2 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Alex Hornung <ahornung@gmail.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/param.h>
35 #include <sys/systm.h>
37 #include <sys/kernel.h>
39 #include <sys/fcntl.h>
42 #include <sys/signalvar.h>
43 #include <sys/vnode.h>
45 #include <sys/mount.h>
47 #include <sys/fcntl.h>
48 #include <sys/namei.h>
49 #include <sys/dirent.h>
50 #include <sys/malloc.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_zone.h>
55 #include <vm/vm_object.h>
56 #include <sys/filio.h>
57 #include <sys/ttycom.h>
59 #include <sys/diskslice.h>
60 #include <sys/sysctl.h>
61 #include <sys/devfs.h>
62 #include <sys/pioctl.h>
64 #include <machine/limits.h>
67 #include <sys/sysref2.h>
68 #include <sys/mplock2.h>
69 #include <vm/vm_page2.h>
71 MALLOC_DECLARE(M_DEVFS);
72 #define DEVFS_BADOP (void *)devfs_badop
74 static int devfs_badop(struct vop_generic_args *);
75 static int devfs_access(struct vop_access_args *);
76 static int devfs_inactive(struct vop_inactive_args *);
77 static int devfs_reclaim(struct vop_reclaim_args *);
78 static int devfs_readdir(struct vop_readdir_args *);
79 static int devfs_getattr(struct vop_getattr_args *);
80 static int devfs_setattr(struct vop_setattr_args *);
81 static int devfs_readlink(struct vop_readlink_args *);
82 static int devfs_print(struct vop_print_args *);
84 static int devfs_nresolve(struct vop_nresolve_args *);
85 static int devfs_nlookupdotdot(struct vop_nlookupdotdot_args *);
86 static int devfs_nsymlink(struct vop_nsymlink_args *);
87 static int devfs_nremove(struct vop_nremove_args *);
89 static int devfs_spec_open(struct vop_open_args *);
90 static int devfs_spec_close(struct vop_close_args *);
91 static int devfs_spec_fsync(struct vop_fsync_args *);
93 static int devfs_spec_read(struct vop_read_args *);
94 static int devfs_spec_write(struct vop_write_args *);
95 static int devfs_spec_ioctl(struct vop_ioctl_args *);
96 static int devfs_spec_poll(struct vop_poll_args *);
97 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
98 static int devfs_spec_strategy(struct vop_strategy_args *);
99 static void devfs_spec_strategy_done(struct bio *);
100 static int devfs_spec_freeblks(struct vop_freeblks_args *);
101 static int devfs_spec_bmap(struct vop_bmap_args *);
102 static int devfs_spec_advlock(struct vop_advlock_args *);
103 static void devfs_spec_getpages_iodone(struct bio *);
104 static int devfs_spec_getpages(struct vop_getpages_args *);
107 static int devfs_specf_close(struct file *);
108 static int devfs_specf_read(struct file *, struct uio *, struct ucred *, int);
109 static int devfs_specf_write(struct file *, struct uio *, struct ucred *, int);
110 static int devfs_specf_stat(struct file *, struct stat *, struct ucred *);
111 static int devfs_specf_kqfilter(struct file *, struct knote *);
112 static int devfs_specf_poll(struct file *, int, struct ucred *);
113 static int devfs_specf_ioctl(struct file *, u_long, caddr_t,
114 struct ucred *, struct sysmsg *);
115 static __inline int sequential_heuristic(struct uio *, struct file *);
117 extern struct lock devfs_lock;
119 static int mpsafe_reads, mpsafe_writes, mplock_reads, mplock_writes;
122 * devfs vnode operations for regular files
124 struct vop_ops devfs_vnode_norm_vops = {
125 .vop_default = vop_defaultop,
126 .vop_access = devfs_access,
127 .vop_advlock = DEVFS_BADOP,
128 .vop_bmap = DEVFS_BADOP,
129 .vop_close = vop_stdclose,
130 .vop_getattr = devfs_getattr,
131 .vop_inactive = devfs_inactive,
132 .vop_ncreate = DEVFS_BADOP,
133 .vop_nresolve = devfs_nresolve,
134 .vop_nlookupdotdot = devfs_nlookupdotdot,
135 .vop_nlink = DEVFS_BADOP,
136 .vop_nmkdir = DEVFS_BADOP,
137 .vop_nmknod = DEVFS_BADOP,
138 .vop_nremove = devfs_nremove,
139 .vop_nrename = DEVFS_BADOP,
140 .vop_nrmdir = DEVFS_BADOP,
141 .vop_nsymlink = devfs_nsymlink,
142 .vop_open = vop_stdopen,
143 .vop_pathconf = vop_stdpathconf,
144 .vop_print = devfs_print,
145 .vop_read = DEVFS_BADOP,
146 .vop_readdir = devfs_readdir,
147 .vop_readlink = devfs_readlink,
148 .vop_reclaim = devfs_reclaim,
149 .vop_setattr = devfs_setattr,
150 .vop_write = DEVFS_BADOP,
151 .vop_ioctl = DEVFS_BADOP
155 * devfs vnode operations for character devices
157 struct vop_ops devfs_vnode_dev_vops = {
158 .vop_default = vop_defaultop,
159 .vop_access = devfs_access,
160 .vop_advlock = devfs_spec_advlock,
161 .vop_bmap = devfs_spec_bmap,
162 .vop_close = devfs_spec_close,
163 .vop_freeblks = devfs_spec_freeblks,
164 .vop_fsync = devfs_spec_fsync,
165 .vop_getattr = devfs_getattr,
166 .vop_getpages = devfs_spec_getpages,
167 .vop_inactive = devfs_inactive,
168 .vop_open = devfs_spec_open,
169 .vop_pathconf = vop_stdpathconf,
170 .vop_print = devfs_print,
171 .vop_poll = devfs_spec_poll,
172 .vop_kqfilter = devfs_spec_kqfilter,
173 .vop_read = devfs_spec_read,
174 .vop_readdir = DEVFS_BADOP,
175 .vop_readlink = DEVFS_BADOP,
176 .vop_reclaim = devfs_reclaim,
177 .vop_setattr = devfs_setattr,
178 .vop_strategy = devfs_spec_strategy,
179 .vop_write = devfs_spec_write,
180 .vop_ioctl = devfs_spec_ioctl
183 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
185 struct fileops devfs_dev_fileops = {
186 .fo_read = devfs_specf_read,
187 .fo_write = devfs_specf_write,
188 .fo_ioctl = devfs_specf_ioctl,
189 .fo_poll = devfs_specf_poll,
190 .fo_kqfilter = devfs_specf_kqfilter,
191 .fo_stat = devfs_specf_stat,
192 .fo_close = devfs_specf_close,
193 .fo_shutdown = nofo_shutdown
197 * These two functions are possibly temporary hacks for
198 * devices (aka the pty code) which want to control the
199 * node attributes themselves.
