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 1.36 2008/03/19 20:18:17 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 <sys/dirent.h>
47 #include <vm/vm_extern.h>
48 #include <vfs/fifofs/fifo.h>
54 /*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
55 static int hammer_vop_fsync(struct vop_fsync_args *);
56 static int hammer_vop_read(struct vop_read_args *);
57 static int hammer_vop_write(struct vop_write_args *);
58 static int hammer_vop_access(struct vop_access_args *);
59 static int hammer_vop_advlock(struct vop_advlock_args *);
60 static int hammer_vop_close(struct vop_close_args *);
61 static int hammer_vop_ncreate(struct vop_ncreate_args *);
62 static int hammer_vop_getattr(struct vop_getattr_args *);
63 static int hammer_vop_nresolve(struct vop_nresolve_args *);
64 static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
65 static int hammer_vop_nlink(struct vop_nlink_args *);
66 static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
67 static int hammer_vop_nmknod(struct vop_nmknod_args *);
68 static int hammer_vop_open(struct vop_open_args *);
69 static int hammer_vop_pathconf(struct vop_pathconf_args *);
70 static int hammer_vop_print(struct vop_print_args *);
71 static int hammer_vop_readdir(struct vop_readdir_args *);
72 static int hammer_vop_readlink(struct vop_readlink_args *);
73 static int hammer_vop_nremove(struct vop_nremove_args *);
74 static int hammer_vop_nrename(struct vop_nrename_args *);
75 static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
76 static int hammer_vop_setattr(struct vop_setattr_args *);
77 static int hammer_vop_strategy(struct vop_strategy_args *);
78 static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
79 static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
80 static int hammer_vop_ioctl(struct vop_ioctl_args *);
81 static int hammer_vop_mountctl(struct vop_mountctl_args *);
83 static int hammer_vop_fifoclose (struct vop_close_args *);
84 static int hammer_vop_fiforead (struct vop_read_args *);
85 static int hammer_vop_fifowrite (struct vop_write_args *);
87 static int hammer_vop_specclose (struct vop_close_args *);
88 static int hammer_vop_specread (struct vop_read_args *);
89 static int hammer_vop_specwrite (struct vop_write_args *);
91 struct vop_ops hammer_vnode_vops = {
92 .vop_default = vop_defaultop,
93 .vop_fsync = hammer_vop_fsync,
94 .vop_getpages = vop_stdgetpages,
95 .vop_putpages = vop_stdputpages,
96 .vop_read = hammer_vop_read,
97 .vop_write = hammer_vop_write,
98 .vop_access = hammer_vop_access,
99 .vop_advlock = hammer_vop_advlock,
100 .vop_close = hammer_vop_close,
101 .vop_ncreate = hammer_vop_ncreate,
102 .vop_getattr = hammer_vop_getattr,
103 .vop_inactive = hammer_vop_inactive,
104 .vop_reclaim = hammer_vop_reclaim,
105 .vop_nresolve = hammer_vop_nresolve,
106 .vop_nlookupdotdot = hammer_vop_nlookupdotdot,
107 .vop_nlink = hammer_vop_nlink,
108 .vop_nmkdir = hammer_vop_nmkdir,
109 .vop_nmknod = hammer_vop_nmknod,
110 .vop_open = hammer_vop_open,
111 .vop_pathconf = hammer_vop_pathconf,
112 .vop_print = hammer_vop_print,
113 .vop_readdir = hammer_vop_readdir,
114 .vop_readlink = hammer_vop_readlink,
115 .vop_nremove = hammer_vop_nremove,
116 .vop_nrename = hammer_vop_nrename,
117 .vop_nrmdir = hammer_vop_nrmdir,
118 .vop_setattr = hammer_vop_setattr,
119 .vop_strategy = hammer_vop_strategy,
120 .vop_nsymlink = hammer_vop_nsymlink,
121 .vop_nwhiteout = hammer_vop_nwhiteout,
122 .vop_ioctl = hammer_vop_ioctl,
123 .vop_mountctl = hammer_vop_mountctl
126 struct vop_ops hammer_spec_vops = {
127 .vop_default = spec_vnoperate,
128 .vop_fsync = hammer_vop_fsync,
129 .vop_read = hammer_vop_specread,
130 .vop_write = hammer_vop_specwrite,
131 .vop_access = hammer_vop_access,
132 .vop_close = hammer_vop_specclose,
133 .vop_getattr = hammer_vop_getattr,
134 .vop_inactive = hammer_vop_inactive,
135 .vop_reclaim = hammer_vop_reclaim,
136 .vop_setattr = hammer_vop_setattr
139 struct vop_ops hammer_fifo_vops = {
140 .vop_default = fifo_vnoperate,
141 .vop_fsync = hammer_vop_fsync,
142 .vop_read = hammer_vop_fiforead,
143 .vop_write = hammer_vop_fifowrite,
144 .vop_access = hammer_vop_access,
145 .vop_close = hammer_vop_fifoclose,
146 .vop_getattr = hammer_vop_getattr,
147 .vop_inactive = hammer_vop_inactive,
148 .vop_reclaim = hammer_vop_reclaim,
149 .vop_setattr = hammer_vop_setattr
152 static int hammer_dounlink(struct nchandle *nch, struct vnode *dvp,
153 struct ucred *cred, int flags);
154 static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
155 static int hammer_vop_strategy_write(struct vop_strategy_args *ap);
160 hammer_vop_vnoperate(struct vop_generic_args *)
162 return (VOCALL(&hammer_vnode_vops, ap));
167 * hammer_vop_fsync { vp, waitfor }
171 hammer_vop_fsync(struct vop_fsync_args *ap)
177 error = hammer_sync_inode(ip, ap->a_waitfor, 0);
182 * hammer_vop_read { vp, uio, ioflag, cred }
186 hammer_vop_read(struct vop_read_args *ap)
188 struct hammer_transaction trans;
197 if (ap->a_vp->v_type != VREG)
201 seqcount = ap->a_ioflag >> 16;
203 hammer_start_transaction(&trans, ip->hmp);
206 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
209 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
210 offset = uio->uio_offset & HAMMER_BUFMASK;
212 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
213 uio->uio_offset - offset, HAMMER_BUFSIZE,
214 MAXBSIZE, seqcount, &bp);
216 error = bread(ap->a_vp, uio->uio_offset - offset,
217 HAMMER_BUFSIZE, &bp);
222 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
223 n = HAMMER_BUFSIZE - offset;
224 if (n > uio->uio_resid)
226 if (n > ip->ino_rec.ino_size - uio->uio_offset)
227 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
228 error = uiomove((char *)bp->b_data + offset, n, uio);
233 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
234 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
235 ip->ino_rec.ino_atime = trans.tid;
236 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
240 hammer_commit_transaction(&trans);
245 * hammer_vop_write { vp, uio, ioflag, cred }
249 hammer_vop_write(struct vop_write_args *ap)
251 struct hammer_transaction trans;
252 struct hammer_inode *ip;
260 if (ap->a_vp->v_type != VREG)
265 if (ip->flags & HAMMER_INODE_RO)
269 * Create a transaction to cover the operations we perform.
