]>
Commit | Line | Data |
---|---|---|
91c1e6ba | 1 | // SPDX-License-Identifier: GPL-2.0 |
2056a782 | 2 | /* |
0fe23479 | 3 | * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk> |
2056a782 | 4 | * |
2056a782 | 5 | */ |
1b0b2836 LC |
6 | |
7 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
8 | ||
2056a782 JA |
9 | #include <linux/kernel.h> |
10 | #include <linux/blkdev.h> | |
11 | #include <linux/blktrace_api.h> | |
12 | #include <linux/percpu.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/mutex.h> | |
5a0e3ad6 | 15 | #include <linux/slab.h> |
2056a782 | 16 | #include <linux/debugfs.h> |
6e5fdeed | 17 | #include <linux/export.h> |
be1c6341 | 18 | #include <linux/time.h> |
939b3669 | 19 | #include <linux/uaccess.h> |
a404d557 | 20 | #include <linux/list.h> |
ca1136c9 | 21 | #include <linux/blk-cgroup.h> |
55782138 | 22 | |
18fbda91 OS |
23 | #include "../../block/blk.h" |
24 | ||
55782138 LZ |
25 | #include <trace/events/block.h> |
26 | ||
2db270a8 | 27 | #include "trace_output.h" |
2056a782 | 28 | |
55782138 LZ |
29 | #ifdef CONFIG_BLK_DEV_IO_TRACE |
30 | ||
2056a782 JA |
31 | static unsigned int blktrace_seq __read_mostly = 1; |
32 | ||
c71a8961 | 33 | static struct trace_array *blk_tr; |
5006ea73 | 34 | static bool blk_tracer_enabled __read_mostly; |
c71a8961 | 35 | |
a404d557 JK |
36 | static LIST_HEAD(running_trace_list); |
37 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(running_trace_lock); | |
38 | ||
c71a8961 | 39 | /* Select an alternative, minimalistic output than the original one */ |
ef18012b | 40 | #define TRACE_BLK_OPT_CLASSIC 0x1 |
ca1136c9 | 41 | #define TRACE_BLK_OPT_CGROUP 0x2 |
69fd5c39 | 42 | #define TRACE_BLK_OPT_CGNAME 0x4 |
c71a8961 ACM |
43 | |
44 | static struct tracer_opt blk_tracer_opts[] = { | |
45 | /* Default disable the minimalistic output */ | |
157f9c00 | 46 | { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) }, |
ca1136c9 SL |
47 | #ifdef CONFIG_BLK_CGROUP |
48 | { TRACER_OPT(blk_cgroup, TRACE_BLK_OPT_CGROUP) }, | |
69fd5c39 | 49 | { TRACER_OPT(blk_cgname, TRACE_BLK_OPT_CGNAME) }, |
ca1136c9 | 50 | #endif |
c71a8961 ACM |
51 | { } |
52 | }; | |
53 | ||
54 | static struct tracer_flags blk_tracer_flags = { | |
55 | .val = 0, | |
56 | .opts = blk_tracer_opts, | |
57 | }; | |
58 | ||
5f3ea37c | 59 | /* Global reference count of probes */ |
a6da0024 JA |
60 | static DEFINE_MUTEX(blk_probe_mutex); |
61 | static int blk_probes_ref; | |
5f3ea37c | 62 | |
3c289ba7 | 63 | static void blk_register_tracepoints(void); |
5f3ea37c ACM |
64 | static void blk_unregister_tracepoints(void); |
65 | ||
be1c6341 OK |
66 | /* |
67 | * Send out a notify message. | |
68 | */ | |
a863055b | 69 | static void trace_note(struct blk_trace *bt, pid_t pid, int action, |
67c0496e | 70 | const void *data, size_t len, u64 cgid) |
be1c6341 OK |
71 | { |
72 | struct blk_io_trace *t; | |
18cea459 | 73 | struct ring_buffer_event *event = NULL; |
13292494 | 74 | struct trace_buffer *buffer = NULL; |
36590c50 | 75 | unsigned int trace_ctx = 0; |
18cea459 LZ |
76 | int cpu = smp_processor_id(); |
77 | bool blk_tracer = blk_tracer_enabled; | |
67c0496e | 78 | ssize_t cgid_len = cgid ? sizeof(cgid) : 0; |
18cea459 LZ |
79 | |
80 | if (blk_tracer) { | |
1c5eb448 | 81 | buffer = blk_tr->array_buffer.buffer; |
36590c50 | 82 | trace_ctx = tracing_gen_ctx_flags(0); |
e77405ad | 83 | event = trace_buffer_lock_reserve(buffer, TRACE_BLK, |
ca1136c9 | 84 | sizeof(*t) + len + cgid_len, |
36590c50 | 85 | trace_ctx); |
18cea459 LZ |
86 | if (!event) |
87 | return; | |
88 | t = ring_buffer_event_data(event); | |
89 | goto record_it; | |
90 | } | |
be1c6341 | 91 | |
c71a8961 ACM |
92 | if (!bt->rchan) |
93 | return; | |
94 | ||
ca1136c9 | 95 | t = relay_reserve(bt->rchan, sizeof(*t) + len + cgid_len); |
d3d9d2a5 | 96 | if (t) { |
d3d9d2a5 | 97 | t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; |
2997c8c4 | 98 | t->time = ktime_to_ns(ktime_get()); |
18cea459 | 99 | record_it: |
d3d9d2a5 | 100 | t->device = bt->dev; |
ca1136c9 | 101 | t->action = action | (cgid ? __BLK_TN_CGROUP : 0); |
d3d9d2a5 JA |
102 | t->pid = pid; |
103 | t->cpu = cpu; | |
ca1136c9 | 104 | t->pdu_len = len + cgid_len; |
67c0496e TH |
105 | if (cgid_len) |
106 | memcpy((void *)t + sizeof(*t), &cgid, cgid_len); | |
ca1136c9 | 107 | memcpy((void *) t + sizeof(*t) + cgid_len, data, len); |
18cea459 LZ |
108 | |
109 | if (blk_tracer) | |
36590c50 | 110 | trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx); |
d3d9d2a5 | 111 | } |
be1c6341 OK |
112 | } |
113 | ||
2056a782 JA |
114 | /* |
115 | * Send out a notify for this process, if we haven't done so since a trace | |
116 | * started | |
117 | */ | |
a404d557 | 118 | static void trace_note_tsk(struct task_struct *tsk) |
2056a782 | 119 | { |
a404d557 JK |
120 | unsigned long flags; |
121 | struct blk_trace *bt; | |
122 | ||
a863055b | 123 | tsk->btrace_seq = blktrace_seq; |
a404d557 JK |
124 | spin_lock_irqsave(&running_trace_lock, flags); |
125 | list_for_each_entry(bt, &running_trace_list, running_list) { | |
126 | trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, | |
67c0496e | 127 | sizeof(tsk->comm), 0); |
a404d557 JK |
128 | } |
129 | spin_unlock_irqrestore(&running_trace_lock, flags); | |
be1c6341 | 130 | } |
2056a782 | 131 | |
be1c6341 OK |
132 | static void trace_note_time(struct blk_trace *bt) |
133 | { | |
59a37f8b | 134 | struct timespec64 now; |
be1c6341 OK |
135 | unsigned long flags; |
136 | u32 words[2]; | |
137 | ||
59a37f8b AB |
138 | /* need to check user space to see if this breaks in y2038 or y2106 */ |
139 | ktime_get_real_ts64(&now); | |
140 | words[0] = (u32)now.tv_sec; | |
be1c6341 OK |
141 | words[1] = now.tv_nsec; |
142 | ||
143 | local_irq_save(flags); | |
67c0496e | 144 | trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words), 0); |
be1c6341 | 145 | local_irq_restore(flags); |
2056a782 JA |
146 | } |
147 | ||
35fe6d76 SL |
148 | void __trace_note_message(struct blk_trace *bt, struct blkcg *blkcg, |
149 | const char *fmt, ...) | |
9d5f09a4 AB |
150 | { |
151 | int n; | |
152 | va_list args; | |
14a73f54 | 153 | unsigned long flags; |
64565911 | 154 | char *buf; |
9d5f09a4 | 155 | |
18cea459 LZ |
156 | if (unlikely(bt->trace_state != Blktrace_running && |
157 | !blk_tracer_enabled)) | |
c71a8961 ACM |
158 | return; |
159 | ||
490da40d TM |
160 | /* |
161 | * If the BLK_TC_NOTIFY action mask isn't set, don't send any note | |
162 | * message to the trace. | |
163 | */ | |
164 | if (!(bt->act_mask & BLK_TC_NOTIFY)) | |
165 | return; | |
166 | ||
14a73f54 | 167 | local_irq_save(flags); |
d8a0349c | 168 | buf = this_cpu_ptr(bt->msg_data); |
9d5f09a4 | 169 | va_start(args, fmt); |
64565911 | 170 | n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args); |
9d5f09a4 AB |
171 | va_end(args); |
172 | ||
35fe6d76 SL |
173 | if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP)) |
174 | blkcg = NULL; | |
175 | #ifdef CONFIG_BLK_CGROUP | |
870c153c | 176 | trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, |
74321038 | 177 | blkcg ? cgroup_id(blkcg->css.cgroup) : 1); |
35fe6d76 | 178 | #else |
870c153c | 179 | trace_note(bt, current->pid, BLK_TN_MESSAGE, buf, n, 0); |
35fe6d76 | 180 | #endif |
14a73f54 | 181 | local_irq_restore(flags); |
9d5f09a4 AB |
182 | } |
183 | EXPORT_SYMBOL_GPL(__trace_note_message); | |
184 | ||
2056a782 JA |
185 | static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector, |
186 | pid_t pid) | |
187 | { | |
188 | if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0) | |
189 | return 1; | |
d0deef5b | 190 | if (sector && (sector < bt->start_lba || sector > bt->end_lba)) |
2056a782 JA |
191 | return 1; |
192 | if (bt->pid && pid != bt->pid) | |
193 | return 1; | |
194 | ||
195 | return 0; | |
196 | } | |
197 | ||
198 | /* | |
199 | * Data direction bit lookup | |
200 | */ | |
e4955c99 LZ |
201 | static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ), |
202 | BLK_TC_ACT(BLK_TC_WRITE) }; | |
2056a782 | 203 | |
7b6d91da | 204 | #define BLK_TC_RAHEAD BLK_TC_AHEAD |
28a8f0d3 | 205 | #define BLK_TC_PREFLUSH BLK_TC_FLUSH |
7b6d91da | 206 | |
35ba8f70 | 207 | /* The ilog2() calls fall out because they're constant */ |
7b6d91da CH |
208 | #define MASK_TC_BIT(rw, __name) ((rw & REQ_ ## __name) << \ |
209 | (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - __REQ_ ## __name)) | |
2056a782 JA |
210 | |
211 | /* | |
212 | * The worker for the various blk_add_trace*() types. Fills out a | |
213 | * blk_io_trace structure and places it in a per-cpu subbuffer. | |
214 | */ | |
5f3ea37c | 215 | static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes, |
1b9a9ab7 | 216 | int op, int op_flags, u32 what, int error, int pdu_len, |
67c0496e | 217 | void *pdu_data, u64 cgid) |
2056a782 JA |
218 | { |
219 | struct task_struct *tsk = current; | |
c71a8961 | 220 | struct ring_buffer_event *event = NULL; |
13292494 | 221 | struct trace_buffer *buffer = NULL; |
2056a782 | 222 | struct blk_io_trace *t; |
0a987751 | 223 | unsigned long flags = 0; |
2056a782 | 224 | unsigned long *sequence; |
36590c50 | 225 | unsigned int trace_ctx = 0; |
2056a782 | 226 | pid_t pid; |
36590c50 | 227 | int cpu; |
18cea459 | 228 | bool blk_tracer = blk_tracer_enabled; |
67c0496e | 229 | ssize_t cgid_len = cgid ? sizeof(cgid) : 0; |
2056a782 | 230 | |
18cea459 | 231 | if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer)) |
