~hc/RK356X_SDK_RELEASE.git

..	..	@@ -1,20 +1,17 @@
	1	+// SPDX-License-Identifier: GPL-2.0
1	2	/*
2	3	* Common code for the NVMe target.
3	4	* 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	5	*/
14	6	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15	7	#include <linux/module.h>
16	8	#include <linux/random.h>
17	9	#include <linux/rculist.h>
	10	+#include <linux/pci-p2pdma.h>
	11	+#include <linux/scatterlist.h>
	12	+
	13	+#define CREATE_TRACE_POINTS
	14	+#include "trace.h"
18	15
19	16	#include "nvmet.h"
20	17
..	..	@@ -44,40 +41,88 @@
44	41	u64 nvmet_ana_chgcnt;
45	42	DECLARE_RWSEM(nvmet_ana_sem);
46	43
	44	+inline u16 errno_to_nvme_status(struct nvmet_req *req, int errno)
	45	+{
	46	+ u16 status;
	47	+
	48	+ switch (errno) {
	49	+ case 0:
	50	+ status = NVME_SC_SUCCESS;
	51	+ break;
	52	+ case -ENOSPC:
	53	+ req->error_loc = offsetof(struct nvme_rw_command, length);
	54	+ status = NVME_SC_CAP_EXCEEDED \| NVME_SC_DNR;
	55	+ break;
	56	+ case -EREMOTEIO:
	57	+ req->error_loc = offsetof(struct nvme_rw_command, slba);
	58	+ status = NVME_SC_LBA_RANGE \| NVME_SC_DNR;
	59	+ break;
	60	+ case -EOPNOTSUPP:
	61	+ req->error_loc = offsetof(struct nvme_common_command, opcode);
	62	+ switch (req->cmd->common.opcode) {
	63	+ case nvme_cmd_dsm:
	64	+ case nvme_cmd_write_zeroes:
	65	+ status = NVME_SC_ONCS_NOT_SUPPORTED \| NVME_SC_DNR;
	66	+ break;
	67	+ default:
	68	+ status = NVME_SC_INVALID_OPCODE \| NVME_SC_DNR;
	69	+ }
	70	+ break;
	71	+ case -ENODATA:
	72	+ req->error_loc = offsetof(struct nvme_rw_command, nsid);
	73	+ status = NVME_SC_ACCESS_DENIED;
	74	+ break;
	75	+ case -EIO:
	76	+ fallthrough;
	77	+ default:
	78	+ req->error_loc = offsetof(struct nvme_common_command, opcode);
	79	+ status = NVME_SC_INTERNAL \| NVME_SC_DNR;
	80	+ }
	81	+
	82	+ return status;
	83	+}
	84	+
47	85	static struct nvmet_subsys nvmet_find_get_subsys(struct nvmet_port port,
48	86	const char *subsysnqn);
49	87
50	88	u16 nvmet_copy_to_sgl(struct nvmet_req req, off_t off, const void buf,
51	89	size_t len)
52	90	{
53		- if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
	91	+ if (sg_pcopy_from_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
	92	+ req->error_loc = offsetof(struct nvme_common_command, dptr);
54	93	return NVME_SC_SGL_INVALID_DATA \| NVME_SC_DNR;
	94	+ }
55	95	return 0;
56	96	}
57	97
58	98	u16 nvmet_copy_from_sgl(struct nvmet_req req, off_t off, void buf, size_t len)
59	99	{
60		- if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len)
	100	+ if (sg_pcopy_to_buffer(req->sg, req->sg_cnt, buf, len, off) != len) {
	101	+ req->error_loc = offsetof(struct nvme_common_command, dptr);
61	102	return NVME_SC_SGL_INVALID_DATA \| NVME_SC_DNR;
	103	+ }
62	104	return 0;
63	105	}
64	106
65	107	u16 nvmet_zero_sgl(struct nvmet_req *req, off_t off, size_t len)
66	108	{
67		- if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len)
	109	+ if (sg_zero_buffer(req->sg, req->sg_cnt, len, off) != len) {
	110	+ req->error_loc = offsetof(struct nvme_common_command, dptr);
68	111	return NVME_SC_SGL_INVALID_DATA \| NVME_SC_DNR;
	112	+ }
69	113	return 0;
70	114	}
71	115
72	116	static unsigned int nvmet_max_nsid(struct nvmet_subsys *subsys)
73	117	{
74		- struct nvmet_ns *ns;
	118	+ unsigned long nsid = 0;
	119	+ struct nvmet_ns *cur;
	120	+ unsigned long idx;
75	121
76		- if (list_empty(&subsys->namespaces))
77		- return 0;
	122	+ xa_for_each(&subsys->namespaces, idx, cur)
	123	+ nsid = cur->nsid;
78	124
79		- ns = list_last_entry(&subsys->namespaces, struct nvmet_ns, dev_link);
80		- return ns->nsid;
	125	+ return nsid;
81	126	}
82	127
83	128	static u32 nvmet_async_event_result(struct nvmet_async_event *aen)
..	..	@@ -85,39 +130,30 @@
85	130	return aen->event_type \| (aen->event_info << 8) \| (aen->log_page << 16);
86	131	}
87	132
88		-static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
	133	+static void nvmet_async_events_failall(struct nvmet_ctrl *ctrl)
89	134	{
	135	+ u16 status = NVME_SC_INTERNAL \| NVME_SC_DNR;
90	136	struct nvmet_req *req;
91	137
92		- while (1) {
93		- mutex_lock(&ctrl->lock);
94		- if (!ctrl->nr_async_event_cmds) {
95		- mutex_unlock(&ctrl->lock);
96		- return;
97		- }
98		-
	138	+ mutex_lock(&ctrl->lock);
	139	+ while (ctrl->nr_async_event_cmds) {
99	140	req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
100	141	mutex_unlock(&ctrl->lock);
101		- nvmet_req_complete(req, NVME_SC_INTERNAL \| NVME_SC_DNR);
