devfs - Change setattr code to use vop_helper fns
[dragonfly.git] / sys / vfs / devfs / devfs_vnops.c
CommitLineData
21864bc5 1/*
9f889dc4
MD
2 * (MPSAFE)
3 *
21864bc5
MD
4 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
5 *
6 * This code is derived from software contributed to The DragonFly Project
7 * by Alex Hornung <ahornung@gmail.com>
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 *
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
18 * distribution.
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36#include <sys/param.h>
37#include <sys/systm.h>
38#include <sys/time.h>
39#include <sys/kernel.h>
40#include <sys/lock.h>
41#include <sys/fcntl.h>
42#include <sys/proc.h>
43#include <sys/priv.h>
44#include <sys/signalvar.h>
45#include <sys/vnode.h>
46#include <sys/uio.h>
47#include <sys/mount.h>
48#include <sys/file.h>
49#include <sys/fcntl.h>
50#include <sys/namei.h>
51#include <sys/dirent.h>
52#include <sys/malloc.h>
53#include <sys/stat.h>
54#include <sys/reg.h>
21864bc5
MD
55#include <vm/vm_pager.h>
56#include <vm/vm_zone.h>
57#include <vm/vm_object.h>
58#include <sys/filio.h>
59#include <sys/ttycom.h>
21864bc5 60#include <sys/tty.h>
2d076755 61#include <sys/diskslice.h>
3a1032a6 62#include <sys/sysctl.h>
2c1e28dd 63#include <sys/devfs.h>
21864bc5 64#include <sys/pioctl.h>
c705e298 65#include <vfs/fifofs/fifo.h>
21864bc5
MD
66
67#include <machine/limits.h>
684a93c4 68
1a54183b
MD
69#include <sys/buf2.h>
70#include <sys/sysref2.h>
684a93c4
MD
71#include <sys/mplock2.h>
72#include <vm/vm_page2.h>
21864bc5
MD
73
74MALLOC_DECLARE(M_DEVFS);
9f889dc4
MD
75#define DEVFS_BADOP (void *)devfs_vop_badop
76
77static int devfs_vop_badop(struct vop_generic_args *);
78static int devfs_vop_access(struct vop_access_args *);
79static int devfs_vop_inactive(struct vop_inactive_args *);
80static int devfs_vop_reclaim(struct vop_reclaim_args *);
81static int devfs_vop_readdir(struct vop_readdir_args *);
82static int devfs_vop_getattr(struct vop_getattr_args *);
83static int devfs_vop_setattr(struct vop_setattr_args *);
84static int devfs_vop_readlink(struct vop_readlink_args *);
85static int devfs_vop_print(struct vop_print_args *);
86
87static int devfs_vop_nresolve(struct vop_nresolve_args *);
88static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
89static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
90static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
91static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
92static int devfs_vop_nremove(struct vop_nremove_args *);
21864bc5
MD
93
94static int devfs_spec_open(struct vop_open_args *);
95static int devfs_spec_close(struct vop_close_args *);
96static int devfs_spec_fsync(struct vop_fsync_args *);
97
98static int devfs_spec_read(struct vop_read_args *);
99static int devfs_spec_write(struct vop_write_args *);
100static int devfs_spec_ioctl(struct vop_ioctl_args *);
21864bc5
MD
101static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
102static int devfs_spec_strategy(struct vop_strategy_args *);
103static void devfs_spec_strategy_done(struct bio *);
104static int devfs_spec_freeblks(struct vop_freeblks_args *);
105static int devfs_spec_bmap(struct vop_bmap_args *);
106static int devfs_spec_advlock(struct vop_advlock_args *);
107static void devfs_spec_getpages_iodone(struct bio *);
108static int devfs_spec_getpages(struct vop_getpages_args *);
109
9f889dc4
MD
110static int devfs_fo_close(struct file *);
111static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
112static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
113static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
114static int devfs_fo_kqfilter(struct file *, struct knote *);
115static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
87baaf0c 116 struct ucred *, struct sysmsg *);
21864bc5 117static __inline int sequential_heuristic(struct uio *, struct file *);
87baaf0c
MD
118
119extern struct lock devfs_lock;
21864bc5
MD
120
121/*
9f889dc4 122 * devfs vnode operations for regular files. All vnode ops are MPSAFE.
21864bc5
MD
123 */
124struct vop_ops devfs_vnode_norm_vops = {
125 .vop_default = vop_defaultop,
9f889dc4 126 .vop_access = devfs_vop_access,
21864bc5 127 .vop_advlock = DEVFS_BADOP,
9f889dc4 128 .vop_bmap = DEVFS_BADOP,
21864bc5 129 .vop_close = vop_stdclose,
9f889dc4
MD
130 .vop_getattr = devfs_vop_getattr,
131 .vop_inactive = devfs_vop_inactive,
21864bc5 132 .vop_ncreate = DEVFS_BADOP,
9f889dc4
MD
133 .vop_nresolve = devfs_vop_nresolve,
134 .vop_nlookupdotdot = devfs_vop_nlookupdotdot,
21864bc5 135 .vop_nlink = DEVFS_BADOP,
9f889dc4 136 .vop_nmkdir = devfs_vop_nmkdir,
21864bc5 137 .vop_nmknod = DEVFS_BADOP,
9f889dc4 138 .vop_nremove = devfs_vop_nremove,
21864bc5 139 .vop_nrename = DEVFS_BADOP,
9f889dc4
MD
140 .vop_nrmdir = devfs_vop_nrmdir,
141 .vop_nsymlink = devfs_vop_nsymlink,
142 .vop_open = vop_stdopen,
21864bc5 143 .vop_pathconf = vop_stdpathconf,
9f889dc4
MD
144 .vop_print = devfs_vop_print,
145 .vop_read = DEVFS_BADOP,
146 .vop_readdir = devfs_vop_readdir,
147 .vop_readlink = devfs_vop_readlink,
148 .vop_reclaim = devfs_vop_reclaim,
149 .vop_setattr = devfs_vop_setattr,
21864bc5
MD
150 .vop_write = DEVFS_BADOP,
151 .vop_ioctl = DEVFS_BADOP
152};
153
154/*
9f889dc4 155 * devfs vnode operations for character devices. All vnode ops are MPSAFE.
21864bc5
MD
156 */
157struct vop_ops devfs_vnode_dev_vops = {
158 .vop_default = vop_defaultop,
9f889dc4 159 .vop_access = devfs_vop_access,
21864bc5 160 .vop_advlock = devfs_spec_advlock,
9f889dc4 161 .vop_bmap = devfs_spec_bmap,
21864bc5
MD
162 .vop_close = devfs_spec_close,
163 .vop_freeblks = devfs_spec_freeblks,
164 .vop_fsync = devfs_spec_fsync,
9f889dc4 165 .vop_getattr = devfs_vop_getattr,
21864bc5 166 .vop_getpages = devfs_spec_getpages,
9f889dc4
MD
167 .vop_inactive = devfs_vop_inactive,
168 .vop_open = devfs_spec_open,
21864bc5 169 .vop_pathconf = vop_stdpathconf,
9f889dc4 170 .vop_print = devfs_vop_print,
21864bc5 171 .vop_kqfilter = devfs_spec_kqfilter,
9f889dc4 172 .vop_read = devfs_spec_read,
21864bc5
MD
173 .vop_readdir = DEVFS_BADOP,
174 .vop_readlink = DEVFS_BADOP,
9f889dc4
MD
175 .vop_reclaim = devfs_vop_reclaim,
176 .vop_setattr = devfs_vop_setattr,
21864bc5
MD
177 .vop_strategy = devfs_spec_strategy,
178 .vop_write = devfs_spec_write,
179 .vop_ioctl = devfs_spec_ioctl
180};
181
9f889dc4
MD
182/*
183 * devfs file pointer operations. All fileops are MPSAFE.
184 */
21864bc5
MD
185struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
186
187struct fileops devfs_dev_fileops = {
9f889dc4
MD
188 .fo_read = devfs_fo_read,
189 .fo_write = devfs_fo_write,
190 .fo_ioctl = devfs_fo_ioctl,
191 .fo_kqfilter = devfs_fo_kqfilter,
192 .fo_stat = devfs_fo_stat,
193 .fo_close = devfs_fo_close,
194 .fo_shutdown = nofo_shutdown
21864bc5
MD
195};
196
4062d050 197/*
9f889dc4
MD
198 * These two functions are possibly temporary hacks for devices (aka
199 * the pty code) which want to control the node attributes themselves.
4062d050
MD
200 *
201 * XXX we may ultimately desire to simply remove the uid/gid/mode
202 * from the node entirely.
9f889dc4
MD
203 *
204 * MPSAFE - sorta. Theoretically the overwrite can compete since they
205 * are loading from the same fields.
