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 1.9 2007/11/30 00:16:56 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>
46 #include <vm/vm_extern.h>
52 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
53 static int hammer_vop_fsync(struct vop_fsync_args *);
54 static int hammer_vop_read(struct vop_read_args *);
55 static int hammer_vop_write(struct vop_write_args *);
56 static int hammer_vop_access(struct vop_access_args *);
57 static int hammer_vop_advlock(struct vop_advlock_args *);
58 static int hammer_vop_close(struct vop_close_args *);
59 static int hammer_vop_ncreate(struct vop_ncreate_args *);
60 static int hammer_vop_getattr(struct vop_getattr_args *);
61 static int hammer_vop_nresolve(struct vop_nresolve_args *);
62 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
63 static int hammer_vop_nlink(struct vop_nlink_args *);
64 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
65 static int hammer_vop_nmknod(struct vop_nmknod_args *);
66 static int hammer_vop_open(struct vop_open_args *);
67 static int hammer_vop_pathconf(struct vop_pathconf_args *);
68 static int hammer_vop_print(struct vop_print_args *);
69 static int hammer_vop_readdir(struct vop_readdir_args *);
70 static int hammer_vop_readlink(struct vop_readlink_args *);
71 static int hammer_vop_nremove(struct vop_nremove_args *);
72 static int hammer_vop_nrename(struct vop_nrename_args *);
73 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
74 static int hammer_vop_setattr(struct vop_setattr_args *);
75 static int hammer_vop_strategy(struct vop_strategy_args *);
76 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
77 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
79 struct vop_ops hammer_vnode_vops = {
80 .vop_default = vop_defaultop,
81 .vop_fsync = hammer_vop_fsync,
82 .vop_getpages = vop_stdgetpages,
83 .vop_putpages = vop_stdputpages,
84 .vop_read = hammer_vop_read,
85 .vop_write = hammer_vop_write,
86 .vop_access = hammer_vop_access,
87 .vop_advlock = hammer_vop_advlock,
88 .vop_close = hammer_vop_close,
89 .vop_ncreate = hammer_vop_ncreate,
90 .vop_getattr = hammer_vop_getattr,
91 .vop_inactive = hammer_vop_inactive,
92 .vop_reclaim = hammer_vop_reclaim,
93 .vop_nresolve = hammer_vop_nresolve,
94 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
95 .vop_nlink = hammer_vop_nlink,
96 .vop_nmkdir = hammer_vop_nmkdir,
97 .vop_nmknod = hammer_vop_nmknod,
98 .vop_open = hammer_vop_open,
99 .vop_pathconf = hammer_vop_pathconf,
100 .vop_print = hammer_vop_print,
101 .vop_readdir = hammer_vop_readdir,
102 .vop_readlink = hammer_vop_readlink,
103 .vop_nremove = hammer_vop_nremove,
104 .vop_nrename = hammer_vop_nrename,
105 .vop_nrmdir = hammer_vop_nrmdir,
106 .vop_setattr = hammer_vop_setattr,
107 .vop_strategy = hammer_vop_strategy,
108 .vop_nsymlink = hammer_vop_nsymlink,
109 .vop_nwhiteout = hammer_vop_nwhiteout
112 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
113 struct ucred *cred, int flags);
114 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
115 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
120 hammer_vop_vnoperate(struct vop_generic_args *)
122 return (VOCALL(&hammer_vnode_vops, ap));
127 * hammer_vop_fsync { vp, waitfor }
131 hammer_vop_fsync(struct vop_fsync_args *ap)
137 error = hammer_sync_inode(ip, ap->a_waitfor, 0);
142 * hammer_vop_read { vp, uio, ioflag, cred }
146 hammer_vop_read(struct vop_read_args *ap)
148 struct hammer_transaction trans;
157 if (ap->a_vp->v_type != VREG)
161 seqcount = ap->a_ioflag >> 16;
163 hammer_start_transaction(&trans, ip->hmp);
166 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
169 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
170 offset = uio->uio_offset & HAMMER_BUFMASK;
172 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
173 uio->uio_offset - offset, HAMMER_BUFSIZE,
174 MAXBSIZE, seqcount, &bp);
176 error = bread(ap->a_vp, uio->uio_offset - offset,
177 HAMMER_BUFSIZE, &bp);
182 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
183 n = HAMMER_BUFSIZE - offset;
184 if (n > uio->uio_resid)
186 if (n > ip->ino_rec.ino_size - uio->uio_offset)
187 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
188 error = uiomove((char *)bp->b_data + offset, n, uio);
193 ip->ino_rec.ino_atime = trans.tid;
194 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
197 hammer_commit_transaction(&trans);
202 * hammer_vop_write { vp, uio, ioflag, cred }
206 hammer_vop_write(struct vop_write_args *ap)
208 struct hammer_transaction trans;
209 struct hammer_inode *ip;
217 if (ap->a_vp->v_type != VREG)
223 * Create a transaction to cover the operations we perform.
