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
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/vnode.h>
45 #include <sys/namei.h>
46 #include <sys/malloc.h>
47 #include <sys/fcntl.h>
49 #include <sys/filedesc.h>
50 #include <sys/module.h>
51 #include <sys/mount.h>
54 #include <vm/vm_extern.h> /* for vnode_pager_setsize */
55 #include <vm/vm_zone.h>
56 #include <vm/vm_object.h> /* for vm cache coherency */
57 #include <miscfs/union/union.h>
61 extern int union_init __P((void));
63 /* must be power of two, otherwise change UNION_HASH() */
66 /* unsigned int ... */
67 #define UNION_HASH(u, l) \
68 (((((uintptr_t) (u)) + ((uintptr_t) l)) >> 8) & (NHASH-1))
70 static LIST_HEAD(unhead, struct union_node) unhead[NHASH];
71 static int unvplock[NHASH];
73 static void union_dircache_r __P((struct vnode *vp, struct vnode ***vppp,
75 static int union_list_lock __P((int ix));
76 static void union_list_unlock __P((int ix));
77 static int union_relookup __P((struct union_mount *um, struct vnode *dvp,
79 struct componentname *cnp,
80 struct componentname *cn, char *path,
82 static void union_updatevp __P((struct union_node *un,
83 struct vnode *uppervp,
84 struct vnode *lowervp));
85 static void union_newlower __P((struct union_node *, struct vnode *));
86 static void union_newupper __P((struct union_node *, struct vnode *));
87 static int union_copyfile __P((struct vnode *, struct vnode *,
88 struct ucred *, struct proc *));
89 static int union_vn_create __P((struct vnode **, struct union_node *,
91 static int union_vn_close __P((struct vnode *, int, struct ucred *,
99 for (i = 0; i < NHASH; i++)
100 LIST_INIT(&unhead[i]);
101 bzero((caddr_t)unvplock, sizeof(unvplock));
109 if (unvplock[ix] & UNVP_LOCKED) {
110 unvplock[ix] |= UNVP_WANT;
111 (void) tsleep((caddr_t) &unvplock[ix], PINOD, "unllck", 0);
114 unvplock[ix] |= UNVP_LOCKED;
119 union_list_unlock(ix)
122 unvplock[ix] &= ~UNVP_LOCKED;
124 if (unvplock[ix] & UNVP_WANT) {
125 unvplock[ix] &= ~UNVP_WANT;
126 wakeup((caddr_t) &unvplock[ix]);
133 * The uppervp, if not NULL, must be referenced and not locked by us
134 * The lowervp, if not NULL, must be referenced.
136 * if uppervp and lowervp match pointers already installed, nothing
137 * happens. The passed vp's (when matching) are not adjusted. This
138 * routine may only be called by union_newupper() and union_newlower().
142 union_updatevp(un, uppervp, lowervp)
143 struct union_node *un;
144 struct vnode *uppervp;
145 struct vnode *lowervp;
147 int ohash = UNION_HASH(un->un_uppervp, un->un_lowervp);
148 int nhash = UNION_HASH(uppervp, lowervp);
149 int docache = (lowervp != NULLVP || uppervp != NULLVP);
153 * Ensure locking is ordered from lower to higher
154 * to avoid deadlocks.
164 if (lhash != uhash) {
165 while (union_list_lock(lhash))
169 while (union_list_lock(uhash))
172 if (ohash != nhash || !docache) {
173 if (un->un_flags & UN_CACHED) {
174 un->un_flags &= ~UN_CACHED;
175 LIST_REMOVE(un, un_cache);
180 union_list_unlock(ohash);
182 if (un->un_lowervp != lowervp) {
183 if (un->un_lowervp) {
184 vrele(un->un_lowervp);
186 free(un->un_path, M_TEMP);
190 un->un_lowervp = lowervp;
191 un->un_lowersz = VNOVAL;
194 if (un->un_uppervp != uppervp) {
196 vrele(un->un_uppervp);
197 un->un_uppervp = uppervp;
198 un->un_uppersz = VNOVAL;
201 if (docache && (ohash != nhash)) {
202 LIST_INSERT_HEAD(&unhead[nhash], un, un_cache);
203 un->un_flags |= UN_CACHED;
206 union_list_unlock(nhash);
210 * Set a new lowervp. The passed lowervp must be referenced and will be
211 * stored in the vp in a referenced state.