201 * XXX we may ultimately desire to simply remove the uid/gid/mode
202 * from the node entirely.
205 node_sync_dev_get(struct devfs_node *node)
209 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
210 node->uid = dev->si_uid;
211 node->gid = dev->si_gid;
212 node->mode = dev->si_perms;
217 node_sync_dev_set(struct devfs_node *node)
221 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
222 dev->si_uid = node->uid;
223 dev->si_gid = node->gid;
224 dev->si_perms = node->mode;
229 * generic entry point for unsupported operations
232 devfs_badop(struct vop_generic_args *ap)
239 devfs_access(struct vop_access_args *ap)
241 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
244 if (!devfs_node_is_accessible(node))
246 node_sync_dev_get(node);
247 error = vop_helper_access(ap, node->uid, node->gid,
248 node->mode, node->flags);
255 devfs_inactive(struct vop_inactive_args *ap)
257 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
259 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
266 devfs_reclaim(struct vop_reclaim_args *ap)
268 struct devfs_node *node;
273 * Check if it is locked already. if not, we acquire the devfs lock
275 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
276 lockmgr(&devfs_lock, LK_EXCLUSIVE);
283 * Get rid of the devfs_node if it is no longer linked into the
287 if ((node = DEVFS_NODE(vp)) != NULL) {
289 if ((node->flags & DEVFS_NODE_LINKED) == 0)
294 lockmgr(&devfs_lock, LK_RELEASE);
297 * v_rdev needs to be properly released using v_release_rdev
298 * Make sure v_data is NULL as well.
307 devfs_readdir(struct vop_readdir_args *ap)
309 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
310 struct devfs_node *node;
319 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
321 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
323 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
326 if (!devfs_node_is_accessible(dnode)) {
331 lockmgr(&devfs_lock, LK_EXCLUSIVE);
333 saveoff = ap->a_uio->uio_offset;
335 if (ap->a_ncookies) {
336 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
339 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
347 nanotime(&dnode->atime);
350 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
355 cookies[cookie_index] = saveoff;
358 if (cookie_index == ncookies)
364 r = vop_write_dirent(&error, ap->a_uio,
365 dnode->parent->d_dir.d_ino,
368 r = vop_write_dirent(&error, ap->a_uio,
375 cookies[cookie_index] = saveoff;
378 if (cookie_index == ncookies)
382 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
383 if ((node->flags & DEVFS_HIDDEN) ||
384 (node->flags & DEVFS_INVISIBLE)) {
389 * If the node type is a valid devfs alias, then we make sure that the
390 * target isn't hidden. If it is, we don't show the link in the
393 if ((node->node_type == Plink) && (node->link_target != NULL) &&
394 (node->link_target->flags & DEVFS_HIDDEN))
397 if (node->cookie < saveoff)
400 saveoff = node->cookie;
402 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
404 node->d_dir.d_namlen,
413 cookies[cookie_index] = node->cookie;
415 if (cookie_index == ncookies)
420 lockmgr(&devfs_lock, LK_RELEASE);
423 ap->a_uio->uio_offset = saveoff;
424 if (error && cookie_index == 0) {
426 kfree(cookies, M_TEMP);
428 *ap->a_cookies = NULL;
432 *ap->a_ncookies = cookie_index;
433 *ap->a_cookies = cookies;
441 devfs_nresolve(struct vop_nresolve_args *ap)
443 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
444 struct devfs_node *node, *found = NULL;
445 struct namecache *ncp;
446 struct vnode *vp = NULL;
451 ncp = ap->a_nch->ncp;
454 if (!devfs_node_is_accessible(dnode))
457 lockmgr(&devfs_lock, LK_EXCLUSIVE);
459 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
461 cache_setvp(ap->a_nch, NULL);
465 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
466 if (len == node->d_dir.d_namlen) {
467 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
476 while ((found->node_type == Plink) && (found->link_target)) {
478 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
482 found = found->link_target;
486 if (!(found->flags & DEVFS_HIDDEN))
487 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
492 cache_setvp(ap->a_nch, NULL);
498 cache_setvp(ap->a_nch, vp);
501 lockmgr(&devfs_lock, LK_RELEASE);
508 devfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
510 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
513 if (!devfs_node_is_accessible(dnode))
516 lockmgr(&devfs_lock, LK_EXCLUSIVE);
517 if (dnode->parent != NULL) {
518 devfs_allocv(ap->a_vpp, dnode->parent);
519 vn_unlock(*ap->a_vpp);
521 lockmgr(&devfs_lock, LK_RELEASE);
523 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
528 devfs_getattr(struct vop_getattr_args *ap)
530 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
531 struct vattr *vap = ap->a_vap;
532 struct partinfo pinfo;
536 if (!devfs_node_is_accessible(node))
539 node_sync_dev_get(node);
541 lockmgr(&devfs_lock, LK_EXCLUSIVE);
543 /* start by zeroing out the attributes */
546 /* next do all the common fields */
547 vap->va_type = ap->a_vp->v_type;
548 vap->va_mode = node->mode;
549 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
550 vap->va_flags = 0; /* XXX: what should this be? */
551 vap->va_blocksize = DEV_BSIZE;
552 vap->va_bytes = vap->va_size = sizeof(struct devfs_node);
554 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
556 vap->va_atime = node->atime;
557 vap->va_mtime = node->mtime;
558 vap->va_ctime = node->ctime;
560 vap->va_nlink = 1; /* number of references to file */
562 vap->va_uid = node->uid;
563 vap->va_gid = node->gid;
568 if ((node->node_type == Pdev) && node->d_dev) {
569 reference_dev(node->d_dev);
570 vap->va_rminor = node->d_dev->si_uminor;
571 release_dev(node->d_dev);
574 /* For a softlink the va_size is the length of the softlink */
575 if (node->symlink_name != 0) {
576 vap->va_bytes = vap->va_size = node->symlink_namelen;
580 * For a disk-type device, va_size is the size of the underlying
581 * device, so that lseek() works properly.
