Commit | Line | Data |
---|---|---|
21d34711 CH |
1 | /* |
2 | * NVM Express device driver | |
3 | * Copyright (c) 2011-2014, Intel Corporation. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify it | |
6 | * under the terms and conditions of the GNU General Public License, | |
7 | * version 2, as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | */ | |
14 | ||
15 | #include <linux/blkdev.h> | |
16 | #include <linux/blk-mq.h> | |
5fd4ce1b | 17 | #include <linux/delay.h> |
21d34711 | 18 | #include <linux/errno.h> |
1673f1f0 | 19 | #include <linux/hdreg.h> |
21d34711 | 20 | #include <linux/kernel.h> |
5bae7f73 CH |
21 | #include <linux/module.h> |
22 | #include <linux/list_sort.h> | |
21d34711 CH |
23 | #include <linux/slab.h> |
24 | #include <linux/types.h> | |
1673f1f0 CH |
25 | #include <linux/pr.h> |
26 | #include <linux/ptrace.h> | |
27 | #include <linux/nvme_ioctl.h> | |
28 | #include <linux/t10-pi.h> | |
29 | #include <scsi/sg.h> | |
30 | #include <asm/unaligned.h> | |
21d34711 CH |
31 | |
32 | #include "nvme.h" | |
33 | ||
5bae7f73 CH |
34 | static int nvme_major; |
35 | module_param(nvme_major, int, 0); | |
36 | ||
1673f1f0 CH |
37 | DEFINE_SPINLOCK(dev_list_lock); |
38 | ||
39 | static void nvme_free_ns(struct kref *kref) | |
40 | { | |
41 | struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref); | |
42 | ||
43 | if (ns->type == NVME_NS_LIGHTNVM) | |
44 | nvme_nvm_unregister(ns->queue, ns->disk->disk_name); | |
45 | ||
46 | spin_lock(&dev_list_lock); | |
47 | ns->disk->private_data = NULL; | |
48 | spin_unlock(&dev_list_lock); | |
49 | ||
50 | nvme_put_ctrl(ns->ctrl); | |
51 | put_disk(ns->disk); | |
52 | kfree(ns); | |
53 | } | |
54 | ||
5bae7f73 | 55 | static void nvme_put_ns(struct nvme_ns *ns) |
1673f1f0 CH |
56 | { |
57 | kref_put(&ns->kref, nvme_free_ns); | |
58 | } | |
59 | ||
60 | static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk) | |
61 | { | |
62 | struct nvme_ns *ns; | |
63 | ||
64 | spin_lock(&dev_list_lock); | |
65 | ns = disk->private_data; | |
66 | if (ns && !kref_get_unless_zero(&ns->kref)) | |
67 | ns = NULL; | |
68 | spin_unlock(&dev_list_lock); | |
69 | ||
70 | return ns; | |
71 | } | |
72 | ||
4160982e CH |
73 | struct request *nvme_alloc_request(struct request_queue *q, |
74 | struct nvme_command *cmd, unsigned int flags) | |
21d34711 CH |
75 | { |
76 | bool write = cmd->common.opcode & 1; | |
21d34711 | 77 | struct request *req; |
21d34711 | 78 | |
4160982e | 79 | req = blk_mq_alloc_request(q, write, flags); |
21d34711 | 80 | if (IS_ERR(req)) |
4160982e | 81 | return req; |
21d34711 CH |
82 | |
83 | req->cmd_type = REQ_TYPE_DRV_PRIV; | |
84 | req->cmd_flags |= REQ_FAILFAST_DRIVER; | |
85 | req->__data_len = 0; | |
86 | req->__sector = (sector_t) -1; | |
87 | req->bio = req->biotail = NULL; | |
88 | ||
21d34711 CH |
89 | req->cmd = (unsigned char *)cmd; |
90 | req->cmd_len = sizeof(struct nvme_command); | |
91 | req->special = (void *)0; | |
92 | ||
4160982e CH |
93 | return req; |
94 | } | |
95 | ||
96 | /* | |
97 | * Returns 0 on success. If the result is negative, it's a Linux error code; | |
98 | * if the result is positive, it's an NVM Express status code | |
99 | */ | |
100 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
101 | void *buffer, unsigned bufflen, u32 *result, unsigned timeout) | |
102 | { | |
103 | struct request *req; | |
104 | int ret; | |
105 | ||
106 | req = nvme_alloc_request(q, cmd, 0); | |
107 | if (IS_ERR(req)) | |
108 | return PTR_ERR(req); | |
109 | ||
110 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
111 | ||
21d34711 CH |
112 | if (buffer && bufflen) { |
113 | ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL); | |
114 | if (ret) | |
115 | goto out; | |
4160982e CH |
116 | } |
117 | ||
118 | blk_execute_rq(req->q, NULL, req, 0); | |
119 | if (result) | |
120 | *result = (u32)(uintptr_t)req->special; | |
121 | ret = req->errors; | |
122 | out: | |
123 | blk_mq_free_request(req); | |
124 | return ret; | |
125 | } | |
126 | ||
127 | int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, | |
128 | void *buffer, unsigned bufflen) | |
129 | { | |
130 | return __nvme_submit_sync_cmd(q, cmd, buffer, bufflen, NULL, 0); | |
131 | } | |
132 | ||
0b7f1f26 KB |
133 | int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd, |
134 | void __user *ubuffer, unsigned bufflen, | |
135 | void __user *meta_buffer, unsigned meta_len, u32 meta_seed, | |
136 | u32 *result, unsigned timeout) | |
4160982e | 137 | { |
0b7f1f26 KB |
138 | bool write = cmd->common.opcode & 1; |
139 | struct nvme_ns *ns = q->queuedata; | |
140 | struct gendisk *disk = ns ? ns->disk : NULL; | |
4160982e | 141 | struct request *req; |
0b7f1f26 KB |
142 | struct bio *bio = NULL; |
143 | void *meta = NULL; | |
4160982e CH |
144 | int ret; |
145 | ||
146 | req = nvme_alloc_request(q, cmd, 0); | |
147 | if (IS_ERR(req)) | |
148 | return PTR_ERR(req); | |
149 | ||
150 | req->timeout = timeout ? timeout : ADMIN_TIMEOUT; | |
151 | ||
152 | if (ubuffer && bufflen) { | |
21d34711 CH |
153 | ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen, |
154 | GFP_KERNEL); | |
155 | if (ret) | |
156 | goto out; | |
157 | bio = req->bio; | |
21d34711 | 158 | |
0b7f1f26 KB |
159 | if (!disk) |
160 | goto submit; | |
161 | bio->bi_bdev = bdget_disk(disk, 0); | |
162 | if (!bio->bi_bdev) { | |
163 | ret = -ENODEV; | |
164 | goto out_unmap; | |
165 | } | |
166 | ||
167 | if (meta_buffer) { | |
168 | struct bio_integrity_payload *bip; | |
169 | ||
170 | meta = kmalloc(meta_len, GFP_KERNEL); | |
171 | if (!meta) { | |
172 | ret = -ENOMEM; | |
173 | goto out_unmap; | |
174 | } | |
175 | ||
176 | if (write) { | |
177 | if (copy_from_user(meta, meta_buffer, | |
178 | meta_len)) { | |
179 | ret = -EFAULT; | |
180 | goto out_free_meta; | |
181 | } | |
182 | } | |
183 | ||
184 | bip = bio_integrity_alloc(bio, GFP_KERNEL, 1); | |
185 | if (!bip) { | |
186 | ret = -ENOMEM; | |
187 | goto out_free_meta; | |
188 | } | |
189 | ||
190 | bip->bip_iter.bi_size = meta_len; | |
191 | bip->bip_iter.bi_sector = meta_seed; | |
192 | ||
193 | ret = bio_integrity_add_page(bio, virt_to_page(meta), | |
194 | meta_len, offset_in_page(meta)); | |
195 | if (ret != meta_len) { | |
196 | ret = -ENOMEM; | |
197 | goto out_free_meta; | |
198 | } | |
199 | } | |
200 | } | |
201 | submit: | |
202 | blk_execute_rq(req->q, disk, req, 0); | |
203 | ret = req->errors; | |
21d34711 CH |
204 | if (result) |
205 | *result = (u32)(uintptr_t)req->special; | |
0b7f1f26 KB |
206 | if (meta && !ret && !write) { |
207 | if (copy_to_user(meta_buffer, meta, meta_len)) | |
208 | ret = -EFAULT; | |
209 | } | |
210 | out_free_meta: | |
211 | kfree(meta); | |
212 | out_unmap: | |
213 | if (bio) { | |
214 | if (disk && bio->bi_bdev) | |
215 | bdput(bio->bi_bdev); | |
216 | blk_rq_unmap_user(bio); | |
217 | } | |
21d34711 CH |
218 | out: |
219 | blk_mq_free_request(req); | |
220 | return ret; | |
221 | } | |
222 | ||
0b7f1f26 KB |
223 | int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd, |
224 | void __user *ubuffer, unsigned bufflen, u32 *result, | |
225 | unsigned timeout) | |
226 | { | |
227 | return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0, | |
228 | result, timeout); | |
229 | } | |
230 | ||
1c63dc66 | 231 | int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id) |
21d34711 CH |
232 | { |
233 | struct nvme_command c = { }; | |
234 | int error; | |
235 | ||
236 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
237 | c.identify.opcode = nvme_admin_identify; | |
238 | c.identify.cns = cpu_to_le32(1); | |
239 | ||
240 | *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL); | |
241 | if (!*id) | |
242 | return -ENOMEM; | |
243 | ||
244 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
245 | sizeof(struct nvme_id_ctrl)); | |
246 | if (error) | |
247 | kfree(*id); | |
248 | return error; | |
249 | } | |
250 | ||
1c63dc66 | 251 | int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid, |
21d34711 CH |
252 | struct nvme_id_ns **id) |
253 | { | |
254 | struct nvme_command c = { }; | |
255 | int error; | |
256 | ||
257 | /* gcc-4.4.4 (at least) has issues with initializers and anon unions */ | |
258 | c.identify.opcode = nvme_admin_identify, | |
259 | c.identify.nsid = cpu_to_le32(nsid), | |
260 | ||
261 | *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL); | |
262 | if (!*id) | |
263 | return -ENOMEM; | |
264 | ||
265 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *id, | |
266 | sizeof(struct nvme_id_ns)); | |
267 | if (error) | |
268 | kfree(*id); | |
269 | return error; | |
270 | } | |
271 | ||
1c63dc66 | 272 | int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid, |
21d34711 CH |
273 | dma_addr_t dma_addr, u32 *result) |
274 | { | |
275 | struct nvme_command c; | |
276 | ||
277 | memset(&c, 0, sizeof(c)); | |
278 | c.features.opcode = nvme_admin_get_features; | |
279 | c.features.nsid = cpu_to_le32(nsid); | |
280 | c.features.prp1 = cpu_to_le64(dma_addr); | |
281 | c.features.fid = cpu_to_le32(fid); | |
282 | ||
4160982e | 283 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0); |
21d34711 CH |
284 | } |
285 | ||
1c63dc66 | 286 | int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11, |
21d34711 CH |
287 | dma_addr_t dma_addr, u32 *result) |
288 | { | |
289 | struct nvme_command c; | |
290 | ||
291 | memset(&c, 0, sizeof(c)); | |
292 | c.features.opcode = nvme_admin_set_features; | |
293 | c.features.prp1 = cpu_to_le64(dma_addr); | |
294 | c.features.fid = cpu_to_le32(fid); | |
295 | c.features.dword11 = cpu_to_le32(dword11); | |
296 | ||
4160982e | 297 | return __nvme_submit_sync_cmd(dev->admin_q, &c, NULL, 0, result, 0); |
21d34711 CH |
298 | } |
299 | ||
1c63dc66 | 300 | int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log) |
21d34711 CH |
301 | { |
302 | struct nvme_command c = { }; | |
303 | int error; | |
304 | ||
305 | c.common.opcode = nvme_admin_get_log_page, | |
306 | c.common.nsid = cpu_to_le32(0xFFFFFFFF), | |
307 | c.common.cdw10[0] = cpu_to_le32( | |
308 | (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) | | |
309 | NVME_LOG_SMART), | |
310 | ||
311 | *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL); | |
312 | if (!*log) | |
313 | return -ENOMEM; | |
314 | ||
315 | error = nvme_submit_sync_cmd(dev->admin_q, &c, *log, | |
316 | sizeof(struct nvme_smart_log)); | |
317 | if (error) | |
318 | kfree(*log); | |
319 | return error; | |
320 | } | |
1673f1f0 CH |
321 | |
322 | static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio) | |
323 | { | |
324 | struct nvme_user_io io; | |
325 | struct nvme_command c; | |
326 | unsigned length, meta_len; | |
327 | void __user *metadata; | |
328 | ||
329 | if (copy_from_user(&io, uio, sizeof(io))) | |
330 | return -EFAULT; | |
331 | ||
332 | switch (io.opcode) { | |
333 | case nvme_cmd_write: | |
334 | case nvme_cmd_read: | |
335 | case nvme_cmd_compare: | |
336 | break; | |
337 | default: | |
338 | return -EINVAL; | |
339 | } | |
340 | ||
341 | length = (io.nblocks + 1) << ns->lba_shift; | |
342 | meta_len = (io.nblocks + 1) * ns->ms; | |
343 | metadata = (void __user *)(uintptr_t)io.metadata; | |
344 | ||
345 | if (ns->ext) { | |
346 | length += meta_len; | |
347 | meta_len = 0; | |
348 | } else if (meta_len) { | |
349 | if ((io.metadata & 3) || !io.metadata) | |
350 | return -EINVAL; | |
351 | } | |
352 | ||
353 | memset(&c, 0, sizeof(c)); | |
354 | c.rw.opcode = io.opcode; | |
355 | c.rw.flags = io.flags; | |
356 | c.rw.nsid = cpu_to_le32(ns->ns_id); | |
357 | c.rw.slba = cpu_to_le64(io.slba); | |
358 | c.rw.length = cpu_to_le16(io.nblocks); | |
359 | c.rw.control = cpu_to_le16(io.control); | |
360 | c.rw.dsmgmt = cpu_to_le32(io.dsmgmt); | |
361 | c.rw.reftag = cpu_to_le32(io.reftag); | |
362 | c.rw.apptag = cpu_to_le16(io.apptag); | |
363 | c.rw.appmask = cpu_to_le16(io.appmask); | |
364 | ||
365 | return __nvme_submit_user_cmd(ns->queue, &c, | |
366 | (void __user *)(uintptr_t)io.addr, length, | |
367 | metadata, meta_len, io.slba, NULL, 0); | |
368 | } | |
369 | ||
370 | int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns, | |
371 | struct nvme_passthru_cmd __user *ucmd) | |
372 | { | |
373 | struct nvme_passthru_cmd cmd; | |
374 | struct nvme_command c; | |
375 | unsigned timeout = 0; | |
376 | int status; | |
377 | ||
378 | if (!capable(CAP_SYS_ADMIN)) | |
379 | return -EACCES; | |
380 | if (copy_from_user(&cmd, ucmd, sizeof(cmd))) | |
381 | return -EFAULT; | |
382 | ||
383 | memset(&c, 0, sizeof(c)); | |
384 | c.common.opcode = cmd.opcode; | |
385 | c.common.flags = cmd.flags; | |
386 | c.common.nsid = cpu_to_le32(cmd.nsid); | |
387 | c.common.cdw2[0] = cpu_to_le32(cmd.cdw2); | |
388 | c.common.cdw2[1] = cpu_to_le32(cmd.cdw3); | |
389 | c.common.cdw10[0] = cpu_to_le32(cmd.cdw10); | |
390 | c.common.cdw10[1] = cpu_to_le32(cmd.cdw11); | |
391 | c.common.cdw10[2] = cpu_to_le32(cmd.cdw12); | |
392 | c.common.cdw10[3] = cpu_to_le32(cmd.cdw13); | |
393 | c.common.cdw10[4] = cpu_to_le32(cmd.cdw14); | |
394 | c.common.cdw10[5] = cpu_to_le32(cmd.cdw15); | |
395 | ||
396 | if (cmd.timeout_ms) | |
397 | timeout = msecs_to_jiffies(cmd.timeout_ms); | |
398 | ||
399 | status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c, | |
400 | (void __user *)cmd.addr, cmd.data_len, | |
401 | &cmd.result, timeout); | |
402 | if (status >= 0) { | |
403 | if (put_user(cmd.result, &ucmd->result)) | |
404 | return -EFAULT; | |
405 | } | |
406 | ||
407 | return status; | |
408 | } | |
409 | ||
410 | static int nvme_ioctl(struct block_device *bdev, fmode_t mode, | |
411 | unsigned int cmd, unsigned long arg) | |
412 | { | |
413 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
414 | ||
415 | switch (cmd) { | |
416 | case NVME_IOCTL_ID: | |
417 | force_successful_syscall_return(); | |
418 | return ns->ns_id; | |
419 | case NVME_IOCTL_ADMIN_CMD: | |
420 | return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg); | |
421 | case NVME_IOCTL_IO_CMD: | |
422 | return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg); | |
423 | case NVME_IOCTL_SUBMIT_IO: | |
424 | return nvme_submit_io(ns, (void __user *)arg); | |
425 | case SG_GET_VERSION_NUM: | |
426 | return nvme_sg_get_version_num((void __user *)arg); | |
427 | case SG_IO: | |
428 | return nvme_sg_io(ns, (void __user *)arg); | |
429 | default: | |
430 | return -ENOTTY; | |
431 | } | |
432 | } | |
433 | ||
434 | #ifdef CONFIG_COMPAT | |
435 | static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
436 | unsigned int cmd, unsigned long arg) | |
437 | { | |
438 | switch (cmd) { | |
439 | case SG_IO: | |
440 | return -ENOIOCTLCMD; | |
441 | } | |
442 | return nvme_ioctl(bdev, mode, cmd, arg); | |
443 | } | |
444 | #else | |
445 | #define nvme_compat_ioctl NULL | |
446 | #endif | |
447 | ||
448 | static int nvme_open(struct block_device *bdev, fmode_t mode) | |
449 | { | |
450 | return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO; | |
451 | } | |
452 | ||
453 | static void nvme_release(struct gendisk *disk, fmode_t mode) | |
454 | { | |
455 | nvme_put_ns(disk->private_data); | |
456 | } | |
457 | ||
458 | static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
459 | { | |
460 | /* some standard values */ | |
461 | geo->heads = 1 << 6; | |
462 | geo->sectors = 1 << 5; | |
463 | geo->cylinders = get_capacity(bdev->bd_disk) >> 11; | |
464 | return 0; | |
465 | } | |
466 | ||
467 | #ifdef CONFIG_BLK_DEV_INTEGRITY | |
468 | static void nvme_init_integrity(struct nvme_ns *ns) | |
469 | { | |
470 | struct blk_integrity integrity; | |
471 | ||
472 | switch (ns->pi_type) { | |
473 | case NVME_NS_DPS_PI_TYPE3: | |
474 | integrity.profile = &t10_pi_type3_crc; | |
475 | break; | |
476 | case NVME_NS_DPS_PI_TYPE1: | |
477 | case NVME_NS_DPS_PI_TYPE2: | |
478 | integrity.profile = &t10_pi_type1_crc; | |
479 | break; | |
480 | default: | |
481 | integrity.profile = NULL; | |
482 | break; | |
483 | } | |
484 | integrity.tuple_size = ns->ms; | |
485 | blk_integrity_register(ns->disk, &integrity); | |
486 | blk_queue_max_integrity_segments(ns->queue, 1); | |
487 | } | |
488 | #else | |
489 | static void nvme_init_integrity(struct nvme_ns *ns) | |
490 | { | |
491 | } | |
492 | #endif /* CONFIG_BLK_DEV_INTEGRITY */ | |
493 | ||
494 | static void nvme_config_discard(struct nvme_ns *ns) | |
495 | { | |
496 | u32 logical_block_size = queue_logical_block_size(ns->queue); | |
497 | ns->queue->limits.discard_zeroes_data = 0; | |
498 | ns->queue->limits.discard_alignment = logical_block_size; | |
499 | ns->queue->limits.discard_granularity = logical_block_size; | |
500 | blk_queue_max_discard_sectors(ns->queue, 0xffffffff); | |
501 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue); | |
502 | } | |
503 | ||
5bae7f73 | 504 | static int nvme_revalidate_disk(struct gendisk *disk) |
1673f1f0 CH |
505 | { |
506 | struct nvme_ns *ns = disk->private_data; | |
507 | struct nvme_id_ns *id; | |
508 | u8 lbaf, pi_type; | |
509 | u16 old_ms; | |
510 | unsigned short bs; | |
511 | ||
512 | if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) { | |
513 | dev_warn(ns->ctrl->dev, "%s: Identify failure nvme%dn%d\n", | |
514 | __func__, ns->ctrl->instance, ns->ns_id); | |
515 | return -ENODEV; | |
516 | } | |
517 | if (id->ncap == 0) { | |
518 | kfree(id); | |
519 | return -ENODEV; | |
520 | } | |
521 | ||
522 | if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) { | |
523 | if (nvme_nvm_register(ns->queue, disk->disk_name)) { | |
524 | dev_warn(ns->ctrl->dev, | |
525 | "%s: LightNVM init failure\n", __func__); | |
526 | kfree(id); | |
527 | return -ENODEV; | |
528 | } | |
529 | ns->type = NVME_NS_LIGHTNVM; | |
530 | } | |
531 | ||
532 | old_ms = ns->ms; | |
533 | lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK; | |
534 | ns->lba_shift = id->lbaf[lbaf].ds; | |
535 | ns->ms = le16_to_cpu(id->lbaf[lbaf].ms); | |
536 | ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT); | |
537 | ||
538 | /* | |
539 | * If identify namespace failed, use default 512 byte block size so | |
540 | * block layer can use before failing read/write for 0 capacity. | |
541 | */ | |
542 | if (ns->lba_shift == 0) | |
543 | ns->lba_shift = 9; | |
544 | bs = 1 << ns->lba_shift; | |
545 | ||
546 | /* XXX: PI implementation requires metadata equal t10 pi tuple size */ | |
547 | pi_type = ns->ms == sizeof(struct t10_pi_tuple) ? | |
548 | id->dps & NVME_NS_DPS_PI_MASK : 0; | |
549 | ||
550 | blk_mq_freeze_queue(disk->queue); | |
551 | if (blk_get_integrity(disk) && (ns->pi_type != pi_type || | |
552 | ns->ms != old_ms || | |
553 | bs != queue_logical_block_size(disk->queue) || | |
554 | (ns->ms && ns->ext))) | |
555 | blk_integrity_unregister(disk); | |
556 | ||
557 | ns->pi_type = pi_type; | |
558 | blk_queue_logical_block_size(ns->queue, bs); | |
559 | ||
560 | if (ns->ms && !ns->ext) | |
561 | nvme_init_integrity(ns); | |
562 | ||
563 | if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk)) | |
564 | set_capacity(disk, 0); | |
565 | else | |
566 | set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9)); | |
567 | ||
568 | if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM) | |
569 | nvme_config_discard(ns); | |
570 | blk_mq_unfreeze_queue(disk->queue); | |
571 | ||
572 | kfree(id); | |
573 | return 0; | |
574 | } | |
575 | ||
576 | static char nvme_pr_type(enum pr_type type) | |
577 | { | |
578 | switch (type) { | |
579 | case PR_WRITE_EXCLUSIVE: | |
580 | return 1; | |
581 | case PR_EXCLUSIVE_ACCESS: | |
582 | return 2; | |
583 | case PR_WRITE_EXCLUSIVE_REG_ONLY: | |
584 | return 3; | |
585 | case PR_EXCLUSIVE_ACCESS_REG_ONLY: | |
586 | return 4; | |
587 | case PR_WRITE_EXCLUSIVE_ALL_REGS: | |
588 | return 5; | |
589 | case PR_EXCLUSIVE_ACCESS_ALL_REGS: | |
590 | return 6; | |
591 | default: | |
592 | return 0; | |
593 | } | |
594 | }; | |
595 | ||
596 | static int nvme_pr_command(struct block_device *bdev, u32 cdw10, | |
597 | u64 key, u64 sa_key, u8 op) | |
598 | { | |
599 | struct nvme_ns *ns = bdev->bd_disk->private_data; | |
600 | struct nvme_command c; | |
601 | u8 data[16] = { 0, }; | |
602 | ||
603 | put_unaligned_le64(key, &data[0]); | |
604 | put_unaligned_le64(sa_key, &data[8]); | |
605 | ||
606 | memset(&c, 0, sizeof(c)); | |
607 | c.common.opcode = op; | |
608 | c.common.nsid = cpu_to_le32(ns->ns_id); | |
609 | c.common.cdw10[0] = cpu_to_le32(cdw10); | |
610 | ||
611 | return nvme_submit_sync_cmd(ns->queue, &c, data, 16); | |
612 | } | |
613 | ||
614 | static int nvme_pr_register(struct block_device *bdev, u64 old, | |
615 | u64 new, unsigned flags) | |
616 | { | |
617 | u32 cdw10; | |
618 | ||
619 | if (flags & ~PR_FL_IGNORE_KEY) | |
620 | return -EOPNOTSUPP; | |
621 | ||
622 | cdw10 = old ? 2 : 0; | |
623 | cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0; | |
624 | cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */ | |
625 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register); | |
626 | } | |
627 | ||
628 | static int nvme_pr_reserve(struct block_device *bdev, u64 key, | |
629 | enum pr_type type, unsigned flags) | |
630 | { | |
631 | u32 cdw10; | |
632 | ||
633 | if (flags & ~PR_FL_IGNORE_KEY) | |
634 | return -EOPNOTSUPP; | |
635 | ||
636 | cdw10 = nvme_pr_type(type) << 8; | |
637 | cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0); | |
638 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire); | |
639 | } | |
640 | ||
641 | static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new, | |
642 | enum pr_type type, bool abort) | |
643 | { | |
644 | u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1; | |
645 | return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire); | |
646 | } | |
647 | ||
648 | static int nvme_pr_clear(struct block_device *bdev, u64 key) | |
649 | { | |
650 | u32 cdw10 = 1 | key ? 1 << 3 : 0; | |
651 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register); | |
652 | } | |
653 | ||
654 | static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
655 | { | |
656 | u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0; | |
657 | return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release); | |
658 | } | |
659 | ||
660 | static const struct pr_ops nvme_pr_ops = { | |
661 | .pr_register = nvme_pr_register, | |
662 | .pr_reserve = nvme_pr_reserve, | |
663 | .pr_release = nvme_pr_release, | |
664 | .pr_preempt = nvme_pr_preempt, | |
665 | .pr_clear = nvme_pr_clear, | |
666 | }; | |
667 | ||
5bae7f73 | 668 | static const struct block_device_operations nvme_fops = { |
1673f1f0 CH |
669 | .owner = THIS_MODULE, |
670 | .ioctl = nvme_ioctl, | |
671 | .compat_ioctl = nvme_compat_ioctl, | |
672 | .open = nvme_open, | |
673 | .release = nvme_release, | |
674 | .getgeo = nvme_getgeo, | |
675 | .revalidate_disk= nvme_revalidate_disk, | |
676 | .pr_ops = &nvme_pr_ops, | |
677 | }; | |
678 | ||
5fd4ce1b CH |
679 | static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled) |
680 | { | |
681 | unsigned long timeout = | |
682 | ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies; | |
683 | u32 csts, bit = enabled ? NVME_CSTS_RDY : 0; | |
684 | int ret; | |
685 | ||
686 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
687 | if ((csts & NVME_CSTS_RDY) == bit) | |
688 | break; | |
689 | ||
690 | msleep(100); | |
691 | if (fatal_signal_pending(current)) | |
692 | return -EINTR; | |
693 | if (time_after(jiffies, timeout)) { | |
694 | dev_err(ctrl->dev, | |
695 | "Device not ready; aborting %s\n", enabled ? | |
696 | "initialisation" : "reset"); | |
697 | return -ENODEV; | |
698 | } | |
699 | } | |
700 | ||
701 | return ret; | |
702 | } | |
703 | ||
704 | /* | |
705 | * If the device has been passed off to us in an enabled state, just clear | |
706 | * the enabled bit. The spec says we should set the 'shutdown notification | |
707 | * bits', but doing so may cause the device to complete commands to the | |
708 | * admin queue ... and we don't know what memory that might be pointing at! | |
709 | */ | |
710 | int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
711 | { | |
712 | int ret; | |
713 | ||
714 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
715 | ctrl->ctrl_config &= ~NVME_CC_ENABLE; | |
716 | ||
717 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
718 | if (ret) | |
719 | return ret; | |
720 | return nvme_wait_ready(ctrl, cap, false); | |
721 | } | |
722 | ||
723 | int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap) | |
724 | { | |
725 | /* | |
726 | * Default to a 4K page size, with the intention to update this | |
727 | * path in the future to accomodate architectures with differing | |
728 | * kernel and IO page sizes. | |
729 | */ | |
730 | unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12; | |
731 | int ret; | |
732 | ||
733 | if (page_shift < dev_page_min) { | |
734 | dev_err(ctrl->dev, | |
735 | "Minimum device page size %u too large for host (%u)\n", | |
736 | 1 << dev_page_min, 1 << page_shift); | |
737 | return -ENODEV; | |
738 | } | |
739 | ||
740 | ctrl->page_size = 1 << page_shift; | |
741 | ||
742 | ctrl->ctrl_config = NVME_CC_CSS_NVM; | |
743 | ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT; | |
744 | ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE; | |
745 | ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES; | |
746 | ctrl->ctrl_config |= NVME_CC_ENABLE; | |
747 | ||
748 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
749 | if (ret) | |
750 | return ret; | |
751 | return nvme_wait_ready(ctrl, cap, true); | |
752 | } | |
753 | ||
754 | int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl) | |
755 | { | |
756 | unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies; | |
757 | u32 csts; | |
758 | int ret; | |
759 | ||
760 | ctrl->ctrl_config &= ~NVME_CC_SHN_MASK; | |
761 | ctrl->ctrl_config |= NVME_CC_SHN_NORMAL; | |
762 | ||
763 | ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config); | |