4062d050
MD
206 */
207static __inline void
208node_sync_dev_get(struct devfs_node *node)
209{
210 cdev_t dev;
211
212 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
213 node->uid = dev->si_uid;
214 node->gid = dev->si_gid;
215 node->mode = dev->si_perms;
216 }
217}
218
219static __inline void
220node_sync_dev_set(struct devfs_node *node)
221{
222 cdev_t dev;
223
224 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
225 dev->si_uid = node->uid;
226 dev->si_gid = node->gid;
227 dev->si_perms = node->mode;
228 }
229}
21864bc5
MD
230
231/*
232 * generic entry point for unsupported operations
233 */
234static int
9f889dc4 235devfs_vop_badop(struct vop_generic_args *ap)
21864bc5 236{
21864bc5
MD
237 return (EIO);
238}
239
240
241static int
9f889dc4 242devfs_vop_access(struct vop_access_args *ap)
21864bc5
MD
243{
244 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
898c91ee 245 int error;
21864bc5 246
894bbb25
AH
247 if (!devfs_node_is_accessible(node))
248 return ENOENT;
4062d050 249 node_sync_dev_get(node);
21864bc5 250 error = vop_helper_access(ap, node->uid, node->gid,
4062d050 251 node->mode, node->flags);
21864bc5 252
21864bc5 253 return error;
21864bc5
MD
254}
255
256
257static int
9f889dc4 258devfs_vop_inactive(struct vop_inactive_args *ap)
21864bc5 259{
ca8d7677 260 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
21864bc5 261
ca8d7677
MD
262 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
263 vrecycle(ap->a_vp);
21864bc5
MD
264 return 0;
265}
266
267
268static int
9f889dc4 269devfs_vop_reclaim(struct vop_reclaim_args *ap)
21864bc5 270{
be6f2e86
MD
271 struct devfs_node *node;
272 struct vnode *vp;
273 int locked;
274
be6f2e86
MD
275 /*
276 * Check if it is locked already. if not, we acquire the devfs lock
277 */
21864bc5
MD
278 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
279 lockmgr(&devfs_lock, LK_EXCLUSIVE);
280 locked = 1;
be6f2e86
MD
281 } else {
282 locked = 0;
21864bc5
MD
283 }
284
be6f2e86
MD
285 /*
286 * Get rid of the devfs_node if it is no longer linked into the
287 * topology.
288 */
289 vp = ap->a_vp;
290 if ((node = DEVFS_NODE(vp)) != NULL) {
4062d050
MD
291 node->v_node = NULL;
292 if ((node->flags & DEVFS_NODE_LINKED) == 0)
be6f2e86 293 devfs_freep(node);
21864bc5
MD
294 }
295
21864bc5
MD
296 if (locked)
297 lockmgr(&devfs_lock, LK_RELEASE);
298
be6f2e86 299 /*
9b823501
AH
300 * v_rdev needs to be properly released using v_release_rdev
301 * Make sure v_data is NULL as well.
be6f2e86
MD
302 */
303 vp->v_data = NULL;
9b823501 304 v_release_rdev(vp);
21864bc5
MD
305 return 0;
306}
307
308
309static int
9f889dc4 310devfs_vop_readdir(struct vop_readdir_args *ap)
21864bc5 311{
898c91ee 312 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
21864bc5 313 struct devfs_node *node;
21864bc5
MD
314 int cookie_index;
315 int ncookies;
898c91ee
MD
316 int error2;
317 int error;
318 int r;
21864bc5
MD
319 off_t *cookies;
320 off_t saveoff;
321
322 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
323
324 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
325 return (EINVAL);
326 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
327 return (error);
328
c512ab96
MD
329 if (!devfs_node_is_accessible(dnode)) {
330 vn_unlock(ap->a_vp);
ca8d7677 331 return ENOENT;
c512ab96 332 }
ca8d7677
MD
333
334 lockmgr(&devfs_lock, LK_EXCLUSIVE);
335
21864bc5
MD
336 saveoff = ap->a_uio->uio_offset;
337
338 if (ap->a_ncookies) {
339 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
340 if (ncookies > 256)
341 ncookies = 256;
342 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
343 cookie_index = 0;
344 } else {
345 ncookies = -1;
346 cookies = NULL;
347 cookie_index = 0;
348 }
349
898c91ee 350 nanotime(&dnode->atime);
21864bc5
MD
351
352 if (saveoff == 0) {
898c91ee
MD
353 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
354 DT_DIR, 1, ".");
21864bc5
MD
355 if (r)
356 goto done;
357 if (cookies)
358 cookies[cookie_index] = saveoff;
359 saveoff++;
360 cookie_index++;
361 if (cookie_index == ncookies)
362 goto done;
363 }
364
365 if (saveoff == 1) {
898c91ee 366 if (dnode->parent) {
21864bc5 367 r = vop_write_dirent(&error, ap->a_uio,
898c91ee 368 dnode->parent->d_dir.d_ino,
21864bc5
MD
369 DT_DIR, 2, "..");
370 } else {
371 r = vop_write_dirent(&error, ap->a_uio,
898c91ee
MD
372 dnode->d_dir.d_ino,
373 DT_DIR, 2, "..");
21864bc5
MD
374 }
375 if (r)
376 goto done;
377 if (cookies)
378 cookies[cookie_index] = saveoff;
379 saveoff++;
380 cookie_index++;
381 if (cookie_index == ncookies)
382 goto done;
383 }
384
898c91ee
MD
385 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
386 if ((node->flags & DEVFS_HIDDEN) ||
387 (node->flags & DEVFS_INVISIBLE)) {
21864bc5 388 continue;
898c91ee 389 }
21864bc5 390
f7e8960c 391 /*
9f889dc4
MD
392 * If the node type is a valid devfs alias, then we make
393 * sure that the target isn't hidden. If it is, we don't
394 * show the link in the directory listing.
f7e8960c
AH
395 */
396 if ((node->node_type == Plink) && (node->link_target != NULL) &&
397 (node->link_target->flags & DEVFS_HIDDEN))
21864bc5 398 continue;
f7e8960c
AH
399
400 if (node->cookie < saveoff)
21864bc5 401 continue;
f7e8960c 402
21864bc5
MD
403 saveoff = node->cookie;
404
898c91ee
MD
405 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
406 node->d_dir.d_type,
407 node->d_dir.d_namlen,
408 node->d_dir.d_name);
21864bc5 409
898c91ee 410 if (error2)
21864bc5
MD
411 break;
412
413 saveoff++;
414
415 if (cookies)
416 cookies[cookie_index] = node->cookie;
417 ++cookie_index;
418 if (cookie_index == ncookies)
419 break;
21864bc5
MD
420 }
421
422done:
ca8d7677 423 lockmgr(&devfs_lock, LK_RELEASE);
21864bc5
MD
424 vn_unlock(ap->a_vp);
425
426 ap->a_uio->uio_offset = saveoff;
427 if (error && cookie_index == 0) {
428 if (cookies) {
429 kfree(cookies, M_TEMP);
430 *ap->a_ncookies = 0;
431 *ap->a_cookies = NULL;
432 }
433 } else {
434 if (cookies) {
435 *ap->a_ncookies = cookie_index;
436 *ap->a_cookies = cookies;
437 }
438 }
439 return (error);
440}
441
442
443static int
9f889dc4 444devfs_vop_nresolve(struct vop_nresolve_args *ap)
21864bc5 445{
898c91ee 446 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5
MD
447 struct devfs_node *node, *found = NULL;
448 struct namecache *ncp;
449 struct vnode *vp = NULL;
21864bc5
MD
450 int error = 0;
451 int len;
260e4e8b 452 int depth;
21864bc5 453
21864bc5
MD
454 ncp = ap->a_nch->ncp;
455 len = ncp->nc_nlen;
456
898c91ee 457 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
458 return ENOENT;
459
21864bc5
MD
460 lockmgr(&devfs_lock, LK_EXCLUSIVE);
461
898c91ee 462 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
e23485a5 463 error = ENOENT;
21864bc5
MD
464 cache_setvp(ap->a_nch, NULL);
465 goto out;
466 }
467
898c91ee 468 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
21864bc5
MD
469 if (len == node->d_dir.d_namlen) {
470 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
21864bc5
MD
471 found = node;
472 break;
473 }
474 }
475 }
476
477 if (found) {
260e4e8b
AH
478 depth = 0;
479 while ((found->node_type == Plink) && (found->link_target)) {
480 if (depth >= 8) {
481 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
482 break;
483 }
484
21864bc5 485 found = found->link_target;
260e4e8b
AH
486 ++depth;
487 }
21864bc5
MD
488
489 if (!(found->flags & DEVFS_HIDDEN))
490 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
21864bc5
MD
491 }
492
21864bc5 493 if (vp == NULL) {
21864bc5
MD
494 error = ENOENT;
495 cache_setvp(ap->a_nch, NULL);
21864bc5
MD
496 goto out;
497
498 }
21864bc5
MD
499 KKASSERT(vp);
500 vn_unlock(vp);
501 cache_setvp(ap->a_nch, vp);
502 vrele(vp);
21864bc5 503out:
21864bc5 504 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 505
21864bc5
MD
506 return error;
507}
508
509
510static int
9f889dc4 511devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
21864bc5 512{
898c91ee 513 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5 514
898c91ee
MD
515 *ap->a_vpp = NULL;
516 if (!devfs_node_is_accessible(dnode))
894bbb25
AH
517 return ENOENT;
518
21864bc5 519 lockmgr(&devfs_lock, LK_EXCLUSIVE);
898c91ee
MD
520 if (dnode->parent != NULL) {
521 devfs_allocv(ap->a_vpp, dnode->parent);
21864bc5
MD
522 vn_unlock(*ap->a_vpp);
523 }
524 lockmgr(&devfs_lock, LK_RELEASE);
525
526 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
527}
528
529
530static int
9f889dc4 531devfs_vop_getattr(struct vop_getattr_args *ap)
21864bc5 532{
21864bc5 533 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
898c91ee 534 struct vattr *vap = ap->a_vap;
2d076755 535 struct partinfo pinfo;
21864bc5
MD
536 int error = 0;
537
952f0188 538#if 0
894bbb25 539 if (!devfs_node_is_accessible(node))
ca8d7677 540 return ENOENT;
952f0188 541#endif
4062d050 542 node_sync_dev_get(node);
ca8d7677 543
ca8d7677 544 lockmgr(&devfs_lock, LK_EXCLUSIVE);
21864bc5
MD
545
546 /* start by zeroing out the attributes */
547 VATTR_NULL(vap);
548
549 /* next do all the common fields */
550 vap->va_type = ap->a_vp->v_type;
551 vap->va_mode = node->mode;
552 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
a3c7ebb9 553 vap->va_flags = 0;
21864bc5 554 vap->va_blocksize = DEV_BSIZE;
a3c7ebb9 555 vap->va_bytes = vap->va_size = 0;
21864bc5 556
21864bc5
MD
557 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
558
21864bc5
MD
559 vap->va_atime = node->atime;
560 vap->va_mtime = node->mtime;
561 vap->va_ctime = node->ctime;
562
563 vap->va_nlink = 1; /* number of references to file */
564
565 vap->va_uid = node->uid;
566 vap->va_gid = node->gid;
567
568 vap->va_rmajor = 0;
569 vap->va_rminor = 0;
570
898c91ee
MD
571 if ((node->node_type == Pdev) && node->d_dev) {
572 reference_dev(node->d_dev);
573 vap->va_rminor = node->d_dev->si_uminor;
574 release_dev(node->d_dev);
21864bc5
MD
575 }
576
577 /* For a softlink the va_size is the length of the softlink */
898c91ee 578 if (node->symlink_name != 0) {
2d076755 579 vap->va_bytes = vap->va_size = node->symlink_namelen;
21864bc5 580 }
2d076755
AH
581
582 /*
583 * For a disk-type device, va_size is the size of the underlying
584 * device, so that lseek() works properly.