271 hammer_start_transaction(&trans, ip->hmp);
277 if (ap->a_ioflag & IO_APPEND)
278 uio->uio_offset = ip->ino_rec.ino_size;
281 * Check for illegal write offsets. Valid range is 0...2^63-1
283 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
284 hammer_commit_transaction(&trans);
289 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
291 while (uio->uio_resid > 0) {
294 offset = uio->uio_offset & HAMMER_BUFMASK;
295 n = HAMMER_BUFSIZE - offset;
296 if (n > uio->uio_resid)
298 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
299 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
303 if (uio->uio_segflg == UIO_NOCOPY) {
305 * Issuing a write with the same data backing the
306 * buffer. Instantiate the buffer to collect the
307 * backing vm pages, then read-in any missing bits.
309 * This case is used by vop_stdputpages().
311 bp = getblk(ap->a_vp, uio->uio_offset - offset,
312 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
313 if ((bp->b_flags & B_CACHE) == 0) {
315 error = bread(ap->a_vp,
316 uio->uio_offset - offset,
317 HAMMER_BUFSIZE, &bp);
319 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
321 * entirely overwrite the buffer
323 bp = getblk(ap->a_vp, uio->uio_offset - offset,
324 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
325 } else if (offset == 0 && uio->uio_offset >= ip->ino_rec.ino_size) {
329 bp = getblk(ap->a_vp, uio->uio_offset - offset,
330 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
334 * Partial overwrite, read in any missing bits then
335 * replace the portion being written.
337 error = bread(ap->a_vp, uio->uio_offset - offset,
338 HAMMER_BUFSIZE, &bp);
343 error = uiomove((char *)bp->b_data + offset, n, uio);
346 * If we screwed up we have to undo any VM size changes we
352 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
357 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
358 if (ip->ino_rec.ino_size < uio->uio_offset) {
359 ip->ino_rec.ino_size = uio->uio_offset;
360 flags = HAMMER_INODE_RDIRTY;
361 vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
365 ip->ino_rec.ino_mtime = trans.tid;
366 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
367 hammer_modify_inode(&trans, ip, flags);
371 * The file write must be tagged with the same TID as the
372 * inode, for consistency in case the inode changed size.
373 * This guarantees the on-disk data records will have a
374 * TID <= the inode TID representing the size change.
376 * If a prior write has not yet flushed, retain its TID.
379 bp->b_tid = ip->last_tid;
382 * For now we can't use ip->last_tid because we may wind
383 * up trying to flush the same buffer with the same TID
384 * (but different data) multiple times, which will cause
388 bp->b_tid = trans.tid;
390 if (ap->a_ioflag & IO_SYNC) {
392 } else if (ap->a_ioflag & IO_DIRECT) {
394 } else if ((ap->a_ioflag >> 16) > 1 &&
395 (uio->uio_offset & HAMMER_BUFMASK) == 0) {
397 * If seqcount indicates sequential operation and
398 * we just finished filling a buffer, push it out
399 * now to prevent the buffer cache from becoming
400 * too full, which would trigger non-optimal
409 hammer_abort_transaction(&trans);
411 hammer_commit_transaction(&trans);
416 * hammer_vop_access { vp, mode, cred }
420 hammer_vop_access(struct vop_access_args *ap)
422 struct hammer_inode *ip = VTOI(ap->a_vp);
427 uid = hammer_to_unix_xid(&ip->ino_data.uid);
428 gid = hammer_to_unix_xid(&ip->ino_data.gid);
430 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
431 ip->ino_data.uflags);
436 * hammer_vop_advlock { vp, id, op, fl, flags }
440 hammer_vop_advlock(struct vop_advlock_args *ap)
442 struct hammer_inode *ip = VTOI(ap->a_vp);
444 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
448 * hammer_vop_close { vp, fflag }
452 hammer_vop_close(struct vop_close_args *ap)
454 return (vop_stdclose(ap));
458 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
460 * The operating system has already ensured that the directory entry
461 * does not exist and done all appropriate namespace locking.
