]>
Commit | Line | Data |
---|---|---|
3a85a5de CH |
1 | /* |
2 | * NVMe over Fabrics loopback device. | |
3 | * Copyright (c) 2015-2016 HGST, a Western Digital Company. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | */ | |
14 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
15 | #include <linux/scatterlist.h> | |
3a85a5de CH |
16 | #include <linux/blk-mq.h> |
17 | #include <linux/nvme.h> | |
18 | #include <linux/module.h> | |
19 | #include <linux/parser.h> | |
3a85a5de CH |
20 | #include "nvmet.h" |
21 | #include "../host/nvme.h" | |
22 | #include "../host/fabrics.h" | |
23 | ||
24 | #define NVME_LOOP_AQ_DEPTH 256 | |
25 | ||
26 | #define NVME_LOOP_MAX_SEGMENTS 256 | |
27 | ||
28 | /* | |
29 | * We handle AEN commands ourselves and don't even let the | |
30 | * block layer know about them. | |
31 | */ | |
32 | #define NVME_LOOP_NR_AEN_COMMANDS 1 | |
33 | #define NVME_LOOP_AQ_BLKMQ_DEPTH \ | |
34 | (NVME_LOOP_AQ_DEPTH - NVME_LOOP_NR_AEN_COMMANDS) | |
35 | ||
36 | struct nvme_loop_iod { | |
d49187e9 | 37 | struct nvme_request nvme_req; |
3a85a5de CH |
38 | struct nvme_command cmd; |
39 | struct nvme_completion rsp; | |
40 | struct nvmet_req req; | |
41 | struct nvme_loop_queue *queue; | |
42 | struct work_struct work; | |
43 | struct sg_table sg_table; | |
44 | struct scatterlist first_sgl[]; | |
45 | }; | |
46 | ||
47 | struct nvme_loop_ctrl { | |
3a85a5de CH |
48 | struct nvme_loop_queue *queues; |
49 | u32 queue_count; | |
50 | ||
51 | struct blk_mq_tag_set admin_tag_set; | |
52 | ||
53 | struct list_head list; | |
54 | u64 cap; | |
55 | struct blk_mq_tag_set tag_set; | |
56 | struct nvme_loop_iod async_event_iod; | |
57 | struct nvme_ctrl ctrl; | |
58 | ||
59 | struct nvmet_ctrl *target_ctrl; | |
60 | struct work_struct delete_work; | |
61 | struct work_struct reset_work; | |
62 | }; | |
63 | ||
64 | static inline struct nvme_loop_ctrl *to_loop_ctrl(struct nvme_ctrl *ctrl) | |
65 | { | |
66 | return container_of(ctrl, struct nvme_loop_ctrl, ctrl); | |
67 | } | |
68 | ||
69 | struct nvme_loop_queue { | |
70 | struct nvmet_cq nvme_cq; | |
71 | struct nvmet_sq nvme_sq; | |
72 | struct nvme_loop_ctrl *ctrl; | |
73 | }; | |
74 | ||
75 | static struct nvmet_port *nvmet_loop_port; | |
76 | ||
77 | static LIST_HEAD(nvme_loop_ctrl_list); | |
78 | static DEFINE_MUTEX(nvme_loop_ctrl_mutex); | |
79 | ||
80 | static void nvme_loop_queue_response(struct nvmet_req *nvme_req); | |
81 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *ctrl); | |
82 | ||
83 | static struct nvmet_fabrics_ops nvme_loop_ops; | |
84 | ||
85 | static inline int nvme_loop_queue_idx(struct nvme_loop_queue *queue) | |
86 | { | |
87 | return queue - queue->ctrl->queues; | |
88 | } | |
89 | ||
90 | static void nvme_loop_complete_rq(struct request *req) | |
91 | { | |
92 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
3a85a5de CH |
93 | |
94 | nvme_cleanup_cmd(req); | |
95 | sg_free_table_chained(&iod->sg_table, true); | |
77f02a7a | 96 | nvme_complete_rq(req); |
3a85a5de | 97 | } |
3a85a5de | 98 | |
3b068376 SG |
99 | static struct blk_mq_tags *nvme_loop_tagset(struct nvme_loop_queue *queue) |
100 | { | |
101 | u32 queue_idx = nvme_loop_queue_idx(queue); | |
3a85a5de | 102 | |
3b068376 SG |
103 | if (queue_idx == 0) |
104 | return queue->ctrl->admin_tag_set.tags[queue_idx]; | |
105 | return queue->ctrl->tag_set.tags[queue_idx - 1]; | |
