2 * Copyright (c) 2007-2008 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 1.52 2008/05/12 21:17:18 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(hammer_transaction_t trans, struct nchandle *nch,
153 struct vnode *dvp, 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)
173 hammer_inode_t ip = VTOI(ap->a_vp);
175 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
176 vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
177 if (ap->a_waitfor == MNT_WAIT)
178 hammer_wait_inode(ip);
183 * hammer_vop_read { vp, uio, ioflag, cred }
187 hammer_vop_read(struct vop_read_args *ap)
189 struct hammer_transaction trans;
198 if (ap->a_vp->v_type != VREG)
202 seqcount = ap->a_ioflag >> 16;
204 hammer_start_transaction(&trans, ip->hmp);
207 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
210 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
211 offset = uio->uio_offset & HAMMER_BUFMASK;
213 error = cluster_read(ap->a_vp, ip->ino_data.size,
214 uio->uio_offset - offset, HAMMER_BUFSIZE,
215 MAXBSIZE, seqcount, &bp);
217 error = bread(ap->a_vp, uio->uio_offset - offset,
218 HAMMER_BUFSIZE, &bp);
223 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
224 n = HAMMER_BUFSIZE - offset;
225 if (n > uio->uio_resid)
227 if (n > ip->ino_data.size - uio->uio_offset)
228 n = (int)(ip->ino_data.size - uio->uio_offset);
229 error = uiomove((char *)bp->b_data + offset, n, uio);
236 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
237 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
238 ip->ino_leaf.atime = trans.time;
239 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
241 hammer_done_transaction(&trans);
246 * hammer_vop_write { vp, uio, ioflag, cred }
250 hammer_vop_write(struct vop_write_args *ap)
252 struct hammer_transaction trans;
253 struct hammer_inode *ip;
262 if (ap->a_vp->v_type != VREG)
267 if (ip->flags & HAMMER_INODE_RO)
271 * Create a transaction to cover the operations we perform.
273 hammer_start_transaction(&trans, ip->hmp);
279 if (ap->a_ioflag & IO_APPEND)
280 uio->uio_offset = ip->ino_data.size;
283 * Check for illegal write offsets. Valid range is 0...2^63-1
285 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
286 hammer_done_transaction(&trans);
291 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
294 while (uio->uio_resid > 0) {
298 * Do not allow huge writes to deadlock the buffer cache
300 if ((++count & 15) == 0) {
302 if ((ap->a_ioflag & IO_NOBWILL) == 0)
304 vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
307 offset = uio->uio_offset & HAMMER_BUFMASK;
308 n = HAMMER_BUFSIZE - offset;
309 if (n > uio->uio_resid)
311 if (uio->uio_offset + n > ip->ino_data.size) {
312 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
316 if (uio->uio_segflg == UIO_NOCOPY) {
318 * Issuing a write with the same data backing the
319 * buffer. Instantiate the buffer to collect the
320 * backing vm pages, then read-in any missing bits.
322 * This case is used by vop_stdputpages().
324 bp = getblk(ap->a_vp, uio->uio_offset - offset,
325 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
326 if ((bp->b_flags & B_CACHE) == 0) {
328 error = bread(ap->a_vp,
329 uio->uio_offset - offset,
330 HAMMER_BUFSIZE, &bp);
332 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
334 * Even though we are entirely overwriting the buffer
335 * we may still have to zero it out to avoid a
336 * mmap/write visibility issue.
338 bp = getblk(ap->a_vp, uio->uio_offset - offset,
339 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
340 if ((bp->b_flags & B_CACHE) == 0)
342 } else if (uio->uio_offset - offset >= ip->ino_data.size) {
344 * If the base offset of the buffer is beyond the
345 * file EOF, we don't have to issue a read.
347 bp = getblk(ap->a_vp, uio->uio_offset - offset,
348 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
352 * Partial overwrite, read in any missing bits then
353 * replace the portion being written.
355 error = bread(ap->a_vp, uio->uio_offset - offset,
356 HAMMER_BUFSIZE, &bp);
361 error = uiomove((char *)bp->b_data + offset, n, uio);
364 * If we screwed up we have to undo any VM size changes we
370 vtruncbuf(ap->a_vp, ip->ino_data.size,
375 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
376 if (ip->ino_data.size < uio->uio_offset) {
377 ip->ino_data.size = uio->uio_offset;
378 flags = HAMMER_INODE_DDIRTY;
379 vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
383 ip->ino_data.mtime = trans.time;
384 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
385 flags |= HAMMER_INODE_DDIRTY; /* XXX mtime */
386 hammer_modify_inode(&trans, ip, flags);
388 if (ap->a_ioflag & IO_SYNC) {
390 } else if (ap->a_ioflag & IO_DIRECT) {
393 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
394 (uio->uio_offset & HAMMER_BUFMASK) == 0) {
396 * XXX HAMMER can only fsync the whole inode,
397 * doing it on every buffer would be a bad idea.
