block: create the request_queue debugfs_dir on registration

We were only creating the request_queue debugfs_dir only for make_request block drivers (multiqueue), but never for request-based block drivers. We did this as we were only creating non-blktrace additional debugfs files on that directory for make_request drivers. However, since blktrace *always* creates that directory anyway, we special-case the use of that directory on blktrace. Other than this being an eye-sore, this exposes request-based block drivers to the same debugfs fragile race that used to exist with make_request block drivers where if we start adding files onto that directory we can later run a race with a double removal of dentries on the directory if we don't deal with this carefully on blktrace. Instead, just simplify things by always creating the request_queue debugfs_dir on request_queue registration. Rename the mutex also to reflect the fact that this is used outside of the blktrace context.

Luis Chamberlain committed 6 years ago

Commit 85e0cbb1 parent b431ef86 files · +39 −48Browse files

Unified · Split

6 files

block/blk-core.c +1 −7
block/blk-mq-debugfs.c +0 −5
block/blk-sysfs.c +9 −0
block/blk.h +0 −2
include/linux/blkdev.h +3 −2
kernel/trace/blktrace.c +26 −32

block/blk-core.c+1−7

@@ -51,9 +51,7 @@ · 50 unmodified lines
51	51	#include "blk-pm.h"
52	52	#include "blk-rq-qos.h"
53	53
54		−#ifdef CONFIG_DEBUG_FS
55	54	struct dentry *blk_debugfs_root;
56		−#endif
57	55
58	56	EXPORT_TRACEPOINT_SYMBOL_GPL(block_bio_remap);
59	57	EXPORT_TRACEPOINT_SYMBOL_GPL(block_rq_remap);
@@ -555,9 +553,7 @@ struct request_queue *__blk_alloc_queue(int node_id) · 495 unmodified lines
555	553
556	554	kobject_init(&q->kobj, &blk_queue_ktype);
557	555
558		−#ifdef CONFIG_BLK_DEV_IO_TRACE
559		− mutex_init(&q->blk_trace_mutex);
560		−#endif
	556	+ mutex_init(&q->debugfs_mutex);
561	557	mutex_init(&q->sysfs_lock);
562	558	mutex_init(&q->sysfs_dir_lock);
563	559	spin_lock_init(&q->queue_lock);
@@ -1931,9 +1927,7 @@ int __init blk_dev_init(void) · 1367 unmodified lines
1931	1927	blk_requestq_cachep = kmem_cache_create("request_queue",
1932	1928	sizeof(struct request_queue), 0, SLAB_PANIC, NULL);
1933	1929
1934		−#ifdef CONFIG_DEBUG_FS
1935	1930	blk_debugfs_root = debugfs_create_dir("block", NULL);
1936		−#endif
1937	1931
1938	1932	return 0;
1939	1933	}

block/blk-mq-debugfs.c+0−5

@@ -824,9 +824,6 @@ void blk_mq_debugfs_register(struct request_queue *q) · 823 unmodified lines
824	824	struct blk_mq_hw_ctx *hctx;
825	825	int i;
826	826
827		− q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
828		− blk_debugfs_root);
829		−
830	827	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
831	828
832	829	/*
@@ -857,9 +854,7 @@ void blk_mq_debugfs_register(struct request_queue *q) · 24 unmodified lines
857	854
858	855	void blk_mq_debugfs_unregister(struct request_queue *q)
859	856	{
860		− debugfs_remove_recursive(q->debugfs_dir);
861	857	q->sched_debugfs_dir = NULL;
862		− q->debugfs_dir = NULL;
863	858	}
864	859
865	860	static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,