215 union_newlower(un, lowervp)
216 struct union_node *un;
217 struct vnode *lowervp;
219 union_updatevp(un, un->un_uppervp, lowervp);
223 * Set a new uppervp. The passed uppervp must be locked and will be
224 * stored in the vp in a locked state. The caller should not unlock
229 union_newupper(un, uppervp)
230 struct union_node *un;
231 struct vnode *uppervp;
233 union_updatevp(un, uppervp, un->un_lowervp);
237 * Keep track of size changes in the underlying vnodes.
238 * If the size changes, then callback to the vm layer
239 * giving priority to the upper layer size.
242 union_newsize(vp, uppersz, lowersz)
244 off_t uppersz, lowersz;
246 struct union_node *un;
249 /* only interested in regular files */
250 if (vp->v_type != VREG)
256 if ((uppersz != VNOVAL) && (un->un_uppersz != uppersz)) {
257 un->un_uppersz = uppersz;
262 if ((lowersz != VNOVAL) && (un->un_lowersz != lowersz)) {
263 un->un_lowersz = lowersz;
269 UDEBUG(("union: %s size now %ld\n",
270 (uppersz != VNOVAL ? "upper" : "lower"), (long)sz));
271 vnode_pager_setsize(vp, sz);
276 * union_allocvp: allocate a union_node and associate it with a
277 * parent union_node and one or two vnodes.
279 * vpp Holds the returned vnode locked and referenced if no
282 * mp Holds the mount point. mp may or may not be busied.
283 * allocvp makes no changes to mp.
285 * dvp Holds the parent union_node to the one we wish to create.
286 * XXX may only be used to traverse an uncopied lowervp-based
289 * dvp may or may not be locked. allocvp makes no changes
292 * upperdvp Holds the parent vnode to uppervp, generally used along
293 * with path component information to create a shadow of
294 * lowervp when uppervp does not exist.
296 * upperdvp is referenced but unlocked on entry, and will be
297 * dereferenced on return.
299 * uppervp Holds the new uppervp vnode to be stored in the
300 * union_node we are allocating. uppervp is referenced but
301 * not locked, and will be dereferenced on return.
303 * lowervp Holds the new lowervp vnode to be stored in the
304 * union_node we are allocating. uppervp is referenced but
305 * not locked, and will be dereferenced on return.
307 * cnp Holds path component information to be coupled with
308 * lowervp and upperdvp to allow unionfs to create an uppervp
309 * later on. Only used if lowervp is valid. The conents
310 * of cnp is only valid for the duration of the call.
312 * docache Determine whether this node should be entered in the
313 * cache or whether it should be destroyed as soon as possible.
315 * all union_nodes are maintained on a singly-linked
316 * list. new nodes are only allocated when they cannot
317 * be found on this list. entries on the list are
318 * removed when the vfs reclaim entry is called.
320 * a single lock is kept for the entire list. this is
321 * needed because the getnewvnode() function can block
322 * waiting for a vnode to become free, in which case there
323 * may be more than one process trying to get the same
324 * vnode. this lock is only taken if we are going to
325 * call getnewvnode, since the kernel itself is single-threaded.
327 * if an entry is found on the list, then call vget() to
328 * take a reference. this is done because there may be
329 * zero references to it and so it needs to removed from
330 * the vnode free list.
