4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
25 * Copyright (c) 1983,1984,1985,1986,1987,1988,1989 AT&T.
26 * All rights reserved.
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/thread.h>
32 #include <sys/t_lock.h>
34 #include <sys/vnode.h>
36 #include <sys/errno.h>
41 #include <sys/debug.h>
42 #include <sys/vmsystm.h>
43 #include <sys/flock.h>
44 #include <sys/share.h>
45 #include <sys/cmn_err.h>
46 #include <sys/tiuser.h>
47 #include <sys/sysmacros.h>
48 #include <sys/callb.h>
50 #include <sys/kstat.h>
51 #include <sys/signal.h>
55 #include <netsmb/smb.h>
56 #include <netsmb/smb_conn.h>
57 #include <netsmb/smb_subr.h>
59 #include <smbfs/smbfs.h>
60 #include <smbfs/smbfs_node.h>
61 #include <smbfs/smbfs_subr.h>
68 #include <vm/seg_map.h>
69 #include <vm/seg_vn.h>
71 #define ATTRCACHE_VALID(vp) (gethrtime() < VTOSMB(vp)->r_attrtime)
73 static int smbfs_getattr_cache(vnode_t
*, smbfattr_t
*);
74 static void smbfattr_to_vattr(vnode_t
*, smbfattr_t
*, vattr_t
*);
75 static void smbfattr_to_xvattr(smbfattr_t
*, vattr_t
*);
76 static int smbfs_getattr_otw(vnode_t
*, struct smbfattr
*, cred_t
*);
80 * The following code provide zone support in order to perform an action
81 * for each smbfs mount in a zone. This is also where we would add
82 * per-zone globals and kernel threads for the smbfs module (since
83 * they must be terminated by the shutdown callback).
87 kmutex_t smg_lock
; /* lock protecting smg_list */
88 list_t smg_list
; /* list of SMBFS mounts in zone */
89 boolean_t smg_destructor_called
;
91 typedef struct smi_globals smi_globals_t
;
93 static zone_key_t smi_list_key
;
98 * Attributes are cached in the smbnode in struct vattr form.
99 * There is a time associated with the cached attributes (r_attrtime)
100 * which tells whether the attributes are valid. The time is initialized
101 * to the difference between current time and the modify time of the vnode
102 * when new attributes are cached. This allows the attributes for
103 * files that have changed recently to be timed out sooner than for files
104 * that have not changed for a long time. There are minimum and maximum
105 * timeout values that can be set per mount point.
109 * Helper for _validate_caches
112 smbfs_waitfor_purge_complete(vnode_t
*vp
)
118 if (np
->r_serial
!= NULL
&& np
->r_serial
!= curthread
) {
119 mutex_enter(&np
->r_statelock
);
120 sigintr(&smask
, VTOSMI(vp
)->smi_flags
& SMI_INT
);
121 while (np
->r_serial
!= NULL
) {
122 if (!cv_wait_sig(&np
->r_cv
, &np
->r_statelock
)) {
124 mutex_exit(&np
->r_statelock
);
129 mutex_exit(&np
->r_statelock
);
135 * Validate caches by checking cached attributes. If the cached
136 * attributes have timed out, then get new attributes from the server.
137 * As a side affect, this will do cache invalidation if the attributes
140 * If the attributes have not timed out and if there is a cache
141 * invalidation being done by some other thread, then wait until that
142 * thread has completed the cache invalidation.
145 smbfs_validate_caches(
152 if (ATTRCACHE_VALID(vp
)) {
153 error
= smbfs_waitfor_purge_complete(vp
);
159 return (smbfs_getattr_otw(vp
, &fa
, cr
));
163 * Purge all of the various data caches.
165 * Here NFS also had a flags arg to control what gets flushed.
166 * We only have the page cache, so no flags arg.
170 smbfs_purge_caches(struct vnode
*vp
, cred_t
*cr
)
174 * Here NFS has: Purge the DNLC for this vp,
175 * Clear any readdir state bits,
176 * the readlink response cache, ...
180 * Flush the page cache.
182 if (vn_has_cached_data(vp
)) {
183 (void) fop_putpage(vp
, 0, 0, B_INVAL
, cr
, NULL
);
187 * Here NFS has: Flush the readdir response cache.
188 * No readdir cache in smbfs.
194 * nfs_purge_rddir_cache()
195 * nfs3_cache_post_op_attr()
196 * nfs3_cache_post_op_vattr()
197 * nfs3_cache_wcc_data()
201 * Check the attribute cache to see if the new attributes match
202 * those cached. If they do, the various `data' caches are
203 * considered to be good. Otherwise, purge the cached data.
