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 1.58 2008/06/02 20:19:03 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);
156 static void hammer_cleanup_write_io(hammer_inode_t ip);
161 hammer_vop_vnoperate(struct vop_generic_args *)
163 return (VOCALL(&hammer_vnode_vops, ap));
168 * hammer_vop_fsync { vp, waitfor }
172 hammer_vop_fsync(struct vop_fsync_args *ap)
174 hammer_inode_t ip = VTOI(ap->a_vp);
176 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
177 vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
178 if (ap->a_waitfor == MNT_WAIT)
179 hammer_wait_inode(ip);
184 * hammer_vop_read { vp, uio, ioflag, cred }
188 hammer_vop_read(struct vop_read_args *ap)
190 struct hammer_transaction trans;
199 if (ap->a_vp->v_type != VREG)
203 seqcount = ap->a_ioflag >> 16;
205 hammer_start_transaction(&trans, ip->hmp);
208 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
211 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
212 offset = uio->uio_offset & HAMMER_BUFMASK;
214 error = cluster_read(ap->a_vp, ip->ino_data.size,
215 uio->uio_offset - offset, HAMMER_BUFSIZE,
216 MAXBSIZE, seqcount, &bp);
218 error = bread(ap->a_vp, uio->uio_offset - offset,
219 HAMMER_BUFSIZE, &bp);
224 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
225 n = HAMMER_BUFSIZE - offset;
226 if (n > uio->uio_resid)
228 if (n > ip->ino_data.size - uio->uio_offset)
229 n = (int)(ip->ino_data.size - uio->uio_offset);
230 error = uiomove((char *)bp->b_data + offset, n, uio);
237 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
238 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
239 ip->ino_leaf.atime = trans.time;
240 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
242 hammer_done_transaction(&trans);
247 * hammer_vop_write { vp, uio, ioflag, cred }
251 hammer_vop_write(struct vop_write_args *ap)
253 struct hammer_transaction trans;
254 struct hammer_inode *ip;
263 if (ap->a_vp->v_type != VREG)
268 if (ip->flags & HAMMER_INODE_RO)
272 * Create a transaction to cover the operations we perform.
274 hammer_start_transaction(&trans, ip->hmp);
280 if (ap->a_ioflag & IO_APPEND)
281 uio->uio_offset = ip->ino_data.size;
284 * Check for illegal write offsets. Valid range is 0...2^63-1
286 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
287 hammer_done_transaction(&trans);
292 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
295 while (uio->uio_resid > 0) {
298 if ((error = hammer_checkspace(trans.hmp)) != 0)
302 * Do not allow huge writes to deadlock the buffer cache
304 if ((++count & 15) == 0) {
306 if ((ap->a_ioflag & IO_NOBWILL) == 0)
308 vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
311 offset = uio->uio_offset & HAMMER_BUFMASK;
312 n = HAMMER_BUFSIZE - offset;
313 if (n > uio->uio_resid)
315 if (uio->uio_offset + n > ip->ino_data.size) {
316 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
320 if (uio->uio_segflg == UIO_NOCOPY) {
322 * Issuing a write with the same data backing the
323 * buffer. Instantiate the buffer to collect the
324 * backing vm pages, then read-in any missing bits.
326 * This case is used by vop_stdputpages().
328 bp = getblk(ap->a_vp, uio->uio_offset - offset,
329 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
330 if ((bp->b_flags & B_CACHE) == 0) {
332 error = bread(ap->a_vp,
333 uio->uio_offset - offset,
334 HAMMER_BUFSIZE, &bp);
336 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
338 * Even though we are entirely overwriting the buffer
339 * we may still have to zero it out to avoid a
340 * mmap/write visibility issue.
342 bp = getblk(ap->a_vp, uio->uio_offset - offset,
343 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
344 if ((bp->b_flags & B_CACHE) == 0)
346 } else if (uio->uio_offset - offset >= ip->ino_data.size) {
348 * If the base offset of the buffer is beyond the
349 * file EOF, we don't have to issue a read.