583 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
584 bzero(&pinfo, sizeof(pinfo));
585 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
586 0, proc0.p_ucred, NULL);
587 if ((error == 0) && (pinfo.media_blksize != 0)) {
588 vap->va_size = pinfo.media_size;
595 lockmgr(&devfs_lock, LK_RELEASE);
602 devfs_setattr(struct vop_setattr_args *ap)
604 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
608 if (!devfs_node_is_accessible(node))
610 node_sync_dev_get(node);
612 lockmgr(&devfs_lock, LK_EXCLUSIVE);
616 if (vap->va_uid != (uid_t)VNOVAL) {
617 if ((ap->a_cred->cr_uid != node->uid) &&
618 (!groupmember(node->gid, ap->a_cred))) {
619 error = priv_check(curthread, PRIV_VFS_CHOWN);
623 node->uid = vap->va_uid;
626 if (vap->va_gid != (uid_t)VNOVAL) {
627 if ((ap->a_cred->cr_uid != node->uid) &&
628 (!groupmember(node->gid, ap->a_cred))) {
629 error = priv_check(curthread, PRIV_VFS_CHOWN);
633 node->gid = vap->va_gid;
636 if (vap->va_mode != (mode_t)VNOVAL) {
637 if (ap->a_cred->cr_uid != node->uid) {
638 error = priv_check(curthread, PRIV_VFS_ADMIN);
642 node->mode = vap->va_mode;
646 node_sync_dev_set(node);
647 nanotime(&node->ctime);
648 lockmgr(&devfs_lock, LK_RELEASE);
655 devfs_readlink(struct vop_readlink_args *ap)
657 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
660 if (!devfs_node_is_accessible(node))
663 lockmgr(&devfs_lock, LK_EXCLUSIVE);
664 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
665 lockmgr(&devfs_lock, LK_RELEASE);
672 devfs_print(struct vop_print_args *ap)
679 devfs_nsymlink(struct vop_nsymlink_args *ap)
681 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
682 struct devfs_node *node;
685 if (!devfs_node_is_accessible(dnode))
688 ap->a_vap->va_type = VLNK;
690 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
693 lockmgr(&devfs_lock, LK_EXCLUSIVE);
694 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
695 ap->a_nch->ncp->nc_name, dnode, NULL);
697 targetlen = strlen(ap->a_target);
699 node = DEVFS_NODE(*ap->a_vpp);
700 node->flags |= DEVFS_USER_CREATED;
701 node->symlink_namelen = targetlen;
702 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
703 memcpy(node->symlink_name, ap->a_target, targetlen);
704 node->symlink_name[targetlen] = '\0';
705 cache_setunresolved(ap->a_nch);
706 cache_setvp(ap->a_nch, *ap->a_vpp);
708 lockmgr(&devfs_lock, LK_RELEASE);
710 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
715 devfs_nremove(struct vop_nremove_args *ap)
717 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
718 struct devfs_node *node;
719 struct namecache *ncp;
722 ncp = ap->a_nch->ncp;
724 if (!devfs_node_is_accessible(dnode))
727 lockmgr(&devfs_lock, LK_EXCLUSIVE);
729 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
732 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
733 if (ncp->nc_nlen != node->d_dir.d_namlen)
735 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
739 * only allow removal of user created stuff (e.g. symlinks)
741 if ((node->flags & DEVFS_USER_CREATED) == 0) {
746 cache_inval_vp(node->v_node, CINV_DESTROY);
753 cache_setunresolved(ap->a_nch);
754 cache_setvp(ap->a_nch, NULL);
757 lockmgr(&devfs_lock, LK_RELEASE);
763 devfs_spec_open(struct vop_open_args *ap)
765 struct vnode *vp = ap->a_vp;
766 struct vnode *orig_vp = NULL;
767 struct devfs_node *node = DEVFS_NODE(vp);
768 struct devfs_node *newnode;
769 cdev_t dev, ndev = NULL;
773 if (node->d_dev == NULL)
775 if (!devfs_node_is_accessible(node))
779 if ((dev = vp->v_rdev) == NULL)
782 if (node && ap->a_fp) {
783 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
784 lockmgr(&devfs_lock, LK_EXCLUSIVE);
786 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
787 ap->a_mode, ap->a_cred);
789 newnode = devfs_create_device_node(
790 DEVFS_MNTDATA(vp->v_mount)->root_node,
792 /* XXX: possibly destroy device if this happens */
794 if (newnode != NULL) {
798 devfs_debug(DEVFS_DEBUG_DEBUG,
799 "parent here is: %s, node is: |%s|\n",
800 ((node->parent->node_type == Proot) ?