400 * If seqcount indicates sequential operation and
401 * we just finished filling a buffer, push it out
402 * now to prevent the buffer cache from becoming
403 * too full, which would trigger non-optimal
412 hammer_done_transaction(&trans);
417 * hammer_vop_access { vp, mode, cred }
421 hammer_vop_access(struct vop_access_args *ap)
423 struct hammer_inode *ip = VTOI(ap->a_vp);
428 uid = hammer_to_unix_xid(&ip->ino_data.uid);
429 gid = hammer_to_unix_xid(&ip->ino_data.gid);
431 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
432 ip->ino_data.uflags);
437 * hammer_vop_advlock { vp, id, op, fl, flags }
441 hammer_vop_advlock(struct vop_advlock_args *ap)
443 struct hammer_inode *ip = VTOI(ap->a_vp);
445 return (lf_advlock(ap, &ip->advlock, ip->ino_data.size));
449 * hammer_vop_close { vp, fflag }
453 hammer_vop_close(struct vop_close_args *ap)
455 return (vop_stdclose(ap));
459 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
461 * The operating system has already ensured that the directory entry
462 * does not exist and done all appropriate namespace locking.
466 hammer_vop_ncreate(struct vop_ncreate_args *ap)
468 struct hammer_transaction trans;
469 struct hammer_inode *dip;
470 struct hammer_inode *nip;
471 struct nchandle *nch;
475 dip = VTOI(ap->a_dvp);
477 if (dip->flags & HAMMER_INODE_RO)
481 * Create a transaction to cover the operations we perform.
483 hammer_start_transaction(&trans, dip->hmp);
486 * Create a new filesystem object of the requested type. The
487 * returned inode will be referenced and shared-locked to prevent
488 * it from being moved to the flusher.
491 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
493 hkprintf("hammer_create_inode error %d\n", error);
494 hammer_done_transaction(&trans);
500 * Add the new filesystem object to the directory. This will also
501 * bump the inode's link count.
503 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
505 hkprintf("hammer_ip_add_directory error %d\n", error);
511 hammer_rel_inode(nip, 0);
512 hammer_done_transaction(&trans);
515 error = hammer_get_vnode(nip, ap->a_vpp);
516 hammer_done_transaction(&trans);
517 hammer_rel_inode(nip, 0);
519 cache_setunresolved(ap->a_nch);
520 cache_setvp(ap->a_nch, *ap->a_vpp);
527 * hammer_vop_getattr { vp, vap }
531 hammer_vop_getattr(struct vop_getattr_args *ap)
533 struct hammer_inode *ip = VTOI(ap->a_vp);
534 struct vattr *vap = ap->a_vap;
537 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
538 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
543 hammer_itimes(ap->a_vp);
546 vap->va_fsid = ip->hmp->fsid_udev;
547 vap->va_fileid = ip->ino_leaf.base.obj_id;
548 vap->va_mode = ip->ino_data.mode;
549 vap->va_nlink = ip->ino_data.nlinks;
550 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
551 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
554 vap->va_size = ip->ino_data.size;
555 hammer_to_timespec(ip->ino_leaf.atime, &vap->va_atime);
556 hammer_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
557 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
558 vap->va_flags = ip->ino_data.uflags;
559 vap->va_gen = 1; /* hammer inums are unique for all time */
560 vap->va_blocksize = HAMMER_BUFSIZE;
561 vap->va_bytes = (ip->ino_data.size + 63) & ~63;
562 vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
563 vap->va_filerev = 0; /* XXX */
564 /* mtime uniquely identifies any adjustments made to the file */
565 vap->va_fsmid = ip->ino_data.mtime;
566 vap->va_uid_uuid = ip->ino_data.uid;
567 vap->va_gid_uuid = ip->ino_data.gid;
568 vap->va_fsid_uuid = ip->hmp->fsid;
569 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
572 switch (ip->ino_data.obj_type) {
573 case HAMMER_OBJTYPE_CDEV:
574 case HAMMER_OBJTYPE_BDEV:
575 vap->va_rmajor = ip->ino_data.rmajor;
576 vap->va_rminor = ip->ino_data.rminor;
586 * hammer_vop_nresolve { nch, dvp, cred }
588 * Locate the requested directory entry.
592 hammer_vop_nresolve(struct vop_nresolve_args *ap)
594 struct hammer_transaction trans;
595 struct namecache *ncp;
599 struct hammer_cursor cursor;
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, &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], dip);
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.