block/blk-sysfs.c+9−0

@@ -11,6 +11,7 @@ · 10 unmodified lines
11	11	#include <linux/blktrace_api.h>
12	12	#include <linux/blk-mq.h>
13	13	#include <linux/blk-cgroup.h>
	14	+#include <linux/debugfs.h>
14	15
15	16	#include "blk.h"
16	17	#include "blk-mq.h"
@@ -917,6 +918,9 @@ static void blk_release_queue(struct kobject *kobj) · 900 unmodified lines
917	918	blk_mq_release(q);
918	919
919	920	blk_trace_shutdown(q);
	921	+ mutex_lock(&q->debugfs_mutex);
	922	+ debugfs_remove_recursive(q->debugfs_dir);
	923	+ mutex_unlock(&q->debugfs_mutex);
920	924
921	925	if (queue_is_mq(q))
922	926	blk_mq_debugfs_unregister(q);
@@ -989,6 +993,11 @@ int blk_register_queue(struct gendisk *disk) · 66 unmodified lines
989	993	goto unlock;
990	994	}
991	995
	996	+ mutex_lock(&q->debugfs_mutex);
	997	+ q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
	998	+ blk_debugfs_root);
	999	+ mutex_unlock(&q->debugfs_mutex);
	1000	+
992	1001	if (queue_is_mq(q)) {
993	1002	__blk_mq_register_dev(dev, q);
994	1003	blk_mq_debugfs_register(q);

block/blk.h+0−2

@@ -14,9 +14,7 @@ · 13 unmodified lines
14	14	/* Max future timer expiry for timeouts */
15	15	#define BLK_MAX_TIMEOUT (5 * HZ)
16	16
17		−#ifdef CONFIG_DEBUG_FS
18	17	extern struct dentry *blk_debugfs_root;
19		−#endif
20	18
21	19	struct blk_flush_queue {
22	20	unsigned int flush_pending_idx:1;

include/linux/blkdev.h+3−2

@@ -528,9 +528,9 @@ struct request_queue { · 527 unmodified lines
528	528	unsigned int sg_timeout;
529	529	unsigned int sg_reserved_size;
530	530	int node;
	531	+ struct mutex debugfs_mutex;
531	532	#ifdef CONFIG_BLK_DEV_IO_TRACE
532	533	struct blk_trace __rcu *blk_trace;
533		− struct mutex blk_trace_mutex;
534	534	#endif
535	535	/*
536	536	* for flush operations
@@ -574,8 +574,9 @@ struct request_queue { · 37 unmodified lines
574	574	struct list_head tag_set_list;
575	575	struct bio_set bio_split;
576	576
577		−#ifdef CONFIG_BLK_DEBUG_FS
578	577	struct dentry *debugfs_dir;
	578	+
	579	+#ifdef CONFIG_BLK_DEBUG_FS
579	580	struct dentry *sched_debugfs_dir;
580	581	struct dentry *rqos_debugfs_dir;
581	582	#endif

