
Jens Axboe2056a782006-03-23 20:00:26 +010062
Olaf Kirchbe1c6342006-12-01 10:39:12 +010063static void trace_note_time(struct blk_trace *bt)
64{
65 struct timespec now;
66 unsigned long flags;
67 u32 words[2];
68
69 getnstimeofday(&now);
70 words[0] = now.tv_sec;
71 words[1] = now.tv_nsec;
72
73 local_irq_save(flags);
74 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
75 local_irq_restore(flags);
Jens Axboe2056a782006-03-23 20:00:26 +010076}
77
78static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
79 pid_t pid)
80{
81 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
82 return 1;
83 if (sector < bt->start_lba || sector > bt->end_lba)
84 return 1;
85 if (bt->pid && pid != bt->pid)
86 return 1;
87
88 return 0;
89}
90
91/*
92 * Data direction bit lookup
93 */
94static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), BLK_TC_ACT(BLK_TC_WRITE) };
95
96/*
97 * Bio action bits of interest
98 */
Jens Axboe7457e6e2006-07-23 02:12:01 +020099static u32 bio_act[9] __read_mostly = { 0, BLK_TC_ACT(BLK_TC_BARRIER), BLK_TC_ACT(BLK_TC_SYNC), 0, BLK_TC_ACT(BLK_TC_AHEAD), 0, 0, 0, BLK_TC_ACT(BLK_TC_META) };
Jens Axboe2056a782006-03-23 20:00:26 +0100100
101/*
102 * More could be added as needed, taking care to increment the decrementer
103 * to get correct indexing
104 */
105#define trace_barrier_bit(rw) \
106 (((rw) & (1 << BIO_RW_BARRIER)) >> (BIO_RW_BARRIER - 0))
107#define trace_sync_bit(rw) \
108 (((rw) & (1 << BIO_RW_SYNC)) >> (BIO_RW_SYNC - 1))
Nathan Scott40359cc2006-07-06 10:03:28 +0200109#define trace_ahead_bit(rw) \
Milton Millerad01b1c2006-07-25 15:04:13 +0200110 (((rw) & (1 << BIO_RW_AHEAD)) << (2 - BIO_RW_AHEAD))
Jens Axboe7457e6e2006-07-23 02:12:01 +0200111#define trace_meta_bit(rw) \
112 (((rw) & (1 << BIO_RW_META)) >> (BIO_RW_META - 3))
Jens Axboe2056a782006-03-23 20:00:26 +0100113
114/*
115 * The worker for the various blk_add_trace*() types. Fills out a
116 * blk_io_trace structure and places it in a per-cpu subbuffer.
117 */
118void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
119 int rw, u32 what, int error, int pdu_len, void *pdu_data)
120{
121 struct task_struct *tsk = current;
122 struct blk_io_trace *t;
123 unsigned long flags;
124 unsigned long *sequence;
125 pid_t pid;
126 int cpu;
127
128 if (unlikely(bt->trace_state != Blktrace_running))
129 return;
130
131 what |= ddir_act[rw & WRITE];
132 what |= bio_act[trace_barrier_bit(rw)];
133 what |= bio_act[trace_sync_bit(rw)];
Nathan Scott40359cc2006-07-06 10:03:28 +0200134 what |= bio_act[trace_ahead_bit(rw)];
Jens Axboe7457e6e2006-07-23 02:12:01 +0200135 what |= bio_act[trace_meta_bit(rw)];
Jens Axboe2056a782006-03-23 20:00:26 +0100136
137 pid = tsk->pid;
138 if (unlikely(act_log_check(bt, what, sector, pid)))
139 return;
140
141 /*
142 * A word about the locking here - we disable interrupts to reserve
143 * some space in the relay per-cpu buffer, to prevent an irq
144 * from coming in and stepping on our toes. Once reserved, it's
145 * enough to get preemption disabled to prevent read of this data
146 * before we are through filling it. get_cpu()/put_cpu() does this
147 * for us
148 */
149 local_irq_save(flags);
150
151 if (unlikely(tsk->btrace_seq != blktrace_seq))
152 trace_note_tsk(bt, tsk);
153
154 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
155 if (t) {
156 cpu = smp_processor_id();
157 sequence = per_cpu_ptr(bt->sequence, cpu);
158
159 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
160 t->sequence = ++(*sequence);
