hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE: NA

--------------------------------

Because slow io or io timeout handler can take a long time, hungtask check is forbidden in order to prevent false positive warnings. However, this also cause kenel to be silence if io really hang.

It's quite complicated to distinguish if io is slow or hanged, this patch add a switch to enable hungtask check, the switch is enabled by default, and can be turn off by:

1) disable config BLK_IO_HUNG_TASK_CHECK 2) add blk_core.io_hung_task_check=0 to boot cmd 3) echo 0 > /sys/module/blk_core/parameters/io_hung_task_check

Noted that user has to be careful to use this with hungtask panic enabeld, since there could be false positive hungtask warnings.

Signed-off-by: Yu Kuai yukuai3@huawei.com --- arch/arm64/configs/openeuler_defconfig | 1 + arch/powerpc/configs/openeuler_defconfig | 1 + arch/x86/configs/openeuler_defconfig | 1 + block/Kconfig | 9 +++++++++ block/bio.c | 2 +- block/blk-core.c | 13 ++++++++++++- block/blk-exec.c | 2 +- block/blk.h | 1 + 8 files changed, 27 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/configs/openeuler_defconfig b/arch/arm64/configs/openeuler_defconfig index 34061d75a0d2..576b61de82a1 100644 --- a/arch/arm64/configs/openeuler_defconfig +++ b/arch/arm64/configs/openeuler_defconfig @@ -951,6 +951,7 @@ CONFIG_BLK_DEBUG_FS=y CONFIG_BLK_DEBUG_FS_ZONED=y # CONFIG_BLK_SED_OPAL is not set # CONFIG_BLK_INLINE_ENCRYPTION is not set +CONFIG_BLK_IO_HUNG_TASK_CHECK=y CONFIG_BLK_DEV_DUMPINFO=y

# diff --git a/arch/powerpc/configs/openeuler_defconfig b/arch/powerpc/configs/openeuler_defconfig index 74118e43ff05..a144065cb5a5 100644 --- a/arch/powerpc/configs/openeuler_defconfig +++ b/arch/powerpc/configs/openeuler_defconfig @@ -687,6 +687,7 @@ CONFIG_BLK_DEBUG_FS_ZONED=y CONFIG_BLK_SED_OPAL=y CONFIG_BLK_INLINE_ENCRYPTION=y # CONFIG_BLK_INLINE_ENCRYPTION_FALLBACK is not set +CONFIG_BLK_IO_HUNG_TASK_CHECK=y

# # Partition Types diff --git a/arch/x86/configs/openeuler_defconfig b/arch/x86/configs/openeuler_defconfig index 91c85e1c229e..c535c1001f6e 100644 --- a/arch/x86/configs/openeuler_defconfig +++ b/arch/x86/configs/openeuler_defconfig @@ -932,6 +932,7 @@ CONFIG_BLK_DEBUG_FS=y CONFIG_BLK_DEBUG_FS_ZONED=y # CONFIG_BLK_SED_OPAL is not set # CONFIG_BLK_INLINE_ENCRYPTION is not set +CONFIG_BLK_IO_HUNG_TASK_CHECK=y CONFIG_BLK_DEV_DUMPINFO=y

# diff --git a/block/Kconfig b/block/Kconfig index e5c965f1ea25..e213002d5e73 100644 --- a/block/Kconfig +++ b/block/Kconfig @@ -209,6 +209,15 @@ config BLK_DEV_DUMPINFO Dump info when open an write opened block device exclusively or open an exclusive opened device for write

+config BLK_IO_HUNG_TASK_CHECK + bool "Enable io hung task check" + depends on DETECT_HUNG_TASK + default y + help + Enabling this lets the block layer detect hungtask for io, noted + if this is set, hungtask will complain about slow io even if such + io is not hanged. Be careful to enable hungtask panic in this case. + menu "Partition Types"

source "block/partitions/Kconfig" diff --git a/block/bio.c b/block/bio.c index 8c64c93e96c8..123b44ba17cb 100644 --- a/block/bio.c +++ b/block/bio.c @@ -1169,7 +1169,7 @@ int submit_bio_wait(struct bio *bio) submit_bio(bio);

/* Prevent hang_check timer from firing at us during very long I/O */ - hang_check = sysctl_hung_task_timeout_secs; + hang_check = sysctl_hung_task_timeout_secs && !io_hung_task_check; if (hang_check) while (!wait_for_completion_io_timeout(&done, hang_check * (HZ/2))) diff --git a/block/blk-core.c b/block/blk-core.c index f91f8e8be482..e291ed8ee01e 100644 --- a/block/blk-core.c +++ b/block/blk-core.c @@ -77,6 +77,17 @@ static int __init precise_iostat_setup(char *str) } __setup("precise_iostat=", precise_iostat_setup);

