]>
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> | |
16 | #include <linux/delay.h> | |
17 | #include <linux/blk-mq.h> | |
18 | #include <linux/nvme.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/parser.h> | |
21 | #include <linux/t10-pi.h> | |
22 | #include "nvmet.h" | |
23 | #include "../host/nvme.h" | |
24 | #include "../host/fabrics.h" | |
25 | ||
26 | #define NVME_LOOP_AQ_DEPTH 256 | |
27 | ||
28 | #define NVME_LOOP_MAX_SEGMENTS 256 | |
29 | ||
30 | /* | |
31 | * We handle AEN commands ourselves and don't even let the | |
32 | * block layer know about them. | |
33 | */ | |
34 | #define NVME_LOOP_NR_AEN_COMMANDS 1 | |
35 | #define NVME_LOOP_AQ_BLKMQ_DEPTH \ | |
36 | (NVME_LOOP_AQ_DEPTH - NVME_LOOP_NR_AEN_COMMANDS) | |
37 | ||
38 | struct nvme_loop_iod { | |
d49187e9 | 39 | struct nvme_request nvme_req; |
3a85a5de CH |
40 | struct nvme_command cmd; |
41 | struct nvme_completion rsp; | |
42 | struct nvmet_req req; | |
43 | struct nvme_loop_queue *queue; | |
44 | struct work_struct work; | |
45 | struct sg_table sg_table; | |
46 | struct scatterlist first_sgl[]; | |
47 | }; | |
48 | ||
49 | struct nvme_loop_ctrl { | |
50 | spinlock_t lock; | |
51 | struct nvme_loop_queue *queues; | |
52 | u32 queue_count; | |
53 | ||
54 | struct blk_mq_tag_set admin_tag_set; | |
55 | ||
56 | struct list_head list; | |
57 | u64 cap; | |
58 | struct blk_mq_tag_set tag_set; | |
59 | struct nvme_loop_iod async_event_iod; | |
60 | struct nvme_ctrl ctrl; | |
61 | ||
62 | struct nvmet_ctrl *target_ctrl; | |
63 | struct work_struct delete_work; | |
64 | struct work_struct reset_work; | |
65 | }; | |
66 | ||
67 | static inline struct nvme_loop_ctrl *to_loop_ctrl(struct nvme_ctrl *ctrl) | |
68 | { | |
69 | return container_of(ctrl, struct nvme_loop_ctrl, ctrl); | |
70 | } | |
71 | ||
72 | struct nvme_loop_queue { | |
73 | struct nvmet_cq nvme_cq; | |
74 | struct nvmet_sq nvme_sq; | |
75 | struct nvme_loop_ctrl *ctrl; | |
76 | }; | |
77 | ||
78 | static struct nvmet_port *nvmet_loop_port; | |
79 | ||
80 | static LIST_HEAD(nvme_loop_ctrl_list); | |
81 | static DEFINE_MUTEX(nvme_loop_ctrl_mutex); | |
82 | ||
83 | static void nvme_loop_queue_response(struct nvmet_req *nvme_req); | |
84 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *ctrl); | |
85 | ||
86 | static struct nvmet_fabrics_ops nvme_loop_ops; | |
87 | ||
88 | static inline int nvme_loop_queue_idx(struct nvme_loop_queue *queue) | |
89 | { | |
90 | return queue - queue->ctrl->queues; | |
91 | } | |
92 | ||
93 | static void nvme_loop_complete_rq(struct request *req) | |
94 | { | |
95 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
96 | int error = 0; | |
97 | ||
98 | nvme_cleanup_cmd(req); | |
99 | sg_free_table_chained(&iod->sg_table, true); | |
100 | ||
101 | if (unlikely(req->errors)) { | |
102 | if (nvme_req_needs_retry(req, req->errors)) { | |
103 | nvme_requeue_req(req); | |
104 | return; | |
105 | } | |
106 | ||
107 | if (req->cmd_type == REQ_TYPE_DRV_PRIV) | |
108 | error = req->errors; | |
109 | else | |
110 | error = nvme_error_status(req->errors); | |
111 | } | |
112 | ||
113 | blk_mq_end_request(req, error); | |
114 | } | |
115 | ||
