2 * Copyright (c) 1994 Jan-Simon Pendry
4 * The Regents of the University of California. All rights reserved.
6 * This code is derived from software contributed to Berkeley by
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
38 * $FreeBSD: src/sys/miscfs/union/union_subr.c,v 1.43.2.2 2001/12/25 01:44:45 dillon Exp $
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/vnode.h>
46 #include <sys/namei.h>
47 #include <sys/malloc.h>
48 #include <sys/fcntl.h>
50 #include <sys/filedesc.h>
51 #include <sys/module.h>
52 #include <sys/mount.h>
55 #include <vm/vm_extern.h> /* for vnode_pager_setsize */
56 #include <vm/vm_zone.h>
57 #include <vm/vm_object.h> /* for vm cache coherency */
60 extern int union_init (void);
62 /* must be power of two, otherwise change UNION_HASH() */
65 /* unsigned int ... */
66 #define UNION_HASH(u, l) \
67 (((((uintptr_t) (u)) + ((uintptr_t) l)) >> 8) & (NHASH-1))
69 static LIST_HEAD(unhead, union_node) unhead[NHASH];
70 static int unvplock[NHASH];
72 static void union_dircache_r (struct vnode *vp, struct vnode ***vppp,
74 static int union_list_lock (int ix);
75 static void union_list_unlock (int ix);
76 static int union_relookup (struct union_mount *um, struct vnode *dvp,
78 struct componentname *cnp,
79 struct componentname *cn, char *path,
81 static void union_updatevp (struct union_node *un,
82 struct vnode *uppervp,
83 struct vnode *lowervp);
84 static void union_newlower (struct union_node *, struct vnode *);
85 static void union_newupper (struct union_node *, struct vnode *);
86 static int union_copyfile (struct vnode *, struct vnode *,
87 struct ucred *, struct thread *);
88 static int union_vn_create (struct vnode **, struct union_node *,
90 static int union_vn_close (struct vnode *, int, struct ucred *);
97 for (i = 0; i < NHASH; i++)
98 LIST_INIT(&unhead[i]);
99 bzero((caddr_t)unvplock, sizeof(unvplock));
104 union_list_lock(int ix)
106 if (unvplock[ix] & UNVP_LOCKED) {
107 unvplock[ix] |= UNVP_WANT;
108 (void) tsleep((caddr_t) &unvplock[ix], 0, "unllck", 0);
111 unvplock[ix] |= UNVP_LOCKED;
116 union_list_unlock(int ix)
118 unvplock[ix] &= ~UNVP_LOCKED;
120 if (unvplock[ix] & UNVP_WANT) {
121 unvplock[ix] &= ~UNVP_WANT;
122 wakeup((caddr_t) &unvplock[ix]);
129 * The uppervp, if not NULL, must be referenced and not locked by us
130 * The lowervp, if not NULL, must be referenced.
132 * if uppervp and lowervp match pointers already installed, nothing
133 * happens. The passed vp's (when matching) are not adjusted. This
134 * routine may only be called by union_newupper() and union_newlower().
138 union_updatevp(struct union_node *un, struct vnode *uppervp,
139 struct vnode *lowervp)
141 int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp);
142 int nhash = UNION_HASH(uppervp, lowervp);
143 int docache = (lowervp != NULLVP || uppervp != NULLVP);
147 * Ensure locking is ordered from lower to higher
148 * to avoid deadlocks.
158 if (lhash != uhash) {
159 while (union_list_lock(lhash))
163 while (union_list_lock(uhash))
166 if (ohash != nhash || !docache) {
167 if (un->un_flags & UN_CACHED) {
168 un->un_flags &= ~UN_CACHED;
169 LIST_REMOVE(un, un_cache);
174 union_list_unlock(ohash);
176 if (un->un_lowervp != lowervp) {
177 if (un->un_lowervp) {
178 vrele(un->un_lowervp);
180 kfree(un->un_path, M_TEMP);
184 un->un_lowervp = lowervp;
185 un->un_lowersz = VNOVAL;
188 if (un->un_uppervp != uppervp) {
190 vrele(un->un_uppervp);
191 un->un_uppervp = uppervp;
192 un->un_uppersz = VNOVAL;
195 if (docache && (ohash != nhash)) {
196 LIST_INSERT_HEAD(&unhead[nhash], un, un_cache);
197 un->un_flags |= UN_CACHED;
200 union_list_unlock(nhash);
204 * Set a new lowervp. The passed lowervp must be referenced and will be
205 * stored in the vp in a referenced state.