683 error = hammer_ip_first(&cursor);
685 error = hammer_ip_resolve_data(&cursor);
688 if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
689 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
690 obj_id = cursor.data->entry.obj_id;
693 error = hammer_ip_next(&cursor);
696 hammer_done_cursor(&cursor);
698 ip = hammer_get_inode(&trans, &dip->cache[1],
699 obj_id, asof, flags, &error);
701 error = hammer_get_vnode(ip, &vp);
702 hammer_rel_inode(ip, 0);
708 cache_setvp(ap->a_nch, vp);
711 } else if (error == ENOENT) {
712 cache_setvp(ap->a_nch, NULL);
715 hammer_done_transaction(&trans);
720 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
722 * Locate the parent directory of a directory vnode.
724 * dvp is referenced but not locked. *vpp must be returned referenced and
725 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
726 * at the root, instead it could indicate that the directory we were in was
729 * NOTE: as-of sequences are not linked into the directory structure. If
730 * we are at the root with a different asof then the mount point, reload
731 * the same directory with the mount point's asof. I'm not sure what this
732 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
733 * get confused, but it hasn't been tested.
737 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
739 struct hammer_transaction trans;
740 struct hammer_inode *dip;
741 struct hammer_inode *ip;
742 int64_t parent_obj_id;
746 dip = VTOI(ap->a_dvp);
747 asof = dip->obj_asof;
748 parent_obj_id = dip->ino_data.parent_obj_id;
750 if (parent_obj_id == 0) {
751 if (dip->obj_id == HAMMER_OBJID_ROOT &&
752 asof != dip->hmp->asof) {
753 parent_obj_id = dip->obj_id;
754 asof = dip->hmp->asof;
755 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
756 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
764 hammer_simple_transaction(&trans, dip->hmp);
766 ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
767 asof, dip->flags, &error);
769 error = hammer_get_vnode(ip, ap->a_vpp);
770 hammer_rel_inode(ip, 0);
774 hammer_done_transaction(&trans);
779 * hammer_vop_nlink { nch, dvp, vp, cred }
783 hammer_vop_nlink(struct vop_nlink_args *ap)
785 struct hammer_transaction trans;
786 struct hammer_inode *dip;
787 struct hammer_inode *ip;
788 struct nchandle *nch;
792 dip = VTOI(ap->a_dvp);
795 if (dip->flags & HAMMER_INODE_RO)
797 if (ip->flags & HAMMER_INODE_RO)
801 * Create a transaction to cover the operations we perform.
803 hammer_start_transaction(&trans, dip->hmp);
806 * Add the filesystem object to the directory. Note that neither
807 * dip nor ip are referenced or locked, but their vnodes are
808 * referenced. This function will bump the inode's link count.
810 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
816 cache_setunresolved(nch);
817 cache_setvp(nch, ap->a_vp);
819 hammer_done_transaction(&trans);
824 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
826 * The operating system has already ensured that the directory entry
827 * does not exist and done all appropriate namespace locking.
831 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
833 struct hammer_transaction trans;
834 struct hammer_inode *dip;
835 struct hammer_inode *nip;
836 struct nchandle *nch;
840 dip = VTOI(ap->a_dvp);
842 if (dip->flags & HAMMER_INODE_RO)
846 * Create a transaction to cover the operations we perform.
848 hammer_start_transaction(&trans, dip->hmp);
851 * Create a new filesystem object of the requested type. The
852 * returned inode will be referenced but not locked.
854 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
856 hkprintf("hammer_mkdir error %d\n", error);
857 hammer_done_transaction(&trans);
862 * Add the new filesystem object to the directory. This will also
863 * bump the inode's link count.
865 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
867 hkprintf("hammer_mkdir (add) error %d\n", error);
873 hammer_rel_inode(nip, 0);
876 error = hammer_get_vnode(nip, ap->a_vpp);
877 hammer_rel_inode(nip, 0);
879 cache_setunresolved(ap->a_nch);
880 cache_setvp(ap->a_nch, *ap->a_vpp);
883 hammer_done_transaction(&trans);
888 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
890 * The operating system has already ensured that the directory entry
891 * does not exist and done all appropriate namespace locking.
895 hammer_vop_nmknod(struct vop_nmknod_args *ap)
897 struct hammer_transaction trans;
898 struct hammer_inode *dip;
899 struct hammer_inode *nip;
900 struct nchandle *nch;
904 dip = VTOI(ap->a_dvp);
906 if (dip->flags & HAMMER_INODE_RO)
910 * Create a transaction to cover the operations we perform.
912 hammer_start_transaction(&trans, dip->hmp);
915 * Create a new filesystem object of the requested type. The
916 * returned inode will be referenced but not locked.
918 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
920 hammer_done_transaction(&trans);
926 * Add the new filesystem object to the directory. This will also
927 * bump the inode's link count.