465 hammer_vop_ncreate(struct vop_ncreate_args *ap)
467 struct hammer_transaction trans;
468 struct hammer_inode *dip;
469 struct hammer_inode *nip;
470 struct nchandle *nch;
474 dip = VTOI(ap->a_dvp);
476 if (dip->flags & HAMMER_INODE_RO)
480 * Create a transaction to cover the operations we perform.
482 hammer_start_transaction(&trans, dip->hmp);
485 * Create a new filesystem object of the requested type. The
486 * returned inode will be referenced but not locked.
489 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
491 kprintf("hammer_create_inode error %d\n", error);
493 hammer_abort_transaction(&trans);
499 * Add the new filesystem object to the directory. This will also
500 * bump the inode's link count.
502 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
504 kprintf("hammer_ip_add_directory error %d\n", error);
510 hammer_rel_inode(nip, 0);
511 hammer_abort_transaction(&trans);
514 hammer_commit_transaction(&trans);
515 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
516 hammer_rel_inode(nip, 0);
518 cache_setunresolved(ap->a_nch);
519 cache_setvp(ap->a_nch, *ap->a_vpp);
526 * hammer_vop_getattr { vp, vap }
530 hammer_vop_getattr(struct vop_getattr_args *ap)
532 struct hammer_inode *ip = VTOI(ap->a_vp);
533 struct vattr *vap = ap->a_vap;
536 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
537 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
542 hammer_itimes(ap->a_vp);
545 vap->va_fsid = ip->hmp->fsid_udev;
546 vap->va_fileid = ip->ino_rec.base.base.obj_id;
547 vap->va_mode = ip->ino_data.mode;
548 vap->va_nlink = ip->ino_rec.ino_nlinks;
549 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
550 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
553 vap->va_size = ip->ino_rec.ino_size;
554 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
555 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
556 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
557 vap->va_flags = ip->ino_data.uflags;
558 vap->va_gen = 1; /* hammer inums are unique for all time */
559 vap->va_blocksize = HAMMER_BUFSIZE;
560 vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
561 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
562 vap->va_filerev = 0; /* XXX */
563 /* mtime uniquely identifies any adjustments made to the file */
564 vap->va_fsmid = ip->ino_rec.ino_mtime;
565 vap->va_uid_uuid = ip->ino_data.uid;
566 vap->va_gid_uuid = ip->ino_data.gid;
567 vap->va_fsid_uuid = ip->hmp->fsid;
568 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
571 switch (ip->ino_rec.base.base.obj_type) {
572 case HAMMER_OBJTYPE_CDEV:
573 case HAMMER_OBJTYPE_BDEV:
574 vap->va_rmajor = ip->ino_data.rmajor;
575 vap->va_rminor = ip->ino_data.rminor;
585 * hammer_vop_nresolve { nch, dvp, cred }
587 * Locate the requested directory entry.
591 hammer_vop_nresolve(struct vop_nresolve_args *ap)
593 struct hammer_transaction trans;
594 struct namecache *ncp;
598 struct hammer_cursor cursor;
599 union hammer_record_ondisk *rec;
609 * Misc initialization, plus handle as-of name extensions. Look for
610 * the '@@' extension. Note that as-of files and directories cannot
613 dip = VTOI(ap->a_dvp);
614 ncp = ap->a_nch->ncp;
615 asof = dip->obj_asof;
619 hammer_simple_transaction(&trans, dip->hmp);
621 for (i = 0; i < nlen; ++i) {
622 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
623 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
624 flags |= HAMMER_INODE_RO;
631 * If there is no path component the time extension is relative to
635 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
636 asof, flags, &error);
638 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
639 hammer_rel_inode(ip, 0);
645 cache_setvp(ap->a_nch, vp);
652 * Calculate the namekey and setup the key range for the scan. This
653 * works kinda like a chained hash table where the lower 32 bits
654 * of the namekey synthesize the chain.
656 * The key range is inclusive of both key_beg and key_end.
658 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
660 error = hammer_init_cursor(&trans, &cursor, &dip->cache[0]);
661 cursor.key_beg.obj_id = dip->obj_id;
662 cursor.key_beg.key = namekey;
663 cursor.key_beg.create_tid = 0;
664 cursor.key_beg.delete_tid = 0;
665 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
666 cursor.key_beg.obj_type = 0;
668 cursor.key_end = cursor.key_beg;
669 cursor.key_end.key |= 0xFFFFFFFFULL;
671 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
674 * Scan all matching records (the chain), locate the one matching
675 * the requested path component.
677 * The hammer_ip_*() functions merge in-memory records with on-disk
678 * records for the purposes of the search.
681 error = hammer_ip_first(&cursor, dip);
687 error = hammer_ip_resolve_data(&cursor);
691 if (nlen == rec->entry.base.data_len &&
692 bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
693 obj_id = rec->entry.obj_id;
696 error = hammer_ip_next(&cursor);
698 hammer_done_cursor(&cursor);
700 ip = hammer_get_inode(&trans, &dip->cache[1],
701 obj_id, asof, flags, &error);
703 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
704 hammer_rel_inode(ip, 0);
710 cache_setvp(ap->a_nch, vp);
713 } else if (error == ENOENT) {
714 cache_setvp(ap->a_nch, NULL);
717 hammer_commit_transaction(&trans);
722 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
724 * Locate the parent directory of a directory vnode.
726 * dvp is referenced but not locked. *vpp must be returned referenced and
727 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
728 * at the root, instead it could indicate that the directory we were in was
731 * NOTE: as-of sequences are not linked into the directory structure. If
732 * we are at the root with a different asof then the mount point, reload
733 * the same directory with the mount point's asof. I'm not sure what this
734 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
735 * get confused, but it hasn't been tested.