351 bp = getblk(ap->a_vp, uio->uio_offset - offset,
352 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
356 * Partial overwrite, read in any missing bits then
357 * replace the portion being written.
359 error = bread(ap->a_vp, uio->uio_offset - offset,
360 HAMMER_BUFSIZE, &bp);
365 error = uiomove((char *)bp->b_data + offset, n, uio);
368 * If we screwed up we have to undo any VM size changes we
374 vtruncbuf(ap->a_vp, ip->ino_data.size,
379 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
380 if (ip->ino_data.size < uio->uio_offset) {
381 ip->ino_data.size = uio->uio_offset;
382 flags = HAMMER_INODE_DDIRTY;
383 vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
387 ip->ino_data.mtime = trans.time;
388 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
389 flags |= HAMMER_INODE_DDIRTY; /* XXX mtime */
390 hammer_modify_inode(&trans, ip, flags);
392 if ((bp->b_flags & B_DIRTY) == 0) {
394 ++ip->hmp->rsv_databufs;
397 if (ap->a_ioflag & IO_SYNC) {
399 } else if (ap->a_ioflag & IO_DIRECT) {
402 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
403 (uio->uio_offset & HAMMER_BUFMASK) == 0) {
405 * XXX HAMMER can only fsync the whole inode,
406 * doing it on every buffer would be a bad idea.
409 * If seqcount indicates sequential operation and
410 * we just finished filling a buffer, push it out
411 * now to prevent the buffer cache from becoming
412 * too full, which would trigger non-optimal
421 hammer_done_transaction(&trans);
426 * hammer_vop_access { vp, mode, cred }
430 hammer_vop_access(struct vop_access_args *ap)
432 struct hammer_inode *ip = VTOI(ap->a_vp);
437 uid = hammer_to_unix_xid(&ip->ino_data.uid);
438 gid = hammer_to_unix_xid(&ip->ino_data.gid);
440 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
441 ip->ino_data.uflags);
446 * hammer_vop_advlock { vp, id, op, fl, flags }
450 hammer_vop_advlock(struct vop_advlock_args *ap)
452 struct hammer_inode *ip = VTOI(ap->a_vp);
454 return (lf_advlock(ap, &ip->advlock, ip->ino_data.size));
458 * hammer_vop_close { vp, fflag }
462 hammer_vop_close(struct vop_close_args *ap)
464 return (vop_stdclose(ap));
468 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
470 * The operating system has already ensured that the directory entry
471 * does not exist and done all appropriate namespace locking.
475 hammer_vop_ncreate(struct vop_ncreate_args *ap)
477 struct hammer_transaction trans;
478 struct hammer_inode *dip;
479 struct hammer_inode *nip;
480 struct nchandle *nch;
484 dip = VTOI(ap->a_dvp);
486 if (dip->flags & HAMMER_INODE_RO)
488 if ((error = hammer_checkspace(dip->hmp)) != 0)
492 * Create a transaction to cover the operations we perform.
494 hammer_start_transaction(&trans, dip->hmp);
497 * Create a new filesystem object of the requested type. The
498 * returned inode will be referenced and shared-locked to prevent
499 * it from being moved to the flusher.
502 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
504 hkprintf("hammer_create_inode error %d\n", error);
505 hammer_done_transaction(&trans);
511 * Add the new filesystem object to the directory. This will also
512 * bump the inode's link count.
514 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
516 hkprintf("hammer_ip_add_directory error %d\n", error);
522 hammer_rel_inode(nip, 0);
523 hammer_done_transaction(&trans);
526 error = hammer_get_vnode(nip, ap->a_vpp);
527 hammer_done_transaction(&trans);
528 hammer_rel_inode(nip, 0);
530 cache_setunresolved(ap->a_nch);
531 cache_setvp(ap->a_nch, *ap->a_vpp);
538 * hammer_vop_getattr { vp, vap }
540 * Retrieve an inode's attribute information. When accessing inodes
541 * historically we fake the atime field to ensure consistent results.