585 */
586 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
587 bzero(&pinfo, sizeof(pinfo));
588 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
589 0, proc0.p_ucred, NULL);
590 if ((error == 0) && (pinfo.media_blksize != 0)) {
591 vap->va_size = pinfo.media_size;
592 } else {
593 vap->va_size = 0;
594 error = 0;
595 }
596 }
597
ca8d7677 598 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 599
894bbb25 600 return (error);
21864bc5
MD
601}
602
603
604static int
9f889dc4 605devfs_vop_setattr(struct vop_setattr_args *ap)
21864bc5 606{
898c91ee 607 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
21864bc5 608 struct vattr *vap;
dffaed1b
AH
609 uid_t cur_uid;
610 gid_t cur_gid;
611 mode_t cur_mode;
21864bc5
MD
612 int error = 0;
613
894bbb25 614 if (!devfs_node_is_accessible(node))
ca8d7677 615 return ENOENT;
4062d050 616 node_sync_dev_get(node);
21864bc5 617
21864bc5
MD
618 lockmgr(&devfs_lock, LK_EXCLUSIVE);
619
620 vap = ap->a_vap;
21864bc5 621
dffaed1b
AH
622 if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
623 cur_uid = node->uid;
624 cur_gid = node->gid;
625 cur_mode = node->mode;
626 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
627 ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
628 if (error)
629 goto out;
630
631 if (node->uid != cur_uid || node->gid != cur_gid) {
632 node->uid = cur_uid;
633 node->gid = cur_gid;
634 node->mode = cur_mode;
21864bc5 635 }
21864bc5
MD
636 }
637
638 if (vap->va_mode != (mode_t)VNOVAL) {
dffaed1b
AH
639 cur_mode = node->mode;
640 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
641 node->uid, node->gid, &cur_mode);
642 if (error == 0 && node->mode != cur_mode) {
643 node->mode = cur_mode;
21864bc5 644 }
21864bc5
MD
645 }
646
647out:
4062d050 648 node_sync_dev_set(node);
07dfa375 649 nanotime(&node->ctime);
21864bc5 650 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 651
21864bc5
MD
652 return error;
653}
654
655
656static int
9f889dc4 657devfs_vop_readlink(struct vop_readlink_args *ap)
21864bc5
MD
658{
659 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
ca8d7677
MD
660 int ret;
661
894bbb25 662 if (!devfs_node_is_accessible(node))
ca8d7677 663 return ENOENT;
21864bc5 664
ca8d7677
MD
665 lockmgr(&devfs_lock, LK_EXCLUSIVE);
666 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
667 lockmgr(&devfs_lock, LK_RELEASE);
668
669 return ret;
21864bc5
MD
670}
671
672
673static int
9f889dc4 674devfs_vop_print(struct vop_print_args *ap)
21864bc5 675{
21864bc5
MD
676 return (0);
677}
678
39a08947 679static int
9f889dc4 680devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
39a08947
AH
681{
682 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
683 struct devfs_node *node;
684
685 if (!devfs_node_is_accessible(dnode))
686 return ENOENT;
687
688 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
689 goto out;
690
691 lockmgr(&devfs_lock, LK_EXCLUSIVE);
692 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
693 ap->a_nch->ncp->nc_name, dnode, NULL);
694
695 if (*ap->a_vpp) {
696 node = DEVFS_NODE(*ap->a_vpp);
697 node->flags |= DEVFS_USER_CREATED;
698 cache_setunresolved(ap->a_nch);
699 cache_setvp(ap->a_nch, *ap->a_vpp);
700 }
701 lockmgr(&devfs_lock, LK_RELEASE);
702out:
703 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
704}
21864bc5
MD
705
706static int
9f889dc4 707devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
21864bc5 708{
898c91ee
MD
709 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
710 struct devfs_node *node;
711 size_t targetlen;
21864bc5 712
898c91ee 713 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
714 return ENOENT;
715
894bbb25
AH
716 ap->a_vap->va_type = VLNK;
717
898c91ee 718 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
21864bc5 719 goto out;
898c91ee 720
21864bc5
MD
721 lockmgr(&devfs_lock, LK_EXCLUSIVE);
722 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
898c91ee 723 ap->a_nch->ncp->nc_name, dnode, NULL);
21864bc5 724
898c91ee 725 targetlen = strlen(ap->a_target);
21864bc5 726 if (*ap->a_vpp) {
898c91ee
MD
727 node = DEVFS_NODE(*ap->a_vpp);
728 node->flags |= DEVFS_USER_CREATED;
729 node->symlink_namelen = targetlen;
730 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
731 memcpy(node->symlink_name, ap->a_target, targetlen);
732 node->symlink_name[targetlen] = '\0';
21864bc5 733 cache_setunresolved(ap->a_nch);
21864bc5
MD
734 cache_setvp(ap->a_nch, *ap->a_vpp);
735 }
736 lockmgr(&devfs_lock, LK_RELEASE);
737out:
738 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
21864bc5
MD
739}
740
39a08947 741static int
9f889dc4 742devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
39a08947
AH
743{
744 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
745 struct devfs_node *node;
746 struct namecache *ncp;
747 int error = ENOENT;
748
749 ncp = ap->a_nch->ncp;
750
751 if (!devfs_node_is_accessible(dnode))
752 return ENOENT;
753
754 lockmgr(&devfs_lock, LK_EXCLUSIVE);
755
756 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
757 goto out;
758
759 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
760 if (ncp->nc_nlen != node->d_dir.d_namlen)
761 continue;
762 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
763 continue;
764
765 /*
766 * only allow removal of user created dirs
767 */
768 if ((node->flags & DEVFS_USER_CREATED) == 0) {
769 error = EPERM;
770 goto out;
771 } else if (node->node_type != Pdir) {
772 error = ENOTDIR;
773 goto out;
774 } else if (node->nchildren > 2) {
775 error = ENOTEMPTY;
776 goto out;
777 } else {
778 if (node->v_node)
779 cache_inval_vp(node->v_node, CINV_DESTROY);
780 devfs_unlinkp(node);
781 error = 0;
782 break;
783 }
784 }
785
786 cache_setunresolved(ap->a_nch);
787 cache_setvp(ap->a_nch, NULL);
788
789out:
790 lockmgr(&devfs_lock, LK_RELEASE);
791 return error;
792}
21864bc5
MD
793
794static int
9f889dc4 795devfs_vop_nremove(struct vop_nremove_args *ap)
21864bc5 796{
898c91ee 797 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5
MD
798 struct devfs_node *node;
799 struct namecache *ncp;
21864bc5
MD
800 int error = ENOENT;
801
21864bc5
MD
802 ncp = ap->a_nch->ncp;
803
898c91ee 804 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
805 return ENOENT;
806
21864bc5
MD
807 lockmgr(&devfs_lock, LK_EXCLUSIVE);
808
898c91ee 809 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
21864bc5 810 goto out;
21864bc5 811
898c91ee
MD
812 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
813 if (ncp->nc_nlen != node->d_dir.d_namlen)
814 continue;
815 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
816 continue;
21864bc5 817
898c91ee
MD
818 /*
819 * only allow removal of user created stuff (e.g. symlinks)
820 */
821 if ((node->flags & DEVFS_USER_CREATED) == 0) {
822 error = EPERM;
823 goto out;
39a08947
AH
824 } else if (node->node_type == Pdir) {