3a85a5de CH |
106 | } |
107 | ||
d49187e9 | 108 | static void nvme_loop_queue_response(struct nvmet_req *req) |
3a85a5de | 109 | { |
3b068376 SG |
110 | struct nvme_loop_queue *queue = |
111 | container_of(req->sq, struct nvme_loop_queue, nvme_sq); | |
112 | struct nvme_completion *cqe = req->rsp; | |
3a85a5de CH |
113 | |
114 | /* | |
115 | * AEN requests are special as they don't time out and can | |
116 | * survive any kind of queue freeze and often don't respond to | |
117 | * aborts. We don't even bother to allocate a struct request | |
118 | * for them but rather special case them here. | |
119 | */ | |
3b068376 | 120 | if (unlikely(nvme_loop_queue_idx(queue) == 0 && |
3a85a5de | 121 | cqe->command_id >= NVME_LOOP_AQ_BLKMQ_DEPTH)) { |
3b068376 | 122 | nvme_complete_async_event(&queue->ctrl->ctrl, cqe->status, |
7bf58533 | 123 | &cqe->result); |
3a85a5de | 124 | } else { |
3b068376 | 125 | struct request *rq; |
3b068376 SG |
126 | |
127 | rq = blk_mq_tag_to_rq(nvme_loop_tagset(queue), cqe->command_id); | |
128 | if (!rq) { | |
129 | dev_err(queue->ctrl->ctrl.device, | |
130 | "tag 0x%x on queue %d not found\n", | |
131 | cqe->command_id, nvme_loop_queue_idx(queue)); | |
132 | return; | |
133 | } | |
3a85a5de | 134 | |
27fa9bc5 | 135 | nvme_end_request(rq, cqe->status, cqe->result); |
3a85a5de CH |
136 | } |
137 | } | |
138 | ||
139 | static void nvme_loop_execute_work(struct work_struct *work) | |
140 | { | |
141 | struct nvme_loop_iod *iod = | |
142 | container_of(work, struct nvme_loop_iod, work); | |
143 | ||
144 | iod->req.execute(&iod->req); | |
145 | } | |
146 | ||
147 | static enum blk_eh_timer_return | |
148 | nvme_loop_timeout(struct request *rq, bool reserved) | |
149 | { | |
150 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(rq); | |
151 | ||
152 | /* queue error recovery */ | |
9a6327d2 | 153 | queue_work(nvme_wq, &iod->queue->ctrl->reset_work); |
3a85a5de CH |
154 | |
155 | /* fail with DNR on admin cmd timeout */ | |
27fa9bc5 | 156 | nvme_req(rq)->status = NVME_SC_ABORT_REQ | NVME_SC_DNR; |
3a85a5de CH |
157 | |
158 | return BLK_EH_HANDLED; | |
159 | } | |
160 | ||
fc17b653 | 161 | static blk_status_t nvme_loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
3a85a5de CH |
162 | const struct blk_mq_queue_data *bd) |
163 | { | |
164 | struct nvme_ns *ns = hctx->queue->queuedata; | |
165 | struct nvme_loop_queue *queue = hctx->driver_data; | |
166 | struct request *req = bd->rq; | |
167 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
fc17b653 | 168 | blk_status_t ret; |
3a85a5de CH |
169 | |
170 | ret = nvme_setup_cmd(ns, req, &iod->cmd); | |
fc17b653 | 171 | if (ret) |
3a85a5de CH |
172 | return ret; |
173 | ||
174 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
175 | iod->req.port = nvmet_loop_port; | |
176 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, | |
177 | &queue->nvme_sq, &nvme_loop_ops)) { | |
178 | nvme_cleanup_cmd(req); | |
179 | blk_mq_start_request(req); | |
180 | nvme_loop_queue_response(&iod->req); | |
fc17b653 | 181 | return BLK_STS_OK; |
3a85a5de CH |
182 | } |
183 | ||
184 | if (blk_rq_bytes(req)) { | |
185 | iod->sg_table.sgl = iod->first_sgl; | |
fc17b653 | 186 | if (sg_alloc_table_chained(&iod->sg_table, |
f9d03f96 | 187 | blk_rq_nr_phys_segments(req), |
fc17b653 CH |
188 | iod->sg_table.sgl)) |
189 | return BLK_STS_RESOURCE; | |
3a85a5de CH |
190 | |