929 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
935 hammer_rel_inode(nip, 0);
938 error = hammer_get_vnode(nip, ap->a_vpp);
939 hammer_rel_inode(nip, 0);
941 cache_setunresolved(ap->a_nch);
942 cache_setvp(ap->a_nch, *ap->a_vpp);
945 hammer_done_transaction(&trans);
950 * hammer_vop_open { vp, mode, cred, fp }
954 hammer_vop_open(struct vop_open_args *ap)
956 if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
959 return(vop_stdopen(ap));
963 * hammer_vop_pathconf { vp, name, retval }
967 hammer_vop_pathconf(struct vop_pathconf_args *ap)
973 * hammer_vop_print { vp }
977 hammer_vop_print(struct vop_print_args *ap)
983 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
987 hammer_vop_readdir(struct vop_readdir_args *ap)
989 struct hammer_transaction trans;
990 struct hammer_cursor cursor;
991 struct hammer_inode *ip;
993 hammer_base_elm_t base;
1001 ip = VTOI(ap->a_vp);
1003 saveoff = uio->uio_offset;
1005 if (ap->a_ncookies) {
1006 ncookies = uio->uio_resid / 16 + 1;
1007 if (ncookies > 1024)
1009 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1017 hammer_simple_transaction(&trans, ip->hmp);
1020 * Handle artificial entries
1024 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1028 cookies[cookie_index] = saveoff;
1031 if (cookie_index == ncookies)
1035 if (ip->ino_data.parent_obj_id) {
1036 r = vop_write_dirent(&error, uio,
1037 ip->ino_data.parent_obj_id,
1040 r = vop_write_dirent(&error, uio,
1041 ip->obj_id, DT_DIR, 2, "..");
1046 cookies[cookie_index] = saveoff;
1049 if (cookie_index == ncookies)
1054 * Key range (begin and end inclusive) to scan. Directory keys
1055 * directly translate to a 64 bit 'seek' position.
1057 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1058 cursor.key_beg.obj_id = ip->obj_id;
1059 cursor.key_beg.create_tid = 0;
1060 cursor.key_beg.delete_tid = 0;
1061 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1062 cursor.key_beg.obj_type = 0;
1063 cursor.key_beg.key = saveoff;
1065 cursor.key_end = cursor.key_beg;
1066 cursor.key_end.key = HAMMER_MAX_KEY;
1067 cursor.asof = ip->obj_asof;
1068 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1070 error = hammer_ip_first(&cursor);
1072 while (error == 0) {
1073 error = hammer_ip_resolve_data(&cursor);
1076 base = &cursor.leaf->base;
1077 saveoff = base->key;
1078 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1080 if (base->obj_id != ip->obj_id)
1081 panic("readdir: bad record at %p", cursor.node);
1083 r = vop_write_dirent(
1084 &error, uio, cursor.data->entry.obj_id,
1085 hammer_get_dtype(cursor.leaf->base.obj_type),
1086 cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1087 (void *)cursor.data->entry.name);
1092 cookies[cookie_index] = base->key;
1094 if (cookie_index == ncookies)
1096 error = hammer_ip_next(&cursor);
1098 hammer_done_cursor(&cursor);
1101 hammer_done_transaction(&trans);
1104 *ap->a_eofflag = (error == ENOENT);
1105 uio->uio_offset = saveoff;
1106 if (error && cookie_index == 0) {
1107 if (error == ENOENT)
1110 kfree(cookies, M_TEMP);
1111 *ap->a_ncookies = 0;
1112 *ap->a_cookies = NULL;
1115 if (error == ENOENT)
1118 *ap->a_ncookies = cookie_index;
1119 *ap->a_cookies = cookies;
1126 * hammer_vop_readlink { vp, uio, cred }
1130 hammer_vop_readlink(struct vop_readlink_args *ap)
1132 struct hammer_transaction trans;
1133 struct hammer_cursor cursor;
1134 struct hammer_inode *ip;
1137 ip = VTOI(ap->a_vp);
1139 hammer_simple_transaction(&trans, ip->hmp);
1141 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1144 * Key range (begin and end inclusive) to scan. Directory keys
1145 * directly translate to a 64 bit 'seek' position.