	142	+ nvmet_req_complete(req, status);
	143	+ mutex_lock(&ctrl->lock);
102	144	}
	145	+ mutex_unlock(&ctrl->lock);
103	146	}
104	147
105		-static void nvmet_async_event_work(struct work_struct *work)
	148	+static void nvmet_async_events_process(struct nvmet_ctrl *ctrl)
106	149	{
107		- struct nvmet_ctrl *ctrl =
108		- container_of(work, struct nvmet_ctrl, async_event_work);
109	150	struct nvmet_async_event *aen;
110	151	struct nvmet_req *req;
111	152
112		- while (1) {
113		- mutex_lock(&ctrl->lock);
114		- aen = list_first_entry_or_null(&ctrl->async_events,
115		- struct nvmet_async_event, entry);
116		- if (!aen \|\| !ctrl->nr_async_event_cmds) {
117		- mutex_unlock(&ctrl->lock);
118		- return;
119		- }
120		-
	153	+ mutex_lock(&ctrl->lock);
	154	+ while (ctrl->nr_async_event_cmds && !list_empty(&ctrl->async_events)) {
	155	+ aen = list_first_entry(&ctrl->async_events,
	156	+ struct nvmet_async_event, entry);
121	157	req = ctrl->async_event_cmds[--ctrl->nr_async_event_cmds];
122	158	nvmet_set_result(req, nvmet_async_event_result(aen));
123	159
..	..	@@ -125,11 +161,34 @@
125	161	kfree(aen);
126	162
127	163	mutex_unlock(&ctrl->lock);
	164	+ trace_nvmet_async_event(ctrl, req->cqe->result.u32);
128	165	nvmet_req_complete(req, 0);
	166	+ mutex_lock(&ctrl->lock);
129	167	}
	168	+ mutex_unlock(&ctrl->lock);
130	169	}
131	170
132		-static void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
	171	+static void nvmet_async_events_free(struct nvmet_ctrl *ctrl)
	172	+{
	173	+ struct nvmet_async_event aen, tmp;
	174	+
	175	+ mutex_lock(&ctrl->lock);
	176	+ list_for_each_entry_safe(aen, tmp, &ctrl->async_events, entry) {
	177	+ list_del(&aen->entry);
	178	+ kfree(aen);
	179	+ }
	180	+ mutex_unlock(&ctrl->lock);
	181	+}
	182	+
	183	+static void nvmet_async_event_work(struct work_struct *work)
	184	+{
	185	+ struct nvmet_ctrl *ctrl =
	186	+ container_of(work, struct nvmet_ctrl, async_event_work);
	187	+
	188	+ nvmet_async_events_process(ctrl);
	189	+}
	190	+
	191	+void nvmet_add_async_event(struct nvmet_ctrl *ctrl, u8 event_type,
133	192	u8 event_info, u8 log_page)
134	193	{
135	194	struct nvmet_async_event *aen;
..	..	@@ -147,13 +206,6 @@
147	206	mutex_unlock(&ctrl->lock);
148	207
149	208	schedule_work(&ctrl->async_event_work);
150		-}
151		-
152		-static bool nvmet_aen_disabled(struct nvmet_ctrl *ctrl, u32 aen)
153		-{
154		- if (!(READ_ONCE(ctrl->aen_enabled) & aen))
155		- return true;
156		- return test_and_set_bit(aen, &ctrl->aen_masked);
157	209	}
158	210
159	211	static void nvmet_add_to_changed_ns_log(struct nvmet_ctrl *ctrl, __le32 nsid)
..	..	@@ -184,9 +236,11 @@
184	236	{
185	237	struct nvmet_ctrl *ctrl;
186	238
	239	+ lockdep_assert_held(&subsys->lock);
	240	+
187	241	list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
188	242	nvmet_add_to_changed_ns_log(ctrl, cpu_to_le32(nsid));
189		- if (nvmet_aen_disabled(ctrl, NVME_AEN_CFG_NS_ATTR))
	243	+ if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_NS_ATTR))
190	244	continue;
191	245	nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
192	246	NVME_AER_NOTICE_NS_CHANGED,
..	..	@@ -203,7 +257,7 @@
203	257	list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
204	258	if (port && ctrl->port != port)
205	259	continue;
206		- if (nvmet_aen_disabled(ctrl, NVME_AEN_CFG_ANA_CHANGE))
	260	+ if (nvmet_aen_bit_disabled(ctrl, NVME_AEN_BIT_ANA_CHANGE))
207	261	continue;
208	262	nvmet_add_async_event(ctrl, NVME_AER_TYPE_NOTICE,
209	263	NVME_AER_NOTICE_ANA, NVME_LOG_ANA);
..	..	@@ -244,6 +298,18 @@
244	298	}
245	299	EXPORT_SYMBOL_GPL(nvmet_unregister_transport);
246	300
	301	+void nvmet_port_del_ctrls(struct nvmet_port port, struct nvmet_subsys subsys)
	302	+{
	303	+ struct nvmet_ctrl *ctrl;
	304	+
	305	+ mutex_lock(&subsys->lock);
	306	+ list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
	307	+ if (ctrl->port == port)
	308	+ ctrl->ops->delete_ctrl(ctrl);
	309	+ }
	310	+ mutex_unlock(&subsys->lock);
	311	+}
	312	+
247	313	int nvmet_enable_port(struct nvmet_port *port)
248	314	{
249	315	const struct nvmet_fabrics_ops *ops;
..	..	@@ -267,18 +333,32 @@
267	333	if (!try_module_get(ops->owner))
268	334	return -EINVAL;
269	335
270		- ret = ops->add_port(port);
271		- if (ret) {
272		- module_put(ops->owner);
273		- return ret;
	336	+ /*
	337	+ * If the user requested PI support and the transport isn't pi capable,
	338	+ * don't enable the port.
