]>
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 | ||
3a85a5de CH |
24 | #define NVME_LOOP_MAX_SEGMENTS 256 |
25 | ||
26 | /* | |
27 | * We handle AEN commands ourselves and don't even let the | |
28 | * block layer know about them. | |
29 | */ | |
30 | #define NVME_LOOP_NR_AEN_COMMANDS 1 | |
31 | #define NVME_LOOP_AQ_BLKMQ_DEPTH \ | |
7aa1f427 | 32 | (NVME_AQ_DEPTH - NVME_LOOP_NR_AEN_COMMANDS) |
3a85a5de CH |
33 | |
34 | struct nvme_loop_iod { | |
d49187e9 | 35 | struct nvme_request nvme_req; |
3a85a5de CH |
36 | struct nvme_command cmd; |
37 | struct nvme_completion rsp; | |
38 | struct nvmet_req req; | |
39 | struct nvme_loop_queue *queue; | |
40 | struct work_struct work; | |
41 | struct sg_table sg_table; | |
42 | struct scatterlist first_sgl[]; | |
43 | }; | |
44 | ||
45 | struct nvme_loop_ctrl { | |
3a85a5de | 46 | struct nvme_loop_queue *queues; |
3a85a5de CH |
47 | |
48 | struct blk_mq_tag_set admin_tag_set; | |
49 | ||
50 | struct list_head list; | |
3a85a5de CH |
51 | struct blk_mq_tag_set tag_set; |
52 | struct nvme_loop_iod async_event_iod; | |
53 | struct nvme_ctrl ctrl; | |
54 | ||
55 | struct nvmet_ctrl *target_ctrl; | |
56 | struct work_struct delete_work; | |
3a85a5de CH |
57 | }; |
58 | ||
59 | static inline struct nvme_loop_ctrl *to_loop_ctrl(struct nvme_ctrl *ctrl) | |
60 | { | |
61 | return container_of(ctrl, struct nvme_loop_ctrl, ctrl); | |
62 | } | |
63 | ||
64 | struct nvme_loop_queue { | |
65 | struct nvmet_cq nvme_cq; | |
66 | struct nvmet_sq nvme_sq; | |
67 | struct nvme_loop_ctrl *ctrl; | |
68 | }; | |
69 | ||
70 | static struct nvmet_port *nvmet_loop_port; | |
71 | ||
72 | static LIST_HEAD(nvme_loop_ctrl_list); | |
73 | static DEFINE_MUTEX(nvme_loop_ctrl_mutex); | |
74 | ||
75 | static void nvme_loop_queue_response(struct nvmet_req *nvme_req); | |
76 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *ctrl); | |
77 | ||
78 | static struct nvmet_fabrics_ops nvme_loop_ops; | |
79 | ||
80 | static inline int nvme_loop_queue_idx(struct nvme_loop_queue *queue) | |
81 | { | |
82 | return queue - queue->ctrl->queues; | |
83 | } | |
84 | ||
85 | static void nvme_loop_complete_rq(struct request *req) | |
86 | { | |
87 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
3a85a5de CH |
88 | |
89 | nvme_cleanup_cmd(req); | |
90 | sg_free_table_chained(&iod->sg_table, true); | |
77f02a7a | 91 | nvme_complete_rq(req); |
3a85a5de | 92 | } |
3a85a5de | 93 | |
3b068376 SG |
94 | static struct blk_mq_tags *nvme_loop_tagset(struct nvme_loop_queue *queue) |
95 | { | |
96 | u32 queue_idx = nvme_loop_queue_idx(queue); | |
3a85a5de | 97 | |
3b068376 SG |
98 | if (queue_idx == 0) |
99 | return queue->ctrl->admin_tag_set.tags[queue_idx]; | |
100 | return queue->ctrl->tag_set.tags[queue_idx - 1]; | |
3a85a5de CH |
101 | } |
102 | ||
d49187e9 | 103 | static void nvme_loop_queue_response(struct nvmet_req *req) |
3a85a5de | 104 | { |
3b068376 SG |
105 | struct nvme_loop_queue *queue = |
106 | container_of(req->sq, struct nvme_loop_queue, nvme_sq); | |
107 | struct nvme_completion *cqe = req->rsp; | |
3a85a5de CH |
108 | |
109 | /* | |
110 | * AEN requests are special as they don't time out and can | |
111 | * survive any kind of queue freeze and often don't respond to | |
