2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.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 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.5 2007/11/20 22:55:40 dillon Exp $
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/fcntl.h>
41 #include <sys/namecache.h>
42 #include <sys/vnode.h>
43 #include <sys/lockf.h>
44 #include <sys/event.h>
51 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
52 static int hammer_vop_fsync(struct vop_fsync_args *);
53 static int hammer_vop_read(struct vop_read_args *);
54 static int hammer_vop_write(struct vop_write_args *);
55 static int hammer_vop_access(struct vop_access_args *);
56 static int hammer_vop_advlock(struct vop_advlock_args *);
57 static int hammer_vop_close(struct vop_close_args *);
58 static int hammer_vop_ncreate(struct vop_ncreate_args *);
59 static int hammer_vop_getattr(struct vop_getattr_args *);
60 static int hammer_vop_nresolve(struct vop_nresolve_args *);
61 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
62 static int hammer_vop_nlink(struct vop_nlink_args *);
63 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
64 static int hammer_vop_nmknod(struct vop_nmknod_args *);
65 static int hammer_vop_open(struct vop_open_args *);
66 static int hammer_vop_pathconf(struct vop_pathconf_args *);
67 static int hammer_vop_print(struct vop_print_args *);
68 static int hammer_vop_readdir(struct vop_readdir_args *);
69 static int hammer_vop_readlink(struct vop_readlink_args *);
70 static int hammer_vop_nremove(struct vop_nremove_args *);
71 static int hammer_vop_nrename(struct vop_nrename_args *);
72 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
73 static int hammer_vop_setattr(struct vop_setattr_args *);
74 static int hammer_vop_strategy(struct vop_strategy_args *);
75 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
76 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
78 struct vop_ops hammer_vnode_vops = {
79 .vop_default = vop_defaultop,
80 .vop_fsync = hammer_vop_fsync,
81 .vop_read = hammer_vop_read,
82 .vop_write = hammer_vop_write,
83 .vop_access = hammer_vop_access,
84 .vop_advlock = hammer_vop_advlock,
85 .vop_close = hammer_vop_close,
86 .vop_ncreate = hammer_vop_ncreate,
87 .vop_getattr = hammer_vop_getattr,
88 .vop_inactive = hammer_vop_inactive,
89 .vop_reclaim = hammer_vop_reclaim,
90 .vop_nresolve = hammer_vop_nresolve,
91 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
92 .vop_nlink = hammer_vop_nlink,
93 .vop_nmkdir = hammer_vop_nmkdir,
94 .vop_nmknod = hammer_vop_nmknod,
95 .vop_open = hammer_vop_open,
96 .vop_pathconf = hammer_vop_pathconf,
97 .vop_print = hammer_vop_print,
98 .vop_readdir = hammer_vop_readdir,
99 .vop_readlink = hammer_vop_readlink,
100 .vop_nremove = hammer_vop_nremove,
101 .vop_nrename = hammer_vop_nrename,
102 .vop_nrmdir = hammer_vop_nrmdir,
103 .vop_setattr = hammer_vop_setattr,
104 .vop_strategy = hammer_vop_strategy,
105 .vop_nsymlink = hammer_vop_nsymlink,
106 .vop_nwhiteout = hammer_vop_nwhiteout
109 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
110 struct ucred *cred, int flags);
111 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
112 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
117 hammer_vop_vnoperate(struct vop_generic_args *)
119 return (VOCALL(&hammer_vnode_vops, ap));
124 * hammer_vop_fsync { vp, waitfor }
128 hammer_vop_fsync(struct vop_fsync_args *ap)
134 * hammer_vop_read { vp, uio, ioflag, cred }
138 hammer_vop_read(struct vop_read_args *ap)
140 struct hammer_transaction trans;
141 struct hammer_inode *ip;
149 if (ap->a_vp->v_type != VREG)
153 seqcount = ap->a_ioflag >> 16;
155 hammer_start_transaction(&trans, ip->hmp);
158 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
161 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
162 offset = uio->uio_offset & HAMMER_BUFMASK;
163 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
164 uio->uio_offset - offset, HAMMER_BUFSIZE,
165 MAXBSIZE, seqcount, &bp);
170 bp->b_flags |= B_CLUSTEROK;
171 n = HAMMER_BUFSIZE - offset;
172 if (n > uio->uio_resid)
174 if (n > ip->ino_rec.ino_size - uio->uio_offset)
175 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
176 error = uiomove((char *)bp->b_data + offset, n, uio);
181 ip->ino_rec.ino_atime = trans.tid;
182 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
185 hammer_commit_transaction(&trans);
190 * hammer_vop_write { vp, uio, ioflag, cred }
194 hammer_vop_write(struct vop_write_args *ap)
196 struct hammer_transaction trans;
197 struct hammer_inode *ip;
204 if (ap->a_vp->v_type != VREG)
210 * Create a transaction to cover the operations we perform.
