Commit | Line | Data |
---|---|---|
77141dc6 | 1 | // SPDX-License-Identifier: GPL-2.0 |
a07b4970 CH |
2 | /* |
3 | * Common code for the NVMe target. | |
4 | * Copyright (c) 2015-2016 HGST, a Western Digital Company. | |
a07b4970 CH |
5 | */ |
6 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
7 | #include <linux/module.h> | |
28b89118 | 8 | #include <linux/random.h> |
b2d09103 | 9 | #include <linux/rculist.h> |
c6925093 | 10 | #include <linux/pci-p2pdma.h> |
a5dffbb6 | 11 | #include <linux/scatterlist.h> |
b2d09103 | 12 | |
a07b4970 CH |
13 | #include "nvmet.h" |
14 | ||
55eb942e | 15 | struct workqueue_struct *buffered_io_wq; |
e929f06d | 16 | static const struct nvmet_fabrics_ops *nvmet_transports[NVMF_TRTYPE_MAX]; |
15fbad96 | 17 | static DEFINE_IDA(cntlid_ida); |
a07b4970 CH |
18 | |
19 | /* | |
20 | * This read/write semaphore is used to synchronize access to configuration | |
21 | * information on a target system that will result in discovery log page | |
22 | * information change for at least one host. | |
23 | * The full list of resources to protected by this semaphore is: | |
24 | * | |
25 | * - subsystems list | |
26 | * - per-subsystem allowed hosts list | |
27 | * - allow_any_host subsystem attribute | |
28 | * - nvmet_genctr | |
29 | * - the nvmet_transports array | |
30 | * | |
31 | * When updating any of those lists/structures write lock should be obtained, | |
32 | * while when reading (popolating discovery log page or checking host-subsystem | |
33 | * link) read lock is obtained to allow concurrent reads. | |
34 | */ | |
35 | DECLARE_RWSEM(nvmet_config_sem); | |
36 | ||
72efd25d CH |
37 | u32 nvmet_ana_group_enabled[NVMET_MAX_ANAGRPS + 1]; |
38 | u64 nvmet_ana_chgcnt; | |
39 | DECLARE_RWSEM(nvmet_ana_sem); | |
40 | ||
c6aa3542 CK |
41 | inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno) |
42 | { | |
43 | u16 status; | |
44 | ||
45 | switch (errno) { | |
46 | case -ENOSPC: | |
47 | req->error_loc = offsetof(struct nvme_rw_command, length); | |
48 | status = NVME_SC_CAP_EXCEEDED | NVME_SC_DNR; | |
49 | break; | |
50 | case -EREMOTEIO: | |
51 | req->error_loc = offsetof(struct nvme_rw_command, slba); | |
52 | status = NVME_SC_LBA_RANGE | NVME_SC_DNR; | |
53 | break; | |
54 | case -EOPNOTSUPP: | |
55 | req->error_loc = offsetof(struct nvme_common_command, opcode); | |
56 | switch (req->cmd->common.opcode) { | |
57 | case nvme_cmd_dsm: | |
58 | case nvme_cmd_write_zeroes: | |
59 | status = NVME_SC_ONCS_NOT_SUPPORTED | NVME_SC_DNR; | |
60 | break; | |
61 | default: | |
62 | status = NVME_SC_INVALID_OPCODE | NVME_SC_DNR; | |
63 | } | |
64 | break; | |
65 | case -ENODATA: | |
66 | req->error_loc = offsetof(struct nvme_rw_command, nsid); | |
67 | status = NVME_SC_ACCESS_DENIED; | |
68 | break; | |
69 | case -EIO: | |
70 | /* FALLTHRU */ | |
71 | default: | |
72 | req->error_loc = offsetof(struct nvme_common_command, opcode); | |
73 | status = NVME_SC_INTERNAL | NVME_SC_DNR; | |
74 | } | |
75 | ||
76 | return status; | |
77 | } | |
78 | ||
a07b4970 CH |
79 | static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, |
80 | const char *subsysnqn); | |
81 | ||
82 | u16 nvmet_copy_to_sgl(struct nvmet_req *req, off_t off, const void *buf, | |
83 | size_t len) | |
84 | { | |
e81446af CK |
85 | if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) { |
86 | req->error_loc = offsetof(struct nvme_common_command, dptr); | |
a07b4970 | 87 | return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; |
e81446af | 88 | } |
a07b4970 CH |
89 | return 0; |
90 | } | |
91 | ||
92 | u16 nvmet_copy_from_sgl(struct nvmet_req *req, off_t off, void *buf, size_t len) | |
93 | { | |
e81446af CK |
94 | if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) { |
95 | req->error_loc = offsetof(struct nvme_common_command, dptr); | |
a07b4970 | 96 | return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; |
e81446af | 97 | } |
a07b4970 CH |
98 | return 0; |
99 | } | |
100 | ||
c7759fff CH |
101 | u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len) |
102 | { | |
e81446af CK |
103 | if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) { |
104 | req->error_loc = offsetof(struct nvme_common_command, dptr); | |
c7759fff | 105 | return NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR; |
e81446af | 106 | } |
c7759fff CH |
107 | return 0; |
108 | } | |
109 | ||
ba2dec35 RS |
110 | static unsigned int nvmet_max_nsid(struct nvmet_subsys *subsys) |
111 | { | |
112 | struct nvmet_ns *ns; | |
113 | ||
114 | if (list_empty(&subsys->namespaces)) | |
115 | return 0; | |
116 | ||
117 | ns = list_last_entry(&subsys->namespaces, struct nvmet_ns, dev_link); | |
118 | return ns->nsid; | |
119 | } | |
120 | ||
a07b4970 CH |
121 | static u32 nvmet_async_event_result(struct nvmet_async_event *aen) |
122 | { | |
123 | return aen->event_type | (aen->event_info << 8) | (aen->log_page << 16); | |
124 | } | |
125 | ||
126 | static void nvmet_async_events_free(struct nvmet_ctrl *ctrl) | |
127 | { | |
128 | struct nvmet_req *req; | |
129 | ||
130 | while (1) { | |
131 | mutex_lock(&ctrl->lock); | |
132 | if (!ctrl->nr_async_event_cmds) { | |
133 | mutex_unlock(&ctrl->lock); | |
134 | return; | |
135 | } | |
136 | ||
137 | req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds]; | |
138 | mutex_unlock(&ctrl->lock); | |
139 | nvmet_req_complete(req, NVME_SC_INTERNAL | NVME_SC_DNR); | |
140 | } | |
141 | } | |
142 | ||
143 | static void nvmet_async_event_work(struct work_struct *work) | |
144 | { | |
145 | struct nvmet_ctrl *ctrl = | |
146 | container_of(work, struct nvmet_ctrl, async_event_work); | |
147 | struct nvmet_async_event *aen; | |
148 | struct nvmet_req *req; | |
149 | ||
150 | while (1) { | |
151 | mutex_lock(&ctrl->lock); | |
152 | aen = list_first_entry_or_null(&ctrl->async_events, | |
153 | struct nvmet_async_event, entry); | |
154 | if (!aen || !ctrl->nr_async_event_cmds) { | |
155 | mutex_unlock(&ctrl->lock); | |
156 | return; | |
157 | } | |
158 | ||
159 | req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds]; | |
160 | nvmet_set_result(req, nvmet_async_event_result(aen)); | |
161 | ||
162 | list_del(&aen->entry); | |
163 | kfree(aen); | |
164 | ||
165 | mutex_unlock(&ctrl->lock); | |
166 | nvmet_req_complete(req, 0); | |
167 | } | |
168 | } | |
169 | ||
b662a078 | 170 | void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type, |
a07b4970 CH |
171 | u8 event_info, u8 log_page) |
172 | { | |
173 | struct nvmet_async_event *aen; | |
174 | ||