	339	+ */
	340	+ if (port->pi_enable && !(ops->flags & NVMF_METADATA_SUPPORTED)) {
	341	+ pr_err("T10-PI is not supported by transport type %d\n",
	342	+ port->disc_addr.trtype);
	343	+ ret = -EINVAL;
	344	+ goto out_put;
274	345	}
	346	+
	347	+ ret = ops->add_port(port);
	348	+ if (ret)
	349	+ goto out_put;
275	350
276	351	/* If the transport didn't set inline_data_size, then disable it. */
277	352	if (port->inline_data_size < 0)
278	353	port->inline_data_size = 0;
279	354
280	355	port->enabled = true;
	356	+ port->tr_ops = ops;
281	357	return 0;
	358	+
	359	+out_put:
	360	+ module_put(ops->owner);
	361	+ return ret;
282	362	}
283	363
284	364	void nvmet_disable_port(struct nvmet_port *port)
..	..	@@ -288,6 +368,7 @@
288	368	lockdep_assert_held(&nvmet_config_sem);
289	369
290	370	port->enabled = false;
	371	+ port->tr_ops = NULL;
291	372
292	373	ops = nvmet_transports[port->disc_addr.trtype];
293	374	ops->remove_port(port);
..	..	@@ -298,6 +379,15 @@
298	379	{
299	380	struct nvmet_ctrl *ctrl = container_of(to_delayed_work(work),
300	381	struct nvmet_ctrl, ka_work);
	382	+ bool reset_tbkas = ctrl->reset_tbkas;
	383	+
	384	+ ctrl->reset_tbkas = false;
	385	+ if (reset_tbkas) {
	386	+ pr_debug("ctrl %d reschedule traffic based keep-alive timer\n",
	387	+ ctrl->cntlid);
	388	+ schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
	389	+ return;
	390	+ }
301	391
302	392	pr_err("ctrl %d keep-alive timer (%d seconds) expired!\n",
303	393	ctrl->cntlid, ctrl->kato);
..	..	@@ -305,7 +395,7 @@
305	395	nvmet_ctrl_fatal_error(ctrl);
306	396	}
307	397
308		-static void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
	398	+void nvmet_start_keep_alive_timer(struct nvmet_ctrl *ctrl)
309	399	{
310	400	if (unlikely(ctrl->kato == 0))
311	401	return;
..	..	@@ -317,7 +407,7 @@
317	407	schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
318	408	}
319	409
320		-static void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
	410	+void nvmet_stop_keep_alive_timer(struct nvmet_ctrl *ctrl)
321	411	{
322	412	if (unlikely(ctrl->kato == 0))
323	413	return;
..	..	@@ -327,28 +417,13 @@
327	417	cancel_delayed_work_sync(&ctrl->ka_work);
328	418	}
329	419
330		-static struct nvmet_ns __nvmet_find_namespace(struct nvmet_ctrl ctrl,
331		- __le32 nsid)
332		-{
333		- struct nvmet_ns *ns;
334		-
335		- list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
336		- if (ns->nsid == le32_to_cpu(nsid))
337		- return ns;
338		- }
339		-
340		- return NULL;
341		-}
342		-
343	420	struct nvmet_ns nvmet_find_namespace(struct nvmet_ctrl ctrl, __le32 nsid)
344	421	{
345	422	struct nvmet_ns *ns;
346	423
347		- rcu_read_lock();
348		- ns = __nvmet_find_namespace(ctrl, nsid);
	424	+ ns = xa_load(&ctrl->subsys->namespaces, le32_to_cpu(nsid));
349	425	if (ns)
350	426	percpu_ref_get(&ns->ref);
351		- rcu_read_unlock();
352	427
353	428	return ns;
354	429	}
..	..	@@ -371,17 +446,121 @@
371	446	nvmet_file_ns_disable(ns);
372	447	}
373	448
	449	+static int nvmet_p2pmem_ns_enable(struct nvmet_ns *ns)
	450	+{
	451	+ int ret;
	452	+ struct pci_dev *p2p_dev;
	453	+
	454	+ if (!ns->use_p2pmem)
	455	+ return 0;
	456	+
	457	+ if (!ns->bdev) {
	458	+ pr_err("peer-to-peer DMA is not supported by non-block device namespaces\n");
	459	+ return -EINVAL;
	460	+ }
	461	+
	462	+ if (!blk_queue_pci_p2pdma(ns->bdev->bd_disk->queue)) {
	463	+ pr_err("peer-to-peer DMA is not supported by the driver of %s\n",
	464	+ ns->device_path);
	465	+ return -EINVAL;
	466	+ }
	467	+
	468	+ if (ns->p2p_dev) {
	469	+ ret = pci_p2pdma_distance(ns->p2p_dev, nvmet_ns_dev(ns), true);
	470	+ if (ret < 0)
	471	+ return -EINVAL;
	472	+ } else {
	473	+ /*
	474	+ * Right now we just check that there is p2pmem available so
	475	+ * we can report an error to the user right away if there
	476	+ * is not. We'll find the actual device to use once we
	477	+ * setup the controller when the port's device is available.
	478	+ */
	479	+
	480	+ p2p_dev = pci_p2pmem_find(nvmet_ns_dev(ns));
	481	+ if (!p2p_dev) {
	482	+ pr_err("no peer-to-peer memory is available for %s\n",
	483	+ ns->device_path);
	484	+ return -EINVAL;
	485	+ }
	486	+
	487	+ pci_dev_put(p2p_dev);
	488	+ }
	489	+
	490	+ return 0;
	491	+}
	492	+
	493	+/*
	494	+ * Note: ctrl->subsys->lock should be held when calling this function
	495	+ */
	496	+static void nvmet_p2pmem_ns_add_p2p(struct nvmet_ctrl *ctrl,
	497	+ struct nvmet_ns *ns)
	498	+{
	499	+ struct device *clients[2];
	500	+ struct pci_dev *p2p_dev;
	501	+ int ret;
	502	+
	503	+ if (!ctrl->p2p_client \|\| !ns->use_p2pmem)
	504	+ return;
	505	+
	506	+ if (ns->p2p_dev) {
	507	+ ret = pci_p2pdma_distance(ns->p2p_dev, ctrl->p2p_client, true);
	508	+ if (ret < 0)
	509	+ return;
	510	+
	511	+ p2p_dev = pci_dev_get(ns->p2p_dev);
	512	+ } else {
	513	+ clients[0] = ctrl->p2p_client;
	514	+ clients[1] = nvmet_ns_dev(ns);
	515	+
	516	+ p2p_dev = pci_p2pmem_find_many(clients, ARRAY_SIZE(clients));
	517	+ if (!p2p_dev) {
	518	+ pr_err("no peer-to-peer memory is available that's supported by %s and %s\n",
	519	+ dev_name(ctrl->p2p_client), ns->device_path);
	520	+ return;
	521	+ }
	522	+ }
	523	+
	524	+ ret = radix_tree_insert(&ctrl->p2p_ns_map, ns->nsid, p2p_dev);
	525	+ if (ret < 0)
	526	+ pci_dev_put(p2p_dev);
	527	+
	528	+ pr_info("using p2pmem on %s for nsid %d\n", pci_name(p2p_dev),
	529	+ ns->nsid);
	530	+}
	531	+
	532	+void nvmet_ns_revalidate(struct nvmet_ns *ns)
	533	+{
	534	+ loff_t oldsize = ns->size;
	535	+
	536	+ if (ns->bdev)
	537	+ nvmet_bdev_ns_revalidate(ns);
	538	+ else
	539	+ nvmet_file_ns_revalidate(ns);
	540	+
	541	+ if (oldsize != ns->size)
	542	+ nvmet_ns_changed(ns->subsys, ns->nsid);
	543	+}
	544	+
374	545	int nvmet_ns_enable(struct nvmet_ns *ns)
375	546	{
376	547	struct nvmet_subsys *subsys = ns->subsys;
	548	+ struct nvmet_ctrl *ctrl;
377	549	int ret;
378	550
379	551	mutex_lock(&subsys->lock);
	552	+ ret = 0;
	553	+
	554	+ if (nvmet_passthru_ctrl(subsys)) {
	555	+ pr_info("cannot enable both passthru and regular namespaces for a single subsystem");
	556	+ goto out_unlock;
	557	+ }
	558	+
	559	+ if (ns->enabled)
	560	+ goto out_unlock;
	561	+
380	562	ret = -EMFILE;
381	563	if (subsys->nr_namespaces == NVMET_MAX_NAMESPACES)
382		- goto out_unlock;
383		- ret = 0;
384		- if (ns->enabled)
385	564	goto out_unlock;
386	565
387	566	ret = nvmet_bdev_ns_enable(ns);
..	..	@@ -389,6 +568,13 @@
389	568	ret = nvmet_file_ns_enable(ns);
390	569	if (ret)
391	570	goto out_unlock;
	571	+
	572	+ ret = nvmet_p2pmem_ns_enable(ns);
	573	+ if (ret)
	574	+ goto out_dev_disable;
	575	+
	576	+ list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
	577	+ nvmet_p2pmem_ns_add_p2p(ctrl, ns);
392	578
393	579	ret = percpu_ref_init(&ns->ref, nvmet_destroy_namespace,
394	580	0, GFP_KERNEL);
..	..	@@ -398,23 +584,10 @@
398	584	if (ns->nsid > subsys->max_nsid)
399	585	subsys->max_nsid = ns->nsid;
400	586
401		- /*
402		- * The namespaces list needs to be sorted to simplify the implementation
403		- * of the Identify Namepace List subcommand.