209 union_newlower(struct union_node *un, struct vnode *lowervp)
211 union_updatevp(un, un->un_uppervp, lowervp);
215 * Set a new uppervp. The passed uppervp must be locked and will be
216 * stored in the vp in a locked state. The caller should not unlock
221 union_newupper(struct union_node *un, struct vnode *uppervp)
223 union_updatevp(un, uppervp, un->un_lowervp);
227 * Keep track of size changes in the underlying vnodes.
228 * If the size changes, then callback to the vm layer
229 * giving priority to the upper layer size.
232 union_newsize(struct vnode *vp, off_t uppersz, off_t lowersz)
234 struct union_node *un;
237 /* only interested in regular files */
238 if (vp->v_type != VREG)
244 if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) {
245 un->un_uppersz = uppersz;
250 if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) {
251 un->un_lowersz = lowersz;
257 UDEBUG(("union: %s size now %ld\n",
258 (uppersz != VNOVAL ? "upper" : "lower"), (long)sz));
259 vnode_pager_setsize(vp, sz);
264 * union_allocvp: allocate a union_node and associate it with a
265 * parent union_node and one or two vnodes.
267 * vpp Holds the returned vnode locked and referenced if no
270 * mp Holds the mount point. mp may or may not be busied.
271 * allocvp makes no changes to mp.
273 * dvp Holds the parent union_node to the one we wish to create.
274 * XXX may only be used to traverse an uncopied lowervp-based
277 * dvp may or may not be locked. allocvp makes no changes
280 * upperdvp Holds the parent vnode to uppervp, generally used along
281 * with path component information to create a shadow of
282 * lowervp when uppervp does not exist.
284 * upperdvp is referenced but unlocked on entry, and will be
285 * dereferenced on return.
287 * uppervp Holds the new uppervp vnode to be stored in the
288 * union_node we are allocating. uppervp is referenced but
289 * not locked, and will be dereferenced on return.
291 * lowervp Holds the new lowervp vnode to be stored in the
292 * union_node we are allocating. lowervp is referenced but
293 * not locked, and will be dereferenced on return.
295 * cnp Holds path component information to be coupled with
296 * lowervp and upperdvp to allow unionfs to create an uppervp
297 * later on. Only used if lowervp is valid. The conents
298 * of cnp is only valid for the duration of the call.
300 * docache Determine whether this node should be entered in the
301 * cache or whether it should be destroyed as soon as possible.
303 * all union_nodes are maintained on a singly-linked
304 * list. new nodes are only allocated when they cannot
305 * be found on this list. entries on the list are
306 * removed when the vfs reclaim entry is called.
308 * a single lock is kept for the entire list. this is
309 * needed because the getnewvnode() function can block
310 * waiting for a vnode to become free, in which case there
311 * may be more than one process trying to get the same
312 * vnode. this lock is only taken if we are going to
313 * call getnewvnode, since the kernel itself is single-threaded.
315 * if an entry is found on the list, then call vget() to
316 * take a reference. this is done because there may be
317 * zero references to it and so it needs to removed from
318 * the vnode free list.