542 * The atime field is stored in the B-Tree element and allowed to be
543 * updated without cycling the element.
547 hammer_vop_getattr(struct vop_getattr_args *ap)
549 struct hammer_inode *ip = VTOI(ap->a_vp);
550 struct vattr *vap = ap->a_vap;
553 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
554 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
559 hammer_itimes(ap->a_vp);
562 vap->va_fsid = ip->hmp->fsid_udev;
563 vap->va_fileid = ip->ino_leaf.base.obj_id;
564 vap->va_mode = ip->ino_data.mode;
565 vap->va_nlink = ip->ino_data.nlinks;
566 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
567 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
570 vap->va_size = ip->ino_data.size;
571 if (ip->flags & HAMMER_INODE_RO)
572 hammer_to_timespec(ip->ino_data.mtime, &vap->va_atime);
574 hammer_to_timespec(ip->ino_leaf.atime, &vap->va_atime);
575 hammer_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
576 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
577 vap->va_flags = ip->ino_data.uflags;
578 vap->va_gen = 1; /* hammer inums are unique for all time */
579 vap->va_blocksize = HAMMER_BUFSIZE;
580 vap->va_bytes = (ip->ino_data.size + 63) & ~63;
581 vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
582 vap->va_filerev = 0; /* XXX */
583 /* mtime uniquely identifies any adjustments made to the file */
584 vap->va_fsmid = ip->ino_data.mtime;
585 vap->va_uid_uuid = ip->ino_data.uid;
586 vap->va_gid_uuid = ip->ino_data.gid;
587 vap->va_fsid_uuid = ip->hmp->fsid;
588 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
591 switch (ip->ino_data.obj_type) {
592 case HAMMER_OBJTYPE_CDEV:
593 case HAMMER_OBJTYPE_BDEV:
594 vap->va_rmajor = ip->ino_data.rmajor;
595 vap->va_rminor = ip->ino_data.rminor;
605 * hammer_vop_nresolve { nch, dvp, cred }
607 * Locate the requested directory entry.
611 hammer_vop_nresolve(struct vop_nresolve_args *ap)
613 struct hammer_transaction trans;
614 struct namecache *ncp;
618 struct hammer_cursor cursor;
628 * Misc initialization, plus handle as-of name extensions. Look for
629 * the '@@' extension. Note that as-of files and directories cannot
632 dip = VTOI(ap->a_dvp);
633 ncp = ap->a_nch->ncp;
634 asof = dip->obj_asof;
638 hammer_simple_transaction(&trans, dip->hmp);
640 for (i = 0; i < nlen; ++i) {
641 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
642 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
643 flags |= HAMMER_INODE_RO;
650 * If there is no path component the time extension is relative to
654 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
655 asof, flags, &error);
657 error = hammer_get_vnode(ip, &vp);
658 hammer_rel_inode(ip, 0);
664 cache_setvp(ap->a_nch, vp);
671 * Calculate the namekey and setup the key range for the scan. This
672 * works kinda like a chained hash table where the lower 32 bits
673 * of the namekey synthesize the chain.
675 * The key range is inclusive of both key_beg and key_end.
677 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
679 error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
680 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
681 cursor.key_beg.obj_id = dip->obj_id;
682 cursor.key_beg.key = namekey;
683 cursor.key_beg.create_tid = 0;
684 cursor.key_beg.delete_tid = 0;
685 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
686 cursor.key_beg.obj_type = 0;
688 cursor.key_end = cursor.key_beg;
689 cursor.key_end.key |= 0xFFFFFFFFULL;
691 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
694 * Scan all matching records (the chain), locate the one matching
695 * the requested path component.
697 * The hammer_ip_*() functions merge in-memory records with on-disk
698 * records for the purposes of the search.