175 | aen = kmalloc(sizeof(*aen), GFP_KERNEL); | |
176 | if (!aen) | |
177 | return; | |
178 | ||
179 | aen->event_type = event_type; | |
180 | aen->event_info = event_info; | |
181 | aen->log_page = log_page; | |
182 | ||
183 | mutex_lock(&ctrl->lock); | |
184 | list_add_tail(&aen->entry, &ctrl->async_events); | |
185 | mutex_unlock(&ctrl->lock); | |
186 | ||
187 | schedule_work(&ctrl->async_event_work); | |
188 | } | |
189 | ||
c16734ea CH |
190 | static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid) |
191 | { | |
192 | u32 i; | |
193 | ||
194 | mutex_lock(&ctrl->lock); | |
195 | if (ctrl->nr_changed_ns > NVME_MAX_CHANGED_NAMESPACES) | |
196 | goto out_unlock; | |
197 | ||
198 | for (i = 0; i < ctrl->nr_changed_ns; i++) { | |
199 | if (ctrl->changed_ns_list[i] == nsid) | |
200 | goto out_unlock; | |
201 | } | |
202 | ||
203 | if (ctrl->nr_changed_ns == NVME_MAX_CHANGED_NAMESPACES) { | |
204 | ctrl->changed_ns_list[0] = cpu_to_le32(0xffffffff); | |
205 | ctrl->nr_changed_ns = U32_MAX; | |
206 | goto out_unlock; | |
207 | } | |
208 | ||
209 | ctrl->changed_ns_list[ctrl->nr_changed_ns++] = nsid; | |
210 | out_unlock: | |
211 | mutex_unlock(&ctrl->lock); | |
212 | } | |
213 | ||
dedf0be5 | 214 | void nvmet_ns_changed(struct nvmet_subsys *subsys, u32 nsid) |
c16734ea CH |
215 | { |
216 | struct nvmet_ctrl *ctrl; | |
217 | ||
013a63ef MG |
218 | lockdep_assert_held(&subsys->lock); |
219 | ||
c16734ea CH |
220 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { |
221 | nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid)); | |
7114ddeb | 222 | if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR)) |
c86b8f7b | 223 | continue; |
c16734ea CH |
224 | nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, |
225 | NVME_AER_NOTICE_NS_CHANGED, | |
226 | NVME_LOG_CHANGED_NS); | |
227 | } | |
228 | } | |
229 | ||
62ac0d32 CH |
230 | void nvmet_send_ana_event(struct nvmet_subsys *subsys, |
231 | struct nvmet_port *port) | |
232 | { | |
233 | struct nvmet_ctrl *ctrl; | |
234 | ||
235 | mutex_lock(&subsys->lock); | |
236 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { | |
237 | if (port && ctrl->port != port) | |
238 | continue; | |
7114ddeb | 239 | if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE)) |
62ac0d32 CH |
240 | continue; |
241 | nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE, | |
242 | NVME_AER_NOTICE_ANA, NVME_LOG_ANA); | |
243 | } | |
244 | mutex_unlock(&subsys->lock); | |
245 | } | |
246 | ||
247 | void nvmet_port_send_ana_event(struct nvmet_port *port) | |
248 | { | |
249 | struct nvmet_subsys_link *p; | |
250 | ||
251 | down_read(&nvmet_config_sem); | |
252 | list_for_each_entry(p, &port->subsystems, entry) | |
253 | nvmet_send_ana_event(p->subsys, port); | |
254 | up_read(&nvmet_config_sem); | |
255 | } | |
256 | ||
e929f06d | 257 | int nvmet_register_transport(const struct nvmet_fabrics_ops *ops) |
a07b4970 CH |
258 | { |
259 | int ret = 0; | |
260 | ||
261 | down_write(&nvmet_config_sem); | |
262 | if (nvmet_transports[ops->type]) | |
263 | ret = -EINVAL; | |
264 | else | |
265 | nvmet_transports[ops->type] = ops; | |
266 | up_write(&nvmet_config_sem); | |
267 | ||
268 | return ret; | |
269 | } | |
270 | EXPORT_SYMBOL_GPL(nvmet_register_transport); | |
271 | ||
e929f06d | 272 | void nvmet_unregister_transport(const struct nvmet_fabrics_ops *ops) |
a07b4970 CH |
273 | { |
274 | down_write(&nvmet_config_sem); | |
275 | nvmet_transports[ops->type] = NULL; | |
276 | up_write(&nvmet_config_sem); | |
277 | } | |
278 | EXPORT_SYMBOL_GPL(nvmet_unregister_transport); | |
279 | ||
280 | int nvmet_enable_port(struct nvmet_port *port) | |
281 | { | |
e929f06d | 282 | const struct nvmet_fabrics_ops *ops; |
a07b4970 CH |
283 | int ret; |
284 | ||
285 | lockdep_assert_held(&nvmet_config_sem); | |
286 | ||
287 | ops = nvmet_transports[port->disc_addr.trtype]; | |
288 | if (!ops) { | |
289 | up_write(&nvmet_config_sem); | |
290 | request_module("nvmet-transport-%d", port->disc_addr.trtype); | |
291 | down_write(&nvmet_config_sem); | |
292 | ops = nvmet_transports[port->disc_addr.trtype]; | |
293 | if (!ops) { | |
294 | pr_err("transport type %d not supported\n", | |
295 | port->disc_addr.trtype); | |
296 | return -EINVAL; | |
297 | } | |
298 | } | |
299 | ||
300 | if (!try_module_get(ops->owner)) | |
301 | return -EINVAL; | |
302 | ||
303 | ret = ops->add_port(port); | |
304 | if (ret) { | |
305 | module_put(ops->owner); | |
306 | return ret; | |
307 | } | |
308 | ||
0d5ee2b2 SW |
309 | /* If the transport didn't set inline_data_size, then disable it. */ |
310 | if (port->inline_data_size < 0) | |
311 | port->inline_data_size = 0; | |
312 | ||
a07b4970 | 313 | port->enabled = true; |
9d09dd8d | 314 | port->tr_ops = ops; |
a07b4970 CH |
315 | return 0; |
316 | } | |
317 | ||
318 | void nvmet_disable_port(struct nvmet_port *port) | |
319 | { | |
e929f06d | 320 | const struct nvmet_fabrics_ops *ops; |
a07b4970 CH |
321 | |
322 | lockdep_assert_held(&nvmet_config_sem); | |
323 | ||
324 | port->enabled = false; | |
9d09dd8d | 325 | port->tr_ops = NULL; |
a07b4970 CH |
326 | |
327 | ops = nvmet_transports[port->disc_addr.trtype]; | |
328 | ops->remove_port(port); | |
329 | module_put(ops->owner); | |
330 | } | |
331 | ||
332 | static void nvmet_keep_alive_timer(struct work_struct *work) | |
333 | { | |
334 | struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work), | |
335 | struct nvmet_ctrl, ka_work); | |
c09305ae SG |
336 | bool cmd_seen = ctrl->cmd_seen; |
337 | ||
338 | ctrl->cmd_seen = false; | |
339 | if (cmd_seen) { | |
340 | pr_debug("ctrl %d reschedule traffic based keep-alive timer\n", | |
341 | ctrl->cntlid); | |
342 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
343 | return; | |
344 | } | |
a07b4970 CH |
345 | |
346 | pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n", | |
347 | ctrl->cntlid, ctrl->kato); | |
348 | ||
23a8ed4a | 349 | nvmet_ctrl_fatal_error(ctrl); |
a07b4970 CH |
350 | } |
351 | ||
352 | static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl) | |
353 | { | |
354 | pr_debug("ctrl %d start keep-alive timer for %d secs\n", | |
355 | ctrl->cntlid, ctrl->kato); | |
356 | ||
357 | INIT_DELAYED_WORK(&ctrl->ka_work, nvmet_keep_alive_timer); | |
358 | schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ); | |
359 | } | |
360 | ||
361 | static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl) | |
362 | { | |
363 | pr_debug("ctrl %d stop keep-alive\n", ctrl->cntlid); | |
364 | ||
365 | cancel_delayed_work_sync(&ctrl->ka_work); | |