739 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
741 struct hammer_transaction trans;
742 struct hammer_inode *dip;
743 struct hammer_inode *ip;
744 int64_t parent_obj_id;
748 dip = VTOI(ap->a_dvp);
749 asof = dip->obj_asof;
750 parent_obj_id = dip->ino_data.parent_obj_id;
752 if (parent_obj_id == 0) {
753 if (dip->obj_id == HAMMER_OBJID_ROOT &&
754 asof != dip->hmp->asof) {
755 parent_obj_id = dip->obj_id;
756 asof = dip->hmp->asof;
757 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
758 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
766 hammer_simple_transaction(&trans, dip->hmp);
768 ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
769 asof, dip->flags, &error);
771 error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
772 hammer_rel_inode(ip, 0);
776 hammer_commit_transaction(&trans);
781 * hammer_vop_nlink { nch, dvp, vp, cred }
785 hammer_vop_nlink(struct vop_nlink_args *ap)
787 struct hammer_transaction trans;
788 struct hammer_inode *dip;
789 struct hammer_inode *ip;
790 struct nchandle *nch;
794 dip = VTOI(ap->a_dvp);
797 if (dip->flags & HAMMER_INODE_RO)
799 if (ip->flags & HAMMER_INODE_RO)
803 * Create a transaction to cover the operations we perform.
805 hammer_start_transaction(&trans, dip->hmp);
808 * Add the filesystem object to the directory. Note that neither
809 * dip nor ip are referenced or locked, but their vnodes are
810 * referenced. This function will bump the inode's link count.
812 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
818 hammer_abort_transaction(&trans);
820 cache_setunresolved(nch);
821 cache_setvp(nch, ap->a_vp);
822 hammer_commit_transaction(&trans);
828 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
830 * The operating system has already ensured that the directory entry
831 * does not exist and done all appropriate namespace locking.
835 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
837 struct hammer_transaction trans;
838 struct hammer_inode *dip;
839 struct hammer_inode *nip;
840 struct nchandle *nch;
844 dip = VTOI(ap->a_dvp);
846 if (dip->flags & HAMMER_INODE_RO)
850 * Create a transaction to cover the operations we perform.
852 hammer_start_transaction(&trans, dip->hmp);
855 * Create a new filesystem object of the requested type. The
856 * returned inode will be referenced but not locked.
858 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
860 kprintf("hammer_mkdir error %d\n", error);
862 hammer_abort_transaction(&trans);
868 * Add the new filesystem object to the directory. This will also
869 * bump the inode's link count.
871 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
873 kprintf("hammer_mkdir (add) error %d\n", error);
879 hammer_rel_inode(nip, 0);
880 hammer_abort_transaction(&trans);
883 hammer_commit_transaction(&trans);
884 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
885 hammer_rel_inode(nip, 0);
887 cache_setunresolved(ap->a_nch);
888 cache_setvp(ap->a_nch, *ap->a_vpp);
895 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
897 * The operating system has already ensured that the directory entry
898 * does not exist and done all appropriate namespace locking.
902 hammer_vop_nmknod(struct vop_nmknod_args *ap)
904 struct hammer_transaction trans;
905 struct hammer_inode *dip;
906 struct hammer_inode *nip;
907 struct nchandle *nch;
911 dip = VTOI(ap->a_dvp);
913 if (dip->flags & HAMMER_INODE_RO)
917 * Create a transaction to cover the operations we perform.
919 hammer_start_transaction(&trans, dip->hmp);
922 * Create a new filesystem object of the requested type. The
923 * returned inode will be referenced but not locked.
925 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
927 hammer_abort_transaction(&trans);
933 * Add the new filesystem object to the directory. This will also
934 * bump the inode's link count.
936 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
942 hammer_rel_inode(nip, 0);
943 hammer_abort_transaction(&trans);
946 hammer_commit_transaction(&trans);
947 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
948 hammer_rel_inode(nip, 0);
950 cache_setunresolved(ap->a_nch);
951 cache_setvp(ap->a_nch, *ap->a_vpp);
958 * hammer_vop_open { vp, mode, cred, fp }
962 hammer_vop_open(struct vop_open_args *ap)
964 if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
967 return(vop_stdopen(ap));
971 * hammer_vop_pathconf { vp, name, retval }
975 hammer_vop_pathconf(struct vop_pathconf_args *ap)
981 * hammer_vop_print { vp }
985 hammer_vop_print(struct vop_print_args *ap)
991 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
995 hammer_vop_readdir(struct vop_readdir_args *ap)
997 struct hammer_transaction trans;
998 struct hammer_cursor cursor;
999 struct hammer_inode *ip;
1001 hammer_record_ondisk_t rec;
1002 hammer_base_elm_t base;
1010 ip = VTOI(ap->a_vp);
1012 saveoff = uio->uio_offset;
1014 if (ap->a_ncookies) {
1015 ncookies = uio->uio_resid / 16 + 1;
1016 if (ncookies > 1024)
1018 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1026 hammer_simple_transaction(&trans, ip->hmp);
1029 * Handle artificial entries
1033 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1037 cookies[cookie_index] = saveoff;
1040 if (cookie_index == ncookies)
1044 if (ip->ino_data.parent_obj_id) {
1045 r = vop_write_dirent(&error, uio,
1046 ip->ino_data.parent_obj_id,
1049 r = vop_write_dirent(&error, uio,
1050 ip->obj_id, DT_DIR, 2, "..");
1055 cookies[cookie_index] = saveoff;
1058 if (cookie_index == ncookies)
1063 * Key range (begin and end inclusive) to scan. Directory keys
1064 * directly translate to a 64 bit 'seek' position.