322 union_allocvp(struct vnode **vpp,
324 struct vnode *dvp, /* parent union vnode */
325 struct vnode *upperdvp, /* parent vnode of uppervp */
326 struct componentname *cnp, /* may be null */
327 struct vnode *uppervp, /* may be null */
328 struct vnode *lowervp, /* may be null */
332 struct union_node *un = NULL;
333 struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
334 struct thread *td = (cnp) ? cnp->cn_td : curthread; /* XXX */
339 if (uppervp == NULLVP && lowervp == NULLVP)
340 panic("union: unidentifiable allocation");
342 if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) {
347 /* detect the root vnode (and aliases) */
349 if ((uppervp == um->um_uppervp) &&
350 ((lowervp == NULLVP) || lowervp == um->um_lowervp)) {
351 if (lowervp == NULLVP) {
352 lowervp = um->um_lowervp;
353 if (lowervp != NULLVP)
362 } else for (try = 0; try < 3; try++) {
365 if (lowervp == NULLVP)
367 hash = UNION_HASH(uppervp, lowervp);
371 if (uppervp == NULLVP)
373 hash = UNION_HASH(uppervp, NULLVP);
377 if (lowervp == NULLVP)
379 hash = UNION_HASH(NULLVP, lowervp);
383 while (union_list_lock(hash))
386 for (un = unhead[hash].lh_first; un != NULL;
387 un = un->un_cache.le_next) {
388 if ((un->un_lowervp == lowervp ||
389 un->un_lowervp == NULLVP) &&
390 (un->un_uppervp == uppervp ||
391 un->un_uppervp == NULLVP) &&
392 (UNIONTOV(un)->v_mount == mp)) {
393 if (vget(UNIONTOV(un), LK_EXCLUSIVE|LK_SLEEPFAIL)) {
394 union_list_unlock(hash);
401 union_list_unlock(hash);
409 * Obtain a lock on the union_node. Everything is unlocked
410 * except for dvp, so check that case. If they match, our
411 * new un is already locked. Otherwise we have to lock our
414 * A potential deadlock situation occurs when we are holding
415 * one lock while trying to get another. We must follow
416 * strict ordering rules to avoid it. We try to locate dvp
417 * by scanning up from un_vnode, since the most likely
418 * scenario is un being under dvp.
421 if (dvp && un->un_vnode != dvp) {
422 struct vnode *scan = un->un_vnode;
425 scan = VTOUNION(scan)->un_pvp;
426 } while (scan && scan->v_tag == VT_UNION && scan != dvp);
429 * our new un is above dvp (we never saw dvp
430 * while moving up the tree).
434 error = vn_lock(un->un_vnode, LK_EXCLUSIVE);
435 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
439 * our new un is under dvp
441 error = vn_lock(un->un_vnode, LK_EXCLUSIVE);
443 } else if (dvp == NULLVP) {
445 * dvp is NULL, we need to lock un.
447 error = vn_lock(un->un_vnode, LK_EXCLUSIVE);
450 * dvp == un->un_vnode, we are already locked.
459 * At this point, the union_node is locked and referenced.
461 * uppervp is locked and referenced or NULL, lowervp is
462 * referenced or NULL.
464 UDEBUG(("Modify existing un %p vn %p upper %p(refs %d) -> %p(refs %d)\n",
465 un, un->un_vnode, un->un_uppervp,
466 (un->un_uppervp ? un->un_uppervp->v_sysref.refcnt : -99),
468 (uppervp ? uppervp->v_sysref.refcnt : -99)
471 if (uppervp != un->un_uppervp) {
472 KASSERT(uppervp == NULL || uppervp->v_sysref.refcnt > 0, ("union_allocvp: too few refs %d (at least 1 required) on uppervp", uppervp->v_sysref.refcnt));
473 union_newupper(un, uppervp);
474 } else if (uppervp) {
475 KASSERT(uppervp->v_sysref.refcnt > 1, ("union_allocvp: too few refs %d (at least 2 required) on uppervp", uppervp->v_sysref.refcnt));
480 * Save information about the lower layer.
481 * This needs to keep track of pathname
482 * and directory information which union_vn_create
485 if (lowervp != un->un_lowervp) {
486 union_newlower(un, lowervp);
487 if (cnp && (lowervp != NULLVP)) {
488 un->un_path = malloc(cnp->cn_namelen+1,
490 bcopy(cnp->cn_nameptr, un->un_path,
492 un->un_path[cnp->cn_namelen] = '\0';
494 } else if (lowervp) {
501 if (upperdvp != un->un_dirvp) {
504 un->un_dirvp = upperdvp;
505 } else if (upperdvp) {
515 * otherwise lock the vp list while we call getnewvnode
516 * since that can block.
518 hash = UNION_HASH(uppervp, lowervp);
520 if (union_list_lock(hash))
525 * Create new node rather then replace old node
528 error = getnewvnode(VT_UNION, mp, vpp, 0, 0);
531 * If an error occurs clear out vnodes.