1147 cursor.key_beg.obj_id = ip->obj_id;
1148 cursor.key_beg.create_tid = 0;
1149 cursor.key_beg.delete_tid = 0;
1150 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1151 cursor.key_beg.obj_type = 0;
1152 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1153 cursor.asof = ip->obj_asof;
1154 cursor.flags |= HAMMER_CURSOR_ASOF;
1156 error = hammer_ip_lookup(&cursor);
1158 error = hammer_ip_resolve_data(&cursor);
1160 KKASSERT(cursor.leaf->data_len >=
1161 HAMMER_SYMLINK_NAME_OFF);
1162 error = uiomove(cursor.data->symlink.name,
1163 cursor.leaf->data_len -
1164 HAMMER_SYMLINK_NAME_OFF,
1168 hammer_done_cursor(&cursor);
1169 hammer_done_transaction(&trans);
1174 * hammer_vop_nremove { nch, dvp, cred }
1178 hammer_vop_nremove(struct vop_nremove_args *ap)
1180 struct hammer_transaction trans;
1183 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1184 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1185 hammer_done_transaction(&trans);
1191 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1195 hammer_vop_nrename(struct vop_nrename_args *ap)
1197 struct hammer_transaction trans;
1198 struct namecache *fncp;
1199 struct namecache *tncp;
1200 struct hammer_inode *fdip;
1201 struct hammer_inode *tdip;
1202 struct hammer_inode *ip;
1203 struct hammer_cursor cursor;
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(&trans, 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, HAMMER_INODE_DDIRTY);
1239 goto failed; /* XXX */
1242 * Locate the record in the originating directory and remove it.
1244 * Calculate the namekey and setup the key range for the scan. This
1245 * works kinda like a chained hash table where the lower 32 bits
1246 * of the namekey synthesize the chain.
1248 * The key range is inclusive of both key_beg and key_end.
1250 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1252 hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1253 cursor.key_beg.obj_id = fdip->obj_id;
1254 cursor.key_beg.key = namekey;
1255 cursor.key_beg.create_tid = 0;
1256 cursor.key_beg.delete_tid = 0;
1257 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1258 cursor.key_beg.obj_type = 0;
1260 cursor.key_end = cursor.key_beg;
1261 cursor.key_end.key |= 0xFFFFFFFFULL;
1262 cursor.asof = fdip->obj_asof;
1263 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1266 * Scan all matching records (the chain), locate the one matching
1267 * the requested path component.
1269 * The hammer_ip_*() functions merge in-memory records with on-disk
1270 * records for the purposes of the search.
1272 error = hammer_ip_first(&cursor);
1273 while (error == 0) {
1274 if (hammer_ip_resolve_data(&cursor) != 0)
1276 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1278 if (fncp->nc_nlen == nlen &&
1279 bcmp(fncp->nc_name, cursor.data->entry.name, 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);
1296 * XXX A deadlock here will break rename's atomicy for the purposes
1297 * of crash recovery.
1299 if (error == EDEADLK) {
1300 hammer_done_cursor(&cursor);
1305 * Cleanup and tell the kernel that the rename succeeded.
1307 hammer_done_cursor(&cursor);
1309 cache_rename(ap->a_fnch, ap->a_tnch);
1312 hammer_done_transaction(&trans);
1317 * hammer_vop_nrmdir { nch, dvp, cred }
1321 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1323 struct hammer_transaction trans;
1326 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1327 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1328 hammer_done_transaction(&trans);
1334 * hammer_vop_setattr { vp, vap, cred }
1338 hammer_vop_setattr(struct vop_setattr_args *ap)
1340 struct hammer_transaction trans;
1342 struct hammer_inode *ip;
1351 ip = ap->a_vp->v_data;
1354 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1356 if (ip->flags & HAMMER_INODE_RO)
1359 hammer_start_transaction(&trans, ip->hmp);
1362 if (vap->va_flags != VNOVAL) {
1363 flags = ip->ino_data.uflags;
1364 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1365 hammer_to_unix_xid(&ip->ino_data.uid),
1368 if (ip->ino_data.uflags != flags) {
1369 ip->ino_data.uflags = flags;
1370 modflags |= HAMMER_INODE_DDIRTY;
1372 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1379 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1383 if (vap->va_uid != (uid_t)VNOVAL) {
1384 hammer_guid_to_uuid(&uuid, vap->va_uid);
1385 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1386 ip->ino_data.uid = uuid;
1387 modflags |= HAMMER_INODE_DDIRTY;
1390 if (vap->va_gid != (uid_t)VNOVAL) {
1391 hammer_guid_to_uuid(&uuid, vap->va_gid);
1392 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1393 ip->ino_data.gid = uuid;
1394 modflags |= HAMMER_INODE_DDIRTY;
1397 while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1398 switch(ap->a_vp->v_type) {
1400 if (vap->va_size == ip->ino_data.size)
1403 * XXX break atomicy, we can deadlock the backend
1404 * if we do not release the lock. Probably not a
1407 if (vap->va_size < ip->ino_data.size) {
1408 vtruncbuf(ap->a_vp, vap->va_size,
1412 vnode_pager_setsize(ap->a_vp, vap->va_size);
1415 ip->ino_data.size = vap->va_size;
1416 modflags |= HAMMER_INODE_DDIRTY;
1417 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1421 * on-media truncation is cached in the inode until
1422 * the inode is synchronized.
