4 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
6 * This code is derived from software contributed to The DragonFly Project
7 * by Alex Hornung <ahornung@gmail.com>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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
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.
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
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/kernel.h>
41 #include <sys/fcntl.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
47 #include <sys/mount.h>
49 #include <sys/fcntl.h>
50 #include <sys/namei.h>
51 #include <sys/dirent.h>
52 #include <sys/malloc.h>
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>
61 #include <sys/diskslice.h>
62 #include <sys/sysctl.h>
63 #include <sys/devfs.h>
64 #include <sys/pioctl.h>
65 #include <vfs/fifofs/fifo.h>
67 #include <machine/limits.h>
70 #include <sys/sysref2.h>
71 #include <sys/mplock2.h>
72 #include <vm/vm_page2.h>
74 #ifndef SPEC_CHAIN_DEBUG
75 #define SPEC_CHAIN_DEBUG 0
78 MALLOC_DECLARE(M_DEVFS);
79 #define DEVFS_BADOP (void *)devfs_vop_badop
81 static int devfs_vop_badop(struct vop_generic_args *);
82 static int devfs_vop_access(struct vop_access_args *);
83 static int devfs_vop_inactive(struct vop_inactive_args *);
84 static int devfs_vop_reclaim(struct vop_reclaim_args *);
85 static int devfs_vop_readdir(struct vop_readdir_args *);
86 static int devfs_vop_getattr(struct vop_getattr_args *);
87 static int devfs_vop_setattr(struct vop_setattr_args *);
88 static int devfs_vop_readlink(struct vop_readlink_args *);
89 static int devfs_vop_print(struct vop_print_args *);
91 static int devfs_vop_nresolve(struct vop_nresolve_args *);
92 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
93 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
94 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
95 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
96 static int devfs_vop_nremove(struct vop_nremove_args *);
98 static int devfs_spec_open(struct vop_open_args *);
99 static int devfs_spec_close(struct vop_close_args *);
100 static int devfs_spec_fsync(struct vop_fsync_args *);
102 static int devfs_spec_read(struct vop_read_args *);
103 static int devfs_spec_write(struct vop_write_args *);
104 static int devfs_spec_ioctl(struct vop_ioctl_args *);
105 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
106 static int devfs_spec_strategy(struct vop_strategy_args *);
107 static void devfs_spec_strategy_done(struct bio *);
108 static int devfs_spec_freeblks(struct vop_freeblks_args *);
109 static int devfs_spec_bmap(struct vop_bmap_args *);
110 static int devfs_spec_advlock(struct vop_advlock_args *);
111 static void devfs_spec_getpages_iodone(struct bio *);
112 static int devfs_spec_getpages(struct vop_getpages_args *);
114 static int devfs_fo_close(struct file *);
115 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
116 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
117 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
118 static int devfs_fo_kqfilter(struct file *, struct knote *);
119 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
120 struct ucred *, struct sysmsg *);
121 static __inline int sequential_heuristic(struct uio *, struct file *);
123 extern struct lock devfs_lock;
126 * devfs vnode operations for regular files. All vnode ops are MPSAFE.
128 struct vop_ops devfs_vnode_norm_vops = {
129 .vop_default = vop_defaultop,
130 .vop_access = devfs_vop_access,
131 .vop_advlock = DEVFS_BADOP,
132 .vop_bmap = DEVFS_BADOP,
133 .vop_close = vop_stdclose,
134 .vop_getattr = devfs_vop_getattr,
135 .vop_inactive = devfs_vop_inactive,
136 .vop_ncreate = DEVFS_BADOP,
137 .vop_nresolve = devfs_vop_nresolve,
138 .vop_nlookupdotdot = devfs_vop_nlookupdotdot,
139 .vop_nlink = DEVFS_BADOP,
140 .vop_nmkdir = devfs_vop_nmkdir,
141 .vop_nmknod = DEVFS_BADOP,
142 .vop_nremove = devfs_vop_nremove,
143 .vop_nrename = DEVFS_BADOP,
144 .vop_nrmdir = devfs_vop_nrmdir,
145 .vop_nsymlink = devfs_vop_nsymlink,
146 .vop_open = vop_stdopen,
147 .vop_pathconf = vop_stdpathconf,
148 .vop_print = devfs_vop_print,
149 .vop_read = DEVFS_BADOP,
150 .vop_readdir = devfs_vop_readdir,
151 .vop_readlink = devfs_vop_readlink,
152 .vop_reclaim = devfs_vop_reclaim,
153 .vop_setattr = devfs_vop_setattr,
154 .vop_write = DEVFS_BADOP,
155 .vop_ioctl = DEVFS_BADOP
159 * devfs vnode operations for character devices. All vnode ops are MPSAFE.
161 struct vop_ops devfs_vnode_dev_vops = {
162 .vop_default = vop_defaultop,
163 .vop_access = devfs_vop_access,
164 .vop_advlock = devfs_spec_advlock,
165 .vop_bmap = devfs_spec_bmap,
166 .vop_close = devfs_spec_close,
167 .vop_freeblks = devfs_spec_freeblks,
168 .vop_fsync = devfs_spec_fsync,
169 .vop_getattr = devfs_vop_getattr,
170 .vop_getpages = devfs_spec_getpages,
171 .vop_inactive = devfs_vop_inactive,
172 .vop_open = devfs_spec_open,
173 .vop_pathconf = vop_stdpathconf,
174 .vop_print = devfs_vop_print,
175 .vop_kqfilter = devfs_spec_kqfilter,
176 .vop_read = devfs_spec_read,
177 .vop_readdir = DEVFS_BADOP,
178 .vop_readlink = DEVFS_BADOP,
179 .vop_reclaim = devfs_vop_reclaim,
180 .vop_setattr = devfs_vop_setattr,
181 .vop_strategy = devfs_spec_strategy,
182 .vop_write = devfs_spec_write,
183 .vop_ioctl = devfs_spec_ioctl
187 * devfs file pointer operations. All fileops are MPSAFE.