543 (*vpp)->v_data = kmalloc(sizeof(struct union_node), M_TEMP, M_WAITOK);
545 vsetflags(*vpp, vflag);
547 (*vpp)->v_type = uppervp->v_type;
549 (*vpp)->v_type = lowervp->v_type;
552 bzero(un, sizeof(*un));
555 un->un_uppervp = uppervp;
556 un->un_uppersz = VNOVAL;
557 un->un_lowervp = lowervp;
558 un->un_lowersz = VNOVAL;
559 un->un_dirvp = upperdvp;
560 un->un_pvp = dvp; /* only parent dir in new allocation */
566 if (cnp && (lowervp != NULLVP)) {
567 un->un_path = kmalloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK);
568 bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen);
569 un->un_path[cnp->cn_namelen] = '\0';
576 LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
577 un->un_flags |= UN_CACHED;
581 * locked refd vpp is returned
586 union_list_unlock(hash);
592 union_freevp(struct vnode *vp)
594 struct union_node *un = VTOUNION(vp);
597 if (un->un_flags & UN_CACHED) {
598 un->un_flags &= ~UN_CACHED;
599 LIST_REMOVE(un, un_cache);
601 if (un->un_pvp != NULLVP) {
605 if (un->un_uppervp != NULLVP) {
606 vrele(un->un_uppervp);
607 un->un_uppervp = NULL;
609 if (un->un_lowervp != NULLVP) {
610 vrele(un->un_lowervp);
611 un->un_lowervp = NULL;
613 if (un->un_dirvp != NULLVP) {
618 kfree(un->un_path, M_TEMP);
626 * copyfile. copy the vnode (fvp) to the vnode (tvp)
627 * using a sequence of reads and writes. both (fvp)
628 * and (tvp) are locked on entry and exit.
630 * fvp and tvp are both exclusive locked on call, but their refcount's
631 * haven't been bumped at all.
634 union_copyfile(struct vnode *fvp, struct vnode *tvp, struct ucred *cred,
644 * allocate a buffer of size MAXBSIZE.
645 * loop doing reads and writes, keeping track
646 * of the current uio offset.
647 * give up at the first sign of trouble.
650 bzero(&uio, sizeof(uio));
653 uio.uio_segflg = UIO_SYSSPACE;
656 buf = kmalloc(MAXBSIZE, M_TEMP, M_WAITOK);
658 /* ugly loop follows... */
660 off_t offset = uio.uio_offset;
670 iov.iov_len = MAXBSIZE;
671 uio.uio_resid = iov.iov_len;
672 uio.uio_rw = UIO_READ;
674 if ((error = VOP_READ(fvp, &uio, 0, cred)) != 0)
678 * Get bytes read, handle read eof case and setup for
681 if ((count = MAXBSIZE - uio.uio_resid) == 0)
686 * Write until an error occurs or our buffer has been
687 * exhausted, then update the offset for the next read.
689 while (bufoffset < count) {
692 iov.iov_base = buf + bufoffset;
693 iov.iov_len = count - bufoffset;
694 uio.uio_offset = offset + bufoffset;
695 uio.uio_rw = UIO_WRITE;
696 uio.uio_resid = iov.iov_len;
698 if ((error = VOP_WRITE(tvp, &uio, 0, cred)) != 0)
700 bufoffset += (count - bufoffset) - uio.uio_resid;
702 uio.uio_offset = offset + bufoffset;
703 } while (error == 0);
711 * un's vnode is assumed to be locked on entry and remains locked on exit.
715 union_copyup(struct union_node *un, int docopy, struct ucred *cred,
719 struct vnode *lvp, *uvp;
722 * If the user does not have read permission, the vnode should not
723 * be copied to upper layer.