366 | } | |
367 | ||
368 | static struct nvmet_ns *__nvmet_find_namespace(struct nvmet_ctrl *ctrl, | |
369 | __le32 nsid) | |
370 | { | |
371 | struct nvmet_ns *ns; | |
372 | ||
373 | list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) { | |
374 | if (ns->nsid == le32_to_cpu(nsid)) | |
375 | return ns; | |
376 | } | |
377 | ||
378 | return NULL; | |
379 | } | |
380 | ||
381 | struct nvmet_ns *nvmet_find_namespace(struct nvmet_ctrl *ctrl, __le32 nsid) | |
382 | { | |
383 | struct nvmet_ns *ns; | |
384 | ||
385 | rcu_read_lock(); | |
386 | ns = __nvmet_find_namespace(ctrl, nsid); | |
387 | if (ns) | |
388 | percpu_ref_get(&ns->ref); | |
389 | rcu_read_unlock(); | |
390 | ||
391 | return ns; | |
392 | } | |
393 | ||
394 | static void nvmet_destroy_namespace(struct percpu_ref *ref) | |
395 | { | |
396 | struct nvmet_ns *ns = container_of(ref, struct nvmet_ns, ref); | |
397 | ||
398 | complete(&ns->disable_done); | |
399 | } | |
400 | ||
401 | void nvmet_put_namespace(struct nvmet_ns *ns) | |
402 | { | |
403 | percpu_ref_put(&ns->ref); | |
404 | } | |
405 | ||
d5eff33e CK |
406 | static void nvmet_ns_dev_disable(struct nvmet_ns *ns) |
407 | { | |
408 | nvmet_bdev_ns_disable(ns); | |
409 | nvmet_file_ns_disable(ns); | |
410 | } | |
411 | ||
c6925093 LG |
412 | static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns) |
413 | { | |
414 | int ret; | |
415 | struct pci_dev *p2p_dev; | |
416 | ||
417 | if (!ns->use_p2pmem) | |
418 | return 0; | |
419 | ||
420 | if (!ns->bdev) { | |
421 | pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n"); | |
422 | return -EINVAL; | |
423 | } | |
424 | ||
425 | if (!blk_queue_pci_p2pdma(ns->bdev->bd_queue)) { | |
426 | pr_err("peer-to-peer DMA is not supported by the driver of %s\n", | |
427 | ns->device_path); | |
428 | return -EINVAL; | |
429 | } | |
430 | ||
431 | if (ns->p2p_dev) { | |
432 | ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true); | |
433 | if (ret < 0) | |
434 | return -EINVAL; | |
435 | } else { | |
436 | /* | |
437 | * Right now we just check that there is p2pmem available so | |
438 | * we can report an error to the user right away if there | |
439 | * is not. We'll find the actual device to use once we | |
440 | * setup the controller when the port's device is available. | |
441 | */ | |
442 | ||
443 | p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns)); | |
444 | if (!p2p_dev) { | |
445 | pr_err("no peer-to-peer memory is available for %s\n", | |
446 | ns->device_path); | |
447 | return -EINVAL; | |
448 | } | |
449 | ||
450 | pci_dev_put(p2p_dev); | |
451 | } | |
452 | ||
453 | return 0; | |
454 | } | |
455 | ||
456 | /* | |
457 | * Note: ctrl->subsys->lock should be held when calling this function | |
458 | */ | |
459 | static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl, | |
460 | struct nvmet_ns *ns) | |
461 | { | |
462 | struct device *clients[2]; | |
463 | struct pci_dev *p2p_dev; | |
464 | int ret; | |
465 | ||
21d3bbdd | 466 | if (!ctrl->p2p_client || !ns->use_p2pmem) |
c6925093 LG |
467 | return; |
468 | ||
469 | if (ns->p2p_dev) { | |
470 | ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true); | |
471 | if (ret < 0) | |
472 | return; | |
473 | ||
474 | p2p_dev = pci_dev_get(ns->p2p_dev); | |
475 | } else { | |
476 | clients[0] = ctrl->p2p_client; | |
477 | clients[1] = nvmet_ns_dev(ns); | |
478 | ||
479 | p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients)); | |
480 | if (!p2p_dev) { | |
481 | pr_err("no peer-to-peer memory is available that's supported by %s and %s\n", | |
482 | dev_name(ctrl->p2p_client), ns->device_path); | |
483 | return; | |
484 | } | |
485 | } | |
486 | ||
487 | ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev); | |
488 | if (ret < 0) | |
489 | pci_dev_put(p2p_dev); | |
490 | ||
491 | pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev), | |
492 | ns->nsid); | |
493 | } | |
494 | ||
a07b4970 CH |
495 | int nvmet_ns_enable(struct nvmet_ns *ns) |
496 | { | |
497 | struct nvmet_subsys *subsys = ns->subsys; | |
c6925093 | 498 | struct nvmet_ctrl *ctrl; |
793c7cfc | 499 | int ret; |
a07b4970 CH |
500 | |
501 | mutex_lock(&subsys->lock); | |
793c7cfc | 502 | ret = 0; |
e4fcf07c | 503 | if (ns->enabled) |
a07b4970 CH |
504 | goto out_unlock; |
505 | ||
e84c2091 MG |
506 | ret = -EMFILE; |
507 | if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES) | |
508 | goto out_unlock; | |
509 | ||
d5eff33e | 510 | ret = nvmet_bdev_ns_enable(ns); |
405a7519 | 511 | if (ret == -ENOTBLK) |
d5eff33e CK |
512 | ret = nvmet_file_ns_enable(ns); |
513 | if (ret) | |
a07b4970 | 514 | goto out_unlock; |
a07b4970 | 515 | |
c6925093 LG |
516 | ret = nvmet_p2pmem_ns_enable(ns); |
517 | if (ret) | |
a536b497 | 518 | goto out_dev_disable; |
c6925093 LG |
519 | |
520 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) | |
521 | nvmet_p2pmem_ns_add_p2p(ctrl, ns); | |
522 | ||
a07b4970 CH |
523 | ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace, |
524 | 0, GFP_KERNEL); | |
525 | if (ret) | |
d5eff33e | 526 | goto out_dev_put; |
a07b4970 CH |
527 | |
528 | if (ns->nsid > subsys->max_nsid) | |
529 | subsys->max_nsid = ns->nsid; | |
530 | ||
531 | /* | |
532 | * The namespaces list needs to be sorted to simplify the implementation | |
533 | * of the Identify Namepace List subcommand. | |
534 | */ | |
535 | if (list_empty(&subsys->namespaces)) { | |
536 | list_add_tail_rcu(&ns->dev_link, &subsys->namespaces); | |
537 | } else { | |
538 | struct nvmet_ns *old; | |
539 | ||
540 | list_for_each_entry_rcu(old, &subsys->namespaces, dev_link) { | |
541 | BUG_ON(ns->nsid == old->nsid); | |
542 | if (ns->nsid < old->nsid) | |
543 | break; | |
544 | } | |
545 | ||
546 | list_add_tail_rcu(&ns->dev_link, &old->dev_link); | |
547 | } | |
793c7cfc | 548 | subsys->nr_namespaces++; |
a07b4970 | 549 | |
c16734ea | 550 | nvmet_ns_changed(subsys, ns->nsid); |
e4fcf07c | 551 | ns->enabled = true; |
a07b4970 CH |
552 | ret = 0; |
553 | out_unlock: | |
554 | mutex_unlock(&subsys->lock); | |
555 | return ret; | |
d5eff33e | 556 | out_dev_put: |
c6925093 LG |
557 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) |
558 | pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid)); | |
a536b497 | 559 | out_dev_disable: |
d5eff33e | 560 | nvmet_ns_dev_disable(ns); |
a07b4970 CH |
561 | goto out_unlock; |
562 | } | |
563 | ||
564 | void nvmet_ns_disable(struct nvmet_ns *ns) | |
565 | { | |
566 | struct nvmet_subsys *subsys = ns->subsys; | |
c6925093 | 567 | struct nvmet_ctrl *ctrl; |