801 "ROOT!" : node->parent->d_dir.d_name),
802 newnode->d_dir.d_name);
803 devfs_debug(DEVFS_DEBUG_DEBUG,
805 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
808 * orig_vp is set to the original vp if we cloned.
810 /* node->flags |= DEVFS_CLONED; */
811 devfs_allocv(&vp, newnode);
816 lockmgr(&devfs_lock, LK_RELEASE);
819 devfs_debug(DEVFS_DEBUG_DEBUG,
820 "devfs_spec_open() called on %s! \n",
824 * Make this field valid before any I/O in ->d_open
826 if (!dev->si_iosize_max)
827 dev->si_iosize_max = DFLTPHYS;
829 if (dev_dflags(dev) & D_TTY)
830 vsetflags(vp, VISTTY);
833 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
834 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
837 * Clean up any cloned vp if we error out.
843 /* orig_vp = NULL; */
849 * This checks if the disk device is going to be opened for writing.
850 * It will be only allowed in the cases where securelevel permits it
851 * and it's not mounted R/W.
853 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
854 (ap->a_cred != FSCRED)) {
856 /* Very secure mode. No open for writing allowed */
857 if (securelevel >= 2)
861 * If it is mounted R/W, do not allow to open for writing.
862 * In the case it's mounted read-only but securelevel
863 * is >= 1, then do not allow opening for writing either.
865 if (vfs_mountedon(vp)) {
866 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
868 else if (securelevel >= 1)
873 if (dev_dflags(dev) & D_TTY) {
878 devfs_debug(DEVFS_DEBUG_DEBUG,
879 "devfs: no t_stop\n");
880 tp->t_stop = nottystop;
886 if (vn_isdisk(vp, NULL)) {
887 if (!dev->si_bsize_phys)
888 dev->si_bsize_phys = DEV_BSIZE;
889 vinitvmio(vp, IDX_TO_OFF(INT_MAX));
895 nanotime(&node->atime);
901 /* Ugly pty magic, to make pty devices appear once they are opened */
902 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
903 node->flags &= ~DEVFS_INVISIBLE;
906 ap->a_fp->f_type = DTYPE_VNODE;
907 ap->a_fp->f_flag = ap->a_mode & FMASK;
908 ap->a_fp->f_ops = &devfs_dev_fileops;
909 ap->a_fp->f_data = vp;
917 devfs_spec_close(struct vop_close_args *ap)
919 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
920 struct proc *p = curproc;
921 struct vnode *vp = ap->a_vp;
922 cdev_t dev = vp->v_rdev;
926 devfs_debug(DEVFS_DEBUG_DEBUG,
927 "devfs_spec_close() called on %s! \n",
931 * A couple of hacks for devices and tty devices. The
932 * vnode ref count cannot be used to figure out the
933 * last close, but we can use v_opencount now that
934 * revoke works properly.
936 * Detect the last close on a controlling terminal and clear
937 * the session (half-close).
942 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
943 p->p_session->s_ttyvp = NULL;
948 * Vnodes can be opened and closed multiple times. Do not really
949 * close the device unless (1) it is being closed forcibly,
950 * (2) the device wants to track closes, or (3) this is the last
951 * vnode doing its last close on the device.
953 * XXX the VXLOCK (force close) case can leave vnodes referencing
954 * a closed device. This might not occur now that our revoke is
957 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
958 if (dev && ((vp->v_flag & VRECLAIMED) ||
959 (dev_dflags(dev) & D_TRACKCLOSE) ||
960 (vp->v_opencount == 1))) {
962 * Unlock around dev_dclose()
965 if (vn_islocked(vp)) {
969 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
972 * Ugly pty magic, to make pty devices disappear again once
975 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
976 node->flags |= DEVFS_INVISIBLE;
979 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
983 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
986 * Track the actual opens and closes on the vnode. The last close
987 * disassociates the rdev. If the rdev is already disassociated or
988 * the opencount is already 0, the vnode might have been revoked
989 * and no further opencount tracking occurs.
993 if (vp->v_opencount > 0)
1001 devfs_specf_close(struct file *fp)
1003 struct vnode *vp = (struct vnode *)fp->f_data;
1007 fp->f_ops = &badfileops;
1008 error = vn_close(vp, fp->f_flag);
1016 * Device-optimized file table vnode read routine.
1018 * This bypasses the VOP table and talks directly to the device. Most
1019 * filesystems just route to specfs and can make this optimization.