191 | iod->req.sg = iod->sg_table.sgl; | |
192 | iod->req.sg_cnt = blk_rq_map_sg(req->q, req, iod->sg_table.sgl); | |
3a85a5de CH |
193 | } |
194 | ||
3a85a5de CH |
195 | blk_mq_start_request(req); |
196 | ||
197 | schedule_work(&iod->work); | |
fc17b653 | 198 | return BLK_STS_OK; |
3a85a5de CH |
199 | } |
200 | ||
201 | static void nvme_loop_submit_async_event(struct nvme_ctrl *arg, int aer_idx) | |
202 | { | |
203 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(arg); | |
204 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
205 | struct nvme_loop_iod *iod = &ctrl->async_event_iod; | |
206 | ||
207 | memset(&iod->cmd, 0, sizeof(iod->cmd)); | |
208 | iod->cmd.common.opcode = nvme_admin_async_event; | |
209 | iod->cmd.common.command_id = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
210 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
211 | ||
212 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, &queue->nvme_sq, | |
213 | &nvme_loop_ops)) { | |
214 | dev_err(ctrl->ctrl.device, "failed async event work\n"); | |
215 | return; | |
216 | } | |
217 | ||
218 | schedule_work(&iod->work); | |
219 | } | |
220 | ||
221 | static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl, | |
222 | struct nvme_loop_iod *iod, unsigned int queue_idx) | |
223 | { | |
3a85a5de CH |
224 | iod->req.cmd = &iod->cmd; |
225 | iod->req.rsp = &iod->rsp; | |
226 | iod->queue = &ctrl->queues[queue_idx]; | |
227 | INIT_WORK(&iod->work, nvme_loop_execute_work); | |
228 | return 0; | |
229 | } | |
230 | ||
d6296d39 CH |
231 | static int nvme_loop_init_request(struct blk_mq_tag_set *set, |
232 | struct request *req, unsigned int hctx_idx, | |
233 | unsigned int numa_node) | |
3a85a5de | 234 | { |
d6296d39 CH |
235 | return nvme_loop_init_iod(set->driver_data, blk_mq_rq_to_pdu(req), |
236 | hctx_idx + 1); | |
3a85a5de CH |
237 | } |
238 | ||
d6296d39 CH |
239 | static int nvme_loop_init_admin_request(struct blk_mq_tag_set *set, |
240 | struct request *req, unsigned int hctx_idx, | |
241 | unsigned int numa_node) | |
3a85a5de | 242 | { |
d6296d39 | 243 | return nvme_loop_init_iod(set->driver_data, blk_mq_rq_to_pdu(req), 0); |
3a85a5de CH |
244 | } |
245 | ||
246 | static int nvme_loop_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
247 | unsigned int hctx_idx) | |
248 | { | |
249 | struct nvme_loop_ctrl *ctrl = data; | |
250 | struct nvme_loop_queue *queue = &ctrl->queues[hctx_idx + 1]; | |
251 | ||
252 | BUG_ON(hctx_idx >= ctrl->queue_count); | |
253 | ||
254 | hctx->driver_data = queue; | |
255 | return 0; | |
256 | } | |
257 | ||
258 | static int nvme_loop_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
259 | unsigned int hctx_idx) | |
260 | { | |
261 | struct nvme_loop_ctrl *ctrl = data; | |
262 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
263 | ||
264 | BUG_ON(hctx_idx != 0); | |
265 | ||
266 | hctx->driver_data = queue; | |
267 | return 0; | |
268 | } | |
269 | ||
f363b089 | 270 | static const struct blk_mq_ops nvme_loop_mq_ops = { |
3a85a5de CH |
271 | .queue_rq = nvme_loop_queue_rq, |
272 | .complete = nvme_loop_complete_rq, | |
3a85a5de CH |
273 | .init_request = nvme_loop_init_request, |
274 | .init_hctx = nvme_loop_init_hctx, | |
275 | .timeout = nvme_loop_timeout, | |
276 | }; | |
277 | ||
f363b089 | 278 | static const struct blk_mq_ops nvme_loop_admin_mq_ops = { |
3a85a5de CH |
279 | .queue_rq = nvme_loop_queue_rq, |