208 struct smbfattr
*fap
,
216 mutex_enter(&np
->r_statelock
);
219 * Compare with NFS macro: CACHE_VALID
220 * If the mtime or size has changed,
223 if (np
->r_attr
.fa_mtime
.tv_sec
!= fap
->fa_mtime
.tv_sec
||
224 np
->r_attr
.fa_mtime
.tv_nsec
!= fap
->fa_mtime
.tv_nsec
)
226 if (np
->r_attr
.fa_size
!= fap
->fa_size
)
229 if (np
->r_attr
.fa_ctime
.tv_sec
!= fap
->fa_ctime
.tv_sec
||
230 np
->r_attr
.fa_ctime
.tv_nsec
!= fap
->fa_ctime
.tv_nsec
)
234 np
->r_sectime
= gethrtime();
237 mutex_exit(&np
->r_statelock
);
240 smbfs_purge_caches(vp
, cr
);
244 * Set attributes cache for given vnode using SMB fattr
245 * and update the attribute cache timeout.
247 * Based on NFS: nfs_attrcache, nfs_attrcache_va
250 smbfs_attrcache_fa(vnode_t
*vp
, struct smbfattr
*fap
)
256 vtype_t vtype
, oldvt
;
263 * We allow v_type to change, so set that here
264 * (and the mode, which depends on the type).
266 if (fap
->fa_attr
& SMB_FA_DIR
) {
268 mode
= smi
->smi_dmode
;
271 mode
= smi
->smi_fmode
;
274 mutex_enter(&np
->r_statelock
);
278 * Delta is the number of nanoseconds that we will
279 * cache the attributes of the file. It is based on
280 * the number of nanoseconds since the last time that
281 * we detected a change. The assumption is that files
282 * that changed recently are likely to change again.
283 * There is a minimum and a maximum for regular files
284 * and for directories which is enforced though.
286 * Using the time since last change was detected
287 * eliminates direct comparison or calculation
288 * using mixed client and server times. SMBFS
289 * does not make any assumptions regarding the
290 * client and server clocks being synchronized.
292 if (fap
->fa_mtime
.tv_sec
!= np
->r_attr
.fa_mtime
.tv_sec
||
293 fap
->fa_mtime
.tv_nsec
!= np
->r_attr
.fa_mtime
.tv_nsec
||
294 fap
->fa_size
!= np
->r_attr
.fa_size
)
297 if ((smi
->smi_flags
& SMI_NOAC
) || (vp
->v_flag
& VNOCACHE
))
300 delta
= now
- np
->r_mtime
;
302 if (delta
< smi
->smi_acdirmin
)
303 delta
= smi
->smi_acdirmin
;
304 else if (delta
> smi
->smi_acdirmax
)
305 delta
= smi
->smi_acdirmax
;
307 if (delta
< smi
->smi_acregmin
)
308 delta
= smi
->smi_acregmin
;
309 else if (delta
> smi
->smi_acregmax
)
310 delta
= smi
->smi_acregmax
;
314 np
->r_attrtime
= now
+ delta
;
321 * Shall we update r_size? (local notion of size)
323 * The real criteria for updating r_size should be:
324 * if the file has grown on the server, or if
325 * the client has not modified the file.
327 * Also deal with the fact that SMB presents
328 * directories as having size=0. Doing that
329 * here and leaving fa_size as returned OtW
330 * avoids fixing the size lots of places.
332 newsize
= fap
->fa_size
;
333 if (vtype
== VDIR
&& newsize
< DEV_BSIZE
)
336 if (np
->r_size
!= newsize
&&
337 (!vn_has_cached_data(vp
) ||
338 (!(np
->r_flags
& RDIRTY
) && np
->r_count
== 0))) {
339 /* OK to set the size. */
340 np
->r_size
= newsize
;
345 * nfs_setswaplike(vp, va);
346 * np->r_flags &= ~RWRITEATTR;
350 np
->n_flag
&= ~NATTRCHANGED
;
351 mutex_exit(&np
->r_statelock
);
353 if (oldvt
!= vtype
) {
354 SMBVDEBUG("vtype change %d to %d\n", oldvt
, vtype
);
359 * Fill in attribute from the cache.
361 * If valid, copy to *fap and return zero,
362 * otherwise return an error.
364 * From NFS: nfs_getattr_cache()
367 smbfs_getattr_cache(vnode_t
*vp
, struct smbfattr
*fap
)
374 mutex_enter(&np
->r_statelock
);
375 if (gethrtime() >= np
->r_attrtime
) {
383 mutex_exit(&np
->r_statelock
);
389 * Get attributes over-the-wire and update attributes cache
390 * if no error occurred in the over-the-wire operation.