225 hammer_start_transaction(&trans, ip->hmp);
231 if (ap->a_ioflag & IO_APPEND)
232 uio->uio_offset = ip->ino_rec.ino_size;
235 * Check for illegal write offsets. Valid range is 0...2^63-1
237 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0)
241 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
243 while (uio->uio_resid > 0) {
244 offset = uio->uio_offset & HAMMER_BUFMASK;
245 if (uio->uio_segflg == UIO_NOCOPY) {
247 * Issuing a write with the same data backing the
248 * buffer. Instantiate the buffer to collect the
249 * backing vm pages, then read-in any missing bits.
251 * This case is used by vop_stdputpages().
253 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
255 if ((bp->b_flags & B_CACHE) == 0) {
257 error = bread(ap->a_vp,
258 uio->uio_offset - offset,
259 HAMMER_BUFSIZE, &bp);
265 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
267 * entirely overwrite the buffer
269 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
271 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
275 bp = getblk(ap->a_vp, uio->uio_offset, HAMMER_BUFSIZE,
280 * Partial overwrite, read in any missing bits then
281 * replace the portion being written.
283 error = bread(ap->a_vp, uio->uio_offset - offset,
284 HAMMER_BUFSIZE, &bp);
290 n = HAMMER_BUFSIZE - offset;
291 if (n > uio->uio_resid)
293 error = uiomove((char *)bp->b_data + offset, n, uio);
298 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
299 if (ip->ino_rec.ino_size < uio->uio_offset) {
300 ip->ino_rec.ino_size = uio->uio_offset;
301 ip->ino_rec.ino_mtime = trans.tid;
302 flags = HAMMER_INODE_RDIRTY | HAMMER_INODE_ITIMES |
304 vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
306 flags = HAMMER_INODE_TID;
308 hammer_modify_inode(&trans, ip, flags);
309 if (ap->a_ioflag & IO_SYNC) {
311 } else if (ap->a_ioflag & IO_DIRECT) {
318 hammer_abort_transaction(&trans);
320 hammer_commit_transaction(&trans);
325 * hammer_vop_access { vp, mode, cred }
329 hammer_vop_access(struct vop_access_args *ap)
331 struct hammer_inode *ip = VTOI(ap->a_vp);
336 uid = hammer_to_unix_xid(&ip->ino_data.uid);
337 gid = hammer_to_unix_xid(&ip->ino_data.gid);
339 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
340 ip->ino_data.uflags);
345 * hammer_vop_advlock { vp, id, op, fl, flags }
349 hammer_vop_advlock(struct vop_advlock_args *ap)
351 struct hammer_inode *ip = VTOI(ap->a_vp);
353 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
357 * hammer_vop_close { vp, fflag }
361 hammer_vop_close(struct vop_close_args *ap)
363 return (vop_stdclose(ap));
367 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
369 * The operating system has already ensured that the directory entry
370 * does not exist and done all appropriate namespace locking.
374 hammer_vop_ncreate(struct vop_ncreate_args *ap)
376 struct hammer_transaction trans;
377 struct hammer_inode *dip;
378 struct hammer_inode *nip;
379 struct nchandle *nch;
383 dip = VTOI(ap->a_dvp);
386 * Create a transaction to cover the operations we perform.
388 hammer_start_transaction(&trans, dip->hmp);
391 * Create a new filesystem object of the requested type. The
392 * returned inode will be referenceds but not locked.
395 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
397 hammer_abort_transaction(&trans);
403 * Add the new filesystem object to the directory. This will also
404 * bump the inode's link count.