1066 hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1067 cursor.key_beg.obj_id = ip->obj_id;
1068 cursor.key_beg.create_tid = 0;
1069 cursor.key_beg.delete_tid = 0;
1070 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1071 cursor.key_beg.obj_type = 0;
1072 cursor.key_beg.key = saveoff;
1074 cursor.key_end = cursor.key_beg;
1075 cursor.key_end.key = HAMMER_MAX_KEY;
1076 cursor.asof = ip->obj_asof;
1077 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1079 error = hammer_ip_first(&cursor, ip);
1081 while (error == 0) {
1082 error = hammer_ip_resolve_record_and_data(&cursor);
1085 rec = cursor.record;
1086 base = &rec->base.base;
1087 saveoff = base->key;
1089 if (base->obj_id != ip->obj_id)
1090 panic("readdir: bad record at %p", cursor.node);
1092 r = vop_write_dirent(
1093 &error, uio, rec->entry.obj_id,
1094 hammer_get_dtype(rec->entry.base.base.obj_type),
1095 rec->entry.base.data_len,
1096 (void *)cursor.data);
1101 cookies[cookie_index] = base->key;
1103 if (cookie_index == ncookies)
1105 error = hammer_ip_next(&cursor);
1107 hammer_done_cursor(&cursor);
1110 hammer_commit_transaction(&trans);
1113 *ap->a_eofflag = (error == ENOENT);
1114 uio->uio_offset = saveoff;
1115 if (error && cookie_index == 0) {
1116 if (error == ENOENT)
1119 kfree(cookies, M_TEMP);
1120 *ap->a_ncookies = 0;
1121 *ap->a_cookies = NULL;
1124 if (error == ENOENT)
1127 *ap->a_ncookies = cookie_index;
1128 *ap->a_cookies = cookies;
1135 * hammer_vop_readlink { vp, uio, cred }
1139 hammer_vop_readlink(struct vop_readlink_args *ap)
1141 struct hammer_transaction trans;
1142 struct hammer_cursor cursor;
1143 struct hammer_inode *ip;
1146 ip = VTOI(ap->a_vp);
1148 hammer_simple_transaction(&trans, ip->hmp);
1150 hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1153 * Key range (begin and end inclusive) to scan. Directory keys
1154 * directly translate to a 64 bit 'seek' position.
1156 cursor.key_beg.obj_id = ip->obj_id;
1157 cursor.key_beg.create_tid = 0;
1158 cursor.key_beg.delete_tid = 0;
1159 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1160 cursor.key_beg.obj_type = 0;
1161 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1162 cursor.asof = ip->obj_asof;
1163 cursor.flags |= HAMMER_CURSOR_ASOF;
1165 error = hammer_ip_lookup(&cursor, ip);
1167 error = hammer_ip_resolve_data(&cursor);
1169 error = uiomove((char *)cursor.data,
1170 cursor.record->base.data_len,
1174 hammer_done_cursor(&cursor);
1175 hammer_commit_transaction(&trans);
1180 * hammer_vop_nremove { nch, dvp, cred }
1184 hammer_vop_nremove(struct vop_nremove_args *ap)
1186 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1190 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1194 hammer_vop_nrename(struct vop_nrename_args *ap)
1196 struct hammer_transaction trans;
1197 struct namecache *fncp;
1198 struct namecache *tncp;
1199 struct hammer_inode *fdip;
1200 struct hammer_inode *tdip;
1201 struct hammer_inode *ip;
1202 struct hammer_cursor cursor;
1203 union hammer_record_ondisk *rec;
1207 fdip = VTOI(ap->a_fdvp);
1208 tdip = VTOI(ap->a_tdvp);
1209 fncp = ap->a_fnch->ncp;
1210 tncp = ap->a_tnch->ncp;
1211 ip = VTOI(fncp->nc_vp);
1212 KKASSERT(ip != NULL);
1214 if (fdip->flags & HAMMER_INODE_RO)
1216 if (tdip->flags & HAMMER_INODE_RO)
1218 if (ip->flags & HAMMER_INODE_RO)
1221 hammer_start_transaction(&trans, fdip->hmp);
1224 * Remove tncp from the target directory and then link ip as
1225 * tncp. XXX pass trans to dounlink
1227 * Force the inode sync-time to match the transaction so it is
1228 * in-sync with the creation of the target directory entry.
1230 error = hammer_dounlink(ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1231 if (error == 0 || error == ENOENT) {
1232 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1234 ip->ino_data.parent_obj_id = tdip->obj_id;
1235 hammer_modify_inode(&trans, ip,
1236 HAMMER_INODE_DDIRTY | HAMMER_INODE_TIDLOCKED);
1240 goto failed; /* XXX */
1243 * Locate the record in the originating directory and remove it.
1245 * Calculate the namekey and setup the key range for the scan. This
1246 * works kinda like a chained hash table where the lower 32 bits
1247 * of the namekey synthesize the chain.
1249 * The key range is inclusive of both key_beg and key_end.
1251 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1253 hammer_init_cursor(&trans, &cursor, &fdip->cache[0]);
1254 cursor.key_beg.obj_id = fdip->obj_id;
1255 cursor.key_beg.key = namekey;
1256 cursor.key_beg.create_tid = 0;
1257 cursor.key_beg.delete_tid = 0;
1258 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1259 cursor.key_beg.obj_type = 0;
1261 cursor.key_end = cursor.key_beg;
1262 cursor.key_end.key |= 0xFFFFFFFFULL;
1263 cursor.asof = fdip->obj_asof;
1264 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1267 * Scan all matching records (the chain), locate the one matching
1268 * the requested path component.