a07b4970 CH |
568 | |
569 | mutex_lock(&subsys->lock); | |
e4fcf07c SA |
570 | if (!ns->enabled) |
571 | goto out_unlock; | |
572 | ||
573 | ns->enabled = false; | |
574 | list_del_rcu(&ns->dev_link); | |
ba2dec35 RS |
575 | if (ns->nsid == subsys->max_nsid) |
576 | subsys->max_nsid = nvmet_max_nsid(subsys); | |
c6925093 LG |
577 | |
578 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) | |
579 | pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid)); | |
580 | ||
a07b4970 CH |
581 | mutex_unlock(&subsys->lock); |
582 | ||
583 | /* | |
584 | * Now that we removed the namespaces from the lookup list, we | |
585 | * can kill the per_cpu ref and wait for any remaining references | |
586 | * to be dropped, as well as a RCU grace period for anyone only | |
587 | * using the namepace under rcu_read_lock(). Note that we can't | |
588 | * use call_rcu here as we need to ensure the namespaces have | |
589 | * been fully destroyed before unloading the module. | |
590 | */ | |
591 | percpu_ref_kill(&ns->ref); | |
592 | synchronize_rcu(); | |
593 | wait_for_completion(&ns->disable_done); | |
594 | percpu_ref_exit(&ns->ref); | |
595 | ||
596 | mutex_lock(&subsys->lock); | |
c6925093 | 597 | |
793c7cfc | 598 | subsys->nr_namespaces--; |
c16734ea | 599 | nvmet_ns_changed(subsys, ns->nsid); |
d5eff33e | 600 | nvmet_ns_dev_disable(ns); |
e4fcf07c | 601 | out_unlock: |
a07b4970 CH |
602 | mutex_unlock(&subsys->lock); |
603 | } | |
604 | ||
605 | void nvmet_ns_free(struct nvmet_ns *ns) | |
606 | { | |
607 | nvmet_ns_disable(ns); | |
608 | ||
72efd25d CH |
609 | down_write(&nvmet_ana_sem); |
610 | nvmet_ana_group_enabled[ns->anagrpid]--; | |
611 | up_write(&nvmet_ana_sem); | |
612 | ||
a07b4970 CH |
613 | kfree(ns->device_path); |
614 | kfree(ns); | |
615 | } | |
616 | ||
617 | struct nvmet_ns *nvmet_ns_alloc(struct nvmet_subsys *subsys, u32 nsid) | |
618 | { | |
619 | struct nvmet_ns *ns; | |
620 | ||
621 | ns = kzalloc(sizeof(*ns), GFP_KERNEL); | |
622 | if (!ns) | |
623 | return NULL; | |
624 | ||
625 | INIT_LIST_HEAD(&ns->dev_link); | |
626 | init_completion(&ns->disable_done); | |
627 | ||
628 | ns->nsid = nsid; | |
629 | ns->subsys = subsys; | |
72efd25d CH |
630 | |
631 | down_write(&nvmet_ana_sem); | |
632 | ns->anagrpid = NVMET_DEFAULT_ANA_GRPID; | |
633 | nvmet_ana_group_enabled[ns->anagrpid]++; | |
634 | up_write(&nvmet_ana_sem); | |
635 | ||
637dc0f3 | 636 | uuid_gen(&ns->uuid); |
55eb942e | 637 | ns->buffered_io = false; |
a07b4970 CH |
638 | |
639 | return ns; | |
640 | } | |
641 | ||
e6a622fd | 642 | static void nvmet_update_sq_head(struct nvmet_req *req) |
a07b4970 | 643 | { |
f9cf2a64 | 644 | if (req->sq->size) { |
e6a622fd SG |
645 | u32 old_sqhd, new_sqhd; |
646 | ||
f9cf2a64 JS |
647 | do { |
648 | old_sqhd = req->sq->sqhd; | |
649 | new_sqhd = (old_sqhd + 1) % req->sq->size; | |
650 | } while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) != | |
651 | old_sqhd); | |
652 | } | |
fc6c9730 | 653 | req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF); |
e6a622fd SG |
654 | } |
655 | ||
76574f37 CK |
656 | static void nvmet_set_error(struct nvmet_req *req, u16 status) |
657 | { | |
658 | struct nvmet_ctrl *ctrl = req->sq->ctrl; | |
659 | struct nvme_error_slot *new_error_slot; | |
660 | unsigned long flags; | |
661 | ||
fc6c9730 | 662 | req->cqe->status = cpu_to_le16(status << 1); |
76574f37 | 663 | |
5698b805 | 664 | if (!ctrl || req->error_loc == NVMET_NO_ERROR_LOC) |
76574f37 CK |
665 | return; |
666 | ||
667 | spin_lock_irqsave(&ctrl->error_lock, flags); | |
668 | ctrl->err_counter++; | |
669 | new_error_slot = | |
670 | &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS]; | |
671 | ||
672 | new_error_slot->error_count = cpu_to_le64(ctrl->err_counter); | |
673 | new_error_slot->sqid = cpu_to_le16(req->sq->qid); | |
674 | new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id); | |
675 | new_error_slot->status_field = cpu_to_le16(status << 1); | |
676 | new_error_slot->param_error_location = cpu_to_le16(req->error_loc); | |
677 | new_error_slot->lba = cpu_to_le64(req->error_slba); | |
678 | new_error_slot->nsid = req->cmd->common.nsid; | |
679 | spin_unlock_irqrestore(&ctrl->error_lock, flags); | |
680 | ||
681 | /* set the more bit for this request */ | |
fc6c9730 | 682 | req->cqe->status |= cpu_to_le16(1 << 14); |
76574f37 CK |
683 | } |
684 | ||
e6a622fd SG |
685 | static void __nvmet_req_complete(struct nvmet_req *req, u16 status) |
686 | { | |
687 | if (!req->sq->sqhd_disabled) | |
688 | nvmet_update_sq_head(req); | |
fc6c9730 MG |
689 | req->cqe->sq_id = cpu_to_le16(req->sq->qid); |
690 | req->cqe->command_id = req->cmd->common.command_id; | |
76574f37 | 691 | |
cb019da3 | 692 | if (unlikely(status)) |
76574f37 | 693 | nvmet_set_error(req, status); |
a07b4970 CH |
694 | if (req->ns) |
695 | nvmet_put_namespace(req->ns); | |
696 | req->ops->queue_response(req); | |
697 | } | |
698 | ||
699 | void nvmet_req_complete(struct nvmet_req *req, u16 status) | |
700 | { | |
701 | __nvmet_req_complete(req, status); | |
702 | percpu_ref_put(&req->sq->ref); | |
703 | } | |
704 | EXPORT_SYMBOL_GPL(nvmet_req_complete); | |
705 | ||
706 | void nvmet_cq_setup(struct nvmet_ctrl *ctrl, struct nvmet_cq *cq, | |
707 | u16 qid, u16 size) | |
708 | { | |
709 | cq->qid = qid; | |
710 | cq->size = size; | |
711 | ||
712 | ctrl->cqs[qid] = cq; | |
713 | } | |
714 | ||
715 | void nvmet_sq_setup(struct nvmet_ctrl *ctrl, struct nvmet_sq *sq, | |
716 | u16 qid, u16 size) | |
717 | { | |
bb1cc747 | 718 | sq->sqhd = 0; |
a07b4970 CH |
719 | sq->qid = qid; |
720 | sq->size = size; | |
721 | ||
722 | ctrl->sqs[qid] = sq; | |
723 | } | |
724 | ||
427242ce SG |
725 | static void nvmet_confirm_sq(struct percpu_ref *ref) |
726 | { | |
727 | struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref); | |
728 | ||
729 | complete(&sq->confirm_done); | |
730 | } | |
731 | ||
a07b4970 CH |
732 | void nvmet_sq_destroy(struct nvmet_sq *sq) |
733 | { | |
734 | /* | |
735 | * If this is the admin queue, complete all AERs so that our | |
736 | * queue doesn't have outstanding requests on it. | |
737 | */ | |
738 | if (sq->ctrl && sq->ctrl->sqs && sq->ctrl->sqs[0] == sq) | |
739 | nvmet_async_events_free(sq->ctrl); | |
427242ce SG |
740 | percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq); |
741 | wait_for_completion(&sq->confirm_done); | |
a07b4970 CH |
742 | wait_for_completion(&sq->free_done); |
743 | percpu_ref_exit(&sq->ref); | |
744 | ||
745 | if (sq->ctrl) { | |
