1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2015 Intel Corporation
4 * Keith Busch <kbusch@kernel.org>
5 */
6 #include <linux/blkdev.h>
7 #include <linux/pr.h>
8 #include <linux/unaligned.h>
9
10 #include "nvme.h"
11
nvme_pr_type_from_blk(enum pr_type type)12 static enum nvme_pr_type nvme_pr_type_from_blk(enum pr_type type)
13 {
14 switch (type) {
15 case PR_WRITE_EXCLUSIVE:
16 return NVME_PR_WRITE_EXCLUSIVE;
17 case PR_EXCLUSIVE_ACCESS:
18 return NVME_PR_EXCLUSIVE_ACCESS;
19 case PR_WRITE_EXCLUSIVE_REG_ONLY:
20 return NVME_PR_WRITE_EXCLUSIVE_REG_ONLY;
21 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
22 return NVME_PR_EXCLUSIVE_ACCESS_REG_ONLY;
23 case PR_WRITE_EXCLUSIVE_ALL_REGS:
24 return NVME_PR_WRITE_EXCLUSIVE_ALL_REGS;
25 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
26 return NVME_PR_EXCLUSIVE_ACCESS_ALL_REGS;
27 }
28
29 return 0;
30 }
31
block_pr_type_from_nvme(enum nvme_pr_type type)32 static enum pr_type block_pr_type_from_nvme(enum nvme_pr_type type)
33 {
34 switch (type) {
35 case NVME_PR_WRITE_EXCLUSIVE:
36 return PR_WRITE_EXCLUSIVE;
37 case NVME_PR_EXCLUSIVE_ACCESS:
38 return PR_EXCLUSIVE_ACCESS;
39 case NVME_PR_WRITE_EXCLUSIVE_REG_ONLY:
40 return PR_WRITE_EXCLUSIVE_REG_ONLY;
41 case NVME_PR_EXCLUSIVE_ACCESS_REG_ONLY:
42 return PR_EXCLUSIVE_ACCESS_REG_ONLY;
43 case NVME_PR_WRITE_EXCLUSIVE_ALL_REGS:
44 return PR_WRITE_EXCLUSIVE_ALL_REGS;
45 case NVME_PR_EXCLUSIVE_ACCESS_ALL_REGS:
46 return PR_EXCLUSIVE_ACCESS_ALL_REGS;
47 }
48
49 return 0;
50 }
51
nvme_send_ns_head_pr_command(struct block_device * bdev,struct nvme_command * c,void * data,unsigned int data_len)52 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
53 struct nvme_command *c, void *data, unsigned int data_len)
54 {
55 struct nvme_ns_head *head = bdev->bd_disk->private_data;
56 int srcu_idx = srcu_read_lock(&head->srcu);
57 struct nvme_ns *ns = nvme_find_path(head);
58 int ret = -EWOULDBLOCK;
59
60 if (ns) {
61 c->common.nsid = cpu_to_le32(ns->head->ns_id);
62 ret = nvme_submit_sync_cmd(ns->queue, c, data, data_len);
63 }
64 srcu_read_unlock(&head->srcu, srcu_idx);
65 return ret;
66 }
67
nvme_send_ns_pr_command(struct nvme_ns * ns,struct nvme_command * c,void * data,unsigned int data_len)68 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
69 void *data, unsigned int data_len)
70 {
71 c->common.nsid = cpu_to_le32(ns->head->ns_id);
72 return nvme_submit_sync_cmd(ns->queue, c, data, data_len);
73 }
74
nvme_status_to_pr_err(int status)75 static int nvme_status_to_pr_err(int status)
76 {
77 if (nvme_is_path_error(status))
78 return PR_STS_PATH_FAILED;
79
80 switch (status & NVME_SCT_SC_MASK) {
81 case NVME_SC_SUCCESS:
82 return PR_STS_SUCCESS;
83 case NVME_SC_RESERVATION_CONFLICT:
84 return PR_STS_RESERVATION_CONFLICT;
85 case NVME_SC_BAD_ATTRIBUTES:
86 case NVME_SC_INVALID_OPCODE:
87 case NVME_SC_INVALID_FIELD:
88 case NVME_SC_INVALID_NS:
89 return -EINVAL;
90 default:
91 return PR_STS_IOERR;
92 }
93 }
94
nvme_send_pr_command(struct block_device * bdev,struct nvme_command * c,void * data,unsigned int data_len)95 static int nvme_send_pr_command(struct block_device *bdev,
96 struct nvme_command *c, void *data, unsigned int data_len)
97 {
98 if (nvme_disk_is_ns_head(bdev->bd_disk))
99 return nvme_send_ns_head_pr_command(bdev, c, data, data_len);
100
101 return nvme_send_ns_pr_command(bdev->bd_disk->private_data, c, data,
102 data_len);
103 }
104