1021 * MPALMOSTSAFE - acquires mplock
1024 devfs_specf_read(struct file *fp, struct uio *uio,
1025 struct ucred *cred, int flags)
1027 struct devfs_node *node;
1033 KASSERT(uio->uio_td == curthread,
1034 ("uio_td %p is not td %p", uio->uio_td, curthread));
1036 if (uio->uio_resid == 0)
1039 vp = (struct vnode *)fp->f_data;
1040 if (vp == NULL || vp->v_type == VBAD)
1043 node = DEVFS_NODE(vp);
1045 if ((dev = vp->v_rdev) == NULL)
1048 /* only acquire mplock for devices that require it */
1049 if (!(dev_dflags(dev) & D_MPSAFE_READ)) {
1050 atomic_add_int(&mplock_reads, 1);
1053 atomic_add_int(&mpsafe_reads, 1);
1058 if ((flags & O_FOFFSET) == 0)
1059 uio->uio_offset = fp->f_offset;
1062 if (flags & O_FBLOCKING) {
1063 /* ioflag &= ~IO_NDELAY; */
1064 } else if (flags & O_FNONBLOCKING) {
1065 ioflag |= IO_NDELAY;
1066 } else if (fp->f_flag & FNONBLOCK) {
1067 ioflag |= IO_NDELAY;
1069 if (flags & O_FBUFFERED) {
1070 /* ioflag &= ~IO_DIRECT; */
1071 } else if (flags & O_FUNBUFFERED) {
1072 ioflag |= IO_DIRECT;
1073 } else if (fp->f_flag & O_DIRECT) {
1074 ioflag |= IO_DIRECT;
1076 ioflag |= sequential_heuristic(uio, fp);
1078 error = dev_dread(dev, uio, ioflag);
1082 nanotime(&node->atime);
1083 if ((flags & O_FOFFSET) == 0)
1084 fp->f_offset = uio->uio_offset;
1085 fp->f_nextoff = uio->uio_offset;
1087 if (!(dev_dflags(dev) & D_MPSAFE_READ))
1095 devfs_specf_write(struct file *fp, struct uio *uio,
1096 struct ucred *cred, int flags)
1098 struct devfs_node *node;
1104 KASSERT(uio->uio_td == curthread,
1105 ("uio_td %p is not p %p", uio->uio_td, curthread));
1107 vp = (struct vnode *)fp->f_data;
1108 if (vp == NULL || vp->v_type == VBAD)
1111 node = DEVFS_NODE(vp);
1113 if (vp->v_type == VREG)
1114 bwillwrite(uio->uio_resid);
1116 vp = (struct vnode *)fp->f_data;
1118 if ((dev = vp->v_rdev) == NULL)
1121 /* only acquire mplock for devices that require it */
1122 if (!(dev_dflags(dev) & D_MPSAFE_WRITE)) {
1123 atomic_add_int(&mplock_writes, 1);
1126 atomic_add_int(&mpsafe_writes, 1);
1131 if ((flags & O_FOFFSET) == 0)
1132 uio->uio_offset = fp->f_offset;
1135 if (vp->v_type == VREG &&
1136 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1137 ioflag |= IO_APPEND;
1140 if (flags & O_FBLOCKING) {
1141 /* ioflag &= ~IO_NDELAY; */
1142 } else if (flags & O_FNONBLOCKING) {
1143 ioflag |= IO_NDELAY;
1144 } else if (fp->f_flag & FNONBLOCK) {
1145 ioflag |= IO_NDELAY;
1147 if (flags & O_FBUFFERED) {
1148 /* ioflag &= ~IO_DIRECT; */
1149 } else if (flags & O_FUNBUFFERED) {
1150 ioflag |= IO_DIRECT;
1151 } else if (fp->f_flag & O_DIRECT) {
1152 ioflag |= IO_DIRECT;
1154 if (flags & O_FASYNCWRITE) {
1155 /* ioflag &= ~IO_SYNC; */
1156 } else if (flags & O_FSYNCWRITE) {
1158 } else if (fp->f_flag & O_FSYNC) {
1162 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1164 ioflag |= sequential_heuristic(uio, fp);
1166 error = dev_dwrite(dev, uio, ioflag);
1170 nanotime(&node->atime);
1171 nanotime(&node->mtime);
1174 if ((flags & O_FOFFSET) == 0)
1175 fp->f_offset = uio->uio_offset;
1176 fp->f_nextoff = uio->uio_offset;
1178 if (!(dev_dflags(dev) & D_MPSAFE_WRITE))
1185 devfs_specf_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1194 vp = (struct vnode *)fp->f_data;
1195 if (vp == NULL || vp->v_type == VBAD)
1198 error = vn_stat(vp, sb, cred);
1203 error = VOP_GETATTR(vp, vap);
1208 * Zero the spare stat fields
1215 * Copy from vattr table ... or not in case it's a cloned device
1217 if (vap->va_fsid != VNOVAL)
1218 sb->st_dev = vap->va_fsid;
1220 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1222 sb->st_ino = vap->va_fileid;
1224 mode = vap->va_mode;
1228 if (vap->va_nlink > (nlink_t)-1)
1229 sb->st_nlink = (nlink_t)-1;
1231 sb->st_nlink = vap->va_nlink;
1233 sb->st_uid = vap->va_uid;
1234 sb->st_gid = vap->va_gid;
1235 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1236 sb->st_size = vap->va_bytes;
1237 sb->st_atimespec = vap->va_atime;
1238 sb->st_mtimespec = vap->va_mtime;
1239 sb->st_ctimespec = vap->va_ctime;
1242 * A VCHR and VBLK device may track the last access and last modified
1243 * time independantly of the filesystem. This is particularly true
1244 * because device read and write calls may bypass the filesystem.
1246 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1249 if (dev->si_lastread) {
1250 sb->st_atimespec.tv_sec = dev->si_lastread;
1251 sb->st_atimespec.tv_nsec = 0;
1253 if (dev->si_lastwrite) {
1254 sb->st_atimespec.tv_sec = dev->si_lastwrite;
1255 sb->st_atimespec.tv_nsec = 0;
1261 * According to www.opengroup.org, the meaning of st_blksize is
1262 * "a filesystem-specific preferred I/O block size for this
1263 * object. In some filesystem types, this may vary from file
1265 * Default to PAGE_SIZE after much discussion.