280 | .complete = nvme_loop_complete_rq, | |
3a85a5de CH |
281 | .init_request = nvme_loop_init_admin_request, |
282 | .init_hctx = nvme_loop_init_admin_hctx, | |
283 | .timeout = nvme_loop_timeout, | |
284 | }; | |
285 | ||
286 | static void nvme_loop_destroy_admin_queue(struct nvme_loop_ctrl *ctrl) | |
287 | { | |
e4c5d376 | 288 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); |
3a85a5de CH |
289 | blk_cleanup_queue(ctrl->ctrl.admin_q); |
290 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
3a85a5de CH |
291 | } |
292 | ||
293 | static void nvme_loop_free_ctrl(struct nvme_ctrl *nctrl) | |
294 | { | |
295 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
296 | ||
297 | if (list_empty(&ctrl->list)) | |
298 | goto free_ctrl; | |
299 | ||
300 | mutex_lock(&nvme_loop_ctrl_mutex); | |
301 | list_del(&ctrl->list); | |
302 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
303 | ||
304 | if (nctrl->tagset) { | |
305 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
306 | blk_mq_free_tag_set(&ctrl->tag_set); | |
307 | } | |
308 | kfree(ctrl->queues); | |
309 | nvmf_free_options(nctrl->opts); | |
310 | free_ctrl: | |
311 | kfree(ctrl); | |
312 | } | |
313 | ||
945dd5ba SG |
314 | static void nvme_loop_destroy_io_queues(struct nvme_loop_ctrl *ctrl) |
315 | { | |
316 | int i; | |
317 | ||
318 | for (i = 1; i < ctrl->queue_count; i++) | |
319 | nvmet_sq_destroy(&ctrl->queues[i].nvme_sq); | |
320 | } | |
321 | ||
322 | static int nvme_loop_init_io_queues(struct nvme_loop_ctrl *ctrl) | |
323 | { | |
324 | struct nvmf_ctrl_options *opts = ctrl->ctrl.opts; | |
325 | unsigned int nr_io_queues; | |
326 | int ret, i; | |
327 | ||
328 | nr_io_queues = min(opts->nr_io_queues, num_online_cpus()); | |
329 | ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues); | |
330 | if (ret || !nr_io_queues) | |
331 | return ret; | |
332 | ||
333 | dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n", nr_io_queues); | |
334 | ||
335 | for (i = 1; i <= nr_io_queues; i++) { | |
336 | ctrl->queues[i].ctrl = ctrl; | |
337 | ret = nvmet_sq_init(&ctrl->queues[i].nvme_sq); | |
338 | if (ret) | |
339 | goto out_destroy_queues; | |
340 | ||
341 | ctrl->queue_count++; | |
342 | } | |
343 | ||
344 | return 0; | |
345 | ||
346 | out_destroy_queues: | |
347 | nvme_loop_destroy_io_queues(ctrl); | |
348 | return ret; | |
349 | } | |
350 | ||
297186d6 SG |
351 | static int nvme_loop_connect_io_queues(struct nvme_loop_ctrl *ctrl) |
352 | { | |
353 | int i, ret; | |
354 | ||
355 | for (i = 1; i < ctrl->queue_count; i++) { | |
356 | ret = nvmf_connect_io_queue(&ctrl->ctrl, i); | |
357 | if (ret) | |
358 | return ret; | |
359 | } | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
3a85a5de CH |
364 | static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) |
365 | { | |
366 | int error; | |
367 | ||
368 | memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set)); | |
369 | ctrl->admin_tag_set.ops = &nvme_loop_admin_mq_ops; | |
370 | ctrl->admin_tag_set.queue_depth = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
371 | ctrl->admin_tag_set.reserved_tags = 2; /* connect + keep-alive */ | |
372 | ctrl->admin_tag_set.numa_node = NUMA_NO_NODE; | |
373 | ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
374 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
375 | ctrl->admin_tag_set.driver_data = ctrl; | |