703 error = hammer_ip_first(&cursor);
705 error = hammer_ip_resolve_data(&cursor);
708 if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
709 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
710 obj_id = cursor.data->entry.obj_id;
713 error = hammer_ip_next(&cursor);
716 hammer_done_cursor(&cursor);
718 ip = hammer_get_inode(&trans, &dip->cache[1],
719 obj_id, asof, flags, &error);
721 error = hammer_get_vnode(ip, &vp);
722 hammer_rel_inode(ip, 0);
728 cache_setvp(ap->a_nch, vp);
731 } else if (error == ENOENT) {
732 cache_setvp(ap->a_nch, NULL);
735 hammer_done_transaction(&trans);
740 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
742 * Locate the parent directory of a directory vnode.
744 * dvp is referenced but not locked. *vpp must be returned referenced and
745 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
746 * at the root, instead it could indicate that the directory we were in was
749 * NOTE: as-of sequences are not linked into the directory structure. If
750 * we are at the root with a different asof then the mount point, reload
751 * the same directory with the mount point's asof. I'm not sure what this
752 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
753 * get confused, but it hasn't been tested.
757 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
759 struct hammer_transaction trans;
760 struct hammer_inode *dip;
761 struct hammer_inode *ip;
762 int64_t parent_obj_id;
766 dip = VTOI(ap->a_dvp);
767 asof = dip->obj_asof;
768 parent_obj_id = dip->ino_data.parent_obj_id;
770 if (parent_obj_id == 0) {
771 if (dip->obj_id == HAMMER_OBJID_ROOT &&
772 asof != dip->hmp->asof) {
773 parent_obj_id = dip->obj_id;
774 asof = dip->hmp->asof;
775 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
776 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
784 hammer_simple_transaction(&trans, dip->hmp);
786 ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
787 asof, dip->flags, &error);
789 error = hammer_get_vnode(ip, ap->a_vpp);
790 hammer_rel_inode(ip, 0);
794 hammer_done_transaction(&trans);
799 * hammer_vop_nlink { nch, dvp, vp, cred }
803 hammer_vop_nlink(struct vop_nlink_args *ap)
805 struct hammer_transaction trans;
806 struct hammer_inode *dip;
807 struct hammer_inode *ip;
808 struct nchandle *nch;
812 dip = VTOI(ap->a_dvp);
815 if (dip->flags & HAMMER_INODE_RO)
817 if (ip->flags & HAMMER_INODE_RO)
819 if ((error = hammer_checkspace(dip->hmp)) != 0)
823 * Create a transaction to cover the operations we perform.
825 hammer_start_transaction(&trans, dip->hmp);
828 * Add the filesystem object to the directory. Note that neither
829 * dip nor ip are referenced or locked, but their vnodes are
830 * referenced. This function will bump the inode's link count.
832 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
838 cache_setunresolved(nch);
839 cache_setvp(nch, ap->a_vp);
841 hammer_done_transaction(&trans);
846 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
848 * The operating system has already ensured that the directory entry
849 * does not exist and done all appropriate namespace locking.
853 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
855 struct hammer_transaction trans;
856 struct hammer_inode *dip;
857 struct hammer_inode *nip;
858 struct nchandle *nch;
862 dip = VTOI(ap->a_dvp);
864 if (dip->flags & HAMMER_INODE_RO)
866 if ((error = hammer_checkspace(dip->hmp)) != 0)
870 * Create a transaction to cover the operations we perform.
872 hammer_start_transaction(&trans, dip->hmp);
875 * Create a new filesystem object of the requested type. The
876 * returned inode will be referenced but not locked.
878 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
880 hkprintf("hammer_mkdir error %d\n", error);
881 hammer_done_transaction(&trans);
886 * Add the new filesystem object to the directory. This will also
887 * bump the inode's link count.
889 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
891 hkprintf("hammer_mkdir (add) error %d\n", error);
897 hammer_rel_inode(nip, 0);
900 error = hammer_get_vnode(nip, ap->a_vpp);
901 hammer_rel_inode(nip, 0);
903 cache_setunresolved(ap->a_nch);
904 cache_setvp(ap->a_nch, *ap->a_vpp);
907 hammer_done_transaction(&trans);
912 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
914 * The operating system has already ensured that the directory entry
915 * does not exist and done all appropriate namespace locking.