334 union_allocvp(vpp, mp, dvp, upperdvp, cnp, uppervp, lowervp, docache)
337 struct vnode *dvp; /* parent union vnode */
338 struct vnode *upperdvp; /* parent vnode of uppervp */
339 struct componentname *cnp; /* may be null */
340 struct vnode *uppervp; /* may be null */
341 struct vnode *lowervp; /* may be null */
345 struct union_node *un = 0;
346 struct vnode *xlowervp = NULLVP;
347 struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
348 struct proc *p = (cnp) ? cnp->cn_proc : curproc;
353 if (uppervp == NULLVP && lowervp == NULLVP)
354 panic("union: unidentifiable allocation");
356 if (uppervp && lowervp && (uppervp->v_type != lowervp->v_type)) {
361 /* detect the root vnode (and aliases) */
363 if ((uppervp == um->um_uppervp) &&
364 ((lowervp == NULLVP) || lowervp == um->um_lowervp)) {
365 if (lowervp == NULLVP) {
366 lowervp = um->um_lowervp;
367 if (lowervp != NULLVP)
376 } else for (try = 0; try < 3; try++) {
379 if (lowervp == NULLVP)
381 hash = UNION_HASH(uppervp, lowervp);
385 if (uppervp == NULLVP)
387 hash = UNION_HASH(uppervp, NULLVP);
391 if (lowervp == NULLVP)
393 hash = UNION_HASH(NULLVP, lowervp);
397 while (union_list_lock(hash))
400 for (un = unhead[hash].lh_first; un != 0;
401 un = un->un_cache.le_next) {
402 if ((un->un_lowervp == lowervp ||
403 un->un_lowervp == NULLVP) &&
404 (un->un_uppervp == uppervp ||
405 un->un_uppervp == NULLVP) &&
406 (UNIONTOV(un)->v_mount == mp)) {
407 if (vget(UNIONTOV(un), 0,
408 cnp ? cnp->cn_proc : NULL)) {
409 union_list_unlock(hash);
416 union_list_unlock(hash);
424 * Obtain a lock on the union_node. Everything is unlocked
425 * except for dvp, so check that case. If they match, our
426 * new un is already locked. Otherwise we have to lock our
429 * A potential deadlock situation occurs when we are holding
430 * one lock while trying to get another. We must follow
431 * strict ordering rules to avoid it. We try to locate dvp
432 * by scanning up from un_vnode, since the most likely
433 * scenario is un being under dvp.
436 if (dvp && un->un_vnode != dvp) {
437 struct vnode *scan = un->un_vnode;
440 scan = VTOUNION(scan)->un_pvp;
441 } while (scan && scan->v_tag == VT_UNION && scan != dvp);
444 * our new un is above dvp (we never saw dvp
445 * while moving up the tree).
448 VOP_UNLOCK(dvp, 0, p);
449 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, p);
450 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
454 * our new un is under dvp
456 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, p);
458 } else if (dvp == NULLVP) {
460 * dvp is NULL, we need to lock un.
462 error = vn_lock(un->un_vnode, LK_EXCLUSIVE, p);
465 * dvp == un->un_vnode, we are already locked.
474 * At this point, the union_node is locked and referenced.
476 * uppervp is locked and referenced or NULL, lowervp is
477 * referenced or NULL.
479 UDEBUG(("Modify existing un %p vn %p upper %p(refs %d) -> %p(refs %d)\n",
480 un, un->un_vnode, un->un_uppervp,
481 (un->un_uppervp ? un->un_uppervp->v_usecount : -99),
483 (uppervp ? uppervp->v_usecount : -99)
486 if (uppervp != un->un_uppervp) {
487 KASSERT(uppervp == NULL || uppervp->v_usecount > 0, ("union_allocvp: too few refs %d (at least 1 required) on uppervp", uppervp->v_usecount));
488 union_newupper(un, uppervp);
489 } else if (uppervp) {
490 KASSERT(uppervp->v_usecount > 1, ("union_allocvp: too few refs %d (at least 2 required) on uppervp", uppervp->v_usecount));
495 * Save information about the lower layer.
496 * This needs to keep track of pathname
497 * and directory information which union_vn_create
500 if (lowervp != un->un_lowervp) {
501 union_newlower(un, lowervp);
502 if (cnp && (lowervp != NULLVP)) {
503 un->un_path = malloc(cnp->cn_namelen+1,
505 bcopy(cnp->cn_nameptr, un->un_path,
507 un->un_path[cnp->cn_namelen] = '\0';
509 } else if (lowervp) {
516 if (upperdvp != un->un_dirvp) {
519 un->un_dirvp = upperdvp;
520 } else if (upperdvp) {
530 * otherwise lock the vp list while we call getnewvnode
531 * since that can block.