376 | ctrl->admin_tag_set.nr_hw_queues = 1; | |
377 | ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT; | |
378 | ||
379 | ctrl->queues[0].ctrl = ctrl; | |
380 | error = nvmet_sq_init(&ctrl->queues[0].nvme_sq); | |
381 | if (error) | |
382 | return error; | |
383 | ctrl->queue_count = 1; | |
384 | ||
385 | error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set); | |
386 | if (error) | |
387 | goto out_free_sq; | |
388 | ||
389 | ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set); | |
390 | if (IS_ERR(ctrl->ctrl.admin_q)) { | |
391 | error = PTR_ERR(ctrl->ctrl.admin_q); | |
392 | goto out_free_tagset; | |
393 | } | |
394 | ||
395 | error = nvmf_connect_admin_queue(&ctrl->ctrl); | |
396 | if (error) | |
397 | goto out_cleanup_queue; | |
398 | ||
399 | error = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->cap); | |
400 | if (error) { | |
401 | dev_err(ctrl->ctrl.device, | |
402 | "prop_get NVME_REG_CAP failed\n"); | |
403 | goto out_cleanup_queue; | |
404 | } | |
405 | ||
406 | ctrl->ctrl.sqsize = | |
096e9e91 | 407 | min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->ctrl.sqsize); |
3a85a5de CH |
408 | |
409 | error = nvme_enable_ctrl(&ctrl->ctrl, ctrl->cap); | |
410 | if (error) | |
411 | goto out_cleanup_queue; | |
412 | ||
413 | ctrl->ctrl.max_hw_sectors = | |
414 | (NVME_LOOP_MAX_SEGMENTS - 1) << (PAGE_SHIFT - 9); | |
415 | ||
416 | error = nvme_init_identify(&ctrl->ctrl); | |
417 | if (error) | |
418 | goto out_cleanup_queue; | |
419 | ||
420 | nvme_start_keep_alive(&ctrl->ctrl); | |
421 | ||
422 | return 0; | |
423 | ||
424 | out_cleanup_queue: | |
425 | blk_cleanup_queue(ctrl->ctrl.admin_q); | |
426 | out_free_tagset: | |
427 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
428 | out_free_sq: | |
429 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); | |
430 | return error; | |
431 | } | |
432 | ||
433 | static void nvme_loop_shutdown_ctrl(struct nvme_loop_ctrl *ctrl) | |
434 | { | |
3a85a5de CH |
435 | nvme_stop_keep_alive(&ctrl->ctrl); |
436 | ||
437 | if (ctrl->queue_count > 1) { | |
438 | nvme_stop_queues(&ctrl->ctrl); | |
439 | blk_mq_tagset_busy_iter(&ctrl->tag_set, | |
440 | nvme_cancel_request, &ctrl->ctrl); | |
945dd5ba | 441 | nvme_loop_destroy_io_queues(ctrl); |
3a85a5de CH |
442 | } |
443 | ||
444 | if (ctrl->ctrl.state == NVME_CTRL_LIVE) | |
445 | nvme_shutdown_ctrl(&ctrl->ctrl); | |
446 | ||
447 | blk_mq_stop_hw_queues(ctrl->ctrl.admin_q); | |
448 | blk_mq_tagset_busy_iter(&ctrl->admin_tag_set, | |
449 | nvme_cancel_request, &ctrl->ctrl); | |
450 | nvme_loop_destroy_admin_queue(ctrl); | |
451 | } | |
452 | ||
453 | static void nvme_loop_del_ctrl_work(struct work_struct *work) | |
454 | { | |
455 | struct nvme_loop_ctrl *ctrl = container_of(work, | |
456 | struct nvme_loop_ctrl, delete_work); | |
457 | ||
3a85a5de | 458 | nvme_uninit_ctrl(&ctrl->ctrl); |
a159c64d | 459 | nvme_loop_shutdown_ctrl(ctrl); |
3a85a5de CH |
460 | nvme_put_ctrl(&ctrl->ctrl); |
461 | } | |
462 | ||
463 | static int __nvme_loop_del_ctrl(struct nvme_loop_ctrl *ctrl) | |
464 | { | |
465 | if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING)) | |
466 | return -EBUSY; | |
467 | ||
9a6327d2 | 468 | if (!queue_work(nvme_wq, &ctrl->delete_work)) |
3a85a5de CH |
469 | return -EBUSY; |
470 | ||
471 | return 0; | |
472 | } | |
473 | ||
474 | static int nvme_loop_del_ctrl(struct nvme_ctrl *nctrl) | |