1268 sb->st_blksize = PAGE_SIZE;
1270 sb->st_flags = vap->va_flags;
1272 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1276 sb->st_gen = (u_int32_t)vap->va_gen;
1278 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1285 devfs_specf_kqfilter(struct file *fp, struct knote *kn)
1293 vp = (struct vnode *)fp->f_data;
1294 if (vp == NULL || vp->v_type == VBAD) {
1298 if ((dev = vp->v_rdev) == NULL) {
1304 error = dev_dkqfilter(dev, kn);
1315 devfs_specf_poll(struct file *fp, int events, struct ucred *cred)
1317 struct devfs_node *node;
1324 vp = (struct vnode *)fp->f_data;
1325 if (vp == NULL || vp->v_type == VBAD) {
1329 node = DEVFS_NODE(vp);
1331 if ((dev = vp->v_rdev) == NULL) {
1336 error = dev_dpoll(dev, events);
1342 nanotime(&node->atime);
1351 * MPALMOSTSAFE - acquires mplock
1354 devfs_specf_ioctl(struct file *fp, u_long com, caddr_t data,
1355 struct ucred *ucred, struct sysmsg *msg)
1357 struct devfs_node *node;
1362 struct fiodname_args *name_args;
1366 vp = ((struct vnode *)fp->f_data);
1368 if ((dev = vp->v_rdev) == NULL)
1369 return EBADF; /* device was revoked */
1373 node = DEVFS_NODE(vp);
1375 devfs_debug(DEVFS_DEBUG_DEBUG,
1376 "devfs_specf_ioctl() called! for dev %s\n",
1379 if (com == FIODTYPE) {
1380 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1383 } else if (com == FIODNAME) {
1384 name_args = (struct fiodname_args *)data;
1385 name = dev->si_name;
1386 namlen = strlen(name) + 1;
1388 devfs_debug(DEVFS_DEBUG_DEBUG,
1389 "ioctl, got: FIODNAME for %s\n", name);
1391 if (namlen <= name_args->len)
1392 error = copyout(dev->si_name, name_args->name, namlen);
1396 devfs_debug(DEVFS_DEBUG_DEBUG,
1397 "ioctl stuff: error: %d\n", error);
1401 /* only acquire mplock for devices that require it */
1402 if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1405 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1409 nanotime(&node->atime);
1410 nanotime(&node->mtime);
1414 if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1417 if (com == TIOCSCTTY) {
1418 devfs_debug(DEVFS_DEBUG_DEBUG,
1419 "devfs_specf_ioctl: got TIOCSCTTY on %s\n",
1422 if (error == 0 && com == TIOCSCTTY) {
1423 struct proc *p = curthread->td_proc;
1424 struct session *sess;
1426 devfs_debug(DEVFS_DEBUG_DEBUG,
1427 "devfs_specf_ioctl: dealing with TIOCSCTTY on %s\n",
1433 sess = p->p_session;
1436 * Do nothing if reassigning same control tty
1438 if (sess->s_ttyvp == vp) {
1444 * Get rid of reference to old control tty
1446 ovp = sess->s_ttyvp;
1455 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_specf_ioctl() finished! \n");
1461 devfs_spec_fsync(struct vop_fsync_args *ap)
1463 struct vnode *vp = ap->a_vp;
1466 if (!vn_isdisk(vp, NULL))
1470 * Flush all dirty buffers associated with a block device.
1472 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1477 devfs_spec_read(struct vop_read_args *ap)
1479 struct devfs_node *node;
1488 node = DEVFS_NODE(vp);
1490 if (dev == NULL) /* device was revoked */
1492 if (uio->uio_resid == 0)
1496 error = dev_dread(dev, uio, ap->a_ioflag);
1497 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1500 nanotime(&node->atime);
1506 * Vnode op for write
1508 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1509 * struct ucred *a_cred)
1512 devfs_spec_write(struct vop_write_args *ap)
1514 struct devfs_node *node;
1523 node = DEVFS_NODE(vp);
1525 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1527 if (dev == NULL) /* device was revoked */
1531 error = dev_dwrite(dev, uio, ap->a_ioflag);
1532 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1535 nanotime(&node->atime);
1536 nanotime(&node->mtime);
1543 * Device ioctl operation.
1545 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1546 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1549 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1551 struct vnode *vp = ap->a_vp;
1552 struct devfs_node *node;
1555 if ((dev = vp->v_rdev) == NULL)
1556 return (EBADF); /* device was revoked */
1557 node = DEVFS_NODE(vp);
1561 nanotime(&node->atime);
1562 nanotime(&node->mtime);
1566 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1567 ap->a_cred, ap->a_sysmsg));
1571 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
1575 devfs_spec_poll(struct vop_poll_args *ap)
1577 struct vnode *vp = ap->a_vp;
1578 struct devfs_node *node;
1581 if ((dev = vp->v_rdev) == NULL)
1582 return (EBADF); /* device was revoked */
1583 node = DEVFS_NODE(vp);
1587 nanotime(&node->atime);
1590 return (dev_dpoll(dev, ap->a_events));
1594 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1598 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1600 struct vnode *vp = ap->a_vp;
1601 struct devfs_node *node;
1604 if ((dev = vp->v_rdev) == NULL)
1605 return (EBADF); /* device was revoked */
1606 node = DEVFS_NODE(vp);
1610 nanotime(&node->atime);
1613 return (dev_dkqfilter(dev, ap->a_kn));
1617 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1618 * calls are not limited to device DMA limits so we have to deal with the
1621 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1624 devfs_spec_strategy(struct vop_strategy_args *ap)
1626 struct bio *bio = ap->a_bio;
1627 struct buf *bp = bio->bio_buf;
1634 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1638 * Collect statistics on synchronous and asynchronous read
1639 * and write counts for disks that have associated filesystems.
1642 KKASSERT(vp->v_rdev != NULL); /* XXX */
1643 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1644 if (bp->b_cmd == BUF_CMD_READ) {
1645 if (bp->b_flags & BIO_SYNC)
1646 mp->mnt_stat.f_syncreads++;
1648 mp->mnt_stat.f_asyncreads++;
1650 if (bp->b_flags & BIO_SYNC)
1651 mp->mnt_stat.f_syncwrites++;
1653 mp->mnt_stat.f_asyncwrites++;
1658 * Device iosize limitations only apply to read and write. Shortcut
1659 * the I/O if it fits.