+/* + * Noted if this is set, hungtask will complain about slow io even if such io is + * not hanged. Be careful to enable hungtask panic in this case. + */ +#ifdef CONFIG_BLK_IO_HUNG_TASK_CHECK +bool io_hung_task_check = true; +#else +bool io_hung_task_check; +#endif +module_param_named(io_hung_task_check, io_hung_task_check, bool, 0644); + /* * For queue allocation */ @@ -1879,7 +1890,7 @@ void blk_io_schedule(void) /* Prevent hang_check timer from firing at us during very long I/O */ unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2;

- if (timeout) + if (timeout && !io_hung_task_check) io_schedule_timeout(timeout); else io_schedule(); diff --git a/block/blk-exec.c b/block/blk-exec.c index b2676de4c6a5..497aa52cd51e 100644 --- a/block/blk-exec.c +++ b/block/blk-exec.c @@ -87,7 +87,7 @@ blk_status_t blk_execute_rq(struct request_queue *q, struct gendisk *bd_disk, blk_execute_rq_nowait(q, bd_disk, rq, at_head, blk_end_sync_rq);

/* Prevent hang_check timer from firing at us during very long I/O */ - hang_check = sysctl_hung_task_timeout_secs; + hang_check = sysctl_hung_task_timeout_secs && !io_hung_task_check; if (hang_check) while (!wait_for_completion_io_timeout(&wait, hang_check * (HZ/2))); else diff --git a/block/blk.h b/block/blk.h index 4bbcc971d4f7..2e487673b3a7 100644 --- a/block/blk.h +++ b/block/blk.h @@ -15,6 +15,7 @@ #define BLK_MAX_TIMEOUT (5 * HZ)

extern struct dentry *blk_debugfs_root; +extern bool io_hung_task_check;

struct blk_flush_queue { unsigned int flush_pending_idx:1;

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE:NA

-------------------------------

The feature that enable default hierarchy for io throttle in cgroup v1 can only be enabled with CONFIG_BLK_DEV_SUPPORT_LEGACY_GLOBAL_LIMIT.

Signed-off-by: Yu Kuai yukuai3@huawei.com --- arch/arm64/configs/openeuler_defconfig | 1 + arch/x86/configs/openeuler_defconfig | 1 + block/Kconfig | 11 +++++++++++ block/blk-throttle.c | 17 ++++++++++++++--- 4 files changed, 27 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/configs/openeuler_defconfig b/arch/arm64/configs/openeuler_defconfig index e360b836280e..10c941186dbe 100644 --- a/arch/arm64/configs/openeuler_defconfig +++ b/arch/arm64/configs/openeuler_defconfig @@ -942,6 +942,7 @@ CONFIG_BLK_DEV_INTEGRITY_T10=m CONFIG_BLK_DEV_ZONED=y CONFIG_BLK_DEV_THROTTLING=y # CONFIG_BLK_DEV_THROTTLING_LOW is not set +CONFIG_BLK_DEV_SUPPORT_LEGACY_GLOBAL_LIMIT=y # CONFIG_BLK_CMDLINE_PARSER is not set CONFIG_BLK_WBT=y # CONFIG_BLK_CGROUP_IOLATENCY is not set diff --git a/arch/x86/configs/openeuler_defconfig b/arch/x86/configs/openeuler_defconfig index de1bd729b75b..8c3ab3abae29 100644 --- a/arch/x86/configs/openeuler_defconfig +++ b/arch/x86/configs/openeuler_defconfig @@ -923,6 +923,7 @@ CONFIG_BLK_DEV_INTEGRITY_T10=m CONFIG_BLK_DEV_ZONED=y CONFIG_BLK_DEV_THROTTLING=y # CONFIG_BLK_DEV_THROTTLING_LOW is not set +CONFIG_BLK_DEV_SUPPORT_LEGACY_GLOBAL_LIMIT=y # CONFIG_BLK_CMDLINE_PARSER is not set CONFIG_BLK_WBT=y # CONFIG_BLK_CGROUP_IOLATENCY is not set diff --git a/block/Kconfig b/block/Kconfig index 1e733b2b14d8..498d18f1d10e 100644 --- a/block/Kconfig +++ b/block/Kconfig @@ -114,6 +114,17 @@ config BLK_DEV_THROTTLING_LOW

Note, this is an experimental interface and could be changed someday.