112 | * aborts. We don't even bother to allocate a struct request | |
113 | * for them but rather special case them here. | |
114 | */ | |
3b068376 | 115 | if (unlikely(nvme_loop_queue_idx(queue) == 0 && |
3a85a5de | 116 | cqe->command_id >= NVME_LOOP_AQ_BLKMQ_DEPTH)) { |
3b068376 | 117 | nvme_complete_async_event(&queue->ctrl->ctrl, cqe->status, |
7bf58533 | 118 | &cqe->result); |
3a85a5de | 119 | } else { |
3b068376 | 120 | struct request *rq; |
3b068376 SG |
121 | |
122 | rq = blk_mq_tag_to_rq(nvme_loop_tagset(queue), cqe->command_id); | |
123 | if (!rq) { | |
124 | dev_err(queue->ctrl->ctrl.device, | |
125 | "tag 0x%x on queue %d not found\n", | |
126 | cqe->command_id, nvme_loop_queue_idx(queue)); | |
127 | return; | |
128 | } | |
3a85a5de | 129 | |
27fa9bc5 | 130 | nvme_end_request(rq, cqe->status, cqe->result); |
3a85a5de CH |
131 | } |
132 | } | |
133 | ||
134 | static void nvme_loop_execute_work(struct work_struct *work) | |
135 | { | |
136 | struct nvme_loop_iod *iod = | |
137 | container_of(work, struct nvme_loop_iod, work); | |
138 | ||
139 | iod->req.execute(&iod->req); | |
140 | } | |
141 | ||
142 | static enum blk_eh_timer_return | |
143 | nvme_loop_timeout(struct request *rq, bool reserved) | |
144 | { | |
145 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(rq); | |
146 | ||
147 | /* queue error recovery */ | |
d86c4d8e | 148 | nvme_reset_ctrl(&iod->queue->ctrl->ctrl); |
3a85a5de CH |
149 | |
150 | /* fail with DNR on admin cmd timeout */ | |
27fa9bc5 | 151 | nvme_req(rq)->status = NVME_SC_ABORT_REQ | NVME_SC_DNR; |
3a85a5de CH |
152 | |
153 | return BLK_EH_HANDLED; | |
154 | } | |
155 | ||
fc17b653 | 156 | static blk_status_t nvme_loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
3a85a5de CH |
157 | const struct blk_mq_queue_data *bd) |
158 | { | |
159 | struct nvme_ns *ns = hctx->queue->queuedata; | |
160 | struct nvme_loop_queue *queue = hctx->driver_data; | |
161 | struct request *req = bd->rq; | |
162 | struct nvme_loop_iod *iod = blk_mq_rq_to_pdu(req); | |
fc17b653 | 163 | blk_status_t ret; |
3a85a5de CH |
164 | |
165 | ret = nvme_setup_cmd(ns, req, &iod->cmd); | |
fc17b653 | 166 | if (ret) |
3a85a5de CH |
167 | return ret; |
168 | ||
169 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
170 | iod->req.port = nvmet_loop_port; | |
171 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, | |
172 | &queue->nvme_sq, &nvme_loop_ops)) { | |
173 | nvme_cleanup_cmd(req); | |
174 | blk_mq_start_request(req); | |
175 | nvme_loop_queue_response(&iod->req); | |
fc17b653 | 176 | return BLK_STS_OK; |
3a85a5de CH |
177 | } |
178 | ||
179 | if (blk_rq_bytes(req)) { | |
180 | iod->sg_table.sgl = iod->first_sgl; | |
fc17b653 | 181 | if (sg_alloc_table_chained(&iod->sg_table, |
f9d03f96 | 182 | blk_rq_nr_phys_segments(req), |
fc17b653 CH |
183 | iod->sg_table.sgl)) |
184 | return BLK_STS_RESOURCE; | |
3a85a5de CH |
185 | |
186 | iod->req.sg = iod->sg_table.sgl; | |
187 | iod->req.sg_cnt = blk_rq_map_sg(req->q, req, iod->sg_table.sgl); | |
3a85a5de CH |
188 | } |
189 | ||
3a85a5de CH |
190 | blk_mq_start_request(req); |
191 | ||
192 | schedule_work(&iod->work); | |
fc17b653 | 193 | return BLK_STS_OK; |
3a85a5de CH |
194 | } |
195 | ||
196 | static void nvme_loop_submit_async_event(struct nvme_ctrl *arg, int aer_idx) | |