2056a782 JA |
232 | return; |
233 | ||
1b9a9ab7 MC |
234 | what |= ddir_act[op_is_write(op) ? WRITE : READ]; |
235 | what |= MASK_TC_BIT(op_flags, SYNC); | |
236 | what |= MASK_TC_BIT(op_flags, RAHEAD); | |
237 | what |= MASK_TC_BIT(op_flags, META); | |
28a8f0d3 | 238 | what |= MASK_TC_BIT(op_flags, PREFLUSH); |
1b9a9ab7 | 239 | what |= MASK_TC_BIT(op_flags, FUA); |
7afafc8a | 240 | if (op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE) |
1b9a9ab7 | 241 | what |= BLK_TC_ACT(BLK_TC_DISCARD); |
3a5e02ce MC |
242 | if (op == REQ_OP_FLUSH) |
243 | what |= BLK_TC_ACT(BLK_TC_FLUSH); | |
ca1136c9 SL |
244 | if (cgid) |
245 | what |= __BLK_TA_CGROUP; | |
2056a782 JA |
246 | |
247 | pid = tsk->pid; | |
d0deef5b | 248 | if (act_log_check(bt, what, sector, pid)) |
2056a782 | 249 | return; |
c71a8961 ACM |
250 | cpu = raw_smp_processor_id(); |
251 | ||
18cea459 | 252 | if (blk_tracer) { |
c71a8961 ACM |
253 | tracing_record_cmdline(current); |
254 | ||
1c5eb448 | 255 | buffer = blk_tr->array_buffer.buffer; |
36590c50 | 256 | trace_ctx = tracing_gen_ctx_flags(0); |
e77405ad | 257 | event = trace_buffer_lock_reserve(buffer, TRACE_BLK, |
ca1136c9 | 258 | sizeof(*t) + pdu_len + cgid_len, |
36590c50 | 259 | trace_ctx); |
c71a8961 ACM |
260 | if (!event) |
261 | return; | |
51a763dd | 262 | t = ring_buffer_event_data(event); |
c71a8961 ACM |
263 | goto record_it; |
264 | } | |
2056a782 | 265 | |
a404d557 JK |
266 | if (unlikely(tsk->btrace_seq != blktrace_seq)) |
267 | trace_note_tsk(tsk); | |
268 | ||
2056a782 JA |
269 | /* |
270 | * A word about the locking here - we disable interrupts to reserve | |
271 | * some space in the relay per-cpu buffer, to prevent an irq | |
14a73f54 | 272 | * from coming in and stepping on our toes. |
2056a782 JA |
273 | */ |
274 | local_irq_save(flags); | |
ca1136c9 | 275 | t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len + cgid_len); |
2056a782 | 276 | if (t) { |
2056a782 JA |
277 | sequence = per_cpu_ptr(bt->sequence, cpu); |
278 | ||
279 | t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION; | |
280 | t->sequence = ++(*sequence); | |
2997c8c4 | 281 | t->time = ktime_to_ns(ktime_get()); |
c71a8961 | 282 | record_it: |
08a06b83 | 283 | /* |
939b3669 ACM |
284 | * These two are not needed in ftrace as they are in the |
285 | * generic trace_entry, filled by tracing_generic_entry_update, | |
286 | * but for the trace_event->bin() synthesizer benefit we do it | |
287 | * here too. | |
288 | */ | |
289 | t->cpu = cpu; | |
290 | t->pid = pid; | |
08a06b83 | 291 | |
2056a782 JA |
292 | t->sector = sector; |
293 | t->bytes = bytes; | |
294 | t->action = what; | |
2056a782 | 295 | t->device = bt->dev; |
2056a782 | 296 | t->error = error; |
ca1136c9 | 297 | t->pdu_len = pdu_len + cgid_len; |
2056a782 | 298 | |
ca1136c9 | 299 | if (cgid_len) |
67c0496e | 300 | memcpy((void *)t + sizeof(*t), &cgid, cgid_len); |
2056a782 | 301 | if (pdu_len) |
ca1136c9 | 302 | memcpy((void *)t + sizeof(*t) + cgid_len, pdu_data, pdu_len); |
c71a8961 | 303 | |
18cea459 | 304 | if (blk_tracer) { |
36590c50 | 305 | trace_buffer_unlock_commit(blk_tr, buffer, event, trace_ctx); |
c71a8961 ACM |
306 | return; |
307 | } | |
2056a782 JA |
308 | } |
309 | ||
310 | local_irq_restore(flags); | |
311 | } | |
312 | ||
ad5dd549 | 313 | static void blk_trace_free(struct blk_trace *bt) |
2056a782 | 314 | { |
f48fc4d3 | 315 | relay_close(bt->rchan); |
39cbb602 | 316 | debugfs_remove(bt->dir); |
2056a782 | 317 | free_percpu(bt->sequence); |
64565911 | 318 | free_percpu(bt->msg_data); |
2056a782 | 319 | kfree(bt); |
ad5dd549 LZ |
320 | } |
321 | ||
a6da0024 JA |
322 | static void get_probe_ref(void) |
323 | { | |
324 | mutex_lock(&blk_probe_mutex); | |
325 | if (++blk_probes_ref == 1) | |
326 | blk_register_tracepoints(); | |
327 | mutex_unlock(&blk_probe_mutex); | |
328 | } | |
329 | ||
330 | static void put_probe_ref(void) | |
331 | { | |
332 | mutex_lock(&blk_probe_mutex); | |
333 | if (!--blk_probes_ref) | |
334 | blk_unregister_tracepoints(); | |
335 | mutex_unlock(&blk_probe_mutex); | |
336 | } | |
337 | ||
ad5dd549 LZ |
338 | static void blk_trace_cleanup(struct blk_trace *bt) |
339 | { | |
c780e86d | 340 | synchronize_rcu(); |
ad5dd549 | 341 | blk_trace_free(bt); |
a6da0024 | 342 | put_probe_ref(); |
2056a782 JA |
343 | } |
344 | ||
1f2cac10 | 345 | static int __blk_trace_remove(struct request_queue *q) |
2056a782 JA |
346 | { |
347 | struct blk_trace *bt; | |
348 | ||
c3dbe541 | 349 | bt = rcu_replace_pointer(q->blk_trace, NULL, |
85e0cbbb | 350 | lockdep_is_held(&q->debugfs_mutex)); |
2056a782 JA |
351 | if (!bt) |
352 | return -EINVAL; | |
353 | ||
55547204 | 354 | if (bt->trace_state != Blktrace_running) |
2056a782 JA |
355 | blk_trace_cleanup(bt); |
356 | ||
357 | return 0; | |
358 | } | |
1f2cac10 JA |
359 | |
360 | int blk_trace_remove(struct request_queue *q) | |
361 | { | |
362 | int ret; | |
363 | ||
85e0cbbb | 364 | mutex_lock(&q->debugfs_mutex); |
1f2cac10 | 365 | ret = __blk_trace_remove(q); |
85e0cbbb | 366 | mutex_unlock(&q->debugfs_mutex); |
1f2cac10 JA |
367 | |
368 | return ret; | |
369 | } | |
6da127ad | 370 | EXPORT_SYMBOL_GPL(blk_trace_remove); |
2056a782 | 371 | |
2056a782 JA |
372 | static ssize_t blk_dropped_read(struct file *filp, char __user *buffer, |
373 | size_t count, loff_t *ppos) | |
374 | { | |
375 | struct blk_trace *bt = filp->private_data; | |
376 | char buf[16]; | |
377 | ||
378 | snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped)); | |
379 | ||
380 | return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf)); | |
381 | } | |
382 | ||
2b8693c0 | 383 | static const struct file_operations blk_dropped_fops = { |
2056a782 | 384 | .owner = THIS_MODULE, |
234e3405 | 385 | .open = simple_open, |
2056a782 | 386 | .read = blk_dropped_read, |
6038f373 | 387 | .llseek = default_llseek, |
2056a782 JA |
388 | }; |
389 | ||
02c62304 AB |
390 | static ssize_t blk_msg_write(struct file *filp, const char __user *buffer, |
391 | size_t count, loff_t *ppos) | |
392 | { | |
393 | char *msg; | |
394 | struct blk_trace *bt; | |
395 | ||
7635b03a | 396 | if (count >= BLK_TN_MAX_MSG) |
02c62304 AB |
397 | return -EINVAL; |
398 | ||
16e5c1fc AV |
399 | msg = memdup_user_nul(buffer, count); |
400 | if (IS_ERR(msg)) | |
401 | return PTR_ERR(msg); | |
02c62304 | 402 | |
02c62304 | 403 | bt = filp->private_data; |
35fe6d76 | 404 | __trace_note_message(bt, NULL, "%s", msg); |
02c62304 AB |
405 | kfree(msg); |
406 | ||
407 | return count; | |
408 | } | |
409 | ||
410 | static const struct file_operations blk_msg_fops = { | |
411 | .owner = THIS_MODULE, | |
234e3405 | 412 | .open = simple_open, |
02c62304 | 413 | .write = blk_msg_write, |
6038f373 | 414 | .llseek = noop_llseek, |
02c62304 AB |
415 | }; |
416 | ||
2056a782 JA |
417 | /* |
418 | * Keep track of how many times we encountered a full subbuffer, to aid | |
419 | * the user space app in telling how many lost events there were. | |
420 | */ | |
421 | static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf, | |
422 | void *prev_subbuf, size_t prev_padding) | |
423 | { | |
424 | struct blk_trace *bt; | |
425 | ||
426 | if (!relay_buf_full(buf)) | |
427 | return 1; | |
428 | ||
429 | bt = buf->chan->private_data; | |
430 | atomic_inc(&bt->dropped); | |
431 | return 0; | |
432 | } | |
433 | ||
434 | static int blk_remove_buf_file_callback(struct dentry *dentry) | |
435 | { | |
436 | debugfs_remove(dentry); | |
f48fc4d3 | 437 | |
2056a782 JA |
438 | return 0; |
439 | } | |
440 | ||
441 | static struct dentry *blk_create_buf_file_callback(const char *filename, | |
442 | struct dentry *parent, | |
f4ae40a6 | 443 | umode_t mode, |
2056a782 JA |
444 | struct rchan_buf *buf, |
445 | int *is_global) | |
446 | { | |
447 | return debugfs_create_file(filename, mode, parent, buf, | |
448 | &relay_file_operations); | |
449 | } | |
450 | ||
abf4e00c | 451 | static const struct rchan_callbacks blk_relay_callbacks = { |
2056a782 JA |
452 | .subbuf_start = blk_subbuf_start_callback, |
453 | .create_buf_file = blk_create_buf_file_callback, | |
454 | .remove_buf_file = blk_remove_buf_file_callback, | |
455 | }; | |
456 | ||
9908c309 LZ |
457 | static void blk_trace_setup_lba(struct blk_trace *bt, |
458 | struct block_device *bdev) | |
459 | { | |
29ff57c6 CH |
460 | if (bdev) { |
461 | bt->start_lba = bdev->bd_start_sect; | |
462 | bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev); | |
9908c309 LZ |
463 | } else { |
464 | bt->start_lba = 0; | |
465 | bt->end_lba = -1ULL; | |
466 | } | |
467 | } | |
468 | ||
2056a782 JA |
469 | /* |
470 | * Setup everything required to start tracing | |
471 | */ | |
a428d314 OS |
472 | static int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev, |
473 | struct block_device *bdev, | |
474 | struct blk_user_trace_setup *buts) | |
2056a782 | 475 | { |
cdea01b2 | 476 | struct blk_trace *bt = NULL; |
2056a782 | 477 | struct dentry *dir = NULL; |
ff14417c | 478 | int ret; |
2056a782 | 479 | |
85e0cbbb | 480 | lockdep_assert_held(&q->debugfs_mutex); |
a67549c8 | 481 | |
171044d4 | 482 | if (!buts->buf_size || !buts->buf_nr) |
2056a782 JA |
483 | return -EINVAL; |
484 | ||
0497b345 JA |
485 | strncpy(buts->name, name, BLKTRACE_BDEV_SIZE); |
486 | buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0'; | |
2056a782 JA |
487 | |
488 | /* | |
489 | * some device names have larger paths - convert the slashes | |