746 | nvmet_ctrl_put(sq->ctrl); | |
747 | sq->ctrl = NULL; /* allows reusing the queue later */ | |
748 | } | |
749 | } | |
750 | EXPORT_SYMBOL_GPL(nvmet_sq_destroy); | |
751 | ||
752 | static void nvmet_sq_free(struct percpu_ref *ref) | |
753 | { | |
754 | struct nvmet_sq *sq = container_of(ref, struct nvmet_sq, ref); | |
755 | ||
756 | complete(&sq->free_done); | |
757 | } | |
758 | ||
759 | int nvmet_sq_init(struct nvmet_sq *sq) | |
760 | { | |
761 | int ret; | |
762 | ||
763 | ret = percpu_ref_init(&sq->ref, nvmet_sq_free, 0, GFP_KERNEL); | |
764 | if (ret) { | |
765 | pr_err("percpu_ref init failed!\n"); | |
766 | return ret; | |
767 | } | |
768 | init_completion(&sq->free_done); | |
427242ce | 769 | init_completion(&sq->confirm_done); |
a07b4970 CH |
770 | |
771 | return 0; | |
772 | } | |
773 | EXPORT_SYMBOL_GPL(nvmet_sq_init); | |
774 | ||
72efd25d CH |
775 | static inline u16 nvmet_check_ana_state(struct nvmet_port *port, |
776 | struct nvmet_ns *ns) | |
777 | { | |
778 | enum nvme_ana_state state = port->ana_state[ns->anagrpid]; | |
779 | ||
780 | if (unlikely(state == NVME_ANA_INACCESSIBLE)) | |
781 | return NVME_SC_ANA_INACCESSIBLE; | |
782 | if (unlikely(state == NVME_ANA_PERSISTENT_LOSS)) | |
783 | return NVME_SC_ANA_PERSISTENT_LOSS; | |
784 | if (unlikely(state == NVME_ANA_CHANGE)) | |
785 | return NVME_SC_ANA_TRANSITION; | |
786 | return 0; | |
787 | } | |
788 | ||
dedf0be5 CK |
789 | static inline u16 nvmet_io_cmd_check_access(struct nvmet_req *req) |
790 | { | |
791 | if (unlikely(req->ns->readonly)) { | |
792 | switch (req->cmd->common.opcode) { | |
793 | case nvme_cmd_read: | |
794 | case nvme_cmd_flush: | |
795 | break; | |
796 | default: | |
797 | return NVME_SC_NS_WRITE_PROTECTED; | |
798 | } | |
799 | } | |
800 | ||
801 | return 0; | |
802 | } | |
803 | ||
d5eff33e CK |
804 | static u16 nvmet_parse_io_cmd(struct nvmet_req *req) |
805 | { | |
806 | struct nvme_command *cmd = req->cmd; | |
807 | u16 ret; | |
808 | ||
809 | ret = nvmet_check_ctrl_status(req, cmd); | |
810 | if (unlikely(ret)) | |
811 | return ret; | |
812 | ||
813 | req->ns = nvmet_find_namespace(req->sq->ctrl, cmd->rw.nsid); | |
e81446af CK |
814 | if (unlikely(!req->ns)) { |
815 | req->error_loc = offsetof(struct nvme_common_command, nsid); | |
d5eff33e | 816 | return NVME_SC_INVALID_NS | NVME_SC_DNR; |
e81446af | 817 | } |
72efd25d | 818 | ret = nvmet_check_ana_state(req->port, req->ns); |
e81446af CK |
819 | if (unlikely(ret)) { |
820 | req->error_loc = offsetof(struct nvme_common_command, nsid); | |
dedf0be5 | 821 | return ret; |
e81446af | 822 | } |
dedf0be5 | 823 | ret = nvmet_io_cmd_check_access(req); |
e81446af CK |
824 | if (unlikely(ret)) { |
825 | req->error_loc = offsetof(struct nvme_common_command, nsid); | |
72efd25d | 826 | return ret; |
e81446af | 827 | } |
d5eff33e CK |
828 | |
829 | if (req->ns->file) | |
830 | return nvmet_file_parse_io_cmd(req); | |
831 | else | |
832 | return nvmet_bdev_parse_io_cmd(req); | |
833 | } | |
834 | ||
a07b4970 | 835 | bool nvmet_req_init(struct nvmet_req *req, struct nvmet_cq *cq, |
e929f06d | 836 | struct nvmet_sq *sq, const struct nvmet_fabrics_ops *ops) |
a07b4970 CH |
837 | { |
838 | u8 flags = req->cmd->common.flags; | |
839 | u16 status; | |
840 | ||
841 | req->cq = cq; | |
842 | req->sq = sq; | |
843 | req->ops = ops; | |
844 | req->sg = NULL; | |
845 | req->sg_cnt = 0; | |
5e62d5c9 | 846 | req->transfer_len = 0; |
fc6c9730 MG |
847 | req->cqe->status = 0; |
848 | req->cqe->sq_head = 0; | |
423b4487 | 849 | req->ns = NULL; |
5698b805 | 850 | req->error_loc = NVMET_NO_ERROR_LOC; |
e4a97625 | 851 | req->error_slba = 0; |
a07b4970 CH |
852 | |
853 | /* no support for fused commands yet */ | |
854 | if (unlikely(flags & (NVME_CMD_FUSE_FIRST | NVME_CMD_FUSE_SECOND))) { | |
e81446af | 855 | req->error_loc = offsetof(struct nvme_common_command, flags); |
a07b4970 CH |
856 | status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; |
857 | goto fail; | |
858 | } | |
859 | ||
bffd2b61 MG |
860 | /* |
861 | * For fabrics, PSDT field shall describe metadata pointer (MPTR) that | |
862 | * contains an address of a single contiguous physical buffer that is | |
863 | * byte aligned. | |
864 | */ | |
865 | if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) { | |
e81446af | 866 | req->error_loc = offsetof(struct nvme_common_command, flags); |
a07b4970 CH |
867 | status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; |
868 | goto fail; | |
869 | } | |
870 | ||
871 | if (unlikely(!req->sq->ctrl)) | |
872 | /* will return an error for any Non-connect command: */ | |
873 | status = nvmet_parse_connect_cmd(req); | |
874 | else if (likely(req->sq->qid != 0)) | |
875 | status = nvmet_parse_io_cmd(req); | |
876 | else if (req->cmd->common.opcode == nvme_fabrics_command) | |
877 | status = nvmet_parse_fabrics_cmd(req); | |
878 | else if (req->sq->ctrl->subsys->type == NVME_NQN_DISC) | |
879 | status = nvmet_parse_discovery_cmd(req); | |
880 | else | |
881 | status = nvmet_parse_admin_cmd(req); | |
882 | ||
883 | if (status) | |
884 | goto fail; | |
885 | ||
886 | if (unlikely(!percpu_ref_tryget_live(&sq->ref))) { | |
887 | status = NVME_SC_INVALID_FIELD | NVME_SC_DNR; | |
888 | goto fail; | |
889 | } | |
890 | ||
c09305ae SG |
891 | if (sq->ctrl) |
892 | sq->ctrl->cmd_seen = true; | |
893 | ||
a07b4970 CH |
894 | return true; |
895 | ||
896 | fail: | |
897 | __nvmet_req_complete(req, status); | |
898 | return false; | |
899 | } | |
900 | EXPORT_SYMBOL_GPL(nvmet_req_init); | |
901 | ||
549f01ae VI |
902 | void nvmet_req_uninit(struct nvmet_req *req) |
903 | { | |
904 | percpu_ref_put(&req->sq->ref); | |
423b4487 SG |
905 | if (req->ns) |
906 | nvmet_put_namespace(req->ns); | |
549f01ae VI |
907 | } |
908 | EXPORT_SYMBOL_GPL(nvmet_req_uninit); | |
909 | ||
5e62d5c9 CH |
910 | void nvmet_req_execute(struct nvmet_req *req) |
911 | { | |
e81446af CK |
912 | if (unlikely(req->data_len != req->transfer_len)) { |
913 | req->error_loc = offsetof(struct nvme_common_command, dptr); | |
5e62d5c9 | 914 | nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA | NVME_SC_DNR); |
e81446af | 915 | } else |
5e62d5c9 CH |
916 | req->execute(req); |
917 | } | |
918 | EXPORT_SYMBOL_GPL(nvmet_req_execute); | |
919 | ||
5b2322e4 LG |
920 | int nvmet_req_alloc_sgl(struct nvmet_req *req) |
921 | { | |
c6925093 LG |
922 | struct pci_dev *p2p_dev = NULL; |
923 | ||
924 | if (IS_ENABLED(CONFIG_PCI_P2PDMA)) { | |
925 | if (req->sq->ctrl && req->ns) | |
926 | p2p_dev = radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, | |
