[Devel] [PATCH 2/2] fs/fuse kio: simplify processing and sending kio requests
Kirill Tkhai
ktkhai at virtuozzo.com
Mon May 27 18:02:57 MSK 2019
On 27.05.2019 14:59, Pavel Butsykin wrote:
> This patch makes functions kpcs_req_classify() and pcs_fuse_submit() easier,
> reducing the number of cases to handle kio requests.
>
> Also, this patch allows us to get rid of the additional context switch, when
> kio request is moved to cc->work_queue list under bg_lock. After this patch,
> execution of kpcs_req_send() is available only without fc->bg_lock.
>
> Signed-off-by: Pavel Butsykin <pbutsykin at virtuozzo.com>
> ---
> fs/fuse/dev.c | 30 ++++++++-------
> fs/fuse/fuse_i.h | 2 +-
> fs/fuse/kio/pcs/pcs_fuse_kdirect.c | 76 +++++++++++++-------------------------
> 3 files changed, 43 insertions(+), 65 deletions(-)
>
> diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c
> index 3e0fecbe8db3..1ed8537b2be9 100644
> --- a/fs/fuse/dev.c
> +++ b/fs/fuse/dev.c
> @@ -393,27 +393,39 @@ void fuse_queue_forget(struct fuse_conn *fc, struct fuse_forget_link *forget,
>
> static void flush_bg_queue(struct fuse_conn *fc, struct fuse_iqueue *fiq)
> {
> + struct fuse_req *req, *next;
> + LIST_HEAD(kio_reqs);
> +
> while (fc->active_background < fc->max_background &&
> !list_empty(&fc->bg_queue)) {
> - struct fuse_req *req;
>
> req = list_first_entry(&fc->bg_queue, struct fuse_req, list);
> - list_del_init(&req->list);
> fc->active_background++;
>
> if (fc->kio.op) {
> int ret = fc->kio.op->req_classify(fc, req, true, true);
> if (likely(!ret)) {
> - fc->kio.op->req_send(fc, req, NULL, true, true);
> + list_move_tail(&req->list, &kio_reqs);
req_classify() unlocks bg_lock while req is still in bg_queue list.
This looks as a racy design.
Also, it looks like it racy in practice with: fuse_kill_requests(..., &fc->bg_queue).
I'd remained list_del_init() in the place like we had.
> continue;
> - } else if (ret < 0)
> + } else if (ret < 0) {
> + list_del_init(&req->list);
> continue;
> + }
> }
> + list_del_init(&req->list);
> +
> spin_lock(&fiq->waitq.lock);
> req->in.h.unique = fuse_get_unique(fiq);
> queue_request(fiq, req);
> spin_unlock(&fiq->waitq.lock);
> }
> +
> + spin_unlock(&fc->bg_lock);
> + list_for_each_entry_safe(req, next, &kio_reqs, list) {
> + list_del_init(&req->list);
> + fc->kio.op->req_send(fc, req, NULL, true);
> + }
> + spin_lock(&fc->bg_lock);
> }
>
> /*
> @@ -563,7 +575,7 @@ static void __fuse_request_send(struct fuse_conn *fc, struct fuse_req *req,
> if (fc->kio.op) {
> int ret = fc->kio.op->req_classify(fc, req, false, false);
> if (likely(!ret))
> - return fc->kio.op->req_send(fc, req, ff, false, false);
> + return fc->kio.op->req_send(fc, req, ff, false);
> else if (ret < 0)
> return;
> }
> @@ -656,14 +668,6 @@ void fuse_request_send_background(struct fuse_conn *fc, struct fuse_req *req)
> {
> WARN_ON(!req->end);
>
> - if (fc->kio.op) {
> - int ret = fc->kio.op->req_classify(fc, req, true, false);
> - if (likely(!ret))
> - return fc->kio.op->req_send(fc, req, NULL, true, false);
> - else if (ret < 0)
> - return;
> - }
> -
> if (!fuse_request_queue_background(fc, req)) {
> if (!req->out.h.error)
> req->out.h.error = -ENOTCONN;
> diff --git a/fs/fuse/fuse_i.h b/fs/fuse/fuse_i.h
> index 7978a1d891d2..34a4317a9689 100644
> --- a/fs/fuse/fuse_i.h
> +++ b/fs/fuse/fuse_i.h
> @@ -530,7 +530,7 @@ struct fuse_kio_ops {
> int (*req_classify)(struct fuse_conn *fc, struct fuse_req *req, bool bg,
> bool locked);
> void (*req_send)(struct fuse_conn *fc, struct fuse_req *req,
> - struct fuse_file *ff, bool bg, bool locked);
> + struct fuse_file *ff, bool bg);
>
> /* Inode scope hooks */
> int (*file_open)(struct fuse_conn *fc, struct file *file,
> diff --git a/fs/fuse/kio/pcs/pcs_fuse_kdirect.c b/fs/fuse/kio/pcs/pcs_fuse_kdirect.c