+config BLK_DEV_SUPPORT_LEGACY_GLOBAL_LIMIT + bool "Block layer global limit in cgroup v1" + depends on BLK_DEV_THROTTLING=y + default n + help + blkio subsytem is not under default hierarchy in cgroup v1 by default, + Enabling this will support globlal limit in cgroup v1. + + Note, a cmdline "blkcg_global_limit=1" is still required to enabled this + feature. + config BLK_CMDLINE_PARSER bool "Block device command line partition parser" help diff --git a/block/blk-throttle.c b/block/blk-throttle.c index 050ddf0ad002..e5da0664b16c 100644 --- a/block/blk-throttle.c +++ b/block/blk-throttle.c @@ -43,9 +43,15 @@ static struct blkcg_policy blkcg_policy_throtl; /* A workqueue to queue throttle related work */ static struct workqueue_struct *kthrotld_workqueue;

+#ifdef BLK_DEV_SUPPORT_LEGACY_GLOBAL_LIMIT /* True if global limit is enabled in cgroup v1 */ static bool global_limit;

+static inline bool blkcg_global_limit_enabled(void) +{ + return global_limit; +} + static int __init setup_global_limit(char *str) { if (!strcmp(str, "1") || !strcmp(str, "Y") || !strcmp(str, "y")) @@ -55,7 +61,12 @@ static int __init setup_global_limit(char *str) }

__setup("blkcg_global_limit=", setup_global_limit); - +#else +static inline bool blkcg_global_limit_enabled(void) +{ + return false; +} +#endif /* * To implement hierarchical throttling, throtl_grps form a tree and bios * are dispatched upwards level by level until they reach the top and get @@ -571,8 +582,8 @@ static void throtl_pd_init(struct blkg_policy_data *pd) * regardless of the position of the group in the hierarchy. */ sq->parent_sq = &td->service_queue; - if ((cgroup_subsys_on_dfl(io_cgrp_subsys) || global_limit) && - blkg->parent) + if ((cgroup_subsys_on_dfl(io_cgrp_subsys) || + blkcg_global_limit_enabled()) && blkg->parent) sq->parent_sq = &blkg_to_tg(blkg->parent)->service_queue; tg->td = td; }

From: Christoph Hellwig hch@lst.de

mainline inclusion from mainline-v5.19-rc4 commit 5cf9c91ba927119fc6606b938b1895bb2459d3bc category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE: NA

Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...

----------------------------------------

Various places like I/O schedulers or the QOS infrastructure try to register debugfs files on demans, which can race with creating and removing the main queue debugfs directory. Use the existing debugfs_mutex to serialize all debugfs operations that rely on q->debugfs_dir or the directories hanging off it.

To make the teardown code a little simpler declare all debugfs dentry pointers and not just the main one uncoditionally in blkdev.h.

Move debugfs_mutex next to the dentries that it protects and document what it is used for.

Signed-off-by: Christoph Hellwig hch@lst.de Link: https://lore.kernel.org/r/20220614074827.458955-3-hch@lst.de Signed-off-by: Jens Axboe axboe@kernel.dk

Conflicts: block/blk-sysfs.c include/linux/blkdev.h kernel/trace/blktrace.c block/blk-mq-debugfs.c block/blk-mq-sched.c Signed-off-by: Yu Kuai yukuai3@huawei.com --- block/blk-mq-debugfs.c | 25 ++++++++++++++++++++----- block/blk-mq-debugfs.h | 5 ----- block/blk-mq-sched.c | 11 +++++++++++ block/blk-rq-qos.c | 2 ++ block/blk-rq-qos.h | 7 ++++++- block/blk-sysfs.c | 20 ++++++++++---------- include/linux/blkdev.h | 8 ++++---- kernel/trace/blktrace.c | 3 --- 8 files changed, 53 insertions(+), 28 deletions(-)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c index a879f94782e4..d8623d9a7b18 100644 --- a/block/blk-mq-debugfs.c +++ b/block/blk-mq-debugfs.c @@ -892,11 +892,6 @@ void blk_mq_debugfs_register(struct request_queue *q) } }

-void blk_mq_debugfs_unregister(struct request_queue *q) -{ - q->sched_debugfs_dir = NULL; -} - static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx) { @@ -930,6 +925,8 @@ void blk_mq_debugfs_register_hctx(struct request_queue *q,

void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx) { + if (!hctx->queue->debugfs_dir) + return; debugfs_remove_recursive(hctx->debugfs_dir); hctx->sched_debugfs_dir = NULL; hctx->debugfs_dir = NULL; @@ -957,6 +954,8 @@ void blk_mq_debugfs_register_sched(struct request_queue *q) { struct elevator_type *e = q->elevator->type;

+ lockdep_assert_held(&q->debugfs_mutex); + /* * If the parent directory has not been created yet, return, we will be * called again later on and the directory/files will be created then. @@ -974,12 +973,18 @@ void blk_mq_debugfs_register_sched(struct request_queue *q)

void blk_mq_debugfs_unregister_sched(struct request_queue *q) { + lockdep_assert_held(&q->debugfs_mutex); + debugfs_remove_recursive(q->sched_debugfs_dir); q->sched_debugfs_dir = NULL; }

void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos) { + lockdep_assert_held(&rqos->q->debugfs_mutex); + + if (!rqos->q->debugfs_dir) + return; debugfs_remove_recursive(rqos->debugfs_dir); rqos->debugfs_dir = NULL; } @@ -989,6 +994,8 @@ void blk_mq_debugfs_register_rqos(struct rq_qos *rqos) struct request_queue *q = rqos->q; const char *dir_name = rq_qos_id_to_name(rqos->id);

