blob: 8d3a277802603dd89c252784ea94f8dcc2d4b6d1 [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
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +020078void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
79{
80 int n;
81 va_list args;
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +020082 unsigned long flags;
Jens Axboe64565912008-05-28 14:45:33 +020083 char *buf;
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +020084
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +020085 local_irq_save(flags);
Jens Axboe64565912008-05-28 14:45:33 +020086 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +020087 va_start(args, fmt);
Jens Axboe64565912008-05-28 14:45:33 +020088 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +020089 va_end(args);
90
Jens Axboe64565912008-05-28 14:45:33 +020091 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +020092 local_irq_restore(flags);
Alan D. Brunelle9d5f09a2008-05-27 14:54:41 +020093}
94EXPORT_SYMBOL_GPL(__trace_note_message);
95
Jens Axboe2056a782006-03-23 20:00:26 +010096static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
97 pid_t pid)
98{
99 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
100 return 1;
101 if (sector < bt->start_lba || sector > bt->end_lba)
102 return 1;
103 if (bt->pid && pid != bt->pid)
104 return 1;
105
106 return 0;
107}
108
109/*
110 * Data direction bit lookup
111 */
112static u32 ddir_act[2] __read_mostly = { BLK_TC_ACT(BLK_TC_READ), BLK_TC_ACT(BLK_TC_WRITE) };
113
114/*
115 * Bio action bits of interest
116 */
Jens Axboe7457e6e2006-07-23 02:12:01 +0200117static 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 +0100118
119/*
120 * More could be added as needed, taking care to increment the decrementer
121 * to get correct indexing
122 */
123#define trace_barrier_bit(rw) \
124 (((rw) & (1 << BIO_RW_BARRIER)) >> (BIO_RW_BARRIER - 0))
125#define trace_sync_bit(rw) \
126 (((rw) & (1 << BIO_RW_SYNC)) >> (BIO_RW_SYNC - 1))
Nathan Scott40359cc2006-07-06 10:03:28 +0200127#define trace_ahead_bit(rw) \
Milton Millerad01b1c2006-07-25 15:04:13 +0200128 (((rw) & (1 << BIO_RW_AHEAD)) << (2 - BIO_RW_AHEAD))
Jens Axboe7457e6e2006-07-23 02:12:01 +0200129#define trace_meta_bit(rw) \
130 (((rw) & (1 << BIO_RW_META)) >> (BIO_RW_META - 3))
Jens Axboe2056a782006-03-23 20:00:26 +0100131
132/*
133 * The worker for the various blk_add_trace*() types. Fills out a
134 * blk_io_trace structure and places it in a per-cpu subbuffer.
135 */
136void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
137 int rw, u32 what, int error, int pdu_len, void *pdu_data)
138{
139 struct task_struct *tsk = current;
140 struct blk_io_trace *t;
141 unsigned long flags;
142 unsigned long *sequence;
143 pid_t pid;
144 int cpu;
145
146 if (unlikely(bt->trace_state != Blktrace_running))
147 return;
148
149 what |= ddir_act[rw & WRITE];
150 what |= bio_act[trace_barrier_bit(rw)];
151 what |= bio_act[trace_sync_bit(rw)];
Nathan Scott40359cc2006-07-06 10:03:28 +0200152 what |= bio_act[trace_ahead_bit(rw)];
Jens Axboe7457e6e2006-07-23 02:12:01 +0200153 what |= bio_act[trace_meta_bit(rw)];
Jens Axboe2056a782006-03-23 20:00:26 +0100154
155 pid = tsk->pid;
156 if (unlikely(act_log_check(bt, what, sector, pid)))
157 return;
158
159 /*
160 * A word about the locking here - we disable interrupts to reserve
161 * some space in the relay per-cpu buffer, to prevent an irq
Carl Henrik Lunde14a73f52008-06-12 20:13:58 +0200162 * from coming in and stepping on our toes.