189 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
191 struct fileops devfs_dev_fileops = {
192 .fo_read = devfs_fo_read,
193 .fo_write = devfs_fo_write,
194 .fo_ioctl = devfs_fo_ioctl,
195 .fo_kqfilter = devfs_fo_kqfilter,
196 .fo_stat = devfs_fo_stat,
197 .fo_close = devfs_fo_close,
198 .fo_shutdown = nofo_shutdown
202 * These two functions are possibly temporary hacks for devices (aka
203 * the pty code) which want to control the node attributes themselves.
205 * XXX we may ultimately desire to simply remove the uid/gid/mode
206 * from the node entirely.
208 * MPSAFE - sorta. Theoretically the overwrite can compete since they
209 * are loading from the same fields.
212 node_sync_dev_get(struct devfs_node *node)
216 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
217 node->uid = dev->si_uid;
218 node->gid = dev->si_gid;
219 node->mode = dev->si_perms;
224 node_sync_dev_set(struct devfs_node *node)
228 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
229 dev->si_uid = node->uid;
230 dev->si_gid = node->gid;
231 dev->si_perms = node->mode;
236 * generic entry point for unsupported operations
239 devfs_vop_badop(struct vop_generic_args *ap)
246 devfs_vop_access(struct vop_access_args *ap)
248 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
251 if (!devfs_node_is_accessible(node))
253 node_sync_dev_get(node);
254 error = vop_helper_access(ap, node->uid, node->gid,
255 node->mode, node->flags);
262 devfs_vop_inactive(struct vop_inactive_args *ap)
264 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
266 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
273 devfs_vop_reclaim(struct vop_reclaim_args *ap)
275 struct devfs_node *node;
280 * Check if it is locked already. if not, we acquire the devfs lock
282 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
283 lockmgr(&devfs_lock, LK_EXCLUSIVE);
290 * Get rid of the devfs_node if it is no longer linked into the
294 if ((node = DEVFS_NODE(vp)) != NULL) {
296 if ((node->flags & DEVFS_NODE_LINKED) == 0)
301 lockmgr(&devfs_lock, LK_RELEASE);
304 * v_rdev needs to be properly released using v_release_rdev
305 * Make sure v_data is NULL as well.
314 devfs_vop_readdir(struct vop_readdir_args *ap)
316 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
317 struct devfs_node *node;
326 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
328 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
330 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
333 if (!devfs_node_is_accessible(dnode)) {
338 lockmgr(&devfs_lock, LK_EXCLUSIVE);
340 saveoff = ap->a_uio->uio_offset;
342 if (ap->a_ncookies) {
343 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
346 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
354 nanotime(&dnode->atime);
357 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
362 cookies[cookie_index] = saveoff;
365 if (cookie_index == ncookies)
371 r = vop_write_dirent(&error, ap->a_uio,
372 dnode->parent->d_dir.d_ino,
375 r = vop_write_dirent(&error, ap->a_uio,
382 cookies[cookie_index] = saveoff;
385 if (cookie_index == ncookies)
389 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
390 if ((node->flags & DEVFS_HIDDEN) ||
391 (node->flags & DEVFS_INVISIBLE)) {
396 * If the node type is a valid devfs alias, then we make
397 * sure that the target isn't hidden. If it is, we don't
398 * show the link in the directory listing.
400 if ((node->node_type == Nlink) && (node->link_target != NULL) &&
401 (node->link_target->flags & DEVFS_HIDDEN))
404 if (node->cookie < saveoff)
407 saveoff = node->cookie;
409 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
411 node->d_dir.d_namlen,
420 cookies[cookie_index] = node->cookie;
422 if (cookie_index == ncookies)
427 lockmgr(&devfs_lock, LK_RELEASE);
430 ap->a_uio->uio_offset = saveoff;
431 if (error && cookie_index == 0) {
433 kfree(cookies, M_TEMP);
435 *ap->a_cookies = NULL;
439 *ap->a_ncookies = cookie_index;
440 *ap->a_cookies = cookies;
448 devfs_vop_nresolve(struct vop_nresolve_args *ap)
450 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
451 struct devfs_node *node, *found = NULL;
452 struct namecache *ncp;
453 struct vnode *vp = NULL;
458 ncp = ap->a_nch->ncp;
461 if (!devfs_node_is_accessible(dnode))
464 lockmgr(&devfs_lock, LK_EXCLUSIVE);
466 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
468 cache_setvp(ap->a_nch, NULL);
472 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
473 if (len == node->d_dir.d_namlen) {
474 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
483 while ((found->node_type == Nlink) && (found->link_target)) {
485 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
489 found = found->link_target;
493 if (!(found->flags & DEVFS_HIDDEN))
494 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
499 cache_setvp(ap->a_nch, NULL);
505 cache_setvp(ap->a_nch, vp);
508 lockmgr(&devfs_lock, LK_RELEASE);
515 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
517 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
520 if (!devfs_node_is_accessible(dnode))
523 lockmgr(&devfs_lock, LK_EXCLUSIVE);
524 if (dnode->parent != NULL) {
525 devfs_allocv(ap->a_vpp, dnode->parent);
526 vn_unlock(*ap->a_vpp);
528 lockmgr(&devfs_lock, LK_RELEASE);
530 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
535 devfs_vop_getattr(struct vop_getattr_args *ap)
537 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
538 struct vattr *vap = ap->a_vap;
539 struct partinfo pinfo;
543 if (!devfs_node_is_accessible(node))
546 node_sync_dev_get(node);
548 lockmgr(&devfs_lock, LK_EXCLUSIVE);
550 /* start by zeroing out the attributes */
553 /* next do all the common fields */
554 vap->va_type = ap->a_vp->v_type;
555 vap->va_mode = node->mode;
556 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
558 vap->va_blocksize = DEV_BSIZE;
559 vap->va_bytes = vap->va_size = 0;
561 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
563 vap->va_atime = node->atime;
564 vap->va_mtime = node->mtime;
565 vap->va_ctime = node->ctime;
567 vap->va_nlink = 1; /* number of references to file */
569 vap->va_uid = node->uid;
570 vap->va_gid = node->gid;
575 if ((node->node_type == Ndev) && node->d_dev) {
576 reference_dev(node->d_dev);
577 vap->va_rminor = node->d_dev->si_uminor;
578 release_dev(node->d_dev);
581 /* For a softlink the va_size is the length of the softlink */
582 if (node->symlink_name != 0) {
583 vap->va_bytes = vap->va_size = node->symlink_namelen;
587 * For a disk-type device, va_size is the size of the underlying
588 * device, so that lseek() works properly.