927 | req->ns->nsid); | |
928 | ||
929 | req->p2p_dev = NULL; | |
930 | if (req->sq->qid && p2p_dev) { | |
931 | req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt, | |
932 | req->transfer_len); | |
933 | if (req->sg) { | |
934 | req->p2p_dev = p2p_dev; | |
935 | return 0; | |
936 | } | |
937 | } | |
938 | ||
939 | /* | |
940 | * If no P2P memory was available we fallback to using | |
941 | * regular memory | |
942 | */ | |
943 | } | |
944 | ||
5b2322e4 LG |
945 | req->sg = sgl_alloc(req->transfer_len, GFP_KERNEL, &req->sg_cnt); |
946 | if (!req->sg) | |
947 | return -ENOMEM; | |
948 | ||
949 | return 0; | |
950 | } | |
951 | EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgl); | |
952 | ||
953 | void nvmet_req_free_sgl(struct nvmet_req *req) | |
954 | { | |
c6925093 LG |
955 | if (req->p2p_dev) |
956 | pci_p2pmem_free_sgl(req->p2p_dev, req->sg); | |
957 | else | |
958 | sgl_free(req->sg); | |
959 | ||
5b2322e4 LG |
960 | req->sg = NULL; |
961 | req->sg_cnt = 0; | |
962 | } | |
963 | EXPORT_SYMBOL_GPL(nvmet_req_free_sgl); | |
964 | ||
a07b4970 CH |
965 | static inline bool nvmet_cc_en(u32 cc) |
966 | { | |
ad4e05b2 | 967 | return (cc >> NVME_CC_EN_SHIFT) & 0x1; |
a07b4970 CH |
968 | } |
969 | ||
970 | static inline u8 nvmet_cc_css(u32 cc) | |
971 | { | |
ad4e05b2 | 972 | return (cc >> NVME_CC_CSS_SHIFT) & 0x7; |
a07b4970 CH |
973 | } |
974 | ||
975 | static inline u8 nvmet_cc_mps(u32 cc) | |
976 | { | |
ad4e05b2 | 977 | return (cc >> NVME_CC_MPS_SHIFT) & 0xf; |
a07b4970 CH |
978 | } |
979 | ||
980 | static inline u8 nvmet_cc_ams(u32 cc) | |
981 | { | |
ad4e05b2 | 982 | return (cc >> NVME_CC_AMS_SHIFT) & 0x7; |
a07b4970 CH |
983 | } |
984 | ||
985 | static inline u8 nvmet_cc_shn(u32 cc) | |
986 | { | |
ad4e05b2 | 987 | return (cc >> NVME_CC_SHN_SHIFT) & 0x3; |
a07b4970 CH |
988 | } |
989 | ||
990 | static inline u8 nvmet_cc_iosqes(u32 cc) | |
991 | { | |
ad4e05b2 | 992 | return (cc >> NVME_CC_IOSQES_SHIFT) & 0xf; |
a07b4970 CH |
993 | } |
994 | ||
995 | static inline u8 nvmet_cc_iocqes(u32 cc) | |
996 | { | |
ad4e05b2 | 997 | return (cc >> NVME_CC_IOCQES_SHIFT) & 0xf; |
a07b4970 CH |
998 | } |
999 | ||
1000 | static void nvmet_start_ctrl(struct nvmet_ctrl *ctrl) | |
1001 | { | |
1002 | lockdep_assert_held(&ctrl->lock); | |
1003 | ||
1004 | if (nvmet_cc_iosqes(ctrl->cc) != NVME_NVM_IOSQES || | |
1005 | nvmet_cc_iocqes(ctrl->cc) != NVME_NVM_IOCQES || | |
1006 | nvmet_cc_mps(ctrl->cc) != 0 || | |
1007 | nvmet_cc_ams(ctrl->cc) != 0 || | |
1008 | nvmet_cc_css(ctrl->cc) != 0) { | |
1009 | ctrl->csts = NVME_CSTS_CFS; | |
1010 | return; | |
1011 | } | |
1012 | ||
1013 | ctrl->csts = NVME_CSTS_RDY; | |
d68a90e1 MG |
1014 | |
1015 | /* | |
1016 | * Controllers that are not yet enabled should not really enforce the | |
1017 | * keep alive timeout, but we still want to track a timeout and cleanup | |
1018 | * in case a host died before it enabled the controller. Hence, simply | |
1019 | * reset the keep alive timer when the controller is enabled. | |
1020 | */ | |
1021 | mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ); | |
a07b4970 CH |
1022 | } |
1023 | ||
1024 | static void nvmet_clear_ctrl(struct nvmet_ctrl *ctrl) | |
1025 | { | |
1026 | lockdep_assert_held(&ctrl->lock); | |
1027 | ||
1028 | /* XXX: tear down queues? */ | |
1029 | ctrl->csts &= ~NVME_CSTS_RDY; | |
1030 | ctrl->cc = 0; | |
1031 | } | |
1032 | ||
1033 | void nvmet_update_cc(struct nvmet_ctrl *ctrl, u32 new) | |
1034 | { | |
1035 | u32 old; | |
1036 | ||
1037 | mutex_lock(&ctrl->lock); | |
1038 | old = ctrl->cc; | |
1039 | ctrl->cc = new; | |
1040 | ||
1041 | if (nvmet_cc_en(new) && !nvmet_cc_en(old)) | |
1042 | nvmet_start_ctrl(ctrl); | |
1043 | if (!nvmet_cc_en(new) && nvmet_cc_en(old)) | |
1044 | nvmet_clear_ctrl(ctrl); | |
1045 | if (nvmet_cc_shn(new) && !nvmet_cc_shn(old)) { | |
1046 | nvmet_clear_ctrl(ctrl); | |
1047 | ctrl->csts |= NVME_CSTS_SHST_CMPLT; | |
1048 | } | |
1049 | if (!nvmet_cc_shn(new) && nvmet_cc_shn(old)) | |
1050 | ctrl->csts &= ~NVME_CSTS_SHST_CMPLT; | |
1051 | mutex_unlock(&ctrl->lock); | |
1052 | } | |
1053 | ||
1054 | static void nvmet_init_cap(struct nvmet_ctrl *ctrl) | |
1055 | { | |
1056 | /* command sets supported: NVMe command set: */ | |
1057 | ctrl->cap = (1ULL << 37); | |
1058 | /* CC.EN timeout in 500msec units: */ | |
1059 | ctrl->cap |= (15ULL << 24); | |
1060 | /* maximum queue entries supported: */ | |
1061 | ctrl->cap |= NVMET_QUEUE_SIZE - 1; | |
1062 | } | |
1063 | ||
1064 | u16 nvmet_ctrl_find_get(const char *subsysnqn, const char *hostnqn, u16 cntlid, | |
1065 | struct nvmet_req *req, struct nvmet_ctrl **ret) | |
1066 | { | |
1067 | struct nvmet_subsys *subsys; | |
1068 | struct nvmet_ctrl *ctrl; | |
1069 | u16 status = 0; | |
1070 | ||
1071 | subsys = nvmet_find_get_subsys(req->port, subsysnqn); | |
1072 | if (!subsys) { | |
1073 | pr_warn("connect request for invalid subsystem %s!\n", | |
1074 | subsysnqn); | |
fc6c9730 | 1075 | req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn); |
a07b4970 CH |
1076 | return NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; |
1077 | } | |
1078 | ||
1079 | mutex_lock(&subsys->lock); | |
1080 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) { | |
1081 | if (ctrl->cntlid == cntlid) { | |
1082 | if (strncmp(hostnqn, ctrl->hostnqn, NVMF_NQN_SIZE)) { | |
1083 | pr_warn("hostnqn mismatch.\n"); | |
1084 | continue; | |
1085 | } | |
1086 | if (!kref_get_unless_zero(&ctrl->ref)) | |
1087 | continue; | |
1088 | ||
1089 | *ret = ctrl; | |
1090 | goto out; | |
1091 | } | |
1092 | } | |
1093 | ||
1094 | pr_warn("could not find controller %d for subsys %s / host %s\n", | |
1095 | cntlid, subsysnqn, hostnqn); | |
fc6c9730 | 1096 | req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid); |
a07b4970 CH |
1097 | status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; |
1098 | ||
1099 | out: | |
1100 | mutex_unlock(&subsys->lock); | |
1101 | nvmet_subsys_put(subsys); | |
1102 | return status; | |
1103 | } | |
1104 | ||
64a0ca88 PP |
1105 | u16 nvmet_check_ctrl_status(struct nvmet_req *req, struct nvme_command *cmd) |
1106 | { | |
1107 | if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) { | |
b40b83e3 | 1108 | pr_err("got cmd %d while CC.EN == 0 on qid = %d\n", |
64a0ca88 PP |
1109 | cmd->common.opcode, req->sq->qid); |
1110 | return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; | |
1111 | } | |
1112 | ||
1113 | if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) { | |