d49187e9 | 116 | static void nvme_loop_queue_response(struct nvmet_req *req) |
3a85a5de CH |
117 | { |
118 | struct nvme_loop_iod *iod = | |
d49187e9 | 119 | container_of(req, struct nvme_loop_iod, req); |
3a85a5de CH |
120 | struct nvme_completion *cqe = &iod->rsp; |
121 | ||
122 | /* | |
123 | * AEN requests are special as they don't time out and can | |
124 | * survive any kind of queue freeze and often don't respond to | |
125 | * aborts. We don't even bother to allocate a struct request | |
126 | * for them but rather special case them here. | |
127 | */ | |
128 | if (unlikely(nvme_loop_queue_idx(iod->queue) == 0 && | |
129 | cqe->command_id >= NVME_LOOP_AQ_BLKMQ_DEPTH)) { | |
7bf58533 CH |
130 | nvme_complete_async_event(&iod->queue->ctrl->ctrl, cqe->status, |
131 | &cqe->result); | |
3a85a5de | 132 | } else { |
d49187e9 | 133 | struct request *rq = blk_mq_rq_from_pdu(iod); |
3a85a5de | 134 | |
d49187e9 CH |
135 | iod->nvme_req.result = cqe->result; |
136 | blk_mq_complete_request(rq, le16_to_cpu(cqe->status) >> 1); | |
3a85a5de CH |
137 | } |
138 | } | |
139 | ||
140 | static void nvme_loop_execute_work(struct work_struct *work) | |
141 | { | |
142 | struct nvme_loop_iod *iod = | |
143 | container_of(work, struct nvme_loop_iod, work); | |
144 | ||
145 | iod->req.execute(&iod->req); | |
146 | } | |
147 | ||
148 | static enum blk_eh_timer_return | |
149 | nvme_loop_timeout(struct request *rq, bool reserved) | |
150 | { | |
151 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(rq); | |
152 | ||
153 | /* queue error recovery */ | |
154 | schedule_work(&iod->queue->ctrl->reset_work); | |
155 | ||
156 | /* fail with DNR on admin cmd timeout */ | |
157 | rq->errors = NVME_SC_ABORT_REQ | NVME_SC_DNR; | |
158 | ||
159 | return BLK_EH_HANDLED; | |
160 | } | |
161 | ||
162 | static int nvme_loop_queue_rq(struct blk_mq_hw_ctx *hctx, | |
163 | const struct blk_mq_queue_data *bd) | |
164 | { | |
165 | struct nvme_ns *ns = hctx->queue->queuedata; | |
166 | struct nvme_loop_queue *queue = hctx->driver_data; | |
167 | struct request *req = bd->rq; | |
168 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
169 | int ret; | |
170 | ||
171 | ret = nvme_setup_cmd(ns, req, &iod->cmd); | |
bac0000a | 172 | if (ret != BLK_MQ_RQ_QUEUE_OK) |
3a85a5de CH |
173 | return ret; |
174 | ||
175 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
176 | iod->req.port = nvmet_loop_port; | |
177 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, | |
178 | &queue->nvme_sq, &nvme_loop_ops)) { | |
179 | nvme_cleanup_cmd(req); | |
180 | blk_mq_start_request(req); | |
181 | nvme_loop_queue_response(&iod->req); | |
bac0000a | 182 | return BLK_MQ_RQ_QUEUE_OK; |
3a85a5de CH |
183 | } |
184 | ||
185 | if (blk_rq_bytes(req)) { | |
186 | iod->sg_table.sgl = iod->first_sgl; | |
187 | ret = sg_alloc_table_chained(&iod->sg_table, | |
f9d03f96 CH |
188 | blk_rq_nr_phys_segments(req), |
189 | iod->sg_table.sgl); | |
3a85a5de CH |
190 | if (ret) |
191 | return BLK_MQ_RQ_QUEUE_BUSY; | |
192 | ||
193 | iod->req.sg = iod->sg_table.sgl; | |
194 | iod->req.sg_cnt = blk_rq_map_sg(req->q, req, iod->sg_table.sgl); | |
3a85a5de CH |
195 | } |
196 | ||
3a85a5de CH |