825 error = EISDIR;
826 goto out;
898c91ee
MD
827 } else {
828 if (node->v_node)
829 cache_inval_vp(node->v_node, CINV_DESTROY);
830 devfs_unlinkp(node);
831 error = 0;
832 break;
21864bc5
MD
833 }
834 }
835
836 cache_setunresolved(ap->a_nch);
837 cache_setvp(ap->a_nch, NULL);
21864bc5
MD
838
839out:
840 lockmgr(&devfs_lock, LK_RELEASE);
21864bc5
MD
841 return error;
842}
843
844
845static int
846devfs_spec_open(struct vop_open_args *ap)
847{
848 struct vnode *vp = ap->a_vp;
ca8d7677 849 struct vnode *orig_vp = NULL;
898c91ee
MD
850 struct devfs_node *node = DEVFS_NODE(vp);
851 struct devfs_node *newnode;
21864bc5 852 cdev_t dev, ndev = NULL;
21864bc5 853 int error = 0;
21864bc5 854
898c91ee
MD
855 if (node) {
856 if (node->d_dev == NULL)
21864bc5 857 return ENXIO;
898c91ee 858 if (!devfs_node_is_accessible(node))
894bbb25 859 return ENOENT;
21864bc5
MD
860 }
861
21864bc5
MD
862 if ((dev = vp->v_rdev) == NULL)
863 return ENXIO;
864
898c91ee 865 if (node && ap->a_fp) {
21864bc5
MD
866 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
867 lockmgr(&devfs_lock, LK_EXCLUSIVE);
07dfa375
AH
868
869 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
870 ap->a_mode, ap->a_cred);
871 if (ndev != NULL) {
898c91ee
MD
872 newnode = devfs_create_device_node(
873 DEVFS_MNTDATA(vp->v_mount)->root_node,
07dfa375
AH
874 ndev, NULL, NULL);
875 /* XXX: possibly destroy device if this happens */
898c91ee 876
07dfa375
AH
877 if (newnode != NULL) {
878 dev = ndev;
879 devfs_link_dev(dev);
21864bc5 880
07dfa375
AH
881 devfs_debug(DEVFS_DEBUG_DEBUG,
882 "parent here is: %s, node is: |%s|\n",
883 ((node->parent->node_type == Proot) ?
884 "ROOT!" : node->parent->d_dir.d_name),
885 newnode->d_dir.d_name);
886 devfs_debug(DEVFS_DEBUG_DEBUG,
887 "test: %s\n",
888 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
889
890 /*
891 * orig_vp is set to the original vp if we cloned.
892 */
893 /* node->flags |= DEVFS_CLONED; */
894 devfs_allocv(&vp, newnode);
895 orig_vp = ap->a_vp;
896 ap->a_vp = vp;
897 }
21864bc5
MD
898 }
899 lockmgr(&devfs_lock, LK_RELEASE);
900 }
901
898c91ee
MD
902 devfs_debug(DEVFS_DEBUG_DEBUG,
903 "devfs_spec_open() called on %s! \n",
904 dev->si_name);
905
21864bc5
MD
906 /*
907 * Make this field valid before any I/O in ->d_open
908 */
909 if (!dev->si_iosize_max)
910 dev->si_iosize_max = DFLTPHYS;
911
912 if (dev_dflags(dev) & D_TTY)
2247fe02 913 vsetflags(vp, VISTTY);
21864bc5 914
9f889dc4
MD
915 /*
916 * Open underlying device
917 */
21864bc5
MD
918 vn_unlock(vp);
919 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
920 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
921
ca8d7677
MD
922 /*
923 * Clean up any cloned vp if we error out.
924 */
21864bc5 925 if (error) {
ca8d7677 926 if (orig_vp) {
21864bc5 927 vput(vp);
ca8d7677
MD
928 ap->a_vp = orig_vp;
929 /* orig_vp = NULL; */
930 }
21864bc5
MD
931 return error;
932 }
933
d894b0eb 934 /*
ddd7de82
AH
935 * This checks if the disk device is going to be opened for writing.
936 * It will be only allowed in the cases where securelevel permits it
937 * and it's not mounted R/W.
d894b0eb 938 */
ddd7de82
AH
939 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
940 (ap->a_cred != FSCRED)) {
941
942 /* Very secure mode. No open for writing allowed */
943 if (securelevel >= 2)
944 return EPERM;
945
946 /*
947 * If it is mounted R/W, do not allow to open for writing.
948 * In the case it's mounted read-only but securelevel
949 * is >= 1, then do not allow opening for writing either.
950 */
951 if (vfs_mountedon(vp)) {
952 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
953 return EBUSY;
954 else if (securelevel >= 1)
d894b0eb
AH
955 return EPERM;
956 }
957 }
21864bc5
MD
958
959 if (dev_dflags(dev) & D_TTY) {
960 if (dev->si_tty) {
961 struct tty *tp;
962 tp = dev->si_tty;
963 if (!tp->t_stop) {
898c91ee
MD
964 devfs_debug(DEVFS_DEBUG_DEBUG,
965 "devfs: no t_stop\n");
21864bc5
MD
966 tp->t_stop = nottystop;
967 }
968 }
969 }
970
971
972 if (vn_isdisk(vp, NULL)) {
973 if (!dev->si_bsize_phys)
974 dev->si_bsize_phys = DEV_BSIZE;
b0d18f7d 975 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
21864bc5
MD
976 }
977
978 vop_stdopen(ap);
07dfa375 979#if 0
898c91ee
MD
980 if (node)
981 nanotime(&node->atime);
07dfa375 982#endif
21864bc5 983
b80b4c32
MD
984 /*
985 * If we replaced the vp the vop_stdopen() call will have loaded
986 * it into fp->f_data and vref()d the vp, giving us two refs. So
987 * instead of just unlocking it here we have to vput() it.
988 */
ca8d7677 989 if (orig_vp)
b80b4c32 990 vput(vp);
21864bc5
MD
991
992 /* Ugly pty magic, to make pty devices appear once they are opened */
898c91ee
MD
993 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
994 node->flags &= ~DEVFS_INVISIBLE;
21864bc5
MD
995
996 if (ap->a_fp) {
b80b4c32 997 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
fb12e9a5 998 KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
21864bc5 999 ap->a_fp->f_ops = &devfs_dev_fileops;
b80b4c32 1000 KKASSERT(ap->a_fp->f_data == (void *)vp);
21864bc5
MD
1001 }
1002
21864bc5
MD
1003 return 0;
1004}
1005
1006
1007static int
1008devfs_spec_close(struct vop_close_args *ap)
1009{
92fb0c6a 1010 struct devfs_node *node;
21864bc5
MD
1011 struct proc *p = curproc;
1012 struct vnode *vp = ap->a_vp;
1013 cdev_t dev = vp->v_rdev;
1014 int error = 0;
1015 int needrelock;
1016
52f98df9
SG
1017 if (dev)
1018 devfs_debug(DEVFS_DEBUG_DEBUG,
1019 "devfs_spec_close() called on %s! \n",
1020 dev->si_name);
1021 else
1022 devfs_debug(DEVFS_DEBUG_DEBUG,
1023 "devfs_spec_close() called, null vode!\n");
21864bc5
MD
1024
1025 /*
1026 * A couple of hacks for devices and tty devices. The
1027 * vnode ref count cannot be used to figure out the
1028 * last close, but we can use v_opencount now that
1029 * revoke works properly.
1030 *
1031 * Detect the last close on a controlling terminal and clear
1032 * the session (half-close).
1033 */
1034 if (dev)
1035 reference_dev(dev);
1036
1037 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1038 p->p_session->s_ttyvp = NULL;
1039 vrele(vp);
1040 }
1041
1042 /*
1043 * Vnodes can be opened and closed multiple times. Do not really
1044 * close the device unless (1) it is being closed forcibly,
1045 * (2) the device wants to track closes, or (3) this is the last
1046 * vnode doing its last close on the device.
1047 *
1048 * XXX the VXLOCK (force close) case can leave vnodes referencing
1049 * a closed device. This might not occur now that our revoke is
1050 * fixed.
1051 */
1052 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1053 if (dev && ((vp->v_flag & VRECLAIMED) ||
1054 (dev_dflags(dev) & D_TRACKCLOSE) ||
1055 (vp->v_opencount == 1))) {
898c91ee 1056 /*
92fb0c6a
MD
1057 * Ugly pty magic, to make pty devices disappear again once
1058 * they are closed.