1661 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1662 devfs_debug(DEVFS_DEBUG_DEBUG,
1663 "%s: si_iosize_max not set!\n",
1664 dev_dname(vp->v_rdev));
1665 maxiosize = MAXPHYS;
1667 #if SPEC_CHAIN_DEBUG & 2
1670 if (bp->b_bcount <= maxiosize ||
1671 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1672 dev_dstrategy_chain(vp->v_rdev, bio);
1677 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1679 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1683 BUF_LOCK(nbp, LK_EXCLUSIVE);
1686 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1687 nbp->b_data = bp->b_data;
1688 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1689 nbp->b_bio1.bio_offset = bio->bio_offset;
1690 nbp->b_bio1.bio_caller_info1.ptr = bio;
1693 * Start the first transfer
1695 if (vn_isdisk(vp, NULL))
1696 chunksize = vp->v_rdev->si_bsize_phys;
1698 chunksize = DEV_BSIZE;
1699 chunksize = maxiosize / chunksize * chunksize;
1700 #if SPEC_CHAIN_DEBUG & 1
1701 devfs_debug(DEVFS_DEBUG_DEBUG,
1702 "spec_strategy chained I/O chunksize=%d\n",
1705 nbp->b_cmd = bp->b_cmd;
1706 nbp->b_bcount = chunksize;
1707 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1708 nbp->b_bio1.bio_caller_info2.index = chunksize;
1710 #if SPEC_CHAIN_DEBUG & 1
1711 devfs_debug(DEVFS_DEBUG_DEBUG,
1712 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1713 bp, 0, bp->b_bcount, nbp->b_bcount);
1716 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1718 if (DEVFS_NODE(vp)) {
1719 nanotime(&DEVFS_NODE(vp)->atime);
1720 nanotime(&DEVFS_NODE(vp)->mtime);
1727 * Chunked up transfer completion routine - chain transfers until done
1731 devfs_spec_strategy_done(struct bio *nbio)
1733 struct buf *nbp = nbio->bio_buf;
1734 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1735 struct buf *bp = bio->bio_buf; /* original bp */
1736 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1737 int boffset = nbp->b_data - bp->b_data;
1739 if (nbp->b_flags & B_ERROR) {
1741 * An error terminates the chain, propogate the error back
1742 * to the original bp
1744 bp->b_flags |= B_ERROR;
1745 bp->b_error = nbp->b_error;
1746 bp->b_resid = bp->b_bcount - boffset +
1747 (nbp->b_bcount - nbp->b_resid);
1748 #if SPEC_CHAIN_DEBUG & 1
1749 devfs_debug(DEVFS_DEBUG_DEBUG,
1750 "spec_strategy: chain %p error %d bcount %d/%d\n",
1751 bp, bp->b_error, bp->b_bcount,
1752 bp->b_bcount - bp->b_resid);
1754 kfree(nbp, M_DEVBUF);
1756 } else if (nbp->b_resid) {
1758 * A short read or write terminates the chain
1760 bp->b_error = nbp->b_error;
1761 bp->b_resid = bp->b_bcount - boffset +
1762 (nbp->b_bcount - nbp->b_resid);
1763 #if SPEC_CHAIN_DEBUG & 1
1764 devfs_debug(DEVFS_DEBUG_DEBUG,
1765 "spec_strategy: chain %p short read(1) "
1767 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1769 kfree(nbp, M_DEVBUF);
1771 } else if (nbp->b_bcount != nbp->b_bufsize) {
1773 * A short read or write can also occur by truncating b_bcount
1775 #if SPEC_CHAIN_DEBUG & 1
1776 devfs_debug(DEVFS_DEBUG_DEBUG,
1777 "spec_strategy: chain %p short read(2) "
1779 bp, nbp->b_bcount + boffset, bp->b_bcount);
1782 bp->b_bcount = nbp->b_bcount + boffset;
1783 bp->b_resid = nbp->b_resid;
1784 kfree(nbp, M_DEVBUF);
1786 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1788 * No more data terminates the chain
1790 #if SPEC_CHAIN_DEBUG & 1
1791 devfs_debug(DEVFS_DEBUG_DEBUG,
1792 "spec_strategy: chain %p finished bcount %d\n",
1797 kfree(nbp, M_DEVBUF);
1801 * Continue the chain
1803 boffset += nbp->b_bcount;
1804 nbp->b_data = bp->b_data + boffset;
1805 nbp->b_bcount = bp->b_bcount - boffset;
1806 if (nbp->b_bcount > chunksize)
1807 nbp->b_bcount = chunksize;
1808 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1809 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1811 #if SPEC_CHAIN_DEBUG & 1
1812 devfs_debug(DEVFS_DEBUG_DEBUG,
1813 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1814 bp, boffset, bp->b_bcount, nbp->b_bcount);
1817 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1822 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1825 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1830 * XXX: This assumes that strategy does the deed right away.
1831 * XXX: this may not be TRTTD.
1833 KKASSERT(ap->a_vp->v_rdev != NULL);
1834 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
1836 bp = geteblk(ap->a_length);
1837 bp->b_cmd = BUF_CMD_FREEBLKS;
1838 bp->b_bio1.bio_offset = ap->a_offset;
1839 bp->b_bcount = ap->a_length;
1840 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1845 * Implement degenerate case where the block requested is the block
1846 * returned, and assume that the entire device is contiguous in regards
1847 * to the contiguous block range (runp and runb).
1849 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1850 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1853 devfs_spec_bmap(struct vop_bmap_args *ap)
1855 if (ap->a_doffsetp != NULL)
1856 *ap->a_doffsetp = ap->a_loffset;
1857 if (ap->a_runp != NULL)
1858 *ap->a_runp = MAXBSIZE;
1859 if (ap->a_runb != NULL) {
1860 if (ap->a_loffset < MAXBSIZE)
1861 *ap->a_runb = (int)ap->a_loffset;
1863 *ap->a_runb = MAXBSIZE;
1870 * Special device advisory byte-level locks.
