1 /* $NetBSD: xenbus_dev.c,v 1.19 2024/02/09 22:08:34 andvar Exp $ */
2 /*
3 * xenbus_dev.c
4 *
5 * Driver giving user-space access to the kernel's xenbus connection
6 * to xenstore.
7 *
8 * Copyright (c) 2005, Christian Limpach
9 * Copyright (c) 2005, Rusty Russell, IBM Corporation
10 *
11 * This file may be distributed separately from the Linux kernel, or
12 * incorporated into other software packages, subject to the following license:
13 *
14 * Permission is hereby granted, free of charge, to any person obtaining a copy
15 * of this source file (the "Software"), to deal in the Software without
16 * restriction, including without limitation the rights to use, copy, modify,
17 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
18 * and to permit persons to whom the Software is furnished to do so, subject to
19 * the following conditions:
20 *
21 * The above copyright notice and this permission notice shall be included in
22 * all copies or substantial portions of the Software.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
25 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
26 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
27 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
28 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
29 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
30 * IN THE SOFTWARE.
31 */
32
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: xenbus_dev.c,v 1.19 2024/02/09 22:08:34 andvar Exp $");
35
36 #include "opt_xen.h"
37
38 #include <sys/types.h>
39 #include <sys/null.h>
40 #include <sys/errno.h>
41 #include <sys/param.h>
42 #include <sys/proc.h>
43 #include <sys/systm.h>
44 #include <sys/dirent.h>
45 #include <sys/stat.h>
46 #include <sys/tree.h>
47 #include <sys/vnode.h>
48 #include <miscfs/specfs/specdev.h>
49 #include <miscfs/kernfs/kernfs.h>
50
51 #include <xen/kernfs_machdep.h>
52
53 #include <xen/intr.h>
54 #include <xen/hypervisor.h>
55 #include <xen/xenbus.h>
56 #include "xenbus_comms.h"
57
58 static int xenbus_dev_read(void *);
59 static int xenbus_dev_write(void *);
60 static int xenbus_dev_open(void *);
61 static int xenbus_dev_close(void *);
62 static int xsd_port_read(void *);
63
64 #define DIR_MODE (S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
65 #define PRIVCMD_MODE (S_IRUSR | S_IWUSR)
66 static const struct kernfs_fileop xenbus_fileops[] = {
67 { .kf_fileop = KERNFS_FILEOP_OPEN, .kf_vop = xenbus_dev_open },
68 { .kf_fileop = KERNFS_FILEOP_CLOSE, .kf_vop = xenbus_dev_close },
69 { .kf_fileop = KERNFS_FILEOP_READ, .kf_vop = xenbus_dev_read },
70 { .kf_fileop = KERNFS_FILEOP_WRITE, .kf_vop = xenbus_dev_write },
71 };
72
73 #define XSD_MODE (S_IRUSR)
74 static const struct kernfs_fileop xsd_port_fileops[] = {
75 { .kf_fileop = KERNFS_FILEOP_READ, .kf_vop = xsd_port_read },
76 };
77
78 static kmutex_t xenbus_dev_open_mtx;
79
80 void
xenbus_kernfs_init(void)81 xenbus_kernfs_init(void)
82 {
83 kernfs_entry_t *dkt;
84 kfstype kfst;
85
86 kfst = KERNFS_ALLOCTYPE(xenbus_fileops);
87 KERNFS_ALLOCENTRY(dkt, KM_SLEEP);
88 KERNFS_INITENTRY(dkt, DT_REG, "xenbus", NULL, kfst, VREG,
89 PRIVCMD_MODE);
90 kernfs_addentry(kernxen_pkt, dkt);
91
92 if (xendomain_is_dom0()) {
93 kfst = KERNFS_ALLOCTYPE(xsd_port_fileops);
94 KERNFS_ALLOCENTRY(dkt, KM_SLEEP);
95 KERNFS_INITENTRY(dkt, DT_REG, "xsd_port", NULL,
96 kfst, VREG, XSD_MODE);
97 kernfs_addentry(kernxen_pkt, dkt);
98 }
99 mutex_init(&xenbus_dev_open_mtx, MUTEX_DEFAULT, IPL_NONE);
100 }
101
102 /*
103 * several process may open /kern/xen/xenbus in parallel.
104 * In a transaction one or more write is followed by one or more read.
105 * Unfortunately we don't get a file descriptor identifier down there,
106 * which we could use to link a read() to a transaction started in a write().
107 * To work around this we keep a list of lwp that opended the xenbus file.
108 * This assumes that a single lwp won't open /kern/xen/xenbus more
109 * than once, and that a transaction started by one lwp won't be ended
110 * by another.
111 * because of this, we also assume that we always got the data before
112 * the read() syscall.
113 */
114
115 struct xenbus_dev_transaction {
116 SLIST_ENTRY(xenbus_dev_transaction) trans_next;
117 struct xenbus_transaction *handle;
118 };
119
120 struct xenbus_dev_lwp {
121 SLIST_ENTRY(xenbus_dev_lwp) lwp_next;
122 SLIST_HEAD(, xenbus_dev_transaction) transactions;
123 lwp_t *lwp;
124 /* Response queue. */
125 #define BUFFER_SIZE (PAGE_SIZE)
126 #define MASK_READ_IDX(idx) ((idx)&(BUFFER_SIZE-1))
127 char read_buffer[BUFFER_SIZE];
128 unsigned int read_cons, read_prod;
129 /* Partial request. */
130 unsigned int len;
131 union {
132 struct xsd_sockmsg msg;
133 char buffer[BUFFER_SIZE];
134 } u;
135 kmutex_t mtx;
136 };
137
138 struct xenbus_dev_data {
139 /* lwps which opended this device */
140 SLIST_HEAD(, xenbus_dev_lwp) lwps;
141 kmutex_t mtx;
142 };
143
144
145 static int
xenbus_dev_read(void * v)146 xenbus_dev_read(void *v)
147 {
148 struct vop_read_args /* {
149 struct vnode *a_vp;
150 struct uio *a_uio;
151 int a_ioflag;
152 struct ucred *a_cred;
153 } */ *ap = v;
154 struct kernfs_node *kfs = VTOKERN(ap->a_vp);
155 struct uio *uio = ap->a_uio;
156 struct xenbus_dev_data *u;
157 struct xenbus_dev_lwp *xlwp;
158 int err;
159 off_t offset;
160
161 mutex_enter(&xenbus_dev_open_mtx);
162 u = kfs->kfs_v;
163 if (u == NULL) {
164 mutex_exit(&xenbus_dev_open_mtx);
165 return EBADF;
166 }
167 mutex_enter(&u->mtx);
168 mutex_exit(&xenbus_dev_open_mtx);
169 SLIST_FOREACH(xlwp, &u->lwps, lwp_next) {
170 if (xlwp->lwp == curlwp) {
171 break;
172 }
173 }
174 if (xlwp == NULL) {
175 mutex_exit(&u->mtx);
176 return EBADF;
177 }
178 mutex_enter(&xlwp->mtx);
179 mutex_exit(&u->mtx);
180
181 if (xlwp->read_prod == xlwp->read_cons) {
182 err = EWOULDBLOCK;
183 goto end;
184 }
185
186 offset = uio->uio_offset;
187 if (xlwp->read_cons > xlwp->read_prod) {
188 err = uiomove(
189 &xlwp->read_buffer[MASK_READ_IDX(xlwp->read_cons)],
190 0U - xlwp->read_cons, uio);
191 if (err)
192 goto end;
193 xlwp->read_cons += (uio->uio_offset - offset);
194 offset = uio->uio_offset;
195 }
196 err = uiomove(&xlwp->read_buffer[MASK_READ_IDX(xlwp->read_cons)],
197 xlwp->read_prod - xlwp->read_cons, uio);
198 if (err == 0)
199 xlwp->read_cons += (uio->uio_offset - offset);
200
201 end:
202 mutex_exit(&xlwp->mtx);
203 return err;
204 }
205
206 static void
queue_reply(struct xenbus_dev_lwp * xlwp,char * data,unsigned int len)207 queue_reply(struct xenbus_dev_lwp *xlwp,
208 char *data, unsigned int len)
209 {
210 int i;
211 KASSERT(mutex_owned(&xlwp->mtx));
212 for (i = 0; i < len; i++, xlwp->read_prod++)
213 xlwp->read_buffer[MASK_READ_IDX(xlwp->read_prod)] = data[i];
214
215 KASSERT((xlwp->read_prod - xlwp->read_cons) <= sizeof(xlwp->read_buffer));
216 }
217
218 static int
xenbus_dev_write(void * v)219 xenbus_dev_write(void *v)
220 {
221 struct vop_write_args /* {
222 struct vnode *a_vp;
223 struct uio *a_uio;
224 int a_ioflag;
225 struct ucred *a_cred;
226 } */ *ap = v;
227 struct kernfs_node *kfs = VTOKERN(ap->a_vp);
228 struct uio *uio = ap->a_uio;
229
230 struct xenbus_dev_data *u;
231 struct xenbus_dev_lwp *xlwp;
232 struct xenbus_dev_transaction *trans;
233 void *reply;
234 int err;
235 size_t size;
236
237 mutex_enter(&xenbus_dev_open_mtx);
238 u = kfs->kfs_v;
239 if (u == NULL) {
240 mutex_exit(&xenbus_dev_open_mtx);
241 return EBADF;
242 }
243 mutex_enter(&u->mtx);
244 mutex_exit(&xenbus_dev_open_mtx);
245 SLIST_FOREACH(xlwp, &u->lwps, lwp_next) {
246 if (xlwp->lwp == curlwp) {
247 break;
248 }
249 }
250 if (xlwp == NULL) {
251 mutex_exit(&u->mtx);
252 return EBADF;
253 }
254 mutex_enter(&xlwp->mtx);
255 mutex_exit(&u->mtx);
256
257 if (uio->uio_offset < 0) {
258 err = EINVAL;
259 goto end;
260 }
261 size = uio->uio_resid;
262
263 if ((size + xlwp->len) > sizeof(xlwp->u.buffer)) {
264 err = EINVAL;
265 goto end;
266 }
267
268 err = uiomove(xlwp->u.buffer + xlwp->len,
269 sizeof(xlwp->u.buffer) - xlwp->len, uio);
270 if (err)
271 goto end;
272
273 xlwp->len += size;
274 if (xlwp->len < (sizeof(xlwp->u.msg) + xlwp->u.msg.len))
275 goto end;
276
277 switch (xlwp->u.msg.type) {
278 case XS_TRANSACTION_START:
279 case XS_TRANSACTION_END:
280 case XS_DIRECTORY:
281 case XS_READ:
282 case XS_GET_PERMS:
283 case XS_RELEASE:
284 case XS_GET_DOMAIN_PATH:
285 case XS_WRITE:
286 case XS_MKDIR:
287 case XS_RM:
288 case XS_SET_PERMS:
289 err = xenbus_dev_request_and_reply(&xlwp->u.msg, &reply);
290 if (err == 0) {
291 if (xlwp->u.msg.type == XS_TRANSACTION_START) {
292 trans = kmem_alloc(sizeof(*trans), KM_SLEEP);
293 trans->handle = (struct xenbus_transaction *)
294 strtoul(reply, NULL, 0);
295 SLIST_INSERT_HEAD(&xlwp->transactions,
296 trans, trans_next);
297 } else if (xlwp->u.msg.type == XS_TRANSACTION_END) {
298 SLIST_FOREACH(trans, &xlwp->transactions,
299 trans_next) {
300 if ((unsigned long)trans->handle ==
301 (unsigned long)xlwp->u.msg.tx_id)
302 break;
303 }
304 if (trans == NULL) {
305 err = EINVAL;
306 goto end;
307 }
308 SLIST_REMOVE(&xlwp->transactions, trans,
309 xenbus_dev_transaction, trans_next);
310 kmem_free(trans, sizeof(*trans));
311 }
312 queue_reply(xlwp, (char *)&xlwp->u.msg,
313 sizeof(xlwp->u.msg));
314 queue_reply(xlwp, (char *)reply, xlwp->u.msg.len);
315
316 xenbus_dev_reply_free(&xlwp->u.msg, reply);
317 }
318 break;
319
320 default:
321 err = EINVAL;
322 break;
323 }
324
325 if (err == 0) {
326 xlwp->len = 0;
327 }
328 end:
329 mutex_exit(&xlwp->mtx);
330 return err;
331 }
332
333 static int
xenbus_dev_open(void * v)334 xenbus_dev_open(void *v)
335 {
336 struct vop_open_args /* {
337 struct vnode *a_vp;
338 int a_mode;
339 struct ucred *a_cred;
340 } */ *ap = v;
341 struct kernfs_node *kfs = VTOKERN(ap->a_vp);
342 struct xenbus_dev_data *u;
343 struct xenbus_dev_lwp *xlwp;
344
345 if (xen_start_info.store_evtchn == 0)
346 return ENOENT;
347
348 mutex_enter(&xenbus_dev_open_mtx);
349 u = kfs->kfs_v;
350 if (u == NULL) {
351 mutex_exit(&xenbus_dev_open_mtx);
352
353 u = kmem_zalloc(sizeof(*u), KM_SLEEP);
354 SLIST_INIT(&u->lwps);
355 mutex_init(&u->mtx, MUTEX_DEFAULT, IPL_NONE);
356
357 mutex_enter(&xenbus_dev_open_mtx);
358 /*
359 * Must re-check if filled while waiting in alloc
360 * by some other lwp.
361 */
362 if (kfs->kfs_v) {
363 kmem_free(u, sizeof(*u));
364 u = kfs->kfs_v;
365 } else {
366 kfs->kfs_v = u;
367 }
368 };
369 mutex_exit(&xenbus_dev_open_mtx);
370
371 mutex_enter(&u->mtx);
372 SLIST_FOREACH(xlwp, &u->lwps, lwp_next) {
373 if (xlwp->lwp == curlwp) {
374 break;
375 }
376 }
377 if (xlwp == NULL) {
378 mutex_exit(&u->mtx);
379
380 xlwp = kmem_zalloc(sizeof(*xlwp), KM_SLEEP);
381 xlwp->lwp = curlwp;
382 SLIST_INIT(&xlwp->transactions);
383 mutex_init(&xlwp->mtx, MUTEX_DEFAULT, IPL_NONE);
384
385 mutex_enter(&u->mtx);
386 /*
387 * While alloc can block, this can't be re-entered with
388 * curlwp, so no need to re-check. Also the node can't
389 * be closed while we are blocked here.
390 */
391 SLIST_INSERT_HEAD(&u->lwps, xlwp, lwp_next);
392 }
393 mutex_exit(&u->mtx);
394
395 return 0;
396 }
397
398 static int
xenbus_dev_close(void * v)399 xenbus_dev_close(void *v)
400 {
401 struct vop_close_args /* {
402 struct vnode *a_vp;
403 int a_fflag;
404 struct ucred *a_cred;
405 } */ *ap = v;
406 struct kernfs_node *kfs = VTOKERN(ap->a_vp);
407
408 struct xenbus_dev_data *u;
409 struct xenbus_dev_lwp *xlwp;
410 struct xenbus_dev_transaction *trans;
411
412 mutex_enter(&xenbus_dev_open_mtx);
413 u = kfs->kfs_v;
414 KASSERT(u != NULL);
415 mutex_enter(&u->mtx);
416 SLIST_FOREACH(xlwp, &u->lwps, lwp_next) {
417 if (xlwp->lwp == curlwp) {
418 break;
419 }
420 }
421 if (xlwp == NULL) {
422 mutex_exit(&u->mtx);
423 mutex_exit(&xenbus_dev_open_mtx);
424 return EBADF;
425 }
426 mutex_enter(&xlwp->mtx);
427 while (!SLIST_EMPTY(&xlwp->transactions)) {
428 trans = SLIST_FIRST(&xlwp->transactions);
429 xenbus_transaction_end(trans->handle, 1);
430 SLIST_REMOVE_HEAD(&xlwp->transactions, trans_next);
431 kmem_free(trans, sizeof(*trans));
432 }
433 mutex_exit(&xlwp->mtx);
434 SLIST_REMOVE(&u->lwps, xlwp, xenbus_dev_lwp, lwp_next);
435 mutex_destroy(&xlwp->mtx);
436
437 if (!SLIST_EMPTY(&u->lwps)) {
438 mutex_exit(&u->mtx);
439 mutex_exit(&xenbus_dev_open_mtx);
440 return 0;
441 }
442 mutex_exit(&u->mtx);
443 mutex_destroy(&u->mtx);
444 kfs->kfs_v = NULL;
445 mutex_exit(&xenbus_dev_open_mtx);
446 kmem_free(xlwp, sizeof(*xlwp));
447 kmem_free(u, sizeof(*u));
448 return 0;
449 }
450
451 #define LD_STRLEN 21 /* a 64bit integer needs 20 digits in base10 */
452
453 static int
xsd_port_read(void * v)454 xsd_port_read(void *v)
455 {
456 struct vop_read_args /* {
457 struct vnode *a_vp;
458 struct uio *a_uio;
459 int a_ioflag;
460 struct ucred *a_cred;
461 } */ *ap = v;
462 struct uio *uio = ap->a_uio;
463 int off, error;
464 size_t len;
465 char strbuf[LD_STRLEN], *bf;
466
467 off = (int)uio->uio_offset;
468 if (off < 0)
469 return EINVAL;
470
471 len = snprintf(strbuf, sizeof(strbuf), "%ld\n",
472 (long)xen_start_info.store_evtchn);
473 if (off >= len) {
474 bf = strbuf;
475 len = 0;
476 } else {
477 bf = &strbuf[off];
478 len -= off;
479 }
480 error = uiomove(bf, len, uio);
481 return error;
482 }
483
484 /*
485 * Local variables:
486 * c-file-style: "linux"
487 * indent-tabs-mode: t
488 * c-indent-level: 8
489 * c-basic-offset: 8
490 * tab-width: 8
491 * End:
492 */
493