725 vn_lock(un->un_lowervp, LK_EXCLUSIVE | LK_RETRY);
726 error = VOP_EACCESS(un->un_lowervp, VREAD, cred);
727 vn_unlock(un->un_lowervp);
731 error = union_vn_create(&uvp, un, td);
735 lvp = un->un_lowervp;
737 KASSERT(uvp->v_sysref.refcnt > 0, ("copy: uvp refcount 0: %d", uvp->v_sysref.refcnt));
740 * XX - should not ignore errors
743 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY);
744 error = VOP_OPEN(lvp, FREAD, cred, NULL);
746 error = union_copyfile(lvp, uvp, cred, td);
748 (void) VOP_CLOSE(lvp, FREAD);
751 UDEBUG(("union: copied up %s\n", un->un_path));
755 union_newupper(un, uvp);
756 KASSERT(uvp->v_sysref.refcnt > 0, ("copy: uvp refcount 0: %d", uvp->v_sysref.refcnt));
757 union_vn_close(uvp, FWRITE, cred);
758 KASSERT(uvp->v_sysref.refcnt > 0, ("copy: uvp refcount 0: %d", uvp->v_sysref.refcnt));
760 * Subsequent IOs will go to the top layer, so
761 * call close on the lower vnode and open on the
762 * upper vnode to ensure that the filesystem keeps
763 * its references counts right. This doesn't do
764 * the right thing with (cred) and (FREAD) though.
765 * Ignoring error returns is not right, either.
770 for (i = 0; i < un->un_openl; i++) {
771 VOP_CLOSE(lvp, FREAD);
772 VOP_OPEN(uvp, FREAD, cred, NULL);
784 * dvp should be locked on entry and will be locked on return. No
785 * net change in the ref count will occur.
787 * If an error is returned, *vpp will be invalid, otherwise it
788 * will hold a locked, referenced vnode. If *vpp == dvp then
789 * remember that only one exclusive lock is held.
793 union_relookup(struct union_mount *um, struct vnode *dvp, struct vnode **vpp,
794 struct componentname *cnp, struct componentname *cn, char *path,
800 * A new componentname structure must be faked up because
801 * there is no way to know where the upper level cnp came
802 * from or what it is being used for. This must duplicate
803 * some of the work done by NDINIT, some of the work done
804 * by namei, some of the work done by lookup and some of
805 * the work done by VOP_LOOKUP when given a CREATE flag.
806 * Conclusion: Horrible.
808 cn->cn_namelen = pathlen;
809 cn->cn_nameptr = objcache_get(namei_oc, M_WAITOK);
810 bcopy(path, cn->cn_nameptr, cn->cn_namelen);
811 cn->cn_nameptr[cn->cn_namelen] = '\0';
813 cn->cn_nameiop = NAMEI_CREATE;
814 cn->cn_flags = CNP_LOCKPARENT;
815 cn->cn_td = cnp->cn_td;
816 if (um->um_op == UNMNT_ABOVE)
817 cn->cn_cred = cnp->cn_cred;
819 cn->cn_cred = um->um_cred;
820 cn->cn_consume = cnp->cn_consume;
826 * Pass dvp unlocked and referenced on call to relookup().
828 * If an error occurs, dvp will be returned unlocked and dereferenced.
831 if ((error = relookup(dvp, vpp, cn)) != 0) {
832 objcache_put(namei_oc, cn->cn_nameptr);
833 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
836 objcache_put(namei_oc, cn->cn_nameptr);
839 * If no error occurs, dvp will be returned locked with the reference
840 * left as before, and vpp will be returned referenced and locked.
842 * We want to return with dvp as it was passed to us, so we get
843 * rid of our reference.
850 * Create a shadow directory in the upper layer.
851 * The new vnode is returned locked.
853 * (um) points to the union mount structure for access to the
854 * the mounting process's credentials.
855 * (dvp) is the directory in which to create the shadow directory,
856 * it is locked (but not ref'd) on entry and return.
857 * (cnp) is the componentname to be created.
858 * (vpp) is the returned newly created shadow directory, which
859 * is returned locked and ref'd
862 union_mkshadow(struct union_mount *um, struct vnode *dvp,
863 struct componentname *cnp, struct vnode **vpp)
867 struct thread *td = cnp->cn_td;
868 struct componentname cn;
870 error = union_relookup(um, dvp, vpp, cnp, &cn,
871 cnp->cn_nameptr, cnp->cn_namelen);
885 * policy: when creating the shadow directory in the
886 * upper layer, create it owned by the user who did
887 * the mount, group from parent directory, and mode
888 * 777 modified by umask (ie mostly identical to the
889 * mkdir syscall). (jsp, kb)
894 va.va_mode = um->um_cmode;
896 error = VOP_MKDIR(dvp, vpp, &cn, &va);
902 * Create a whiteout entry in the upper layer.