391 * Return 0 if successful, otherwise error.
392 * From NFS: nfs_getattr_otw
395 smbfs_getattr_otw(vnode_t
*vp
, struct smbfattr
*fap
, cred_t
*cr
)
398 struct smb_cred scred
;
404 * Here NFS uses the ACL RPC (if smi_flags & SMI_ACL)
405 * With SMB, getting the ACL is a significantly more
406 * expensive operation, so we do that only when asked
407 * for the uid/gid. See smbfsgetattr().
410 /* Shared lock for (possible) n_fid use. */
411 if (smbfs_rw_enter_sig(&np
->r_lkserlock
, RW_READER
, SMBINTR(vp
)))
413 smb_credinit(&scred
, cr
);
415 bzero(fap
, sizeof (*fap
));
416 error
= smbfs_smb_getfattr(np
, fap
, &scred
);
418 smb_credrele(&scred
);
419 smbfs_rw_exit(&np
->r_lkserlock
);
422 /* Here NFS has: PURGE_STALE_FH(error, vp, cr) */
423 smbfs_attrcache_remove(np
);
424 if (error
== ENOENT
|| error
== ENOTDIR
) {
426 * Getattr failed because the object was
427 * removed or renamed by another client.
428 * Remove any cached attributes under it.
430 smbfs_attrcache_prune(np
);
436 * Here NFS has: nfs_cache_fattr(vap, fa, vap, t, cr);
437 * which did: fattr_to_vattr, nfs_attr_cache.
438 * We cache the fattr form, so just do the
439 * cache check and store the attributes.
441 smbfs_cache_check(vp
, fap
, cr
);
442 smbfs_attrcache_fa(vp
, fap
);
448 * Return either cached or remote attributes. If we get remote attrs,
449 * use them to check and invalidate caches, then cache the new attributes.
451 * From NFS: nfsgetattr()
454 smbfsgetattr(vnode_t
*vp
, struct vattr
*vap
, cred_t
*cr
)
463 ASSERT(curproc
->p_zone
== smi
->smi_zone_ref
.zref_zone
);
466 * If asked for UID or GID, update n_uid, n_gid.
469 if (vap
->va_mask
& (VATTR_UID
| VATTR_GID
)) {
470 if (smi
->smi_flags
& SMI_ACL
)
471 (void) smbfs_acl_getids(vp
, cr
);
472 /* else leave as set in make_smbnode */
474 mask
&= ~(VATTR_UID
| VATTR_GID
);
478 * If we've got cached attributes, just use them;
479 * otherwise go to the server to get attributes,
480 * which will update the cache in the process.
482 error
= smbfs_getattr_cache(vp
, &fa
);
484 error
= smbfs_getattr_otw(vp
, &fa
, cr
);
487 vap
->va_mask
|= mask
;
490 * Re. client's view of the file size, see:
491 * smbfs_attrcache_fa, smbfs_getattr_otw
493 smbfattr_to_vattr(vp
, &fa
, vap
);
494 if (vap
->va_mask
& VATTR_XVATTR
)
495 smbfattr_to_xvattr(&fa
, vap
);
502 * Convert SMB over the wire attributes to vnode form.
503 * Returns 0 for success, error if failed (overflow, etc).
504 * From NFS: nattr_to_vattr()
507 smbfattr_to_vattr(vnode_t
*vp
, struct smbfattr
*fa
, struct vattr
*vap
)
509 struct smbnode
*np
= VTOSMB(vp
);
512 * Take type, mode, uid, gid from the smbfs node,
513 * which has have been updated by _getattr_otw.
515 vap
->va_type
= vp
->v_type
;
516 vap
->va_mode
= np
->n_mode
;
518 vap
->va_uid
= np
->n_uid
;
519 vap
->va_gid
= np
->n_gid
;
521 vap
->va_fsid
= vp
->v_vfsp
->vfs_dev
;
522 vap
->va_nodeid
= np
->n_ino
;
526 * Difference from NFS here: We cache attributes as
527 * reported by the server, so r_attr.fa_size is the
528 * server's idea of the file size. This is called
529 * for getattr, so we want to return the client's
530 * idea of the file size. NFS deals with that in
531 * nfsgetattr(), the equivalent of our caller.
533 vap
->va_size
= np
->r_size
;
536 * Times. Note, already converted from NT to
537 * Unix form (in the unmarshalling code).
539 vap
->va_atime
= fa
->fa_atime
;
540 vap
->va_mtime
= fa
->fa_mtime
;
541 vap
->va_ctime
= fa
->fa_ctime
;
544 * rdev, blksize, seq are made up.