490 | * to underscores for this to work as expected | |
491 | */ | |
ff14417c | 492 | strreplace(buts->name, '/', '_'); |
2056a782 | 493 | |
1b0b2836 LC |
494 | /* |
495 | * bdev can be NULL, as with scsi-generic, this is a helpful as | |
496 | * we can be. | |
497 | */ | |
c3dbe541 | 498 | if (rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 499 | lockdep_is_held(&q->debugfs_mutex))) { |
1b0b2836 LC |
500 | pr_warn("Concurrent blktraces are not allowed on %s\n", |
501 | buts->name); | |
502 | return -EBUSY; | |
503 | } | |
504 | ||
2056a782 JA |
505 | bt = kzalloc(sizeof(*bt), GFP_KERNEL); |
506 | if (!bt) | |
ad5dd549 | 507 | return -ENOMEM; |
2056a782 | 508 | |
ad5dd549 | 509 | ret = -ENOMEM; |
2056a782 JA |
510 | bt->sequence = alloc_percpu(unsigned long); |
511 | if (!bt->sequence) | |
512 | goto err; | |
513 | ||
313e458f | 514 | bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); |
64565911 JA |
515 | if (!bt->msg_data) |
516 | goto err; | |
517 | ||
bad8e64f | 518 | /* |
85e0cbbb LC |
519 | * When tracing the whole disk reuse the existing debugfs directory |
520 | * created by the block layer on init. For partitions block devices, | |
bad8e64f LC |
521 | * and scsi-generic block devices we create a temporary new debugfs |
522 | * directory that will be removed once the trace ends. | |
523 | */ | |
fa01b1e9 | 524 | if (bdev && !bdev_is_partition(bdev)) |
bad8e64f LC |
525 | dir = q->debugfs_dir; |
526 | else | |
6ac93117 | 527 | bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root); |
2056a782 | 528 | |
b431ef83 LC |
529 | /* |
530 | * As blktrace relies on debugfs for its interface the debugfs directory | |
531 | * is required, contrary to the usual mantra of not checking for debugfs | |
532 | * files or directories. | |
533 | */ | |
534 | if (IS_ERR_OR_NULL(dir)) { | |
535 | pr_warn("debugfs_dir not present for %s so skipping\n", | |
536 | buts->name); | |
537 | ret = -ENOENT; | |
538 | goto err; | |
539 | } | |
540 | ||
6da127ad | 541 | bt->dev = dev; |
2056a782 | 542 | atomic_set(&bt->dropped, 0); |
a404d557 | 543 | INIT_LIST_HEAD(&bt->running_list); |
2056a782 JA |
544 | |
545 | ret = -EIO; | |
c0ea5760 GKH |
546 | debugfs_create_file("dropped", 0444, dir, bt, &blk_dropped_fops); |
547 | debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops); | |
02c62304 | 548 | |
171044d4 AB |
549 | bt->rchan = relay_open("trace", dir, buts->buf_size, |
550 | buts->buf_nr, &blk_relay_callbacks, bt); | |
2056a782 JA |
551 | if (!bt->rchan) |
552 | goto err; | |
2056a782 | 553 | |
171044d4 | 554 | bt->act_mask = buts->act_mask; |
2056a782 JA |
555 | if (!bt->act_mask) |
556 | bt->act_mask = (u16) -1; | |
557 | ||
9908c309 | 558 | blk_trace_setup_lba(bt, bdev); |
2056a782 | 559 | |
d0deef5b SD |
560 | /* overwrite with user settings */ |
561 | if (buts->start_lba) | |
562 | bt->start_lba = buts->start_lba; | |
563 | if (buts->end_lba) | |
564 | bt->end_lba = buts->end_lba; | |
565 | ||
171044d4 | 566 | bt->pid = buts->pid; |
2056a782 JA |
567 | bt->trace_state = Blktrace_setup; |
568 | ||
c3dbe541 | 569 | rcu_assign_pointer(q->blk_trace, bt); |
a6da0024 | 570 | get_probe_ref(); |
cbe28296 | 571 | |
6ac93117 | 572 | ret = 0; |
2056a782 | 573 | err: |
6ac93117 OS |
574 | if (ret) |
575 | blk_trace_free(bt); | |
2056a782 JA |
576 | return ret; |
577 | } | |
171044d4 | 578 | |
1f2cac10 JA |
579 | static int __blk_trace_setup(struct request_queue *q, char *name, dev_t dev, |
580 | struct block_device *bdev, char __user *arg) | |
171044d4 AB |
581 | { |
582 | struct blk_user_trace_setup buts; | |
583 | int ret; | |
584 | ||
585 | ret = copy_from_user(&buts, arg, sizeof(buts)); | |
586 | if (ret) | |
587 | return -EFAULT; | |
588 | ||
d0deef5b | 589 | ret = do_blk_trace_setup(q, name, dev, bdev, &buts); |
171044d4 AB |
590 | if (ret) |
591 | return ret; | |
592 | ||
9a8c28c8 | 593 | if (copy_to_user(arg, &buts, sizeof(buts))) { |
2967acbb | 594 | __blk_trace_remove(q); |
171044d4 | 595 | return -EFAULT; |
9a8c28c8 | 596 | } |
171044d4 AB |
597 | return 0; |
598 | } | |
1f2cac10 JA |
599 | |
600 | int blk_trace_setup(struct request_queue *q, char *name, dev_t dev, | |
601 | struct block_device *bdev, | |
602 | char __user *arg) | |
603 | { | |
604 | int ret; | |
605 | ||
85e0cbbb | 606 | mutex_lock(&q->debugfs_mutex); |
1f2cac10 | 607 | ret = __blk_trace_setup(q, name, dev, bdev, arg); |
85e0cbbb | 608 | mutex_unlock(&q->debugfs_mutex); |
1f2cac10 JA |
609 | |
610 | return ret; | |
611 | } | |
6da127ad | 612 | EXPORT_SYMBOL_GPL(blk_trace_setup); |
2056a782 | 613 | |
62c2a7d9 AB |
614 | #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64) |
615 | static int compat_blk_trace_setup(struct request_queue *q, char *name, | |
616 | dev_t dev, struct block_device *bdev, | |
617 | char __user *arg) | |
618 | { | |
619 | struct blk_user_trace_setup buts; | |
620 | struct compat_blk_user_trace_setup cbuts; | |
621 | int ret; | |
622 | ||
623 | if (copy_from_user(&cbuts, arg, sizeof(cbuts))) | |
624 | return -EFAULT; | |
625 | ||
626 | buts = (struct blk_user_trace_setup) { | |
627 | .act_mask = cbuts.act_mask, | |
628 | .buf_size = cbuts.buf_size, | |
629 | .buf_nr = cbuts.buf_nr, | |
630 | .start_lba = cbuts.start_lba, | |
631 | .end_lba = cbuts.end_lba, | |
632 | .pid = cbuts.pid, | |
633 | }; | |
62c2a7d9 AB |
634 | |
635 | ret = do_blk_trace_setup(q, name, dev, bdev, &buts); | |
636 | if (ret) | |
637 | return ret; | |
638 | ||
f8c5e944 | 639 | if (copy_to_user(arg, &buts.name, ARRAY_SIZE(buts.name))) { |
2967acbb | 640 | __blk_trace_remove(q); |
62c2a7d9 AB |
641 | return -EFAULT; |
642 | } | |
643 | ||
644 | return 0; | |
645 | } | |
646 | #endif | |
647 | ||
1f2cac10 | 648 | static int __blk_trace_startstop(struct request_queue *q, int start) |
2056a782 | 649 | { |
2056a782 | 650 | int ret; |
c780e86d | 651 | struct blk_trace *bt; |
2056a782 | 652 | |
c780e86d | 653 | bt = rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 654 | lockdep_is_held(&q->debugfs_mutex)); |
939b3669 | 655 | if (bt == NULL) |
2056a782 JA |
656 | return -EINVAL; |
657 | ||
658 | /* | |
659 | * For starting a trace, we can transition from a setup or stopped | |
660 | * trace. For stopping a trace, the state must be running | |
661 | */ | |
662 | ret = -EINVAL; | |
663 | if (start) { | |
664 | if (bt->trace_state == Blktrace_setup || | |
665 | bt->trace_state == Blktrace_stopped) { | |
666 | blktrace_seq++; | |
667 | smp_mb(); | |
668 | bt->trace_state = Blktrace_running; | |
a404d557 JK |
669 | spin_lock_irq(&running_trace_lock); |
670 | list_add(&bt->running_list, &running_trace_list); | |
671 | spin_unlock_irq(&running_trace_lock); | |
be1c6341 OK |
672 | |
673 | trace_note_time(bt); | |
2056a782 JA |
674 | ret = 0; |
675 | } | |
676 | } else { | |
677 | if (bt->trace_state == Blktrace_running) { | |
678 | bt->trace_state = Blktrace_stopped; | |
a404d557 JK |
679 | spin_lock_irq(&running_trace_lock); |
680 | list_del_init(&bt->running_list); | |
681 | spin_unlock_irq(&running_trace_lock); | |
2056a782 JA |
682 | relay_flush(bt->rchan); |
683 | ret = 0; | |
684 | } | |
685 | } | |
686 | ||
687 | return ret; | |
688 | } | |
1f2cac10 JA |
689 | |
690 | int blk_trace_startstop(struct request_queue *q, int start) | |
691 | { | |
692 | int ret; | |
693 | ||
85e0cbbb | 694 | mutex_lock(&q->debugfs_mutex); |
1f2cac10 | 695 | ret = __blk_trace_startstop(q, start); |
85e0cbbb | 696 | mutex_unlock(&q->debugfs_mutex); |
1f2cac10 JA |
697 | |
698 | return ret; | |
699 | } | |
6da127ad | 700 | EXPORT_SYMBOL_GPL(blk_trace_startstop); |
2056a782 | 701 | |
5acb3cc2 WL |
702 | /* |
703 | * When reading or writing the blktrace sysfs files, the references to the | |
704 | * opened sysfs or device files should prevent the underlying block device | |
705 | * from being removed. So no further delete protection is really needed. | |
706 | */ | |
707 | ||
2056a782 JA |
708 | /** |
709 | * blk_trace_ioctl: - handle the ioctls associated with tracing | |
710 | * @bdev: the block device | |
ef18012b | 711 | * @cmd: the ioctl cmd |
2056a782 JA |
712 | * @arg: the argument data, if any |
713 | * | |
714 | **/ | |
715 | int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg) | |
716 | { | |
165125e1 | 717 | struct request_queue *q; |
2056a782 | 718 | int ret, start = 0; |
6da127ad | 719 | char b[BDEVNAME_SIZE]; |
2056a782 JA |
720 | |
721 | q = bdev_get_queue(bdev); | |
722 | if (!q) | |
723 | return -ENXIO; | |
724 | ||
85e0cbbb | 725 | mutex_lock(&q->debugfs_mutex); |
2056a782 JA |
726 | |
727 | switch (cmd) { | |
728 | case BLKTRACESETUP: | |
f36f21ec | 729 | bdevname(bdev, b); |
1f2cac10 | 730 | ret = __blk_trace_setup(q, b, bdev->bd_dev, bdev, arg); |
2056a782 | 731 | break; |
62c2a7d9 AB |
732 | #if defined(CONFIG_COMPAT) && defined(CONFIG_X86_64) |
733 | case BLKTRACESETUP32: | |
734 | bdevname(bdev, b); | |
735 | ret = compat_blk_trace_setup(q, b, bdev->bd_dev, bdev, arg); | |
736 | break; | |
737 | #endif | |
2056a782 JA |
738 | case BLKTRACESTART: |
739 | start = 1; | |
df561f66 | 740 | fallthrough; |
2056a782 | 741 | case BLKTRACESTOP: |
1f2cac10 | 742 | ret = __blk_trace_startstop(q, start); |
2056a782 JA |
743 | break; |
744 | case BLKTRACETEARDOWN: | |
1f2cac10 | 745 | ret = __blk_trace_remove(q); |
2056a782 JA |
746 | break; |
747 | default: | |
748 | ret = -ENOTTY; | |
749 | break; | |