590 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
591 bzero(&pinfo, sizeof(pinfo));
592 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
593 0, proc0.p_ucred, NULL);
594 if ((error == 0) && (pinfo.media_blksize != 0)) {
595 vap->va_size = pinfo.media_size;
602 lockmgr(&devfs_lock, LK_RELEASE);
609 devfs_vop_setattr(struct vop_setattr_args *ap)
611 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
618 if (!devfs_node_is_accessible(node))
620 node_sync_dev_get(node);
622 lockmgr(&devfs_lock, LK_EXCLUSIVE);
626 if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
629 cur_mode = node->mode;
630 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
631 ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
635 if (node->uid != cur_uid || node->gid != cur_gid) {
638 node->mode = cur_mode;
642 if (vap->va_mode != (mode_t)VNOVAL) {
643 cur_mode = node->mode;
644 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
645 node->uid, node->gid, &cur_mode);
646 if (error == 0 && node->mode != cur_mode) {
647 node->mode = cur_mode;
652 node_sync_dev_set(node);
653 nanotime(&node->ctime);
654 lockmgr(&devfs_lock, LK_RELEASE);
661 devfs_vop_readlink(struct vop_readlink_args *ap)
663 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
666 if (!devfs_node_is_accessible(node))
669 lockmgr(&devfs_lock, LK_EXCLUSIVE);
670 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
671 lockmgr(&devfs_lock, LK_RELEASE);
678 devfs_vop_print(struct vop_print_args *ap)
684 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
686 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
687 struct devfs_node *node;
689 if (!devfs_node_is_accessible(dnode))
692 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
695 lockmgr(&devfs_lock, LK_EXCLUSIVE);
696 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
697 ap->a_nch->ncp->nc_name, dnode, NULL);
700 node = DEVFS_NODE(*ap->a_vpp);
701 node->flags |= DEVFS_USER_CREATED;
702 cache_setunresolved(ap->a_nch);
703 cache_setvp(ap->a_nch, *ap->a_vpp);
705 lockmgr(&devfs_lock, LK_RELEASE);
707 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
711 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
713 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
714 struct devfs_node *node;
717 if (!devfs_node_is_accessible(dnode))
720 ap->a_vap->va_type = VLNK;
722 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
725 lockmgr(&devfs_lock, LK_EXCLUSIVE);
726 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
727 ap->a_nch->ncp->nc_name, dnode, NULL);
729 targetlen = strlen(ap->a_target);
731 node = DEVFS_NODE(*ap->a_vpp);
732 node->flags |= DEVFS_USER_CREATED;
733 node->symlink_namelen = targetlen;
734 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
735 memcpy(node->symlink_name, ap->a_target, targetlen);
736 node->symlink_name[targetlen] = '\0';
737 cache_setunresolved(ap->a_nch);
738 cache_setvp(ap->a_nch, *ap->a_vpp);
740 lockmgr(&devfs_lock, LK_RELEASE);
742 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
746 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
748 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
749 struct devfs_node *node;
750 struct namecache *ncp;
753 ncp = ap->a_nch->ncp;
755 if (!devfs_node_is_accessible(dnode))
758 lockmgr(&devfs_lock, LK_EXCLUSIVE);
760 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
763 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
764 if (ncp->nc_nlen != node->d_dir.d_namlen)
766 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
770 * only allow removal of user created dirs
772 if ((node->flags & DEVFS_USER_CREATED) == 0) {
775 } else if (node->node_type != Ndir) {
778 } else if (node->nchildren > 2) {
783 cache_inval_vp(node->v_node, CINV_DESTROY);
790 cache_unlink(ap->a_nch);
792 lockmgr(&devfs_lock, LK_RELEASE);
797 devfs_vop_nremove(struct vop_nremove_args *ap)
799 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
800 struct devfs_node *node;
801 struct namecache *ncp;
804 ncp = ap->a_nch->ncp;
806 if (!devfs_node_is_accessible(dnode))
809 lockmgr(&devfs_lock, LK_EXCLUSIVE);
811 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
814 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
815 if (ncp->nc_nlen != node->d_dir.d_namlen)
817 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
821 * only allow removal of user created stuff (e.g. symlinks)
823 if ((node->flags & DEVFS_USER_CREATED) == 0) {
826 } else if (node->node_type == Ndir) {
831 cache_inval_vp(node->v_node, CINV_DESTROY);
838 cache_unlink(ap->a_nch);
840 lockmgr(&devfs_lock, LK_RELEASE);
846 devfs_spec_open(struct vop_open_args *ap)
848 struct vnode *vp = ap->a_vp;
849 struct vnode *orig_vp = NULL;
850 struct devfs_node *node = DEVFS_NODE(vp);
851 struct devfs_node *newnode;
852 cdev_t dev, ndev = NULL;
856 if (node->d_dev == NULL)
858 if (!devfs_node_is_accessible(node))
862 if ((dev = vp->v_rdev) == NULL)
865 if (node && ap->a_fp) {
866 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
867 lockmgr(&devfs_lock, LK_EXCLUSIVE);
869 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
870 ap->a_mode, ap->a_cred);
872 newnode = devfs_create_device_node(
873 DEVFS_MNTDATA(vp->v_mount)->root_node,
875 /* XXX: possibly destroy device if this happens */
877 if (newnode != NULL) {
881 devfs_debug(DEVFS_DEBUG_DEBUG,
882 "parent here is: %s, node is: |%s|\n",
883 ((node->parent->node_type == Nroot) ?