533 hash = UNION_HASH(uppervp, lowervp);
535 if (union_list_lock(hash))
540 * Create new node rather then replace old node
543 error = getnewvnode(VT_UNION, mp, union_vnodeop_p, vpp);
546 * If an error occurs clear out vnodes.
558 MALLOC((*vpp)->v_data, void *, sizeof(struct union_node),
561 (*vpp)->v_flag |= vflag;
563 (*vpp)->v_type = uppervp->v_type;
565 (*vpp)->v_type = lowervp->v_type;
568 bzero(un, sizeof(*un));
570 lockinit(&un->un_lock, PVFS, "unlock", 0, 0);
571 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, p);
574 un->un_uppervp = uppervp;
575 un->un_uppersz = VNOVAL;
576 un->un_lowervp = lowervp;
577 un->un_lowersz = VNOVAL;
578 un->un_dirvp = upperdvp;
579 un->un_pvp = dvp; /* only parent dir in new allocation */
585 if (cnp && (lowervp != NULLVP)) {
586 un->un_path = malloc(cnp->cn_namelen+1, M_TEMP, M_WAITOK);
587 bcopy(cnp->cn_nameptr, un->un_path, cnp->cn_namelen);
588 un->un_path[cnp->cn_namelen] = '\0';
595 LIST_INSERT_HEAD(&unhead[hash], un, un_cache);
596 un->un_flags |= UN_CACHED;
604 union_list_unlock(hash);
613 struct union_node *un = VTOUNION(vp);
615 if (un->un_flags & UN_CACHED) {
616 un->un_flags &= ~UN_CACHED;
617 LIST_REMOVE(un, un_cache);
620 if (un->un_pvp != NULLVP) {
624 if (un->un_uppervp != NULLVP) {
625 vrele(un->un_uppervp);
626 un->un_uppervp = NULL;
628 if (un->un_lowervp != NULLVP) {
629 vrele(un->un_lowervp);
630 un->un_lowervp = NULL;
632 if (un->un_dirvp != NULLVP) {
637 free(un->un_path, M_TEMP);
641 FREE(vp->v_data, M_TEMP);
648 * copyfile. copy the vnode (fvp) to the vnode (tvp)
649 * using a sequence of reads and writes. both (fvp)
650 * and (tvp) are locked on entry and exit.
652 * fvp and tvp are both exclusive locked on call, but their refcount's
653 * haven't been bumped at all.
656 union_copyfile(fvp, tvp, cred, p)
669 * allocate a buffer of size MAXBSIZE.
670 * loop doing reads and writes, keeping track
671 * of the current uio offset.
672 * give up at the first sign of trouble.
675 bzero(&uio, sizeof(uio));
678 uio.uio_segflg = UIO_SYSSPACE;
681 VOP_LEASE(fvp, p, cred, LEASE_READ);
682 VOP_LEASE(tvp, p, cred, LEASE_WRITE);
684 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
686 /* ugly loop follows... */
688 off_t offset = uio.uio_offset;
698 iov.iov_len = MAXBSIZE;
699 uio.uio_resid = iov.iov_len;
700 uio.uio_rw = UIO_READ;
702 if ((error = VOP_READ(fvp, &uio, 0, cred)) != 0)
706 * Get bytes read, handle read eof case and setup for
709 if ((count = MAXBSIZE - uio.uio_resid) == 0)
714 * Write until an error occurs or our buffer has been
715 * exhausted, then update the offset for the next read.