1270 * The hammer_ip_*() functions merge in-memory records with on-disk
1271 * records for the purposes of the search.
1273 error = hammer_ip_first(&cursor, fdip);
1274 while (error == 0) {
1275 if (hammer_ip_resolve_data(&cursor) != 0)
1277 rec = cursor.record;
1278 if (fncp->nc_nlen == rec->entry.base.data_len &&
1279 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1282 error = hammer_ip_next(&cursor);
1286 * If all is ok we have to get the inode so we can adjust nlinks.
1288 * WARNING: hammer_ip_del_directory() may have to terminate the
1289 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1293 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1294 hammer_done_cursor(&cursor);
1296 cache_rename(ap->a_fnch, ap->a_tnch);
1297 if (error == EDEADLK)
1301 hammer_commit_transaction(&trans);
1303 hammer_abort_transaction(&trans);
1309 * hammer_vop_nrmdir { nch, dvp, cred }
1313 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1315 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, 0));
1319 * hammer_vop_setattr { vp, vap, cred }
1323 hammer_vop_setattr(struct vop_setattr_args *ap)
1325 struct hammer_transaction trans;
1327 struct hammer_inode *ip;
1331 int64_t aligned_size;
1336 ip = ap->a_vp->v_data;
1339 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1341 if (ip->flags & HAMMER_INODE_RO)
1344 hammer_start_transaction(&trans, ip->hmp);
1347 if (vap->va_flags != VNOVAL) {
1348 flags = ip->ino_data.uflags;
1349 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1350 hammer_to_unix_xid(&ip->ino_data.uid),
1353 if (ip->ino_data.uflags != flags) {
1354 ip->ino_data.uflags = flags;
1355 modflags |= HAMMER_INODE_DDIRTY;
1357 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1364 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1368 if (vap->va_uid != (uid_t)VNOVAL) {
1369 hammer_guid_to_uuid(&uuid, vap->va_uid);
1370 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1371 ip->ino_data.uid = uuid;
1372 modflags |= HAMMER_INODE_DDIRTY;
1375 if (vap->va_gid != (uid_t)VNOVAL) {
1376 hammer_guid_to_uuid(&uuid, vap->va_gid);
1377 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1378 ip->ino_data.gid = uuid;
1379 modflags |= HAMMER_INODE_DDIRTY;
1382 while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1383 switch(ap->a_vp->v_type) {
1385 if (vap->va_size == ip->ino_rec.ino_size)
1387 if (vap->va_size < ip->ino_rec.ino_size) {
1388 vtruncbuf(ap->a_vp, vap->va_size,
1392 vnode_pager_setsize(ap->a_vp, vap->va_size);
1395 ip->ino_rec.ino_size = vap->va_size;
1396 modflags |= HAMMER_INODE_RDIRTY;
1397 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1398 ~(int64_t)HAMMER_BUFMASK;
1401 error = hammer_ip_delete_range(&trans, ip,
1403 0x7FFFFFFFFFFFFFFFLL);
1406 * If truncating we have to clean out a portion of
1407 * the last block on-disk.
1409 if (truncating && error == 0 &&
1410 vap->va_size < aligned_size) {
1414 offset = vap->va_size & HAMMER_BUFMASK;
1415 error = bread(ap->a_vp,
1416 aligned_size - HAMMER_BUFSIZE,
1417 HAMMER_BUFSIZE, &bp);
1419 bzero(bp->b_data + offset,
1420 HAMMER_BUFSIZE - offset);
1422 bp->b_tid = trans.tid;
1430 error = hammer_ip_delete_range(&trans, ip,
1432 0x7FFFFFFFFFFFFFFFLL);
1433 ip->ino_rec.ino_size = vap->va_size;
1434 modflags |= HAMMER_INODE_RDIRTY;
1442 if (vap->va_atime.tv_sec != VNOVAL) {
1443 ip->ino_rec.ino_atime =
1444 hammer_timespec_to_transid(&vap->va_atime);
1445 modflags |= HAMMER_INODE_ITIMES;
1447 if (vap->va_mtime.tv_sec != VNOVAL) {
1448 ip->ino_rec.ino_mtime =
1449 hammer_timespec_to_transid(&vap->va_mtime);
1450 modflags |= HAMMER_INODE_ITIMES;
1452 if (vap->va_mode != (mode_t)VNOVAL) {
1453 if (ip->ino_data.mode != vap->va_mode) {
1454 ip->ino_data.mode = vap->va_mode;
1455 modflags |= HAMMER_INODE_DDIRTY;
1460 hammer_abort_transaction(&trans);
1462 hammer_modify_inode(&trans, ip, modflags);
1463 hammer_commit_transaction(&trans);
1469 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1473 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1475 struct hammer_transaction trans;
1476 struct hammer_inode *dip;
1477 struct hammer_inode *nip;
1478 struct nchandle *nch;
1479 hammer_record_t record;
1483 ap->a_vap->va_type = VLNK;
1486 dip = VTOI(ap->a_dvp);
1488 if (dip->flags & HAMMER_INODE_RO)
1492 * Create a transaction to cover the operations we perform.
1494 hammer_start_transaction(&trans, dip->hmp);
1497 * Create a new filesystem object of the requested type. The
1498 * returned inode will be referenced but not locked.
1501 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1503 hammer_abort_transaction(&trans);
1509 * Add the new filesystem object to the directory. This will also
1510 * bump the inode's link count.