Jens Axboe2056a782006-03-23 20:00:26 +0100163 */
164 local_irq_save(flags);
165
166 if (unlikely(tsk->btrace_seq != blktrace_seq))
167 trace_note_tsk(bt, tsk);
168
169 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
170 if (t) {
171 cpu = smp_processor_id();
172 sequence = per_cpu_ptr(bt->sequence, cpu);
173
174 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
175 t->sequence = ++(*sequence);
Ingo Molnar2997c8c2008-01-11 13:35:54 +0100176 t->time = ktime_to_ns(ktime_get());
Jens Axboe2056a782006-03-23 20:00:26 +0100177 t->sector = sector;
178 t->bytes = bytes;
179 t->action = what;
180 t->pid = pid;
181 t->device = bt->dev;
182 t->cpu = cpu;
183 t->error = error;
184 t->pdu_len = pdu_len;
185
186 if (pdu_len)
187 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
188 }
189
190 local_irq_restore(flags);
191}
192
193EXPORT_SYMBOL_GPL(__blk_add_trace);
194
195static struct dentry *blk_tree_root;
Jens Axboe11a57152008-01-11 13:37:01 +0100196static DEFINE_MUTEX(blk_tree_mutex);
Jens Axboe2056a782006-03-23 20:00:26 +0100197static unsigned int root_users;
198
199static inline void blk_remove_root(void)
200{
201 if (blk_tree_root) {
202 debugfs_remove(blk_tree_root);
203 blk_tree_root = NULL;
204 }
205}
206
207static void blk_remove_tree(struct dentry *dir)
208{
209 mutex_lock(&blk_tree_mutex);
210 debugfs_remove(dir);
211 if (--root_users == 0)
212 blk_remove_root();
213 mutex_unlock(&blk_tree_mutex);
214}
215
216static struct dentry *blk_create_tree(const char *blk_name)
217{
218 struct dentry *dir = NULL;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100219 int created = 0;
Jens Axboe2056a782006-03-23 20:00:26 +0100220
221 mutex_lock(&blk_tree_mutex);
222
223 if (!blk_tree_root) {
224 blk_tree_root = debugfs_create_dir("block", NULL);
225 if (!blk_tree_root)
226 goto err;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100227 created = 1;
Jens Axboe2056a782006-03-23 20:00:26 +0100228 }
229
230 dir = debugfs_create_dir(blk_name, blk_tree_root);
231 if (dir)
232 root_users++;
Aneesh Kumar K.V35fc51e2007-11-21 12:25:41 +0100233 else {
234 /* Delete root only if we created it */
235 if (created)
236 blk_remove_root();
237 }
Jens Axboe2056a782006-03-23 20:00:26 +0100238
239err:
240 mutex_unlock(&blk_tree_mutex);
241 return dir;
242}
243
244static void blk_trace_cleanup(struct blk_trace *bt)
245{
246 relay_close(bt->rchan);
247 debugfs_remove(bt->dropped_file);
248 blk_remove_tree(bt->dir);
249 free_percpu(bt->sequence);
Jens Axboe64565912008-05-28 14:45:33 +0200250 free_percpu(bt->msg_data);
Jens Axboe2056a782006-03-23 20:00:26 +0100251 kfree(bt);
252}
253
Christof Schmitt6da127a2008-01-11 10:09:43 +0100254int blk_trace_remove(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100255{
256 struct blk_trace *bt;
257
258 bt = xchg(&q->blk_trace, NULL);
259 if (!bt)
260 return -EINVAL;
261
262 if (bt->trace_state == Blktrace_setup ||
263 bt->trace_state == Blktrace_stopped)
264 blk_trace_cleanup(bt);
265
266 return 0;
267}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100268EXPORT_SYMBOL_GPL(blk_trace_remove);
Jens Axboe2056a782006-03-23 20:00:26 +0100269
270static int blk_dropped_open(struct inode *inode, struct file *filp)
271{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700272 filp->private_data = inode->i_private;
Jens Axboe2056a782006-03-23 20:00:26 +0100273
274 return 0;
275}
276
277static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
278 size_t count, loff_t *ppos)
279{
280 struct blk_trace *bt = filp->private_data;
281 char buf[16];
282
283 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
284
285 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
286}
287
Arjan van de Ven2b8693c2007-02-12 00:55:32 -0800288static const struct file_operations blk_dropped_fops = {
Jens Axboe2056a782006-03-23 20:00:26 +0100289 .owner = THIS_MODULE,
290 .open = blk_dropped_open,
291 .read = blk_dropped_read,
292};
293
294/*
295 * Keep track of how many times we encountered a full subbuffer, to aid
296 * the user space app in telling how many lost events there were.