717 while (bufoffset < count) {
720 iov.iov_base = buf + bufoffset;
721 iov.iov_len = count - bufoffset;
722 uio.uio_offset = offset + bufoffset;
723 uio.uio_rw = UIO_WRITE;
724 uio.uio_resid = iov.iov_len;
726 if ((error = VOP_WRITE(tvp, &uio, 0, cred)) != 0)
728 bufoffset += (count - bufoffset) - uio.uio_resid;
730 uio.uio_offset = offset + bufoffset;
731 } while (error == 0);
739 * un's vnode is assumed to be locked on entry and remains locked on exit.
743 union_copyup(un, docopy, cred, p)
744 struct union_node *un;
750 struct vnode *lvp, *uvp;
753 * If the user does not have read permission, the vnode should not
754 * be copied to upper layer.
756 vn_lock(un->un_lowervp, LK_EXCLUSIVE | LK_RETRY, p);
757 error = VOP_ACCESS(un->un_lowervp, VREAD, cred, p);
758 VOP_UNLOCK(un->un_lowervp, 0, p);
762 error = union_vn_create(&uvp, un, p);
766 lvp = un->un_lowervp;
768 KASSERT(uvp->v_usecount > 0, ("copy: uvp refcount 0: %d", uvp->v_usecount));
771 * XX - should not ignore errors
774 vn_lock(lvp, LK_EXCLUSIVE | LK_RETRY, p);
775 error = VOP_OPEN(lvp, FREAD, cred, p);
776 if (error == 0 && vn_canvmio(lvp) == TRUE)
777 error = vfs_object_create(lvp, p, cred);
779 error = union_copyfile(lvp, uvp, cred, p);
780 VOP_UNLOCK(lvp, 0, p);
781 (void) VOP_CLOSE(lvp, FREAD, cred, p);
784 UDEBUG(("union: copied up %s\n", un->un_path));
787 VOP_UNLOCK(uvp, 0, p);
788 union_newupper(un, uvp);
789 KASSERT(uvp->v_usecount > 0, ("copy: uvp refcount 0: %d", uvp->v_usecount));
790 union_vn_close(uvp, FWRITE, cred, p);
791 KASSERT(uvp->v_usecount > 0, ("copy: uvp refcount 0: %d", uvp->v_usecount));
793 * Subsequent IOs will go to the top layer, so
794 * call close on the lower vnode and open on the
795 * upper vnode to ensure that the filesystem keeps
796 * its references counts right. This doesn't do
797 * the right thing with (cred) and (FREAD) though.
798 * Ignoring error returns is not right, either.
803 for (i = 0; i < un->un_openl; i++) {
804 (void) VOP_CLOSE(lvp, FREAD, cred, p);
805 (void) VOP_OPEN(uvp, FREAD, cred, p);
808 if (vn_canvmio(uvp) == TRUE)
809 error = vfs_object_create(uvp, p, cred);
821 * dvp should be locked on entry and will be locked on return. No
822 * net change in the ref count will occur.
824 * If an error is returned, *vpp will be invalid, otherwise it
825 * will hold a locked, referenced vnode. If *vpp == dvp then
826 * remember that only one exclusive lock is held.
830 union_relookup(um, dvp, vpp, cnp, cn, path, pathlen)
831 struct union_mount *um;
834 struct componentname *cnp;
835 struct componentname *cn;
842 * A new componentname structure must be faked up because
843 * there is no way to know where the upper level cnp came
844 * from or what it is being used for. This must duplicate
845 * some of the work done by NDINIT, some of the work done
846 * by namei, some of the work done by lookup and some of
847 * the work done by VOP_LOOKUP when given a CREATE flag.
848 * Conclusion: Horrible.
850 cn->cn_namelen = pathlen;
851 cn->cn_pnbuf = zalloc(namei_zone);
852 bcopy(path, cn->cn_pnbuf, cn->cn_namelen);
853 cn->cn_pnbuf[cn->cn_namelen] = '\0';
855 cn->cn_nameiop = CREATE;
856 cn->cn_flags = (LOCKPARENT|LOCKLEAF|HASBUF|SAVENAME|ISLASTCN);
857 cn->cn_proc = cnp->cn_proc;
858 if (um->um_op == UNMNT_ABOVE)
859 cn->cn_cred = cnp->cn_cred;
861 cn->cn_cred = um->um_cred;
862 cn->cn_nameptr = cn->cn_pnbuf;
863 cn->cn_consume = cnp->cn_consume;
866 VOP_UNLOCK(dvp, 0, cnp->cn_proc);
869 * Pass dvp unlocked and referenced on call to relookup().