197 | { | |
198 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(arg); | |
199 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
200 | struct nvme_loop_iod *iod = &ctrl->async_event_iod; | |
201 | ||
202 | memset(&iod->cmd, 0, sizeof(iod->cmd)); | |
203 | iod->cmd.common.opcode = nvme_admin_async_event; | |
204 | iod->cmd.common.command_id = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
205 | iod->cmd.common.flags |= NVME_CMD_SGL_METABUF; | |
206 | ||
207 | if (!nvmet_req_init(&iod->req, &queue->nvme_cq, &queue->nvme_sq, | |
208 | &nvme_loop_ops)) { | |
209 | dev_err(ctrl->ctrl.device, "failed async event work\n"); | |
210 | return; | |
211 | } | |
212 | ||
213 | schedule_work(&iod->work); | |
214 | } | |
215 | ||
216 | static int nvme_loop_init_iod(struct nvme_loop_ctrl *ctrl, | |
217 | struct nvme_loop_iod *iod, unsigned int queue_idx) | |
218 | { | |
3a85a5de CH |
219 | iod->req.cmd = &iod->cmd; |
220 | iod->req.rsp = &iod->rsp; | |
221 | iod->queue = &ctrl->queues[queue_idx]; | |
222 | INIT_WORK(&iod->work, nvme_loop_execute_work); | |
223 | return 0; | |
224 | } | |
225 | ||
d6296d39 CH |
226 | static int nvme_loop_init_request(struct blk_mq_tag_set *set, |
227 | struct request *req, unsigned int hctx_idx, | |
228 | unsigned int numa_node) | |
3a85a5de | 229 | { |
62b83b18 | 230 | struct nvme_loop_ctrl *ctrl = set->driver_data; |
3a85a5de | 231 | |
62b83b18 CH |
232 | return nvme_loop_init_iod(ctrl, blk_mq_rq_to_pdu(req), |
233 | (set == &ctrl->tag_set) ? hctx_idx + 1 : 0); | |
3a85a5de CH |
234 | } |
235 | ||
236 | static int nvme_loop_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
237 | unsigned int hctx_idx) | |
238 | { | |
239 | struct nvme_loop_ctrl *ctrl = data; | |
240 | struct nvme_loop_queue *queue = &ctrl->queues[hctx_idx + 1]; | |
241 | ||
d858e5f0 | 242 | BUG_ON(hctx_idx >= ctrl->ctrl.queue_count); |
3a85a5de CH |
243 | |
244 | hctx->driver_data = queue; | |
245 | return 0; | |
246 | } | |
247 | ||
248 | static int nvme_loop_init_admin_hctx(struct blk_mq_hw_ctx *hctx, void *data, | |
249 | unsigned int hctx_idx) | |
250 | { | |
251 | struct nvme_loop_ctrl *ctrl = data; | |
252 | struct nvme_loop_queue *queue = &ctrl->queues[0]; | |
253 | ||
254 | BUG_ON(hctx_idx != 0); | |
255 | ||
256 | hctx->driver_data = queue; | |
257 | return 0; | |
258 | } | |
259 | ||
f363b089 | 260 | static const struct blk_mq_ops nvme_loop_mq_ops = { |
3a85a5de CH |
261 | .queue_rq = nvme_loop_queue_rq, |
262 | .complete = nvme_loop_complete_rq, | |
3a85a5de CH |
263 | .init_request = nvme_loop_init_request, |
264 | .init_hctx = nvme_loop_init_hctx, | |
265 | .timeout = nvme_loop_timeout, | |
266 | }; | |
267 | ||
f363b089 | 268 | static const struct blk_mq_ops nvme_loop_admin_mq_ops = { |
3a85a5de CH |
269 | .queue_rq = nvme_loop_queue_rq, |
270 | .complete = nvme_loop_complete_rq, | |
62b83b18 | 271 | .init_request = nvme_loop_init_request, |
3a85a5de CH |
272 | .init_hctx = nvme_loop_init_admin_hctx, |
273 | .timeout = nvme_loop_timeout, | |
274 | }; | |
275 | ||
276 | static void nvme_loop_destroy_admin_queue(struct nvme_loop_ctrl *ctrl) | |
277 | { | |
e4c5d376 | 278 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); |
3a85a5de CH |