919 hammer_vop_nmknod(struct vop_nmknod_args *ap)
921 struct hammer_transaction trans;
922 struct hammer_inode *dip;
923 struct hammer_inode *nip;
924 struct nchandle *nch;
928 dip = VTOI(ap->a_dvp);
930 if (dip->flags & HAMMER_INODE_RO)
932 if ((error = hammer_checkspace(dip->hmp)) != 0)
936 * Create a transaction to cover the operations we perform.
938 hammer_start_transaction(&trans, dip->hmp);
941 * Create a new filesystem object of the requested type. The
942 * returned inode will be referenced but not locked.
944 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
946 hammer_done_transaction(&trans);
952 * Add the new filesystem object to the directory. This will also
953 * bump the inode's link count.
955 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
961 hammer_rel_inode(nip, 0);
964 error = hammer_get_vnode(nip, ap->a_vpp);
965 hammer_rel_inode(nip, 0);
967 cache_setunresolved(ap->a_nch);
968 cache_setvp(ap->a_nch, *ap->a_vpp);
971 hammer_done_transaction(&trans);
976 * hammer_vop_open { vp, mode, cred, fp }
980 hammer_vop_open(struct vop_open_args *ap)
982 if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
985 return(vop_stdopen(ap));
989 * hammer_vop_pathconf { vp, name, retval }
993 hammer_vop_pathconf(struct vop_pathconf_args *ap)
999 * hammer_vop_print { vp }
1003 hammer_vop_print(struct vop_print_args *ap)
1009 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
1013 hammer_vop_readdir(struct vop_readdir_args *ap)
1015 struct hammer_transaction trans;
1016 struct hammer_cursor cursor;
1017 struct hammer_inode *ip;
1019 hammer_base_elm_t base;
1027 ip = VTOI(ap->a_vp);
1029 saveoff = uio->uio_offset;
1031 if (ap->a_ncookies) {
1032 ncookies = uio->uio_resid / 16 + 1;
1033 if (ncookies > 1024)
1035 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1043 hammer_simple_transaction(&trans, ip->hmp);
1046 * Handle artificial entries
1050 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1054 cookies[cookie_index] = saveoff;
1057 if (cookie_index == ncookies)
1061 if (ip->ino_data.parent_obj_id) {
1062 r = vop_write_dirent(&error, uio,
1063 ip->ino_data.parent_obj_id,
1066 r = vop_write_dirent(&error, uio,
1067 ip->obj_id, DT_DIR, 2, "..");
1072 cookies[cookie_index] = saveoff;
1075 if (cookie_index == ncookies)
1080 * Key range (begin and end inclusive) to scan. Directory keys
1081 * directly translate to a 64 bit 'seek' position.
1083 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1084 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1085 cursor.key_beg.obj_id = ip->obj_id;
1086 cursor.key_beg.create_tid = 0;
1087 cursor.key_beg.delete_tid = 0;
1088 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1089 cursor.key_beg.obj_type = 0;
1090 cursor.key_beg.key = saveoff;
1092 cursor.key_end = cursor.key_beg;
1093 cursor.key_end.key = HAMMER_MAX_KEY;
1094 cursor.asof = ip->obj_asof;
1095 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1097 error = hammer_ip_first(&cursor);
1099 while (error == 0) {
1100 error = hammer_ip_resolve_data(&cursor);
1103 base = &cursor.leaf->base;
1104 saveoff = base->key;
1105 KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);
1107 if (base->obj_id != ip->obj_id)
1108 panic("readdir: bad record at %p", cursor.node);
1110 r = vop_write_dirent(
1111 &error, uio, cursor.data->entry.obj_id,
1112 hammer_get_dtype(cursor.leaf->base.obj_type),
1113 cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
1114 (void *)cursor.data->entry.name);
1119 cookies[cookie_index] = base->key;
1121 if (cookie_index == ncookies)
1123 error = hammer_ip_next(&cursor);
1125 hammer_done_cursor(&cursor);
1128 hammer_done_transaction(&trans);
1131 *ap->a_eofflag = (error == ENOENT);
1132 uio->uio_offset = saveoff;
1133 if (error && cookie_index == 0) {
1134 if (error == ENOENT)
1137 kfree(cookies, M_TEMP);
1138 *ap->a_ncookies = 0;
1139 *ap->a_cookies = NULL;
1142 if (error == ENOENT)
1145 *ap->a_ncookies = cookie_index;
1146 *ap->a_cookies = cookies;
1153 * hammer_vop_readlink { vp, uio, cred }
1157 hammer_vop_readlink(struct vop_readlink_args *ap)
1159 struct hammer_transaction trans;
1160 struct hammer_cursor cursor;
1161 struct hammer_inode *ip;
1164 ip = VTOI(ap->a_vp);
1167 * Shortcut if the symlink data was stuffed into ino_data.