871 * If an error occurs, dvp will be returned unlocked and dereferenced.
874 if ((error = relookup(dvp, vpp, cn)) != 0) {
875 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, cnp->cn_proc);
880 * If no error occurs, dvp will be returned locked with the reference
881 * left as before, and vpp will be returned referenced and locked.
883 * We want to return with dvp as it was passed to us, so we get
884 * rid of our reference.
891 * Create a shadow directory in the upper layer.
892 * The new vnode is returned locked.
894 * (um) points to the union mount structure for access to the
895 * the mounting process's credentials.
896 * (dvp) is the directory in which to create the shadow directory,
897 * it is locked (but not ref'd) on entry and return.
898 * (cnp) is the componentname to be created.
899 * (vpp) is the returned newly created shadow directory, which
900 * is returned locked and ref'd
903 union_mkshadow(um, dvp, cnp, vpp)
904 struct union_mount *um;
906 struct componentname *cnp;
911 struct proc *p = cnp->cn_proc;
912 struct componentname cn;
914 error = union_relookup(um, dvp, vpp, cnp, &cn,
915 cnp->cn_nameptr, cnp->cn_namelen);
920 if (cn.cn_flags & HASBUF) {
921 zfree(namei_zone, cn.cn_pnbuf);
922 cn.cn_flags &= ~HASBUF;
933 * policy: when creating the shadow directory in the
934 * upper layer, create it owned by the user who did
935 * the mount, group from parent directory, and mode
936 * 777 modified by umask (ie mostly identical to the
937 * mkdir syscall). (jsp, kb)
942 va.va_mode = um->um_cmode;
944 /* VOP_LEASE: dvp is locked */
945 VOP_LEASE(dvp, p, cn.cn_cred, LEASE_WRITE);
947 error = VOP_MKDIR(dvp, vpp, &cn, &va);
948 if (cn.cn_flags & HASBUF) {
949 zfree(namei_zone, cn.cn_pnbuf);
950 cn.cn_flags &= ~HASBUF;
957 * Create a whiteout entry in the upper layer.
959 * (um) points to the union mount structure for access to the
960 * the mounting process's credentials.
961 * (dvp) is the directory in which to create the whiteout.
962 * it is locked on entry and return.
963 * (cnp) is the componentname to be created.
966 union_mkwhiteout(um, dvp, cnp, path)
967 struct union_mount *um;
969 struct componentname *cnp;
973 struct proc *p = cnp->cn_proc;
975 struct componentname cn;
977 error = union_relookup(um, dvp, &wvp, cnp, &cn, path, strlen(path));
982 if (cn.cn_flags & HASBUF) {
983 zfree(namei_zone, cn.cn_pnbuf);
984 cn.cn_flags &= ~HASBUF;
993 /* VOP_LEASE: dvp is locked */
994 VOP_LEASE(dvp, p, p->p_ucred, LEASE_WRITE);
996 error = VOP_WHITEOUT(dvp, &cn, CREATE);
997 if (cn.cn_flags & HASBUF) {
998 zfree(namei_zone, cn.cn_pnbuf);
999 cn.cn_flags &= ~HASBUF;
1005 * union_vn_create: creates and opens a new shadow file
1006 * on the upper union layer. this function is similar
1007 * in spirit to calling vn_open but it avoids calling namei().
1008 * the problem with calling namei is that a) it locks too many
1009 * things, and b) it doesn't start at the "right" directory,
1010 * whereas relookup is told where to start.
1012 * On entry, the vnode associated with un is locked. It remains locked
1015 * If no error occurs, *vpp contains a locked referenced vnode for your
1016 * use. If an error occurs *vpp iis undefined.