750 | } | |
751 | ||
85e0cbbb | 752 | mutex_unlock(&q->debugfs_mutex); |
2056a782 JA |
753 | return ret; |
754 | } | |
755 | ||
756 | /** | |
757 | * blk_trace_shutdown: - stop and cleanup trace structures | |
758 | * @q: the request queue associated with the device | |
759 | * | |
760 | **/ | |
165125e1 | 761 | void blk_trace_shutdown(struct request_queue *q) |
2056a782 | 762 | { |
85e0cbbb | 763 | mutex_lock(&q->debugfs_mutex); |
c780e86d | 764 | if (rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 765 | lockdep_is_held(&q->debugfs_mutex))) { |
1f2cac10 JA |
766 | __blk_trace_startstop(q, 0); |
767 | __blk_trace_remove(q); | |
6c5c9341 | 768 | } |
1f2cac10 | 769 | |
85e0cbbb | 770 | mutex_unlock(&q->debugfs_mutex); |
2056a782 | 771 | } |
5f3ea37c | 772 | |
ca1136c9 | 773 | #ifdef CONFIG_BLK_CGROUP |
67c0496e | 774 | static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio) |
ca1136c9 | 775 | { |
c780e86d | 776 | struct blk_trace *bt; |
ca1136c9 | 777 | |
c780e86d JK |
778 | /* We don't use the 'bt' value here except as an optimization... */ |
779 | bt = rcu_dereference_protected(q->blk_trace, 1); | |
ca1136c9 | 780 | if (!bt || !(blk_tracer_flags.val & TRACE_BLK_OPT_CGROUP)) |
67c0496e | 781 | return 0; |
ca1136c9 | 782 | |
db6638d7 | 783 | if (!bio->bi_blkg) |
67c0496e | 784 | return 0; |
74321038 | 785 | return cgroup_id(bio_blkcg(bio)->css.cgroup); |
ca1136c9 SL |
786 | } |
787 | #else | |
e75ad2cc | 788 | static u64 blk_trace_bio_get_cgid(struct request_queue *q, struct bio *bio) |
ca1136c9 | 789 | { |
67c0496e | 790 | return 0; |
ca1136c9 SL |
791 | } |
792 | #endif | |
793 | ||
67c0496e | 794 | static u64 |
a54895fa | 795 | blk_trace_request_get_cgid(struct request *rq) |
ca1136c9 SL |
796 | { |
797 | if (!rq->bio) | |
67c0496e | 798 | return 0; |
ca1136c9 | 799 | /* Use the first bio */ |
a54895fa | 800 | return blk_trace_bio_get_cgid(rq->q, rq->bio); |
ca1136c9 SL |
801 | } |
802 | ||
5f3ea37c ACM |
803 | /* |
804 | * blktrace probes | |
805 | */ | |
806 | ||
807 | /** | |
808 | * blk_add_trace_rq - Add a trace for a request oriented action | |
5f3ea37c | 809 | * @rq: the source request |
caf7df12 | 810 | * @error: return status to log |
af5040da | 811 | * @nr_bytes: number of completed bytes |
5f3ea37c | 812 | * @what: the action |
ca1136c9 | 813 | * @cgid: the cgroup info |
5f3ea37c ACM |
814 | * |
815 | * Description: | |
816 | * Records an action against a request. Will log the bio offset + size. | |
817 | * | |
818 | **/ | |
caf7df12 | 819 | static void blk_add_trace_rq(struct request *rq, int error, |
67c0496e | 820 | unsigned int nr_bytes, u32 what, u64 cgid) |
5f3ea37c | 821 | { |
c780e86d | 822 | struct blk_trace *bt; |
5f3ea37c | 823 | |
c780e86d JK |
824 | rcu_read_lock(); |
825 | bt = rcu_dereference(rq->q->blk_trace); | |
826 | if (likely(!bt)) { | |
827 | rcu_read_unlock(); | |
5f3ea37c | 828 | return; |
c780e86d | 829 | } |
5f3ea37c | 830 | |
57292b58 | 831 | if (blk_rq_is_passthrough(rq)) |
5f3ea37c | 832 | what |= BLK_TC_ACT(BLK_TC_PC); |
48b77ad6 | 833 | else |
5f3ea37c | 834 | what |= BLK_TC_ACT(BLK_TC_FS); |
48b77ad6 CH |
835 | |
836 | __blk_add_trace(bt, blk_rq_trace_sector(rq), nr_bytes, req_op(rq), | |
ca1136c9 | 837 | rq->cmd_flags, what, error, 0, NULL, cgid); |
c780e86d | 838 | rcu_read_unlock(); |
5f3ea37c ACM |
839 | } |
840 | ||
a54895fa | 841 | static void blk_add_trace_rq_insert(void *ignore, struct request *rq) |
5f3ea37c | 842 | { |
ca1136c9 | 843 | blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_INSERT, |
a54895fa | 844 | blk_trace_request_get_cgid(rq)); |
5f3ea37c ACM |
845 | } |
846 | ||
a54895fa | 847 | static void blk_add_trace_rq_issue(void *ignore, struct request *rq) |
5f3ea37c | 848 | { |
ca1136c9 | 849 | blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_ISSUE, |
a54895fa | 850 | blk_trace_request_get_cgid(rq)); |
5f3ea37c ACM |
851 | } |
852 | ||
a54895fa | 853 | static void blk_add_trace_rq_merge(void *ignore, struct request *rq) |
f3bdc62f JK |
854 | { |
855 | blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_BACKMERGE, | |
a54895fa | 856 | blk_trace_request_get_cgid(rq)); |
f3bdc62f JK |
857 | } |
858 | ||
a54895fa | 859 | static void blk_add_trace_rq_requeue(void *ignore, struct request *rq) |
5f3ea37c | 860 | { |
ca1136c9 | 861 | blk_add_trace_rq(rq, 0, blk_rq_bytes(rq), BLK_TA_REQUEUE, |
a54895fa | 862 | blk_trace_request_get_cgid(rq)); |
5f3ea37c ACM |
863 | } |
864 | ||
caf7df12 CH |
865 | static void blk_add_trace_rq_complete(void *ignore, struct request *rq, |
866 | int error, unsigned int nr_bytes) | |
5f3ea37c | 867 | { |
ca1136c9 | 868 | blk_add_trace_rq(rq, error, nr_bytes, BLK_TA_COMPLETE, |
a54895fa | 869 | blk_trace_request_get_cgid(rq)); |
5f3ea37c ACM |
870 | } |
871 | ||
872 | /** | |
873 | * blk_add_trace_bio - Add a trace for a bio oriented action | |
874 | * @q: queue the io is for | |
875 | * @bio: the source bio | |
876 | * @what: the action | |
797a455d | 877 | * @error: error, if any |
5f3ea37c ACM |
878 | * |
879 | * Description: | |
880 | * Records an action against a bio. Will log the bio offset + size. | |
881 | * | |
882 | **/ | |
883 | static void blk_add_trace_bio(struct request_queue *q, struct bio *bio, | |
1690102d | 884 | u32 what, int error) |
5f3ea37c | 885 | { |
c780e86d | 886 | struct blk_trace *bt; |
5f3ea37c | 887 | |
c780e86d JK |
888 | rcu_read_lock(); |
889 | bt = rcu_dereference(q->blk_trace); | |
890 | if (likely(!bt)) { | |
891 | rcu_read_unlock(); | |
5f3ea37c | 892 | return; |
c780e86d | 893 | } |
5f3ea37c | 894 | |
4f024f37 | 895 | __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size, |
1690102d MPS |
896 | bio_op(bio), bio->bi_opf, what, error, 0, NULL, |
897 | blk_trace_bio_get_cgid(q, bio)); | |
c780e86d | 898 | rcu_read_unlock(); |
5f3ea37c ACM |
899 | } |
900 | ||
e8a676d6 | 901 | static void blk_add_trace_bio_bounce(void *ignore, struct bio *bio) |
5f3ea37c | 902 | { |
309dca30 | 903 | blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BOUNCE, 0); |
5f3ea37c ACM |
904 | } |
905 | ||
0a82a8d1 | 906 | static void blk_add_trace_bio_complete(void *ignore, |
d24de76a | 907 | struct request_queue *q, struct bio *bio) |
5f3ea37c | 908 | { |
d24de76a CH |
909 | blk_add_trace_bio(q, bio, BLK_TA_COMPLETE, |
910 | blk_status_to_errno(bio->bi_status)); | |
5f3ea37c ACM |
911 | } |
912 | ||
e8a676d6 | 913 | static void blk_add_trace_bio_backmerge(void *ignore, struct bio *bio) |
5f3ea37c | 914 | { |
309dca30 CH |
915 | blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_BACKMERGE, |
916 | 0); | |
5f3ea37c ACM |
917 | } |
918 | ||
e8a676d6 | 919 | static void blk_add_trace_bio_frontmerge(void *ignore, struct bio *bio) |
5f3ea37c | 920 | { |
309dca30 CH |
921 | blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_FRONTMERGE, |
922 | 0); | |
5f3ea37c ACM |
923 | } |
924 | ||
e8a676d6 | 925 | static void blk_add_trace_bio_queue(void *ignore, struct bio *bio) |
5f3ea37c | 926 | { |
309dca30 | 927 | blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_QUEUE, 0); |
5f3ea37c ACM |
928 | } |
929 | ||
e8a676d6 | 930 | static void blk_add_trace_getrq(void *ignore, struct bio *bio) |
5f3ea37c | 931 | { |
309dca30 | 932 | blk_add_trace_bio(bio->bi_bdev->bd_disk->queue, bio, BLK_TA_GETRQ, 0); |
5f3ea37c ACM |
933 | } |
934 | ||
38516ab5 | 935 | static void blk_add_trace_plug(void *ignore, struct request_queue *q) |
5f3ea37c | 936 | { |
c780e86d | 937 | struct blk_trace *bt; |
5f3ea37c | 938 | |
c780e86d JK |
939 | rcu_read_lock(); |
940 | bt = rcu_dereference(q->blk_trace); | |
5f3ea37c | 941 | if (bt) |
67c0496e | 942 | __blk_add_trace(bt, 0, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL, 0); |
c780e86d | 943 | rcu_read_unlock(); |
5f3ea37c ACM |
944 | } |
945 | ||
49cac01e JA |
946 | static void blk_add_trace_unplug(void *ignore, struct request_queue *q, |
947 | unsigned int depth, bool explicit) | |
5f3ea37c | 948 | { |
c780e86d | 949 | struct blk_trace *bt; |
5f3ea37c | 950 | |
c780e86d JK |
951 | rcu_read_lock(); |
952 | bt = rcu_dereference(q->blk_trace); | |
5f3ea37c | 953 | if (bt) { |
94b5eb28 | 954 | __be64 rpdu = cpu_to_be64(depth); |
49cac01e | 955 | u32 what; |
5f3ea37c | 956 | |
49cac01e JA |
957 | if (explicit) |
958 | what = BLK_TA_UNPLUG_IO; | |
959 | else | |
960 | what = BLK_TA_UNPLUG_TIMER; | |
961 | ||
67c0496e | 962 | __blk_add_trace(bt, 0, 0, 0, 0, what, 0, sizeof(rpdu), &rpdu, 0); |
5f3ea37c | 963 | } |
c780e86d | 964 | rcu_read_unlock(); |
5f3ea37c ACM |
965 | } |
966 | ||
eb6f7f7c | 967 | static void blk_add_trace_split(void *ignore, struct bio *bio, unsigned int pdu) |
5f3ea37c | 968 | { |
309dca30 | 969 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
c780e86d | 970 | struct blk_trace *bt; |
5f3ea37c | 971 | |
c780e86d JK |
972 | rcu_read_lock(); |
973 | bt = rcu_dereference(q->blk_trace); | |
5f3ea37c ACM |
974 | if (bt) { |
975 | __be64 rpdu = cpu_to_be64(pdu); | |
976 | ||
4f024f37 | 977 | __blk_add_trace(bt, bio->bi_iter.bi_sector, |
1eff9d32 | 978 | bio->bi_iter.bi_size, bio_op(bio), bio->bi_opf, |
48bc3cd3 CK |
979 | BLK_TA_SPLIT, |
980 | blk_status_to_errno(bio->bi_status), | |
981 | sizeof(rpdu), &rpdu, | |
982 | blk_trace_bio_get_cgid(q, bio)); | |
5f3ea37c | 983 | } |
c780e86d | 984 | rcu_read_unlock(); |
5f3ea37c ACM |
985 | } |
986 | ||
987 | /** | |
d07335e5 | 988 | * blk_add_trace_bio_remap - Add a trace for a bio-remap operation |