> index a101d3950373..c15b690b0d30 100644
> --- a/fs/fuse/kio/pcs/pcs_fuse_kdirect.c
> +++ b/fs/fuse/kio/pcs/pcs_fuse_kdirect.c
> @@ -947,7 +947,7 @@ out:
> }
>
> static void pcs_fuse_submit(struct pcs_fuse_cluster *pfc, struct fuse_req *req,
> - struct fuse_file *ff, bool async, bool lk)
> + struct fuse_file *ff)
> {
> struct pcs_fuse_req *r = pcs_req_from_fuse(req);
> struct fuse_inode *fi = get_fuse_inode(req->io_inode);
> @@ -1040,11 +1040,7 @@ static void pcs_fuse_submit(struct pcs_fuse_cluster *pfc, struct fuse_req *req,
> error:
> DTRACE("do fuse_request_end req:%p op:%d err:%d\n", req, req->in.h.opcode, req->out.h.error);
>
> - if (lk)
> - spin_unlock(&pfc->fc->bg_lock);
> request_end(pfc->fc, req);
> - if (lk)
> - spin_lock(&pfc->fc->bg_lock);
> return;
>
> submit:
> @@ -1052,11 +1048,7 @@ submit:
> req->out.h.error = -EIO;
> goto error;
> }
> -
> - if (async)
> - pcs_cc_submit(ireq->cc, ireq);
> - else
> - ireq_process(ireq);
> + ireq_process(ireq);
> }
>
> static void kpcs_setattr_end(struct fuse_conn *fc, struct fuse_req *req)
> @@ -1119,7 +1111,7 @@ static void _pcs_shrink_end(struct fuse_conn *fc, struct fuse_req *req)
>
> TRACE("resubmit %p\n", &r->req);
> list_del_init(&ireq->list);
> - pcs_fuse_submit(pfc, &r->req, r->req.ff, true, false);
> + pcs_fuse_submit(pfc, &r->req, r->req.ff);
> }
> }
>
> @@ -1232,64 +1224,46 @@ static int kpcs_req_classify(struct fuse_conn* fc, struct fuse_req *req,
> return 1;
>
> BUG_ON(!pfc);
> - /* At this point request can not belongs to any list
> - * so we can avoid grab fc->lock here at all.
> - */
> - BUG_ON(!list_empty(&req->list));
> -
> DTRACE("Classify req:%p op:%d end:%p bg:%d lk:%d\n", req, req->in.h.opcode,
> req->end, bg, lk);
> ret = pcs_kio_classify_req(fc, req, lk);
> - if (ret) {
> - if (ret < 0) {
> - if (!bg)
> - atomic_inc(&req->count);
> - __clear_bit(FR_PENDING, &req->flags);
> - req->out.h.error = ret;
> - if (lk)
> - spin_unlock(&fc->bg_lock);
> - request_end(fc, req);
> - if (lk)
> - spin_lock(&fc->bg_lock);
> - return ret;
> - }
> - return 1;
> - }
> + if (likely(!ret))
> + return 0;
>
> - if (!lk) {
> - spin_lock(&fc->bg_lock);
> - if (fc->num_background + 1 >= fc->max_background ||
> - !fc->connected) {
> + if (ret < 0) {
> + if (!bg)
> + atomic_inc(&req->count);
> + __clear_bit(FR_PENDING, &req->flags);
> + req->out.h.error = ret;
> + if (lk)
> spin_unlock(&fc->bg_lock);
> - return 1;
> - }
> - fc->num_background++;
> - fc->active_background++;
> -
> - if (fc->num_background == fc->congestion_threshold &&
> - fc->bdi_initialized) {
> - set_bdi_congested(&fc->bdi, BLK_RW_SYNC);
> - set_bdi_congested(&fc->bdi, BLK_RW_ASYNC);
> - }
> - spin_unlock(&fc->bg_lock);
> + request_end(fc, req);
> + if (lk)
> + spin_lock(&fc->bg_lock);
> + return ret;
> }
> - return 0;
> + return 1;
> }
>
> static void kpcs_req_send(struct fuse_conn* fc, struct fuse_req *req,
> - struct fuse_file *ff, bool bg, bool lk)
> + struct fuse_file *ff, bool bg)
> {
> struct pcs_fuse_cluster *pfc = (struct pcs_fuse_cluster*)fc->kio.ctx;
>
> - TRACE("Send req:%p op:%d end:%p bg:%d lk:%d\n",
> - req, req->in.h.opcode, req->end, bg, lk);
> + /* At this point request can not belongs to any list
> + * so we can avoid grab fc->lock here at all.
> + */
> + BUG_ON(!list_empty(&req->list));
> +
> + TRACE("Send req:%p op:%d end:%p bg:%d\n",
> + req, req->in.h.opcode, req->end, bg);
>
> /* request_end below will do fuse_put_request() */
> if (!bg)
> atomic_inc(&req->count);
> __clear_bit(FR_PENDING, &req->flags);
>
> - pcs_fuse_submit(pfc, req, ff ? : req->ff, lk, lk);
> + pcs_fuse_submit(pfc, req, ff ? : req->ff);
> if (!bg)
> wait_event(req->waitq,
> test_bit(FR_FINISHED, &req->flags) && !req->end);
>
More information about the Devel
mailing list