1512 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1515 * Add a record representing the symlink. symlink stores the link
1516 * as pure data, not a string, and is no \0 terminated.
1519 record = hammer_alloc_mem_record(nip);
1520 bytes = strlen(ap->a_target);
1522 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1523 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1524 record->rec.base.data_len = bytes;
1525 record->data = (void *)ap->a_target;
1526 /* will be reallocated by routine below */
1527 error = hammer_ip_add_record(&trans, record);
1530 * Set the file size to the length of the link.
1533 nip->ino_rec.ino_size = bytes;
1534 hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1542 hammer_rel_inode(nip, 0);
1543 hammer_abort_transaction(&trans);
1546 hammer_commit_transaction(&trans);
1547 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1548 hammer_rel_inode(nip, 0);
1550 cache_setunresolved(ap->a_nch);
1551 cache_setvp(ap->a_nch, *ap->a_vpp);
1558 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1562 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1564 return(hammer_dounlink(ap->a_nch, ap->a_dvp, ap->a_cred, ap->a_flags));
1568 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1572 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1574 struct hammer_inode *ip = ap->a_vp->v_data;
1576 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1577 ap->a_fflag, ap->a_cred));
1582 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1587 mp = ap->a_head.a_ops->head.vv_mount;
1590 case MOUNTCTL_SET_EXPORT:
1591 if (ap->a_ctllen != sizeof(struct export_args))
1593 error = hammer_vfs_export(mp, ap->a_op,
1594 (const struct export_args *)ap->a_ctl);
1597 error = journal_mountctl(ap);
1604 * hammer_vop_strategy { vp, bio }
1606 * Strategy call, used for regular file read & write only. Note that the
1607 * bp may represent a cluster.
1609 * To simplify operation and allow better optimizations in the future,
1610 * this code does not make any assumptions with regards to buffer alignment
1615 hammer_vop_strategy(struct vop_strategy_args *ap)
1620 bp = ap->a_bio->bio_buf;
1624 error = hammer_vop_strategy_read(ap);
1627 error = hammer_vop_strategy_write(ap);
1633 bp->b_error = error;
1635 bp->b_flags |= B_ERROR;
1641 * Read from a regular file. Iterate the related records and fill in the
1642 * BIO/BUF. Gaps are zero-filled.
1644 * The support code in hammer_object.c should be used to deal with mixed
1645 * in-memory and on-disk records.
1651 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1653 struct hammer_transaction trans;
1654 struct hammer_inode *ip;
1655 struct hammer_cursor cursor;
1656 hammer_record_ondisk_t rec;
1657 hammer_base_elm_t base;
1670 ip = ap->a_vp->v_data;
1672 hammer_simple_transaction(&trans, ip->hmp);
1673 hammer_init_cursor(&trans, &cursor, &ip->cache[0]);
1676 * Key range (begin and end inclusive) to scan. Note that the key's
1677 * stored in the actual records represent BASE+LEN, not BASE. The
1678 * first record containing bio_offset will have a key > bio_offset.
1680 cursor.key_beg.obj_id = ip->obj_id;
1681 cursor.key_beg.create_tid = 0;
1682 cursor.key_beg.delete_tid = 0;
1683 cursor.key_beg.obj_type = 0;
1684 cursor.key_beg.key = bio->bio_offset + 1;
1685 cursor.asof = ip->obj_asof;
1686 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1688 cursor.key_end = cursor.key_beg;
1689 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1690 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1691 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1692 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1694 ran_end = bio->bio_offset + bp->b_bufsize;
1695 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1696 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1697 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1698 if (tmp64 < ran_end)
1699 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1701 cursor.key_end.key = ran_end + MAXPHYS + 1;
1703 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1705 error = hammer_ip_first(&cursor, ip);
1708 while (error == 0) {
1709 error = hammer_ip_resolve_data(&cursor);
1712 rec = cursor.record;
1713 base = &rec->base.base;
1715 rec_offset = base->key - rec->data.base.data_len;
1718 * Calculate the gap, if any, and zero-fill it.
1720 n = (int)(rec_offset - (bio->bio_offset + boff));
1722 if (n > bp->b_bufsize - boff)
1723 n = bp->b_bufsize - boff;
1724 bzero((char *)bp->b_data + boff, n);
1730 * Calculate the data offset in the record and the number
1731 * of bytes we can copy.
1733 * Note there is a degenerate case here where boff may
1734 * already be at bp->b_bufsize.
1737 n = rec->data.base.data_len - roff;
1739 if (n > bp->b_bufsize - boff)
1740 n = bp->b_bufsize - boff;
1741 bcopy((char *)cursor.data + roff,
1742 (char *)bp->b_data + boff, n);
1744 if (boff == bp->b_bufsize)
1746 error = hammer_ip_next(&cursor);
1748 hammer_done_cursor(&cursor);
1749 hammer_commit_transaction(&trans);
1752 * There may have been a gap after the last record
1754 if (error == ENOENT)
1756 if (error == 0 && boff != bp->b_bufsize) {
1757 KKASSERT(boff < bp->b_bufsize);
1758 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1759 /* boff = bp->b_bufsize; */
1766 * Write to a regular file. Iterate the related records and mark for
1767 * deletion. If existing edge records (left and right side) overlap our
1768 * write they have to be marked deleted and new records created, usually
1769 * referencing a portion of the original data. Then add a record to
1770 * represent the buffer.
1772 * The support code in hammer_object.c should be used to deal with mixed
1773 * in-memory and on-disk records.