546cf44a | 989 | * @ignore: trace callback data parameter (not used) |
5f3ea37c | 990 | * @bio: the source bio |
1c02fca6 | 991 | * @dev: source device |
a42aaa3b | 992 | * @from: source sector |
5f3ea37c | 993 | * |
1c02fca6 | 994 | * Called after a bio is remapped to a different device and/or sector. |
5f3ea37c | 995 | **/ |
1c02fca6 CH |
996 | static void blk_add_trace_bio_remap(void *ignore, struct bio *bio, dev_t dev, |
997 | sector_t from) | |
5f3ea37c | 998 | { |
309dca30 | 999 | struct request_queue *q = bio->bi_bdev->bd_disk->queue; |
c780e86d | 1000 | struct blk_trace *bt; |
5f3ea37c ACM |
1001 | struct blk_io_trace_remap r; |
1002 | ||
c780e86d JK |
1003 | rcu_read_lock(); |
1004 | bt = rcu_dereference(q->blk_trace); | |
1005 | if (likely(!bt)) { | |
1006 | rcu_read_unlock(); | |
5f3ea37c | 1007 | return; |
c780e86d | 1008 | } |
5f3ea37c | 1009 | |
a42aaa3b | 1010 | r.device_from = cpu_to_be32(dev); |
74d46992 | 1011 | r.device_to = cpu_to_be32(bio_dev(bio)); |
a42aaa3b | 1012 | r.sector_from = cpu_to_be64(from); |
5f3ea37c | 1013 | |
4f024f37 | 1014 | __blk_add_trace(bt, bio->bi_iter.bi_sector, bio->bi_iter.bi_size, |
48bc3cd3 CK |
1015 | bio_op(bio), bio->bi_opf, BLK_TA_REMAP, |
1016 | blk_status_to_errno(bio->bi_status), | |
ca1136c9 | 1017 | sizeof(r), &r, blk_trace_bio_get_cgid(q, bio)); |
c780e86d | 1018 | rcu_read_unlock(); |
5f3ea37c ACM |
1019 | } |
1020 | ||
b0da3f0d JN |
1021 | /** |
1022 | * blk_add_trace_rq_remap - Add a trace for a request-remap operation | |
546cf44a | 1023 | * @ignore: trace callback data parameter (not used) |
b0da3f0d JN |
1024 | * @rq: the source request |
1025 | * @dev: target device | |
1026 | * @from: source sector | |
1027 | * | |
1028 | * Description: | |
1029 | * Device mapper remaps request to other devices. | |
1030 | * Add a trace for that action. | |
1031 | * | |
1032 | **/ | |
a54895fa | 1033 | static void blk_add_trace_rq_remap(void *ignore, struct request *rq, dev_t dev, |
b0da3f0d JN |
1034 | sector_t from) |
1035 | { | |
c780e86d | 1036 | struct blk_trace *bt; |
b0da3f0d JN |
1037 | struct blk_io_trace_remap r; |
1038 | ||
c780e86d | 1039 | rcu_read_lock(); |
a54895fa | 1040 | bt = rcu_dereference(rq->q->blk_trace); |
c780e86d JK |
1041 | if (likely(!bt)) { |
1042 | rcu_read_unlock(); | |
b0da3f0d | 1043 | return; |
c780e86d | 1044 | } |
b0da3f0d JN |
1045 | |
1046 | r.device_from = cpu_to_be32(dev); | |
1047 | r.device_to = cpu_to_be32(disk_devt(rq->rq_disk)); | |
1048 | r.sector_from = cpu_to_be64(from); | |
1049 | ||
1050 | __blk_add_trace(bt, blk_rq_pos(rq), blk_rq_bytes(rq), | |
caf7df12 | 1051 | rq_data_dir(rq), 0, BLK_TA_REMAP, 0, |
a54895fa | 1052 | sizeof(r), &r, blk_trace_request_get_cgid(rq)); |
c780e86d | 1053 | rcu_read_unlock(); |
b0da3f0d JN |
1054 | } |
1055 | ||
5f3ea37c ACM |
1056 | /** |
1057 | * blk_add_driver_data - Add binary message with driver-specific data | |
5f3ea37c ACM |
1058 | * @rq: io request |
1059 | * @data: driver-specific data | |
1060 | * @len: length of driver-specific data | |
1061 | * | |
1062 | * Description: | |
1063 | * Some drivers might want to write driver-specific data per request. | |
1064 | * | |
1065 | **/ | |
a54895fa | 1066 | void blk_add_driver_data(struct request *rq, void *data, size_t len) |
5f3ea37c | 1067 | { |
c780e86d | 1068 | struct blk_trace *bt; |
5f3ea37c | 1069 | |
c780e86d | 1070 | rcu_read_lock(); |
a54895fa | 1071 | bt = rcu_dereference(rq->q->blk_trace); |
c780e86d JK |
1072 | if (likely(!bt)) { |
1073 | rcu_read_unlock(); | |
5f3ea37c | 1074 | return; |
c780e86d | 1075 | } |
5f3ea37c | 1076 | |
48b77ad6 | 1077 | __blk_add_trace(bt, blk_rq_trace_sector(rq), blk_rq_bytes(rq), 0, 0, |
ca1136c9 | 1078 | BLK_TA_DRV_DATA, 0, len, data, |
a54895fa | 1079 | blk_trace_request_get_cgid(rq)); |
c780e86d | 1080 | rcu_read_unlock(); |
5f3ea37c ACM |
1081 | } |
1082 | EXPORT_SYMBOL_GPL(blk_add_driver_data); | |
1083 | ||
3c289ba7 | 1084 | static void blk_register_tracepoints(void) |
5f3ea37c ACM |
1085 | { |
1086 | int ret; | |
1087 | ||
38516ab5 | 1088 | ret = register_trace_block_rq_insert(blk_add_trace_rq_insert, NULL); |
5f3ea37c | 1089 | WARN_ON(ret); |
38516ab5 | 1090 | ret = register_trace_block_rq_issue(blk_add_trace_rq_issue, NULL); |
5f3ea37c | 1091 | WARN_ON(ret); |
f3bdc62f JK |
1092 | ret = register_trace_block_rq_merge(blk_add_trace_rq_merge, NULL); |
1093 | WARN_ON(ret); | |
38516ab5 | 1094 | ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL); |
5f3ea37c | 1095 | WARN_ON(ret); |
38516ab5 | 1096 | ret = register_trace_block_rq_complete(blk_add_trace_rq_complete, NULL); |
5f3ea37c | 1097 | WARN_ON(ret); |
38516ab5 | 1098 | ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL); |
5f3ea37c | 1099 | WARN_ON(ret); |
38516ab5 | 1100 | ret = register_trace_block_bio_complete(blk_add_trace_bio_complete, NULL); |
5f3ea37c | 1101 | WARN_ON(ret); |
38516ab5 | 1102 | ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL); |
5f3ea37c | 1103 | WARN_ON(ret); |
38516ab5 | 1104 | ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL); |
5f3ea37c | 1105 | WARN_ON(ret); |
38516ab5 | 1106 | ret = register_trace_block_bio_queue(blk_add_trace_bio_queue, NULL); |
5f3ea37c | 1107 | WARN_ON(ret); |
38516ab5 | 1108 | ret = register_trace_block_getrq(blk_add_trace_getrq, NULL); |
5f3ea37c | 1109 | WARN_ON(ret); |
38516ab5 | 1110 | ret = register_trace_block_plug(blk_add_trace_plug, NULL); |
5f3ea37c | 1111 | WARN_ON(ret); |
49cac01e | 1112 | ret = register_trace_block_unplug(blk_add_trace_unplug, NULL); |
5f3ea37c | 1113 | WARN_ON(ret); |
38516ab5 | 1114 | ret = register_trace_block_split(blk_add_trace_split, NULL); |
5f3ea37c | 1115 | WARN_ON(ret); |
d07335e5 | 1116 | ret = register_trace_block_bio_remap(blk_add_trace_bio_remap, NULL); |
5f3ea37c | 1117 | WARN_ON(ret); |
38516ab5 | 1118 | ret = register_trace_block_rq_remap(blk_add_trace_rq_remap, NULL); |
b0da3f0d | 1119 | WARN_ON(ret); |
5f3ea37c ACM |
1120 | } |
1121 | ||
1122 | static void blk_unregister_tracepoints(void) | |
1123 | { | |
38516ab5 | 1124 | unregister_trace_block_rq_remap(blk_add_trace_rq_remap, NULL); |
d07335e5 | 1125 | unregister_trace_block_bio_remap(blk_add_trace_bio_remap, NULL); |
38516ab5 | 1126 | unregister_trace_block_split(blk_add_trace_split, NULL); |
49cac01e | 1127 | unregister_trace_block_unplug(blk_add_trace_unplug, NULL); |
38516ab5 | 1128 | unregister_trace_block_plug(blk_add_trace_plug, NULL); |
38516ab5 SR |
1129 | unregister_trace_block_getrq(blk_add_trace_getrq, NULL); |
1130 | unregister_trace_block_bio_queue(blk_add_trace_bio_queue, NULL); | |
1131 | unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge, NULL); | |
1132 | unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge, NULL); | |
1133 | unregister_trace_block_bio_complete(blk_add_trace_bio_complete, NULL); | |
1134 | unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce, NULL); | |
1135 | unregister_trace_block_rq_complete(blk_add_trace_rq_complete, NULL); | |
1136 | unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue, NULL); | |
f3bdc62f | 1137 | unregister_trace_block_rq_merge(blk_add_trace_rq_merge, NULL); |
38516ab5 SR |
1138 | unregister_trace_block_rq_issue(blk_add_trace_rq_issue, NULL); |
1139 | unregister_trace_block_rq_insert(blk_add_trace_rq_insert, NULL); | |
5f3ea37c ACM |
1140 | |
1141 | tracepoint_synchronize_unregister(); | |
1142 | } | |
c71a8961 ACM |
1143 | |
1144 | /* | |
1145 | * struct blk_io_tracer formatting routines | |
1146 | */ | |
1147 | ||
1148 | static void fill_rwbs(char *rwbs, const struct blk_io_trace *t) | |
1149 | { | |
157f9c00 | 1150 | int i = 0; |
65796348 | 1151 | int tc = t->action >> BLK_TC_SHIFT; |
157f9c00 | 1152 | |
ca1136c9 | 1153 | if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) { |
18cea459 LZ |
1154 | rwbs[i++] = 'N'; |
1155 | goto out; | |
1156 | } | |
1157 | ||
c09c47ca NK |
1158 | if (tc & BLK_TC_FLUSH) |
1159 | rwbs[i++] = 'F'; | |
1160 | ||
65796348 | 1161 | if (tc & BLK_TC_DISCARD) |
157f9c00 | 1162 | rwbs[i++] = 'D'; |
65796348 | 1163 | else if (tc & BLK_TC_WRITE) |
157f9c00 ACM |
1164 | rwbs[i++] = 'W'; |
1165 | else if (t->bytes) | |
1166 | rwbs[i++] = 'R'; | |
1167 | else | |
1168 | rwbs[i++] = 'N'; | |
1169 | ||
c09c47ca NK |
1170 | if (tc & BLK_TC_FUA) |
1171 | rwbs[i++] = 'F'; | |
65796348 | 1172 | if (tc & BLK_TC_AHEAD) |
157f9c00 | 1173 | rwbs[i++] = 'A'; |
65796348 | 1174 | if (tc & BLK_TC_SYNC) |
157f9c00 | 1175 | rwbs[i++] = 'S'; |
65796348 | 1176 | if (tc & BLK_TC_META) |
157f9c00 | 1177 | rwbs[i++] = 'M'; |
18cea459 | 1178 | out: |
157f9c00 | 1179 | rwbs[i] = '\0'; |
c71a8961 ACM |
1180 | } |
1181 | ||
1182 | static inline | |
1183 | const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent) | |
1184 | { | |
1185 | return (const struct blk_io_trace *)ent; | |
1186 | } | |
1187 | ||
ca1136c9 SL |
1188 | static inline const void *pdu_start(const struct trace_entry *ent, bool has_cg) |
1189 | { | |
67c0496e | 1190 | return (void *)(te_blk_io_trace(ent) + 1) + (has_cg ? sizeof(u64) : 0); |
ca1136c9 SL |
1191 | } |
1192 | ||
67c0496e | 1193 | static inline u64 t_cgid(const struct trace_entry *ent) |
c71a8961 | 1194 | { |
67c0496e | 1195 | return *(u64 *)(te_blk_io_trace(ent) + 1); |
ca1136c9 SL |
1196 | } |
1197 | ||
1198 | static inline int pdu_real_len(const struct trace_entry *ent, bool has_cg) | |