kernel/trace/blktrace.c+26−32

@@ -348,7 +348,7 @@ static int __blk_trace_remove(struct request_queue *q) · 347 unmodified lines
348	348	struct blk_trace *bt;
349	349
350	350	bt = rcu_replace_pointer(q->blk_trace, NULL,
351		− lockdep_is_held(&q->blk_trace_mutex));
	351	+ lockdep_is_held(&q->debugfs_mutex));
352	352	if (!bt)
353	353	return -EINVAL;
354	354
@@ -362,9 +362,9 @@ int blk_trace_remove(struct request_queue *q) · 7 unmodified lines
362	362	{
363	363	int ret;
364	364
365		− mutex_lock(&q->blk_trace_mutex);
	365	+ mutex_lock(&q->debugfs_mutex);
366	366	ret = __blk_trace_remove(q);
367		− mutex_unlock(&q->blk_trace_mutex);
	367	+ mutex_unlock(&q->debugfs_mutex);
368	368
369	369	return ret;
370	370	}
@@ -483,14 +483,11 @@ static int do_blk_trace_setup(struct request_queue q, char name, dev_t dev, · 112 unmodified lines
483	483	struct dentry *dir = NULL;
484	484	int ret;
485	485
486		− lockdep_assert_held(&q->blk_trace_mutex);
	486	+ lockdep_assert_held(&q->debugfs_mutex);
487	487
488	488	if (!buts->buf_size \|\| !buts->buf_nr)
489	489	return -EINVAL;
490	490
491		− if (!blk_debugfs_root)
492		− return -ENOENT;
493		−
494	491	strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
495	492	buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
496	493
@@ -505,7 +502,7 @@ static int do_blk_trace_setup(struct request_queue q, char name, dev_t dev, · 8 unmodified lines
505	502	* we can be.
506	503	*/
507	504	if (rcu_dereference_protected(q->blk_trace,
508		− lockdep_is_held(&q->blk_trace_mutex))) {
	505	+ lockdep_is_held(&q->debugfs_mutex))) {
509	506	pr_warn("Concurrent blktraces are not allowed on %s\n",
510	507	buts->name);
511	508	return -EBUSY;
@@ -524,18 +521,15 @@ static int do_blk_trace_setup(struct request_queue q, char name, dev_t dev, · 12 unmodified lines
524	521	if (!bt->msg_data)
525	522	goto err;
526	523
527		−#ifdef CONFIG_BLK_DEBUG_FS
528	524	/*
529		− * When tracing whole make_request drivers (multiqueue) block devices,
530		− * reuse the existing debugfs directory created by the block layer on
531		− * init. For request-based block devices, all partitions block devices,
	525	+ * When tracing the whole disk reuse the existing debugfs directory
	526	+ * created by the block layer on init. For partitions block devices,
532	527	* and scsi-generic block devices we create a temporary new debugfs
533	528	* directory that will be removed once the trace ends.
534	529	*/
535		− if (queue_is_mq(q) && bdev && bdev == bdev->bd_contains)
	530	+ if (bdev && bdev == bdev->bd_contains)
536	531	dir = q->debugfs_dir;
537	532	else
538		−#endif
539	533	bt->dir = dir = debugfs_create_dir(buts->name, blk_debugfs_root);
540	534
541	535	/*
@@ -617,9 +611,9 @@ int blk_trace_setup(struct request_queue q, char name, dev_t dev, · 75 unmodified lines
617	611	{
618	612	int ret;
619	613
620		− mutex_lock(&q->blk_trace_mutex);
	614	+ mutex_lock(&q->debugfs_mutex);
621	615	ret = __blk_trace_setup(q, name, dev, bdev, arg);
622		− mutex_unlock(&q->blk_trace_mutex);
	616	+ mutex_unlock(&q->debugfs_mutex);
623	617
624	618	return ret;
625	619	}
@@ -665,7 +659,7 @@ static int __blk_trace_startstop(struct request_queue *q, int start) · 39 unmodified lines
665	659	struct blk_trace *bt;
666	660
667	661	bt = rcu_dereference_protected(q->blk_trace,
668		− lockdep_is_held(&q->blk_trace_mutex));
	662	+ lockdep_is_held(&q->debugfs_mutex));
669	663	if (bt == NULL)
670	664	return -EINVAL;
671	665
@@ -705,9 +699,9 @@ int blk_trace_startstop(struct request_queue *q, int start) · 33 unmodified lines
705	699	{
706	700	int ret;
707	701
708		− mutex_lock(&q->blk_trace_mutex);
	702	+ mutex_lock(&q->debugfs_mutex);
709	703	ret = __blk_trace_startstop(q, start);