279 | blk_cleanup_queue(ctrl->ctrl.admin_q); |
280 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
3a85a5de CH |
281 | } |
282 | ||
283 | static void nvme_loop_free_ctrl(struct nvme_ctrl *nctrl) | |
284 | { | |
285 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
286 | ||
287 | if (list_empty(&ctrl->list)) | |
288 | goto free_ctrl; | |
289 | ||
290 | mutex_lock(&nvme_loop_ctrl_mutex); | |
291 | list_del(&ctrl->list); | |
292 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
293 | ||
294 | if (nctrl->tagset) { | |
295 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
296 | blk_mq_free_tag_set(&ctrl->tag_set); | |
297 | } | |
298 | kfree(ctrl->queues); | |
299 | nvmf_free_options(nctrl->opts); | |
300 | free_ctrl: | |
301 | kfree(ctrl); | |
302 | } | |
303 | ||
945dd5ba SG |
304 | static void nvme_loop_destroy_io_queues(struct nvme_loop_ctrl *ctrl) |
305 | { | |
306 | int i; | |
307 | ||
d858e5f0 | 308 | for (i = 1; i < ctrl->ctrl.queue_count; i++) |
945dd5ba SG |
309 | nvmet_sq_destroy(&ctrl->queues[i].nvme_sq); |
310 | } | |
311 | ||
312 | static int nvme_loop_init_io_queues(struct nvme_loop_ctrl *ctrl) | |
313 | { | |
314 | struct nvmf_ctrl_options *opts = ctrl->ctrl.opts; | |
315 | unsigned int nr_io_queues; | |
316 | int ret, i; | |
317 | ||
318 | nr_io_queues = min(opts->nr_io_queues, num_online_cpus()); | |
319 | ret = nvme_set_queue_count(&ctrl->ctrl, &nr_io_queues); | |
320 | if (ret || !nr_io_queues) | |
321 | return ret; | |
322 | ||
323 | dev_info(ctrl->ctrl.device, "creating %d I/O queues.\n", nr_io_queues); | |
324 | ||
325 | for (i = 1; i <= nr_io_queues; i++) { | |
326 | ctrl->queues[i].ctrl = ctrl; | |
327 | ret = nvmet_sq_init(&ctrl->queues[i].nvme_sq); | |
328 | if (ret) | |
329 | goto out_destroy_queues; | |
330 | ||
d858e5f0 | 331 | ctrl->ctrl.queue_count++; |
945dd5ba SG |
332 | } |
333 | ||
334 | return 0; | |
335 | ||
336 | out_destroy_queues: | |
337 | nvme_loop_destroy_io_queues(ctrl); | |
338 | return ret; | |
339 | } | |
340 | ||
297186d6 SG |
341 | static int nvme_loop_connect_io_queues(struct nvme_loop_ctrl *ctrl) |
342 | { | |
343 | int i, ret; | |
344 | ||
d858e5f0 | 345 | for (i = 1; i < ctrl->ctrl.queue_count; i++) { |
297186d6 SG |
346 | ret = nvmf_connect_io_queue(&ctrl->ctrl, i); |
347 | if (ret) | |
348 | return ret; | |
349 | } | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
3a85a5de CH |
354 | static int nvme_loop_configure_admin_queue(struct nvme_loop_ctrl *ctrl) |
355 | { | |
356 | int error; | |
357 | ||
358 | memset(&ctrl->admin_tag_set, 0, sizeof(ctrl->admin_tag_set)); | |
359 | ctrl->admin_tag_set.ops = &nvme_loop_admin_mq_ops; | |
360 | ctrl->admin_tag_set.queue_depth = NVME_LOOP_AQ_BLKMQ_DEPTH; | |
361 | ctrl->admin_tag_set.reserved_tags = 2; /* connect + keep-alive */ | |
362 | ctrl->admin_tag_set.numa_node = NUMA_NO_NODE; | |
363 | ctrl->admin_tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
364 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
365 | ctrl->admin_tag_set.driver_data = ctrl; | |
366 | ctrl->admin_tag_set.nr_hw_queues = 1; | |
367 | ctrl->admin_tag_set.timeout = ADMIN_TIMEOUT; | |
368 | ||
369 | ctrl->queues[0].ctrl = ctrl; | |
370 | error = nvmet_sq_init(&ctrl->queues[0].nvme_sq); | |
371 | if (error) | |