1199 | { | |
67c0496e | 1200 | return te_blk_io_trace(ent)->pdu_len - (has_cg ? sizeof(u64) : 0); |
c71a8961 ACM |
1201 | } |
1202 | ||
66de7792 LZ |
1203 | static inline u32 t_action(const struct trace_entry *ent) |
1204 | { | |
1205 | return te_blk_io_trace(ent)->action; | |
1206 | } | |
1207 | ||
1208 | static inline u32 t_bytes(const struct trace_entry *ent) | |
1209 | { | |
1210 | return te_blk_io_trace(ent)->bytes; | |
1211 | } | |
1212 | ||
c71a8961 ACM |
1213 | static inline u32 t_sec(const struct trace_entry *ent) |
1214 | { | |
1215 | return te_blk_io_trace(ent)->bytes >> 9; | |
1216 | } | |
1217 | ||
1218 | static inline unsigned long long t_sector(const struct trace_entry *ent) | |
1219 | { | |
1220 | return te_blk_io_trace(ent)->sector; | |
1221 | } | |
1222 | ||
1223 | static inline __u16 t_error(const struct trace_entry *ent) | |
1224 | { | |
e0dc81be | 1225 | return te_blk_io_trace(ent)->error; |
c71a8961 ACM |
1226 | } |
1227 | ||
ca1136c9 | 1228 | static __u64 get_pdu_int(const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1229 | { |
71df3fd8 | 1230 | const __be64 *val = pdu_start(ent, has_cg); |
c71a8961 ACM |
1231 | return be64_to_cpu(*val); |
1232 | } | |
1233 | ||
ca1136c9 SL |
1234 | typedef void (blk_log_action_t) (struct trace_iterator *iter, const char *act, |
1235 | bool has_cg); | |
b6a4b0c3 | 1236 | |
ca1136c9 SL |
1237 | static void blk_log_action_classic(struct trace_iterator *iter, const char *act, |
1238 | bool has_cg) | |
c71a8961 | 1239 | { |
c09c47ca | 1240 | char rwbs[RWBS_LEN]; |
35ac51bf LZ |
1241 | unsigned long long ts = iter->ts; |
1242 | unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC); | |
c71a8961 | 1243 | unsigned secs = (unsigned long)ts; |
b6a4b0c3 | 1244 | const struct blk_io_trace *t = te_blk_io_trace(iter->ent); |
c71a8961 ACM |
1245 | |
1246 | fill_rwbs(rwbs, t); | |
1247 | ||
f4a1d08c SRRH |
1248 | trace_seq_printf(&iter->seq, |
1249 | "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ", | |
1250 | MAJOR(t->device), MINOR(t->device), iter->cpu, | |
1251 | secs, nsec_rem, iter->ent->pid, act, rwbs); | |
c71a8961 ACM |
1252 | } |
1253 | ||
ca1136c9 SL |
1254 | static void blk_log_action(struct trace_iterator *iter, const char *act, |
1255 | bool has_cg) | |
c71a8961 | 1256 | { |
c09c47ca | 1257 | char rwbs[RWBS_LEN]; |
b6a4b0c3 LZ |
1258 | const struct blk_io_trace *t = te_blk_io_trace(iter->ent); |
1259 | ||
c71a8961 | 1260 | fill_rwbs(rwbs, t); |
ca1136c9 | 1261 | if (has_cg) { |
67c0496e | 1262 | u64 id = t_cgid(iter->ent); |
ca1136c9 | 1263 | |
69fd5c39 SL |
1264 | if (blk_tracer_flags.val & TRACE_BLK_OPT_CGNAME) { |
1265 | char blkcg_name_buf[NAME_MAX + 1] = "<...>"; | |
1266 | ||
1267 | cgroup_path_from_kernfs_id(id, blkcg_name_buf, | |
1268 | sizeof(blkcg_name_buf)); | |
1269 | trace_seq_printf(&iter->seq, "%3d,%-3d %s %2s %3s ", | |
1270 | MAJOR(t->device), MINOR(t->device), | |
1271 | blkcg_name_buf, act, rwbs); | |
40430452 TH |
1272 | } else { |
1273 | /* | |
1274 | * The cgid portion used to be "INO,GEN". Userland | |
1275 | * builds a FILEID_INO32_GEN fid out of them and | |
1276 | * opens the cgroup using open_by_handle_at(2). | |
1277 | * While 32bit ino setups are still the same, 64bit | |
1278 | * ones now use the 64bit ino as the whole ID and | |
1279 | * no longer use generation. | |
1280 | * | |
2b5894cc | 1281 | * Regardless of the content, always output |
40430452 TH |
1282 | * "LOW32,HIGH32" so that FILEID_INO32_GEN fid can |
1283 | * be mapped back to @id on both 64 and 32bit ino | |
1284 | * setups. See __kernfs_fh_to_dentry(). | |
1285 | */ | |
69fd5c39 | 1286 | trace_seq_printf(&iter->seq, |
40430452 | 1287 | "%3d,%-3d %llx,%-llx %2s %3s ", |
ca1136c9 | 1288 | MAJOR(t->device), MINOR(t->device), |
40430452 TH |
1289 | id & U32_MAX, id >> 32, act, rwbs); |
1290 | } | |
ca1136c9 SL |
1291 | } else |
1292 | trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ", | |
1293 | MAJOR(t->device), MINOR(t->device), act, rwbs); | |
c71a8961 ACM |
1294 | } |
1295 | ||
ca1136c9 SL |
1296 | static void blk_log_dump_pdu(struct trace_seq *s, |
1297 | const struct trace_entry *ent, bool has_cg) | |
66de7792 | 1298 | { |
04986257 | 1299 | const unsigned char *pdu_buf; |
66de7792 | 1300 | int pdu_len; |
f4a1d08c | 1301 | int i, end; |
66de7792 | 1302 | |
ca1136c9 SL |
1303 | pdu_buf = pdu_start(ent, has_cg); |
1304 | pdu_len = pdu_real_len(ent, has_cg); | |
66de7792 LZ |
1305 | |
1306 | if (!pdu_len) | |
f4a1d08c | 1307 | return; |
66de7792 LZ |
1308 | |
1309 | /* find the last zero that needs to be printed */ | |
1310 | for (end = pdu_len - 1; end >= 0; end--) | |
1311 | if (pdu_buf[end]) | |
1312 | break; | |
1313 | end++; | |
1314 | ||
f4a1d08c | 1315 | trace_seq_putc(s, '('); |
66de7792 LZ |
1316 | |
1317 | for (i = 0; i < pdu_len; i++) { | |
1318 | ||
f4a1d08c SRRH |
1319 | trace_seq_printf(s, "%s%02x", |
1320 | i == 0 ? "" : " ", pdu_buf[i]); | |
66de7792 LZ |
1321 | |
1322 | /* | |
2b5894cc | 1323 | * stop when the rest is just zeros and indicate so |
66de7792 LZ |
1324 | * with a ".." appended |
1325 | */ | |
f4a1d08c SRRH |
1326 | if (i == end && end != pdu_len - 1) { |
1327 | trace_seq_puts(s, " ..) "); | |
1328 | return; | |
1329 | } | |
66de7792 LZ |
1330 | } |
1331 | ||
f4a1d08c | 1332 | trace_seq_puts(s, ") "); |
66de7792 LZ |
1333 | } |
1334 | ||
ca1136c9 | 1335 | static void blk_log_generic(struct trace_seq *s, const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1336 | { |
4ca53085 SR |
1337 | char cmd[TASK_COMM_LEN]; |
1338 | ||
1339 | trace_find_cmdline(ent->pid, cmd); | |
c71a8961 | 1340 | |
66de7792 | 1341 | if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { |
f4a1d08c | 1342 | trace_seq_printf(s, "%u ", t_bytes(ent)); |
ca1136c9 | 1343 | blk_log_dump_pdu(s, ent, has_cg); |
f4a1d08c | 1344 | trace_seq_printf(s, "[%s]\n", cmd); |
66de7792 LZ |
1345 | } else { |
1346 | if (t_sec(ent)) | |
f4a1d08c | 1347 | trace_seq_printf(s, "%llu + %u [%s]\n", |
66de7792 | 1348 | t_sector(ent), t_sec(ent), cmd); |
f4a1d08c SRRH |
1349 | else |
1350 | trace_seq_printf(s, "[%s]\n", cmd); | |
66de7792 | 1351 | } |
c71a8961 ACM |
1352 | } |
1353 | ||
f4a1d08c | 1354 | static void blk_log_with_error(struct trace_seq *s, |
ca1136c9 | 1355 | const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1356 | { |
66de7792 | 1357 | if (t_action(ent) & BLK_TC_ACT(BLK_TC_PC)) { |
ca1136c9 | 1358 | blk_log_dump_pdu(s, ent, has_cg); |
f4a1d08c | 1359 | trace_seq_printf(s, "[%d]\n", t_error(ent)); |
66de7792 LZ |
1360 | } else { |
1361 | if (t_sec(ent)) | |
f4a1d08c SRRH |
1362 | trace_seq_printf(s, "%llu + %u [%d]\n", |
1363 | t_sector(ent), | |
1364 | t_sec(ent), t_error(ent)); | |
1365 | else | |
1366 | trace_seq_printf(s, "%llu [%d]\n", | |
1367 | t_sector(ent), t_error(ent)); | |
66de7792 | 1368 | } |
c71a8961 ACM |
1369 | } |
1370 | ||
ca1136c9 | 1371 | static void blk_log_remap(struct trace_seq *s, const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1372 | { |
5aec598c | 1373 | const struct blk_io_trace_remap *__r = pdu_start(ent, has_cg); |
c71a8961 | 1374 | |
f4a1d08c SRRH |
1375 | trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n", |
1376 | t_sector(ent), t_sec(ent), | |
5aec598c CK |
1377 | MAJOR(be32_to_cpu(__r->device_from)), |
1378 | MINOR(be32_to_cpu(__r->device_from)), | |
1379 | be64_to_cpu(__r->sector_from)); | |
c71a8961 ACM |
1380 | } |
1381 | ||
ca1136c9 | 1382 | static void blk_log_plug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1383 | { |
4ca53085 SR |
1384 | char cmd[TASK_COMM_LEN]; |
1385 | ||
1386 | trace_find_cmdline(ent->pid, cmd); | |
1387 | ||
f4a1d08c | 1388 | trace_seq_printf(s, "[%s]\n", cmd); |
c71a8961 ACM |
1389 | } |
1390 | ||
ca1136c9 | 1391 | static void blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1392 | { |
4ca53085 SR |
1393 | char cmd[TASK_COMM_LEN]; |
1394 | ||
1395 | trace_find_cmdline(ent->pid, cmd); | |
1396 | ||
ca1136c9 | 1397 | trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent, has_cg)); |
c71a8961 ACM |
1398 | } |
1399 | ||
ca1136c9 | 1400 | static void blk_log_split(struct trace_seq *s, const struct trace_entry *ent, bool has_cg) |
c71a8961 | 1401 | { |
4ca53085 SR |
1402 | char cmd[TASK_COMM_LEN]; |
1403 | ||
1404 | trace_find_cmdline(ent->pid, cmd); | |
1405 | ||
f4a1d08c | 1406 | trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent), |
ca1136c9 | 1407 | get_pdu_int(ent, has_cg), cmd); |
c71a8961 ACM |
1408 | } |
1409 | ||
ca1136c9 SL |
1410 | static void blk_log_msg(struct trace_seq *s, const struct trace_entry *ent, |
1411 | bool has_cg) | |
18cea459 | 1412 | { |
18cea459 | 1413 | |
ca1136c9 SL |
1414 | trace_seq_putmem(s, pdu_start(ent, has_cg), |
1415 | pdu_real_len(ent, has_cg)); | |
f4a1d08c | 1416 | trace_seq_putc(s, '\n'); |
18cea459 LZ |
1417 | } |
1418 | ||
c71a8961 ACM |
1419 | /* |
1420 | * struct tracer operations | |
1421 | */ | |
1422 | ||
1423 | static void blk_tracer_print_header(struct seq_file *m) | |
1424 | { | |
1425 | if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) | |
1426 | return; | |
1427 | seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n" | |
1428 | "# | | | | | |\n"); | |
1429 | } | |
1430 | ||
1431 | static void blk_tracer_start(struct trace_array *tr) | |
1432 | { | |
ad5dd549 | 1433 | blk_tracer_enabled = true; |
c71a8961 ACM |
1434 | } |
1435 | ||