406 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
412 hammer_rel_inode(nip, 0);
413 hammer_abort_transaction(&trans);
416 hammer_commit_transaction(&trans);
417 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
418 hammer_rel_inode(nip, 0);
420 cache_setunresolved(ap->a_nch);
421 cache_setvp(ap->a_nch, *ap->a_vpp);
428 * hammer_vop_getattr { vp, vap }
432 hammer_vop_getattr(struct vop_getattr_args *ap)
434 struct hammer_inode *ip = VTOI(ap->a_vp);
435 struct vattr *vap = ap->a_vap;
438 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
439 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
444 hammer_itimes(ap->a_vp);
447 vap->va_fsid = ip->hmp->fsid_udev;
448 vap->va_fileid = ip->ino_rec.base.base.obj_id;
449 vap->va_mode = ip->ino_data.mode;
450 vap->va_nlink = ip->ino_rec.ino_nlinks;
451 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
452 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
455 vap->va_size = ip->ino_rec.ino_size;
456 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
457 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
458 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
459 vap->va_flags = ip->ino_data.uflags;
460 vap->va_gen = 1; /* hammer inums are unique for all time */
461 vap->va_blocksize = 32768; /* XXX - extract from root volume */
462 vap->va_bytes = ip->ino_rec.ino_size;
463 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
464 vap->va_filerev = 0; /* XXX */
465 /* mtime uniquely identifies any adjustments made to the file */
466 vap->va_fsmid = ip->ino_rec.ino_mtime;
467 vap->va_uid_uuid = ip->ino_data.uid;
468 vap->va_gid_uuid = ip->ino_data.gid;
469 vap->va_fsid_uuid = ip->hmp->fsid;
470 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
476 * hammer_vop_nresolve { nch, dvp, cred }
478 * Locate the requested directory entry.
482 hammer_vop_nresolve(struct vop_nresolve_args *ap)
484 struct namecache *ncp;
488 struct hammer_cursor cursor;
489 union hammer_record_ondisk *rec;
497 * Misc initialization, plus handle as-of name extensions. Look for
498 * the '@@' extension. Note that as-of files and directories cannot
503 dip = VTOI(ap->a_dvp);
504 ncp = ap->a_nch->ncp;
505 asof = dip->obj_asof;
508 for (i = 0; i < nlen; ++i) {
509 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
510 asof = hammer_now_tid() -
511 strtoq(ncp->nc_name + i + 2, NULL, 0) *
513 kprintf("ASOF %016llx\n", asof);
520 * Calculate the namekey and setup the key range for the scan. This
521 * works kinda like a chained hash table where the lower 32 bits
522 * of the namekey synthesize the chain.
524 * The key range is inclusive of both key_beg and key_end.
526 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
528 hammer_init_cursor_ip(&cursor, dip);
529 cursor.key_beg.obj_id = dip->obj_id;
530 cursor.key_beg.key = namekey;
531 cursor.key_beg.create_tid = asof;
532 cursor.key_beg.delete_tid = 0;
533 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
534 cursor.key_beg.obj_type = 0;
536 cursor.key_end = cursor.key_beg;
537 cursor.key_end.key |= 0xFFFFFFFFULL;
540 * Scan all matching records (the chain), locate the one matching
541 * the requested path component.
543 * The hammer_ip_*() functions merge in-memory records with on-disk
544 * records for the purposes of the search.
546 error = hammer_ip_first(&cursor, dip);
548 error = hammer_ip_resolve_data(&cursor);
552 if (nlen == rec->entry.base.data_len &&
553 bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
556 error = hammer_ip_next(&cursor);
559 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id,
562 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
563 hammer_rel_inode(ip, 0);
569 cache_setvp(ap->a_nch, vp);
572 } else if (error == ENOENT) {
573 cache_setvp(ap->a_nch, NULL);
575 hammer_done_cursor(&cursor);
580 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
582 * Locate the parent directory of a directory vnode.
584 * dvp is referenced but not locked. *vpp must be returned referenced and
585 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
586 * at the root, instead it could indicate that the directory we were in was
591 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
593 struct hammer_inode *dip;
594 u_int64_t parent_obj_id;
596 dip = VTOI(ap->a_dvp);
597 if ((parent_obj_id = dip->ino_data.parent_obj_id) == 0) {
601 return(hammer_vfs_vget(dip->hmp->mp, parent_obj_id, ap->a_vpp));
605 * hammer_vop_nlink { nch, dvp, vp, cred }
609 hammer_vop_nlink(struct vop_nlink_args *ap)
611 struct hammer_transaction trans;
612 struct hammer_inode *dip;
613 struct hammer_inode *ip;
614 struct nchandle *nch;
618 dip = VTOI(ap->a_dvp);
622 * Create a transaction to cover the operations we perform.