884 "ROOT!" : node->parent->d_dir.d_name),
885 newnode->d_dir.d_name);
886 devfs_debug(DEVFS_DEBUG_DEBUG,
888 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
891 * orig_vp is set to the original vp if we cloned.
893 /* node->flags |= DEVFS_CLONED; */
894 devfs_allocv(&vp, newnode);
899 lockmgr(&devfs_lock, LK_RELEASE);
902 devfs_debug(DEVFS_DEBUG_DEBUG,
903 "devfs_spec_open() called on %s! \n",
907 * Make this field valid before any I/O in ->d_open
909 if (!dev->si_iosize_max)
910 /* XXX: old DFLTPHYS == 64KB dependency */
911 dev->si_iosize_max = min(MAXPHYS,64*1024);
913 if (dev_dflags(dev) & D_TTY)
914 vsetflags(vp, VISTTY);
917 * Open underlying device
920 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
921 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
924 * Clean up any cloned vp if we error out.
930 /* orig_vp = NULL; */
936 * This checks if the disk device is going to be opened for writing.
937 * It will be only allowed in the cases where securelevel permits it
938 * and it's not mounted R/W.
940 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
941 (ap->a_cred != FSCRED)) {
943 /* Very secure mode. No open for writing allowed */
944 if (securelevel >= 2)
948 * If it is mounted R/W, do not allow to open for writing.
949 * In the case it's mounted read-only but securelevel
950 * is >= 1, then do not allow opening for writing either.
952 if (vfs_mountedon(vp)) {
953 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
955 else if (securelevel >= 1)
960 if (dev_dflags(dev) & D_TTY) {
965 devfs_debug(DEVFS_DEBUG_DEBUG,
966 "devfs: no t_stop\n");
967 tp->t_stop = nottystop;
973 if (vn_isdisk(vp, NULL)) {
974 if (!dev->si_bsize_phys)
975 dev->si_bsize_phys = DEV_BSIZE;
976 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
982 nanotime(&node->atime);
986 * If we replaced the vp the vop_stdopen() call will have loaded
987 * it into fp->f_data and vref()d the vp, giving us two refs. So
988 * instead of just unlocking it here we have to vput() it.
993 /* Ugly pty magic, to make pty devices appear once they are opened */
994 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
995 node->flags &= ~DEVFS_INVISIBLE;
998 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
999 KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1000 ap->a_fp->f_ops = &devfs_dev_fileops;
1001 KKASSERT(ap->a_fp->f_data == (void *)vp);
1009 devfs_spec_close(struct vop_close_args *ap)
1011 struct devfs_node *node;
1012 struct proc *p = curproc;
1013 struct vnode *vp = ap->a_vp;
1014 cdev_t dev = vp->v_rdev;
1019 devfs_debug(DEVFS_DEBUG_DEBUG,
1020 "devfs_spec_close() called on %s! \n",
1023 devfs_debug(DEVFS_DEBUG_DEBUG,
1024 "devfs_spec_close() called, null vode!\n");
1027 * A couple of hacks for devices and tty devices. The
1028 * vnode ref count cannot be used to figure out the
1029 * last close, but we can use v_opencount now that
1030 * revoke works properly.
1032 * Detect the last close on a controlling terminal and clear
1033 * the session (half-close).
1038 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1039 p->p_session->s_ttyvp = NULL;
1044 * Vnodes can be opened and closed multiple times. Do not really
1045 * close the device unless (1) it is being closed forcibly,
1046 * (2) the device wants to track closes, or (3) this is the last
1047 * vnode doing its last close on the device.
1049 * XXX the VXLOCK (force close) case can leave vnodes referencing
1050 * a closed device. This might not occur now that our revoke is
1053 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1054 if (dev && ((vp->v_flag & VRECLAIMED) ||
1055 (dev_dflags(dev) & D_TRACKCLOSE) ||
1056 (vp->v_opencount == 1))) {
1058 * Ugly pty magic, to make pty devices disappear again once
1061 node = DEVFS_NODE(ap->a_vp);
1062 if (node && (node->flags & DEVFS_PTY))
1063 node->flags |= DEVFS_INVISIBLE;
1066 * Unlock around dev_dclose(), unless the vnode is
1067 * undergoing a vgone/reclaim (during umount).
1070 if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1076 * WARNING! If the device destroys itself the devfs node
1077 * can disappear here.
1079 * WARNING! vn_lock() will fail if the vp is in a VRECLAIM,
1080 * which can occur during umount.