404		- */
405		- if (list_empty(&subsys->namespaces)) {
406		- list_add_tail_rcu(&ns->dev_link, &subsys->namespaces);
407		- } else {
408		- struct nvmet_ns *old;
	587	+ ret = xa_insert(&subsys->namespaces, ns->nsid, ns, GFP_KERNEL);
	588	+ if (ret)
	589	+ goto out_restore_subsys_maxnsid;
409	590
410		- list_for_each_entry_rcu(old, &subsys->namespaces, dev_link) {
411		- BUG_ON(ns->nsid == old->nsid);
412		- if (ns->nsid < old->nsid)
413		- break;
414		- }
415		-
416		- list_add_tail_rcu(&ns->dev_link, &old->dev_link);
417		- }
418	591	subsys->nr_namespaces++;
419	592
420	593	nvmet_ns_changed(subsys, ns->nsid);
..	..	@@ -423,7 +596,14 @@
423	596	out_unlock:
424	597	mutex_unlock(&subsys->lock);
425	598	return ret;
	599	+
	600	+out_restore_subsys_maxnsid:
	601	+ subsys->max_nsid = nvmet_max_nsid(subsys);
	602	+ percpu_ref_exit(&ns->ref);
426	603	out_dev_put:
	604	+ list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
	605	+ pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
	606	+out_dev_disable:
427	607	nvmet_ns_dev_disable(ns);
428	608	goto out_unlock;
429	609	}
..	..	@@ -431,15 +611,20 @@
431	611	void nvmet_ns_disable(struct nvmet_ns *ns)
432	612	{
433	613	struct nvmet_subsys *subsys = ns->subsys;
	614	+ struct nvmet_ctrl *ctrl;
434	615
435	616	mutex_lock(&subsys->lock);
436	617	if (!ns->enabled)
437	618	goto out_unlock;
438	619
439	620	ns->enabled = false;
440		- list_del_rcu(&ns->dev_link);
	621	+ xa_erase(&ns->subsys->namespaces, ns->nsid);
441	622	if (ns->nsid == subsys->max_nsid)
442	623	subsys->max_nsid = nvmet_max_nsid(subsys);
	624	+
	625	+ list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry)
	626	+ pci_dev_put(radix_tree_delete(&ctrl->p2p_ns_map, ns->nsid));
	627	+
443	628	mutex_unlock(&subsys->lock);
444	629
445	630	/*
..	..	@@ -456,6 +641,7 @@
456	641	percpu_ref_exit(&ns->ref);
457	642
458	643	mutex_lock(&subsys->lock);
	644	+
459	645	subsys->nr_namespaces--;
460	646	nvmet_ns_changed(subsys, ns->nsid);
461	647	nvmet_ns_dev_disable(ns);
..	..	@@ -483,7 +669,6 @@
483	669	if (!ns)
484	670	return NULL;
485	671
486		- INIT_LIST_HEAD(&ns->dev_link);
487	672	init_completion(&ns->disable_done);
488	673
489	674	ns->nsid = nsid;
..	..	@@ -500,35 +685,74 @@
500	685	return ns;
501	686	}
502	687
503		-static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
	688	+static void nvmet_update_sq_head(struct nvmet_req *req)
504	689	{
505		- u32 old_sqhd, new_sqhd;
506		- u16 sqhd;
507		-
508		- if (status)
509		- nvmet_set_status(req, status);
510		-
511	690	if (req->sq->size) {
	691	+ u32 old_sqhd, new_sqhd;
	692	+
512	693	do {
513	694	old_sqhd = req->sq->sqhd;
514	695	new_sqhd = (old_sqhd + 1) % req->sq->size;
515	696	} while (cmpxchg(&req->sq->sqhd, old_sqhd, new_sqhd) !=
516	697	old_sqhd);
517	698	}
518		- sqhd = req->sq->sqhd & 0x0000FFFF;
519		- req->rsp->sq_head = cpu_to_le16(sqhd);
520		- req->rsp->sq_id = cpu_to_le16(req->sq->qid);
521		- req->rsp->command_id = req->cmd->common.command_id;
	699	+ req->cqe->sq_head = cpu_to_le16(req->sq->sqhd & 0x0000FFFF);
	700	+}
522	701
523		- if (req->ns)
524		- nvmet_put_namespace(req->ns);
	702	+static void nvmet_set_error(struct nvmet_req *req, u16 status)
	703	+{
	704	+ struct nvmet_ctrl *ctrl = req->sq->ctrl;
	705	+ struct nvme_error_slot *new_error_slot;
	706	+ unsigned long flags;
	707	+
	708	+ req->cqe->status = cpu_to_le16(status << 1);
	709	+
	710	+ if (!ctrl \|\| req->error_loc == NVMET_NO_ERROR_LOC)
	711	+ return;
	712	+
	713	+ spin_lock_irqsave(&ctrl->error_lock, flags);
	714	+ ctrl->err_counter++;
	715	+ new_error_slot =
	716	+ &ctrl->slots[ctrl->err_counter % NVMET_ERROR_LOG_SLOTS];
	717	+
	718	+ new_error_slot->error_count = cpu_to_le64(ctrl->err_counter);
	719	+ new_error_slot->sqid = cpu_to_le16(req->sq->qid);
	720	+ new_error_slot->cmdid = cpu_to_le16(req->cmd->common.command_id);
	721	+ new_error_slot->status_field = cpu_to_le16(status << 1);
	722	+ new_error_slot->param_error_location = cpu_to_le16(req->error_loc);
	723	+ new_error_slot->lba = cpu_to_le64(req->error_slba);
	724	+ new_error_slot->nsid = req->cmd->common.nsid;