545 * va_nblocks is 512 byte blocks.
547 vap
->va_rdev
= vp
->v_rdev
;
548 vap
->va_blksize
= MAXBSIZE
;
549 vap
->va_nblocks
= (fsblkcnt64_t
)btod(np
->r_attr
.fa_allocsz
);
554 * smbfattr_to_xvattr: like smbfattr_to_vattr but for
555 * Extensible system attributes (PSARC 2007/315)
558 smbfattr_to_xvattr(struct smbfattr
*fa
, struct vattr
*vap
)
560 xvattr_t
*xvap
= (xvattr_t
*)vap
; /* *vap may be xvattr_t */
561 xoptattr_t
*xoap
= NULL
;
563 if ((xoap
= xva_getxoptattr(xvap
)) == NULL
)
566 if (XVA_ISSET_REQ(xvap
, XAT_CREATETIME
)) {
567 xoap
->xoa_createtime
= fa
->fa_createtime
;
568 XVA_SET_RTN(xvap
, XAT_CREATETIME
);
571 if (XVA_ISSET_REQ(xvap
, XAT_ARCHIVE
)) {
573 ((fa
->fa_attr
& SMB_FA_ARCHIVE
) != 0);
574 XVA_SET_RTN(xvap
, XAT_ARCHIVE
);
577 if (XVA_ISSET_REQ(xvap
, XAT_SYSTEM
)) {
579 ((fa
->fa_attr
& SMB_FA_SYSTEM
) != 0);
580 XVA_SET_RTN(xvap
, XAT_SYSTEM
);
583 if (XVA_ISSET_REQ(xvap
, XAT_READONLY
)) {
585 ((fa
->fa_attr
& SMB_FA_RDONLY
) != 0);
586 XVA_SET_RTN(xvap
, XAT_READONLY
);
589 if (XVA_ISSET_REQ(xvap
, XAT_HIDDEN
)) {
591 ((fa
->fa_attr
& SMB_FA_HIDDEN
) != 0);
592 XVA_SET_RTN(xvap
, XAT_HIDDEN
);
598 * nfs_async_... stuff
599 * which we're not using (no async I/O), and:
602 * nfs_invalidate_pages()
603 * which we have in smbfs_vnops.c, and
610 * Helper function for smbfs_sync
612 * Walk the per-zone list of smbfs mounts, calling smbfs_rflush
613 * on each one. This is a little tricky because we need to exit
614 * the list mutex before each _rflush call and then try to resume
615 * where we were in the list after re-entering the mutex.
618 smbfs_flushall(cred_t
*cr
)
621 smbmntinfo_t
*tmp_smi
, *cur_smi
, *next_smi
;
623 smg
= zone_getspecific(smi_list_key
, crgetzone(cr
));
626 mutex_enter(&smg
->smg_lock
);
627 cur_smi
= list_head(&smg
->smg_list
);
628 if (cur_smi
== NULL
) {
629 mutex_exit(&smg
->smg_lock
);
632 VFS_HOLD(cur_smi
->smi_vfsp
);
633 mutex_exit(&smg
->smg_lock
);
636 smbfs_rflush(cur_smi
->smi_vfsp
, cr
);
638 mutex_enter(&smg
->smg_lock
);
640 * Resume after cur_smi if that's still on the list,
641 * otherwise restart at the head.
643 for (tmp_smi
= list_head(&smg
->smg_list
);
645 tmp_smi
= list_next(&smg
->smg_list
, tmp_smi
))
646 if (tmp_smi
== cur_smi
)
649 next_smi
= list_next(&smg
->smg_list
, tmp_smi
);
651 next_smi
= list_head(&smg
->smg_list
);
653 if (next_smi
!= NULL
)
654 VFS_HOLD(next_smi
->smi_vfsp
);
655 VFS_RELE(cur_smi
->smi_vfsp
);
657 mutex_exit(&smg
->smg_lock
);
659 if (next_smi
!= NULL
) {
666 * SMB Client initialization and cleanup.
667 * Much of it is per-zone now.
673 smbfs_zone_init(zoneid_t zoneid
)
677 smg
= kmem_alloc(sizeof (*smg
), KM_SLEEP
);
678 mutex_init(&smg
->smg_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
679 list_create(&smg
->smg_list
, sizeof (smbmntinfo_t
),
680 offsetof(smbmntinfo_t
, smi_zone_node
));
681 smg
->smg_destructor_called
= B_FALSE
;
686 * Callback routine to tell all SMBFS mounts in the zone to stop creating new
687 * threads. Existing threads should exit.