1425 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1426 ip->flags |= HAMMER_INODE_TRUNCATED;
1427 ip->trunc_off = vap->va_size;
1428 } else if (ip->trunc_off > vap->va_size) {
1429 ip->trunc_off = vap->va_size;
1434 * If truncating we have to clean out a portion of
1435 * the last block on-disk. We do this in the
1436 * front-end buffer cache.
1438 if (truncating && vap->va_size < aligned_size) {
1442 offset = vap->va_size & HAMMER_BUFMASK;
1443 error = bread(ap->a_vp,
1444 aligned_size - HAMMER_BUFSIZE,
1445 HAMMER_BUFSIZE, &bp);
1447 bzero(bp->b_data + offset,
1448 HAMMER_BUFSIZE - offset);
1456 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1457 ip->flags |= HAMMER_INODE_TRUNCATED;
1458 ip->trunc_off = vap->va_size;
1459 } else if (ip->trunc_off > vap->va_size) {
1460 ip->trunc_off = vap->va_size;
1462 ip->ino_data.size = vap->va_size;
1463 modflags |= HAMMER_INODE_DDIRTY;
1471 if (vap->va_atime.tv_sec != VNOVAL) {
1472 ip->ino_leaf.atime =
1473 hammer_timespec_to_transid(&vap->va_atime);
1474 modflags |= HAMMER_INODE_ITIMES;
1476 if (vap->va_mtime.tv_sec != VNOVAL) {
1477 ip->ino_data.mtime =
1478 hammer_timespec_to_transid(&vap->va_mtime);
1479 modflags |= HAMMER_INODE_ITIMES;
1481 if (vap->va_mode != (mode_t)VNOVAL) {
1482 if (ip->ino_data.mode != vap->va_mode) {
1483 ip->ino_data.mode = vap->va_mode;
1484 modflags |= HAMMER_INODE_DDIRTY;
1489 hammer_modify_inode(&trans, ip, modflags);
1490 hammer_done_transaction(&trans);
1495 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1499 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1501 struct hammer_transaction trans;
1502 struct hammer_inode *dip;
1503 struct hammer_inode *nip;
1504 struct nchandle *nch;
1505 hammer_record_t record;
1509 ap->a_vap->va_type = VLNK;
1512 dip = VTOI(ap->a_dvp);
1514 if (dip->flags & HAMMER_INODE_RO)
1518 * Create a transaction to cover the operations we perform.
1520 hammer_start_transaction(&trans, dip->hmp);
1523 * Create a new filesystem object of the requested type. The
1524 * returned inode will be referenced but not locked.
1527 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1529 hammer_done_transaction(&trans);
1535 * Add a record representing the symlink. symlink stores the link
1536 * as pure data, not a string, and is no \0 terminated.
1539 bytes = strlen(ap->a_target);
1540 record = hammer_alloc_mem_record(nip, bytes);
1541 record->type = HAMMER_MEM_RECORD_GENERAL;
1543 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1544 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1545 record->leaf.data_len = bytes;
1546 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1547 bcopy(ap->a_target, record->data->symlink.name, bytes);
1548 error = hammer_ip_add_record(&trans, record);
1551 * Set the file size to the length of the link.
1554 nip->ino_data.size = bytes;
1555 hammer_modify_inode(&trans, nip, HAMMER_INODE_DDIRTY);
1559 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1565 hammer_rel_inode(nip, 0);
1568 error = hammer_get_vnode(nip, ap->a_vpp);
1569 hammer_rel_inode(nip, 0);
1571 cache_setunresolved(ap->a_nch);
1572 cache_setvp(ap->a_nch, *ap->a_vpp);
1575 hammer_done_transaction(&trans);
1580 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1584 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1586 struct hammer_transaction trans;
1589 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1590 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1591 ap->a_cred, ap->a_flags);
1592 hammer_done_transaction(&trans);
1598 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1602 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1604 struct hammer_inode *ip = ap->a_vp->v_data;
1606 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1607 ap->a_fflag, ap->a_cred));
1612 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1617 mp = ap->a_head.a_ops->head.vv_mount;
1620 case MOUNTCTL_SET_EXPORT:
1621 if (ap->a_ctllen != sizeof(struct export_args))
1623 error = hammer_vfs_export(mp, ap->a_op,
1624 (const struct export_args *)ap->a_ctl);
1627 error = journal_mountctl(ap);
1634 * hammer_vop_strategy { vp, bio }
1636 * Strategy call, used for regular file read & write only. Note that the
1637 * bp may represent a cluster.
1639 * To simplify operation and allow better optimizations in the future,
1640 * this code does not make any assumptions with regards to buffer alignment
1645 hammer_vop_strategy(struct vop_strategy_args *ap)
1650 bp = ap->a_bio->bio_buf;
1654 error = hammer_vop_strategy_read(ap);
1657 error = hammer_vop_strategy_write(ap);
1660 bp->b_error = error = EINVAL;
1661 bp->b_flags |= B_ERROR;
1669 * Read from a regular file. Iterate the related records and fill in the
1670 * BIO/BUF. Gaps are zero-filled.