212 hammer_start_transaction(&trans, ip->hmp);
218 if (ap->a_ioflag & IO_APPEND)
219 uio->uio_offset = ip->ino_rec.ino_size;
222 * Check for illegal write offsets. Valid range is 0...2^63-1
224 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0)
228 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
230 while (uio->uio_resid > 0) {
231 offset = uio->uio_offset & HAMMER_BUFMASK;
232 if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
233 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
235 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
236 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
240 error = bread(ap->a_vp, uio->uio_offset - offset,
241 HAMMER_BUFSIZE, &bp);
247 n = HAMMER_BUFSIZE - offset;
248 if (n > uio->uio_resid)
250 error = uiomove((char *)bp->b_data + offset, n, uio);
255 bp->b_flags |= B_CLUSTEROK;
256 if (ip->ino_rec.ino_size < uio->uio_offset) {
257 ip->ino_rec.ino_size = uio->uio_offset;
258 ip->ino_rec.ino_mtime = trans.tid;
259 hammer_modify_inode(&trans, ip,
260 HAMMER_INODE_RDIRTY | HAMMER_INODE_ITIMES);
262 if (ap->a_ioflag & IO_SYNC) {
264 } else if (ap->a_ioflag & IO_DIRECT) {
271 hammer_abort_transaction(&trans);
273 hammer_commit_transaction(&trans);
278 * hammer_vop_access { vp, mode, cred }
282 hammer_vop_access(struct vop_access_args *ap)
284 struct hammer_inode *ip = VTOI(ap->a_vp);
289 uid = hammer_to_unix_xid(&ip->ino_data.uid);
290 gid = hammer_to_unix_xid(&ip->ino_data.gid);
292 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
293 ip->ino_data.uflags);
298 * hammer_vop_advlock { vp, id, op, fl, flags }
302 hammer_vop_advlock(struct vop_advlock_args *ap)
304 struct hammer_inode *ip = VTOI(ap->a_vp);
306 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
310 * hammer_vop_close { vp, fflag }
314 hammer_vop_close(struct vop_close_args *ap)
316 return (vop_stdclose(ap));
320 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
322 * The operating system has already ensured that the directory entry
323 * does not exist and done all appropriate namespace locking.
327 hammer_vop_ncreate(struct vop_ncreate_args *ap)
329 struct hammer_transaction trans;
330 struct hammer_inode *dip;
331 struct hammer_inode *nip;
332 struct nchandle *nch;
336 dip = VTOI(ap->a_dvp);
339 * Create a transaction to cover the operations we perform.
341 hammer_start_transaction(&trans, dip->hmp);
344 * Create a new filesystem object of the requested type. The
345 * returned inode will be referenceds but not locked.
348 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
350 hammer_abort_transaction(&trans);
356 * Add the new filesystem object to the directory. This will also
357 * bump the inode's link count.