197 | blk_mq_start_request(req); |
198 | ||
199 | schedule_work(&iod->work); | |
bac0000a | 200 | return BLK_MQ_RQ_QUEUE_OK; |
3a85a5de CH |
201 | } |
202 | ||
203 | static void nvme_loop_submit_async_event(struct nvme_ctrl *arg, int aer_idx) | |
204 | { | |
205 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(arg); | |
206 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
207 | struct nvme_loop_iod *iod = &ctrl->async_event_iod; | |
208 | ||
209 | memset(&iod->cmd, 0, sizeof(iod->cmd)); | |
210 | iod->cmd.common.opcode = nvme_admin_async_event; | |
211 | iod->cmd.common.command_id = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
212 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
213 | ||
214 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, &queue->nvme_sq, | |
215 | &nvme_loop_ops)) { | |
216 | dev_err(ctrl->ctrl.device, "failed async event work\n"); | |
217 | return; | |
218 | } | |
219 | ||
220 | schedule_work(&iod->work); | |
221 | } | |
222 | ||
223 | static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl, | |
224 | struct nvme_loop_iod *iod, unsigned int queue_idx) | |
225 | { | |
226 | BUG_ON(queue_idx >= ctrl->queue_count); | |
227 | ||
228 | iod->req.cmd = &iod->cmd; | |
229 | iod->req.rsp = &iod->rsp; | |
230 | iod->queue = &ctrl->queues[queue_idx]; | |
231 | INIT_WORK(&iod->work, nvme_loop_execute_work); | |
232 | return 0; | |
233 | } | |
234 | ||
235 | static int nvme_loop_init_request(void *data, struct request *req, | |
236 | unsigned int hctx_idx, unsigned int rq_idx, | |
237 | unsigned int numa_node) | |
238 | { | |
239 | return nvme_loop_init_iod(data, blk_mq_rq_to_pdu(req), hctx_idx + 1); | |
240 | } | |
241 | ||
242 | static int nvme_loop_init_admin_request(void *data, struct request *req, | |
243 | unsigned int hctx_idx, unsigned int rq_idx, | |
244 | unsigned int numa_node) | |
245 | { | |
246 | return nvme_loop_init_iod(data, blk_mq_rq_to_pdu(req), 0); | |
247 | } | |
248 | ||
249 | static int nvme_loop_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
250 | unsigned int hctx_idx) | |
251 | { | |
252 | struct nvme_loop_ctrl *ctrl = data; | |
253 | struct nvme_loop_queue *queue = &ctrl->queues[hctx_idx + 1]; | |
254 | ||
255 | BUG_ON(hctx_idx >= ctrl->queue_count); | |
256 | ||
257 | hctx->driver_data = queue; | |
258 | return 0; | |
259 | } | |
260 | ||
261 | static int nvme_loop_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
262 | unsigned int hctx_idx) | |
263 | { | |
264 | struct nvme_loop_ctrl *ctrl = data; | |
265 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
266 | ||
267 | BUG_ON(hctx_idx != 0); | |
268 | ||
269 | hctx->driver_data = queue; | |
270 | return 0; | |
271 | } | |
272 | ||
273 | static struct blk_mq_ops nvme_loop_mq_ops = { | |
274 | .queue_rq = nvme_loop_queue_rq, | |
275 | .complete = nvme_loop_complete_rq, | |
3a85a5de CH |
276 | .init_request = nvme_loop_init_request, |
277 | .init_hctx = nvme_loop_init_hctx, | |
278 | .timeout = nvme_loop_timeout, | |
279 | }; | |
280 | ||
281 | static struct blk_mq_ops nvme_loop_admin_mq_ops = { | |
282 | .queue_rq = nvme_loop_queue_rq, | |
283 | .complete = nvme_loop_complete_rq, | |