297 */
298static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
299 void *prev_subbuf, size_t prev_padding)
300{
301 struct blk_trace *bt;
302
303 if (!relay_buf_full(buf))
304 return 1;
305
306 bt = buf->chan->private_data;
307 atomic_inc(&bt->dropped);
308 return 0;
309}
310
311static int blk_remove_buf_file_callback(struct dentry *dentry)
312{
313 debugfs_remove(dentry);
314 return 0;
315}
316
317static struct dentry *blk_create_buf_file_callback(const char *filename,
318 struct dentry *parent,
319 int mode,
320 struct rchan_buf *buf,
321 int *is_global)
322{
323 return debugfs_create_file(filename, mode, parent, buf,
324 &relay_file_operations);
325}
326
327static struct rchan_callbacks blk_relay_callbacks = {
328 .subbuf_start = blk_subbuf_start_callback,
329 .create_buf_file = blk_create_buf_file_callback,
330 .remove_buf_file = blk_remove_buf_file_callback,
331};
332
333/*
334 * Setup everything required to start tracing
335 */
Christof Schmitt6da127a2008-01-11 10:09:43 +0100336int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
Arnd Bergmann171044d42007-10-09 13:23:53 +0200337 struct blk_user_trace_setup *buts)
Jens Axboe2056a782006-03-23 20:00:26 +0100338{
Jens Axboe2056a782006-03-23 20:00:26 +0100339 struct blk_trace *old_bt, *bt = NULL;
340 struct dentry *dir = NULL;
Jens Axboe2056a782006-03-23 20:00:26 +0100341 int ret, i;
342
Arnd Bergmann171044d42007-10-09 13:23:53 +0200343 if (!buts->buf_size || !buts->buf_nr)
Jens Axboe2056a782006-03-23 20:00:26 +0100344 return -EINVAL;
345
Christof Schmitt6da127a2008-01-11 10:09:43 +0100346 strcpy(buts->name, name);
Jens Axboe2056a782006-03-23 20:00:26 +0100347
348 /*
349 * some device names have larger paths - convert the slashes
350 * to underscores for this to work as expected
351 */
Arnd Bergmann171044d42007-10-09 13:23:53 +0200352 for (i = 0; i < strlen(buts->name); i++)
353 if (buts->name[i] == '/')
354 buts->name[i] = '_';
Jens Axboe2056a782006-03-23 20:00:26 +0100355
356 ret = -ENOMEM;
357 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
358 if (!bt)
359 goto err;
360
361 bt->sequence = alloc_percpu(unsigned long);
362 if (!bt->sequence)
363 goto err;
364
Jens Axboe64565912008-05-28 14:45:33 +0200365 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG);
366 if (!bt->msg_data)
367 goto err;
368
Jens Axboe2056a782006-03-23 20:00:26 +0100369 ret = -ENOENT;
Arnd Bergmann171044d42007-10-09 13:23:53 +0200370 dir = blk_create_tree(buts->name);
Jens Axboe2056a782006-03-23 20:00:26 +0100371 if (!dir)
372 goto err;
373
374 bt->dir = dir;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100375 bt->dev = dev;
Jens Axboe2056a782006-03-23 20:00:26 +0100376 atomic_set(&bt->dropped, 0);
377
378 ret = -EIO;
379 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops);
380 if (!bt->dropped_file)
381 goto err;
382
Arnd Bergmann171044d42007-10-09 13:23:53 +0200383 bt->rchan = relay_open("trace", dir, buts->buf_size,
384 buts->buf_nr, &blk_relay_callbacks, bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100385 if (!bt->rchan)
386 goto err;
Jens Axboe2056a782006-03-23 20:00:26 +0100387
Arnd Bergmann171044d42007-10-09 13:23:53 +0200388 bt->act_mask = buts->act_mask;
Jens Axboe2056a782006-03-23 20:00:26 +0100389 if (!bt->act_mask)
390 bt->act_mask = (u16) -1;
391
Arnd Bergmann171044d42007-10-09 13:23:53 +0200392 bt->start_lba = buts->start_lba;
393 bt->end_lba = buts->end_lba;
Jens Axboe2056a782006-03-23 20:00:26 +0100394 if (!bt->end_lba)
395 bt->end_lba = -1ULL;
396
Arnd Bergmann171044d42007-10-09 13:23:53 +0200397 bt->pid = buts->pid;
Jens Axboe2056a782006-03-23 20:00:26 +0100398 bt->trace_state = Blktrace_setup;
399
400 ret = -EBUSY;
401 old_bt = xchg(&q->blk_trace, bt);
402 if (old_bt) {
403 (void) xchg(&q->blk_trace, old_bt);
404 goto err;
405 }
406
407 return 0;
408err:
409 if (dir)