624 hammer_start_transaction(&trans, dip->hmp);
627 * Add the filesystem object to the directory. Note that neither
628 * dip nor ip are referenced or locked, but their vnodes are
629 * referenced. This function will bump the inode's link count.
631 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
637 hammer_abort_transaction(&trans);
639 cache_setunresolved(nch);
640 cache_setvp(nch, ap->a_vp);
641 hammer_commit_transaction(&trans);
647 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
649 * The operating system has already ensured that the directory entry
650 * does not exist and done all appropriate namespace locking.
654 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
656 struct hammer_transaction trans;
657 struct hammer_inode *dip;
658 struct hammer_inode *nip;
659 struct nchandle *nch;
663 dip = VTOI(ap->a_dvp);
666 * Create a transaction to cover the operations we perform.
668 hammer_start_transaction(&trans, dip->hmp);
671 * Create a new filesystem object of the requested type. The
672 * returned inode will be referenced but not locked.
674 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
676 hammer_abort_transaction(&trans);
682 * Add the new filesystem object to the directory. This will also
683 * bump the inode's link count.
685 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
691 hammer_rel_inode(nip, 0);
692 hammer_abort_transaction(&trans);
695 hammer_commit_transaction(&trans);
696 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
697 hammer_rel_inode(nip, 0);
699 cache_setunresolved(ap->a_nch);
700 cache_setvp(ap->a_nch, *ap->a_vpp);
707 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
709 * The operating system has already ensured that the directory entry
710 * does not exist and done all appropriate namespace locking.
714 hammer_vop_nmknod(struct vop_nmknod_args *ap)
716 struct hammer_transaction trans;
717 struct hammer_inode *dip;
718 struct hammer_inode *nip;
719 struct nchandle *nch;
723 dip = VTOI(ap->a_dvp);
726 * Create a transaction to cover the operations we perform.
728 hammer_start_transaction(&trans, dip->hmp);
731 * Create a new filesystem object of the requested type. The
732 * returned inode will be referenced but not locked.
734 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
736 hammer_abort_transaction(&trans);
742 * Add the new filesystem object to the directory. This will also
743 * bump the inode's link count.
745 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
751 hammer_rel_inode(nip, 0);
752 hammer_abort_transaction(&trans);
755 hammer_commit_transaction(&trans);
756 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
757 hammer_rel_inode(nip, 0);
759 cache_setunresolved(ap->a_nch);
760 cache_setvp(ap->a_nch, *ap->a_vpp);
767 * hammer_vop_open { vp, mode, cred, fp }
771 hammer_vop_open(struct vop_open_args *ap)
773 return(vop_stdopen(ap));
777 * hammer_vop_pathconf { vp, name, retval }
781 hammer_vop_pathconf(struct vop_pathconf_args *ap)
787 * hammer_vop_print { vp }
791 hammer_vop_print(struct vop_print_args *ap)
797 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
801 hammer_vop_readdir(struct vop_readdir_args *ap)
803 struct hammer_cursor cursor;
804 struct hammer_inode *ip;
806 hammer_record_ondisk_t rec;
807 hammer_base_elm_t base;
817 hammer_init_cursor_ip(&cursor, ip);
820 * Key range (begin and end inclusive) to scan. Directory keys
821 * directly translate to a 64 bit 'seek' position.