372 | return error; | |
d858e5f0 | 373 | ctrl->ctrl.queue_count = 1; |
3a85a5de CH |
374 | |
375 | error = blk_mq_alloc_tag_set(&ctrl->admin_tag_set); | |
376 | if (error) | |
377 | goto out_free_sq; | |
378 | ||
379 | ctrl->ctrl.admin_q = blk_mq_init_queue(&ctrl->admin_tag_set); | |
380 | if (IS_ERR(ctrl->ctrl.admin_q)) { | |
381 | error = PTR_ERR(ctrl->ctrl.admin_q); | |
382 | goto out_free_tagset; | |
383 | } | |
384 | ||
385 | error = nvmf_connect_admin_queue(&ctrl->ctrl); | |
386 | if (error) | |
387 | goto out_cleanup_queue; | |
388 | ||
20d0dfe6 | 389 | error = nvmf_reg_read64(&ctrl->ctrl, NVME_REG_CAP, &ctrl->ctrl.cap); |
3a85a5de CH |
390 | if (error) { |
391 | dev_err(ctrl->ctrl.device, | |
392 | "prop_get NVME_REG_CAP failed\n"); | |
393 | goto out_cleanup_queue; | |
394 | } | |
395 | ||
396 | ctrl->ctrl.sqsize = | |
20d0dfe6 | 397 | min_t(int, NVME_CAP_MQES(ctrl->ctrl.cap), ctrl->ctrl.sqsize); |
3a85a5de | 398 | |
20d0dfe6 | 399 | error = nvme_enable_ctrl(&ctrl->ctrl, ctrl->ctrl.cap); |
3a85a5de CH |
400 | if (error) |
401 | goto out_cleanup_queue; | |
402 | ||
403 | ctrl->ctrl.max_hw_sectors = | |
404 | (NVME_LOOP_MAX_SEGMENTS - 1) << (PAGE_SHIFT - 9); | |
405 | ||
406 | error = nvme_init_identify(&ctrl->ctrl); | |
407 | if (error) | |
408 | goto out_cleanup_queue; | |
409 | ||
3a85a5de CH |
410 | return 0; |
411 | ||
412 | out_cleanup_queue: | |
413 | blk_cleanup_queue(ctrl->ctrl.admin_q); | |
414 | out_free_tagset: | |
415 | blk_mq_free_tag_set(&ctrl->admin_tag_set); | |
416 | out_free_sq: | |
417 | nvmet_sq_destroy(&ctrl->queues[0].nvme_sq); | |
418 | return error; | |
419 | } | |
420 | ||
421 | static void nvme_loop_shutdown_ctrl(struct nvme_loop_ctrl *ctrl) | |
422 | { | |
d858e5f0 | 423 | if (ctrl->ctrl.queue_count > 1) { |
3a85a5de CH |
424 | nvme_stop_queues(&ctrl->ctrl); |
425 | blk_mq_tagset_busy_iter(&ctrl->tag_set, | |
426 | nvme_cancel_request, &ctrl->ctrl); | |
945dd5ba | 427 | nvme_loop_destroy_io_queues(ctrl); |
3a85a5de CH |
428 | } |
429 | ||
430 | if (ctrl->ctrl.state == NVME_CTRL_LIVE) | |
431 | nvme_shutdown_ctrl(&ctrl->ctrl); | |
432 | ||
c1c0ffff | 433 | blk_mq_quiesce_queue(ctrl->ctrl.admin_q); |
3a85a5de CH |
434 | blk_mq_tagset_busy_iter(&ctrl->admin_tag_set, |
435 | nvme_cancel_request, &ctrl->ctrl); | |
c1c0ffff | 436 | blk_mq_unquiesce_queue(ctrl->ctrl.admin_q); |
3a85a5de CH |
437 | nvme_loop_destroy_admin_queue(ctrl); |
438 | } | |
439 | ||
440 | static void nvme_loop_del_ctrl_work(struct work_struct *work) | |
441 | { | |
442 | struct nvme_loop_ctrl *ctrl = container_of(work, | |
443 | struct nvme_loop_ctrl, delete_work); | |
444 | ||
d09f2b45 SG |
445 | nvme_stop_ctrl(&ctrl->ctrl); |
446 | nvme_remove_namespaces(&ctrl->ctrl); | |
a159c64d | 447 | nvme_loop_shutdown_ctrl(ctrl); |
d09f2b45 | 448 | nvme_uninit_ctrl(&ctrl->ctrl); |
3a85a5de CH |
449 | nvme_put_ctrl(&ctrl->ctrl); |
450 | } | |
451 | ||
452 | static int __nvme_loop_del_ctrl(struct nvme_loop_ctrl *ctrl) | |
453 | { | |
454 | if (!nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_DELETING)) | |
455 | return -EBUSY; | |
456 | ||
9a6327d2 | 457 | if (!queue_work(nvme_wq, &ctrl->delete_work)) |
3a85a5de CH |
458 | return -EBUSY; |
459 | ||
460 | return 0; | |
461 | } | |
462 | ||