904 * (um) points to the union mount structure for access to the
905 * the mounting process's credentials.
906 * (dvp) is the directory in which to create the whiteout.
907 * it is locked on entry and return.
908 * (cnp) is the componentname to be created.
911 union_mkwhiteout(struct union_mount *um, struct vnode *dvp,
912 struct componentname *cnp, char *path)
915 struct thread *td = cnp->cn_td;
917 struct componentname cn;
920 KKASSERT(td->td_proc);
921 cred = td->td_proc->p_ucred;
923 error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path));
935 error = VOP_WHITEOUT(dvp, &cn, NAMEI_CREATE);
940 * union_vn_create: creates and opens a new shadow file
941 * on the upper union layer. this function is similar
942 * in spirit to calling vn_open but it avoids calling namei().
943 * the problem with calling namei is that a) it locks too many
944 * things, and b) it doesn't start at the "right" directory,
945 * whereas relookup is told where to start.
947 * On entry, the vnode associated with un is locked. It remains locked
950 * If no error occurs, *vpp contains a locked referenced vnode for your
951 * use. If an error occurs *vpp iis undefined.
954 union_vn_create(struct vnode **vpp, struct union_node *un, struct thread *td)
959 struct vattr *vap = &vat;
960 int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL);
963 struct componentname cn;
965 KKASSERT(td->td_proc);
966 cred = td->td_proc->p_ucred;
967 cmode = UN_FILEMODE & ~td->td_proc->p_fd->fd_cmask;
972 * Build a new componentname structure (for the same
973 * reasons outlines in union_mkshadow).
974 * The difference here is that the file is owned by
975 * the current user, rather than by the person who
976 * did the mount, since the current user needs to be
977 * able to write the file (that's why it is being
978 * copied in the first place).
980 cn.cn_namelen = strlen(un->un_path);
981 cn.cn_nameptr = objcache_get(namei_oc, M_WAITOK);
982 bcopy(un->un_path, cn.cn_nameptr, cn.cn_namelen+1);
983 cn.cn_nameiop = NAMEI_CREATE;
984 cn.cn_flags = CNP_LOCKPARENT;
990 * Pass dvp unlocked and referenced on call to relookup().
992 * If an error occurs, dvp will be returned unlocked and dereferenced.
995 error = relookup(un->un_dirvp, &vp, &cn);
996 objcache_put(namei_oc, cn.cn_nameptr);
1001 * If no error occurs, dvp will be returned locked with the reference
1002 * left as before, and vpp will be returned referenced and locked.
1006 if (vp == un->un_dirvp)
1014 * Good - there was no race to create the file
1015 * so go ahead and create it. The permissions
1016 * on the file will be 0666 modified by the
1017 * current user's umask. Access to the file, while
1018 * it is unioned, will require access to the top *and*
1019 * bottom files. Access when not unioned will simply
1020 * require access to the top-level file.
1021 * TODO: confirm choice of access permissions.
1024 vap->va_type = VREG;
1025 vap->va_mode = cmode;
1026 error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap);
1031 error = VOP_OPEN(vp, fmode, cred, NULL);
1041 union_vn_close(struct vnode *vp, int fmode, struct ucred *cred)
1043 return (VOP_CLOSE(vp, fmode));
1049 * union_removed_upper:
1051 * called with union_node unlocked. XXX
1055 union_removed_upper(struct union_node *un)
1057 struct thread *td = curthread; /* XXX */
1061 * Do not set the uppervp to NULLVP. If lowervp is NULLVP,
1062 * union node will have neither uppervp nor lowervp. We remove
1063 * the union node from cache, so that it will not be referrenced.
1065 union_newupper(un, NULLVP);
1066 if (un->un_dircache != 0) {
1067 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1069 kfree(un->un_dircache, M_TEMP);
1070 un->un_dircache = 0;
1073 if (un->un_flags & UN_CACHED) {
1074 un->un_flags &= ~UN_CACHED;
1075 LIST_REMOVE(un, un_cache);
1082 * determine whether a whiteout is needed
1083 * during a remove/rmdir operation.