+ lockdep_assert_held(&q->debugfs_mutex); + if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs) return;

@@ -1004,6 +1011,8 @@ void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)

void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q) { + lockdep_assert_held(&q->debugfs_mutex); + debugfs_remove_recursive(q->rqos_debugfs_dir); q->rqos_debugfs_dir = NULL; } @@ -1013,6 +1022,8 @@ void blk_mq_debugfs_register_sched_hctx(struct request_queue *q, { struct elevator_type *e = q->elevator->type;

+ lockdep_assert_held(&q->debugfs_mutex); + if (!e->hctx_debugfs_attrs) return;

@@ -1024,6 +1035,10 @@ void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,

void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx) { + lockdep_assert_held(&hctx->queue->debugfs_mutex); + + if (!hctx->queue->debugfs_dir) + return; debugfs_remove_recursive(hctx->sched_debugfs_dir); hctx->sched_debugfs_dir = NULL; } diff --git a/block/blk-mq-debugfs.h b/block/blk-mq-debugfs.h index a68aa6041a10..891c3af6f611 100644 --- a/block/blk-mq-debugfs.h +++ b/block/blk-mq-debugfs.h @@ -19,7 +19,6 @@ int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq); int blk_mq_debugfs_rq_show(struct seq_file *m, void *v);

void blk_mq_debugfs_register(struct request_queue *q); -void blk_mq_debugfs_unregister(struct request_queue *q); void blk_mq_debugfs_register_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx); void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx); @@ -40,10 +39,6 @@ static inline void blk_mq_debugfs_register(struct request_queue *q) { }

-static inline void blk_mq_debugfs_unregister(struct request_queue *q) -{ -} - static inline void blk_mq_debugfs_register_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx) { diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c index c92d25b71a72..8620a5d75c62 100644 --- a/block/blk-mq-sched.c +++ b/block/blk-mq-sched.c @@ -610,7 +610,9 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) if (ret) goto err_free_map_and_rqs;

+ mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_register_sched(q); + mutex_unlock(&q->debugfs_mutex);

queue_for_each_hw_ctx(q, hctx, i) { if (e->ops.init_hctx) { @@ -623,7 +625,9 @@ int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) return ret; } } + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_register_sched_hctx(q, hctx); + mutex_unlock(&q->debugfs_mutex); }

return 0; @@ -664,14 +668,21 @@ void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) unsigned int flags = 0;

queue_for_each_hw_ctx(q, hctx, i) { + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_unregister_sched_hctx(hctx); + mutex_unlock(&q->debugfs_mutex); + if (e->type->ops.exit_hctx && hctx->sched_data) { e->type->ops.exit_hctx(hctx, i); hctx->sched_data = NULL; } flags = hctx->flags; } + + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_unregister_sched(q); + mutex_unlock(&q->debugfs_mutex); + if (e->type->ops.exit_sched) e->type->ops.exit_sched(e); blk_mq_sched_tags_teardown(q, flags); diff --git a/block/blk-rq-qos.c b/block/blk-rq-qos.c index e83af7bc7591..249a6f05dd3b 100644 --- a/block/blk-rq-qos.c +++ b/block/blk-rq-qos.c @@ -294,7 +294,9 @@ void rq_qos_wait(struct rq_wait *rqw, void *private_data,

void rq_qos_exit(struct request_queue *q) { + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_unregister_queue_rqos(q); + mutex_unlock(&q->debugfs_mutex);

while (q->rq_qos) { struct rq_qos *rqos = q->rq_qos; diff --git a/block/blk-rq-qos.h b/block/blk-rq-qos.h index 37c59d7d6ba7..af1c2ca157d7 100644 --- a/block/blk-rq-qos.h +++ b/block/blk-rq-qos.h @@ -118,8 +118,11 @@ static inline int rq_qos_add(struct request_queue *q, struct rq_qos *rqos)

blk_mq_unfreeze_queue(q);

- if (rqos->ops->debugfs_attrs) + if (rqos->ops->debugfs_attrs) { + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_register_rqos(rqos); + mutex_unlock(&q->debugfs_mutex); + }

return 0; ebusy: @@ -150,7 +153,9 @@ static inline void rq_qos_del(struct request_queue *q, struct rq_qos *rqos)

blk_mq_unfreeze_queue(q);

+ mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_unregister_rqos(rqos); + mutex_unlock(&q->debugfs_mutex); }

typedef bool (acquire_inflight_cb_t)(struct rq_wait *rqw, void *private_data); diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c index c95be9626a09..9e74668f98cc 100644 --- a/block/blk-sysfs.c +++ b/block/blk-sysfs.c @@ -804,14 +804,13 @@ static void blk_release_queue(struct kobject *kobj) if (queue_is_mq(q)) blk_mq_release(q);