Ingo Molnar2997c8c2008-01-11 13:35:54 +0100161 t->time = ktime_to_ns(ktime_get());
Jens Axboe2056a782006-03-23 20:00:26 +0100162 t->sector = sector;
163 t->bytes = bytes;
164 t->action = what;
165 t->pid = pid;
166 t->device = bt->dev;
167 t->cpu = cpu;
168 t->error = error;
169 t->pdu_len = pdu_len;
170
171 if (pdu_len)
172 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
173 }
174
175 local_irq_restore(flags);
176}
177
178EXPORT_SYMBOL_GPL(__blk_add_trace);
179
180static struct dentry *blk_tree_root;
Jens Axboe11a57152008-01-11 13:37:01 +0100181static DEFINE_MUTEX(blk_tree_mutex);
Jens Axboe2056a782006-03-23 20:00:26 +0100182static unsigned int root_users;
183
184static inline void blk_remove_root(void)
185{
186 if (blk_tree_root) {
187 debugfs_remove(blk_tree_root);
188 blk_tree_root = NULL;
189 }
190}
191
192static void blk_remove_tree(struct dentry *dir)
193{
194 mutex_lock(&blk_tree_mutex);
195 debugfs_remove(dir);
196 if (--root_users == 0)
197 blk_remove_root();
198 mutex_unlock(&blk_tree_mutex);
199}
200
201static struct dentry *blk_create_tree(const char *blk_name)
202{
203 struct dentry *dir = NULL;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100204 int created = 0;
Jens Axboe2056a782006-03-23 20:00:26 +0100205
206 mutex_lock(&blk_tree_mutex);
207
208 if (!blk_tree_root) {
209 blk_tree_root = debugfs_create_dir("block", NULL);
210 if (!blk_tree_root)
211 goto err;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100212 created = 1;
Jens Axboe2056a782006-03-23 20:00:26 +0100213 }
214
215 dir = debugfs_create_dir(blk_name, blk_tree_root);
216 if (dir)
217 root_users++;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100218 else {
219 /* Delete root only if we created it */
220 if (created)
221 blk_remove_root();
222 }
Jens Axboe2056a782006-03-23 20:00:26 +0100223
224err:
225 mutex_unlock(&blk_tree_mutex);
226 return dir;
227}
228
229static void blk_trace_cleanup(struct blk_trace *bt)
230{
231 relay_close(bt->rchan);
232 debugfs_remove(bt->dropped_file);
233 blk_remove_tree(bt->dir);
234 free_percpu(bt->sequence);
235 kfree(bt);
236}
237
Christof Schmitt6da127a2008-01-11 10:09:43 +0100238int blk_trace_remove(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100239{
240 struct blk_trace *bt;
241
242 bt = xchg(&q->blk_trace, NULL);
243 if (!bt)
244 return -EINVAL;
245
246 if (bt->trace_state == Blktrace_setup ||
247 bt->trace_state == Blktrace_stopped)
248 blk_trace_cleanup(bt);
249
250 return 0;
251}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100252EXPORT_SYMBOL_GPL(blk_trace_remove);
Jens Axboe2056a782006-03-23 20:00:26 +0100253
254static int blk_dropped_open(struct inode *inode, struct file *filp)
255{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700256 filp->private_data = inode->i_private;
Jens Axboe2056a782006-03-23 20:00:26 +0100257
258 return 0;
259}
260
261static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
262 size_t count, loff_t *ppos)
263{
264 struct blk_trace *bt = filp->private_data;
265 char buf[16];
266
267 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
268
269 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
270}
271
Arjan van de Ven2b8693c2007-02-12 00:55:32 -0800272static const struct file_operations blk_dropped_fops = {
Jens Axboe2056a782006-03-23 20:00:26 +0100273 .owner = THIS_MODULE,
274 .open = blk_dropped_open,
275 .read = blk_dropped_read,
276};
277
278/*
279 * Keep track of how many times we encountered a full subbuffer, to aid
280 * the user space app in telling how many lost events there were.