1059 */
1060 node = DEVFS_NODE(ap->a_vp);
1061 if (node && (node->flags & DEVFS_PTY))
1062 node->flags |= DEVFS_INVISIBLE;
1063
1064 /*
898c91ee
MD
1065 * Unlock around dev_dclose()
1066 */
21864bc5
MD
1067 needrelock = 0;
1068 if (vn_islocked(vp)) {
1069 needrelock = 1;
1070 vn_unlock(vp);
1071 }
898c91ee
MD
1072
1073 /*
92fb0c6a
MD
1074 * WARNING! If the device destroys itself the devfs node
1075 * can disappear here.
898c91ee 1076 */
92fb0c6a
MD
1077 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1078 /* node is now stale */
21864bc5
MD
1079
1080 if (needrelock)
1081 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1082 } else {
1083 error = 0;
1084 }
1085 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
898c91ee 1086
21864bc5
MD
1087 /*
1088 * Track the actual opens and closes on the vnode. The last close
898c91ee
MD
1089 * disassociates the rdev. If the rdev is already disassociated or
1090 * the opencount is already 0, the vnode might have been revoked
1091 * and no further opencount tracking occurs.
21864bc5 1092 */
898c91ee 1093 if (dev)
21864bc5 1094 release_dev(dev);
898c91ee 1095 if (vp->v_opencount > 0)
21864bc5 1096 vop_stdclose(ap);
21864bc5
MD
1097 return(error);
1098
1099}
1100
1101
1102static int
9f889dc4 1103devfs_fo_close(struct file *fp)
21864bc5 1104{
21864bc5 1105 struct vnode *vp = (struct vnode *)fp->f_data;
898c91ee 1106 int error;
21864bc5 1107
21864bc5 1108 fp->f_ops = &badfileops;
21864bc5 1109 error = vn_close(vp, fp->f_flag);
21864bc5
MD
1110
1111 return (error);
1112}
1113
1114
1115/*
1116 * Device-optimized file table vnode read routine.
1117 *
1118 * This bypasses the VOP table and talks directly to the device. Most
1119 * filesystems just route to specfs and can make this optimization.
1120 *
1121 * MPALMOSTSAFE - acquires mplock
1122 */
1123static int
9f889dc4 1124devfs_fo_read(struct file *fp, struct uio *uio,
898c91ee 1125 struct ucred *cred, int flags)
21864bc5 1126{
898c91ee 1127 struct devfs_node *node;
21864bc5
MD
1128 struct vnode *vp;
1129 int ioflag;
1130 int error;
1131 cdev_t dev;
1132
21864bc5
MD
1133 KASSERT(uio->uio_td == curthread,
1134 ("uio_td %p is not td %p", uio->uio_td, curthread));
1135
3a1032a6
AH
1136 if (uio->uio_resid == 0)
1137 return 0;
1138
21864bc5 1139 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1140 if (vp == NULL || vp->v_type == VBAD)
1141 return EBADF;
1142
898c91ee 1143 node = DEVFS_NODE(vp);
21864bc5 1144
3a1032a6
AH
1145 if ((dev = vp->v_rdev) == NULL)
1146 return EBADF;
1147
21864bc5
MD
1148 reference_dev(dev);
1149
21864bc5
MD
1150 if ((flags & O_FOFFSET) == 0)
1151 uio->uio_offset = fp->f_offset;
1152
1153 ioflag = 0;
1154 if (flags & O_FBLOCKING) {
1155 /* ioflag &= ~IO_NDELAY; */
1156 } else if (flags & O_FNONBLOCKING) {
1157 ioflag |= IO_NDELAY;
1158 } else if (fp->f_flag & FNONBLOCK) {
1159 ioflag |= IO_NDELAY;
1160 }
1161 if (flags & O_FBUFFERED) {
1162 /* ioflag &= ~IO_DIRECT; */
1163 } else if (flags & O_FUNBUFFERED) {
1164 ioflag |= IO_DIRECT;
1165 } else if (fp->f_flag & O_DIRECT) {
1166 ioflag |= IO_DIRECT;
1167 }
1168 ioflag |= sequential_heuristic(uio, fp);
1169
1170 error = dev_dread(dev, uio, ioflag);
1171
1172 release_dev(dev);
898c91ee
MD
1173 if (node)
1174 nanotime(&node->atime);
21864bc5
MD
1175 if ((flags & O_FOFFSET) == 0)
1176 fp->f_offset = uio->uio_offset;
1177 fp->f_nextoff = uio->uio_offset;
3a1032a6 1178
21864bc5
MD
1179 return (error);
1180}
1181
1182
1183static int
9f889dc4 1184devfs_fo_write(struct file *fp, struct uio *uio,
898c91ee 1185 struct ucred *cred, int flags)
21864bc5 1186{
898c91ee 1187 struct devfs_node *node;
21864bc5
MD
1188 struct vnode *vp;
1189 int ioflag;
1190 int error;
1191 cdev_t dev;
1192
21864bc5
MD
1193 KASSERT(uio->uio_td == curthread,
1194 ("uio_td %p is not p %p", uio->uio_td, curthread));
1195
1196 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1197 if (vp == NULL || vp->v_type == VBAD)
1198 return EBADF;
1199
898c91ee 1200 node = DEVFS_NODE(vp);
3a1032a6 1201
21864bc5
MD
1202 if (vp->v_type == VREG)
1203 bwillwrite(uio->uio_resid);
3a1032a6 1204
21864bc5
MD
1205 vp = (struct vnode *)fp->f_data;
1206
3a1032a6
AH
1207 if ((dev = vp->v_rdev) == NULL)
1208 return EBADF;
1209
21864bc5
MD
1210 reference_dev(dev);
1211
1212 if ((flags & O_FOFFSET) == 0)
1213 uio->uio_offset = fp->f_offset;
1214
1215 ioflag = IO_UNIT;
1216 if (vp->v_type == VREG &&
1217 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1218 ioflag |= IO_APPEND;
1219 }
1220
1221 if (flags & O_FBLOCKING) {
1222 /* ioflag &= ~IO_NDELAY; */
1223 } else if (flags & O_FNONBLOCKING) {
1224 ioflag |= IO_NDELAY;
1225 } else if (fp->f_flag & FNONBLOCK) {
1226 ioflag |= IO_NDELAY;
1227 }
1228 if (flags & O_FBUFFERED) {
1229 /* ioflag &= ~IO_DIRECT; */
1230 } else if (flags & O_FUNBUFFERED) {
1231 ioflag |= IO_DIRECT;
1232 } else if (fp->f_flag & O_DIRECT) {
1233 ioflag |= IO_DIRECT;
1234 }
1235 if (flags & O_FASYNCWRITE) {
1236 /* ioflag &= ~IO_SYNC; */
1237 } else if (flags & O_FSYNCWRITE) {
1238 ioflag |= IO_SYNC;
1239 } else if (fp->f_flag & O_FSYNC) {
1240 ioflag |= IO_SYNC;
1241 }
1242
1243 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1244 ioflag |= IO_SYNC;
1245 ioflag |= sequential_heuristic(uio, fp);
1246
1247 error = dev_dwrite(dev, uio, ioflag);
1248
1249 release_dev(dev);
07dfa375
AH
1250 if (node) {
1251 nanotime(&node->atime);
898c91ee 1252 nanotime(&node->mtime);
07dfa375 1253 }
21864bc5
MD
1254
1255 if ((flags & O_FOFFSET) == 0)
1256 fp->f_offset = uio->uio_offset;
1257 fp->f_nextoff = uio->uio_offset;
3a1032a6 1258
21864bc5
MD
1259 return (error);
1260}
1261
1262
1263static int
9f889dc4 1264devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
21864bc5
MD
1265{
1266 struct vnode *vp;
3a1032a6
AH
1267 struct vattr vattr;
1268 struct vattr *vap;
1269 u_short mode;
1270 cdev_t dev;
21864bc5
MD
1271 int error;
1272
21864bc5 1273 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1274 if (vp == NULL || vp->v_type == VBAD)
1275 return EBADF;
1276
21864bc5 1277 error = vn_stat(vp, sb, cred);
3a1032a6 1278 if (error)
21864bc5 1279 return (error);
21864bc5
MD
1280
1281 vap = &vattr;
1282 error = VOP_GETATTR(vp, vap);
3a1032a6 1283 if (error)
21864bc5 1284 return (error);
21864bc5
MD
1285
1286 /*
1287 * Zero the spare stat fields
1288 */
1289 sb->st_lspare = 0;
d98152a8
MD
1290 sb->st_qspare1 = 0;
1291 sb->st_qspare2 = 0;
21864bc5
MD
1292
1293 /*
1294 * Copy from vattr table ... or not in case it's a cloned device
1295 */
1296 if (vap->va_fsid != VNOVAL)
1297 sb->st_dev = vap->va_fsid;
1298 else
1299 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1300
1301 sb->st_ino = vap->va_fileid;
1302
1303 mode = vap->va_mode;
1304 mode |= S_IFCHR;
1305 sb->st_mode = mode;
1306
1307 if (vap->va_nlink > (nlink_t)-1)
1308 sb->st_nlink = (nlink_t)-1;
1309 else
1310 sb->st_nlink = vap->va_nlink;
3a1032a6 1311
21864bc5
MD
1312 sb->st_uid = vap->va_uid;
1313 sb->st_gid = vap->va_gid;
ca8d7677 1314 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
2d076755 1315 sb->st_size = vap->va_bytes;
21864bc5
MD
1316 sb->st_atimespec = vap->va_atime;
1317 sb->st_mtimespec = vap->va_mtime;
1318 sb->st_ctimespec = vap->va_ctime;
1319
1320 /*
1321 * A VCHR and VBLK device may track the last access and last modified
1322 * time independantly of the filesystem. This is particularly true
1323 * because device read and write calls may bypass the filesystem.