710		− mutex_unlock(&q->blk_trace_mutex);
	704	+ mutex_unlock(&q->debugfs_mutex);
711	705
712	706	return ret;
713	707	}
@@ -736,7 +730,7 @@ int blk_trace_ioctl(struct block_device bdev, unsigned cmd, char __user arg) · 22 unmodified lines
736	730	if (!q)
737	731	return -ENXIO;
738	732
739		− mutex_lock(&q->blk_trace_mutex);
	733	+ mutex_lock(&q->debugfs_mutex);
740	734
741	735	switch (cmd) {
742	736	case BLKTRACESETUP:
@@ -763,7 +757,7 @@ int blk_trace_ioctl(struct block_device bdev, unsigned cmd, char __user arg) · 20 unmodified lines
763	757	break;
764	758	}
765	759
766		− mutex_unlock(&q->blk_trace_mutex);
	760	+ mutex_unlock(&q->debugfs_mutex);
767	761	return ret;
768	762	}
769	763
@@ -774,14 +768,14 @@ int blk_trace_ioctl(struct block_device bdev, unsigned cmd, char __user arg) · 4 unmodified lines
774	768	**/
775	769	void blk_trace_shutdown(struct request_queue *q)
776	770	{
777		− mutex_lock(&q->blk_trace_mutex);
	771	+ mutex_lock(&q->debugfs_mutex);
778	772	if (rcu_dereference_protected(q->blk_trace,
779		− lockdep_is_held(&q->blk_trace_mutex))) {
	773	+ lockdep_is_held(&q->debugfs_mutex))) {
780	774	__blk_trace_startstop(q, 0);
781	775	__blk_trace_remove(q);
782	776	}
783	777
784		− mutex_unlock(&q->blk_trace_mutex);
	778	+ mutex_unlock(&q->debugfs_mutex);
785	779	}
786	780
787	781	#ifdef CONFIG_BLK_CGROUP
@@ -1662,7 +1656,7 @@ static int blk_trace_remove_queue(struct request_queue *q) · 874 unmodified lines
1662	1656	struct blk_trace *bt;
1663	1657
1664	1658	bt = rcu_replace_pointer(q->blk_trace, NULL,
1665		− lockdep_is_held(&q->blk_trace_mutex));
	1659	+ lockdep_is_held(&q->debugfs_mutex));
1666	1660	if (bt == NULL)
1667	1661	return -EINVAL;
1668	1662
@@ -1837,10 +1831,10 @@ static ssize_t sysfs_blk_trace_attr_show(struct device *dev, · 168 unmodified lines
1837	1831	if (q == NULL)
1838	1832	goto out_bdput;
1839	1833
1840		− mutex_lock(&q->blk_trace_mutex);
	1834	+ mutex_lock(&q->debugfs_mutex);
1841	1835
1842	1836	bt = rcu_dereference_protected(q->blk_trace,
1843		− lockdep_is_held(&q->blk_trace_mutex));
	1837	+ lockdep_is_held(&q->debugfs_mutex));
1844	1838	if (attr == &dev_attr_enable) {
1845	1839	ret = sprintf(buf, "%u\n", !!bt);
1846	1840	goto out_unlock_bdev;
@@ -1858,7 +1852,7 @@ static ssize_t sysfs_blk_trace_attr_show(struct device *dev, · 11 unmodified lines
1858	1852	ret = sprintf(buf, "%llu\n", bt->end_lba);
1859	1853
1860	1854	out_unlock_bdev:
1861		− mutex_unlock(&q->blk_trace_mutex);
	1855	+ mutex_unlock(&q->debugfs_mutex);
1862	1856	out_bdput:
1863	1857	bdput(bdev);
1864	1858	out:
@@ -1901,10 +1895,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev, · 36 unmodified lines
1901	1895	if (q == NULL)
1902	1896	goto out_bdput;
1903	1897
1904		− mutex_lock(&q->blk_trace_mutex);
	1898	+ mutex_lock(&q->debugfs_mutex);
1905	1899
1906	1900	bt = rcu_dereference_protected(q->blk_trace,
1907		− lockdep_is_held(&q->blk_trace_mutex));
	1901	+ lockdep_is_held(&q->debugfs_mutex));
1908	1902	if (attr == &dev_attr_enable) {
1909	1903	if (!!value == !!bt) {
1910	1904	ret = 0;
@@ -1921,7 +1915,7 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev, · 10 unmodified lines
1921	1915	if (bt == NULL) {
1922	1916	ret = blk_trace_setup_queue(q, bdev);
1923	1917	bt = rcu_dereference_protected(q->blk_trace,
1924		− lockdep_is_held(&q->blk_trace_mutex));
	1918	+ lockdep_is_held(&q->debugfs_mutex));
1925	1919	}
1926	1920
1927	1921	if (ret == 0) {
@@ -1936,7 +1930,7 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev, · 8 unmodified lines
1936	1930	}
1937	1931
1938	1932	out_unlock_bdev:
1939		− mutex_unlock(&q->blk_trace_mutex);
	1933	+ mutex_unlock(&q->debugfs_mutex);
1940	1934	out_bdput:
1941	1935	bdput(bdev);
1942	1936	out: