1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2011-2014, Intel Corporation.
4 * Copyright (c) 2017-2021 Christoph Hellwig.
5 */
6 #include <linux/ptrace.h> /* for force_successful_syscall_return */
7 #include <linux/nvme_ioctl.h>
8 #include "nvme.h"
9
10 /*
11 * Convert integer values from ioctl structures to user pointers, silently
12 * ignoring the upper bits in the compat case to match behaviour of 32-bit
13 * kernels.
14 */
nvme_to_user_ptr(uintptr_t ptrval)15 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
16 {
17 if (in_compat_syscall())
18 ptrval = (compat_uptr_t)ptrval;
19 return (void __user *)ptrval;
20 }
21
nvme_add_user_metadata(struct bio * bio,void __user * ubuf,unsigned len,u32 seed,bool write)22 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
23 unsigned len, u32 seed, bool write)
24 {
25 struct bio_integrity_payload *bip;
26 int ret = -ENOMEM;
27 void *buf;
28
29 buf = kmalloc(len, GFP_KERNEL);
30 if (!buf)
31 goto out;
32
33 ret = -EFAULT;
34 if (write && copy_from_user(buf, ubuf, len))
35 goto out_free_meta;
36
37 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
38 if (IS_ERR(bip)) {
39 ret = PTR_ERR(bip);
40 goto out_free_meta;
41 }
42
43 bip->bip_iter.bi_size = len;
44 bip->bip_iter.bi_sector = seed;
45 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
46 offset_in_page(buf));
47 if (ret == len)
48 return buf;
49 ret = -ENOMEM;
50 out_free_meta:
51 kfree(buf);
52 out:
53 return ERR_PTR(ret);
54 }
55
nvme_submit_user_cmd(struct request_queue * q,struct nvme_command * cmd,void __user * ubuffer,unsigned bufflen,void __user * meta_buffer,unsigned meta_len,u32 meta_seed,u64 * result,unsigned timeout)56 static int nvme_submit_user_cmd(struct request_queue *q,
57 struct nvme_command *cmd, void __user *ubuffer,
58 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
59 u32 meta_seed, u64 *result, unsigned timeout)
60 {
61 bool write = nvme_is_write(cmd);
62 struct nvme_ns *ns = q->queuedata;
63 struct block_device *bdev = ns ? ns->disk->part0 : NULL;
64 struct request *req;
65 struct bio *bio = NULL;
66 void *meta = NULL;
67 int ret;
68
69 req = nvme_alloc_request(q, cmd, 0);
70 if (IS_ERR(req))
71 return PTR_ERR(req);
72
73 if (timeout)
74 req->timeout = timeout;
75 nvme_req(req)->flags |= NVME_REQ_USERCMD;
76
77 if (ubuffer && bufflen) {
78 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
79 GFP_KERNEL);
80 if (ret)
81 goto out;
82 bio = req->bio;
83 if (bdev)
84 bio_set_dev(bio, bdev);
85 if (bdev && meta_buffer && meta_len) {
86 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
87 meta_seed, write);
88 if (IS_ERR(meta)) {
89 ret = PTR_ERR(meta);
90 goto out_unmap;
91 }
92 req->cmd_flags |= REQ_INTEGRITY;
93 }
94 }
95
96 ret = nvme_execute_passthru_rq(req);
97 if (result)
98 *result = le64_to_cpu(nvme_req(req)->result.u64);
99 if (meta && !ret && !write) {
100 if (copy_to_user(meta_buffer, meta, meta_len))
101 ret = -EFAULT;
102 }
103 kfree(meta);
104 out_unmap:
105 if (bio)
106 blk_rq_unmap_user(bio);
107 out:
108 blk_mq_free_request(req);
109 return ret;
110 }
111
112
nvme_submit_io(struct nvme_ns * ns,struct nvme_user_io __user * uio)113 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
114 {
115 struct nvme_user_io io;
116 struct nvme_command c;
117 unsigned length, meta_len;
118 void __user *metadata;
119
120 if (copy_from_user(&io, uio, sizeof(io)))
121 return -EFAULT;
122 if (io.flags)
123 return -EINVAL;
124
125 switch (io.opcode) {
126 case nvme_cmd_write:
127 case nvme_cmd_read:
128 case nvme_cmd_compare:
129 break;
130 default:
131 return -EINVAL;
132 }
133
134 length = (io.nblocks + 1) << ns->lba_shift;
135
136 if ((io.control & NVME_RW_PRINFO_PRACT) &&
137 ns->ms == sizeof(struct t10_pi_tuple)) {
138 /*
139 * Protection information is stripped/inserted by the
140 * controller.
141 */
142 if (nvme_to_user_ptr(io.metadata))
143 return -EINVAL;
144 meta_len = 0;
145 metadata = NULL;
146 } else {
147 meta_len = (io.nblocks + 1) * ns->ms;
148 metadata = nvme_to_user_ptr(io.metadata);
149 }
150
151 if (ns->features & NVME_NS_EXT_LBAS) {
152 length += meta_len;
153 meta_len = 0;
154 } else if (meta_len) {
155 if ((io.metadata & 3) || !io.metadata)
156 return -EINVAL;
157 }
158
159 memset(&c, 0, sizeof(c));
160 c.rw.opcode = io.opcode;
161 c.rw.flags = io.flags;
162 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
163 c.rw.slba = cpu_to_le64(io.slba);
164 c.rw.length = cpu_to_le16(io.nblocks);
165 c.rw.control = cpu_to_le16(io.control);
166 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
167 c.rw.reftag = cpu_to_le32(io.reftag);
168 c.rw.apptag = cpu_to_le16(io.apptag);
169 c.rw.appmask = cpu_to_le16(io.appmask);
170
171 return nvme_submit_user_cmd(ns->queue, &c,
172 nvme_to_user_ptr(io.addr), length,
173 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
174 }
175
nvme_validate_passthru_nsid(struct nvme_ctrl * ctrl,struct nvme_ns * ns,__u32 nsid)176 static bool nvme_validate_passthru_nsid(struct nvme_ctrl *ctrl,
177 struct nvme_ns *ns, __u32 nsid)
178 {
179 if (ns && nsid != ns->head->ns_id) {
180 dev_err(ctrl->device,
181 "%s: nsid (%u) in cmd does not match nsid (%u)"
182 "of namespace\n",
183 current->comm, nsid, ns->head->ns_id);
184 return false;
185 }
186
187 return true;
188 }
189
nvme_user_cmd(struct nvme_ctrl * ctrl,struct nvme_ns * ns,struct nvme_passthru_cmd __user * ucmd)190 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
191 struct nvme_passthru_cmd __user *ucmd)
192 {
193 struct nvme_passthru_cmd cmd;
194 struct nvme_command c;
195 unsigned timeout = 0;
196 u64 result;
197 int status;
198
199 if (!capable(CAP_SYS_ADMIN))
200 return -EACCES;
201 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
202 return -EFAULT;
203 if (cmd.flags)
204 return -EINVAL;
205 if (!nvme_validate_passthru_nsid(ctrl, ns, cmd.nsid))
206 return -EINVAL;
207
208 memset(&c, 0, sizeof(c));
209 c.common.opcode = cmd.opcode;
210 c.common.flags = cmd.flags;
211 c.common.nsid = cpu_to_le32(cmd.nsid);
212 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
213 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
214 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
215 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
216 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
217 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
218 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
219 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
220
221 if (cmd.timeout_ms)
222 timeout = msecs_to_jiffies(cmd.timeout_ms);
223
224 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
225 nvme_to_user_ptr(cmd.addr), cmd.data_len,
226 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
227 0, &result, timeout);
228
229 if (status >= 0) {
230 if (put_user(result, &ucmd->result))
231 return -EFAULT;
232 }
233
234 return status;
235 }
236
nvme_user_cmd64(struct nvme_ctrl * ctrl,struct nvme_ns * ns,struct nvme_passthru_cmd64 __user * ucmd)237 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
238 struct nvme_passthru_cmd64 __user *ucmd)
239 {
240 struct nvme_passthru_cmd64 cmd;
241 struct nvme_command c;
242 unsigned timeout = 0;
243 int status;
244
245 if (!capable(CAP_SYS_ADMIN))
246 return -EACCES;
247 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
248 return -EFAULT;
249 if (cmd.flags)
250 return -EINVAL;
251 if (!nvme_validate_passthru_nsid(ctrl, ns, cmd.nsid))
252 return -EINVAL;
253
254 memset(&c, 0, sizeof(c));
255 c.common.opcode = cmd.opcode;
256 c.common.flags = cmd.flags;
257 c.common.nsid = cpu_to_le32(cmd.nsid);
258 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
259 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
260 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
261 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
262 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
263 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
264 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
265 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
266
267 if (cmd.timeout_ms)
268 timeout = msecs_to_jiffies(cmd.timeout_ms);
269
270 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
271 nvme_to_user_ptr(cmd.addr), cmd.data_len,
272 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
273 0, &cmd.result, timeout);
274
275 if (status >= 0) {
276 if (put_user(cmd.result, &ucmd->result))
277 return -EFAULT;
278 }
279
280 return status;
281 }
282
is_ctrl_ioctl(unsigned int cmd)283 static bool is_ctrl_ioctl(unsigned int cmd)
284 {
285 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
286 return true;
287 if (is_sed_ioctl(cmd))
288 return true;
289 return false;
290 }
291
nvme_ctrl_ioctl(struct nvme_ctrl * ctrl,unsigned int cmd,void __user * argp)292 static int nvme_ctrl_ioctl(struct nvme_ctrl *ctrl, unsigned int cmd,
293 void __user *argp)
294 {
295 switch (cmd) {
296 case NVME_IOCTL_ADMIN_CMD:
297 return nvme_user_cmd(ctrl, NULL, argp);
298 case NVME_IOCTL_ADMIN64_CMD:
299 return nvme_user_cmd64(ctrl, NULL, argp);
300 default:
301 return sed_ioctl(ctrl->opal_dev, cmd, argp);
302 }
303 }
304
305 #ifdef COMPAT_FOR_U64_ALIGNMENT
306 struct nvme_user_io32 {
307 __u8 opcode;
308 __u8 flags;
309 __u16 control;
310 __u16 nblocks;
311 __u16 rsvd;
312 __u64 metadata;
313 __u64 addr;
314 __u64 slba;
315 __u32 dsmgmt;
316 __u32 reftag;
317 __u16 apptag;
318 __u16 appmask;
319 } __attribute__((__packed__));
320 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
321 #endif /* COMPAT_FOR_U64_ALIGNMENT */
322
nvme_ns_ioctl(struct nvme_ns * ns,unsigned int cmd,void __user * argp)323 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned int cmd,
324 void __user *argp)
325 {
326 switch (cmd) {
327 case NVME_IOCTL_ID:
328 force_successful_syscall_return();
329 return ns->head->ns_id;
330 case NVME_IOCTL_IO_CMD:
331 return nvme_user_cmd(ns->ctrl, ns, argp);
332 /*
333 * struct nvme_user_io can have different padding on some 32-bit ABIs.
334 * Just accept the compat version as all fields that are used are the
335 * same size and at the same offset.
336 */
337 #ifdef COMPAT_FOR_U64_ALIGNMENT
338 case NVME_IOCTL_SUBMIT_IO32:
339 #endif
340 case NVME_IOCTL_SUBMIT_IO:
341 return nvme_submit_io(ns, argp);
342 case NVME_IOCTL_IO64_CMD:
343 return nvme_user_cmd64(ns->ctrl, ns, argp);
344 default:
345 return -ENOTTY;
346 }
347 }
348
__nvme_ioctl(struct nvme_ns * ns,unsigned int cmd,void __user * arg)349 static int __nvme_ioctl(struct nvme_ns *ns, unsigned int cmd, void __user *arg)
350 {
351 if (is_ctrl_ioctl(cmd))
352 return nvme_ctrl_ioctl(ns->ctrl, cmd, arg);
353 return nvme_ns_ioctl(ns, cmd, arg);
354 }
355
nvme_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)356 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
357 unsigned int cmd, unsigned long arg)
358 {
359 struct nvme_ns *ns = bdev->bd_disk->private_data;
360
361 return __nvme_ioctl(ns, cmd, (void __user *)arg);
362 }
363
nvme_ns_chr_ioctl(struct file * file,unsigned int cmd,unsigned long arg)364 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
365 {
366 struct nvme_ns *ns =
367 container_of(file_inode(file)->i_cdev, struct nvme_ns, cdev);
368
369 return __nvme_ioctl(ns, cmd, (void __user *)arg);
370 }
371
372 #ifdef CONFIG_NVME_MULTIPATH
nvme_ns_head_ctrl_ioctl(struct nvme_ns * ns,unsigned int cmd,void __user * argp,struct nvme_ns_head * head,int srcu_idx)373 static int nvme_ns_head_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
374 void __user *argp, struct nvme_ns_head *head, int srcu_idx)
375 __releases(&head->srcu)
376 {
377 struct nvme_ctrl *ctrl = ns->ctrl;
378 int ret;
379
380 nvme_get_ctrl(ns->ctrl);
381 srcu_read_unlock(&head->srcu, srcu_idx);
382 ret = nvme_ctrl_ioctl(ns->ctrl, cmd, argp);
383
384 nvme_put_ctrl(ctrl);
385 return ret;
386 }
387
nvme_ns_head_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)388 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
389 unsigned int cmd, unsigned long arg)
390 {
391 struct nvme_ns_head *head = bdev->bd_disk->private_data;
392 void __user *argp = (void __user *)arg;
393 struct nvme_ns *ns;
394 int srcu_idx, ret = -EWOULDBLOCK;
395
396 srcu_idx = srcu_read_lock(&head->srcu);
397 ns = nvme_find_path(head);
398 if (!ns)
399 goto out_unlock;
400
401 /*
402 * Handle ioctls that apply to the controller instead of the namespace
403 * seperately and drop the ns SRCU reference early. This avoids a
404 * deadlock when deleting namespaces using the passthrough interface.
405 */
406 if (is_ctrl_ioctl(cmd))
407 return nvme_ns_head_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
408
409 ret = nvme_ns_ioctl(ns, cmd, argp);
410 out_unlock:
411 srcu_read_unlock(&head->srcu, srcu_idx);
412 return ret;
413 }
414
nvme_ns_head_chr_ioctl(struct file * file,unsigned int cmd,unsigned long arg)415 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
416 unsigned long arg)
417 {
418 struct cdev *cdev = file_inode(file)->i_cdev;
419 struct nvme_ns_head *head =
420 container_of(cdev, struct nvme_ns_head, cdev);
421 void __user *argp = (void __user *)arg;
422 struct nvme_ns *ns;
423 int srcu_idx, ret = -EWOULDBLOCK;
424
425 srcu_idx = srcu_read_lock(&head->srcu);
426 ns = nvme_find_path(head);
427 if (!ns)
428 goto out_unlock;
429
430 if (is_ctrl_ioctl(cmd))
431 return nvme_ns_head_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
432
433 ret = nvme_ns_ioctl(ns, cmd, argp);
434 out_unlock:
435 srcu_read_unlock(&head->srcu, srcu_idx);
436 return ret;
437 }
438 #endif /* CONFIG_NVME_MULTIPATH */
439
nvme_dev_user_cmd(struct nvme_ctrl * ctrl,void __user * argp)440 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
441 {
442 struct nvme_ns *ns;
443 int ret;
444
445 down_read(&ctrl->namespaces_rwsem);
446 if (list_empty(&ctrl->namespaces)) {
447 ret = -ENOTTY;
448 goto out_unlock;
449 }
450
451 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
452 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
453 dev_warn(ctrl->device,
454 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
455 ret = -EINVAL;
456 goto out_unlock;
457 }
458
459 dev_warn(ctrl->device,
460 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
461 kref_get(&ns->kref);
462 up_read(&ctrl->namespaces_rwsem);
463
464 ret = nvme_user_cmd(ctrl, ns, argp);
465 nvme_put_ns(ns);
466 return ret;
467
468 out_unlock:
469 up_read(&ctrl->namespaces_rwsem);
470 return ret;
471 }
472
nvme_dev_ioctl(struct file * file,unsigned int cmd,unsigned long arg)473 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
474 unsigned long arg)
475 {
476 struct nvme_ctrl *ctrl = file->private_data;
477 void __user *argp = (void __user *)arg;
478
479 switch (cmd) {
480 case NVME_IOCTL_ADMIN_CMD:
481 return nvme_user_cmd(ctrl, NULL, argp);
482 case NVME_IOCTL_ADMIN64_CMD:
483 return nvme_user_cmd64(ctrl, NULL, argp);
484 case NVME_IOCTL_IO_CMD:
485 return nvme_dev_user_cmd(ctrl, argp);
486 case NVME_IOCTL_RESET:
487 if (!capable(CAP_SYS_ADMIN))
488 return -EACCES;
489 dev_warn(ctrl->device, "resetting controller\n");
490 return nvme_reset_ctrl_sync(ctrl);
491 case NVME_IOCTL_SUBSYS_RESET:
492 if (!capable(CAP_SYS_ADMIN))
493 return -EACCES;
494 return nvme_reset_subsystem(ctrl);
495 case NVME_IOCTL_RESCAN:
496 if (!capable(CAP_SYS_ADMIN))
497 return -EACCES;
498 nvme_queue_scan(ctrl);
499 return 0;
500 default:
501 return -ENOTTY;
502 }
503 }
504