1324 */
1325 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1326 dev = vp->v_rdev;
1327 if (dev != NULL) {
1328 if (dev->si_lastread) {
1329 sb->st_atimespec.tv_sec = dev->si_lastread;
1330 sb->st_atimespec.tv_nsec = 0;
1331 }
1332 if (dev->si_lastwrite) {
1333 sb->st_atimespec.tv_sec = dev->si_lastwrite;
1334 sb->st_atimespec.tv_nsec = 0;
1335 }
1336 }
1337 }
1338
1339 /*
1340 * According to www.opengroup.org, the meaning of st_blksize is
1341 * "a filesystem-specific preferred I/O block size for this
1342 * object. In some filesystem types, this may vary from file
1343 * to file"
1344 * Default to PAGE_SIZE after much discussion.
1345 */
1346
1347 sb->st_blksize = PAGE_SIZE;
1348
1349 sb->st_flags = vap->va_flags;
1350
1351 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1352 if (error)
1353 sb->st_gen = 0;
1354 else
1355 sb->st_gen = (u_int32_t)vap->va_gen;
1356
1357 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
21864bc5 1358
21864bc5
MD
1359 return (0);
1360}
1361
1362
1363static int
9f889dc4 1364devfs_fo_kqfilter(struct file *fp, struct knote *kn)
21864bc5
MD
1365{
1366 struct vnode *vp;
21864bc5
MD
1367 int error;
1368 cdev_t dev;
1369
21864bc5
MD
1370 vp = (struct vnode *)fp->f_data;
1371 if (vp == NULL || vp->v_type == VBAD) {
1372 error = EBADF;
1373 goto done;
1374 }
21864bc5
MD
1375 if ((dev = vp->v_rdev) == NULL) {
1376 error = EBADF;
1377 goto done;
1378 }
1379 reference_dev(dev);
1380
1381 error = dev_dkqfilter(dev, kn);
1382
1383 release_dev(dev);
1384
21864bc5 1385done:
b287d649 1386 return (error);
21864bc5
MD
1387}
1388
21864bc5
MD
1389/*
1390 * MPALMOSTSAFE - acquires mplock
1391 */
1392static int
9f889dc4 1393devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
87baaf0c 1394 struct ucred *ucred, struct sysmsg *msg)
21864bc5 1395{
898c91ee
MD
1396 struct devfs_node *node;
1397 struct vnode *vp;
21864bc5 1398 struct vnode *ovp;
21864bc5
MD
1399 cdev_t dev;
1400 int error;
1401 struct fiodname_args *name_args;
1402 size_t namlen;
1403 const char *name;
1404
898c91ee 1405 vp = ((struct vnode *)fp->f_data);
3a1032a6
AH
1406
1407 if ((dev = vp->v_rdev) == NULL)
1408 return EBADF; /* device was revoked */
1409
1410 reference_dev(dev);
21864bc5 1411
898c91ee
MD
1412 node = DEVFS_NODE(vp);
1413
1414 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1415 "devfs_fo_ioctl() called! for dev %s\n",
898c91ee 1416 dev->si_name);
21864bc5
MD
1417
1418 if (com == FIODTYPE) {
1419 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1420 error = 0;
1421 goto out;
1422 } else if (com == FIODNAME) {
1423 name_args = (struct fiodname_args *)data;
1424 name = dev->si_name;
1425 namlen = strlen(name) + 1;
1426
898c91ee
MD
1427 devfs_debug(DEVFS_DEBUG_DEBUG,
1428 "ioctl, got: FIODNAME for %s\n", name);
21864bc5
MD
1429
1430 if (namlen <= name_args->len)
1431 error = copyout(dev->si_name, name_args->name, namlen);
1432 else
1433 error = EINVAL;
1434
898c91ee
MD
1435 devfs_debug(DEVFS_DEBUG_DEBUG,
1436 "ioctl stuff: error: %d\n", error);
21864bc5
MD
1437 goto out;
1438 }
3a1032a6 1439
87baaf0c 1440 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
3a1032a6 1441
07dfa375 1442#if 0
898c91ee
MD
1443 if (node) {
1444 nanotime(&node->atime);
1445 nanotime(&node->mtime);
21864bc5 1446 }
07dfa375 1447#endif
898c91ee
MD
1448 if (com == TIOCSCTTY) {
1449 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1450 "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
898c91ee
MD
1451 dev->si_name);
1452 }
21864bc5 1453 if (error == 0 && com == TIOCSCTTY) {
21864bc5
MD
1454 struct proc *p = curthread->td_proc;
1455 struct session *sess;
898c91ee
MD
1456
1457 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1458 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
898c91ee
MD
1459 dev->si_name);
1460 if (p == NULL) {
21864bc5
MD
1461 error = ENOTTY;
1462 goto out;
1463 }
1464 sess = p->p_session;
898c91ee
MD
1465
1466 /*
1467 * Do nothing if reassigning same control tty
1468 */
21864bc5
MD
1469 if (sess->s_ttyvp == vp) {
1470 error = 0;
1471 goto out;
1472 }
898c91ee
MD
1473
1474 /*
1475 * Get rid of reference to old control tty
1476 */
21864bc5
MD
1477 ovp = sess->s_ttyvp;
1478 vref(vp);
1479 sess->s_ttyvp = vp;
1480 if (ovp)
1481 vrele(ovp);
1482 }
1483
1484out:
3a1032a6 1485 release_dev(dev);
9f889dc4 1486 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
21864bc5
MD
1487 return (error);
1488}
1489
1490
1491static int
1492devfs_spec_fsync(struct vop_fsync_args *ap)
1493{
1494 struct vnode *vp = ap->a_vp;
1495 int error;
1496
1497 if (!vn_isdisk(vp, NULL))
1498 return (0);
1499
1500 /*
1501 * Flush all dirty buffers associated with a block device.
1502 */
1503 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1504 return (error);
1505}
1506
21864bc5
MD
1507static int
1508devfs_spec_read(struct vop_read_args *ap)
1509{
898c91ee 1510 struct devfs_node *node;
21864bc5
MD
1511 struct vnode *vp;
1512 struct uio *uio;
1513 cdev_t dev;
1514 int error;
1515
1516 vp = ap->a_vp;
1517 dev = vp->v_rdev;
1518 uio = ap->a_uio;
898c91ee 1519 node = DEVFS_NODE(vp);
21864bc5
MD
1520
1521 if (dev == NULL) /* device was revoked */
1522 return (EBADF);
1523 if (uio->uio_resid == 0)
1524 return (0);
1525
1526 vn_unlock(vp);
1527 error = dev_dread(dev, uio, ap->a_ioflag);
1528 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1529
898c91ee
MD
1530 if (node)
1531 nanotime(&node->atime);
21864bc5
MD
1532
1533 return (error);
1534}
1535
1536/*
1537 * Vnode op for write
1538 *
1539 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1540 * struct ucred *a_cred)
1541 */
21864bc5
MD
1542static int
1543devfs_spec_write(struct vop_write_args *ap)
1544{
898c91ee 1545 struct devfs_node *node;
21864bc5
MD
1546 struct vnode *vp;
1547 struct uio *uio;
1548 cdev_t dev;
1549 int error;
1550
1551 vp = ap->a_vp;
1552 dev = vp->v_rdev;
1553 uio = ap->a_uio;
898c91ee 1554 node = DEVFS_NODE(vp);
21864bc5
MD
1555
1556 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1557
1558 if (dev == NULL) /* device was revoked */
1559 return (EBADF);
1560
1561 vn_unlock(vp);
1562 error = dev_dwrite(dev, uio, ap->a_ioflag);
1563 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1564
07dfa375
AH
1565 if (node) {
1566 nanotime(&node->atime);
898c91ee 1567 nanotime(&node->mtime);
07dfa375 1568 }
21864bc5
MD
1569
1570 return (error);
1571}
1572
1573/*
1574 * Device ioctl operation.