3a85a5de CH |
284 | .init_request = nvme_loop_init_admin_request, |
285 | .init_hctx = nvme_loop_init_admin_hctx, | |
286 | .timeout = nvme_loop_timeout, | |
287 | }; | |
288 | ||
289 | static void nvme_loop_destroy_admin_queue(struct nvme_loop_ctrl *ctrl) | |
290 | { | |
291 | blk_cleanup_queue(ctrl->ctrl.admin_q); | |
292 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
293 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); | |
294 | } | |
295 | ||
296 | static void nvme_loop_free_ctrl(struct nvme_ctrl *nctrl) | |
297 | { | |
298 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
299 | ||
300 | if (list_empty(&ctrl->list)) | |
301 | goto free_ctrl; | |
302 | ||
303 | mutex_lock(&nvme_loop_ctrl_mutex); | |
304 | list_del(&ctrl->list); | |
305 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
306 | ||
307 | if (nctrl->tagset) { | |
308 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
309 | blk_mq_free_tag_set(&ctrl->tag_set); | |
310 | } | |
311 | kfree(ctrl->queues); | |
312 | nvmf_free_options(nctrl->opts); | |
313 | free_ctrl: | |
314 | kfree(ctrl); | |
315 | } | |
316 | ||
317 | static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) | |
318 | { | |
319 | int error; | |
320 | ||
321 | memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set)); | |
322 | ctrl->admin_tag_set.ops = &nvme_loop_admin_mq_ops; | |
323 | ctrl->admin_tag_set.queue_depth = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
324 | ctrl->admin_tag_set.reserved_tags = 2; /* connect + keep-alive */ | |
325 | ctrl->admin_tag_set.numa_node = NUMA_NO_NODE; | |
326 | ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
327 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
328 | ctrl->admin_tag_set.driver_data = ctrl; | |
329 | ctrl->admin_tag_set.nr_hw_queues = 1; | |
330 | ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT; | |
331 | ||
332 | ctrl->queues[0].ctrl = ctrl; | |
333 | error = nvmet_sq_init(&ctrl->queues[0].nvme_sq); | |
334 | if (error) | |
335 | return error; | |
336 | ctrl->queue_count = 1; | |
337 | ||
338 | error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set); | |
339 | if (error) | |
340 | goto out_free_sq; | |
341 | ||
342 | ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set); | |
343 | if (IS_ERR(ctrl->ctrl.admin_q)) { | |
344 | error = PTR_ERR(ctrl->ctrl.admin_q); | |
345 | goto out_free_tagset; | |
346 | } | |
347 | ||
348 | error = nvmf_connect_admin_queue(&ctrl->ctrl); | |
349 | if (error) | |
350 | goto out_cleanup_queue; | |
351 | ||
352 | error = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->cap); | |
353 | if (error) { | |
354 | dev_err(ctrl->ctrl.device, | |
355 | "prop_get NVME_REG_CAP failed\n"); | |
356 | goto out_cleanup_queue; | |
357 | } | |
358 | ||
359 | ctrl->ctrl.sqsize = | |
360 | min_t(int, NVME_CAP_MQES(ctrl->cap) + 1, ctrl->ctrl.sqsize); | |
361 | ||
362 | error = nvme_enable_ctrl(&ctrl->ctrl, ctrl->cap); | |
363 | if (error) | |
364 | goto out_cleanup_queue; | |
365 | ||
366 | ctrl->ctrl.max_hw_sectors = | |
367 | (NVME_LOOP_MAX_SEGMENTS - 1) << (PAGE_SHIFT - 9); | |
368 | ||
369 | error = nvme_init_identify(&ctrl->ctrl); | |