	725	+ spin_unlock_irqrestore(&ctrl->error_lock, flags);
	726	+
	727	+ /* set the more bit for this request */
	728	+ req->cqe->status \|= cpu_to_le16(1 << 14);
	729	+}
	730	+
	731	+static void __nvmet_req_complete(struct nvmet_req *req, u16 status)
	732	+{
	733	+ struct nvmet_ns *ns = req->ns;
	734	+
	735	+ if (!req->sq->sqhd_disabled)
	736	+ nvmet_update_sq_head(req);
	737	+ req->cqe->sq_id = cpu_to_le16(req->sq->qid);
	738	+ req->cqe->command_id = req->cmd->common.command_id;
	739	+
	740	+ if (unlikely(status))
	741	+ nvmet_set_error(req, status);
	742	+
	743	+ trace_nvmet_req_complete(req);
	744	+
525	745	req->ops->queue_response(req);
	746	+ if (ns)
	747	+ nvmet_put_namespace(ns);
526	748	}
527	749
528	750	void nvmet_req_complete(struct nvmet_req *req, u16 status)
529	751	{
	752	+ struct nvmet_sq *sq = req->sq;
	753	+
530	754	__nvmet_req_complete(req, status);
531		- percpu_ref_put(&req->sq->ref);
	755	+ percpu_ref_put(&sq->ref);
532	756	}
533	757	EXPORT_SYMBOL_GPL(nvmet_req_complete);
534	758
..	..	@@ -537,8 +761,6 @@
537	761	{
538	762	cq->qid = qid;
539	763	cq->size = size;
540		-
541		- ctrl->cqs[qid] = cq;
542	764	}
543	765
544	766	void nvmet_sq_setup(struct nvmet_ctrl ctrl, struct nvmet_sq sq,
..	..	@@ -560,19 +782,28 @@
560	782
561	783	void nvmet_sq_destroy(struct nvmet_sq *sq)
562	784	{
	785	+ struct nvmet_ctrl *ctrl = sq->ctrl;
	786	+
563	787	/*
564	788	* If this is the admin queue, complete all AERs so that our
565	789	* queue doesn't have outstanding requests on it.
566	790	*/
567		- if (sq->ctrl && sq->ctrl->sqs && sq->ctrl->sqs[0] == sq)
568		- nvmet_async_events_free(sq->ctrl);
	791	+ if (ctrl && ctrl->sqs && ctrl->sqs[0] == sq)
	792	+ nvmet_async_events_failall(ctrl);
569	793	percpu_ref_kill_and_confirm(&sq->ref, nvmet_confirm_sq);
570	794	wait_for_completion(&sq->confirm_done);
571	795	wait_for_completion(&sq->free_done);
572	796	percpu_ref_exit(&sq->ref);
573	797
574		- if (sq->ctrl) {
575		- nvmet_ctrl_put(sq->ctrl);
	798	+ if (ctrl) {
	799	+ /*
	800	+ * The teardown flow may take some time, and the host may not
	801	+ * send us keep-alive during this period, hence reset the
	802	+ * traffic based keep-alive timer so we don't trigger a
	803	+ * controller teardown as a result of a keep-alive expiration.
	804	+ */
	805	+ ctrl->reset_tbkas = true;
	806	+ nvmet_ctrl_put(ctrl);
576	807	sq->ctrl = NULL; /* allows reusing the queue later */
577	808	}
578	809	}
..	..	@@ -639,15 +870,24 @@
639	870	if (unlikely(ret))
640	871	return ret;
641	872
	873	+ if (nvmet_req_passthru_ctrl(req))
	874	+ return nvmet_parse_passthru_io_cmd(req);
	875	+
642	876	req->ns = nvmet_find_namespace(req->sq->ctrl, cmd->rw.nsid);
643		- if (unlikely(!req->ns))
	877	+ if (unlikely(!req->ns)) {
	878	+ req->error_loc = offsetof(struct nvme_common_command, nsid);
644	879	return NVME_SC_INVALID_NS \| NVME_SC_DNR;
	880	+ }
645	881	ret = nvmet_check_ana_state(req->port, req->ns);
646		- if (unlikely(ret))
	882	+ if (unlikely(ret)) {
	883	+ req->error_loc = offsetof(struct nvme_common_command, nsid);
647	884	return ret;
	885	+ }
648	886	ret = nvmet_io_cmd_check_access(req);
649		- if (unlikely(ret))
	887	+ if (unlikely(ret)) {
	888	+ req->error_loc = offsetof(struct nvme_common_command, nsid);
650	889	return ret;
	890	+ }
651	891
652	892	if (req->ns->file)
653	893	return nvmet_file_parse_io_cmd(req);
..	..	@@ -665,13 +905,20 @@
665	905	req->sq = sq;
666	906	req->ops = ops;
667	907	req->sg = NULL;
	908	+ req->metadata_sg = NULL;
668	909	req->sg_cnt = 0;
	910	+ req->metadata_sg_cnt = 0;
669	911	req->transfer_len = 0;
670		- req->rsp->status = 0;
	912	+ req->metadata_len = 0;
	913	+ req->cqe->status = 0;
	914	+ req->cqe->sq_head = 0;
671	915	req->ns = NULL;
	916	+ req->error_loc = NVMET_NO_ERROR_LOC;
	917	+ req->error_slba = 0;
672	918
673	919	/* no support for fused commands yet */
674	920	if (unlikely(flags & (NVME_CMD_FUSE_FIRST \| NVME_CMD_FUSE_SECOND))) {
	921	+ req->error_loc = offsetof(struct nvme_common_command, flags);
675	922	status = NVME_SC_INVALID_FIELD \| NVME_SC_DNR;
676	923	goto fail;
677	924	}
..	..	@@ -682,29 +929,31 @@
682	929	* byte aligned.