475 | { | |
476 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
477 | int ret; | |
478 | ||
479 | ret = __nvme_loop_del_ctrl(ctrl); | |
480 | if (ret) | |
481 | return ret; | |
482 | ||
483 | flush_work(&ctrl->delete_work); | |
484 | ||
485 | return 0; | |
486 | } | |
487 | ||
488 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *nctrl) | |
489 | { | |
490 | struct nvme_loop_ctrl *ctrl; | |
491 | ||
492 | mutex_lock(&nvme_loop_ctrl_mutex); | |
493 | list_for_each_entry(ctrl, &nvme_loop_ctrl_list, list) { | |
494 | if (ctrl->ctrl.cntlid == nctrl->cntlid) | |
495 | __nvme_loop_del_ctrl(ctrl); | |
496 | } | |
497 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
498 | } | |
499 | ||
500 | static void nvme_loop_reset_ctrl_work(struct work_struct *work) | |
501 | { | |
502 | struct nvme_loop_ctrl *ctrl = container_of(work, | |
503 | struct nvme_loop_ctrl, reset_work); | |
504 | bool changed; | |
297186d6 | 505 | int ret; |
3a85a5de CH |
506 | |
507 | nvme_loop_shutdown_ctrl(ctrl); | |
508 | ||
509 | ret = nvme_loop_configure_admin_queue(ctrl); | |
510 | if (ret) | |
511 | goto out_disable; | |
512 | ||
945dd5ba SG |
513 | ret = nvme_loop_init_io_queues(ctrl); |
514 | if (ret) | |
515 | goto out_destroy_admin; | |
3a85a5de | 516 | |
297186d6 SG |
517 | ret = nvme_loop_connect_io_queues(ctrl); |
518 | if (ret) | |
519 | goto out_destroy_io; | |
3a85a5de CH |
520 | |
521 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); | |
522 | WARN_ON_ONCE(!changed); | |
523 | ||
524 | nvme_queue_scan(&ctrl->ctrl); | |
525 | nvme_queue_async_events(&ctrl->ctrl); | |
526 | ||
527 | nvme_start_queues(&ctrl->ctrl); | |
528 | ||
529 | return; | |
530 | ||
945dd5ba SG |
531 | out_destroy_io: |
532 | nvme_loop_destroy_io_queues(ctrl); | |
533 | out_destroy_admin: | |
3a85a5de CH |
534 | nvme_loop_destroy_admin_queue(ctrl); |
535 | out_disable: | |
536 | dev_warn(ctrl->ctrl.device, "Removing after reset failure\n"); | |
3a85a5de CH |
537 | nvme_uninit_ctrl(&ctrl->ctrl); |
538 | nvme_put_ctrl(&ctrl->ctrl); | |
539 | } | |
540 | ||
541 | static int nvme_loop_reset_ctrl(struct nvme_ctrl *nctrl) | |
542 | { | |
543 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
544 | ||
545 | if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RESETTING)) | |
546 | return -EBUSY; | |
547 | ||
9a6327d2 | 548 | if (!queue_work(nvme_wq, &ctrl->reset_work)) |
3a85a5de CH |
549 | return -EBUSY; |
550 | ||
551 | flush_work(&ctrl->reset_work); | |
552 | ||
553 | return 0; | |
554 | } | |
555 | ||
556 | static const struct nvme_ctrl_ops nvme_loop_ctrl_ops = { | |
557 | .name = "loop", | |
558 | .module = THIS_MODULE, | |
d3d5b87d | 559 | .flags = NVME_F_FABRICS, |
3a85a5de CH |
560 | .reg_read32 = nvmf_reg_read32, |
561 | .reg_read64 = nvmf_reg_read64, | |
562 | .reg_write32 = nvmf_reg_write32, | |
563 | .reset_ctrl = nvme_loop_reset_ctrl, | |
564 | .free_ctrl = nvme_loop_free_ctrl, | |
565 | .submit_async_event = nvme_loop_submit_async_event, | |
566 | .delete_ctrl = nvme_loop_del_ctrl, | |
567 | .get_subsysnqn = nvmf_get_subsysnqn, | |
568 | }; | |
569 | ||
570 | static int nvme_loop_create_io_queues(struct nvme_loop_ctrl *ctrl) | |
571 | { | |
297186d6 | 572 | int ret; |
3a85a5de | 573 | |
945dd5ba SG |
574 | ret = nvme_loop_init_io_queues(ctrl); |
575 | if (ret) | |
3a85a5de CH |
576 | return ret; |
577 | ||
3a85a5de CH |