410 blk_remove_tree(dir);
411 if (bt) {
412 if (bt->dropped_file)
413 debugfs_remove(bt->dropped_file);
Alan Sterna1205862006-12-06 20:32:37 -0800414 free_percpu(bt->sequence);
Jens Axboe64565912008-05-28 14:45:33 +0200415 free_percpu(bt->msg_data);
Jens Axboe2056a782006-03-23 20:00:26 +0100416 if (bt->rchan)
417 relay_close(bt->rchan);
418 kfree(bt);
419 }
420 return ret;
421}
422
Christof Schmitt6da127a2008-01-11 10:09:43 +0100423int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
424 char __user *arg)
Arnd Bergmann171044d42007-10-09 13:23:53 +0200425{
426 struct blk_user_trace_setup buts;
427 int ret;
428
429 ret = copy_from_user(&buts, arg, sizeof(buts));
430 if (ret)
431 return -EFAULT;
432
Christof Schmitt6da127a2008-01-11 10:09:43 +0100433 ret = do_blk_trace_setup(q, name, dev, &buts);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200434 if (ret)
435 return ret;
436
437 if (copy_to_user(arg, &buts, sizeof(buts)))
438 return -EFAULT;
439
440 return 0;
441}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100442EXPORT_SYMBOL_GPL(blk_trace_setup);
Arnd Bergmann171044d42007-10-09 13:23:53 +0200443
Christof Schmitt6da127a2008-01-11 10:09:43 +0100444int blk_trace_startstop(struct request_queue *q, int start)
Jens Axboe2056a782006-03-23 20:00:26 +0100445{
446 struct blk_trace *bt;
447 int ret;
448
449 if ((bt = q->blk_trace) == NULL)
450 return -EINVAL;
451
452 /*
453 * For starting a trace, we can transition from a setup or stopped
454 * trace. For stopping a trace, the state must be running
455 */
456 ret = -EINVAL;
457 if (start) {
458 if (bt->trace_state == Blktrace_setup ||
459 bt->trace_state == Blktrace_stopped) {
460 blktrace_seq++;
461 smp_mb();
462 bt->trace_state = Blktrace_running;
Olaf Kirchbe1c6342006-12-01 10:39:12 +0100463
464 trace_note_time(bt);
Jens Axboe2056a782006-03-23 20:00:26 +0100465 ret = 0;
466 }
467 } else {
468 if (bt->trace_state == Blktrace_running) {
469 bt->trace_state = Blktrace_stopped;
470 relay_flush(bt->rchan);
471 ret = 0;
472 }
473 }
474
475 return ret;
476}
Christof Schmitt6da127a2008-01-11 10:09:43 +0100477EXPORT_SYMBOL_GPL(blk_trace_startstop);
Jens Axboe2056a782006-03-23 20:00:26 +0100478
479/**
480 * blk_trace_ioctl: - handle the ioctls associated with tracing
481 * @bdev: the block device
482 * @cmd: the ioctl cmd
483 * @arg: the argument data, if any
484 *
485 **/
486int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
487{
Jens Axboe165125e2007-07-24 09:28:11 +0200488 struct request_queue *q;
Jens Axboe2056a782006-03-23 20:00:26 +0100489 int ret, start = 0;
Christof Schmitt6da127a2008-01-11 10:09:43 +0100490 char b[BDEVNAME_SIZE];
Jens Axboe2056a782006-03-23 20:00:26 +0100491
492 q = bdev_get_queue(bdev);
493 if (!q)
494 return -ENXIO;
495
496 mutex_lock(&bdev->bd_mutex);
497
498 switch (cmd) {
499 case BLKTRACESETUP:
Jean Delvaref36f21e2008-05-12 14:02:33 -0700500 bdevname(bdev, b);
Christof Schmitt6da127a2008-01-11 10:09:43 +0100501 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
Jens Axboe2056a782006-03-23 20:00:26 +0100502 break;
503 case BLKTRACESTART:
504 start = 1;
505 case BLKTRACESTOP:
506 ret = blk_trace_startstop(q, start);
507 break;
508 case BLKTRACETEARDOWN:
509 ret = blk_trace_remove(q);
510 break;
511 default:
512 ret = -ENOTTY;
513 break;
514 }
515
516 mutex_unlock(&bdev->bd_mutex);
517 return ret;
518}
519
520/**
521 * blk_trace_shutdown: - stop and cleanup trace structures
522 * @q: the request queue associated with the device
523 *
524 **/
Jens Axboe165125e2007-07-24 09:28:11 +0200525void blk_trace_shutdown(struct request_queue *q)
Jens Axboe2056a782006-03-23 20:00:26 +0100526{
Alexey Dobriyan6c5c9342006-09-29 01:59:40 -0700527 if (q->blk_trace) {
528 blk_trace_startstop(q, 0);
529 blk_trace_remove(q);
530 }
Jens Axboe2056a782006-03-23 20:00:26 +0100531}