823 cursor.key_beg.obj_id = ip->obj_id;
824 cursor.key_beg.create_tid = ip->obj_asof;
825 cursor.key_beg.delete_tid = 0;
826 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
827 cursor.key_beg.obj_type = 0;
828 cursor.key_beg.key = uio->uio_offset;
830 cursor.key_end = cursor.key_beg;
831 cursor.key_end.key = HAMMER_MAX_KEY;
833 if (ap->a_ncookies) {
834 ncookies = uio->uio_resid / 16 + 1;
837 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
845 saveoff = cursor.key_beg.key;
846 error = hammer_ip_first(&cursor, ip);
849 error = hammer_ip_resolve_data(&cursor);
853 base = &rec->base.base;
856 r = vop_write_dirent(
857 &error, uio, rec->entry.obj_id,
858 rec->entry.base.data_len,
859 hammer_get_dtype(rec->entry.base.base.obj_type),
860 (void *)cursor.data);
865 cookies[cookie_index] = base->key;
867 if (cookie_index == ncookies)
869 error = hammer_ip_next(&cursor);
871 hammer_done_cursor(&cursor);
874 *ap->a_eofflag = (error == ENOENT);
877 uio->uio_offset = saveoff;
878 if (error && cookie_index == 0) {
880 kfree(cookies, M_TEMP);
882 *ap->a_cookies = NULL;
887 *ap->a_ncookies = cookie_index;
888 *ap->a_cookies = cookies;
895 * hammer_vop_readlink { vp, uio, cred }
899 hammer_vop_readlink(struct vop_readlink_args *ap)
905 * hammer_vop_nremove { nch, dvp, cred }
909 hammer_vop_nremove(struct vop_nremove_args *ap)
911 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
915 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
919 hammer_vop_nrename(struct vop_nrename_args *ap)
921 struct hammer_transaction trans;
922 struct namecache *fncp;
923 struct namecache *tncp;
924 struct hammer_inode *fdip;
925 struct hammer_inode *tdip;
926 struct hammer_inode *ip;
927 struct hammer_cursor cursor;
928 union hammer_record_ondisk *rec;
932 fdip = VTOI(ap->a_fdvp);
933 tdip = VTOI(ap->a_tdvp);
934 fncp = ap->a_fnch->ncp;
935 tncp = ap->a_tnch->ncp;
936 hammer_start_transaction(&trans, fdip->hmp);
939 * Extract the hammer_inode from fncp and add link to the target
942 ip = VTOI(fncp->nc_vp);
943 KKASSERT(ip != NULL);
945 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
948 * Locate the record in the originating directory and remove it.
950 * Calculate the namekey and setup the key range for the scan. This
951 * works kinda like a chained hash table where the lower 32 bits
952 * of the namekey synthesize the chain.
954 * The key range is inclusive of both key_beg and key_end.
956 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
958 hammer_init_cursor_ip(&cursor, fdip);
959 cursor.key_beg.obj_id = fdip->obj_id;
960 cursor.key_beg.key = namekey;
961 cursor.key_beg.create_tid = fdip->obj_asof;
962 cursor.key_beg.delete_tid = 0;
963 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
964 cursor.key_beg.obj_type = 0;
966 cursor.key_end = cursor.key_beg;
967 cursor.key_end.key |= 0xFFFFFFFFULL;
970 * Scan all matching records (the chain), locate the one matching
971 * the requested path component.
973 * The hammer_ip_*() functions merge in-memory records with on-disk
974 * records for the purposes of the search.
976 error = hammer_ip_first(&cursor, fdip);
978 if (hammer_ip_resolve_data(&cursor) != 0)
981 if (fncp->nc_nlen == rec->entry.base.data_len &&
982 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
985 error = hammer_ip_next(&cursor);
989 * If all is ok we have to get the inode so we can adjust nlinks.
993 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
995 cache_rename(ap->a_fnch, ap->a_tnch);
996 cache_setvp(ap->a_tnch, ip->vp);
999 hammer_done_cursor(&cursor);
1001 hammer_commit_transaction(&trans);
1003 hammer_abort_transaction(&trans);
1009 * hammer_vop_nrmdir { nch, dvp, cred }
1013 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1015 /* XXX check that directory is empty */
1017 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1021 * hammer_vop_setattr { vp, vap, cred }
1025 hammer_vop_setattr(struct vop_setattr_args *ap)
1027 struct hammer_transaction trans;
1029 struct hammer_inode *ip;
1032 int64_t aligned_size;
1037 ip = ap->a_vp->v_data;
1040 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1043 hammer_start_transaction(&trans, ip->hmp);
1046 if (vap->va_flags != VNOVAL) {
1047 flags = ip->ino_data.uflags;
1048 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1049 hammer_to_unix_xid(&ip->ino_data.uid),
1052 if (ip->ino_data.uflags != flags) {
1053 ip->ino_data.uflags = flags;
1054 modflags |= HAMMER_INODE_DDIRTY;