- blk_trace_shutdown(q); mutex_lock(&q->debugfs_mutex); + blk_trace_shutdown(q); debugfs_remove_recursive(q->debugfs_dir); + q->debugfs_dir = NULL; + q->sched_debugfs_dir = NULL; mutex_unlock(&q->debugfs_mutex);

- if (queue_is_mq(q)) - blk_mq_debugfs_unregister(q); - bioset_exit(&q->bio_split);

ida_simple_remove(&blk_queue_ida, q->id); @@ -881,17 +880,18 @@ int blk_register_queue(struct gendisk *disk) goto unlock; }

+ if (queue_is_mq(q)) + __blk_mq_register_dev(dev, q); + mutex_lock(&q->sysfs_lock); + mutex_lock(&q->debugfs_mutex); q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent), blk_debugfs_root); - mutex_unlock(&q->debugfs_mutex); - - if (queue_is_mq(q)) { - __blk_mq_register_dev(dev, q); + if (queue_is_mq(q)) blk_mq_debugfs_register(q); - }

- mutex_lock(&q->sysfs_lock); + mutex_unlock(&q->debugfs_mutex); + if (q->elevator) { ret = elv_register_queue(q, false); if (ret) { diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index 50b4fd0a0687..8208922552b5 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -540,7 +540,6 @@ struct request_queue { unsigned int sg_timeout; unsigned int sg_reserved_size; int node; - struct mutex debugfs_mutex; #ifdef CONFIG_BLK_DEV_IO_TRACE struct blk_trace __rcu *blk_trace; #endif @@ -588,11 +587,12 @@ struct request_queue { struct bio_set bio_split;

struct dentry *debugfs_dir; - -#ifdef CONFIG_BLK_DEBUG_FS struct dentry *sched_debugfs_dir; struct dentry *rqos_debugfs_dir; -#endif + /* + * Serializes all debugfs metadata operations using the above dentries. + */ + struct mutex debugfs_mutex;

bool mq_sysfs_init_done;

diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c index bc98d5e4e033..246ed13b49ca 100644 --- a/kernel/trace/blktrace.c +++ b/kernel/trace/blktrace.c @@ -773,12 +773,9 @@ int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg) **/ void blk_trace_shutdown(struct request_queue *q) { - mutex_lock(&q->debugfs_mutex); if (rcu_dereference_protected(q->blk_trace, lockdep_is_held(&q->debugfs_mutex))) __blk_trace_remove(q); - - mutex_unlock(&q->debugfs_mutex); }

#ifdef CONFIG_BLK_CGROUP

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE: NA

----------------------------------------

All operations to create and remove files under 'q->debugfs_dir' should be protected by 'q->debugfs_mutex'.

Signed-off-by: Yu Kuai yukuai3@huawei.com --- block/blk-mq-debugfs.c | 8 ++++++++ block/blk-mq.c | 2 ++ 2 files changed, 10 insertions(+)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c index 1e4a5ff73bd8..e2b8942311c9 100644 --- a/block/blk-mq-debugfs.c +++ b/block/blk-mq-debugfs.c @@ -911,6 +911,8 @@ void blk_mq_debugfs_register_hctx(struct request_queue *q, char name[20]; int i;

+ lockdep_assert_held(&q->debugfs_mutex); + if (!q->debugfs_dir) return;

@@ -925,6 +927,8 @@ void blk_mq_debugfs_register_hctx(struct request_queue *q,

void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx) { + lockdep_assert_held(&hctx->queue->debugfs_mutex); + if (!hctx->queue->debugfs_dir) return; debugfs_remove_recursive(hctx->debugfs_dir); @@ -937,8 +941,10 @@ void blk_mq_debugfs_register_hctxs(struct request_queue *q) struct blk_mq_hw_ctx *hctx; int i;

+ mutex_lock(&q->debugfs_mutex); queue_for_each_hw_ctx(q, hctx, i) blk_mq_debugfs_register_hctx(q, hctx); + mutex_unlock(&q->debugfs_mutex); }

void blk_mq_debugfs_unregister_hctxs(struct request_queue *q) @@ -946,8 +952,10 @@ void blk_mq_debugfs_unregister_hctxs(struct request_queue *q) struct blk_mq_hw_ctx *hctx; int i;

+ mutex_lock(&q->debugfs_mutex); queue_for_each_hw_ctx(q, hctx, i) blk_mq_debugfs_unregister_hctx(hctx); + mutex_unlock(&q->debugfs_mutex); }

void blk_mq_debugfs_register_sched(struct request_queue *q) diff --git a/block/blk-mq.c b/block/blk-mq.c index a28957dfb757..5670dfeac85a 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -2873,7 +2873,9 @@ static void blk_mq_exit_hw_queues(struct request_queue *q, queue_for_each_hw_ctx(q, hctx, i) { if (i == nr_queue) break; + mutex_lock(&q->debugfs_mutex); blk_mq_debugfs_unregister_hctx(hctx); + mutex_unlock(&q->debugfs_mutex); blk_mq_exit_hctx(q, set, hctx, i); } }

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE: NA

----------------------------------------

After a disk is created, debugfs inode and dentry will be created together, and the memory used for debugfs can't be freed until disk removal.