359 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
365 hammer_rel_inode(nip, 0);
366 hammer_abort_transaction(&trans);
369 hammer_commit_transaction(&trans);
370 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
371 hammer_rel_inode(nip, 0);
373 cache_setunresolved(ap->a_nch);
374 cache_setvp(ap->a_nch, *ap->a_vpp);
381 * hammer_vop_getattr { vp, vap }
385 hammer_vop_getattr(struct vop_getattr_args *ap)
387 struct hammer_inode *ip = VTOI(ap->a_vp);
388 struct vattr *vap = ap->a_vap;
391 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
392 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
397 hammer_itimes(ap->a_vp);
400 vap->va_fsid = ip->hmp->fsid_udev;
401 vap->va_fileid = ip->ino_rec.base.base.obj_id;
402 vap->va_mode = ip->ino_data.mode;
403 vap->va_nlink = ip->ino_rec.ino_nlinks;
404 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
405 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
408 vap->va_size = ip->ino_rec.ino_size;
409 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
410 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
411 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
412 vap->va_flags = ip->ino_data.uflags;
413 vap->va_gen = 1; /* hammer inums are unique for all time */
414 vap->va_blocksize = 32768; /* XXX - extract from root volume */
415 vap->va_bytes = ip->ino_rec.ino_size;
416 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
417 vap->va_filerev = 0; /* XXX */
418 /* mtime uniquely identifies any adjustments made to the file */
419 vap->va_fsmid = ip->ino_rec.ino_mtime;
420 vap->va_uid_uuid = ip->ino_data.uid;
421 vap->va_gid_uuid = ip->ino_data.gid;
422 vap->va_fsid_uuid = ip->hmp->fsid;
423 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
429 * hammer_vop_nresolve { nch, dvp, cred }
431 * Locate the requested directory entry.
435 hammer_vop_nresolve(struct vop_nresolve_args *ap)
437 struct namecache *ncp;
438 struct hammer_inode *dip;
439 struct hammer_cursor cursor;
440 union hammer_record_ondisk *rec;
446 * Calculate the namekey and setup the key range for the scan. This
447 * works kinda like a chained hash table where the lower 32 bits
448 * of the namekey synthesize the chain.
450 * The key range is inclusive of both key_beg and key_end.
452 dip = VTOI(ap->a_dvp);
453 ncp = ap->a_nch->ncp;
454 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
455 kprintf("hammer_vop_nresolve %s dip %p\n", ncp->nc_name, dip);
457 hammer_init_cursor_ip(&cursor, dip);
458 cursor.key_beg.obj_id = dip->obj_id;
459 cursor.key_beg.key = namekey;
460 cursor.key_beg.create_tid = dip->obj_asof;
461 cursor.key_beg.delete_tid = 0;
462 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
463 cursor.key_beg.obj_type = 0;
465 cursor.key_end = cursor.key_beg;
466 cursor.key_end.key |= 0xFFFFFFFFULL;
469 * Scan all matching records (the chain), locate the one matching
470 * the requested path component.
472 * The hammer_ip_*() functions merge in-memory records with on-disk
473 * records for the purposes of the search.
475 error = hammer_ip_first(&cursor, dip);
477 error = hammer_ip_resolve_data(&cursor);
481 if (ncp->nc_nlen == rec->entry.base.data_len &&
482 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
485 error = hammer_ip_next(&cursor);
488 error = hammer_vfs_vget(dip->hmp->mp, rec->entry.obj_id, &vp);
491 cache_setvp(ap->a_nch, vp);
494 } else if (error == ENOENT) {
495 cache_setvp(ap->a_nch, NULL);
497 hammer_done_cursor(&cursor);
498 kprintf("hammer_vop_nresolve error %d\n", error);
503 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
505 * Locate the parent directory of a directory vnode.
507 * dvp is referenced but not locked. *vpp must be returned referenced and
508 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
509 * at the root, instead it could indicate that the directory we were in was
514 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
516 struct hammer_inode *dip;
517 u_int64_t parent_obj_id;
519 dip = VTOI(ap->a_dvp);
520 if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
524 return(hammer_vfs_vget(dip->hmp->mp, parent_obj_id, ap->a_vpp));
528 * hammer_vop_nlink { nch, dvp, vp, cred }
532 hammer_vop_nlink(struct vop_nlink_args *ap)
534 struct hammer_transaction trans;
535 struct hammer_inode *dip;
536 struct hammer_inode *ip;
537 struct nchandle *nch;
541 dip = VTOI(ap->a_dvp);
545 * Create a transaction to cover the operations we perform.