281 */
282static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
283 void *prev_subbuf, size_t prev_padding)
284{
285 struct blk_trace *bt;
286
287 if (!relay_buf_full(buf))
288 return 1;
289
290 bt = buf->chan->private_data;
291 atomic_inc(&bt->dropped);
292 return 0;
293}
294
295static int blk_remove_buf_file_callback(struct dentry *dentry)
296{
297 debugfs_remove(dentry);
298 return 0;
299}
300
301static struct dentry *blk_create_buf_file_callback(const char *filename,
302 struct dentry *parent,
303 int mode,
304 struct rchan_buf *buf,
305 int *is_global)
306{
307 return debugfs_create_file(filename, mode, parent, buf,
308 &relay_file_operations);
309}
310
311static struct rchan_callbacks blk_relay_callbacks = {
312 .subbuf_start = blk_subbuf_start_callback,
313 .create_buf_file = blk_create_buf_file_callback,
314 .remove_buf_file = blk_remove_buf_file_callback,
315};
316
317/*
318 * Setup everything required to start tracing
319 */
Christof Schmitt6da127a2008-01-11 10:09:43 +0100320int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
Arnd Bergmann171044d42007-10-09 13:23:53 +0200321 struct blk_user_trace_setup *buts)
Jens Axboe2056a782006-03-23 20:00:26 +0100322{
Jens Axboe2056a782006-03-23 20:00:26 +0100323 struct blk_trace *old_bt, *bt = NULL;
324 struct dentry *dir = NULL;
Jens Axboe2056a782006-03-23 20:00:26 +0100325 int ret, i;
326
Arnd Bergmann171044d42007-10-09 13:23:53 +0200327 if (!buts->buf_size || !buts->buf_nr)
Jens Axboe2056a782006-03-23 20:00:26 +0100328 return -EINVAL;
329
Christof Schmitt6da127a2008-01-11 10:09:43 +0100330 strcpy(buts->name, name);
Jens Axboe2056a782006-03-23 20:00:26 +0100331
332 /*
333 * some device names have larger paths - convert the slashes
334 * to underscores for this to work as expected
335 */
Arnd Bergmann171044d42007-10-09 13:23:53 +0200336 for (i = 0; i < strlen(buts->name); i++)
337 if (buts->name[i] == '/')
338 buts->name[i] = '_';
Jens Axboe2056a782006-03-23 20:00:26 +0100339
340 ret = -ENOMEM;
341 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
342 if (!bt)
343 goto err;
344
345 bt->sequence = alloc_percpu(unsigned long);
346 if (!bt->sequence)
347 goto err;
348
349 ret = -ENOENT;
Arnd Bergmann171044d42007-10-09 13:23:53 +0200350 dir = blk_create_tree(buts->name);
Jens Axboe2056a782006-03-23 20:00:26 +0100351 if (!dir)
352 goto err;
353
354 bt->dir = dir;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100355 bt->dev = dev;
Jens Axboe2056a782006-03-23 20:00:26 +0100356 atomic_set(&bt->dropped, 0);
357
358 ret = -EIO;
359 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
360 if (!bt->dropped_file)
361 goto err;
362
Arnd Bergmann171044d42007-10-09 13:23:53 +0200363 bt->rchan = relay_open("trace", dir, buts->buf_size,
364 buts->buf_nr, &blk_relay_callbacks, bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100365 if (!bt->rchan)
366 goto err;
Jens Axboe2056a782006-03-23 20:00:26 +0100367
Arnd Bergmann171044d42007-10-09 13:23:53 +0200368 bt->act_mask = buts->act_mask;
Jens Axboe2056a782006-03-23 20:00:26 +0100369 if (!bt->act_mask)
370 bt->act_mask = (u16) -1;
371
Arnd Bergmann171044d42007-10-09 13:23:53 +0200372 bt->start_lba = buts->start_lba;
373 bt->end_lba = buts->end_lba;
Jens Axboe2056a782006-03-23 20:00:26 +0100374 if (!bt->end_lba)
375 bt->end_lba = -1ULL;
376
Arnd Bergmann171044d42007-10-09 13:23:53 +0200377 bt->pid = buts->pid;
Jens Axboe2056a782006-03-23 20:00:26 +0100378 bt->trace_state = Blktrace_setup;
379
380 ret = -EBUSY;
381 old_bt = xchg(&q->blk_trace, bt);
382 if (old_bt) {
383 (void) xchg(&q->blk_trace, old_bt);
384 goto err;
385 }
386
387 return 0;
388err:
389 if (dir)
390 blk_remove_tree(dir);
391 if (bt) {
392 if (bt->dropped_file)