463 | static int nvme_loop_del_ctrl(struct nvme_ctrl *nctrl) | |
464 | { | |
465 | struct nvme_loop_ctrl *ctrl = to_loop_ctrl(nctrl); | |
466 | int ret; | |
467 | ||
468 | ret = __nvme_loop_del_ctrl(ctrl); | |
469 | if (ret) | |
470 | return ret; | |
471 | ||
472 | flush_work(&ctrl->delete_work); | |
473 | ||
474 | return 0; | |
475 | } | |
476 | ||
477 | static void nvme_loop_delete_ctrl(struct nvmet_ctrl *nctrl) | |
478 | { | |
479 | struct nvme_loop_ctrl *ctrl; | |
480 | ||
481 | mutex_lock(&nvme_loop_ctrl_mutex); | |
482 | list_for_each_entry(ctrl, &nvme_loop_ctrl_list, list) { | |
483 | if (ctrl->ctrl.cntlid == nctrl->cntlid) | |
484 | __nvme_loop_del_ctrl(ctrl); | |
485 | } | |
486 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
487 | } | |
488 | ||
489 | static void nvme_loop_reset_ctrl_work(struct work_struct *work) | |
490 | { | |
d86c4d8e CH |
491 | struct nvme_loop_ctrl *ctrl = |
492 | container_of(work, struct nvme_loop_ctrl, ctrl.reset_work); | |
3a85a5de | 493 | bool changed; |
297186d6 | 494 | int ret; |
3a85a5de | 495 | |
d09f2b45 | 496 | nvme_stop_ctrl(&ctrl->ctrl); |
3a85a5de CH |
497 | nvme_loop_shutdown_ctrl(ctrl); |
498 | ||
499 | ret = nvme_loop_configure_admin_queue(ctrl); | |
500 | if (ret) | |
501 | goto out_disable; | |
502 | ||
945dd5ba SG |
503 | ret = nvme_loop_init_io_queues(ctrl); |
504 | if (ret) | |
505 | goto out_destroy_admin; | |
3a85a5de | 506 | |
297186d6 SG |
507 | ret = nvme_loop_connect_io_queues(ctrl); |
508 | if (ret) | |
509 | goto out_destroy_io; | |
3a85a5de | 510 | |
4368c39b SG |
511 | blk_mq_update_nr_hw_queues(&ctrl->tag_set, |
512 | ctrl->ctrl.queue_count - 1); | |
513 | ||
3a85a5de CH |
514 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); |
515 | WARN_ON_ONCE(!changed); | |
516 | ||
d09f2b45 | 517 | nvme_start_ctrl(&ctrl->ctrl); |
3a85a5de CH |
518 | |
519 | return; | |
520 | ||
945dd5ba SG |
521 | out_destroy_io: |
522 | nvme_loop_destroy_io_queues(ctrl); | |
523 | out_destroy_admin: | |
3a85a5de CH |
524 | nvme_loop_destroy_admin_queue(ctrl); |
525 | out_disable: | |
526 | dev_warn(ctrl->ctrl.device, "Removing after reset failure\n"); | |
3a85a5de CH |
527 | nvme_uninit_ctrl(&ctrl->ctrl); |
528 | nvme_put_ctrl(&ctrl->ctrl); | |
529 | } | |
530 | ||
3a85a5de CH |
531 | static const struct nvme_ctrl_ops nvme_loop_ctrl_ops = { |
532 | .name = "loop", | |
533 | .module = THIS_MODULE, | |
d3d5b87d | 534 | .flags = NVME_F_FABRICS, |
3a85a5de CH |
535 | .reg_read32 = nvmf_reg_read32, |
536 | .reg_read64 = nvmf_reg_read64, | |
537 | .reg_write32 = nvmf_reg_write32, | |
3a85a5de CH |
538 | .free_ctrl = nvme_loop_free_ctrl, |
539 | .submit_async_event = nvme_loop_submit_async_event, | |
540 | .delete_ctrl = nvme_loop_del_ctrl, | |
3a85a5de CH |
541 | }; |
542 | ||
543 | static int nvme_loop_create_io_queues(struct nvme_loop_ctrl *ctrl) | |
544 | { | |
297186d6 | 545 | int ret; |
3a85a5de | 546 | |
945dd5ba SG |
547 | ret = nvme_loop_init_io_queues(ctrl); |
548 | if (ret) | |
3a85a5de CH |
549 | return ret; |
550 | ||
3a85a5de CH |
551 | memset(&ctrl->tag_set, 0, sizeof(ctrl->tag_set)); |
552 | ctrl->tag_set.ops = &nvme_loop_mq_ops; | |
eadb7cf4 | 553 | ctrl->tag_set.queue_depth = ctrl->ctrl.opts->queue_size; |
3a85a5de CH |