547 hammer_start_transaction(&trans, dip->hmp);
550 * Add the filesystem object to the directory. Note that neither
551 * dip nor ip are referenced or locked, but their vnodes are
552 * referenced. This function will bump the inode's link count.
554 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
560 hammer_abort_transaction(&trans);
562 cache_setunresolved(nch);
563 cache_setvp(nch, ap->a_vp);
564 hammer_commit_transaction(&trans);
570 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
572 * The operating system has already ensured that the directory entry
573 * does not exist and done all appropriate namespace locking.
577 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
579 struct hammer_transaction trans;
580 struct hammer_inode *dip;
581 struct hammer_inode *nip;
582 struct nchandle *nch;
586 dip = VTOI(ap->a_dvp);
589 * Create a transaction to cover the operations we perform.
591 hammer_start_transaction(&trans, dip->hmp);
594 * Create a new filesystem object of the requested type. The
595 * returned inode will be referenced but not locked.
597 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
599 hammer_abort_transaction(&trans);
605 * Add the new filesystem object to the directory. This will also
606 * bump the inode's link count.
608 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
614 hammer_rel_inode(nip, 0);
615 hammer_abort_transaction(&trans);
618 hammer_commit_transaction(&trans);
619 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
620 hammer_rel_inode(nip, 0);
622 cache_setunresolved(ap->a_nch);
623 cache_setvp(ap->a_nch, *ap->a_vpp);
630 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
632 * The operating system has already ensured that the directory entry
633 * does not exist and done all appropriate namespace locking.
637 hammer_vop_nmknod(struct vop_nmknod_args *ap)
639 struct hammer_transaction trans;
640 struct hammer_inode *dip;
641 struct hammer_inode *nip;
642 struct nchandle *nch;
646 dip = VTOI(ap->a_dvp);
649 * Create a transaction to cover the operations we perform.
651 hammer_start_transaction(&trans, dip->hmp);
654 * Create a new filesystem object of the requested type. The
655 * returned inode will be referenced but not locked.
657 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
659 hammer_abort_transaction(&trans);
665 * Add the new filesystem object to the directory. This will also
666 * bump the inode's link count.
668 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
674 hammer_rel_inode(nip, 0);
675 hammer_abort_transaction(&trans);
678 hammer_commit_transaction(&trans);
679 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
680 hammer_rel_inode(nip, 0);
682 cache_setunresolved(ap->a_nch);
683 cache_setvp(ap->a_nch, *ap->a_vpp);
690 * hammer_vop_open { vp, mode, cred, fp }
694 hammer_vop_open(struct vop_open_args *ap)
696 kprintf("hammer_vop_open\n");
697 return(vop_stdopen(ap));
701 * hammer_vop_pathconf { vp, name, retval }
705 hammer_vop_pathconf(struct vop_pathconf_args *ap)
711 * hammer_vop_print { vp }
715 hammer_vop_print(struct vop_print_args *ap)
721 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
725 hammer_vop_readdir(struct vop_readdir_args *ap)
727 struct hammer_cursor cursor;
728 struct hammer_inode *ip;
730 hammer_record_ondisk_t rec;
731 hammer_base_elm_t base;
741 hammer_init_cursor_ip(&cursor, ip);
744 * Key range (begin and end inclusive) to scan. Directory keys
745 * directly translate to a 64 bit 'seek' position.