The number of debugfs inode and dentry is based on how many cpus and hctxs. For example, testing on a 128-core environemt, with default module parameters, each loop device will cost 1679KB memory, and debugfs will cost 336KB(20%).

The memory cost for debugfs for a disk seems little, but if a big machine contains thousands of disks, the cost will be xxGB. This memory overhead can be avoided by disabling CONFIG_BLK_DEBUG_FS.

This patch add a disk level switch that can enable/disable debugfs dynamically, so that user can disable debugfs if they care about the memory overhead, in the meantime, debugfs can be enabled again in demand.

Signed-off-by: Yu Kuai yukuai3@huawei.com --- block/blk-mq-debugfs.c | 64 +++++++++++++++++++++++++++++++++++++----- block/blk-mq-debugfs.h | 5 ++++ block/blk-sysfs.c | 54 +++++++++++++++++++++++++++++++++++ include/linux/blkdev.h | 4 ++- 4 files changed, 119 insertions(+), 8 deletions(-)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c index e2b8942311c9..f600ac9677a5 100644 --- a/block/blk-mq-debugfs.c +++ b/block/blk-mq-debugfs.c @@ -859,11 +859,20 @@ static void debugfs_create_files(struct dentry *parent, void *data, (void *)attr, &blk_mq_debugfs_fops); }

+static bool blk_mq_debugfs_enabled(struct request_queue *q) +{ + return !IS_ERR_OR_NULL(q->debugfs_dir) && + test_bit(QUEUE_FLAG_DEBUGFS, &q->queue_flags); +} + void blk_mq_debugfs_register(struct request_queue *q) { struct blk_mq_hw_ctx *hctx; int i;

+ if (!blk_mq_debugfs_enabled(q)) + return; + debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);

/* @@ -892,6 +901,46 @@ void blk_mq_debugfs_register(struct request_queue *q) } }

+static void debugfs_remove_files(struct dentry *parent, + const struct blk_mq_debugfs_attr *attr) +{ + if (IS_ERR_OR_NULL(parent)) + return; + + for (; attr->name; attr++) + debugfs_lookup_and_remove(attr->name, parent); +} + +void blk_mq_debugfs_unregister(struct request_queue *q) +{ + struct blk_mq_hw_ctx *hctx; + unsigned long i; + + spin_lock(&q->queue_lock); + if (q->rq_qos) { + struct rq_qos *rqos = q->rq_qos; + + while (rqos) { + rqos->debugfs_dir = NULL; + rqos = rqos->next; + } + } + spin_unlock(&q->queue_lock); + + debugfs_remove_recursive(q->rqos_debugfs_dir); + q->rqos_debugfs_dir = NULL; + + queue_for_each_hw_ctx(q, hctx, i) { + if (hctx->debugfs_dir) + blk_mq_debugfs_unregister_hctx(hctx); + } + + if (q->sched_debugfs_dir) + blk_mq_debugfs_unregister_sched(q); + + debugfs_remove_files(q->debugfs_dir, blk_mq_debugfs_queue_attrs); +} + static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx) { @@ -913,7 +962,7 @@ void blk_mq_debugfs_register_hctx(struct request_queue *q,

lockdep_assert_held(&q->debugfs_mutex);

- if (!q->debugfs_dir) + if (!blk_mq_debugfs_enabled(q)) return;

snprintf(name, sizeof(name), "hctx%u", hctx->queue_num); @@ -929,7 +978,7 @@ void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx) { lockdep_assert_held(&hctx->queue->debugfs_mutex);

- if (!hctx->queue->debugfs_dir) + if (!blk_mq_debugfs_enabled(hctx->queue)) return; debugfs_remove_recursive(hctx->debugfs_dir); hctx->sched_debugfs_dir = NULL; @@ -968,7 +1017,7 @@ void blk_mq_debugfs_register_sched(struct request_queue *q) * If the parent directory has not been created yet, return, we will be * called again later on and the directory/files will be created then. */ - if (!q->debugfs_dir) + if (!blk_mq_debugfs_enabled(q)) return;

if (!e->queue_debugfs_attrs) @@ -991,7 +1040,7 @@ void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos) { lockdep_assert_held(&rqos->q->debugfs_mutex);

- if (!rqos->q->debugfs_dir) + if (!blk_mq_debugfs_enabled(rqos->q)) return; debugfs_remove_recursive(rqos->debugfs_dir); rqos->debugfs_dir = NULL; @@ -1004,7 +1053,8 @@ void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)

lockdep_assert_held(&q->debugfs_mutex);