1086 union_dowhiteout(struct union_node *un, struct ucred *cred, struct thread *td)
1090 if (un->un_lowervp != NULLVP)
1093 if (VOP_GETATTR(un->un_uppervp, &va) == 0 &&
1094 (va.va_flags & OPAQUE))
1101 union_dircache_r(struct vnode *vp, struct vnode ***vppp, int *cntp)
1103 struct union_node *un;
1105 if (vp->v_tag != VT_UNION) {
1110 panic("union: dircache table too small");
1119 if (un->un_uppervp != NULLVP)
1120 union_dircache_r(un->un_uppervp, vppp, cntp);
1121 if (un->un_lowervp != NULLVP)
1122 union_dircache_r(un->un_lowervp, vppp, cntp);
1126 union_dircache(struct vnode *vp, struct thread *td)
1131 struct vnode **dircache;
1132 struct union_node *un;
1135 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1136 dircache = VTOUNION(vp)->un_dircache;
1140 if (dircache == NULL) {
1142 union_dircache_r(vp, 0, &cnt);
1144 dircache = malloc(cnt * sizeof(struct vnode *),
1147 union_dircache_r(vp, &vpp, &cnt);
1153 if (*vpp++ == VTOUNION(vp)->un_uppervp)
1155 } while (*vpp != NULLVP);
1161 /*vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);*/
1162 UDEBUG(("ALLOCVP-3 %p ref %d\n", *vpp, (*vpp ? (*vpp)->v_sysref.refcnt : -99)));
1164 error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, NULL, *vpp, NULLVP, 0);
1165 UDEBUG(("ALLOCVP-3B %p ref %d\n", nvp, (*vpp ? (*vpp)->v_sysref.refcnt : -99)));
1169 VTOUNION(vp)->un_dircache = 0;
1171 un->un_dircache = dircache;
1179 * Guarentee coherency with the VM cache by invalidating any clean VM pages
1180 * associated with this write and updating any dirty VM pages. Since our
1181 * vnode is locked, other processes will not be able to read the pages in
1182 * again until after our write completes.
1184 * We also have to be coherent with reads, by flushing any pending dirty
1185 * pages prior to issuing the read.
1187 * XXX this is somewhat of a hack at the moment. To support this properly
1188 * we would have to be able to run VOP_READ and VOP_WRITE through the VM
1189 * cache. Then we wouldn't need to worry about coherency.
1193 union_vm_coherency(struct vnode *vp, struct uio *uio, int cleanfls)
1200 if ((object = vp->v_object) == NULL)
1203 pgoff = uio->uio_offset & PAGE_MASK;
1204 pstart = uio->uio_offset / PAGE_SIZE;
1205 pend = pstart + (uio->uio_resid + pgoff + PAGE_MASK) / PAGE_SIZE;
1207 vm_object_page_clean(object, pstart, pend, OBJPC_SYNC);
1209 vm_object_page_remove(object, pstart, pend, TRUE);
1213 * Module glue to remove #ifdef UNION from vfs_syscalls.c
1216 union_dircheck(struct thread *td, struct vnode **vp, struct file *fp)
1220 if ((*vp)->v_tag == VT_UNION) {
1223 lvp = union_dircache(*vp, td);
1224 if (lvp != NULLVP) {
1228 * If the directory is opaque,
1229 * then don't show lower entries
1231 error = VOP_GETATTR(*vp, &va);
1232 if (va.va_flags & OPAQUE) {
1238 if (lvp != NULLVP) {
1239 error = VOP_OPEN(lvp, FREAD, fp->f_cred, NULL);
1247 error = vn_close(*vp, FREAD);
1251 return -1; /* goto unionread */
1258 union_modevent(module_t mod, int type, void *data)
1262 union_dircheckp = union_dircheck;
1265 union_dircheckp = NULL;
1273 static moduledata_t union_mod = {
1279 DECLARE_MODULE(union_dircheck, union_mod, SI_SUB_VFS, SI_ORDER_ANY);