578 | memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set)); |
579 | ctrl->tag_set.ops = &nvme_loop_mq_ops; | |
eadb7cf4 | 580 | ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size; |
3a85a5de CH |
581 | ctrl->tag_set.reserved_tags = 1; /* fabric connect */ |
582 | ctrl->tag_set.numa_node = NUMA_NO_NODE; | |
583 | ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
584 | ctrl->tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
585 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
586 | ctrl->tag_set.driver_data = ctrl; | |
587 | ctrl->tag_set.nr_hw_queues = ctrl->queue_count - 1; | |
588 | ctrl->tag_set.timeout = NVME_IO_TIMEOUT; | |
589 | ctrl->ctrl.tagset = &ctrl->tag_set; | |
590 | ||
591 | ret = blk_mq_alloc_tag_set(&ctrl->tag_set); | |
592 | if (ret) | |
593 | goto out_destroy_queues; | |
594 | ||
595 | ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set); | |
596 | if (IS_ERR(ctrl->ctrl.connect_q)) { | |
597 | ret = PTR_ERR(ctrl->ctrl.connect_q); | |
598 | goto out_free_tagset; | |
599 | } | |
600 | ||
297186d6 SG |
601 | ret = nvme_loop_connect_io_queues(ctrl); |
602 | if (ret) | |
603 | goto out_cleanup_connect_q; | |
3a85a5de CH |
604 | |
605 | return 0; | |
606 | ||
607 | out_cleanup_connect_q: | |
608 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
609 | out_free_tagset: | |
610 | blk_mq_free_tag_set(&ctrl->tag_set); | |
611 | out_destroy_queues: | |
945dd5ba | 612 | nvme_loop_destroy_io_queues(ctrl); |
3a85a5de CH |
613 | return ret; |
614 | } | |
615 | ||
616 | static struct nvme_ctrl *nvme_loop_create_ctrl(struct device *dev, | |
617 | struct nvmf_ctrl_options *opts) | |
618 | { | |
619 | struct nvme_loop_ctrl *ctrl; | |
620 | bool changed; | |
621 | int ret; | |
622 | ||
623 | ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL); | |
624 | if (!ctrl) | |
625 | return ERR_PTR(-ENOMEM); | |
626 | ctrl->ctrl.opts = opts; | |
627 | INIT_LIST_HEAD(&ctrl->list); | |
628 | ||
629 | INIT_WORK(&ctrl->delete_work, nvme_loop_del_ctrl_work); | |
630 | INIT_WORK(&ctrl->reset_work, nvme_loop_reset_ctrl_work); | |
631 | ||
632 | ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_loop_ctrl_ops, | |
633 | 0 /* no quirks, we're perfect! */); | |
634 | if (ret) | |
635 | goto out_put_ctrl; | |
636 | ||
3a85a5de CH |
637 | ret = -ENOMEM; |
638 | ||
eadb7cf4 | 639 | ctrl->ctrl.sqsize = opts->queue_size - 1; |
3a85a5de CH |
640 | ctrl->ctrl.kato = opts->kato; |
641 | ||
642 | ctrl->queues = kcalloc(opts->nr_io_queues + 1, sizeof(*ctrl->queues), | |
643 | GFP_KERNEL); | |
644 | if (!ctrl->queues) | |
645 | goto out_uninit_ctrl; | |
646 | ||
647 | ret = nvme_loop_configure_admin_queue(ctrl); | |
648 | if (ret) | |
649 | goto out_free_queues; | |
650 | ||
651 | if (opts->queue_size > ctrl->ctrl.maxcmd) { | |
652 | /* warn if maxcmd is lower than queue_size */ | |
653 | dev_warn(ctrl->ctrl.device, | |
654 | "queue_size %zu > ctrl maxcmd %u, clamping down\n", | |
655 | opts->queue_size, ctrl->ctrl.maxcmd); | |
656 | opts->queue_size = ctrl->ctrl.maxcmd; | |
657 | } | |
658 | ||
659 | if (opts->nr_io_queues) { | |
660 | ret = nvme_loop_create_io_queues(ctrl); | |
661 | if (ret) | |
662 | goto out_remove_admin_queue; | |
663 | } | |
664 | ||
665 | nvme_loop_init_iod(ctrl, &ctrl->async_event_iod, 0); | |
666 | ||
667 | dev_info(ctrl->ctrl.device, | |
668 | "new ctrl: \"%s\"\n", ctrl->ctrl.opts->subsysnqn); | |