747 cursor.key_beg.obj_id = ip->obj_id;
748 cursor.key_beg.create_tid = ip->obj_asof;
749 cursor.key_beg.delete_tid = 0;
750 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
751 cursor.key_beg.obj_type = 0;
752 cursor.key_beg.key = uio->uio_offset;
754 cursor.key_end = cursor.key_beg;
755 cursor.key_end.key = HAMMER_MAX_KEY;
757 if (ap->a_ncookies) {
758 ncookies = uio->uio_resid / 16 + 1;
761 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
769 saveoff = cursor.key_beg.key;
770 error = hammer_ip_first(&cursor, ip);
773 error = hammer_ip_resolve_data(&cursor);
777 base = &rec->base.base;
780 r = vop_write_dirent(
781 &error, uio, rec->entry.obj_id,
782 rec->entry.base.data_len,
783 hammer_get_dtype(rec->entry.base.base.obj_type),
784 (void *)cursor.data);
789 cookies[cookie_index] = base->key;
791 if (cookie_index == ncookies)
793 error = hammer_ip_next(&cursor);
795 hammer_done_cursor(&cursor);
798 *ap->a_eofflag = (error == ENOENT);
801 uio->uio_offset = saveoff;
802 if (error && cookie_index == 0) {
804 kfree(cookies, M_TEMP);
806 *ap->a_cookies = NULL;
811 *ap->a_ncookies = cookie_index;
812 *ap->a_cookies = cookies;
819 * hammer_vop_readlink { vp, uio, cred }
823 hammer_vop_readlink(struct vop_readlink_args *ap)
829 * hammer_vop_nremove { nch, dvp, cred }
833 hammer_vop_nremove(struct vop_nremove_args *ap)
835 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
839 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
843 hammer_vop_nrename(struct vop_nrename_args *ap)
845 struct hammer_transaction trans;
846 struct namecache *fncp;
847 struct namecache *tncp;
848 struct hammer_inode *fdip;
849 struct hammer_inode *tdip;
850 struct hammer_inode *ip;
851 struct hammer_cursor cursor;
852 union hammer_record_ondisk *rec;
856 fdip = VTOI(ap->a_fdvp);
857 tdip = VTOI(ap->a_tdvp);
858 fncp = ap->a_fnch->ncp;
859 tncp = ap->a_tnch->ncp;
860 hammer_start_transaction(&trans, fdip->hmp);
863 * Extract the hammer_inode from fncp and add link to the target
866 ip = VTOI(fncp->nc_vp);
867 KKASSERT(ip != NULL);
869 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
872 * Locate the record in the originating directory and remove it.
874 * Calculate the namekey and setup the key range for the scan. This
875 * works kinda like a chained hash table where the lower 32 bits
876 * of the namekey synthesize the chain.
878 * The key range is inclusive of both key_beg and key_end.
880 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
882 hammer_init_cursor_ip(&cursor, fdip);
883 cursor.key_beg.obj_id = fdip->obj_id;
884 cursor.key_beg.key = namekey;
885 cursor.key_beg.create_tid = fdip->obj_asof;
886 cursor.key_beg.delete_tid = 0;
887 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
888 cursor.key_beg.obj_type = 0;
890 cursor.key_end = cursor.key_beg;
891 cursor.key_end.key |= 0xFFFFFFFFULL;
894 * Scan all matching records (the chain), locate the one matching
895 * the requested path component.
897 * The hammer_ip_*() functions merge in-memory records with on-disk
898 * records for the purposes of the search.
900 error = hammer_ip_first(&cursor, fdip);
902 if (hammer_ip_resolve_data(&cursor) != 0)
905 if (fncp->nc_nlen == rec->entry.base.data_len &&
906 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
909 error = hammer_ip_next(&cursor);
913 * If all is ok we have to get the inode so we can adjust nlinks.