1019 union_vn_create(vpp, un, p)
1021 struct union_node *un;
1025 struct ucred *cred = p->p_ucred;
1027 struct vattr *vap = &vat;
1028 int fmode = FFLAGS(O_WRONLY|O_CREAT|O_TRUNC|O_EXCL);
1030 int cmode = UN_FILEMODE & ~p->p_fd->fd_cmask;
1031 struct componentname cn;
1036 * Build a new componentname structure (for the same
1037 * reasons outlines in union_mkshadow).
1038 * The difference here is that the file is owned by
1039 * the current user, rather than by the person who
1040 * did the mount, since the current user needs to be
1041 * able to write the file (that's why it is being
1042 * copied in the first place).
1044 cn.cn_namelen = strlen(un->un_path);
1045 cn.cn_pnbuf = zalloc(namei_zone);
1046 bcopy(un->un_path, cn.cn_pnbuf, cn.cn_namelen+1);
1047 cn.cn_nameiop = CREATE;
1048 cn.cn_flags = (LOCKPARENT|LOCKLEAF|HASBUF|SAVENAME|ISLASTCN);
1050 cn.cn_cred = p->p_ucred;
1051 cn.cn_nameptr = cn.cn_pnbuf;
1055 * Pass dvp unlocked and referenced on call to relookup().
1057 * If an error occurs, dvp will be returned unlocked and dereferenced.
1060 error = relookup(un->un_dirvp, &vp, &cn);
1065 * If no error occurs, dvp will be returned locked with the reference
1066 * left as before, and vpp will be returned referenced and locked.
1070 if (cn.cn_flags & HASBUF) {
1071 zfree(namei_zone, cn.cn_pnbuf);
1072 cn.cn_flags &= ~HASBUF;
1074 if (vp == un->un_dirvp)
1082 * Good - there was no race to create the file
1083 * so go ahead and create it. The permissions
1084 * on the file will be 0666 modified by the
1085 * current user's umask. Access to the file, while
1086 * it is unioned, will require access to the top *and*
1087 * bottom files. Access when not unioned will simply
1088 * require access to the top-level file.
1089 * TODO: confirm choice of access permissions.
1092 vap->va_type = VREG;
1093 vap->va_mode = cmode;
1094 VOP_LEASE(un->un_dirvp, p, cred, LEASE_WRITE);
1095 error = VOP_CREATE(un->un_dirvp, &vp, &cn, vap);
1096 if (cn.cn_flags & HASBUF) {
1097 zfree(namei_zone, cn.cn_pnbuf);
1098 cn.cn_flags &= ~HASBUF;
1104 error = VOP_OPEN(vp, fmode, cred, p);
1105 if (error == 0 && vn_canvmio(vp) == TRUE)
1106 error = vfs_object_create(vp, p, cred);
1117 union_vn_close(vp, fmode, cred, p)
1126 return (VOP_CLOSE(vp, fmode, cred, p));
1132 * union_removed_upper:
1134 * called with union_node unlocked. XXX
1138 union_removed_upper(un)
1139 struct union_node *un;
1141 struct proc *p = curproc; /* XXX */
1145 * Do not set the uppervp to NULLVP. If lowervp is NULLVP,
1146 * union node will have neither uppervp nor lowervp. We remove
1147 * the union node from cache, so that it will not be referrenced.
1149 union_newupper(un, NULLVP);
1150 if (un->un_dircache != 0) {
1151 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1153 free(un->un_dircache, M_TEMP);
1154 un->un_dircache = 0;
1157 if (un->un_flags & UN_CACHED) {
1158 un->un_flags &= ~UN_CACHED;
1159 LIST_REMOVE(un, un_cache);
1166 * determine whether a whiteout is needed
1167 * during a remove/rmdir operation.