- if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs) + if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs || + !blk_mq_debugfs_enabled(q)) return;

if (!q->rqos_debugfs_dir) @@ -1024,7 +1074,7 @@ void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,

lockdep_assert_held(&q->debugfs_mutex);

- if (!e->hctx_debugfs_attrs) + if (!e->hctx_debugfs_attrs || !blk_mq_debugfs_enabled(q)) return;

hctx->sched_debugfs_dir = debugfs_create_dir("sched", @@ -1037,7 +1087,7 @@ void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx) { lockdep_assert_held(&hctx->queue->debugfs_mutex);

- if (!hctx->queue->debugfs_dir) + if (!blk_mq_debugfs_enabled(hctx->queue)) return; debugfs_remove_recursive(hctx->sched_debugfs_dir); hctx->sched_debugfs_dir = NULL; diff --git a/block/blk-mq-debugfs.h b/block/blk-mq-debugfs.h index f6898560b1f3..3a2c43a9a0ae 100644 --- a/block/blk-mq-debugfs.h +++ b/block/blk-mq-debugfs.h @@ -19,6 +19,7 @@ int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq); int blk_mq_debugfs_rq_show(struct seq_file *m, void *v);

void blk_mq_debugfs_register(struct request_queue *q); +void blk_mq_debugfs_unregister(struct request_queue *q); void blk_mq_debugfs_register_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx); void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx); @@ -38,6 +39,10 @@ static inline void blk_mq_debugfs_register(struct request_queue *q) { }

+static inline void blk_mq_debugfs_unregister(struct request_queue *q) +{ +} + static inline void blk_mq_debugfs_register_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx) { diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c index dbe5d15c511f..ea2d62ace573 100644 --- a/block/blk-sysfs.c +++ b/block/blk-sysfs.c @@ -619,6 +619,57 @@ QUEUE_RW_ENTRY(queue_iostats, "iostats"); QUEUE_RW_ENTRY(queue_random, "add_random"); QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");

+#ifdef CONFIG_BLK_DEBUG_FS +static ssize_t queue_debugfs_show(struct request_queue *q, char *page) +{ + return queue_var_show(test_bit(QUEUE_FLAG_DEBUGFS, &q->queue_flags), + page); +} + +static ssize_t queue_debugfs_store(struct request_queue *q, const char *page, + size_t count) +{ + unsigned long val; + ssize_t ret; + bool enabled; + int err; + + if (!queue_is_mq(q)) + return count; + + if (!blk_queue_registered(q)) + return -ENODEV; + + ret = queue_var_store(&val, page, count); + if (ret < 0) + return ret; + + err = blk_queue_enter(q, 0); + if (err) + return err; + + mutex_lock(&q->debugfs_mutex); + enabled = test_bit(QUEUE_FLAG_DEBUGFS, &q->queue_flags); + if (!!val == enabled) + goto unlock; + + if (val) { + blk_queue_flag_set(QUEUE_FLAG_DEBUGFS, q); + blk_mq_debugfs_register(q); + } else { + blk_mq_debugfs_unregister(q); + blk_queue_flag_clear(QUEUE_FLAG_DEBUGFS, q); + } + +unlock: + mutex_unlock(&q->debugfs_mutex); + blk_queue_exit(q); + return ret; +} + +QUEUE_RW_ENTRY(queue_debugfs, "debugfs"); +#endif + static struct attribute *queue_attrs[] = { &queue_requests_entry.attr, &queue_ra_entry.attr, @@ -661,6 +712,9 @@ static struct attribute *queue_attrs[] = { &queue_io_timeout_entry.attr, #ifdef CONFIG_BLK_DEV_THROTTLING_LOW &blk_throtl_sample_time_entry.attr, +#endif +#ifdef CONFIG_BLK_DEBUG_FS + &queue_debugfs_entry.attr, #endif NULL, }; diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index 8208922552b5..5ccb9d5f1c29 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -643,10 +643,12 @@ struct request_queue { #define QUEUE_FLAG_NOWAIT 29 /* device supports NOWAIT */ /*at least one blk-mq hctx can't get driver tag */ #define QUEUE_FLAG_HCTX_WAIT 30 +#define QUEUE_FLAG_DEBUGFS 31 /* supports debugfs */

#define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \ (1 << QUEUE_FLAG_SAME_COMP) | \ - (1 << QUEUE_FLAG_NOWAIT)) + (1 << QUEUE_FLAG_NOWAIT) | \ + (1 << QUEUE_FLAG_DEBUGFS))

void blk_queue_flag_set(unsigned int flag, struct request_queue *q); void blk_queue_flag_clear(unsigned int flag, struct request_queue *q);

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I9QJ1S CVE: NA

----------------------------------------

All block layer debugfs is enabled by default, and it can be disabled by adding "blk_mq.enable_debugfs=0" to boot cmd. Noted that user can still enable or disable debugfs by sysfs for each disk.