370 | if (error) | |
371 | goto out_cleanup_queue; | |
372 | ||
373 | nvme_start_keep_alive(&ctrl->ctrl); | |
374 | ||
375 | return 0; | |
376 | ||
377 | out_cleanup_queue: | |
378 | blk_cleanup_queue(ctrl->ctrl.admin_q); | |
379 | out_free_tagset: | |
380 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
381 | out_free_sq: | |
382 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); | |
383 | return error; | |
384 | } | |
385 | ||
386 | static void nvme_loop_shutdown_ctrl(struct nvme_loop_ctrl *ctrl) | |
387 | { | |
388 | int i; | |
389 | ||
390 | nvme_stop_keep_alive(&ctrl->ctrl); | |
391 | ||
392 | if (ctrl->queue_count > 1) { | |
393 | nvme_stop_queues(&ctrl->ctrl); | |
394 | blk_mq_tagset_busy_iter(&ctrl->tag_set, | |
395 | nvme_cancel_request, &ctrl->ctrl); | |
396 | ||
397 | for (i = 1; i < ctrl->queue_count; i++) | |
398 | nvmet_sq_destroy(&ctrl->queues[i].nvme_sq); | |
399 | } | |
400 | ||
401 | if (ctrl->ctrl.state == NVME_CTRL_LIVE) | |
402 | nvme_shutdown_ctrl(&ctrl->ctrl); | |
403 | ||
404 | blk_mq_stop_hw_queues(ctrl->ctrl.admin_q); | |
405 | blk_mq_tagset_busy_iter(&ctrl->admin_tag_set, | |
406 | nvme_cancel_request, &ctrl->ctrl); | |
407 | nvme_loop_destroy_admin_queue(ctrl); | |
408 | } | |
409 | ||
410 | static void nvme_loop_del_ctrl_work(struct work_struct *work) | |
411 | { | |
412 | struct nvme_loop_ctrl *ctrl = container_of(work, | |
413 | struct nvme_loop_ctrl, delete_work); | |
414 | ||
3a85a5de | 415 | nvme_uninit_ctrl(&ctrl->ctrl); |
a159c64d | 416 | nvme_loop_shutdown_ctrl(ctrl); |
3a85a5de CH |
417 | nvme_put_ctrl(&ctrl->ctrl); |
418 | } | |
419 | ||
420 | static int __nvme_loop_del_ctrl(struct nvme_loop_ctrl *ctrl) | |
421 | { | |
422 | if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING)) | |
423 | return -EBUSY; | |
424 | ||
425 | if (!schedule_work(&ctrl->delete_work)) | |
426 | return -EBUSY; | |
427 | ||
428 | return 0; | |
429 | } | |
430 | ||
431 | static int nvme_loop_del_ctrl(struct nvme_ctrl *nctrl) | |
432 | { | |
433 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
434 | int ret; | |
435 | ||
436 | ret = __nvme_loop_del_ctrl(ctrl); | |
437 | if (ret) | |
438 | return ret; | |
439 | ||
440 | flush_work(&ctrl->delete_work); | |
441 | ||
442 | return 0; | |
443 | } | |
444 | ||
445 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *nctrl) | |
446 | { | |
447 | struct nvme_loop_ctrl *ctrl; | |
448 | ||
449 | mutex_lock(&nvme_loop_ctrl_mutex); | |
450 | list_for_each_entry(ctrl, &nvme_loop_ctrl_list, list) { | |
451 | if (ctrl->ctrl.cntlid == nctrl->cntlid) | |
452 | __nvme_loop_del_ctrl(ctrl); | |
453 | } | |
454 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
455 | } | |
456 | ||
457 | static void nvme_loop_reset_ctrl_work(struct work_struct *work) | |
458 | { | |
459 | struct nvme_loop_ctrl *ctrl = container_of(work, | |
460 | struct nvme_loop_ctrl, reset_work); | |
461 | bool changed; | |
462 | int i, ret; | |
463 | ||
464 | nvme_loop_shutdown_ctrl(ctrl); | |
465 | ||
466 | ret = nvme_loop_configure_admin_queue(ctrl); | |
467 | if (ret) | |
468 | goto out_disable; | |
469 | ||
470 | for (i = 1; i <= ctrl->ctrl.opts->nr_io_queues; i++) { | |
471 | ctrl->queues[i].ctrl = ctrl; | |
472 | ret = nvmet_sq_init(&ctrl->queues[i].nvme_sq); | |
473 | if (ret) | |
474 | goto out_free_queues; | |
475 | ||
476 | ctrl->queue_count++; | |
477 | } | |
478 | ||
479 | for (i = 1; i <= ctrl->ctrl.opts->nr_io_queues; i++) { | |
480 | ret = nvmf_connect_io_queue(&ctrl->ctrl, i); | |
481 | if (ret) | |
482 | goto out_free_queues; | |
483 | } | |
484 | ||
485 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); | |
486 | WARN_ON_ONCE(!changed); | |
487 | ||
488 | nvme_queue_scan(&ctrl->ctrl); | |
489 | nvme_queue_async_events(&ctrl->ctrl); | |
490 | ||
491 | nvme_start_queues(&ctrl->ctrl); | |
492 | ||
493 | return; | |
494 | ||
495 | out_free_queues: | |
496 | for (i = 1; i < ctrl->queue_count; i++) | |
497 | nvmet_sq_destroy(&ctrl->queues[i].nvme_sq); | |
498 | nvme_loop_destroy_admin_queue(ctrl); | |
499 | out_disable: | |
500 | dev_warn(ctrl->ctrl.device, "Removing after reset failure\n"); | |
3a85a5de CH |
501 | nvme_uninit_ctrl(&ctrl->ctrl); |
502 | nvme_put_ctrl(&ctrl->ctrl); | |
503 | } | |
504 | ||
505 | static int nvme_loop_reset_ctrl(struct nvme_ctrl *nctrl) | |
506 | { | |
507 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
508 | ||
509 | if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_RESETTING)) | |
510 | return -EBUSY; | |
511 | ||
512 | if (!schedule_work(&ctrl->reset_work)) | |
513 | return -EBUSY; | |
514 | ||
515 | flush_work(&ctrl->reset_work); | |
516 | ||
517 | return 0; | |
518 | } | |
519 | ||
520 | static const struct nvme_ctrl_ops nvme_loop_ctrl_ops = { | |
521 | .name = "loop", | |
522 | .module = THIS_MODULE, | |
523 | .is_fabrics = true, | |
524 | .reg_read32 = nvmf_reg_read32, | |
525 | .reg_read64 = nvmf_reg_read64, | |
526 | .reg_write32 = nvmf_reg_write32, | |
527 | .reset_ctrl = nvme_loop_reset_ctrl, | |
528 | .free_ctrl = nvme_loop_free_ctrl, | |
529 | .submit_async_event = nvme_loop_submit_async_event, | |
530 | .delete_ctrl = nvme_loop_del_ctrl, | |
531 | .get_subsysnqn = nvmf_get_subsysnqn, | |
532 | }; | |
533 | ||
534 | static int nvme_loop_create_io_queues(struct nvme_loop_ctrl *ctrl) | |
535 | { | |
536 | struct nvmf_ctrl_options *opts = ctrl->ctrl.opts; | |
537 | int ret, i; | |
538 | ||
539 | ret = nvme_set_queue_count(&ctrl->ctrl, &opts->nr_io_queues); | |
540 | if (ret || !opts->nr_io_queues) | |
541 | return ret; | |
542 | ||
543 | dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n", | |
544 | opts->nr_io_queues); | |
545 | ||
546 | for (i = 1; i <= opts->nr_io_queues; i++) { | |
547 | ctrl->queues[i].ctrl = ctrl; | |
548 | ret = nvmet_sq_init(&ctrl->queues[i].nvme_sq); | |
549 | if (ret) | |
550 | goto out_destroy_queues; | |
551 | ||
552 | ctrl->queue_count++; | |
553 | } | |
554 | ||
555 | memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set)); | |
556 | ctrl->tag_set.ops = &nvme_loop_mq_ops; | |
eadb7cf4 | 557 | ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size; |
3a85a5de CH |
558 | ctrl->tag_set.reserved_tags = 1; /* fabric connect */ |
559 | ctrl->tag_set.numa_node = NUMA_NO_NODE; | |
560 | ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
561 | ctrl->tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