1170 union_dowhiteout(un, cred, p)
1171 struct union_node *un;
1177 if (un->un_lowervp != NULLVP)
1180 if (VOP_GETATTR(un->un_uppervp, &va, cred, p) == 0 &&
1181 (va.va_flags & OPAQUE))
1188 union_dircache_r(vp, vppp, cntp)
1190 struct vnode ***vppp;
1193 struct union_node *un;
1195 if (vp->v_op != union_vnodeop_p) {
1200 panic("union: dircache table too small");
1209 if (un->un_uppervp != NULLVP)
1210 union_dircache_r(un->un_uppervp, vppp, cntp);
1211 if (un->un_lowervp != NULLVP)
1212 union_dircache_r(un->un_lowervp, vppp, cntp);
1216 union_dircache(vp, p)
1223 struct vnode **dircache;
1224 struct union_node *un;
1227 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1228 dircache = VTOUNION(vp)->un_dircache;
1232 if (dircache == NULL) {
1234 union_dircache_r(vp, 0, &cnt);
1236 dircache = malloc(cnt * sizeof(struct vnode *),
1239 union_dircache_r(vp, &vpp, &cnt);
1245 if (*vpp++ == VTOUNION(vp)->un_uppervp)
1247 } while (*vpp != NULLVP);
1253 /*vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY, p);*/
1254 UDEBUG(("ALLOCVP-3 %p ref %d\n", *vpp, (*vpp ? (*vpp)->v_usecount : -99)));
1256 error = union_allocvp(&nvp, vp->v_mount, NULLVP, NULLVP, NULL, *vpp, NULLVP, 0);
1257 UDEBUG(("ALLOCVP-3B %p ref %d\n", nvp, (*vpp ? (*vpp)->v_usecount : -99)));
1261 VTOUNION(vp)->un_dircache = 0;
1263 un->un_dircache = dircache;
1266 VOP_UNLOCK(vp, 0, p);
1271 * Guarentee coherency with the VM cache by invalidating any clean VM pages
1272 * associated with this write and updating any dirty VM pages. Since our
1273 * vnode is locked, other processes will not be able to read the pages in
1274 * again until after our write completes.
1276 * We also have to be coherent with reads, by flushing any pending dirty
1277 * pages prior to issuing the read.
1279 * XXX this is somewhat of a hack at the moment. To support this properly
1280 * we would have to be able to run VOP_READ and VOP_WRITE through the VM
1281 * cache. Then we wouldn't need to worry about coherency.
1285 union_vm_coherency(struct vnode *vp, struct uio *uio, int cleanfls)
1292 if ((object = vp->v_object) == NULL)
1295 pgoff = uio->uio_offset & PAGE_MASK;
1296 pstart = uio->uio_offset / PAGE_SIZE;
1297 pend = pstart + (uio->uio_resid + pgoff + PAGE_MASK) / PAGE_SIZE;
1299 vm_object_page_clean(object, pstart, pend, OBJPC_SYNC);
1301 vm_object_page_remove(object, pstart, pend, TRUE);
1305 * Module glue to remove #ifdef UNION from vfs_syscalls.c
1308 union_dircheck(struct proc *p, struct vnode **vp, struct file *fp)
1312 if ((*vp)->v_op == union_vnodeop_p) {
1315 lvp = union_dircache(*vp, p);
1316 if (lvp != NULLVP) {
1320 * If the directory is opaque,
1321 * then don't show lower entries
1323 error = VOP_GETATTR(*vp, &va, fp->f_cred, p);
1324 if (va.va_flags & OPAQUE) {
1330 if (lvp != NULLVP) {
1331 error = VOP_OPEN(lvp, FREAD, fp->f_cred, p);
1332 if (error == 0 && vn_canvmio(lvp) == TRUE)
1333 error = vfs_object_create(lvp, p, fp->f_cred);
1338 VOP_UNLOCK(lvp, 0, p);
1339 fp->f_data = (caddr_t) lvp;
1341 error = vn_close(*vp, FREAD, fp->f_cred, p);
1345 return -1; /* goto unionread */
1352 union_modevent(module_t mod, int type, void *data)
1356 union_dircheckp = union_dircheck;
1359 union_dircheckp = NULL;
1367 static moduledata_t union_mod = {
1373 DECLARE_MODULE(union_dircheck, union_mod, SI_SUB_VFS, SI_ORDER_ANY);