1777 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1779 struct hammer_transaction trans;
1787 ip = ap->a_vp->v_data;
1789 if (ip->flags & HAMMER_INODE_RO)
1793 * Start a transaction using the TID stored with the bp.
1795 KKASSERT(bp->b_tid != 0);
1796 hammer_start_transaction_tid(&trans, ip->hmp, bp->b_tid);
1799 * Delete any records overlapping our range. This function will
1800 * (eventually) properly truncate partial overlaps.
1802 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1803 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1806 error = hammer_ip_delete_range(&trans, ip, bio->bio_offset,
1812 * Add a single record to cover the write. We can write a record
1813 * with only the actual file data - for example, a small 200 byte
1814 * file does not have to write out a 16K record.
1816 * While the data size does not have to be aligned, we still do it
1817 * to reduce fragmentation in a future allocation model.
1822 if (ip->ino_rec.ino_size - bio->bio_offset > bp->b_bufsize) {
1823 limit_size = bp->b_bufsize;
1825 limit_size = (int)(ip->ino_rec.ino_size -
1827 KKASSERT(limit_size >= 0);
1828 limit_size = (limit_size + 63) & ~63;
1830 error = hammer_ip_sync_data(&trans, ip, bio->bio_offset,
1831 bp->b_data, limit_size);
1835 * If an error occured abort the transaction
1838 /* XXX undo deletion */
1839 hammer_abort_transaction(&trans);
1840 bp->b_resid = bp->b_bufsize;
1842 hammer_commit_transaction(&trans);
1850 * dounlink - disconnect a directory entry
1852 * XXX whiteout support not really in yet
1855 hammer_dounlink(struct nchandle *nch, struct vnode *dvp, struct ucred *cred,
1858 struct hammer_transaction trans;
1859 struct namecache *ncp;
1862 hammer_record_ondisk_t rec;
1863 struct hammer_cursor cursor;
1868 * Calculate the namekey and setup the key range for the scan. This
1869 * works kinda like a chained hash table where the lower 32 bits
1870 * of the namekey synthesize the chain.
1872 * The key range is inclusive of both key_beg and key_end.
1877 if (dip->flags & HAMMER_INODE_RO)
1880 hammer_start_transaction(&trans, dip->hmp);
1882 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1884 hammer_init_cursor(&trans, &cursor, &dip->cache[0]);
1885 cursor.key_beg.obj_id = dip->obj_id;
1886 cursor.key_beg.key = namekey;
1887 cursor.key_beg.create_tid = 0;
1888 cursor.key_beg.delete_tid = 0;
1889 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1890 cursor.key_beg.obj_type = 0;
1892 cursor.key_end = cursor.key_beg;
1893 cursor.key_end.key |= 0xFFFFFFFFULL;
1894 cursor.asof = dip->obj_asof;
1895 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1898 * Scan all matching records (the chain), locate the one matching
1899 * the requested path component. info->last_error contains the
1900 * error code on search termination and could be 0, ENOENT, or
1903 * The hammer_ip_*() functions merge in-memory records with on-disk
1904 * records for the purposes of the search.
1906 error = hammer_ip_first(&cursor, dip);
1907 while (error == 0) {
1908 error = hammer_ip_resolve_data(&cursor);
1911 rec = cursor.record;
1912 if (ncp->nc_nlen == rec->entry.base.data_len &&
1913 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
1916 error = hammer_ip_next(&cursor);
1920 * If all is ok we have to get the inode so we can adjust nlinks.
1922 * If the target is a directory, it must be empty.
1925 ip = hammer_get_inode(&trans, &dip->cache[1],
1927 dip->hmp->asof, 0, &error);
1928 if (error == ENOENT) {
1929 kprintf("obj_id %016llx\n", rec->entry.obj_id);
1930 Debugger("ENOENT unlinking object that should exist, cont to sync");
1931 hammer_sync_hmp(dip->hmp, MNT_NOWAIT);
1932 Debugger("ENOENT - sync done");
1934 if (error == 0 && ip->ino_rec.base.base.obj_type ==
1935 HAMMER_OBJTYPE_DIRECTORY) {
1936 error = hammer_ip_check_directory_empty(&trans, ip);
1939 * WARNING: hammer_ip_del_directory() may have to terminate
1940 * the cursor to avoid a lock recursion. It's ok to call
1941 * hammer_done_cursor() twice.
1944 error = hammer_ip_del_directory(&trans, &cursor, dip, ip);
1946 cache_setunresolved(nch);
1947 cache_setvp(nch, NULL);
1950 cache_inval_vp(ip->vp, CINV_DESTROY);
1952 hammer_rel_inode(ip, 0);
1954 hammer_done_cursor(&cursor);
1955 if (error == EDEADLK)
1959 hammer_commit_transaction(&trans);
1961 hammer_abort_transaction(&trans);
1965 /************************************************************************
1966 * FIFO AND SPECFS OPS *
1967 ************************************************************************
1972 hammer_vop_fifoclose (struct vop_close_args *ap)
1974 /* XXX update itimes */
1975 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
1979 hammer_vop_fiforead (struct vop_read_args *ap)
1983 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1984 /* XXX update access time */
1989 hammer_vop_fifowrite (struct vop_write_args *ap)
1993 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
1994 /* XXX update access time */
1999 hammer_vop_specclose (struct vop_close_args *ap)
2001 /* XXX update itimes */
2002 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2006 hammer_vop_specread (struct vop_read_args *ap)
2008 /* XXX update access time */
2009 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2013 hammer_vop_specwrite (struct vop_write_args *ap)
2015 /* XXX update last change time */
2016 return (VOCALL(&spec_vnode_vops, &ap->a_head));