393 debugfs_remove(bt->dropped_file);
Alan Sterna1205862006-12-06 20:32:37 -0800394 free_percpu(bt->sequence);
Jens Axboe2056a782006-03-23 20:00:26 +0100395 if (bt->rchan)
396 relay_close(bt->rchan);
397 kfree(bt);
398 }
399 return ret;
400}
401
Christof Schmitt6da127a2008-01-11 10:09:43 +0100402int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
403 char __user *arg)
Arnd Bergmann171044d42007-10-09 13:23:53 +0200404{
405 struct blk_user_trace_setup buts;
406 int ret;
407
408 ret = copy_from_user(&buts, arg, sizeof(buts));
409 if (ret)
410 return -EFAULT;
411
Christof Schmitt6da127a2008-01-11 10:09:43 +0100412 ret = do_blk_trace_setup(q, name, dev, &buts);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200413 if (ret)
414 return ret;
415
416 if (copy_to_user(arg, &buts, sizeof(buts)))
417 return -EFAULT;
418
419 return 0;
420}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100421EXPORT_SYMBOL_GPL(blk_trace_setup);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200422
Christof Schmitt6da127a2008-01-11 10:09:43 +0100423int blk_trace_startstop(struct request_queue *q, int start)
Jens Axboe2056a782006-03-23 20:00:26 +0100424{
425 struct blk_trace *bt;
426 int ret;
427
428 if ((bt = q->blk_trace) == NULL)
429 return -EINVAL;
430
431 /*
432 * For starting a trace, we can transition from a setup or stopped
433 * trace. For stopping a trace, the state must be running
434 */
435 ret = -EINVAL;
436 if (start) {
437 if (bt->trace_state == Blktrace_setup ||
438 bt->trace_state == Blktrace_stopped) {
439 blktrace_seq++;
440 smp_mb();
441 bt->trace_state = Blktrace_running;
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100442
443 trace_note_time(bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100444 ret = 0;
445 }
446 } else {
447 if (bt->trace_state == Blktrace_running) {
448 bt->trace_state = Blktrace_stopped;
449 relay_flush(bt->rchan);
450 ret = 0;
451 }
452 }
453
454 return ret;
455}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100456EXPORT_SYMBOL_GPL(blk_trace_startstop);
Jens Axboe2056a782006-03-23 20:00:26 +0100457
458/**
459 * blk_trace_ioctl: - handle the ioctls associated with tracing
460 * @bdev: the block device
461 * @cmd: the ioctl cmd
462 * @arg: the argument data, if any
463 *
464 **/
465int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
466{
Jens Axboe165125e2007-07-24 09:28:11 +0200467 struct request_queue *q;
Jens Axboe2056a782006-03-23 20:00:26 +0100468 int ret, start = 0;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100469 char b[BDEVNAME_SIZE];
Jens Axboe2056a782006-03-23 20:00:26 +0100470
471 q = bdev_get_queue(bdev);
472 if (!q)
473 return -ENXIO;
474
475 mutex_lock(&bdev->bd_mutex);
476
477 switch (cmd) {
478 case BLKTRACESETUP:
Jean Delvaref36f21e2008-05-12 14:02:33 -0700479 bdevname(bdev, b);
Christof Schmitt6da127a2008-01-11 10:09:43 +0100480 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
Jens Axboe2056a782006-03-23 20:00:26 +0100481 break;
482 case BLKTRACESTART:
483 start = 1;
484 case BLKTRACESTOP:
485 ret = blk_trace_startstop(q, start);
486 break;
487 case BLKTRACETEARDOWN:
488 ret = blk_trace_remove(q);
489 break;
490 default:
491 ret = -ENOTTY;
492 break;
493 }
494
495 mutex_unlock(&bdev->bd_mutex);
496 return ret;
497}
498
499/**
500 * blk_trace_shutdown: - stop and cleanup trace structures
501 * @q: the request queue associated with the device
502 *
503 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200504void blk_trace_shutdown(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100505{
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -0700506 if (q->blk_trace) {
507 blk_trace_startstop(q, 0);
508 blk_trace_remove(q);
509 }
Jens Axboe2056a782006-03-23 20:00:26 +0100510}