562 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
563 | ctrl->tag_set.driver_data = ctrl; | |
564 | ctrl->tag_set.nr_hw_queues = ctrl->queue_count - 1; | |
565 | ctrl->tag_set.timeout = NVME_IO_TIMEOUT; | |
566 | ctrl->ctrl.tagset = &ctrl->tag_set; | |
567 | ||
568 | ret = blk_mq_alloc_tag_set(&ctrl->tag_set); | |
569 | if (ret) | |
570 | goto out_destroy_queues; | |
571 | ||
572 | ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set); | |
573 | if (IS_ERR(ctrl->ctrl.connect_q)) { | |
574 | ret = PTR_ERR(ctrl->ctrl.connect_q); | |
575 | goto out_free_tagset; | |
576 | } | |
577 | ||
578 | for (i = 1; i <= opts->nr_io_queues; i++) { | |
579 | ret = nvmf_connect_io_queue(&ctrl->ctrl, i); | |
580 | if (ret) | |
581 | goto out_cleanup_connect_q; | |
582 | } | |
583 | ||
584 | return 0; | |
585 | ||
586 | out_cleanup_connect_q: | |
587 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
588 | out_free_tagset: | |
589 | blk_mq_free_tag_set(&ctrl->tag_set); | |
590 | out_destroy_queues: | |
591 | for (i = 1; i < ctrl->queue_count; i++) | |
592 | nvmet_sq_destroy(&ctrl->queues[i].nvme_sq); | |
593 | return ret; | |
594 | } | |
595 | ||
596 | static struct nvme_ctrl *nvme_loop_create_ctrl(struct device *dev, | |
597 | struct nvmf_ctrl_options *opts) | |
598 | { | |
599 | struct nvme_loop_ctrl *ctrl; | |
600 | bool changed; | |
601 | int ret; | |
602 | ||
603 | ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL); | |
604 | if (!ctrl) | |
605 | return ERR_PTR(-ENOMEM); | |
606 | ctrl->ctrl.opts = opts; | |
607 | INIT_LIST_HEAD(&ctrl->list); | |
608 | ||
609 | INIT_WORK(&ctrl->delete_work, nvme_loop_del_ctrl_work); | |
610 | INIT_WORK(&ctrl->reset_work, nvme_loop_reset_ctrl_work); | |
611 | ||
612 | ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_loop_ctrl_ops, | |
613 | 0 /* no quirks, we're perfect! */); | |
614 | if (ret) | |
615 | goto out_put_ctrl; | |
616 | ||
617 | spin_lock_init(&ctrl->lock); | |
618 | ||
619 | ret = -ENOMEM; | |
620 | ||
eadb7cf4 | 621 | ctrl->ctrl.sqsize = opts->queue_size - 1; |
3a85a5de CH |
622 | ctrl->ctrl.kato = opts->kato; |
623 | ||
624 | ctrl->queues = kcalloc(opts->nr_io_queues + 1, sizeof(*ctrl->queues), | |
625 | GFP_KERNEL); | |
626 | if (!ctrl->queues) | |
627 | goto out_uninit_ctrl; | |
628 | ||
629 | ret = nvme_loop_configure_admin_queue(ctrl); | |
630 | if (ret) | |
631 | goto out_free_queues; | |
632 | ||
633 | if (opts->queue_size > ctrl->ctrl.maxcmd) { | |
634 | /* warn if maxcmd is lower than queue_size */ | |
635 | dev_warn(ctrl->ctrl.device, | |
636 | "queue_size %zu > ctrl maxcmd %u, clamping down\n", | |
637 | opts->queue_size, ctrl->ctrl.maxcmd); | |
638 | opts->queue_size = ctrl->ctrl.maxcmd; | |
639 | } | |
640 | ||
641 | if (opts->nr_io_queues) { | |
642 | ret = nvme_loop_create_io_queues(ctrl); | |
643 | if (ret) | |
644 | goto out_remove_admin_queue; | |
645 | } | |
646 | ||
647 | nvme_loop_init_iod(ctrl, &ctrl->async_event_iod, 0); | |
648 | ||
649 | dev_info(ctrl->ctrl.device, | |
650 | "new ctrl: \"%s\"\n", ctrl->ctrl.opts->subsysnqn); | |
651 | ||
652 | kref_get(&ctrl->ctrl.kref); | |
653 | ||