669 | ||
670 | kref_get(&ctrl->ctrl.kref); | |
671 | ||
672 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); | |
673 | WARN_ON_ONCE(!changed); | |
674 | ||
675 | mutex_lock(&nvme_loop_ctrl_mutex); | |
676 | list_add_tail(&ctrl->list, &nvme_loop_ctrl_list); | |
677 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
678 | ||
679 | if (opts->nr_io_queues) { | |
680 | nvme_queue_scan(&ctrl->ctrl); | |
681 | nvme_queue_async_events(&ctrl->ctrl); | |
682 | } | |
683 | ||
684 | return &ctrl->ctrl; | |
685 | ||
686 | out_remove_admin_queue: | |
687 | nvme_loop_destroy_admin_queue(ctrl); | |
688 | out_free_queues: | |
689 | kfree(ctrl->queues); | |
690 | out_uninit_ctrl: | |
691 | nvme_uninit_ctrl(&ctrl->ctrl); | |
692 | out_put_ctrl: | |
693 | nvme_put_ctrl(&ctrl->ctrl); | |
694 | if (ret > 0) | |
695 | ret = -EIO; | |
696 | return ERR_PTR(ret); | |
697 | } | |
698 | ||
699 | static int nvme_loop_add_port(struct nvmet_port *port) | |
700 | { | |
701 | /* | |
702 | * XXX: disalow adding more than one port so | |
703 | * there is no connection rejections when a | |
704 | * a subsystem is assigned to a port for which | |
705 | * loop doesn't have a pointer. | |
706 | * This scenario would be possible if we allowed | |
707 | * more than one port to be added and a subsystem | |
708 | * was assigned to a port other than nvmet_loop_port. | |
709 | */ | |
710 | ||
711 | if (nvmet_loop_port) | |
712 | return -EPERM; | |
713 | ||
714 | nvmet_loop_port = port; | |
715 | return 0; | |
716 | } | |
717 | ||
718 | static void nvme_loop_remove_port(struct nvmet_port *port) | |
719 | { | |
720 | if (port == nvmet_loop_port) | |
721 | nvmet_loop_port = NULL; | |
722 | } | |
723 | ||
724 | static struct nvmet_fabrics_ops nvme_loop_ops = { | |
725 | .owner = THIS_MODULE, | |
726 | .type = NVMF_TRTYPE_LOOP, | |
727 | .add_port = nvme_loop_add_port, | |
728 | .remove_port = nvme_loop_remove_port, | |
729 | .queue_response = nvme_loop_queue_response, | |
730 | .delete_ctrl = nvme_loop_delete_ctrl, | |
731 | }; | |
732 | ||
733 | static struct nvmf_transport_ops nvme_loop_transport = { | |
734 | .name = "loop", | |
735 | .create_ctrl = nvme_loop_create_ctrl, | |
736 | }; | |
737 | ||
738 | static int __init nvme_loop_init_module(void) | |
739 | { | |
740 | int ret; | |
741 | ||
742 | ret = nvmet_register_transport(&nvme_loop_ops); | |
743 | if (ret) | |
744 | return ret; | |
d19eef02 SG |
745 | |
746 | ret = nvmf_register_transport(&nvme_loop_transport); | |
747 | if (ret) | |
748 | nvmet_unregister_transport(&nvme_loop_ops); | |
749 | ||
750 | return ret; | |
3a85a5de CH |
751 | } |
752 | ||
753 | static void __exit nvme_loop_cleanup_module(void) | |
754 | { | |
755 | struct nvme_loop_ctrl *ctrl, *next; | |
756 | ||
757 | nvmf_unregister_transport(&nvme_loop_transport); | |
758 | nvmet_unregister_transport(&nvme_loop_ops); | |
759 | ||
760 | mutex_lock(&nvme_loop_ctrl_mutex); | |
761 | list_for_each_entry_safe(ctrl, next, &nvme_loop_ctrl_list, list) | |
762 | __nvme_loop_del_ctrl(ctrl); | |
763 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
764 | ||
9a6327d2 | 765 | flush_workqueue(nvme_wq); |
3a85a5de CH |
766 | } |
767 | ||
768 | module_init(nvme_loop_init_module); | |
769 | module_exit(nvme_loop_cleanup_module); | |
770 | ||
771 | MODULE_LICENSE("GPL v2"); | |
772 | MODULE_ALIAS("nvmet-transport-254"); /* 254 == NVMF_TRTYPE_LOOP */ |