1575 *
1576 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
87baaf0c 1577 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
21864bc5 1578 */
21864bc5
MD
1579static int
1580devfs_spec_ioctl(struct vop_ioctl_args *ap)
1581{
21864bc5 1582 struct vnode *vp = ap->a_vp;
898c91ee
MD
1583 struct devfs_node *node;
1584 cdev_t dev;
21864bc5
MD
1585
1586 if ((dev = vp->v_rdev) == NULL)
1587 return (EBADF); /* device was revoked */
898c91ee 1588 node = DEVFS_NODE(vp);
21864bc5 1589
07dfa375 1590#if 0
898c91ee
MD
1591 if (node) {
1592 nanotime(&node->atime);
1593 nanotime(&node->mtime);
21864bc5 1594 }
07dfa375 1595#endif
21864bc5 1596
87baaf0c
MD
1597 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1598 ap->a_cred, ap->a_sysmsg));
21864bc5
MD
1599}
1600
1601/*
21864bc5
MD
1602 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1603 */
1604/* ARGSUSED */
1605static int
1606devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1607{
21864bc5 1608 struct vnode *vp = ap->a_vp;
898c91ee
MD
1609 struct devfs_node *node;
1610 cdev_t dev;
21864bc5
MD
1611
1612 if ((dev = vp->v_rdev) == NULL)
b287d649 1613 return (EBADF); /* device was revoked (EBADF) */
898c91ee 1614 node = DEVFS_NODE(vp);
21864bc5 1615
07dfa375 1616#if 0
898c91ee
MD
1617 if (node)
1618 nanotime(&node->atime);
07dfa375 1619#endif
21864bc5 1620
b287d649 1621 return (dev_dkqfilter(dev, ap->a_kn));
21864bc5
MD
1622}
1623
21864bc5
MD
1624/*
1625 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1626 * calls are not limited to device DMA limits so we have to deal with the
1627 * case.
1628 *
1629 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1630 */
1631static int
1632devfs_spec_strategy(struct vop_strategy_args *ap)
1633{
1634 struct bio *bio = ap->a_bio;
1635 struct buf *bp = bio->bio_buf;
1636 struct buf *nbp;
1637 struct vnode *vp;
1638 struct mount *mp;
1639 int chunksize;
1640 int maxiosize;
1641
1642 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1643 buf_start(bp);
1644
1645 /*
1646 * Collect statistics on synchronous and asynchronous read
1647 * and write counts for disks that have associated filesystems.
1648 */
1649 vp = ap->a_vp;
1650 KKASSERT(vp->v_rdev != NULL); /* XXX */
1651 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1652 if (bp->b_cmd == BUF_CMD_READ) {
21864bc5
MD
1653 if (bp->b_flags & BIO_SYNC)
1654 mp->mnt_stat.f_syncreads++;
1655 else
1656 mp->mnt_stat.f_asyncreads++;
1657 } else {
1658 if (bp->b_flags & BIO_SYNC)
1659 mp->mnt_stat.f_syncwrites++;
1660 else
1661 mp->mnt_stat.f_asyncwrites++;
1662 }
1663 }
1664
1665 /*
1666 * Device iosize limitations only apply to read and write. Shortcut
1667 * the I/O if it fits.
1668 */
1669 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
898c91ee
MD
1670 devfs_debug(DEVFS_DEBUG_DEBUG,
1671 "%s: si_iosize_max not set!\n",
1672 dev_dname(vp->v_rdev));
21864bc5
MD
1673 maxiosize = MAXPHYS;
1674 }
1675#if SPEC_CHAIN_DEBUG & 2
1676 maxiosize = 4096;
1677#endif
1678 if (bp->b_bcount <= maxiosize ||
1679 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1680 dev_dstrategy_chain(vp->v_rdev, bio);
1681 return (0);
1682 }
1683
1684 /*
1685 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1686 */
1687 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1688 initbufbio(nbp);
1689 buf_dep_init(nbp);
21864bc5
MD
1690 BUF_LOCK(nbp, LK_EXCLUSIVE);
1691 BUF_KERNPROC(nbp);
1692 nbp->b_vp = vp;
1693 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1694 nbp->b_data = bp->b_data;
1695 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1696 nbp->b_bio1.bio_offset = bio->bio_offset;
1697 nbp->b_bio1.bio_caller_info1.ptr = bio;
1698
1699 /*
1700 * Start the first transfer
1701 */
1702 if (vn_isdisk(vp, NULL))
1703 chunksize = vp->v_rdev->si_bsize_phys;
1704 else
1705 chunksize = DEV_BSIZE;
1706 chunksize = maxiosize / chunksize * chunksize;
1707#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1708 devfs_debug(DEVFS_DEBUG_DEBUG,
1709 "spec_strategy chained I/O chunksize=%d\n",
1710 chunksize);
21864bc5
MD
1711#endif
1712 nbp->b_cmd = bp->b_cmd;
1713 nbp->b_bcount = chunksize;
1714 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1715 nbp->b_bio1.bio_caller_info2.index = chunksize;
1716
1717#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1718 devfs_debug(DEVFS_DEBUG_DEBUG,
1719 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1720 bp, 0, bp->b_bcount, nbp->b_bcount);
21864bc5
MD
1721#endif
1722
1723 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1724
1725 if (DEVFS_NODE(vp)) {
1726 nanotime(&DEVFS_NODE(vp)->atime);
1727 nanotime(&DEVFS_NODE(vp)->mtime);
1728 }
1729
1730 return (0);
1731}
1732
1733/*
1734 * Chunked up transfer completion routine - chain transfers until done
77912481
MD
1735 *
1736 * NOTE: MPSAFE callback.
21864bc5
MD
1737 */
1738static
1739void
1740devfs_spec_strategy_done(struct bio *nbio)
1741{
1742 struct buf *nbp = nbio->bio_buf;
1743 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1744 struct buf *bp = bio->bio_buf; /* original bp */
1745 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1746 int boffset = nbp->b_data - bp->b_data;
1747
1748 if (nbp->b_flags & B_ERROR) {
1749 /*
1750 * An error terminates the chain, propogate the error back
1751 * to the original bp
1752 */
1753 bp->b_flags |= B_ERROR;
1754 bp->b_error = nbp->b_error;
1755 bp->b_resid = bp->b_bcount - boffset +
1756 (nbp->b_bcount - nbp->b_resid);
1757#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1758 devfs_debug(DEVFS_DEBUG_DEBUG,
1759 "spec_strategy: chain %p error %d bcount %d/%d\n",
1760 bp, bp->b_error, bp->b_bcount,
1761 bp->b_bcount - bp->b_resid);
21864bc5 1762#endif
21864bc5
MD
1763 } else if (nbp->b_resid) {
1764 /*
1765 * A short read or write terminates the chain
1766 */
1767 bp->b_error = nbp->b_error;
1768 bp->b_resid = bp->b_bcount - boffset +
1769 (nbp->b_bcount - nbp->b_resid);
1770#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1771 devfs_debug(DEVFS_DEBUG_DEBUG,
1772 "spec_strategy: chain %p short read(1) "
1773 "bcount %d/%d\n",
1774 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
21864bc5 1775#endif
21864bc5
MD
1776 } else if (nbp->b_bcount != nbp->b_bufsize) {
1777 /*
1778 * A short read or write can also occur by truncating b_bcount
1779 */
1780#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1781 devfs_debug(DEVFS_DEBUG_DEBUG,
1782 "spec_strategy: chain %p short read(2) "
1783 "bcount %d/%d\n",
1784 bp, nbp->b_bcount + boffset, bp->b_bcount);
21864bc5
MD
1785#endif
1786 bp->b_error = 0;
1787 bp->b_bcount = nbp->b_bcount + boffset;
1788 bp->b_resid = nbp->b_resid;
21864bc5
MD
1789 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1790 /*
1791 * No more data terminates the chain
1792 */
1793#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1794 devfs_debug(DEVFS_DEBUG_DEBUG,
1795 "spec_strategy: chain %p finished bcount %d\n",
1796 bp, bp->b_bcount);
21864bc5
MD
1797#endif
1798 bp->b_error = 0;
1799 bp->b_resid = 0;
21864bc5
MD
1800 } else {
1801 /*
1802 * Continue the chain
1803 */
1804 boffset += nbp->b_bcount;
1805 nbp->b_data = bp->b_data + boffset;
1806 nbp->b_bcount = bp->b_bcount - boffset;
1807 if (nbp->b_bcount > chunksize)
1808 nbp->b_bcount = chunksize;
1809 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1810 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1811
1812#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1813 devfs_debug(DEVFS_DEBUG_DEBUG,
1814 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1815 bp, boffset, bp->b_bcount, nbp->b_bcount);
21864bc5
MD
1816#endif
1817
1818 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
b5d7061d 1819 return;
21864bc5 1820 }
b5d7061d
MD
1821
1822 /*
1823 * Fall through to here on termination. biodone(bp) and
1824 * clean up and free nbp.
1825 */
1826 biodone(bio);
1827 BUF_UNLOCK(nbp);
1828 uninitbufbio(nbp);
1829 kfree(nbp, M_DEVBUF);
21864bc5
MD
1830}
1831
1832/*
1833 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1834 */
1835static int
1836devfs_spec_freeblks(struct vop_freeblks_args *ap)
1837{
1838 struct buf *bp;
1839
1840 /*
1841 * XXX: This assumes that strategy does the deed right away.