554 | ctrl->tag_set.reserved_tags = 1; /* fabric connect */ |
555 | ctrl->tag_set.numa_node = NUMA_NO_NODE; | |
556 | ctrl->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; | |
557 | ctrl->tag_set.cmd_size = sizeof(struct nvme_loop_iod) + | |
558 | SG_CHUNK_SIZE * sizeof(struct scatterlist); | |
559 | ctrl->tag_set.driver_data = ctrl; | |
d858e5f0 | 560 | ctrl->tag_set.nr_hw_queues = ctrl->ctrl.queue_count - 1; |
3a85a5de CH |
561 | ctrl->tag_set.timeout = NVME_IO_TIMEOUT; |
562 | ctrl->ctrl.tagset = &ctrl->tag_set; | |
563 | ||
564 | ret = blk_mq_alloc_tag_set(&ctrl->tag_set); | |
565 | if (ret) | |
566 | goto out_destroy_queues; | |
567 | ||
568 | ctrl->ctrl.connect_q = blk_mq_init_queue(&ctrl->tag_set); | |
569 | if (IS_ERR(ctrl->ctrl.connect_q)) { | |
570 | ret = PTR_ERR(ctrl->ctrl.connect_q); | |
571 | goto out_free_tagset; | |
572 | } | |
573 | ||
297186d6 SG |
574 | ret = nvme_loop_connect_io_queues(ctrl); |
575 | if (ret) | |
576 | goto out_cleanup_connect_q; | |
3a85a5de CH |
577 | |
578 | return 0; | |
579 | ||
580 | out_cleanup_connect_q: | |
581 | blk_cleanup_queue(ctrl->ctrl.connect_q); | |
582 | out_free_tagset: | |
583 | blk_mq_free_tag_set(&ctrl->tag_set); | |
584 | out_destroy_queues: | |
945dd5ba | 585 | nvme_loop_destroy_io_queues(ctrl); |
3a85a5de CH |
586 | return ret; |
587 | } | |
588 | ||
589 | static struct nvme_ctrl *nvme_loop_create_ctrl(struct device *dev, | |
590 | struct nvmf_ctrl_options *opts) | |
591 | { | |
592 | struct nvme_loop_ctrl *ctrl; | |
593 | bool changed; | |
594 | int ret; | |
595 | ||
596 | ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL); | |
597 | if (!ctrl) | |
598 | return ERR_PTR(-ENOMEM); | |
599 | ctrl->ctrl.opts = opts; | |
600 | INIT_LIST_HEAD(&ctrl->list); | |
601 | ||
602 | INIT_WORK(&ctrl->delete_work, nvme_loop_del_ctrl_work); | |
d86c4d8e | 603 | INIT_WORK(&ctrl->ctrl.reset_work, nvme_loop_reset_ctrl_work); |
3a85a5de CH |
604 | |
605 | ret = nvme_init_ctrl(&ctrl->ctrl, dev, &nvme_loop_ctrl_ops, | |
606 | 0 /* no quirks, we're perfect! */); | |
607 | if (ret) | |
608 | goto out_put_ctrl; | |
609 | ||
3a85a5de CH |
610 | ret = -ENOMEM; |
611 | ||
eadb7cf4 | 612 | ctrl->ctrl.sqsize = opts->queue_size - 1; |
3a85a5de CH |
613 | ctrl->ctrl.kato = opts->kato; |
614 | ||
615 | ctrl->queues = kcalloc(opts->nr_io_queues + 1, sizeof(*ctrl->queues), | |
616 | GFP_KERNEL); | |
617 | if (!ctrl->queues) | |
618 | goto out_uninit_ctrl; | |
619 | ||
620 | ret = nvme_loop_configure_admin_queue(ctrl); | |
621 | if (ret) | |
622 | goto out_free_queues; | |
623 | ||
624 | if (opts->queue_size > ctrl->ctrl.maxcmd) { | |
625 | /* warn if maxcmd is lower than queue_size */ | |
626 | dev_warn(ctrl->ctrl.device, | |
627 | "queue_size %zu > ctrl maxcmd %u, clamping down\n", | |
628 | opts->queue_size, ctrl->ctrl.maxcmd); | |
629 | opts->queue_size = ctrl->ctrl.maxcmd; | |
630 | } | |
631 | ||
632 | if (opts->nr_io_queues) { | |
633 | ret = nvme_loop_create_io_queues(ctrl); | |
634 | if (ret) | |
635 | goto out_remove_admin_queue; | |
636 | } | |
637 | ||
638 | nvme_loop_init_iod(ctrl, &ctrl->async_event_iod, 0); | |
639 | ||
640 | dev_info(ctrl->ctrl.device, | |
641 | "new ctrl: \"%s\"\n", ctrl->ctrl.opts->subsysnqn); | |
642 | ||
643 | kref_get(&ctrl->ctrl.kref); | |
644 | ||