1082 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1083 /* node is now stale */
1086 if (vn_lock(vp, LK_EXCLUSIVE | LK_RETRY) != 0) {
1087 panic("devfs_spec_close: vnode %p "
1088 "unexpectedly could not be relocked",
1095 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1098 * Track the actual opens and closes on the vnode. The last close
1099 * disassociates the rdev. If the rdev is already disassociated or
1100 * the opencount is already 0, the vnode might have been revoked
1101 * and no further opencount tracking occurs.
1105 if (vp->v_opencount > 0)
1113 devfs_fo_close(struct file *fp)
1115 struct vnode *vp = (struct vnode *)fp->f_data;
1118 fp->f_ops = &badfileops;
1119 error = vn_close(vp, fp->f_flag);
1126 * Device-optimized file table vnode read routine.
1128 * This bypasses the VOP table and talks directly to the device. Most
1129 * filesystems just route to specfs and can make this optimization.
1131 * MPALMOSTSAFE - acquires mplock
1134 devfs_fo_read(struct file *fp, struct uio *uio,
1135 struct ucred *cred, int flags)
1137 struct devfs_node *node;
1143 KASSERT(uio->uio_td == curthread,
1144 ("uio_td %p is not td %p", uio->uio_td, curthread));
1146 if (uio->uio_resid == 0)
1149 vp = (struct vnode *)fp->f_data;
1150 if (vp == NULL || vp->v_type == VBAD)
1153 node = DEVFS_NODE(vp);
1155 if ((dev = vp->v_rdev) == NULL)
1160 if ((flags & O_FOFFSET) == 0)
1161 uio->uio_offset = fp->f_offset;
1164 if (flags & O_FBLOCKING) {
1165 /* ioflag &= ~IO_NDELAY; */
1166 } else if (flags & O_FNONBLOCKING) {
1167 ioflag |= IO_NDELAY;
1168 } else if (fp->f_flag & FNONBLOCK) {
1169 ioflag |= IO_NDELAY;
1171 if (flags & O_FBUFFERED) {
1172 /* ioflag &= ~IO_DIRECT; */
1173 } else if (flags & O_FUNBUFFERED) {
1174 ioflag |= IO_DIRECT;
1175 } else if (fp->f_flag & O_DIRECT) {
1176 ioflag |= IO_DIRECT;
1178 ioflag |= sequential_heuristic(uio, fp);
1180 error = dev_dread(dev, uio, ioflag);
1184 nanotime(&node->atime);
1185 if ((flags & O_FOFFSET) == 0)
1186 fp->f_offset = uio->uio_offset;
1187 fp->f_nextoff = uio->uio_offset;
1194 devfs_fo_write(struct file *fp, struct uio *uio,
1195 struct ucred *cred, int flags)
1197 struct devfs_node *node;
1203 KASSERT(uio->uio_td == curthread,
1204 ("uio_td %p is not p %p", uio->uio_td, curthread));
1206 vp = (struct vnode *)fp->f_data;
1207 if (vp == NULL || vp->v_type == VBAD)
1210 node = DEVFS_NODE(vp);
1212 if (vp->v_type == VREG)
1213 bwillwrite(uio->uio_resid);
1215 vp = (struct vnode *)fp->f_data;
1217 if ((dev = vp->v_rdev) == NULL)
1222 if ((flags & O_FOFFSET) == 0)
1223 uio->uio_offset = fp->f_offset;
1226 if (vp->v_type == VREG &&
1227 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1228 ioflag |= IO_APPEND;
1231 if (flags & O_FBLOCKING) {
1232 /* ioflag &= ~IO_NDELAY; */
1233 } else if (flags & O_FNONBLOCKING) {
1234 ioflag |= IO_NDELAY;
1235 } else if (fp->f_flag & FNONBLOCK) {
1236 ioflag |= IO_NDELAY;
1238 if (flags & O_FBUFFERED) {
1239 /* ioflag &= ~IO_DIRECT; */
1240 } else if (flags & O_FUNBUFFERED) {
1241 ioflag |= IO_DIRECT;
1242 } else if (fp->f_flag & O_DIRECT) {
1243 ioflag |= IO_DIRECT;
1245 if (flags & O_FASYNCWRITE) {
1246 /* ioflag &= ~IO_SYNC; */
1247 } else if (flags & O_FSYNCWRITE) {
1249 } else if (fp->f_flag & O_FSYNC) {
1253 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1255 ioflag |= sequential_heuristic(uio, fp);
1257 error = dev_dwrite(dev, uio, ioflag);
1261 nanotime(&node->atime);
1262 nanotime(&node->mtime);
1265 if ((flags & O_FOFFSET) == 0)
1266 fp->f_offset = uio->uio_offset;
1267 fp->f_nextoff = uio->uio_offset;
1274 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1283 vp = (struct vnode *)fp->f_data;
1284 if (vp == NULL || vp->v_type == VBAD)
1287 error = vn_stat(vp, sb, cred);
1292 error = VOP_GETATTR(vp, vap);
1297 * Zero the spare stat fields
1304 * Copy from vattr table ... or not in case it's a cloned device
1306 if (vap->va_fsid != VNOVAL)
1307 sb->st_dev = vap->va_fsid;
1309 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1311 sb->st_ino = vap->va_fileid;
1313 mode = vap->va_mode;
1317 if (vap->va_nlink > (nlink_t)-1)
1318 sb->st_nlink = (nlink_t)-1;
1320 sb->st_nlink = vap->va_nlink;
1322 sb->st_uid = vap->va_uid;
1323 sb->st_gid = vap->va_gid;
1324 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1325 sb->st_size = vap->va_bytes;
1326 sb->st_atimespec = vap->va_atime;
1327 sb->st_mtimespec = vap->va_mtime;
1328 sb->st_ctimespec = vap->va_ctime;
1331 * A VCHR and VBLK device may track the last access and last modified
1332 * time independantly of the filesystem. This is particularly true
1333 * because device read and write calls may bypass the filesystem.