b40b83e3 | 1114 | pr_err("got cmd %d while CSTS.RDY == 0 on qid = %d\n", |
64a0ca88 | 1115 | cmd->common.opcode, req->sq->qid); |
64a0ca88 PP |
1116 | return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR; |
1117 | } | |
1118 | return 0; | |
1119 | } | |
1120 | ||
253928ee | 1121 | bool nvmet_host_allowed(struct nvmet_subsys *subsys, const char *hostnqn) |
a07b4970 CH |
1122 | { |
1123 | struct nvmet_host_link *p; | |
1124 | ||
253928ee SG |
1125 | lockdep_assert_held(&nvmet_config_sem); |
1126 | ||
a07b4970 CH |
1127 | if (subsys->allow_any_host) |
1128 | return true; | |
1129 | ||
253928ee SG |
1130 | if (subsys->type == NVME_NQN_DISC) /* allow all access to disc subsys */ |
1131 | return true; | |
1132 | ||
a07b4970 CH |
1133 | list_for_each_entry(p, &subsys->hosts, entry) { |
1134 | if (!strcmp(nvmet_host_name(p->host), hostnqn)) | |
1135 | return true; | |
1136 | } | |
1137 | ||
1138 | return false; | |
1139 | } | |
1140 | ||
c6925093 LG |
1141 | /* |
1142 | * Note: ctrl->subsys->lock should be held when calling this function | |
1143 | */ | |
1144 | static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl, | |
1145 | struct nvmet_req *req) | |
1146 | { | |
1147 | struct nvmet_ns *ns; | |
1148 | ||
1149 | if (!req->p2p_client) | |
1150 | return; | |
1151 | ||
1152 | ctrl->p2p_client = get_device(req->p2p_client); | |
1153 | ||
1154 | list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) | |
1155 | nvmet_p2pmem_ns_add_p2p(ctrl, ns); | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * Note: ctrl->subsys->lock should be held when calling this function | |
1160 | */ | |
1161 | static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl) | |
1162 | { | |
1163 | struct radix_tree_iter iter; | |
1164 | void __rcu **slot; | |
1165 | ||
1166 | radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0) | |
1167 | pci_dev_put(radix_tree_deref_slot(slot)); | |
1168 | ||
1169 | put_device(ctrl->p2p_client); | |
1170 | } | |
1171 | ||
d11de63f YY |
1172 | static void nvmet_fatal_error_handler(struct work_struct *work) |
1173 | { | |
1174 | struct nvmet_ctrl *ctrl = | |
1175 | container_of(work, struct nvmet_ctrl, fatal_err_work); | |
1176 | ||
1177 | pr_err("ctrl %d fatal error occurred!\n", ctrl->cntlid); | |
1178 | ctrl->ops->delete_ctrl(ctrl); | |
1179 | } | |
1180 | ||
a07b4970 CH |
1181 | u16 nvmet_alloc_ctrl(const char *subsysnqn, const char *hostnqn, |
1182 | struct nvmet_req *req, u32 kato, struct nvmet_ctrl **ctrlp) | |
1183 | { | |
1184 | struct nvmet_subsys *subsys; | |
1185 | struct nvmet_ctrl *ctrl; | |
1186 | int ret; | |
1187 | u16 status; | |
1188 | ||
1189 | status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; | |
1190 | subsys = nvmet_find_get_subsys(req->port, subsysnqn); | |
1191 | if (!subsys) { | |
1192 | pr_warn("connect request for invalid subsystem %s!\n", | |
1193 | subsysnqn); | |
fc6c9730 | 1194 | req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn); |
a07b4970 CH |
1195 | goto out; |
1196 | } | |
1197 | ||
1198 | status = NVME_SC_CONNECT_INVALID_PARAM | NVME_SC_DNR; | |
1199 | down_read(&nvmet_config_sem); | |
253928ee | 1200 | if (!nvmet_host_allowed(subsys, hostnqn)) { |
a07b4970 CH |
1201 | pr_info("connect by host %s for subsystem %s not allowed\n", |
1202 | hostnqn, subsysnqn); | |
fc6c9730 | 1203 | req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn); |
a07b4970 | 1204 | up_read(&nvmet_config_sem); |
130c24b5 | 1205 | status = NVME_SC_CONNECT_INVALID_HOST | NVME_SC_DNR; |
a07b4970 CH |
1206 | goto out_put_subsystem; |
1207 | } | |
1208 | up_read(&nvmet_config_sem); | |
1209 | ||
1210 | status = NVME_SC_INTERNAL; | |
1211 | ctrl = kzalloc(sizeof(*ctrl), GFP_KERNEL); | |
1212 | if (!ctrl) | |
1213 | goto out_put_subsystem; | |
1214 | mutex_init(&ctrl->lock); | |
1215 | ||
1216 | nvmet_init_cap(ctrl); | |
1217 | ||
4ee43280 CH |
1218 | ctrl->port = req->port; |
1219 | ||
a07b4970 CH |
1220 | INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work); |
1221 | INIT_LIST_HEAD(&ctrl->async_events); | |
c6925093 | 1222 | INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL); |
d11de63f | 1223 | INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler); |
a07b4970 CH |
1224 | |
1225 | memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE); | |
1226 | memcpy(ctrl->hostnqn, hostnqn, NVMF_NQN_SIZE); | |
1227 | ||
1228 | kref_init(&ctrl->ref); | |
1229 | ctrl->subsys = subsys; | |
c86b8f7b | 1230 | WRITE_ONCE(ctrl->aen_enabled, NVMET_AEN_CFG_OPTIONAL); |
a07b4970 | 1231 | |
c16734ea CH |
1232 | ctrl->changed_ns_list = kmalloc_array(NVME_MAX_CHANGED_NAMESPACES, |
1233 | sizeof(__le32), GFP_KERNEL); | |
1234 | if (!ctrl->changed_ns_list) | |
1235 | goto out_free_ctrl; | |
1236 | ||
a07b4970 CH |
1237 | ctrl->cqs = kcalloc(subsys->max_qid + 1, |
1238 | sizeof(struct nvmet_cq *), | |
1239 | GFP_KERNEL); | |
1240 | if (!ctrl->cqs) | |
c16734ea | 1241 | goto out_free_changed_ns_list; |
a07b4970 CH |
1242 | |
1243 | ctrl->sqs = kcalloc(subsys->max_qid + 1, | |
1244 | sizeof(struct nvmet_sq *), | |
1245 | GFP_KERNEL); | |
1246 | if (!ctrl->sqs) | |
1247 | goto out_free_cqs; | |
1248 | ||
15fbad96 | 1249 | ret = ida_simple_get(&cntlid_ida, |
a07b4970 CH |
1250 | NVME_CNTLID_MIN, NVME_CNTLID_MAX, |
1251 | GFP_KERNEL); | |
1252 | if (ret < 0) { | |
1253 | status = NVME_SC_CONNECT_CTRL_BUSY | NVME_SC_DNR; | |
1254 | goto out_free_sqs; | |
1255 | } | |
1256 | ctrl->cntlid = ret; | |
1257 | ||
1258 | ctrl->ops = req->ops; | |
a07b4970 | 1259 | |
f9362ac1 JS |
1260 | /* |
1261 | * Discovery controllers may use some arbitrary high value | |
1262 | * in order to cleanup stale discovery sessions | |
1263 | */ | |
1264 | if ((ctrl->subsys->type == NVME_NQN_DISC) && !kato) | |
1265 | kato = NVMET_DISC_KATO_MS; | |
1266 | ||
1267 | /* keep-alive timeout in seconds */ | |
1268 | ctrl->kato = DIV_ROUND_UP(kato, 1000); | |
1269 | ||
e4a97625 CK |
1270 | ctrl->err_counter = 0; |
1271 | spin_lock_init(&ctrl->error_lock); | |
1272 | ||
a07b4970 CH |
1273 | nvmet_start_keep_alive_timer(ctrl); |
1274 | ||
1275 | mutex_lock(&subsys->lock); | |
1276 | list_add_tail(&ctrl->subsys_entry, &subsys->ctrls); | |
c6925093 | 1277 | nvmet_setup_p2p_ns_map(ctrl, req); |
a07b4970 CH |
1278 | mutex_unlock(&subsys->lock); |
1279 | ||
1280 | *ctrlp = ctrl; | |
1281 | return 0; | |
1282 | ||
1283 | out_free_sqs: | |
1284 | kfree(ctrl->sqs); | |
1285 | out_free_cqs: | |
1286 | kfree(ctrl->cqs); | |
c16734ea CH |
1287 | out_free_changed_ns_list: |