1056 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1063 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1067 if (vap->va_uid != (uid_t)VNOVAL) {
1068 hammer_guid_to_uuid(&uuid, vap->va_uid);
1069 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1070 ip->ino_data.uid = uuid;
1071 modflags |= HAMMER_INODE_DDIRTY;
1074 if (vap->va_gid != (uid_t)VNOVAL) {
1075 hammer_guid_to_uuid(&uuid, vap->va_gid);
1076 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1077 ip->ino_data.gid = uuid;
1078 modflags |= HAMMER_INODE_DDIRTY;
1081 if (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1082 switch(ap->a_vp->v_type) {
1084 if (vap->va_size < ip->ino_rec.ino_size) {
1085 vtruncbuf(ap->a_vp, vap->va_size,
1087 } else if (vap->va_size > ip->ino_rec.ino_size) {
1088 vnode_pager_setsize(ap->a_vp, vap->va_size);
1090 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1091 ~(int64_t)HAMMER_BUFMASK;
1092 error = hammer_ip_delete_range(&trans, ip,
1094 0x7FFFFFFFFFFFFFFFLL);
1095 ip->ino_rec.ino_size = vap->va_size;
1096 modflags |= HAMMER_INODE_RDIRTY;
1099 error = hammer_ip_delete_range(&trans, ip,
1101 0x7FFFFFFFFFFFFFFFLL);
1102 ip->ino_rec.ino_size = vap->va_size;
1103 modflags |= HAMMER_INODE_RDIRTY;
1110 if (vap->va_atime.tv_sec != VNOVAL) {
1111 ip->ino_rec.ino_atime =
1112 hammer_timespec_to_transid(&vap->va_atime);
1113 modflags |= HAMMER_INODE_ITIMES;
1115 if (vap->va_mtime.tv_sec != VNOVAL) {
1116 ip->ino_rec.ino_mtime =
1117 hammer_timespec_to_transid(&vap->va_mtime);
1118 modflags |= HAMMER_INODE_ITIMES;
1120 if (vap->va_mode != (mode_t)VNOVAL) {
1121 if (ip->ino_data.mode != vap->va_mode) {
1122 ip->ino_data.mode = vap->va_mode;
1123 modflags |= HAMMER_INODE_DDIRTY;
1128 hammer_abort_transaction(&trans);
1130 if (modflags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY))
1131 modflags |= HAMMER_INODE_TID;
1132 hammer_modify_inode(&trans, ip, modflags);
1133 hammer_commit_transaction(&trans);
1139 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1143 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1149 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1153 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1155 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1159 * hammer_vop_strategy { vp, bio }
1161 * Strategy call, used for regular file read & write only. Note that the
1162 * bp may represent a cluster.
1164 * To simplify operation and allow better optimizations in the future,
1165 * this code does not make any assumptions with regards to buffer alignment
1170 hammer_vop_strategy(struct vop_strategy_args *ap)
1175 bp = ap->a_bio->bio_buf;
1179 error = hammer_vop_strategy_read(ap);
1182 error = hammer_vop_strategy_write(ap);
1188 bp->b_error = error;
1190 bp->b_flags |= B_ERROR;
1196 * Read from a regular file. Iterate the related records and fill in the
1197 * BIO/BUF. Gaps are zero-filled.
1199 * The support code in hammer_object.c should be used to deal with mixed
1200 * in-memory and on-disk records.
1206 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1208 struct hammer_inode *ip = ap->a_vp->v_data;
1209 struct hammer_cursor cursor;
1210 hammer_record_ondisk_t rec;
1211 hammer_base_elm_t base;
1225 hammer_init_cursor_ip(&cursor, ip);
1228 * Key range (begin and end inclusive) to scan. Note that the key's
1229 * stored in the actual records represent BASE+LEN, not BASE. The
1230 * first record containing bio_offset will have a key > bio_offset.
1232 cursor.key_beg.obj_id = ip->obj_id;
1233 cursor.key_beg.create_tid = ip->obj_asof;
1234 cursor.key_beg.delete_tid = 0;
1235 cursor.key_beg.obj_type = 0;
1236 cursor.key_beg.key = bio->bio_offset + 1;
1238 cursor.key_end = cursor.key_beg;
1239 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1240 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1241 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1242 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1244 ran_end = bio->bio_offset + bp->b_bufsize;
1245 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1246 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1247 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1248 if (tmp64 < ran_end)
1249 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1251 cursor.key_end.key = ran_end + MAXPHYS + 1;
1254 error = hammer_ip_first(&cursor, ip);
1257 while (error == 0) {
1258 error = hammer_ip_resolve_data(&cursor);
1261 rec = cursor.record;
1262 base = &rec->base.base;
1264 rec_offset = base->key - rec->data.base.data_len;
1267 * Calculate the gap, if any, and zero-fill it.