1169 if (ip->ino_data.size <= HAMMER_INODE_BASESYMLEN) {
1170 error = uiomove(ip->ino_data.ext.symlink,
1171 ip->ino_data.size, ap->a_uio);
1178 hammer_simple_transaction(&trans, ip->hmp);
1179 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1182 * Key range (begin and end inclusive) to scan. Directory keys
1183 * directly translate to a 64 bit 'seek' position.
1185 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC; /* XXX */
1186 cursor.key_beg.obj_id = ip->obj_id;
1187 cursor.key_beg.create_tid = 0;
1188 cursor.key_beg.delete_tid = 0;
1189 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1190 cursor.key_beg.obj_type = 0;
1191 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1192 cursor.asof = ip->obj_asof;
1193 cursor.flags |= HAMMER_CURSOR_ASOF;
1195 error = hammer_ip_lookup(&cursor);
1197 error = hammer_ip_resolve_data(&cursor);
1199 KKASSERT(cursor.leaf->data_len >=
1200 HAMMER_SYMLINK_NAME_OFF);
1201 error = uiomove(cursor.data->symlink.name,
1202 cursor.leaf->data_len -
1203 HAMMER_SYMLINK_NAME_OFF,
1207 hammer_done_cursor(&cursor);
1208 hammer_done_transaction(&trans);
1213 * hammer_vop_nremove { nch, dvp, cred }
1217 hammer_vop_nremove(struct vop_nremove_args *ap)
1219 struct hammer_transaction trans;
1220 struct hammer_inode *dip;
1223 dip = VTOI(ap->a_dvp);
1225 if (hammer_nohistory(dip) == 0 &&
1226 (error = hammer_checkspace(dip->hmp)) != 0) {
1230 hammer_start_transaction(&trans, dip->hmp);
1231 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1232 hammer_done_transaction(&trans);
1238 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1242 hammer_vop_nrename(struct vop_nrename_args *ap)
1244 struct hammer_transaction trans;
1245 struct namecache *fncp;
1246 struct namecache *tncp;
1247 struct hammer_inode *fdip;
1248 struct hammer_inode *tdip;
1249 struct hammer_inode *ip;
1250 struct hammer_cursor cursor;
1254 fdip = VTOI(ap->a_fdvp);
1255 tdip = VTOI(ap->a_tdvp);
1256 fncp = ap->a_fnch->ncp;
1257 tncp = ap->a_tnch->ncp;
1258 ip = VTOI(fncp->nc_vp);
1259 KKASSERT(ip != NULL);
1261 if (fdip->flags & HAMMER_INODE_RO)
1263 if (tdip->flags & HAMMER_INODE_RO)
1265 if (ip->flags & HAMMER_INODE_RO)
1267 if ((error = hammer_checkspace(fdip->hmp)) != 0)
1270 hammer_start_transaction(&trans, fdip->hmp);
1273 * Remove tncp from the target directory and then link ip as
1274 * tncp. XXX pass trans to dounlink
1276 * Force the inode sync-time to match the transaction so it is
1277 * in-sync with the creation of the target directory entry.