683	930	*/
684	931	if (unlikely((flags & NVME_CMD_SGL_ALL) != NVME_CMD_SGL_METABUF)) {
	932	+ req->error_loc = offsetof(struct nvme_common_command, flags);
685	933	status = NVME_SC_INVALID_FIELD \| NVME_SC_DNR;
686	934	goto fail;
687	935	}
688	936
689	937	if (unlikely(!req->sq->ctrl))
690		- /* will return an error for any Non-connect command: */
	938	+ /* will return an error for any non-connect command: */
691	939	status = nvmet_parse_connect_cmd(req);
692	940	else if (likely(req->sq->qid != 0))
693	941	status = nvmet_parse_io_cmd(req);
694		- else if (req->cmd->common.opcode == nvme_fabrics_command)
695		- status = nvmet_parse_fabrics_cmd(req);
696		- else if (req->sq->ctrl->subsys->type == NVME_NQN_DISC)
697		- status = nvmet_parse_discovery_cmd(req);
698	942	else
699	943	status = nvmet_parse_admin_cmd(req);
700	944
701	945	if (status)
702	946	goto fail;
703	947
	948	+ trace_nvmet_req_init(req, req->cmd);
	949	+
704	950	if (unlikely(!percpu_ref_tryget_live(&sq->ref))) {
705	951	status = NVME_SC_INVALID_FIELD \| NVME_SC_DNR;
706	952	goto fail;
707	953	}
	954	+
	955	+ if (sq->ctrl)
	956	+ sq->ctrl->reset_tbkas = true;
708	957
709	958	return true;
710	959
..	..	@@ -722,14 +971,112 @@
722	971	}
723	972	EXPORT_SYMBOL_GPL(nvmet_req_uninit);
724	973
725		-void nvmet_req_execute(struct nvmet_req *req)
	974	+bool nvmet_check_transfer_len(struct nvmet_req *req, size_t len)
726	975	{
727		- if (unlikely(req->data_len != req->transfer_len))
	976	+ if (unlikely(len != req->transfer_len)) {
	977	+ req->error_loc = offsetof(struct nvme_common_command, dptr);
728	978	nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA \| NVME_SC_DNR);
729		- else
730		- req->execute(req);
	979	+ return false;
	980	+ }
	981	+
	982	+ return true;
731	983	}
732		-EXPORT_SYMBOL_GPL(nvmet_req_execute);
	984	+EXPORT_SYMBOL_GPL(nvmet_check_transfer_len);
	985	+
	986	+bool nvmet_check_data_len_lte(struct nvmet_req *req, size_t data_len)
	987	+{
	988	+ if (unlikely(data_len > req->transfer_len)) {
	989	+ req->error_loc = offsetof(struct nvme_common_command, dptr);
	990	+ nvmet_req_complete(req, NVME_SC_SGL_INVALID_DATA \| NVME_SC_DNR);
	991	+ return false;
	992	+ }
	993	+
	994	+ return true;
	995	+}
	996	+
	997	+static unsigned int nvmet_data_transfer_len(struct nvmet_req *req)
	998	+{
	999	+ return req->transfer_len - req->metadata_len;
	1000	+}
	1001	+
	1002	+static int nvmet_req_alloc_p2pmem_sgls(struct pci_dev *p2p_dev,
	1003	+ struct nvmet_req *req)
	1004	+{
	1005	+ req->sg = pci_p2pmem_alloc_sgl(p2p_dev, &req->sg_cnt,
	1006	+ nvmet_data_transfer_len(req));
	1007	+ if (!req->sg)
	1008	+ goto out_err;
	1009	+
	1010	+ if (req->metadata_len) {
	1011	+ req->metadata_sg = pci_p2pmem_alloc_sgl(p2p_dev,
	1012	+ &req->metadata_sg_cnt, req->metadata_len);
	1013	+ if (!req->metadata_sg)
	1014	+ goto out_free_sg;
	1015	+ }
	1016	+
	1017	+ req->p2p_dev = p2p_dev;
	1018	+
	1019	+ return 0;
	1020	+out_free_sg:
	1021	+ pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
	1022	+out_err:
	1023	+ return -ENOMEM;
	1024	+}
	1025	+
	1026	+static struct pci_dev nvmet_req_find_p2p_dev(struct nvmet_req req)
	1027	+{
	1028	+ if (!IS_ENABLED(CONFIG_PCI_P2PDMA) \|\|
	1029	+ !req->sq->ctrl \|\| !req->sq->qid \|\| !req->ns)
	1030	+ return NULL;
	1031	+ return radix_tree_lookup(&req->sq->ctrl->p2p_ns_map, req->ns->nsid);
	1032	+}
	1033	+
	1034	+int nvmet_req_alloc_sgls(struct nvmet_req *req)
	1035	+{
	1036	+ struct pci_dev *p2p_dev = nvmet_req_find_p2p_dev(req);
	1037	+
	1038	+ if (p2p_dev && !nvmet_req_alloc_p2pmem_sgls(p2p_dev, req))
	1039	+ return 0;
	1040	+
	1041	+ req->sg = sgl_alloc(nvmet_data_transfer_len(req), GFP_KERNEL,
	1042	+ &req->sg_cnt);
	1043	+ if (unlikely(!req->sg))
	1044	+ goto out;
	1045	+
	1046	+ if (req->metadata_len) {
	1047	+ req->metadata_sg = sgl_alloc(req->metadata_len, GFP_KERNEL,
	1048	+ &req->metadata_sg_cnt);
	1049	+ if (unlikely(!req->metadata_sg))
	1050	+ goto out_free;
	1051	+ }
	1052	+
	1053	+ return 0;
	1054	+out_free:
	1055	+ sgl_free(req->sg);
	1056	+out:
	1057	+ return -ENOMEM;
	1058	+}
	1059	+EXPORT_SYMBOL_GPL(nvmet_req_alloc_sgls);
	1060	+
	1061	+void nvmet_req_free_sgls(struct nvmet_req *req)
	1062	+{
	1063	+ if (req->p2p_dev) {
	1064	+ pci_p2pmem_free_sgl(req->p2p_dev, req->sg);
	1065	+ if (req->metadata_sg)
	1066	+ pci_p2pmem_free_sgl(req->p2p_dev, req->metadata_sg);
	1067	+ req->p2p_dev = NULL;
	1068	+ } else {
	1069	+ sgl_free(req->sg);
	1070	+ if (req->metadata_sg)
	1071	+ sgl_free(req->metadata_sg);
	1072	+ }
	1073	+
	1074	+ req->sg = NULL;
	1075	+ req->metadata_sg = NULL;
	1076	+ req->sg_cnt = 0;
	1077	+ req->metadata_sg_cnt = 0;
	1078	+}
	1079	+EXPORT_SYMBOL_GPL(nvmet_req_free_sgls);
733	1080
734	1081	static inline bool nvmet_cc_en(u32 cc)
735	1082	{
..	..	@@ -853,7 +1200,7 @@
853	1200	if (!subsys) {
854	1201	pr_warn("connect request for invalid subsystem %s!\n",
855	1202	subsysnqn);
856		- req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
	1203	+ req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
857	1204	return NVME_SC_CONNECT_INVALID_PARAM \| NVME_SC_DNR;
858	1205	}
859	1206
..	..	@@ -874,7 +1221,7 @@
874	1221
875	1222	pr_warn("could not find controller %d for subsys %s / host %s\n",
876	1223	cntlid, subsysnqn, hostnqn);
877		- req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
	1224	+ req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(cntlid);