1872 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1873 * struct flock *a_fl, int a_flags)
1877 devfs_spec_advlock(struct vop_advlock_args *ap)
1879 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1883 devfs_spec_getpages_iodone(struct bio *bio)
1885 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1886 wakeup(bio->bio_buf);
1890 * spec_getpages() - get pages associated with device vnode.
1892 * Note that spec_read and spec_write do not use the buffer cache, so we
1893 * must fully implement getpages here.
1896 devfs_spec_getpages(struct vop_getpages_args *ap)
1900 int i, pcount, size;
1903 vm_ooffset_t offset;
1904 int toff, nextoff, nread;
1905 struct vnode *vp = ap->a_vp;
1910 pcount = round_page(ap->a_count) / PAGE_SIZE;
1913 * Calculate the offset of the transfer and do sanity check.
1915 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1918 * Round up physical size for real devices. We cannot round using
1919 * v_mount's block size data because v_mount has nothing to do with
1920 * the device. i.e. it's usually '/dev'. We need the physical block
1921 * size for the device itself.
1923 * We can't use v_rdev->si_mountpoint because it only exists when the
1924 * block device is mounted. However, we can use v_rdev.
1926 if (vn_isdisk(vp, NULL))
1927 blksiz = vp->v_rdev->si_bsize_phys;
1931 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1934 kva = (vm_offset_t)bp->b_data;
1937 * Map the pages to be read into the kva.
1939 pmap_qenter(kva, ap->a_m, pcount);
1941 /* Build a minimal buffer header. */
1942 bp->b_cmd = BUF_CMD_READ;
1943 bp->b_bcount = size;
1945 bp->b_runningbufspace = size;
1947 runningbufspace += bp->b_runningbufspace;
1951 bp->b_bio1.bio_offset = offset;
1952 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1954 mycpu->gd_cnt.v_vnodein++;
1955 mycpu->gd_cnt.v_vnodepgsin += pcount;
1958 vn_strategy(ap->a_vp, &bp->b_bio1);
1962 /* We definitely need to be at splbio here. */
1963 while (bp->b_cmd != BUF_CMD_DONE)
1964 tsleep(bp, 0, "spread", 0);
1968 if (bp->b_flags & B_ERROR) {
1970 error = bp->b_error;
1976 * If EOF is encountered we must zero-extend the result in order
1977 * to ensure that the page does not contain garabge. When no
1978 * error occurs, an early EOF is indicated if b_bcount got truncated.
1979 * b_resid is relative to b_bcount and should be 0, but some devices
1980 * might indicate an EOF with b_resid instead of truncating b_bcount.
1982 nread = bp->b_bcount - bp->b_resid;
1983 if (nread < ap->a_count)
1984 bzero((caddr_t)kva + nread, ap->a_count - nread);
1985 pmap_qremove(kva, pcount);
1988 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
1989 nextoff = toff + PAGE_SIZE;
1992 m->flags &= ~PG_ZERO;
1995 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
1996 * pmap modified bit. pmap modified bit should have
1997 * already been cleared.
1999 if (nextoff <= nread) {
2000 m->valid = VM_PAGE_BITS_ALL;
2002 } else if (toff < nread) {
2004 * Since this is a VM request, we have to supply the
2005 * unaligned offset to allow vm_page_set_valid()
2006 * to zero sub-DEV_BSIZE'd portions of the page.
2008 vm_page_set_valid(m, 0, nread - toff);
2009 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2015 if (i != ap->a_reqpage) {
2017 * Just in case someone was asking for this page we
2018 * now tell them that it is ok to use.
2020 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2022 if (m->flags & PG_WANTED) {
2023 vm_page_activate(m);
2025 vm_page_deactivate(m);
2034 } else if (m->valid) {
2037 * Since this is a VM request, we need to make the
2038 * entire page presentable by zeroing invalid sections.
2040 if (m->valid != VM_PAGE_BITS_ALL)
2041 vm_page_zero_invalid(m, FALSE);
2045 m = ap->a_m[ap->a_reqpage];
2046 devfs_debug(DEVFS_DEBUG_WARNING,
2047 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2048 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2049 devfs_debug(DEVFS_DEBUG_WARNING,
2050 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2051 size, bp->b_resid, ap->a_count, m->valid);
2052 devfs_debug(DEVFS_DEBUG_WARNING,
2053 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2054 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2056 * Free the buffer header back to the swap buffer pool.
2059 return VM_PAGER_ERROR;
2062 * Free the buffer header back to the swap buffer pool.
2065 if (DEVFS_NODE(ap->a_vp))
2066 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2072 sequential_heuristic(struct uio *uio, struct file *fp)
2075 * Sequential heuristic - detect sequential operation
2077 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2078 uio->uio_offset == fp->f_nextoff) {
2080 * XXX we assume that the filesystem block size is
2081 * the default. Not true, but still gives us a pretty
2082 * good indicator of how sequential the read operations
2085 int tmpseq = fp->f_seqcount;
2087 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2088 if (tmpseq > IO_SEQMAX)
2090 fp->f_seqcount = tmpseq;
2091 return(fp->f_seqcount << IO_SEQSHIFT);
2095 * Not sequential, quick draw-down of seqcount
2097 if (fp->f_seqcount > 1)
2104 extern SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2106 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_writes, CTLFLAG_RD, &mpsafe_writes,
2107 0, "mpsafe writes");
2108 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_writes, CTLFLAG_RD, &mplock_writes,
2109 0, "non-mpsafe writes");
2110 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_reads, CTLFLAG_RD, &mpsafe_reads,
2112 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_reads, CTLFLAG_RD, &mplock_reads,
2113 0, "non-mpsafe reads");