Signed-off-by: Yu Kuai yukuai3@huawei.com --- arch/arm64/configs/openeuler_defconfig | 1 + arch/powerpc/configs/openeuler_defconfig | 1 + arch/x86/configs/openeuler_defconfig | 1 + block/Kconfig | 11 +++++++++++ block/blk-mq.c | 10 ++++++++++ 5 files changed, 24 insertions(+)

diff --git a/arch/arm64/configs/openeuler_defconfig b/arch/arm64/configs/openeuler_defconfig index 10c941186dbe..50e67e81d7d8 100644 --- a/arch/arm64/configs/openeuler_defconfig +++ b/arch/arm64/configs/openeuler_defconfig @@ -950,6 +950,7 @@ CONFIG_BLK_WBT=y CONFIG_BLK_WBT_MQ=y CONFIG_BLK_DEBUG_FS=y CONFIG_BLK_DEBUG_FS_ZONED=y +CONFIG_BLK_DEBUG_FS_SWITCH=y # CONFIG_BLK_SED_OPAL is not set # CONFIG_BLK_INLINE_ENCRYPTION is not set # CONFIG_BLK_IO_HUNG_TASK_CHECK is not set diff --git a/arch/powerpc/configs/openeuler_defconfig b/arch/powerpc/configs/openeuler_defconfig index 68f642dd09f2..61cd711798b0 100644 --- a/arch/powerpc/configs/openeuler_defconfig +++ b/arch/powerpc/configs/openeuler_defconfig @@ -683,6 +683,7 @@ CONFIG_BLK_CGROUP_IOLATENCY=y CONFIG_BLK_CGROUP_IOCOST=y CONFIG_BLK_WBT_MQ=y CONFIG_BLK_DEBUG_FS=y +CONFIG_BLK_DEBUG_FS_SWITCH=y CONFIG_BLK_DEBUG_FS_ZONED=y CONFIG_BLK_SED_OPAL=y CONFIG_BLK_INLINE_ENCRYPTION=y diff --git a/arch/x86/configs/openeuler_defconfig b/arch/x86/configs/openeuler_defconfig index 8c3ab3abae29..91ab9038c77b 100644 --- a/arch/x86/configs/openeuler_defconfig +++ b/arch/x86/configs/openeuler_defconfig @@ -930,6 +930,7 @@ CONFIG_BLK_WBT=y # CONFIG_BLK_CGROUP_IOCOST is not set CONFIG_BLK_WBT_MQ=y CONFIG_BLK_DEBUG_FS=y +CONFIG_BLK_DEBUG_FS_SWITCH=y CONFIG_BLK_DEBUG_FS_ZONED=y # CONFIG_BLK_SED_OPAL is not set # CONFIG_BLK_INLINE_ENCRYPTION is not set diff --git a/block/Kconfig b/block/Kconfig index 498d18f1d10e..6818b1738dd7 100644 --- a/block/Kconfig +++ b/block/Kconfig @@ -185,6 +185,17 @@ config BLK_DEBUG_FS Unless you are building a kernel for a tiny system, you should say Y here.

+config BLK_DEBUG_FS_SWITCH + bool "Disable/Enable block layer debugfs by demands" + default y + depends on BLK_DEBUG_FS + help + All block layer debugfs is enabled by default, enabling this option + allow user to disable all debugfs by module parameters. Noted that + user can still enable or disable debugfs by sysfs for each disk. + + If memory resources is limited, you should say Y here. + config BLK_DEBUG_FS_ZONED bool default BLK_DEBUG_FS && BLK_DEV_ZONED diff --git a/block/blk-mq.c b/block/blk-mq.c index 5670dfeac85a..ea97231a25ee 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -47,6 +47,11 @@ bool mq_unfair_dtag = true; module_param_named(unfair_dtag, mq_unfair_dtag, bool, 0444);

+#ifdef CONFIG_BLK_DEBUG_FS_SWITCH +bool enable_debugfs = true; +module_param_named(enable_debugfs, enable_debugfs, bool, 0444); +#endif + static DEFINE_PER_CPU(struct list_head, blk_cpu_done);

static void blk_mq_poll_stats_start(struct request_queue *q); @@ -3540,6 +3545,11 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, q->tag_set = set;

q->queue_flags |= QUEUE_FLAG_MQ_DEFAULT; + +#ifdef CONFIG_BLK_DEBUG_FS_SWITCH + if (!enable_debugfs) + blk_queue_flag_clear(QUEUE_FLAG_DEBUGFS, q); +#endif if (set->nr_maps > HCTX_TYPE_POLL && set->map[HCTX_TYPE_POLL].nr_queues) blk_queue_flag_set(QUEUE_FLAG_POLL, q);