878	1225	status = NVME_SC_CONNECT_INVALID_PARAM \| NVME_SC_DNR;
879	1226
880	1227	out:
..	..	@@ -899,12 +1246,16 @@
899	1246	return 0;
900	1247	}
901	1248
902		-static bool __nvmet_host_allowed(struct nvmet_subsys *subsys,
903		- const char *hostnqn)
	1249	+bool nvmet_host_allowed(struct nvmet_subsys subsys, const char hostnqn)
904	1250	{
905	1251	struct nvmet_host_link *p;
906	1252
	1253	+ lockdep_assert_held(&nvmet_config_sem);
	1254	+
907	1255	if (subsys->allow_any_host)
	1256	+ return true;
	1257	+
	1258	+ if (subsys->type == NVME_NQN_DISC) /* allow all access to disc subsys */
908	1259	return true;
909	1260
910	1261	list_for_each_entry(p, &subsys->hosts, entry) {
..	..	@@ -915,28 +1266,36 @@
915	1266	return false;
916	1267	}
917	1268
918		-static bool nvmet_host_discovery_allowed(struct nvmet_req *req,
919		- const char *hostnqn)
	1269	+/*
	1270	+ * Note: ctrl->subsys->lock should be held when calling this function
	1271	+ */
	1272	+static void nvmet_setup_p2p_ns_map(struct nvmet_ctrl *ctrl,
	1273	+ struct nvmet_req *req)
920	1274	{
921		- struct nvmet_subsys_link *s;
	1275	+ struct nvmet_ns *ns;
	1276	+ unsigned long idx;
922	1277
923		- list_for_each_entry(s, &req->port->subsystems, entry) {
924		- if (__nvmet_host_allowed(s->subsys, hostnqn))
925		- return true;
926		- }
	1278	+ if (!req->p2p_client)
	1279	+ return;
927	1280
928		- return false;
	1281	+ ctrl->p2p_client = get_device(req->p2p_client);
	1282	+
	1283	+ xa_for_each(&ctrl->subsys->namespaces, idx, ns)
	1284	+ nvmet_p2pmem_ns_add_p2p(ctrl, ns);
929	1285	}
930	1286
931		-bool nvmet_host_allowed(struct nvmet_req req, struct nvmet_subsys subsys,
932		- const char *hostnqn)
	1287	+/*
	1288	+ * Note: ctrl->subsys->lock should be held when calling this function
	1289	+ */
	1290	+static void nvmet_release_p2p_ns_map(struct nvmet_ctrl *ctrl)
933	1291	{
934		- lockdep_assert_held(&nvmet_config_sem);
	1292	+ struct radix_tree_iter iter;
	1293	+ void __rcu **slot;
935	1294
936		- if (subsys->type == NVME_NQN_DISC)
937		- return nvmet_host_discovery_allowed(req, hostnqn);
938		- else
939		- return __nvmet_host_allowed(subsys, hostnqn);
	1295	+ radix_tree_for_each_slot(slot, &ctrl->p2p_ns_map, &iter, 0)
	1296	+ pci_dev_put(radix_tree_deref_slot(slot));
	1297	+
	1298	+ put_device(ctrl->p2p_client);
940	1299	}
941	1300
942	1301	static void nvmet_fatal_error_handler(struct work_struct *work)
..	..	@@ -961,16 +1320,16 @@
961	1320	if (!subsys) {
962	1321	pr_warn("connect request for invalid subsystem %s!\n",
963	1322	subsysnqn);
964		- req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
	1323	+ req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(subsysnqn);
965	1324	goto out;
966	1325	}
967	1326
968	1327	status = NVME_SC_CONNECT_INVALID_PARAM \| NVME_SC_DNR;
969	1328	down_read(&nvmet_config_sem);
970		- if (!nvmet_host_allowed(req, subsys, hostnqn)) {
	1329	+ if (!nvmet_host_allowed(subsys, hostnqn)) {
971	1330	pr_info("connect by host %s for subsystem %s not allowed\n",
972	1331	hostnqn, subsysnqn);
973		- req->rsp->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
	1332	+ req->cqe->result.u32 = IPO_IATTR_CONNECT_DATA(hostnqn);
974	1333	up_read(&nvmet_config_sem);
975	1334	status = NVME_SC_CONNECT_INVALID_HOST \| NVME_SC_DNR;
976	1335	goto out_put_subsystem;
..	..	@@ -989,6 +1348,7 @@
989	1348
990	1349	INIT_WORK(&ctrl->async_event_work, nvmet_async_event_work);
991	1350	INIT_LIST_HEAD(&ctrl->async_events);
	1351	+ INIT_RADIX_TREE(&ctrl->p2p_ns_map, GFP_KERNEL);
992	1352	INIT_WORK(&ctrl->fatal_err_work, nvmet_fatal_error_handler);
993	1353
994	1354	memcpy(ctrl->subsysnqn, subsysnqn, NVMF_NQN_SIZE);
..	..	@@ -1003,20 +1363,17 @@
1003	1363	if (!ctrl->changed_ns_list)
1004	1364	goto out_free_ctrl;
1005	1365
1006		- ctrl->cqs = kcalloc(subsys->max_qid + 1,
1007		- sizeof(struct nvmet_cq *),
1008		- GFP_KERNEL);
1009		- if (!ctrl->cqs)
1010		- goto out_free_changed_ns_list;
1011		-
1012	1366	ctrl->sqs = kcalloc(subsys->max_qid + 1,
1013	1367	sizeof(struct nvmet_sq *),
1014	1368	GFP_KERNEL);
1015	1369	if (!ctrl->sqs)
1016		- goto out_free_cqs;
	1370	+ goto out_free_changed_ns_list;
	1371	+
	1372	+ if (subsys->cntlid_min > subsys->cntlid_max)
	1373	+ goto out_free_sqs;
1017	1374
1018	1375	ret = ida_simple_get(&cntlid_ida,
1019		- NVME_CNTLID_MIN, NVME_CNTLID_MAX,
	1376	+ subsys->cntlid_min, subsys->cntlid_max,
1020	1377	GFP_KERNEL);
1021	1378	if (ret < 0) {
1022	1379	status = NVME_SC_CONNECT_CTRL_BUSY \| NVME_SC_DNR;
..	..	@@ -1025,46 +1382,32 @@
1025	1382	ctrl->cntlid = ret;
1026	1383
1027	1384	ctrl->ops = req->ops;
1028		- if (ctrl->subsys->type == NVME_NQN_DISC) {
1029		- /* Don't accept keep-alive timeout for discovery controllers */
1030		- if (kato) {
1031		- status = NVME_SC_INVALID_FIELD \| NVME_SC_DNR;
1032		- goto out_remove_ida;
1033		- }
1034	1385
1035		- /*
1036		- * Discovery controllers use some arbitrary high value in order
1037		- * to cleanup stale discovery sessions
1038		- *
1039		- * From the latest base diff RC:
1040		- * "The Keep Alive command is not supported by
1041		- * Discovery controllers. A transport may specify a
1042		- * fixed Discovery controller activity timeout value
1043		- * (e.g., 2 minutes). If no commands are received
1044		- * by a Discovery controller within that time
1045		- * period, the controller may perform the
1046		- * actions for Keep Alive Timer expiration".