1335 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1338 if (dev->si_lastread) {
1339 sb->st_atimespec.tv_sec = dev->si_lastread;
1340 sb->st_atimespec.tv_nsec = 0;
1342 if (dev->si_lastwrite) {
1343 sb->st_atimespec.tv_sec = dev->si_lastwrite;
1344 sb->st_atimespec.tv_nsec = 0;
1350 * According to www.opengroup.org, the meaning of st_blksize is
1351 * "a filesystem-specific preferred I/O block size for this
1352 * object. In some filesystem types, this may vary from file
1354 * Default to PAGE_SIZE after much discussion.
1357 sb->st_blksize = PAGE_SIZE;
1359 sb->st_flags = vap->va_flags;
1361 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1365 sb->st_gen = (u_int32_t)vap->va_gen;
1367 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1374 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1380 vp = (struct vnode *)fp->f_data;
1381 if (vp == NULL || vp->v_type == VBAD) {
1385 if ((dev = vp->v_rdev) == NULL) {
1391 error = dev_dkqfilter(dev, kn);
1400 * MPALMOSTSAFE - acquires mplock
1403 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1404 struct ucred *ucred, struct sysmsg *msg)
1407 struct devfs_node *node;
1413 struct fiodname_args *name_args;
1417 vp = ((struct vnode *)fp->f_data);
1419 if ((dev = vp->v_rdev) == NULL)
1420 return EBADF; /* device was revoked */
1425 node = DEVFS_NODE(vp);
1428 devfs_debug(DEVFS_DEBUG_DEBUG,
1429 "devfs_fo_ioctl() called! for dev %s\n",
1432 if (com == FIODTYPE) {
1433 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1436 } else if (com == FIODNAME) {
1437 name_args = (struct fiodname_args *)data;
1438 name = dev->si_name;
1439 namlen = strlen(name) + 1;
1441 devfs_debug(DEVFS_DEBUG_DEBUG,
1442 "ioctl, got: FIODNAME for %s\n", name);
1444 if (namlen <= name_args->len)
1445 error = copyout(dev->si_name, name_args->name, namlen);
1449 devfs_debug(DEVFS_DEBUG_DEBUG,
1450 "ioctl stuff: error: %d\n", error);
1454 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1458 nanotime(&node->atime);
1459 nanotime(&node->mtime);
1462 if (com == TIOCSCTTY) {
1463 devfs_debug(DEVFS_DEBUG_DEBUG,
1464 "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1467 if (error == 0 && com == TIOCSCTTY) {
1468 struct proc *p = curthread->td_proc;
1469 struct session *sess;
1471 devfs_debug(DEVFS_DEBUG_DEBUG,
1472 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1478 sess = p->p_session;
1481 * Do nothing if reassigning same control tty
1483 if (sess->s_ttyvp == vp) {
1489 * Get rid of reference to old control tty
1491 ovp = sess->s_ttyvp;
1500 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1506 devfs_spec_fsync(struct vop_fsync_args *ap)
1508 struct vnode *vp = ap->a_vp;
1511 if (!vn_isdisk(vp, NULL))
1515 * Flush all dirty buffers associated with a block device.
1517 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1522 devfs_spec_read(struct vop_read_args *ap)
1524 struct devfs_node *node;
1533 node = DEVFS_NODE(vp);
1535 if (dev == NULL) /* device was revoked */
1537 if (uio->uio_resid == 0)
1541 error = dev_dread(dev, uio, ap->a_ioflag);
1542 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1545 nanotime(&node->atime);
1551 * Vnode op for write
1553 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1554 * struct ucred *a_cred)
1557 devfs_spec_write(struct vop_write_args *ap)
1559 struct devfs_node *node;
1568 node = DEVFS_NODE(vp);
1570 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1572 if (dev == NULL) /* device was revoked */
1576 error = dev_dwrite(dev, uio, ap->a_ioflag);
1577 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1580 nanotime(&node->atime);
1581 nanotime(&node->mtime);
1588 * Device ioctl operation.
1590 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1591 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1594 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1596 struct vnode *vp = ap->a_vp;
1598 struct devfs_node *node;
1602 if ((dev = vp->v_rdev) == NULL)
1603 return (EBADF); /* device was revoked */
1605 node = DEVFS_NODE(vp);
1608 nanotime(&node->atime);
1609 nanotime(&node->mtime);
1613 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1614 ap->a_cred, ap->a_sysmsg));
1618 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1622 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1624 struct vnode *vp = ap->a_vp;
1626 struct devfs_node *node;
1630 if ((dev = vp->v_rdev) == NULL)
1631 return (EBADF); /* device was revoked (EBADF) */
1633 node = DEVFS_NODE(vp);
1636 nanotime(&node->atime);
1639 return (dev_dkqfilter(dev, ap->a_kn));
1643 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1644 * calls are not limited to device DMA limits so we have to deal with the
1647 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1650 devfs_spec_strategy(struct vop_strategy_args *ap)
1652 struct bio *bio = ap->a_bio;
1653 struct buf *bp = bio->bio_buf;
1660 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1664 * Collect statistics on synchronous and asynchronous read
1665 * and write counts for disks that have associated filesystems.
1668 KKASSERT(vp->v_rdev != NULL); /* XXX */
1669 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1670 if (bp->b_cmd == BUF_CMD_READ) {
1671 if (bp->b_flags & BIO_SYNC)
1672 mp->mnt_stat.f_syncreads++;
1674 mp->mnt_stat.f_asyncreads++;
1676 if (bp->b_flags & BIO_SYNC)
1677 mp->mnt_stat.f_syncwrites++;
1679 mp->mnt_stat.f_asyncwrites++;
1684 * Device iosize limitations only apply to read and write. Shortcut
1685 * the I/O if it fits.