1288 | kfree(ctrl->changed_ns_list); | |
a07b4970 CH |
1289 | out_free_ctrl: |
1290 | kfree(ctrl); | |
1291 | out_put_subsystem: | |
1292 | nvmet_subsys_put(subsys); | |
1293 | out: | |
1294 | return status; | |
1295 | } | |
1296 | ||
1297 | static void nvmet_ctrl_free(struct kref *ref) | |
1298 | { | |
1299 | struct nvmet_ctrl *ctrl = container_of(ref, struct nvmet_ctrl, ref); | |
1300 | struct nvmet_subsys *subsys = ctrl->subsys; | |
1301 | ||
a07b4970 | 1302 | mutex_lock(&subsys->lock); |
c6925093 | 1303 | nvmet_release_p2p_ns_map(ctrl); |
a07b4970 CH |
1304 | list_del(&ctrl->subsys_entry); |
1305 | mutex_unlock(&subsys->lock); | |
1306 | ||
6b1943af IR |
1307 | nvmet_stop_keep_alive_timer(ctrl); |
1308 | ||
06406d81 SG |
1309 | flush_work(&ctrl->async_event_work); |
1310 | cancel_work_sync(&ctrl->fatal_err_work); | |
1311 | ||
15fbad96 | 1312 | ida_simple_remove(&cntlid_ida, ctrl->cntlid); |
a07b4970 CH |
1313 | |
1314 | kfree(ctrl->sqs); | |
1315 | kfree(ctrl->cqs); | |
c16734ea | 1316 | kfree(ctrl->changed_ns_list); |
a07b4970 | 1317 | kfree(ctrl); |
6b1943af IR |
1318 | |
1319 | nvmet_subsys_put(subsys); | |
a07b4970 CH |
1320 | } |
1321 | ||
1322 | void nvmet_ctrl_put(struct nvmet_ctrl *ctrl) | |
1323 | { | |
1324 | kref_put(&ctrl->ref, nvmet_ctrl_free); | |
1325 | } | |
1326 | ||
a07b4970 CH |
1327 | void nvmet_ctrl_fatal_error(struct nvmet_ctrl *ctrl) |
1328 | { | |
8242ddac SG |
1329 | mutex_lock(&ctrl->lock); |
1330 | if (!(ctrl->csts & NVME_CSTS_CFS)) { | |
1331 | ctrl->csts |= NVME_CSTS_CFS; | |
8242ddac SG |
1332 | schedule_work(&ctrl->fatal_err_work); |
1333 | } | |
1334 | mutex_unlock(&ctrl->lock); | |
a07b4970 CH |
1335 | } |
1336 | EXPORT_SYMBOL_GPL(nvmet_ctrl_fatal_error); | |
1337 | ||
1338 | static struct nvmet_subsys *nvmet_find_get_subsys(struct nvmet_port *port, | |
1339 | const char *subsysnqn) | |
1340 | { | |
1341 | struct nvmet_subsys_link *p; | |
1342 | ||
1343 | if (!port) | |
1344 | return NULL; | |
1345 | ||
43a6f8fb | 1346 | if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) { |
a07b4970 CH |
1347 | if (!kref_get_unless_zero(&nvmet_disc_subsys->ref)) |
1348 | return NULL; | |
1349 | return nvmet_disc_subsys; | |
1350 | } | |
1351 | ||
1352 | down_read(&nvmet_config_sem); | |
1353 | list_for_each_entry(p, &port->subsystems, entry) { | |
1354 | if (!strncmp(p->subsys->subsysnqn, subsysnqn, | |
1355 | NVMF_NQN_SIZE)) { | |
1356 | if (!kref_get_unless_zero(&p->subsys->ref)) | |
1357 | break; | |
1358 | up_read(&nvmet_config_sem); | |
1359 | return p->subsys; | |
1360 | } | |
1361 | } | |
1362 | up_read(&nvmet_config_sem); | |
1363 | return NULL; | |
1364 | } | |
1365 | ||
1366 | struct nvmet_subsys *nvmet_subsys_alloc(const char *subsysnqn, | |
1367 | enum nvme_subsys_type type) | |
1368 | { | |
1369 | struct nvmet_subsys *subsys; | |
1370 | ||
1371 | subsys = kzalloc(sizeof(*subsys), GFP_KERNEL); | |
1372 | if (!subsys) | |
6b7e631b | 1373 | return ERR_PTR(-ENOMEM); |
a07b4970 | 1374 | |
637dc0f3 | 1375 | subsys->ver = NVME_VS(1, 3, 0); /* NVMe 1.3.0 */ |
2e7f5d2a JT |
1376 | /* generate a random serial number as our controllers are ephemeral: */ |
1377 | get_random_bytes(&subsys->serial, sizeof(subsys->serial)); | |
a07b4970 CH |
1378 | |
1379 | switch (type) { | |
1380 | case NVME_NQN_NVME: | |
1381 | subsys->max_qid = NVMET_NR_QUEUES; | |
1382 | break; | |
1383 | case NVME_NQN_DISC: | |
1384 | subsys->max_qid = 0; | |
1385 | break; | |
1386 | default: | |
1387 | pr_err("%s: Unknown Subsystem type - %d\n", __func__, type); | |
1388 | kfree(subsys); | |
6b7e631b | 1389 | return ERR_PTR(-EINVAL); |
a07b4970 CH |
1390 | } |
1391 | subsys->type = type; | |
1392 | subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE, | |
1393 | GFP_KERNEL); | |
69555af2 | 1394 | if (!subsys->subsysnqn) { |
a07b4970 | 1395 | kfree(subsys); |
6b7e631b | 1396 | return ERR_PTR(-ENOMEM); |
a07b4970 CH |
1397 | } |
1398 | ||
1399 | kref_init(&subsys->ref); | |
1400 | ||
1401 | mutex_init(&subsys->lock); | |
1402 | INIT_LIST_HEAD(&subsys->namespaces); | |
1403 | INIT_LIST_HEAD(&subsys->ctrls); | |
a07b4970 CH |
1404 | INIT_LIST_HEAD(&subsys->hosts); |
1405 | ||
1406 | return subsys; | |
1407 | } | |
1408 | ||
1409 | static void nvmet_subsys_free(struct kref *ref) | |
1410 | { | |
1411 | struct nvmet_subsys *subsys = | |
1412 | container_of(ref, struct nvmet_subsys, ref); | |
1413 | ||
1414 | WARN_ON_ONCE(!list_empty(&subsys->namespaces)); | |
1415 | ||
a07b4970 CH |
1416 | kfree(subsys->subsysnqn); |
1417 | kfree(subsys); | |
1418 | } | |
1419 | ||
344770b0 SG |
1420 | void nvmet_subsys_del_ctrls(struct nvmet_subsys *subsys) |
1421 | { | |
1422 | struct nvmet_ctrl *ctrl; | |
1423 | ||
1424 | mutex_lock(&subsys->lock); | |
1425 | list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) | |
1426 | ctrl->ops->delete_ctrl(ctrl); | |
1427 | mutex_unlock(&subsys->lock); | |
1428 | } | |
1429 | ||
a07b4970 CH |
1430 | void nvmet_subsys_put(struct nvmet_subsys *subsys) |
1431 | { | |
1432 | kref_put(&subsys->ref, nvmet_subsys_free); | |
1433 | } | |
1434 | ||
1435 | static int __init nvmet_init(void) | |
1436 | { | |
1437 | int error; | |
1438 | ||
72efd25d CH |
1439 | nvmet_ana_group_enabled[NVMET_DEFAULT_ANA_GRPID] = 1; |
1440 | ||
55eb942e CK |
1441 | buffered_io_wq = alloc_workqueue("nvmet-buffered-io-wq", |
1442 | WQ_MEM_RECLAIM, 0); | |
1443 | if (!buffered_io_wq) { | |
1444 | error = -ENOMEM; | |
1445 | goto out; | |
1446 | } | |
72efd25d | 1447 | |
a07b4970 CH |
1448 | error = nvmet_init_discovery(); |
1449 | if (error) | |
04db0e5e | 1450 | goto out_free_work_queue; |
a07b4970 CH |
1451 | |
1452 | error = nvmet_init_configfs(); | |
1453 | if (error) | |
1454 | goto out_exit_discovery; | |
1455 | return 0; | |
1456 | ||
1457 | out_exit_discovery: | |
1458 | nvmet_exit_discovery(); | |
04db0e5e CK |
1459 | out_free_work_queue: |
1460 | destroy_workqueue(buffered_io_wq); | |
a07b4970 CH |
1461 | out: |
1462 | return error; | |
1463 | } | |
1464 | ||
1465 | static void __exit nvmet_exit(void) | |
1466 | { | |
1467 | nvmet_exit_configfs(); | |
1468 | nvmet_exit_discovery(); | |
15fbad96 | 1469 | ida_destroy(&cntlid_ida); |
55eb942e | 1470 | destroy_workqueue(buffered_io_wq); |
a07b4970 CH |
1471 | |
1472 | BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_entry) != 1024); | |
1473 | BUILD_BUG_ON(sizeof(struct nvmf_disc_rsp_page_hdr) != 1024); | |
1474 | } | |
1475 | ||
1476 | module_init(nvmet_init); | |
1477 | module_exit(nvmet_exit); | |
1478 | ||
1479 | MODULE_LICENSE("GPL v2"); |