nvme_pr_command(struct block_device * bdev,u32 cdw10,u64 key,u64 sa_key,u8 op)105 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
106 u64 key, u64 sa_key, u8 op)
107 {
108 struct nvme_command c = { };
109 u8 data[16] = { 0, };
110 int ret;
111
112 put_unaligned_le64(key, &data[0]);
113 put_unaligned_le64(sa_key, &data[8]);
114
115 c.common.opcode = op;
116 c.common.cdw10 = cpu_to_le32(cdw10);
117
118 ret = nvme_send_pr_command(bdev, &c, data, sizeof(data));
119 if (ret < 0)
120 return ret;
121
122 return nvme_status_to_pr_err(ret);
123 }
124
nvme_pr_register(struct block_device * bdev,u64 old,u64 new,unsigned flags)125 static int nvme_pr_register(struct block_device *bdev, u64 old,
126 u64 new, unsigned flags)
127 {
128 u32 cdw10;
129
130 if (flags & ~PR_FL_IGNORE_KEY)
131 return -EOPNOTSUPP;
132
133 cdw10 = old ? 2 : 0;
134 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
135 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
136 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
137 }
138
nvme_pr_reserve(struct block_device * bdev,u64 key,enum pr_type type,unsigned flags)139 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
140 enum pr_type type, unsigned flags)
141 {
142 u32 cdw10;
143
144 if (flags & ~PR_FL_IGNORE_KEY)
145 return -EOPNOTSUPP;
146
147 cdw10 = nvme_pr_type_from_blk(type) << 8;
148 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
149 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
150 }
151
nvme_pr_preempt(struct block_device * bdev,u64 old,u64 new,enum pr_type type,bool abort)152 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
153 enum pr_type type, bool abort)
154 {
155 u32 cdw10 = nvme_pr_type_from_blk(type) << 8 | (abort ? 2 : 1);
156
157 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
158 }
159
nvme_pr_clear(struct block_device * bdev,u64 key)160 static int nvme_pr_clear(struct block_device *bdev, u64 key)
161 {
162 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
163
164 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
165 }
166
nvme_pr_release(struct block_device * bdev,u64 key,enum pr_type type)167 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
168 {
169 u32 cdw10 = nvme_pr_type_from_blk(type) << 8 | (key ? 0 : 1 << 3);
170
171 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
172 }
173
nvme_pr_resv_report(struct block_device * bdev,void * data,u32 data_len,bool * eds)174 static int nvme_pr_resv_report(struct block_device *bdev, void *data,
175 u32 data_len, bool *eds)
176 {
177 struct nvme_command c = { };
178 int ret;
179
180 c.common.opcode = nvme_cmd_resv_report;
181 c.common.cdw10 = cpu_to_le32(nvme_bytes_to_numd(data_len));
182 c.common.cdw11 = cpu_to_le32(NVME_EXTENDED_DATA_STRUCT);
183 *eds = true;
184
185 retry:
186 ret = nvme_send_pr_command(bdev, &c, data, data_len);
187 if (ret == NVME_SC_HOST_ID_INCONSIST &&
188 c.common.cdw11 == cpu_to_le32(NVME_EXTENDED_DATA_STRUCT)) {
189 c.common.cdw11 = 0;
190 *eds = false;
191 goto retry;
192 }
193
194 if (ret < 0)
195 return ret;
196
197 return nvme_status_to_pr_err(ret);
198 }
199
nvme_pr_read_keys(struct block_device * bdev,struct pr_keys * keys_info)200 static int nvme_pr_read_keys(struct block_device *bdev,
201 struct pr_keys *keys_info)
202 {
203 u32 rse_len, num_keys = keys_info->num_keys;
204 struct nvme_reservation_status_ext *rse;
205 int ret, i;
206 bool eds;
207
208 /*
209 * Assume we are using 128-bit host IDs and allocate a buffer large
210 * enough to get enough keys to fill the return keys buffer.