917 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
919 cache_rename(ap->a_fnch, ap->a_tnch);
920 cache_setvp(ap->a_tnch, ip->vp);
924 hammer_commit_transaction(&trans);
926 hammer_abort_transaction(&trans);
928 hammer_done_cursor(&cursor);
933 * hammer_vop_nrmdir { nch, dvp, cred }
937 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
939 /* XXX check that directory is empty */
941 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
945 * hammer_vop_setattr { vp, vap, cred }
949 hammer_vop_setattr(struct vop_setattr_args *ap)
951 struct hammer_transaction trans;
953 struct hammer_inode *ip;
960 ip = ap->a_vp->v_data;
963 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
966 hammer_start_transaction(&trans, ip->hmp);
969 if (vap->va_flags != VNOVAL) {
970 flags = ip->ino_data.uflags;
971 error = vop_helper_setattr_flags(&flags, vap->va_flags,
972 hammer_to_unix_xid(&ip->ino_data.uid),
975 if (ip->ino_data.uflags != flags) {
976 ip->ino_data.uflags = flags;
977 modflags |= HAMMER_INODE_DDIRTY;
979 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
986 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
990 if (vap->va_uid != (uid_t)VNOVAL) {
991 hammer_guid_to_uuid(&uuid, vap->va_uid);
992 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
993 ip->ino_data.uid = uuid;
994 modflags |= HAMMER_INODE_DDIRTY;
997 if (vap->va_gid != (uid_t)VNOVAL) {
998 hammer_guid_to_uuid(&uuid, vap->va_gid);
999 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1000 ip->ino_data.gid = uuid;
1001 modflags |= HAMMER_INODE_DDIRTY;
1004 if (vap->va_size != VNOVAL) {
1005 switch(ap->a_vp->v_type) {
1008 error = hammer_ip_delete_range(&trans, ip,
1010 0x7FFFFFFFFFFFFFFFLL);
1017 if (vap->va_atime.tv_sec != VNOVAL) {
1018 ip->ino_rec.ino_atime =
1019 hammer_timespec_to_transid(&vap->va_atime);
1020 modflags |= HAMMER_INODE_ITIMES;
1022 if (vap->va_mtime.tv_sec != VNOVAL) {
1023 ip->ino_rec.ino_mtime =
1024 hammer_timespec_to_transid(&vap->va_mtime);
1025 modflags |= HAMMER_INODE_ITIMES;
1027 if (vap->va_mode != (mode_t)VNOVAL) {
1028 if (ip->ino_data.mode != vap->va_mode) {
1029 ip->ino_data.mode = vap->va_mode;
1030 modflags |= HAMMER_INODE_DDIRTY;
1035 hammer_abort_transaction(&trans);
1038 hammer_modify_inode(&trans, ip, modflags);
1039 hammer_commit_transaction(&trans);
1045 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1049 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1055 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1059 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1061 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1065 * hammer_vop_strategy { vp, bio }
1067 * Strategy call, used for regular file read & write only. Note that the
1068 * bp may represent a cluster.
1070 * To simplify operation and allow better optimizations in the future,
1071 * this code does not make any assumptions with regards to buffer alignment
1076 hammer_vop_strategy(struct vop_strategy_args *ap)
1081 bp = ap->a_bio->bio_buf;
1085 error = hammer_vop_strategy_read(ap);
1088 error = hammer_vop_strategy_write(ap);
1094 bp->b_error = error;
1096 bp->b_flags |= B_ERROR;
1102 * Read from a regular file. Iterate the related records and fill in the
1103 * BIO/BUF. Gaps are zero-filled.
1105 * The support code in hammer_object.c should be used to deal with mixed
1106 * in-memory and on-disk records.
1112 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1114 struct hammer_inode *ip = ap->a_vp->v_data;
1115 struct hammer_cursor cursor;
1116 hammer_record_ondisk_t rec;
1117 hammer_base_elm_t base;
1130 hammer_init_cursor_ip(&cursor, ip);
1133 * Key range (begin and end inclusive) to scan. Note that the key's
1134 * stored in the actual records represent the
1136 cursor.key_beg.obj_id = ip->obj_id;
1137 cursor.key_beg.create_tid = ip->obj_asof;
1138 cursor.key_beg.delete_tid = 0;
1139 cursor.key_beg.obj_type = 0;
1140 cursor.key_beg.key = bio->bio_offset;
1142 cursor.key_end = cursor.key_beg;
1143 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1144 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1145 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1146 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1148 ran_end = bio->bio_offset + bp->b_bufsize - 1;
1149 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1150 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1151 if (ran_end + MAXPHYS < ran_end)
1152 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1154 cursor.key_end.key = ran_end + MAXPHYS;
1157 error = hammer_ip_first(&cursor, ip);
1160 while (error == 0) {
1161 error = hammer_ip_resolve_data(&cursor);
1164 rec = cursor.record;
1165 base = &rec->base.base;
1167 rec_offset = base->key - rec->data.base.data_len + 1;
1170 * Calculate the gap, if any, and zero-fill it.