654 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); | |
655 | WARN_ON_ONCE(!changed); | |
656 | ||
657 | mutex_lock(&nvme_loop_ctrl_mutex); | |
658 | list_add_tail(&ctrl->list, &nvme_loop_ctrl_list); | |
659 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
660 | ||
661 | if (opts->nr_io_queues) { | |
662 | nvme_queue_scan(&ctrl->ctrl); | |
663 | nvme_queue_async_events(&ctrl->ctrl); | |
664 | } | |
665 | ||
666 | return &ctrl->ctrl; | |
667 | ||
668 | out_remove_admin_queue: | |
669 | nvme_loop_destroy_admin_queue(ctrl); | |
670 | out_free_queues: | |
671 | kfree(ctrl->queues); | |
672 | out_uninit_ctrl: | |
673 | nvme_uninit_ctrl(&ctrl->ctrl); | |
674 | out_put_ctrl: | |
675 | nvme_put_ctrl(&ctrl->ctrl); | |
676 | if (ret > 0) | |
677 | ret = -EIO; | |
678 | return ERR_PTR(ret); | |
679 | } | |
680 | ||
681 | static int nvme_loop_add_port(struct nvmet_port *port) | |
682 | { | |
683 | /* | |
684 | * XXX: disalow adding more than one port so | |
685 | * there is no connection rejections when a | |
686 | * a subsystem is assigned to a port for which | |
687 | * loop doesn't have a pointer. | |
688 | * This scenario would be possible if we allowed | |
689 | * more than one port to be added and a subsystem | |
690 | * was assigned to a port other than nvmet_loop_port. | |
691 | */ | |
692 | ||
693 | if (nvmet_loop_port) | |
694 | return -EPERM; | |
695 | ||
696 | nvmet_loop_port = port; | |
697 | return 0; | |
698 | } | |
699 | ||
700 | static void nvme_loop_remove_port(struct nvmet_port *port) | |
701 | { | |
702 | if (port == nvmet_loop_port) | |
703 | nvmet_loop_port = NULL; | |
704 | } | |
705 | ||
706 | static struct nvmet_fabrics_ops nvme_loop_ops = { | |
707 | .owner = THIS_MODULE, | |
708 | .type = NVMF_TRTYPE_LOOP, | |
709 | .add_port = nvme_loop_add_port, | |
710 | .remove_port = nvme_loop_remove_port, | |
711 | .queue_response = nvme_loop_queue_response, | |
712 | .delete_ctrl = nvme_loop_delete_ctrl, | |
713 | }; | |
714 | ||
715 | static struct nvmf_transport_ops nvme_loop_transport = { | |
716 | .name = "loop", | |
717 | .create_ctrl = nvme_loop_create_ctrl, | |
718 | }; | |
719 | ||
720 | static int __init nvme_loop_init_module(void) | |
721 | { | |
722 | int ret; | |
723 | ||
724 | ret = nvmet_register_transport(&nvme_loop_ops); | |
725 | if (ret) | |
726 | return ret; | |
727 | nvmf_register_transport(&nvme_loop_transport); | |
728 | return 0; | |
729 | } | |
730 | ||
731 | static void __exit nvme_loop_cleanup_module(void) | |
732 | { | |
733 | struct nvme_loop_ctrl *ctrl, *next; | |
734 | ||
735 | nvmf_unregister_transport(&nvme_loop_transport); | |
736 | nvmet_unregister_transport(&nvme_loop_ops); | |
737 | ||
738 | mutex_lock(&nvme_loop_ctrl_mutex); | |
739 | list_for_each_entry_safe(ctrl, next, &nvme_loop_ctrl_list, list) | |
740 | __nvme_loop_del_ctrl(ctrl); | |
741 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
742 | ||
743 | flush_scheduled_work(); | |
744 | } | |
745 | ||
746 | module_init(nvme_loop_init_module); | |
747 | module_exit(nvme_loop_cleanup_module); | |
748 | ||
749 | MODULE_LICENSE("GPL v2"); | |
750 | MODULE_ALIAS("nvmet-transport-254"); /* 254 == NVMF_TRTYPE_LOOP */ |