211 */
212 rse_len = struct_size(rse, regctl_eds, num_keys);
213 rse = kzalloc(rse_len, GFP_KERNEL);
214 if (!rse)
215 return -ENOMEM;
216
217 ret = nvme_pr_resv_report(bdev, rse, rse_len, &eds);
218 if (ret)
219 goto free_rse;
220
221 keys_info->generation = le32_to_cpu(rse->gen);
222 keys_info->num_keys = get_unaligned_le16(&rse->regctl);
223
224 num_keys = min(num_keys, keys_info->num_keys);
225 for (i = 0; i < num_keys; i++) {
226 if (eds) {
227 keys_info->keys[i] =
228 le64_to_cpu(rse->regctl_eds[i].rkey);
229 } else {
230 struct nvme_reservation_status *rs;
231
232 rs = (struct nvme_reservation_status *)rse;
233 keys_info->keys[i] = le64_to_cpu(rs->regctl_ds[i].rkey);
234 }
235 }
236
237 free_rse:
238 kfree(rse);
239 return ret;
240 }
241
nvme_pr_read_reservation(struct block_device * bdev,struct pr_held_reservation * resv)242 static int nvme_pr_read_reservation(struct block_device *bdev,
243 struct pr_held_reservation *resv)
244 {
245 struct nvme_reservation_status_ext tmp_rse, *rse;
246 int ret, i, num_regs;
247 u32 rse_len;
248 bool eds;
249
250 get_num_regs:
251 /*
252 * Get the number of registrations so we know how big to allocate
253 * the response buffer.
254 */
255 ret = nvme_pr_resv_report(bdev, &tmp_rse, sizeof(tmp_rse), &eds);
256 if (ret)
257 return ret;
258
259 num_regs = get_unaligned_le16(&tmp_rse.regctl);
260 if (!num_regs) {
261 resv->generation = le32_to_cpu(tmp_rse.gen);
262 return 0;
263 }
264
265 rse_len = struct_size(rse, regctl_eds, num_regs);
266 rse = kzalloc(rse_len, GFP_KERNEL);
267 if (!rse)
268 return -ENOMEM;
269
270 ret = nvme_pr_resv_report(bdev, rse, rse_len, &eds);
271 if (ret)
272 goto free_rse;
273
274 if (num_regs != get_unaligned_le16(&rse->regctl)) {
275 kfree(rse);
276 goto get_num_regs;
277 }
278
279 resv->generation = le32_to_cpu(rse->gen);
280 resv->type = block_pr_type_from_nvme(rse->rtype);
281
282 for (i = 0; i < num_regs; i++) {
283 if (eds) {
284 if (rse->regctl_eds[i].rcsts) {
285 resv->key = le64_to_cpu(rse->regctl_eds[i].rkey);
286 break;
287 }
288 } else {
289 struct nvme_reservation_status *rs;
290
291 rs = (struct nvme_reservation_status *)rse;
292 if (rs->regctl_ds[i].rcsts) {
293 resv->key = le64_to_cpu(rs->regctl_ds[i].rkey);
294 break;
295 }
296 }
297 }
298
299 free_rse:
300 kfree(rse);
301 return ret;
302 }
303
304 const struct pr_ops nvme_pr_ops = {
305 .pr_register = nvme_pr_register,
306 .pr_reserve = nvme_pr_reserve,
307 .pr_release = nvme_pr_release,
308 .pr_preempt = nvme_pr_preempt,
309 .pr_clear = nvme_pr_clear,
310 .pr_read_keys = nvme_pr_read_keys,
311 .pr_read_reservation = nvme_pr_read_reservation,
312 };
313