645 | changed = nvme_change_ctrl_state(&ctrl->ctrl, NVME_CTRL_LIVE); | |
646 | WARN_ON_ONCE(!changed); | |
647 | ||
648 | mutex_lock(&nvme_loop_ctrl_mutex); | |
649 | list_add_tail(&ctrl->list, &nvme_loop_ctrl_list); | |
650 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
651 | ||
d09f2b45 | 652 | nvme_start_ctrl(&ctrl->ctrl); |
3a85a5de CH |
653 | |
654 | return &ctrl->ctrl; | |
655 | ||
656 | out_remove_admin_queue: | |
657 | nvme_loop_destroy_admin_queue(ctrl); | |
658 | out_free_queues: | |
659 | kfree(ctrl->queues); | |
660 | out_uninit_ctrl: | |
661 | nvme_uninit_ctrl(&ctrl->ctrl); | |
662 | out_put_ctrl: | |
663 | nvme_put_ctrl(&ctrl->ctrl); | |
664 | if (ret > 0) | |
665 | ret = -EIO; | |
666 | return ERR_PTR(ret); | |
667 | } | |
668 | ||
669 | static int nvme_loop_add_port(struct nvmet_port *port) | |
670 | { | |
671 | /* | |
672 | * XXX: disalow adding more than one port so | |
673 | * there is no connection rejections when a | |
674 | * a subsystem is assigned to a port for which | |
675 | * loop doesn't have a pointer. | |
676 | * This scenario would be possible if we allowed | |
677 | * more than one port to be added and a subsystem | |
678 | * was assigned to a port other than nvmet_loop_port. | |
679 | */ | |
680 | ||
681 | if (nvmet_loop_port) | |
682 | return -EPERM; | |
683 | ||
684 | nvmet_loop_port = port; | |
685 | return 0; | |
686 | } | |
687 | ||
688 | static void nvme_loop_remove_port(struct nvmet_port *port) | |
689 | { | |
690 | if (port == nvmet_loop_port) | |
691 | nvmet_loop_port = NULL; | |
692 | } | |
693 | ||
694 | static struct nvmet_fabrics_ops nvme_loop_ops = { | |
695 | .owner = THIS_MODULE, | |
696 | .type = NVMF_TRTYPE_LOOP, | |
697 | .add_port = nvme_loop_add_port, | |
698 | .remove_port = nvme_loop_remove_port, | |
699 | .queue_response = nvme_loop_queue_response, | |
700 | .delete_ctrl = nvme_loop_delete_ctrl, | |
701 | }; | |
702 | ||
703 | static struct nvmf_transport_ops nvme_loop_transport = { | |
704 | .name = "loop", | |
705 | .create_ctrl = nvme_loop_create_ctrl, | |
706 | }; | |
707 | ||
708 | static int __init nvme_loop_init_module(void) | |
709 | { | |
710 | int ret; | |
711 | ||
712 | ret = nvmet_register_transport(&nvme_loop_ops); | |
713 | if (ret) | |
714 | return ret; | |
d19eef02 SG |
715 | |
716 | ret = nvmf_register_transport(&nvme_loop_transport); | |
717 | if (ret) | |
718 | nvmet_unregister_transport(&nvme_loop_ops); | |
719 | ||
720 | return ret; | |
3a85a5de CH |
721 | } |
722 | ||
723 | static void __exit nvme_loop_cleanup_module(void) | |
724 | { | |
725 | struct nvme_loop_ctrl *ctrl, *next; | |
726 | ||
727 | nvmf_unregister_transport(&nvme_loop_transport); | |
728 | nvmet_unregister_transport(&nvme_loop_ops); | |
729 | ||
730 | mutex_lock(&nvme_loop_ctrl_mutex); | |
731 | list_for_each_entry_safe(ctrl, next, &nvme_loop_ctrl_list, list) | |
732 | __nvme_loop_del_ctrl(ctrl); | |
733 | mutex_unlock(&nvme_loop_ctrl_mutex); | |
734 | ||
9a6327d2 | 735 | flush_workqueue(nvme_wq); |
3a85a5de CH |
736 | } |
737 | ||
738 | module_init(nvme_loop_init_module); | |
739 | module_exit(nvme_loop_cleanup_module); | |
740 | ||
741 | MODULE_LICENSE("GPL v2"); | |
742 | MODULE_ALIAS("nvmet-transport-254"); /* 254 == NVMF_TRTYPE_LOOP */ |