1047		- */
1048		- ctrl->kato = NVMET_DISC_KATO;
1049		- } else {
1050		- /* keep-alive timeout in seconds */
1051		- ctrl->kato = DIV_ROUND_UP(kato, 1000);
1052		- }
	1386	+ /*
	1387	+ * Discovery controllers may use some arbitrary high value
	1388	+ * in order to cleanup stale discovery sessions
	1389	+ */
	1390	+ if ((ctrl->subsys->type == NVME_NQN_DISC) && !kato)
	1391	+ kato = NVMET_DISC_KATO_MS;
	1392	+
	1393	+ /* keep-alive timeout in seconds */
	1394	+ ctrl->kato = DIV_ROUND_UP(kato, 1000);
	1395	+
	1396	+ ctrl->err_counter = 0;
	1397	+ spin_lock_init(&ctrl->error_lock);
	1398	+
1053	1399	nvmet_start_keep_alive_timer(ctrl);
1054	1400
1055	1401	mutex_lock(&subsys->lock);
1056	1402	list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
	1403	+ nvmet_setup_p2p_ns_map(ctrl, req);
1057	1404	mutex_unlock(&subsys->lock);
1058	1405
1059	1406	*ctrlp = ctrl;
1060	1407	return 0;
1061	1408
1062		-out_remove_ida:
1063		- ida_simple_remove(&cntlid_ida, ctrl->cntlid);
1064	1409	out_free_sqs:
1065	1410	kfree(ctrl->sqs);
1066		-out_free_cqs:
1067		- kfree(ctrl->cqs);
1068	1411	out_free_changed_ns_list:
1069	1412	kfree(ctrl->changed_ns_list);
1070	1413	out_free_ctrl:
..	..	@@ -1081,6 +1424,7 @@
1081	1424	struct nvmet_subsys *subsys = ctrl->subsys;
1082	1425
1083	1426	mutex_lock(&subsys->lock);
	1427	+ nvmet_release_p2p_ns_map(ctrl);
1084	1428	list_del(&ctrl->subsys_entry);
1085	1429	mutex_unlock(&subsys->lock);
1086	1430
..	..	@@ -1091,8 +1435,8 @@
1091	1435
1092	1436	ida_simple_remove(&cntlid_ida, ctrl->cntlid);
1093	1437
	1438	+ nvmet_async_events_free(ctrl);
1094	1439	kfree(ctrl->sqs);
1095		- kfree(ctrl->cqs);
1096	1440	kfree(ctrl->changed_ns_list);
1097	1441	kfree(ctrl);
1098	1442
..	..	@@ -1123,8 +1467,7 @@
1123	1467	if (!port)
1124	1468	return NULL;
1125	1469
1126		- if (!strncmp(NVME_DISC_SUBSYS_NAME, subsysnqn,
1127		- NVMF_NQN_SIZE)) {
	1470	+ if (!strcmp(NVME_DISC_SUBSYS_NAME, subsysnqn)) {
1128	1471	if (!kref_get_unless_zero(&nvmet_disc_subsys->ref))
1129	1472	return NULL;
1130	1473	return nvmet_disc_subsys;
..	..	@@ -1151,9 +1494,9 @@
1151	1494
1152	1495	subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
1153	1496	if (!subsys)
1154		- return NULL;
	1497	+ return ERR_PTR(-ENOMEM);
1155	1498
1156		- subsys->ver = NVME_VS(1, 3, 0); /* NVMe 1.3.0 */
	1499	+ subsys->ver = NVMET_DEFAULT_VS;
1157	1500	/* generate a random serial number as our controllers are ephemeral: */
1158	1501	get_random_bytes(&subsys->serial, sizeof(subsys->serial));
1159	1502
..	..	@@ -1167,20 +1510,21 @@
1167	1510	default:
1168	1511	pr_err("%s: Unknown Subsystem type - %d\n", __func__, type);
1169	1512	kfree(subsys);
1170		- return NULL;
	1513	+ return ERR_PTR(-EINVAL);
1171	1514	}
1172	1515	subsys->type = type;
1173	1516	subsys->subsysnqn = kstrndup(subsysnqn, NVMF_NQN_SIZE,
1174	1517	GFP_KERNEL);
1175	1518	if (!subsys->subsysnqn) {
1176	1519	kfree(subsys);
1177		- return NULL;
	1520	+ return ERR_PTR(-ENOMEM);
1178	1521	}
1179		-
	1522	+ subsys->cntlid_min = NVME_CNTLID_MIN;
	1523	+ subsys->cntlid_max = NVME_CNTLID_MAX;
1180	1524	kref_init(&subsys->ref);
1181	1525
1182	1526	mutex_init(&subsys->lock);
1183		- INIT_LIST_HEAD(&subsys->namespaces);
	1527	+ xa_init(&subsys->namespaces);
1184	1528	INIT_LIST_HEAD(&subsys->ctrls);
1185	1529	INIT_LIST_HEAD(&subsys->hosts);
1186	1530
..	..	@@ -1192,9 +1536,13 @@
1192	1536	struct nvmet_subsys *subsys =
1193	1537	container_of(ref, struct nvmet_subsys, ref);
1194	1538
1195		- WARN_ON_ONCE(!list_empty(&subsys->namespaces));
	1539	+ WARN_ON_ONCE(!xa_empty(&subsys->namespaces));
	1540	+
	1541	+ xa_destroy(&subsys->namespaces);
	1542	+ nvmet_passthru_subsys_free(subsys);
1196	1543
1197	1544	kfree(subsys->subsysnqn);
	1545	+ kfree_rcu(subsys->model, rcuhead);
1198	1546	kfree(subsys);
1199	1547	}
1200	1548
..	..	@@ -1258,3 +1606,4 @@
1258	1606	module_exit(nvmet_exit);
1259	1607
1260	1608	MODULE_LICENSE("GPL v2");
	1609	+MODULE_IMPORT_NS(VFS_internal_I_am_really_a_filesystem_and_am_NOT_a_driver);