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 <asm/unaligned.h> |
9 | |
10 | #include "nvme.h" |
11 | |
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 | |
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 | |
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(ssp: &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(q: ns->queue, cmd: c, buf: data, bufflen: data_len); |
63 | } |
64 | srcu_read_unlock(ssp: &head->srcu, idx: srcu_idx); |
65 | return ret; |
66 | } |
67 | |
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(q: ns->queue, cmd: c, buf: data, bufflen: data_len); |
73 | } |
74 | |
75 | static int nvme_sc_to_pr_err(int nvme_sc) |
76 | { |
77 | if (nvme_is_path_error(status: nvme_sc)) |
78 | return PR_STS_PATH_FAILED; |
79 | |
80 | switch (nvme_sc) { |
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_ONCS_NOT_SUPPORTED: |
86 | return -EOPNOTSUPP; |
87 | case NVME_SC_BAD_ATTRIBUTES: |
88 | case NVME_SC_INVALID_OPCODE: |
89 | case NVME_SC_INVALID_FIELD: |
90 | case NVME_SC_INVALID_NS: |
91 | return -EINVAL; |
92 | default: |
93 | return PR_STS_IOERR; |
94 | } |
95 | } |
96 | |
97 | static int nvme_send_pr_command(struct block_device *bdev, |
98 | struct nvme_command *c, void *data, unsigned int data_len) |
99 | { |
100 | if (nvme_disk_is_ns_head(disk: bdev->bd_disk)) |
101 | return nvme_send_ns_head_pr_command(bdev, c, data, data_len); |
102 | |
103 | return nvme_send_ns_pr_command(ns: bdev->bd_disk->private_data, c, data, |
104 | data_len); |
105 | } |
106 | |
107 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, |
108 | u64 key, u64 sa_key, u8 op) |
109 | { |
110 | struct nvme_command c = { }; |
111 | u8 data[16] = { 0, }; |
112 | int ret; |
113 | |
114 | put_unaligned_le64(val: key, p: &data[0]); |
115 | put_unaligned_le64(val: sa_key, p: &data[8]); |
116 | |
117 | c.common.opcode = op; |
118 | c.common.cdw10 = cpu_to_le32(cdw10); |
119 | |
120 | ret = nvme_send_pr_command(bdev, c: &c, data, data_len: sizeof(data)); |
121 | if (ret < 0) |
122 | return ret; |
123 | |
124 | return nvme_sc_to_pr_err(nvme_sc: ret); |
125 | } |
126 | |
127 | static int nvme_pr_register(struct block_device *bdev, u64 old, |
128 | u64 new, unsigned flags) |
129 | { |
130 | u32 cdw10; |
131 | |
132 | if (flags & ~PR_FL_IGNORE_KEY) |
133 | return -EOPNOTSUPP; |
134 | |
135 | cdw10 = old ? 2 : 0; |
136 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; |
137 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ |
138 | return nvme_pr_command(bdev, cdw10, key: old, sa_key: new, op: nvme_cmd_resv_register); |
139 | } |
140 | |
141 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, |
142 | enum pr_type type, unsigned flags) |
143 | { |
144 | u32 cdw10; |
145 | |
146 | if (flags & ~PR_FL_IGNORE_KEY) |
147 | return -EOPNOTSUPP; |
148 | |
149 | cdw10 = nvme_pr_type_from_blk(type) << 8; |
150 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); |
151 | return nvme_pr_command(bdev, cdw10, key, sa_key: 0, op: nvme_cmd_resv_acquire); |
152 | } |
153 | |
154 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, |
155 | enum pr_type type, bool abort) |
156 | { |
157 | u32 cdw10 = nvme_pr_type_from_blk(type) << 8 | (abort ? 2 : 1); |
158 | |
159 | return nvme_pr_command(bdev, cdw10, key: old, sa_key: new, op: nvme_cmd_resv_acquire); |
160 | } |
161 | |
162 | static int nvme_pr_clear(struct block_device *bdev, u64 key) |
163 | { |
164 | u32 cdw10 = 1 | (key ? 0 : 1 << 3); |
165 | |
166 | return nvme_pr_command(bdev, cdw10, key, sa_key: 0, op: nvme_cmd_resv_release); |
167 | } |
168 | |
169 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) |
170 | { |
171 | u32 cdw10 = nvme_pr_type_from_blk(type) << 8 | (key ? 0 : 1 << 3); |
172 | |
173 | return nvme_pr_command(bdev, cdw10, key, sa_key: 0, op: nvme_cmd_resv_release); |
174 | } |
175 | |
176 | static int nvme_pr_resv_report(struct block_device *bdev, void *data, |
177 | u32 data_len, bool *eds) |
178 | { |
179 | struct nvme_command c = { }; |
180 | int ret; |
181 | |
182 | c.common.opcode = nvme_cmd_resv_report; |
183 | c.common.cdw10 = cpu_to_le32(nvme_bytes_to_numd(data_len)); |
184 | c.common.cdw11 = cpu_to_le32(NVME_EXTENDED_DATA_STRUCT); |
185 | *eds = true; |
186 | |
187 | retry: |
188 | ret = nvme_send_pr_command(bdev, c: &c, data, data_len); |
189 | if (ret == NVME_SC_HOST_ID_INCONSIST && |
190 | c.common.cdw11 == cpu_to_le32(NVME_EXTENDED_DATA_STRUCT)) { |
191 | c.common.cdw11 = 0; |
192 | *eds = false; |
193 | goto retry; |
194 | } |
195 | |
196 | if (ret < 0) |
197 | return ret; |
198 | |
199 | return nvme_sc_to_pr_err(nvme_sc: ret); |
200 | } |
201 | |
202 | static int nvme_pr_read_keys(struct block_device *bdev, |
203 | struct pr_keys *keys_info) |
204 | { |
205 | u32 rse_len, num_keys = keys_info->num_keys; |
206 | struct nvme_reservation_status_ext *rse; |
207 | int ret, i; |
208 | bool eds; |
209 | |
210 | /* |
211 | * Assume we are using 128-bit host IDs and allocate a buffer large |
212 | * enough to get enough keys to fill the return keys buffer. |
213 | */ |
214 | rse_len = struct_size(rse, regctl_eds, num_keys); |
215 | rse = kzalloc(size: rse_len, GFP_KERNEL); |
216 | if (!rse) |
217 | return -ENOMEM; |
218 | |
219 | ret = nvme_pr_resv_report(bdev, data: rse, data_len: rse_len, eds: &eds); |
220 | if (ret) |
221 | goto free_rse; |
222 | |
223 | keys_info->generation = le32_to_cpu(rse->gen); |
224 | keys_info->num_keys = get_unaligned_le16(p: &rse->regctl); |
225 | |
226 | num_keys = min(num_keys, keys_info->num_keys); |
227 | for (i = 0; i < num_keys; i++) { |
228 | if (eds) { |
229 | keys_info->keys[i] = |
230 | le64_to_cpu(rse->regctl_eds[i].rkey); |
231 | } else { |
232 | struct nvme_reservation_status *rs; |
233 | |
234 | rs = (struct nvme_reservation_status *)rse; |
235 | keys_info->keys[i] = le64_to_cpu(rs->regctl_ds[i].rkey); |
236 | } |
237 | } |
238 | |
239 | free_rse: |
240 | kfree(objp: rse); |
241 | return ret; |
242 | } |
243 | |
244 | static int nvme_pr_read_reservation(struct block_device *bdev, |
245 | struct pr_held_reservation *resv) |
246 | { |
247 | struct nvme_reservation_status_ext tmp_rse, *rse; |
248 | int ret, i, num_regs; |
249 | u32 rse_len; |
250 | bool eds; |
251 | |
252 | get_num_regs: |
253 | /* |
254 | * Get the number of registrations so we know how big to allocate |
255 | * the response buffer. |
256 | */ |
257 | ret = nvme_pr_resv_report(bdev, data: &tmp_rse, data_len: sizeof(tmp_rse), eds: &eds); |
258 | if (ret) |
259 | return ret; |
260 | |
261 | num_regs = get_unaligned_le16(p: &tmp_rse.regctl); |
262 | if (!num_regs) { |
263 | resv->generation = le32_to_cpu(tmp_rse.gen); |
264 | return 0; |
265 | } |
266 | |
267 | rse_len = struct_size(rse, regctl_eds, num_regs); |
268 | rse = kzalloc(size: rse_len, GFP_KERNEL); |
269 | if (!rse) |
270 | return -ENOMEM; |
271 | |
272 | ret = nvme_pr_resv_report(bdev, data: rse, data_len: rse_len, eds: &eds); |
273 | if (ret) |
274 | goto free_rse; |
275 | |
276 | if (num_regs != get_unaligned_le16(p: &rse->regctl)) { |
277 | kfree(objp: rse); |
278 | goto get_num_regs; |
279 | } |
280 | |
281 | resv->generation = le32_to_cpu(rse->gen); |
282 | resv->type = block_pr_type_from_nvme(type: rse->rtype); |
283 | |
284 | for (i = 0; i < num_regs; i++) { |
285 | if (eds) { |
286 | if (rse->regctl_eds[i].rcsts) { |
287 | resv->key = le64_to_cpu(rse->regctl_eds[i].rkey); |
288 | break; |
289 | } |
290 | } else { |
291 | struct nvme_reservation_status *rs; |
292 | |
293 | rs = (struct nvme_reservation_status *)rse; |
294 | if (rs->regctl_ds[i].rcsts) { |
295 | resv->key = le64_to_cpu(rs->regctl_ds[i].rkey); |
296 | break; |
297 | } |
298 | } |
299 | } |
300 | |
301 | free_rse: |
302 | kfree(objp: rse); |
303 | return ret; |
304 | } |
305 | |
306 | const struct pr_ops nvme_pr_ops = { |
307 | .pr_register = nvme_pr_register, |
308 | .pr_reserve = nvme_pr_reserve, |
309 | .pr_release = nvme_pr_release, |
310 | .pr_preempt = nvme_pr_preempt, |
311 | .pr_clear = nvme_pr_clear, |
312 | .pr_read_keys = nvme_pr_read_keys, |
313 | .pr_read_reservation = nvme_pr_read_reservation, |
314 | }; |
315 | |