1436 | static int blk_tracer_init(struct trace_array *tr) | |
1437 | { | |
1438 | blk_tr = tr; | |
1439 | blk_tracer_start(tr); | |
c71a8961 ACM |
1440 | return 0; |
1441 | } | |
1442 | ||
1443 | static void blk_tracer_stop(struct trace_array *tr) | |
1444 | { | |
ad5dd549 | 1445 | blk_tracer_enabled = false; |
c71a8961 ACM |
1446 | } |
1447 | ||
1448 | static void blk_tracer_reset(struct trace_array *tr) | |
1449 | { | |
c71a8961 ACM |
1450 | blk_tracer_stop(tr); |
1451 | } | |
1452 | ||
e4955c99 | 1453 | static const struct { |
c71a8961 | 1454 | const char *act[2]; |
ca1136c9 SL |
1455 | void (*print)(struct trace_seq *s, const struct trace_entry *ent, |
1456 | bool has_cg); | |
e4955c99 | 1457 | } what2act[] = { |
ef18012b | 1458 | [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic }, |
c71a8961 ACM |
1459 | [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic }, |
1460 | [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic }, | |
1461 | [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic }, | |
1462 | [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic }, | |
1463 | [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error }, | |
1464 | [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic }, | |
1465 | [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error }, | |
1466 | [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug }, | |
1467 | [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug }, | |
49cac01e | 1468 | [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug }, |
c71a8961 ACM |
1469 | [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic }, |
1470 | [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split }, | |
1471 | [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic }, | |
1472 | [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap }, | |
1473 | }; | |
1474 | ||
b6a4b0c3 LZ |
1475 | static enum print_line_t print_one_line(struct trace_iterator *iter, |
1476 | bool classic) | |
c71a8961 | 1477 | { |
983f938a | 1478 | struct trace_array *tr = iter->tr; |
2c9b238e | 1479 | struct trace_seq *s = &iter->seq; |
b6a4b0c3 LZ |
1480 | const struct blk_io_trace *t; |
1481 | u16 what; | |
b6a4b0c3 LZ |
1482 | bool long_act; |
1483 | blk_log_action_t *log_action; | |
ca1136c9 | 1484 | bool has_cg; |
c71a8961 | 1485 | |
b6a4b0c3 | 1486 | t = te_blk_io_trace(iter->ent); |
ca1136c9 | 1487 | what = (t->action & ((1 << BLK_TC_SHIFT) - 1)) & ~__BLK_TA_CGROUP; |
983f938a | 1488 | long_act = !!(tr->trace_flags & TRACE_ITER_VERBOSE); |
b6a4b0c3 | 1489 | log_action = classic ? &blk_log_action_classic : &blk_log_action; |
ca1136c9 | 1490 | has_cg = t->action & __BLK_TA_CGROUP; |
08a06b83 | 1491 | |
ca1136c9 SL |
1492 | if ((t->action & ~__BLK_TN_CGROUP) == BLK_TN_MESSAGE) { |
1493 | log_action(iter, long_act ? "message" : "m", has_cg); | |
1494 | blk_log_msg(s, iter->ent, has_cg); | |
b7d7641e | 1495 | return trace_handle_return(s); |
18cea459 LZ |
1496 | } |
1497 | ||
eb08f8eb | 1498 | if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act))) |
f4a1d08c | 1499 | trace_seq_printf(s, "Unknown action %x\n", what); |
c71a8961 | 1500 | else { |
ca1136c9 SL |
1501 | log_action(iter, what2act[what].act[long_act], has_cg); |
1502 | what2act[what].print(s, iter->ent, has_cg); | |
c71a8961 | 1503 | } |
f4a1d08c SRRH |
1504 | |
1505 | return trace_handle_return(s); | |
c71a8961 ACM |
1506 | } |
1507 | ||
b6a4b0c3 | 1508 | static enum print_line_t blk_trace_event_print(struct trace_iterator *iter, |
a9a57763 | 1509 | int flags, struct trace_event *event) |
b6a4b0c3 | 1510 | { |
b6a4b0c3 LZ |
1511 | return print_one_line(iter, false); |
1512 | } | |
1513 | ||
f4a1d08c | 1514 | static void blk_trace_synthesize_old_trace(struct trace_iterator *iter) |
08a06b83 ACM |
1515 | { |
1516 | struct trace_seq *s = &iter->seq; | |
1517 | struct blk_io_trace *t = (struct blk_io_trace *)iter->ent; | |
1518 | const int offset = offsetof(struct blk_io_trace, sector); | |
1519 | struct blk_io_trace old = { | |
1520 | .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION, | |
6c051ce0 | 1521 | .time = iter->ts, |
08a06b83 ACM |
1522 | }; |
1523 | ||
f4a1d08c SRRH |
1524 | trace_seq_putmem(s, &old, offset); |
1525 | trace_seq_putmem(s, &t->sector, | |
1526 | sizeof(old) - offset + t->pdu_len); | |
08a06b83 ACM |
1527 | } |
1528 | ||
ae7462b4 | 1529 | static enum print_line_t |
a9a57763 SR |
1530 | blk_trace_event_print_binary(struct trace_iterator *iter, int flags, |
1531 | struct trace_event *event) | |
08a06b83 | 1532 | { |
f4a1d08c SRRH |
1533 | blk_trace_synthesize_old_trace(iter); |
1534 | ||
1535 | return trace_handle_return(&iter->seq); | |
08a06b83 ACM |
1536 | } |
1537 | ||
c71a8961 ACM |
1538 | static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter) |
1539 | { | |
c71a8961 ACM |
1540 | if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC)) |
1541 | return TRACE_TYPE_UNHANDLED; | |
1542 | ||
b6a4b0c3 | 1543 | return print_one_line(iter, true); |
c71a8961 ACM |
1544 | } |
1545 | ||
8c1a49ae SRRH |
1546 | static int |
1547 | blk_tracer_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set) | |
f3948f88 LZ |
1548 | { |
1549 | /* don't output context-info for blk_classic output */ | |
1550 | if (bit == TRACE_BLK_OPT_CLASSIC) { | |
1551 | if (set) | |
983f938a | 1552 | tr->trace_flags &= ~TRACE_ITER_CONTEXT_INFO; |
f3948f88 | 1553 | else |
983f938a | 1554 | tr->trace_flags |= TRACE_ITER_CONTEXT_INFO; |
f3948f88 LZ |
1555 | } |
1556 | return 0; | |
1557 | } | |
1558 | ||
c71a8961 ACM |
1559 | static struct tracer blk_tracer __read_mostly = { |
1560 | .name = "blk", | |
1561 | .init = blk_tracer_init, | |
1562 | .reset = blk_tracer_reset, | |
1563 | .start = blk_tracer_start, | |
1564 | .stop = blk_tracer_stop, | |
1565 | .print_header = blk_tracer_print_header, | |
1566 | .print_line = blk_tracer_print_line, | |
1567 | .flags = &blk_tracer_flags, | |
f3948f88 | 1568 | .set_flag = blk_tracer_set_flag, |
c71a8961 ACM |
1569 | }; |
1570 | ||
a9a57763 | 1571 | static struct trace_event_functions trace_blk_event_funcs = { |
c71a8961 | 1572 | .trace = blk_trace_event_print, |
08a06b83 | 1573 | .binary = blk_trace_event_print_binary, |
c71a8961 ACM |
1574 | }; |
1575 | ||
a9a57763 SR |
1576 | static struct trace_event trace_blk_event = { |
1577 | .type = TRACE_BLK, | |
1578 | .funcs = &trace_blk_event_funcs, | |
1579 | }; | |
1580 | ||
c71a8961 ACM |
1581 | static int __init init_blk_tracer(void) |
1582 | { | |
9023c930 | 1583 | if (!register_trace_event(&trace_blk_event)) { |
a395d6a7 | 1584 | pr_warn("Warning: could not register block events\n"); |
c71a8961 ACM |
1585 | return 1; |
1586 | } | |
1587 | ||
1588 | if (register_tracer(&blk_tracer) != 0) { | |
a395d6a7 | 1589 | pr_warn("Warning: could not register the block tracer\n"); |
9023c930 | 1590 | unregister_trace_event(&trace_blk_event); |
c71a8961 ACM |
1591 | return 1; |
1592 | } | |
1593 | ||
1594 | return 0; | |
1595 | } | |
1596 | ||
1597 | device_initcall(init_blk_tracer); | |
1598 | ||
1599 | static int blk_trace_remove_queue(struct request_queue *q) | |
1600 | { | |
1601 | struct blk_trace *bt; | |
1602 | ||
c3dbe541 | 1603 | bt = rcu_replace_pointer(q->blk_trace, NULL, |
85e0cbbb | 1604 | lockdep_is_held(&q->debugfs_mutex)); |
c71a8961 ACM |
1605 | if (bt == NULL) |
1606 | return -EINVAL; | |
1607 | ||
5afedf67 ZC |
1608 | if (bt->trace_state == Blktrace_running) { |
1609 | bt->trace_state = Blktrace_stopped; | |
1610 | spin_lock_irq(&running_trace_lock); | |
1611 | list_del_init(&bt->running_list); | |
1612 | spin_unlock_irq(&running_trace_lock); | |
1613 | relay_flush(bt->rchan); | |
1614 | } | |
1615 | ||
a6da0024 | 1616 | put_probe_ref(); |
c780e86d | 1617 | synchronize_rcu(); |
ad5dd549 | 1618 | blk_trace_free(bt); |
c71a8961 ACM |
1619 | return 0; |
1620 | } | |
1621 | ||
1622 | /* | |
1623 | * Setup everything required to start tracing | |
1624 | */ | |
9908c309 LZ |
1625 | static int blk_trace_setup_queue(struct request_queue *q, |
1626 | struct block_device *bdev) | |
c71a8961 | 1627 | { |
cdea01b2 | 1628 | struct blk_trace *bt = NULL; |
18cea459 | 1629 | int ret = -ENOMEM; |
c71a8961 | 1630 | |
c71a8961 ACM |
1631 | bt = kzalloc(sizeof(*bt), GFP_KERNEL); |
1632 | if (!bt) | |
15152e44 | 1633 | return -ENOMEM; |
c71a8961 | 1634 | |
18cea459 LZ |
1635 | bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char)); |
1636 | if (!bt->msg_data) | |
1637 | goto free_bt; | |
1638 | ||
9908c309 | 1639 | bt->dev = bdev->bd_dev; |
c71a8961 | 1640 | bt->act_mask = (u16)-1; |
9908c309 LZ |
1641 | |
1642 | blk_trace_setup_lba(bt, bdev); | |
c71a8961 | 1643 | |
c3dbe541 | 1644 | rcu_assign_pointer(q->blk_trace, bt); |
a6da0024 | 1645 | get_probe_ref(); |
c71a8961 | 1646 | return 0; |
18cea459 LZ |
1647 | |
1648 | free_bt: | |
1649 | blk_trace_free(bt); | |
1650 | return ret; | |
c71a8961 ACM |
1651 | } |
1652 | ||
1653 | /* | |
1654 | * sysfs interface to enable and configure tracing | |
1655 | */ | |
1656 | ||
c71a8961 ACM |
1657 | static ssize_t sysfs_blk_trace_attr_show(struct device *dev, |
1658 | struct device_attribute *attr, | |
1659 | char *buf); | |
1660 | static ssize_t sysfs_blk_trace_attr_store(struct device *dev, | |
1661 | struct device_attribute *attr, | |
1662 | const char *buf, size_t count); | |
1663 | #define BLK_TRACE_DEVICE_ATTR(_name) \ | |
1664 | DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \ | |
1665 | sysfs_blk_trace_attr_show, \ | |
1666 | sysfs_blk_trace_attr_store) | |
1667 | ||
cd649b8b | 1668 | static BLK_TRACE_DEVICE_ATTR(enable); |
c71a8961 ACM |
1669 | static BLK_TRACE_DEVICE_ATTR(act_mask); |
1670 | static BLK_TRACE_DEVICE_ATTR(pid); | |
1671 | static BLK_TRACE_DEVICE_ATTR(start_lba); | |
1672 | static BLK_TRACE_DEVICE_ATTR(end_lba); | |
1673 | ||
1674 | static struct attribute *blk_trace_attrs[] = { | |
1675 | &dev_attr_enable.attr, | |
1676 | &dev_attr_act_mask.attr, | |
1677 | &dev_attr_pid.attr, | |
1678 | &dev_attr_start_lba.attr, | |
1679 | &dev_attr_end_lba.attr, | |
1680 | NULL | |
1681 | }; | |
1682 | ||
1683 | struct attribute_group blk_trace_attr_group = { | |
1684 | .name = "trace", | |
1685 | .attrs = blk_trace_attrs, | |
1686 | }; | |
1687 | ||
09341997 LZ |
1688 | static const struct { |
1689 | int mask; | |
1690 | const char *str; | |
1691 | } mask_maps[] = { | |
1692 | { BLK_TC_READ, "read" }, | |
1693 | { BLK_TC_WRITE, "write" }, | |
c09c47ca | 1694 | { BLK_TC_FLUSH, "flush" }, |
09341997 LZ |
1695 | { BLK_TC_SYNC, "sync" }, |
1696 | { BLK_TC_QUEUE, "queue" }, | |
1697 | { BLK_TC_REQUEUE, "requeue" }, | |
1698 | { BLK_TC_ISSUE, "issue" }, | |
1699 | { BLK_TC_COMPLETE, "complete" }, | |
1700 | { BLK_TC_FS, "fs" }, | |
1701 | { BLK_TC_PC, "pc" }, | |
8d1547e0 | 1702 | { BLK_TC_NOTIFY, "notify" }, |
09341997 LZ |
1703 | { BLK_TC_AHEAD, "ahead" }, |
1704 | { BLK_TC_META, "meta" }, | |
1705 | { BLK_TC_DISCARD, "discard" }, | |
1706 | { BLK_TC_DRV_DATA, "drv_data" }, | |
c09c47ca | 1707 | { BLK_TC_FUA, "fua" }, |
09341997 LZ |
1708 | }; |
1709 | ||
1710 | static int blk_trace_str2mask(const char *str) | |
c71a8961 | 1711 | { |
09341997 | 1712 | int i; |
c71a8961 | 1713 | int mask = 0; |
9eb85125 | 1714 | char *buf, *s, *token; |
c71a8961 | 1715 | |
9eb85125 LZ |
1716 | buf = kstrdup(str, GFP_KERNEL); |
1717 | if (buf == NULL) | |
c71a8961 | 1718 | return -ENOMEM; |
9eb85125 | 1719 | s = strstrip(buf); |
c71a8961 ACM |
1720 | |
1721 | while (1) { | |
09341997 LZ |
1722 | token = strsep(&s, ","); |
1723 | if (token == NULL) | |
c71a8961 ACM |
1724 | break; |
1725 | ||
09341997 LZ |
1726 | if (*token == '\0') |
1727 | continue; | |
1728 | ||
1729 | for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { | |
1730 | if (strcasecmp(token, mask_maps[i].str) == 0) { | |
1731 | mask |= mask_maps[i].mask; | |
1732 | break; | |
1733 | } | |
1734 | } | |
1735 | if (i == ARRAY_SIZE(mask_maps)) { | |
1736 | mask = -EINVAL; | |
1737 | break; | |
1738 | } | |
c71a8961 | 1739 | } |
9eb85125 | 1740 | kfree(buf); |
c71a8961 ACM |
1741 | |
1742 | return mask; | |
1743 | } | |
1744 | ||
09341997 LZ |
1745 | static ssize_t blk_trace_mask2str(char *buf, int mask) |
1746 | { | |
1747 | int i; | |
1748 | char *p = buf; | |
1749 | ||
1750 | for (i = 0; i < ARRAY_SIZE(mask_maps); i++) { | |
1751 | if (mask & mask_maps[i].mask) { | |
1752 | p += sprintf(p, "%s%s", | |
1753 | (p == buf) ? "" : ",", mask_maps[i].str); | |
1754 | } | |
1755 | } | |
1756 | *p++ = '\n'; | |
1757 | ||
1758 | return p - buf; | |
1759 | } | |
1760 | ||
c71a8961 ACM |
1761 | static ssize_t sysfs_blk_trace_attr_show(struct device *dev, |
1762 | struct device_attribute *attr, | |
1763 | char *buf) | |
1764 | { | |
0d02129e CH |
1765 | struct block_device *bdev = dev_to_bdev(dev); |
1766 | struct request_queue *q = bdev_get_queue(bdev); | |
c780e86d | 1767 | struct blk_trace *bt; |
c71a8961 ACM |
1768 | ssize_t ret = -ENXIO; |
1769 | ||
85e0cbbb | 1770 | mutex_lock(&q->debugfs_mutex); |
cd649b8b | 1771 | |
c780e86d | 1772 | bt = rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 1773 | lockdep_is_held(&q->debugfs_mutex)); |
cd649b8b | 1774 | if (attr == &dev_attr_enable) { |
c780e86d | 1775 | ret = sprintf(buf, "%u\n", !!bt); |
cd649b8b LZ |
1776 | goto out_unlock_bdev; |
1777 | } | |
1778 | ||
c780e86d | 1779 | if (bt == NULL) |
c71a8961 ACM |
1780 | ret = sprintf(buf, "disabled\n"); |
1781 | else if (attr == &dev_attr_act_mask) | |
c780e86d | 1782 | ret = blk_trace_mask2str(buf, bt->act_mask); |
c71a8961 | 1783 | else if (attr == &dev_attr_pid) |
c780e86d | 1784 | ret = sprintf(buf, "%u\n", bt->pid); |
c71a8961 | 1785 | else if (attr == &dev_attr_start_lba) |
c780e86d | 1786 | ret = sprintf(buf, "%llu\n", bt->start_lba); |
c71a8961 | 1787 | else if (attr == &dev_attr_end_lba) |
c780e86d | 1788 | ret = sprintf(buf, "%llu\n", bt->end_lba); |
cd649b8b LZ |
1789 | |
1790 | out_unlock_bdev: | |
85e0cbbb | 1791 | mutex_unlock(&q->debugfs_mutex); |
c71a8961 ACM |
1792 | return ret; |
1793 | } | |
1794 | ||
1795 | static ssize_t sysfs_blk_trace_attr_store(struct device *dev, | |
1796 | struct device_attribute *attr, | |
1797 | const char *buf, size_t count) | |
1798 | { | |
0d02129e CH |
1799 | struct block_device *bdev = dev_to_bdev(dev); |
1800 | struct request_queue *q = bdev_get_queue(bdev); | |
c780e86d | 1801 | struct blk_trace *bt; |
c71a8961 | 1802 | u64 value; |
09341997 | 1803 | ssize_t ret = -EINVAL; |
c71a8961 ACM |
1804 | |
1805 | if (count == 0) | |
1806 | goto out; | |
1807 | ||
1808 | if (attr == &dev_attr_act_mask) { | |
5f339453 | 1809 | if (kstrtoull(buf, 0, &value)) { |
c71a8961 | 1810 | /* Assume it is a list of trace category names */ |
09341997 LZ |
1811 | ret = blk_trace_str2mask(buf); |
1812 | if (ret < 0) | |
c71a8961 | 1813 | goto out; |
09341997 | 1814 | value = ret; |
c71a8961 | 1815 | } |
0d02129e CH |
1816 | } else { |
1817 | if (kstrtoull(buf, 0, &value)) | |
1818 | goto out; | |
1819 | } | |
c71a8961 | 1820 | |
85e0cbbb | 1821 | mutex_lock(&q->debugfs_mutex); |
cd649b8b | 1822 | |
c780e86d | 1823 | bt = rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 1824 | lockdep_is_held(&q->debugfs_mutex)); |
cd649b8b | 1825 | if (attr == &dev_attr_enable) { |
c780e86d | 1826 | if (!!value == !!bt) { |
757d9140 SRV |
1827 | ret = 0; |
1828 | goto out_unlock_bdev; | |
1829 | } | |
cd649b8b | 1830 | if (value) |
9908c309 | 1831 | ret = blk_trace_setup_queue(q, bdev); |
cd649b8b LZ |
1832 | else |
1833 | ret = blk_trace_remove_queue(q); | |
1834 | goto out_unlock_bdev; | |
1835 | } | |
1836 | ||
c71a8961 | 1837 | ret = 0; |
153031a3 | 1838 | if (bt == NULL) { |
9908c309 | 1839 | ret = blk_trace_setup_queue(q, bdev); |
153031a3 | 1840 | bt = rcu_dereference_protected(q->blk_trace, |
85e0cbbb | 1841 | lockdep_is_held(&q->debugfs_mutex)); |
153031a3 | 1842 | } |
c71a8961 ACM |
1843 | |
1844 | if (ret == 0) { | |
1845 | if (attr == &dev_attr_act_mask) | |
c780e86d | 1846 | bt->act_mask = value; |
c71a8961 | 1847 | else if (attr == &dev_attr_pid) |
c780e86d | 1848 | bt->pid = value; |
c71a8961 | 1849 | else if (attr == &dev_attr_start_lba) |
c780e86d | 1850 | bt->start_lba = value; |
c71a8961 | 1851 | else if (attr == &dev_attr_end_lba) |
c780e86d | 1852 | bt->end_lba = value; |
c71a8961 | 1853 | } |
cd649b8b LZ |
1854 | |
1855 | out_unlock_bdev: | |
85e0cbbb | 1856 | mutex_unlock(&q->debugfs_mutex); |
c71a8961 | 1857 | out: |
cd649b8b | 1858 | return ret ? ret : count; |
c71a8961 | 1859 | } |
cd649b8b | 1860 | |
1d54ad6d LZ |
1861 | int blk_trace_init_sysfs(struct device *dev) |
1862 | { | |
1863 | return sysfs_create_group(&dev->kobj, &blk_trace_attr_group); | |
1864 | } | |
1865 | ||
48c0d4d4 ZK |
1866 | void blk_trace_remove_sysfs(struct device *dev) |
1867 | { | |
1868 | sysfs_remove_group(&dev->kobj, &blk_trace_attr_group); | |
1869 | } | |
1870 | ||
55782138 LZ |
1871 | #endif /* CONFIG_BLK_DEV_IO_TRACE */ |
1872 | ||
1873 | #ifdef CONFIG_EVENT_TRACING | |
1874 | ||
1f83bb4b CK |
1875 | /** |
1876 | * blk_fill_rwbs - Fill the buffer rwbs by mapping op to character string. | |
94d4bffd | 1877 | * @rwbs: buffer to be filled |
1f83bb4b CK |
1878 | * @op: REQ_OP_XXX for the tracepoint |
1879 | * | |
1880 | * Description: | |
1881 | * Maps the REQ_OP_XXX to character and fills the buffer provided by the | |
1882 | * caller with resulting string. | |
1883 | * | |
1884 | **/ | |
179d1600 | 1885 | void blk_fill_rwbs(char *rwbs, unsigned int op) |
55782138 LZ |
1886 | { |
1887 | int i = 0; | |
1888 | ||
ef295ecf | 1889 | if (op & REQ_PREFLUSH) |
c09c47ca NK |
1890 | rwbs[i++] = 'F'; |
1891 | ||
ef295ecf | 1892 | switch (op & REQ_OP_MASK) { |
1b9a9ab7 MC |
1893 | case REQ_OP_WRITE: |
1894 | case REQ_OP_WRITE_SAME: | |
55782138 | 1895 | rwbs[i++] = 'W'; |
1b9a9ab7 MC |
1896 | break; |
1897 | case REQ_OP_DISCARD: | |
55782138 | 1898 | rwbs[i++] = 'D'; |
1b9a9ab7 | 1899 | break; |
288dab8a CH |
1900 | case REQ_OP_SECURE_ERASE: |
1901 | rwbs[i++] = 'D'; | |
1902 | rwbs[i++] = 'E'; | |
1903 | break; | |
3a5e02ce MC |
1904 | case REQ_OP_FLUSH: |
1905 | rwbs[i++] = 'F'; | |
1906 | break; | |
1b9a9ab7 | 1907 | case REQ_OP_READ: |
55782138 | 1908 | rwbs[i++] = 'R'; |
1b9a9ab7 MC |
1909 | break; |
1910 | default: | |
55782138 | 1911 | rwbs[i++] = 'N'; |
1b9a9ab7 | 1912 | } |
55782138 | 1913 | |
ef295ecf | 1914 | if (op & REQ_FUA) |
c09c47ca | 1915 | rwbs[i++] = 'F'; |
ef295ecf | 1916 | if (op & REQ_RAHEAD) |
55782138 | 1917 | rwbs[i++] = 'A'; |
ef295ecf | 1918 | if (op & REQ_SYNC) |
55782138 | 1919 | rwbs[i++] = 'S'; |
ef295ecf | 1920 | if (op & REQ_META) |
55782138 LZ |
1921 | rwbs[i++] = 'M'; |
1922 | ||
1923 | rwbs[i] = '\0'; | |
1924 | } | |
9ca8f8e5 | 1925 | EXPORT_SYMBOL_GPL(blk_fill_rwbs); |
55782138 | 1926 | |
55782138 LZ |
1927 | #endif /* CONFIG_EVENT_TRACING */ |
1928 |