1672 * The support code in hammer_object.c should be used to deal with mixed
1673 * in-memory and on-disk records.
1679 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1681 struct hammer_transaction trans;
1682 struct hammer_inode *ip;
1683 struct hammer_cursor cursor;
1684 hammer_base_elm_t base;
1697 ip = ap->a_vp->v_data;
1699 hammer_simple_transaction(&trans, ip->hmp);
1700 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1703 * Key range (begin and end inclusive) to scan. Note that the key's
1704 * stored in the actual records represent BASE+LEN, not BASE. The
1705 * first record containing bio_offset will have a key > bio_offset.
1707 cursor.key_beg.obj_id = ip->obj_id;
1708 cursor.key_beg.create_tid = 0;
1709 cursor.key_beg.delete_tid = 0;
1710 cursor.key_beg.obj_type = 0;
1711 cursor.key_beg.key = bio->bio_offset + 1;
1712 cursor.asof = ip->obj_asof;
1713 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1715 cursor.key_end = cursor.key_beg;
1716 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
1718 if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1719 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1720 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1721 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1725 ran_end = bio->bio_offset + bp->b_bufsize;
1726 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1727 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1728 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1729 if (tmp64 < ran_end)
1730 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1732 cursor.key_end.key = ran_end + MAXPHYS + 1;
1734 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1736 error = hammer_ip_first(&cursor);
1739 while (error == 0) {
1740 error = hammer_ip_resolve_data(&cursor);
1743 base = &cursor.leaf->base;
1745 rec_offset = base->key - cursor.leaf->data_len;
1748 * Calculate the gap, if any, and zero-fill it.
1750 n = (int)(rec_offset - (bio->bio_offset + boff));
1752 if (n > bp->b_bufsize - boff)
1753 n = bp->b_bufsize - boff;
1754 bzero((char *)bp->b_data + boff, n);
1760 * Calculate the data offset in the record and the number
1761 * of bytes we can copy.
1763 * Note there is a degenerate case here where boff may
1764 * already be at bp->b_bufsize.
1768 n = cursor.leaf->data_len - roff;
1770 if (n > bp->b_bufsize - boff)
1771 n = bp->b_bufsize - boff;
1774 * If we cached a truncation point on our front-end the
1775 * on-disk version may still have physical records beyond
1776 * that point. Truncate visibility.
1778 if (ip->trunc_off <= rec_offset)
1780 else if (ip->trunc_off < rec_offset + n)
1781 n = (int)(ip->trunc_off - rec_offset);
1787 bcopy((char *)cursor.data + roff,
1788 (char *)bp->b_data + boff, n);
1791 if (boff == bp->b_bufsize)
1793 error = hammer_ip_next(&cursor);
1795 hammer_done_cursor(&cursor);
1796 hammer_done_transaction(&trans);
1799 * There may have been a gap after the last record
1801 if (error == ENOENT)
1803 if (error == 0 && boff != bp->b_bufsize) {
1804 KKASSERT(boff < bp->b_bufsize);
1805 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1806 /* boff = bp->b_bufsize; */
1809 bp->b_error = error;
1811 bp->b_flags |= B_ERROR;
1817 * Write to a regular file. Because this is a strategy call the OS is
1818 * trying to actually sync data to the media. HAMMER can only flush
1819 * the entire inode (so the TID remains properly synchronized).
1821 * Basically all we do here is place the bio on the inode's flush queue
1822 * and activate the flusher.
1826 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1834 ip = ap->a_vp->v_data;
1836 if (ip->flags & HAMMER_INODE_RO) {
1837 bp->b_error = EROFS;
1838 bp->b_flags |= B_ERROR;
1844 * If the inode is being flushed we cannot re-queue buffers
1845 * it may have already flushed, or it could result in duplicate
1846 * records in the database.
1849 if (ip->flags & HAMMER_INODE_WRITE_ALT)
1850 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1852 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1854 hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1856 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1861 * If the write was not part of an integrated flush operation then
1864 if (ip->flush_state != HAMMER_FST_FLUSH ||
1865 (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1866 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1873 * Backend code which actually performs the write to the media. This
1874 * routine is typically called from the flusher. The bio will be disposed
1875 * of (biodone'd) by this routine.
1877 * Iterate the related records and mark for deletion. If existing edge
1878 * records (left and right side) overlap our write they have to be marked
1879 * deleted and new records created, usually referencing a portion of the
1880 * original data. Then add a record to represent the buffer.
1883 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
1885 struct buf *bp = bio->bio_buf;
1888 KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1891 * If the inode is going or gone, just throw away any frontend
1894 if (ip->flags & HAMMER_INODE_DELETED) {
1902 * Delete any records overlapping our range. This function will
1903 * (eventually) properly truncate partial overlaps.