1687 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1688 devfs_debug(DEVFS_DEBUG_DEBUG,
1689 "%s: si_iosize_max not set!\n",
1690 dev_dname(vp->v_rdev));
1691 maxiosize = MAXPHYS;
1693 #if SPEC_CHAIN_DEBUG & 2
1696 if (bp->b_bcount <= maxiosize ||
1697 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1698 dev_dstrategy_chain(vp->v_rdev, bio);
1703 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1705 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1708 BUF_LOCK(nbp, LK_EXCLUSIVE);
1711 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1712 nbp->b_data = bp->b_data;
1713 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1714 nbp->b_bio1.bio_offset = bio->bio_offset;
1715 nbp->b_bio1.bio_caller_info1.ptr = bio;
1718 * Start the first transfer
1720 if (vn_isdisk(vp, NULL))
1721 chunksize = vp->v_rdev->si_bsize_phys;
1723 chunksize = DEV_BSIZE;
1724 chunksize = maxiosize / chunksize * chunksize;
1725 #if SPEC_CHAIN_DEBUG & 1
1726 devfs_debug(DEVFS_DEBUG_DEBUG,
1727 "spec_strategy chained I/O chunksize=%d\n",
1730 nbp->b_cmd = bp->b_cmd;
1731 nbp->b_bcount = chunksize;
1732 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1733 nbp->b_bio1.bio_caller_info2.index = chunksize;
1735 #if SPEC_CHAIN_DEBUG & 1
1736 devfs_debug(DEVFS_DEBUG_DEBUG,
1737 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1738 bp, 0, bp->b_bcount, nbp->b_bcount);
1741 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1743 if (DEVFS_NODE(vp)) {
1744 nanotime(&DEVFS_NODE(vp)->atime);
1745 nanotime(&DEVFS_NODE(vp)->mtime);
1752 * Chunked up transfer completion routine - chain transfers until done
1754 * NOTE: MPSAFE callback.
1758 devfs_spec_strategy_done(struct bio *nbio)
1760 struct buf *nbp = nbio->bio_buf;
1761 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1762 struct buf *bp = bio->bio_buf; /* original bp */
1763 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1764 int boffset = nbp->b_data - bp->b_data;
1766 if (nbp->b_flags & B_ERROR) {
1768 * An error terminates the chain, propogate the error back
1769 * to the original bp
1771 bp->b_flags |= B_ERROR;
1772 bp->b_error = nbp->b_error;
1773 bp->b_resid = bp->b_bcount - boffset +
1774 (nbp->b_bcount - nbp->b_resid);
1775 #if SPEC_CHAIN_DEBUG & 1
1776 devfs_debug(DEVFS_DEBUG_DEBUG,
1777 "spec_strategy: chain %p error %d bcount %d/%d\n",
1778 bp, bp->b_error, bp->b_bcount,
1779 bp->b_bcount - bp->b_resid);
1781 } else if (nbp->b_resid) {
1783 * A short read or write terminates the chain
1785 bp->b_error = nbp->b_error;
1786 bp->b_resid = bp->b_bcount - boffset +
1787 (nbp->b_bcount - nbp->b_resid);
1788 #if SPEC_CHAIN_DEBUG & 1
1789 devfs_debug(DEVFS_DEBUG_DEBUG,
1790 "spec_strategy: chain %p short read(1) "
1792 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1794 } else if (nbp->b_bcount != nbp->b_bufsize) {
1796 * A short read or write can also occur by truncating b_bcount
1798 #if SPEC_CHAIN_DEBUG & 1
1799 devfs_debug(DEVFS_DEBUG_DEBUG,
1800 "spec_strategy: chain %p short read(2) "
1802 bp, nbp->b_bcount + boffset, bp->b_bcount);
1805 bp->b_bcount = nbp->b_bcount + boffset;
1806 bp->b_resid = nbp->b_resid;
1807 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1809 * No more data terminates the chain
1811 #if SPEC_CHAIN_DEBUG & 1
1812 devfs_debug(DEVFS_DEBUG_DEBUG,
1813 "spec_strategy: chain %p finished bcount %d\n",
1820 * Continue the chain
1822 boffset += nbp->b_bcount;
1823 nbp->b_data = bp->b_data + boffset;
1824 nbp->b_bcount = bp->b_bcount - boffset;
1825 if (nbp->b_bcount > chunksize)
1826 nbp->b_bcount = chunksize;
1827 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1828 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1830 #if SPEC_CHAIN_DEBUG & 1
1831 devfs_debug(DEVFS_DEBUG_DEBUG,
1832 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1833 bp, boffset, bp->b_bcount, nbp->b_bcount);
1836 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1841 * Fall through to here on termination. biodone(bp) and
1842 * clean up and free nbp.
1847 kfree(nbp, M_DEVBUF);
1851 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1854 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1859 * XXX: This assumes that strategy does the deed right away.
1860 * XXX: this may not be TRTTD.
1862 KKASSERT(ap->a_vp->v_rdev != NULL);
1863 if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1865 bp = geteblk(ap->a_length);
1866 bp->b_cmd = BUF_CMD_FREEBLKS;
1867 bp->b_bio1.bio_offset = ap->a_offset;
1868 bp->b_bcount = ap->a_length;
1869 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1874 * Implement degenerate case where the block requested is the block
1875 * returned, and assume that the entire device is contiguous in regards
1876 * to the contiguous block range (runp and runb).