764 | if (ret) | |
765 | return ret; | |
766 | ||
767 | while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) { | |
768 | if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT) | |
769 | break; | |
770 | ||
771 | msleep(100); | |
772 | if (fatal_signal_pending(current)) | |
773 | return -EINTR; | |
774 | if (time_after(jiffies, timeout)) { | |
775 | dev_err(ctrl->dev, | |
776 | "Device shutdown incomplete; abort shutdown\n"); | |
777 | return -ENODEV; | |
778 | } | |
779 | } | |
780 | ||
781 | return ret; | |
782 | } | |
783 | ||
7fd8930f CH |
784 | /* |
785 | * Initialize the cached copies of the Identify data and various controller | |
786 | * register in our nvme_ctrl structure. This should be called as soon as | |
787 | * the admin queue is fully up and running. | |
788 | */ | |
789 | int nvme_init_identify(struct nvme_ctrl *ctrl) | |
790 | { | |
791 | struct nvme_id_ctrl *id; | |
792 | u64 cap; | |
793 | int ret, page_shift; | |
794 | ||
795 | ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap); | |
796 | if (ret) { | |
797 | dev_err(ctrl->dev, "Reading CAP failed (%d)\n", ret); | |
798 | return ret; | |
799 | } | |
800 | page_shift = NVME_CAP_MPSMIN(cap) + 12; | |
801 | ||
802 | ret = nvme_identify_ctrl(ctrl, &id); | |
803 | if (ret) { | |
804 | dev_err(ctrl->dev, "Identify Controller failed (%d)\n", ret); | |
805 | return -EIO; | |
806 | } | |
807 | ||
808 | ctrl->oncs = le16_to_cpup(&id->oncs); | |
809 | ctrl->abort_limit = id->acl + 1; | |
810 | ctrl->vwc = id->vwc; | |
811 | memcpy(ctrl->serial, id->sn, sizeof(id->sn)); | |
812 | memcpy(ctrl->model, id->mn, sizeof(id->mn)); | |
813 | memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr)); | |
814 | if (id->mdts) | |
815 | ctrl->max_hw_sectors = 1 << (id->mdts + page_shift - 9); | |
816 | else | |
817 | ctrl->max_hw_sectors = UINT_MAX; | |
818 | ||
819 | if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) { | |
820 | unsigned int max_hw_sectors; | |
821 | ||
822 | ctrl->stripe_size = 1 << (id->vs[3] + page_shift); | |
823 | max_hw_sectors = ctrl->stripe_size >> (page_shift - 9); | |
824 | if (ctrl->max_hw_sectors) { | |
825 | ctrl->max_hw_sectors = min(max_hw_sectors, | |
826 | ctrl->max_hw_sectors); | |
827 | } else { | |
828 | ctrl->max_hw_sectors = max_hw_sectors; | |
829 | } | |
830 | } | |
831 | ||
832 | kfree(id); | |
833 | return 0; | |
834 | } | |
835 | ||
1673f1f0 CH |
836 | static void nvme_free_ctrl(struct kref *kref) |
837 | { | |
838 | struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref); | |
839 | ||
840 | ctrl->ops->free_ctrl(ctrl); | |
841 | } | |
842 | ||
843 | void nvme_put_ctrl(struct nvme_ctrl *ctrl) | |
844 | { | |
845 | kref_put(&ctrl->kref, nvme_free_ctrl); | |
846 | } | |
847 | ||
5bae7f73 CH |
848 | static int ns_cmp(void *priv, struct list_head *a, struct list_head *b) |
849 | { | |
850 | struct nvme_ns *nsa = container_of(a, struct nvme_ns, list); | |
851 | struct nvme_ns *nsb = container_of(b, struct nvme_ns, list); | |
852 | ||
853 | return nsa->ns_id - nsb->ns_id; | |
854 | } | |
855 | ||
856 | static struct nvme_ns *nvme_find_ns(struct nvme_ctrl *ctrl, unsigned nsid) | |
857 | { | |
858 | struct nvme_ns *ns; | |
859 | ||
860 | list_for_each_entry(ns, &ctrl->namespaces, list) { | |
861 | if (ns->ns_id == nsid) | |
862 | return ns; | |
863 | if (ns->ns_id > nsid) | |
864 | break; | |
865 | } | |
866 | return NULL; | |
867 | } | |
868 | ||
869 | static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid) | |
870 | { | |
871 | struct nvme_ns *ns; | |
872 | struct gendisk *disk; | |
873 | int node = dev_to_node(ctrl->dev); | |
874 | ||
875 | ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node); | |
876 | if (!ns) | |
877 | return; | |
878 | ||
879 | ns->queue = blk_mq_init_queue(ctrl->tagset); | |
880 | if (IS_ERR(ns->queue)) | |
881 | goto out_free_ns; | |
882 | queue_flag_set_unlocked(QUEUE_FLAG_NOMERGES, ns->queue); | |
883 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue); | |
884 | ns->queue->queuedata = ns; | |
885 | ns->ctrl = ctrl; | |
886 | ||
887 | disk = alloc_disk_node(0, node); | |
888 | if (!disk) | |
889 | goto out_free_queue; | |
890 | ||
891 | kref_init(&ns->kref); | |
892 | ns->ns_id = nsid; | |
893 | ns->disk = disk; | |
894 | ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */ | |
895 | list_add_tail(&ns->list, &ctrl->namespaces); | |
896 | ||
897 | blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift); | |
898 | if (ctrl->max_hw_sectors) { | |
899 | blk_queue_max_hw_sectors(ns->queue, ctrl->max_hw_sectors); | |
900 | blk_queue_max_segments(ns->queue, | |
901 | (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1); | |
902 | } | |
903 | if (ctrl->stripe_size) | |
904 | blk_queue_chunk_sectors(ns->queue, ctrl->stripe_size >> 9); | |
905 | if (ctrl->vwc & NVME_CTRL_VWC_PRESENT) | |
906 | blk_queue_flush(ns->queue, REQ_FLUSH | REQ_FUA); | |
907 | blk_queue_virt_boundary(ns->queue, ctrl->page_size - 1); | |
908 | ||
909 | disk->major = nvme_major; | |
910 | disk->first_minor = 0; | |
911 | disk->fops = &nvme_fops; | |
912 | disk->private_data = ns; | |
913 | disk->queue = ns->queue; | |
914 | disk->driverfs_dev = ctrl->device; | |
915 | disk->flags = GENHD_FL_EXT_DEVT; | |
916 | sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, nsid); | |
917 | ||
918 | /* | |
919 | * Initialize capacity to 0 until we establish the namespace format and | |
920 | * setup integrity extentions if necessary. The revalidate_disk after | |
921 | * add_disk allows the driver to register with integrity if the format | |
922 | * requires it. | |
923 | */ | |
924 | set_capacity(disk, 0); | |
925 | if (nvme_revalidate_disk(ns->disk)) | |
926 | goto out_free_disk; | |
927 | ||
928 | kref_get(&ctrl->kref); | |
929 | if (ns->type != NVME_NS_LIGHTNVM) { | |
930 | add_disk(ns->disk); | |
931 | if (ns->ms) { | |
932 | struct block_device *bd = bdget_disk(ns->disk, 0); | |
933 | if (!bd) | |
934 | return; | |
935 | if (blkdev_get(bd, FMODE_READ, NULL)) { | |
936 | bdput(bd); | |
937 | return; | |
938 | } | |
939 | blkdev_reread_part(bd); | |
940 | blkdev_put(bd, FMODE_READ); | |
941 | } | |
942 | } | |
943 | ||
944 | return; | |
945 | out_free_disk: | |
946 | kfree(disk); | |
947 | list_del(&ns->list); | |
948 | out_free_queue: | |
949 | blk_cleanup_queue(ns->queue); | |
950 | out_free_ns: | |
951 | kfree(ns); | |
952 | } | |
953 | ||
954 | static void nvme_ns_remove(struct nvme_ns *ns) | |
955 | { | |
956 | bool kill = nvme_io_incapable(ns->ctrl) && | |
957 | !blk_queue_dying(ns->queue); | |
958 | ||
959 | if (kill) | |
960 | blk_set_queue_dying(ns->queue); | |
961 | if (ns->disk->flags & GENHD_FL_UP) { | |
962 | if (blk_get_integrity(ns->disk)) | |
963 | blk_integrity_unregister(ns->disk); | |
964 | del_gendisk(ns->disk); | |
965 | } | |
966 | if (kill || !blk_queue_dying(ns->queue)) { | |
967 | blk_mq_abort_requeue_list(ns->queue); | |
968 | blk_cleanup_queue(ns->queue); | |
969 | } | |
970 | list_del_init(&ns->list); | |
971 | nvme_put_ns(ns); | |
972 | } | |
973 | ||
974 | static void __nvme_scan_namespaces(struct nvme_ctrl *ctrl, unsigned nn) | |
975 | { | |
976 | struct nvme_ns *ns, *next; | |
977 | unsigned i; | |
978 | ||
979 | for (i = 1; i <= nn; i++) { | |
980 | ns = nvme_find_ns(ctrl, i); | |
981 | if (ns) { | |
982 | if (revalidate_disk(ns->disk)) | |
983 | nvme_ns_remove(ns); | |
984 | } else | |
985 | nvme_alloc_ns(ctrl, i); | |
986 | } | |
987 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) { | |
988 | if (ns->ns_id > nn) | |
989 | nvme_ns_remove(ns); | |
990 | } | |
991 | list_sort(NULL, &ctrl->namespaces, ns_cmp); | |
992 | } | |
993 | ||
994 | void nvme_scan_namespaces(struct nvme_ctrl *ctrl) | |
995 | { | |
996 | struct nvme_id_ctrl *id; | |
997 | ||
998 | if (nvme_identify_ctrl(ctrl, &id)) | |
999 | return; | |
1000 | __nvme_scan_namespaces(ctrl, le32_to_cpup(&id->nn)); | |
1001 | kfree(id); | |
1002 | } | |
1003 | ||
1004 | void nvme_remove_namespaces(struct nvme_ctrl *ctrl) | |
1005 | { | |
1006 | struct nvme_ns *ns, *next; | |
1007 | ||
1008 | list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) | |
1009 | nvme_ns_remove(ns); | |
1010 | } | |
1011 | ||
1012 | int __init nvme_core_init(void) | |
1013 | { | |
1014 | int result; | |
1015 | ||
1016 | result = register_blkdev(nvme_major, "nvme"); | |
1017 | if (result < 0) | |
1018 | return result; | |
1019 | else if (result > 0) | |
1020 | nvme_major = result; | |
1021 | ||
1022 | return 0; | |
1023 | } | |
1024 | ||
1025 | void nvme_core_exit(void) | |
1026 | { | |
1027 | unregister_blkdev(nvme_major, "nvme"); | |
1028 | } |