1842 * XXX: this may not be TRTTD.
1843 */
1844 KKASSERT(ap->a_vp->v_rdev != NULL);
bf390b25 1845 if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
21864bc5
MD
1846 return (0);
1847 bp = geteblk(ap->a_length);
1848 bp->b_cmd = BUF_CMD_FREEBLKS;
1849 bp->b_bio1.bio_offset = ap->a_offset;
1850 bp->b_bcount = ap->a_length;
1851 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1852 return (0);
1853}
1854
1855/*
1856 * Implement degenerate case where the block requested is the block
1857 * returned, and assume that the entire device is contiguous in regards
1858 * to the contiguous block range (runp and runb).
1859 *
1860 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1861 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1862 */
1863static int
1864devfs_spec_bmap(struct vop_bmap_args *ap)
1865{
1866 if (ap->a_doffsetp != NULL)
1867 *ap->a_doffsetp = ap->a_loffset;
1868 if (ap->a_runp != NULL)
1869 *ap->a_runp = MAXBSIZE;
1870 if (ap->a_runb != NULL) {
1871 if (ap->a_loffset < MAXBSIZE)
1872 *ap->a_runb = (int)ap->a_loffset;
1873 else
1874 *ap->a_runb = MAXBSIZE;
1875 }
1876 return (0);
1877}
1878
1879
1880/*
1881 * Special device advisory byte-level locks.
1882 *
1883 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1884 * struct flock *a_fl, int a_flags)
1885 */
1886/* ARGSUSED */
1887static int
1888devfs_spec_advlock(struct vop_advlock_args *ap)
1889{
1890 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1891}
1892
77912481
MD
1893/*
1894 * NOTE: MPSAFE callback.
1895 */
21864bc5
MD
1896static void
1897devfs_spec_getpages_iodone(struct bio *bio)
1898{
1899 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1900 wakeup(bio->bio_buf);
1901}
1902
1903/*
1904 * spec_getpages() - get pages associated with device vnode.
1905 *
1906 * Note that spec_read and spec_write do not use the buffer cache, so we
1907 * must fully implement getpages here.
1908 */
1909static int
1910devfs_spec_getpages(struct vop_getpages_args *ap)
1911{
1912 vm_offset_t kva;
1913 int error;
1914 int i, pcount, size;
1915 struct buf *bp;
1916 vm_page_t m;
1917 vm_ooffset_t offset;
1918 int toff, nextoff, nread;
1919 struct vnode *vp = ap->a_vp;
1920 int blksiz;
1921 int gotreqpage;
1922
1923 error = 0;
1924 pcount = round_page(ap->a_count) / PAGE_SIZE;
1925
1926 /*
1927 * Calculate the offset of the transfer and do sanity check.
1928 */
1929 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1930
1931 /*
1932 * Round up physical size for real devices. We cannot round using
1933 * v_mount's block size data because v_mount has nothing to do with
1934 * the device. i.e. it's usually '/dev'. We need the physical block
1935 * size for the device itself.
1936 *
1937 * We can't use v_rdev->si_mountpoint because it only exists when the
1938 * block device is mounted. However, we can use v_rdev.
1939 */
21864bc5
MD
1940 if (vn_isdisk(vp, NULL))
1941 blksiz = vp->v_rdev->si_bsize_phys;
1942 else
1943 blksiz = DEV_BSIZE;
1944
1945 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1946
ad8b1a17 1947 bp = getpbuf_kva(NULL);
21864bc5
MD
1948 kva = (vm_offset_t)bp->b_data;
1949
1950 /*
1951 * Map the pages to be read into the kva.
1952 */
1953 pmap_qenter(kva, ap->a_m, pcount);
1954
1955 /* Build a minimal buffer header. */
1956 bp->b_cmd = BUF_CMD_READ;
1957 bp->b_bcount = size;
1958 bp->b_resid = 0;
77912481 1959 bsetrunningbufspace(bp, size);
21864bc5
MD
1960
1961 bp->b_bio1.bio_offset = offset;
1962 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1963
1964 mycpu->gd_cnt.v_vnodein++;
1965 mycpu->gd_cnt.v_vnodepgsin += pcount;
1966
1967 /* Do the input. */
1968 vn_strategy(ap->a_vp, &bp->b_bio1);
1969
1970 crit_enter();
1971
1972 /* We definitely need to be at splbio here. */
1973 while (bp->b_cmd != BUF_CMD_DONE)
1974 tsleep(bp, 0, "spread", 0);
1975
1976 crit_exit();
1977
1978 if (bp->b_flags & B_ERROR) {
1979 if (bp->b_error)
1980 error = bp->b_error;
1981 else
1982 error = EIO;
1983 }
1984
1985 /*
1986 * If EOF is encountered we must zero-extend the result in order
1987 * to ensure that the page does not contain garabge. When no
1988 * error occurs, an early EOF is indicated if b_bcount got truncated.
1989 * b_resid is relative to b_bcount and should be 0, but some devices
1990 * might indicate an EOF with b_resid instead of truncating b_bcount.
1991 */
1992 nread = bp->b_bcount - bp->b_resid;
1993 if (nread < ap->a_count)
1994 bzero((caddr_t)kva + nread, ap->a_count - nread);
1995 pmap_qremove(kva, pcount);
1996
1997 gotreqpage = 0;
1998 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
1999 nextoff = toff + PAGE_SIZE;
2000 m = ap->a_m[i];
2001
2002 m->flags &= ~PG_ZERO;
2003
cb1cf930
MD
2004 /*
2005 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2006 * pmap modified bit. pmap modified bit should have
2007 * already been cleared.
2008 */
21864bc5
MD
2009 if (nextoff <= nread) {
2010 m->valid = VM_PAGE_BITS_ALL;
2011 vm_page_undirty(m);
2012 } else if (toff < nread) {
2013 /*
2014 * Since this is a VM request, we have to supply the
cb1cf930 2015 * unaligned offset to allow vm_page_set_valid()
21864bc5
MD
2016 * to zero sub-DEV_BSIZE'd portions of the page.
2017 */
1a54183b
MD
2018 vm_page_set_valid(m, 0, nread - toff);
2019 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
21864bc5
MD
2020 } else {
2021 m->valid = 0;
2022 vm_page_undirty(m);
2023 }
2024
2025 if (i != ap->a_reqpage) {
2026 /*
2027 * Just in case someone was asking for this page we
2028 * now tell them that it is ok to use.
2029 */
2030 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2031 if (m->valid) {
2032 if (m->flags & PG_WANTED) {
2033 vm_page_activate(m);
2034 } else {
2035 vm_page_deactivate(m);
2036 }
2037 vm_page_wakeup(m);
2038 } else {
2039 vm_page_free(m);
2040 }
2041 } else {
2042 vm_page_free(m);
2043 }
2044 } else if (m->valid) {
2045 gotreqpage = 1;
2046 /*
2047 * Since this is a VM request, we need to make the
2048 * entire page presentable by zeroing invalid sections.
2049 */
2050 if (m->valid != VM_PAGE_BITS_ALL)
2051 vm_page_zero_invalid(m, FALSE);
2052 }
2053 }
2054 if (!gotreqpage) {
2055 m = ap->a_m[ap->a_reqpage];
2056 devfs_debug(DEVFS_DEBUG_WARNING,
2057 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2058 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2059 devfs_debug(DEVFS_DEBUG_WARNING,
2060 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2061 size, bp->b_resid, ap->a_count, m->valid);
2062 devfs_debug(DEVFS_DEBUG_WARNING,
2063 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2064 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2065 /*
2066 * Free the buffer header back to the swap buffer pool.
2067 */
2068 relpbuf(bp, NULL);
2069 return VM_PAGER_ERROR;
2070 }
2071 /*
2072 * Free the buffer header back to the swap buffer pool.
2073 */
2074 relpbuf(bp, NULL);
07dfa375
AH
2075 if (DEVFS_NODE(ap->a_vp))
2076 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
21864bc5
MD
2077 return VM_PAGER_OK;
2078}
2079
21864bc5
MD
2080static __inline
2081int
2082sequential_heuristic(struct uio *uio, struct file *fp)
2083{
2084 /*
2085 * Sequential heuristic - detect sequential operation
2086 */
2087 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2088 uio->uio_offset == fp->f_nextoff) {
21864bc5
MD
2089 /*
2090 * XXX we assume that the filesystem block size is
2091 * the default. Not true, but still gives us a pretty
2092 * good indicator of how sequential the read operations
2093 * are.
2094 */
898c91ee
MD
2095 int tmpseq = fp->f_seqcount;
2096
21864bc5
MD
2097 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2098 if (tmpseq > IO_SEQMAX)
2099 tmpseq = IO_SEQMAX;
2100 fp->f_seqcount = tmpseq;
2101 return(fp->f_seqcount << IO_SEQSHIFT);
2102 }
2103
2104 /*
2105 * Not sequential, quick draw-down of seqcount
2106 */
2107 if (fp->f_seqcount > 1)
2108 fp->f_seqcount = 1;
2109 else
2110 fp->f_seqcount = 0;
2111 return(0);
2112}