1269 n = (int)(rec_offset - (bio->bio_offset + boff));
1271 if (n > bp->b_bufsize - boff)
1272 n = bp->b_bufsize - boff;
1273 kprintf("zfill %d bytes\n", n);
1274 bzero((char *)bp->b_data + boff, n);
1280 * Calculate the data offset in the record and the number
1281 * of bytes we can copy.
1283 * Note there is a degenerate case here where boff may
1284 * already be at bp->b_bufsize.
1287 n = rec->data.base.data_len - roff;
1289 if (n > bp->b_bufsize - boff)
1290 n = bp->b_bufsize - boff;
1291 bcopy((char *)cursor.data + roff, (char *)bp->b_data + boff, n);
1293 if (boff == bp->b_bufsize)
1295 error = hammer_ip_next(&cursor);
1297 hammer_done_cursor(&cursor);
1300 * There may have been a gap after the last record
1302 if (error == ENOENT)
1304 if (error == 0 && boff != bp->b_bufsize) {
1305 KKASSERT(boff < bp->b_bufsize);
1306 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1307 /* boff = bp->b_bufsize; */
1314 * Write to a regular file. Iterate the related records and mark for
1315 * deletion. If existing edge records (left and right side) overlap our
1316 * write they have to be marked deleted and new records created, usually
1317 * referencing a portion of the original data. Then add a record to
1318 * represent the buffer.
1320 * The support code in hammer_object.c should be used to deal with mixed
1321 * in-memory and on-disk records.
1325 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1327 struct hammer_transaction trans;
1335 ip = ap->a_vp->v_data;
1336 hammer_start_transaction(&trans, ip->hmp);
1339 * Delete any records overlapping our range. This function will
1342 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1343 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1346 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1347 bio->bio_offset + bp->b_bufsize - 1);
1351 * Add a single record to cover the write
1354 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1355 bp->b_data, bp->b_bufsize);
1359 * If an error occured abort the transaction
1362 /* XXX undo deletion */
1363 hammer_abort_transaction(&trans);
1364 bp->b_resid = bp->b_bufsize;
1366 hammer_commit_transaction(&trans);
1373 * dounlink - disconnect a directory entry
1375 * XXX whiteout support not really in yet
1378 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1381 struct hammer_transaction trans;
1382 struct namecache *ncp;
1385 hammer_record_ondisk_t rec;
1386 struct hammer_cursor cursor;
1391 * Calculate the namekey and setup the key range for the scan. This
1392 * works kinda like a chained hash table where the lower 32 bits
1393 * of the namekey synthesize the chain.
1395 * The key range is inclusive of both key_beg and key_end.
1399 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1401 hammer_init_cursor_ip(&cursor, dip);
1402 cursor.key_beg.obj_id = dip->obj_id;
1403 cursor.key_beg.key = namekey;
1404 cursor.key_beg.create_tid = dip->obj_asof;
1405 cursor.key_beg.delete_tid = 0;
1406 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1407 cursor.key_beg.obj_type = 0;
1409 cursor.key_end = cursor.key_beg;
1410 cursor.key_end.key |= 0xFFFFFFFFULL;
1412 hammer_start_transaction(&trans, dip->hmp);
1415 * Scan all matching records (the chain), locate the one matching
1416 * the requested path component. info->last_error contains the
1417 * error code on search termination and could be 0, ENOENT, or
1420 * The hammer_ip_*() functions merge in-memory records with on-disk
1421 * records for the purposes of the search.
1423 error = hammer_ip_first(&cursor, dip);
1424 while (error == 0) {
1425 error = hammer_ip_resolve_data(&cursor);
1428 rec = cursor.record;
1429 if (ncp->nc_nlen == rec->entry.base.data_len &&
1430 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1433 error = hammer_ip_next(&cursor);
1437 * If all is ok we have to get the inode so we can adjust nlinks.
1440 ip = hammer_get_inode(dip->hmp, rec->entry.obj_id,
1441 dip->hmp->asof, &error);
1443 error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1445 cache_setunresolved(nch);
1446 cache_setvp(nch, NULL);
1449 cache_inval_vp(ip->vp, CINV_DESTROY);
1451 hammer_rel_inode(ip, 0);
1454 hammer_commit_transaction(&trans);
1456 hammer_abort_transaction(&trans);
1459 hammer_done_cursor(&cursor);