1172 n = (int)(rec_offset - (bio->bio_offset + boff));
1174 if (n > bp->b_bufsize - boff)
1175 n = bp->b_bufsize - boff;
1176 kprintf("zfill %d bytes\n", n);
1177 bzero((char *)bp->b_data + boff, n);
1183 * Calculate the data offset in the record and the number
1184 * of bytes we can copy.
1186 * Note there is a degenerate case here where boff may
1187 * already be at bp->b_bufsize.
1190 n = rec->data.base.data_len - roff;
1192 if (n > bp->b_bufsize - boff)
1193 n = bp->b_bufsize - boff;
1194 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1196 if (boff == bp->b_bufsize)
1198 error = hammer_ip_next(&cursor);
1200 hammer_done_cursor(&cursor);
1203 * There may have been a gap after the last record
1205 if (error == ENOENT)
1207 if (error == 0 && boff != bp->b_bufsize) {
1208 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1209 /* boff = bp->b_bufsize; */
1216 * Write to a regular file. Iterate the related records and mark for
1217 * deletion. If existing edge records (left and right side) overlap our
1218 * write they have to be marked deleted and new records created, usually
1219 * referencing a portion of the original data. Then add a record to
1220 * represent the buffer.
1222 * The support code in hammer_object.c should be used to deal with mixed
1223 * in-memory and on-disk records.
1227 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1229 struct hammer_transaction trans;
1237 ip = ap->a_vp->v_data;
1238 hammer_start_transaction(&trans, ip->hmp);
1241 * Delete any records overlapping our range. This function will
1244 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1245 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1248 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1249 bio->bio_offset + bp->b_bufsize - 1);
1253 * Add a single record to cover the write
1256 error = hammer_ip_add_data(&trans, ip, bio->bio_offset,
1257 bp->b_data, bp->b_bufsize);
1261 * If an error occured abort the transaction
1264 /* XXX undo deletion */
1265 hammer_abort_transaction(&trans);
1266 bp->b_resid = bp->b_bufsize;
1268 hammer_commit_transaction(&trans);
1275 * dounlink - disconnect a directory entry
1277 * XXX whiteout support not really in yet
1280 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1283 struct hammer_transaction trans;
1284 struct namecache *ncp;
1287 hammer_record_ondisk_t rec;
1288 struct hammer_cursor cursor;
1293 * Calculate the namekey and setup the key range for the scan. This
1294 * works kinda like a chained hash table where the lower 32 bits
1295 * of the namekey synthesize the chain.
1297 * The key range is inclusive of both key_beg and key_end.
1301 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1303 hammer_init_cursor_ip(&cursor, dip);
1304 cursor.key_beg.obj_id = dip->obj_id;
1305 cursor.key_beg.key = namekey;
1306 cursor.key_beg.create_tid = dip->obj_asof;
1307 cursor.key_beg.delete_tid = 0;
1308 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1309 cursor.key_beg.obj_type = 0;
1311 cursor.key_end = cursor.key_beg;
1312 cursor.key_end.key |= 0xFFFFFFFFULL;
1314 hammer_start_transaction(&trans, dip->hmp);
1317 * Scan all matching records (the chain), locate the one matching
1318 * the requested path component. info->last_error contains the
1319 * error code on search termination and could be 0, ENOENT, or
1322 * The hammer_ip_*() functions merge in-memory records with on-disk
1323 * records for the purposes of the search.
1325 error = hammer_ip_first(&cursor, dip);
1326 while (error == 0) {
1327 error = hammer_ip_resolve_data(&cursor);
1330 rec = cursor.record;
1331 if (ncp->nc_nlen == rec->entry.base.data_len &&
1332 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1335 error = hammer_ip_next(&cursor);
1339 * If all is ok we have to get the inode so we can adjust nlinks.
1342 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id, &error);
1344 error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1346 cache_setunresolved(nch);
1347 cache_setvp(nch, NULL);
1350 cache_inval_vp(ip->vp, CINV_DESTROY);
1352 hammer_rel_inode(ip, 0);
1355 hammer_commit_transaction(&trans);
1357 hammer_abort_transaction(&trans);
1360 hammer_done_cursor(&cursor);