1878 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1879 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1882 devfs_spec_bmap(struct vop_bmap_args *ap)
1884 if (ap->a_doffsetp != NULL)
1885 *ap->a_doffsetp = ap->a_loffset;
1886 if (ap->a_runp != NULL)
1887 *ap->a_runp = MAXBSIZE;
1888 if (ap->a_runb != NULL) {
1889 if (ap->a_loffset < MAXBSIZE)
1890 *ap->a_runb = (int)ap->a_loffset;
1892 *ap->a_runb = MAXBSIZE;
1899 * Special device advisory byte-level locks.
1901 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1902 * struct flock *a_fl, int a_flags)
1906 devfs_spec_advlock(struct vop_advlock_args *ap)
1908 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1912 * NOTE: MPSAFE callback.
1915 devfs_spec_getpages_iodone(struct bio *bio)
1917 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1918 wakeup(bio->bio_buf);
1922 * spec_getpages() - get pages associated with device vnode.
1924 * Note that spec_read and spec_write do not use the buffer cache, so we
1925 * must fully implement getpages here.
1928 devfs_spec_getpages(struct vop_getpages_args *ap)
1932 int i, pcount, size;
1935 vm_ooffset_t offset;
1936 int toff, nextoff, nread;
1937 struct vnode *vp = ap->a_vp;
1942 pcount = round_page(ap->a_count) / PAGE_SIZE;
1945 * Calculate the offset of the transfer and do sanity check.
1947 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1950 * Round up physical size for real devices. We cannot round using
1951 * v_mount's block size data because v_mount has nothing to do with
1952 * the device. i.e. it's usually '/dev'. We need the physical block
1953 * size for the device itself.
1955 * We can't use v_rdev->si_mountpoint because it only exists when the
1956 * block device is mounted. However, we can use v_rdev.
1958 if (vn_isdisk(vp, NULL))
1959 blksiz = vp->v_rdev->si_bsize_phys;
1963 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1965 bp = getpbuf_kva(NULL);
1966 kva = (vm_offset_t)bp->b_data;
1969 * Map the pages to be read into the kva.
1971 pmap_qenter(kva, ap->a_m, pcount);
1973 /* Build a minimal buffer header. */
1974 bp->b_cmd = BUF_CMD_READ;
1975 bp->b_bcount = size;
1977 bsetrunningbufspace(bp, size);
1979 bp->b_bio1.bio_offset = offset;
1980 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1982 mycpu->gd_cnt.v_vnodein++;
1983 mycpu->gd_cnt.v_vnodepgsin += pcount;
1986 vn_strategy(ap->a_vp, &bp->b_bio1);
1990 /* We definitely need to be at splbio here. */
1991 while (bp->b_cmd != BUF_CMD_DONE)
1992 tsleep(bp, 0, "spread", 0);
1996 if (bp->b_flags & B_ERROR) {
1998 error = bp->b_error;
2004 * If EOF is encountered we must zero-extend the result in order
2005 * to ensure that the page does not contain garabge. When no
2006 * error occurs, an early EOF is indicated if b_bcount got truncated.
2007 * b_resid is relative to b_bcount and should be 0, but some devices
2008 * might indicate an EOF with b_resid instead of truncating b_bcount.
2010 nread = bp->b_bcount - bp->b_resid;
2011 if (nread < ap->a_count)
2012 bzero((caddr_t)kva + nread, ap->a_count - nread);
2013 pmap_qremove(kva, pcount);
2016 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2017 nextoff = toff + PAGE_SIZE;
2020 m->flags &= ~PG_ZERO;
2023 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2024 * pmap modified bit. pmap modified bit should have
2025 * already been cleared.
2027 if (nextoff <= nread) {
2028 m->valid = VM_PAGE_BITS_ALL;
2030 } else if (toff < nread) {
2032 * Since this is a VM request, we have to supply the
2033 * unaligned offset to allow vm_page_set_valid()
2034 * to zero sub-DEV_BSIZE'd portions of the page.
2036 vm_page_set_valid(m, 0, nread - toff);
2037 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2043 if (i != ap->a_reqpage) {
2045 * Just in case someone was asking for this page we
2046 * now tell them that it is ok to use.
2048 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2050 if (m->flags & PG_REFERENCED) {
2051 vm_page_activate(m);
2053 vm_page_deactivate(m);
2062 } else if (m->valid) {
2065 * Since this is a VM request, we need to make the
2066 * entire page presentable by zeroing invalid sections.
2068 if (m->valid != VM_PAGE_BITS_ALL)
2069 vm_page_zero_invalid(m, FALSE);
2073 m = ap->a_m[ap->a_reqpage];
2074 devfs_debug(DEVFS_DEBUG_WARNING,
2075 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2076 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2077 devfs_debug(DEVFS_DEBUG_WARNING,
2078 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2079 size, bp->b_resid, ap->a_count, m->valid);
2080 devfs_debug(DEVFS_DEBUG_WARNING,
2081 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2082 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2084 * Free the buffer header back to the swap buffer pool.
2087 return VM_PAGER_ERROR;
2090 * Free the buffer header back to the swap buffer pool.
2093 if (DEVFS_NODE(ap->a_vp))
2094 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2100 sequential_heuristic(struct uio *uio, struct file *fp)
2103 * Sequential heuristic - detect sequential operation
2105 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2106 uio->uio_offset == fp->f_nextoff) {
2108 * XXX we assume that the filesystem block size is
2109 * the default. Not true, but still gives us a pretty
2110 * good indicator of how sequential the read operations
2113 int tmpseq = fp->f_seqcount;
2115 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2116 if (tmpseq > IO_SEQMAX)
2118 fp->f_seqcount = tmpseq;
2119 return(fp->f_seqcount << IO_SEQSHIFT);
2123 * Not sequential, quick draw-down of seqcount
2125 if (fp->f_seqcount > 1)