1905 if (ip->sync_ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1906 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1909 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1915 * Add a single record to cover the write. We can write a record
1916 * with only the actual file data - for example, a small 200 byte
1917 * file does not have to write out a 16K record.
1919 * While the data size does not have to be aligned, we still do it
1920 * to reduce fragmentation in a future allocation model.
1925 if (ip->sync_ino_data.size - bio->bio_offset >
1927 limit_size = bp->b_bufsize;
1929 limit_size = (int)(ip->sync_ino_data.size -
1931 KKASSERT(limit_size >= 0);
1932 limit_size = (limit_size + 63) & ~63;
1935 error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
1936 bp->b_data, limit_size);
1940 Debugger("hammer_dowrite: error");
1943 bp->b_resid = bp->b_bufsize;
1944 bp->b_error = error;
1945 bp->b_flags |= B_ERROR;
1955 * dounlink - disconnect a directory entry
1957 * XXX whiteout support not really in yet
1960 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1961 struct vnode *dvp, struct ucred *cred, int flags)
1963 struct namecache *ncp;
1966 struct hammer_cursor cursor;
1971 * Calculate the namekey and setup the key range for the scan. This
1972 * works kinda like a chained hash table where the lower 32 bits
1973 * of the namekey synthesize the chain.
1975 * The key range is inclusive of both key_beg and key_end.
1980 if (dip->flags & HAMMER_INODE_RO)
1983 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1985 hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
1986 cursor.key_beg.obj_id = dip->obj_id;
1987 cursor.key_beg.key = namekey;
1988 cursor.key_beg.create_tid = 0;
1989 cursor.key_beg.delete_tid = 0;
1990 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1991 cursor.key_beg.obj_type = 0;
1993 cursor.key_end = cursor.key_beg;
1994 cursor.key_end.key |= 0xFFFFFFFFULL;
1995 cursor.asof = dip->obj_asof;
1996 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1999 * Scan all matching records (the chain), locate the one matching
2000 * the requested path component. info->last_error contains the
2001 * error code on search termination and could be 0, ENOENT, or
2004 * The hammer_ip_*() functions merge in-memory records with on-disk
2005 * records for the purposes of the search.
2007 error = hammer_ip_first(&cursor);
2009 while (error == 0) {
2010 error = hammer_ip_resolve_data(&cursor);
2013 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2015 if (ncp->nc_nlen == nlen &&
2016 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2019 error = hammer_ip_next(&cursor);
2023 * If all is ok we have to get the inode so we can adjust nlinks.
2025 * If the target is a directory, it must be empty.
2028 ip = hammer_get_inode(trans, &dip->cache[1],
2029 cursor.data->entry.obj_id,
2030 dip->hmp->asof, 0, &error);
2031 if (error == ENOENT) {
2032 kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2033 Debugger("ENOENT unlinking object that should exist");
2037 * If we are trying to remove a directory the directory must
2040 * WARNING: hammer_ip_check_directory_empty() may have to
2041 * terminate the cursor to avoid a deadlock. It is ok to
2042 * call hammer_done_cursor() twice.
2044 if (error == 0 && ip->ino_data.obj_type ==
2045 HAMMER_OBJTYPE_DIRECTORY) {
2046 error = hammer_ip_check_directory_empty(trans, &cursor,
2051 * Delete the directory entry.
2053 * WARNING: hammer_ip_del_directory() may have to terminate
2054 * the cursor to avoid a deadlock. It is ok to call
2055 * hammer_done_cursor() twice.
2058 error = hammer_ip_del_directory(trans, &cursor,
2062 cache_setunresolved(nch);
2063 cache_setvp(nch, NULL);
2066 cache_inval_vp(ip->vp, CINV_DESTROY);
2068 hammer_rel_inode(ip, 0);
2070 hammer_done_cursor(&cursor);
2071 if (error == EDEADLK)
2077 /************************************************************************
2078 * FIFO AND SPECFS OPS *
2079 ************************************************************************
2084 hammer_vop_fifoclose (struct vop_close_args *ap)
2086 /* XXX update itimes */
2087 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2091 hammer_vop_fiforead (struct vop_read_args *ap)
2095 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2096 /* XXX update access time */
2101 hammer_vop_fifowrite (struct vop_write_args *ap)
2105 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2106 /* XXX update access time */
2111 hammer_vop_specclose (struct vop_close_args *ap)
2113 /* XXX update itimes */
2114 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2118 hammer_vop_specread (struct vop_read_args *ap)
2120 /* XXX update access time */
2121 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2125 hammer_vop_specwrite (struct vop_write_args *ap)
2127 /* XXX update last change time */
2128 return (VOCALL(&spec_vnode_vops, &ap->a_head));