691 smbfs_zone_shutdown(zoneid_t zoneid
, void *data
)
693 smi_globals_t
*smg
= data
;
698 mutex_enter(&smg
->smg_lock
);
699 for (smi
= list_head(&smg
->smg_list
); smi
!= NULL
;
700 smi
= list_next(&smg
->smg_list
, smi
)) {
703 * If we've done the shutdown work for this FS, skip.
704 * Once we go off the end of the list, we're done.
706 if (smi
->smi_flags
& SMI_DEAD
)
710 * We will do work, so not done. Get a hold on the FS.
712 VFS_HOLD(smi
->smi_vfsp
);
714 mutex_enter(&smi
->smi_lock
);
715 smi
->smi_flags
|= SMI_DEAD
;
716 mutex_exit(&smi
->smi_lock
);
719 * Drop lock and release FS, which may change list, then repeat.
720 * We're done when every mi has been done or the list is empty.
722 mutex_exit(&smg
->smg_lock
);
723 VFS_RELE(smi
->smi_vfsp
);
726 mutex_exit(&smg
->smg_lock
);
730 smbfs_zone_free_globals(smi_globals_t
*smg
)
732 list_destroy(&smg
->smg_list
); /* makes sure the list is empty */
733 mutex_destroy(&smg
->smg_lock
);
734 kmem_free(smg
, sizeof (*smg
));
740 smbfs_zone_destroy(zoneid_t zoneid
, void *data
)
742 smi_globals_t
*smg
= data
;
745 mutex_enter(&smg
->smg_lock
);
746 if (list_head(&smg
->smg_list
) != NULL
) {
747 /* Still waiting for VFS_FREEVFS() */
748 smg
->smg_destructor_called
= B_TRUE
;
749 mutex_exit(&smg
->smg_lock
);
752 smbfs_zone_free_globals(smg
);
756 * Add an SMBFS mount to the per-zone list of SMBFS mounts.
759 smbfs_zonelist_add(smbmntinfo_t
*smi
)
763 smg
= zone_getspecific(smi_list_key
, smi
->smi_zone_ref
.zref_zone
);
764 mutex_enter(&smg
->smg_lock
);
765 list_insert_head(&smg
->smg_list
, smi
);
766 mutex_exit(&smg
->smg_lock
);
770 * Remove an SMBFS mount from the per-zone list of SMBFS mounts.
773 smbfs_zonelist_remove(smbmntinfo_t
*smi
)
777 smg
= zone_getspecific(smi_list_key
, smi
->smi_zone_ref
.zref_zone
);
778 mutex_enter(&smg
->smg_lock
);
779 list_remove(&smg
->smg_list
, smi
);
781 * We can be called asynchronously by VFS_FREEVFS() after the zone
782 * shutdown/destroy callbacks have executed; if so, clean up the zone's
785 if (list_head(&smg
->smg_list
) == NULL
&&
786 smg
->smg_destructor_called
== B_TRUE
) {
787 smbfs_zone_free_globals(smg
);
790 mutex_exit(&smg
->smg_lock
);
794 #ifdef NEED_SMBFS_CALLBACKS
796 * Call-back hooks for netsmb, in case we want them.
797 * Apple's VFS wants them. We may not need them.
800 static void smbfs_dead(smb_share_t
*ssp
)
803 * Walk the mount list, finding all mounts
804 * using this share...
809 static void smbfs_cb_nop(smb_share_t
*ss
)
814 smb_fscb_t smbfs_cb
= {
815 .fscb_disconn
= smbfs_dead
,
816 .fscb_connect
= smbfs_cb_nop
,
817 .fscb_down
= smbfs_cb_nop
,
818 .fscb_up
= smbfs_cb_nop
};
820 #endif /* NEED_SMBFS_CALLBACKS */
823 * SMBFS Client initialization routine. This routine should only be called
824 * once. It performs the following tasks:
825 * - Initalize all global locks
826 * - Call sub-initialization routines (localize access to variables)
832 zone_key_create(&smi_list_key
, smbfs_zone_init
, smbfs_zone_shutdown
,
834 #ifdef NEED_SMBFS_CALLBACKS
835 (void) smb_fscb_set(&smbfs_cb
);
836 #endif /* NEED_SMBFS_CALLBACKS */
841 * This routine is called when the modunload is called. This will cleanup
842 * the previously allocated/initialized nodes.
847 #ifdef NEED_SMBFS_CALLBACKS
848 (void) smb_fscb_set(NULL
);
849 #endif /* NEED_SMBFS_CALLBACKS */
850 (void) zone_key_delete(smi_list_key
);