1279 error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1280 if (error == 0 || error == ENOENT) {
1281 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1283 ip->ino_data.parent_obj_id = tdip->obj_id;
1284 hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1288 goto failed; /* XXX */
1291 * Locate the record in the originating directory and remove it.
1293 * Calculate the namekey and setup the key range for the scan. This
1294 * works kinda like a chained hash table where the lower 32 bits
1295 * of the namekey synthesize the chain.
1297 * The key range is inclusive of both key_beg and key_end.
1299 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1301 hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1302 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1303 cursor.key_beg.obj_id = fdip->obj_id;
1304 cursor.key_beg.key = namekey;
1305 cursor.key_beg.create_tid = 0;
1306 cursor.key_beg.delete_tid = 0;
1307 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1308 cursor.key_beg.obj_type = 0;
1310 cursor.key_end = cursor.key_beg;
1311 cursor.key_end.key |= 0xFFFFFFFFULL;
1312 cursor.asof = fdip->obj_asof;
1313 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1316 * Scan all matching records (the chain), locate the one matching
1317 * the requested path component.
1319 * The hammer_ip_*() functions merge in-memory records with on-disk
1320 * records for the purposes of the search.
1322 error = hammer_ip_first(&cursor);
1323 while (error == 0) {
1324 if (hammer_ip_resolve_data(&cursor) != 0)
1326 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
1328 if (fncp->nc_nlen == nlen &&
1329 bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
1332 error = hammer_ip_next(&cursor);
1336 * If all is ok we have to get the inode so we can adjust nlinks.
1338 * WARNING: hammer_ip_del_directory() may have to terminate the
1339 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1343 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1346 * XXX A deadlock here will break rename's atomicy for the purposes
1347 * of crash recovery.
1349 if (error == EDEADLK) {
1350 hammer_done_cursor(&cursor);
1355 * Cleanup and tell the kernel that the rename succeeded.
1357 hammer_done_cursor(&cursor);
1359 cache_rename(ap->a_fnch, ap->a_tnch);
1362 hammer_done_transaction(&trans);
1367 * hammer_vop_nrmdir { nch, dvp, cred }
1371 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1373 struct hammer_transaction trans;
1374 struct hammer_inode *dip;
1377 dip = VTOI(ap->a_dvp);
1379 if (hammer_nohistory(dip) == 0 &&
1380 (error = hammer_checkspace(dip->hmp)) != 0) {
1384 hammer_start_transaction(&trans, dip->hmp);
1385 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1386 hammer_done_transaction(&trans);
1392 * hammer_vop_setattr { vp, vap, cred }
1396 hammer_vop_setattr(struct vop_setattr_args *ap)
1398 struct hammer_transaction trans;
1400 struct hammer_inode *ip;
1408 ip = ap->a_vp->v_data;
1411 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1413 if (ip->flags & HAMMER_INODE_RO)
1415 if (hammer_nohistory(ip) == 0 &&
1416 (error = hammer_checkspace(ip->hmp)) != 0) {
1420 hammer_start_transaction(&trans, ip->hmp);
1423 if (vap->va_flags != VNOVAL) {
1424 flags = ip->ino_data.uflags;
1425 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1426 hammer_to_unix_xid(&ip->ino_data.uid),
1429 if (ip->ino_data.uflags != flags) {
1430 ip->ino_data.uflags = flags;
1431 modflags |= HAMMER_INODE_DDIRTY;
1433 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1440 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1444 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
1445 mode_t cur_mode = ip->ino_data.mode;
1446 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1447 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1451 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
1453 &cur_uid, &cur_gid, &cur_mode);
1455 hammer_guid_to_uuid(&uuid_uid, cur_uid);
1456 hammer_guid_to_uuid(&uuid_gid, cur_gid);
1457 if (bcmp(&uuid_uid, &ip->ino_data.uid,
1458 sizeof(uuid_uid)) ||
1459 bcmp(&uuid_gid, &ip->ino_data.gid,
1460 sizeof(uuid_gid)) ||
1461 ip->ino_data.mode != cur_mode
1463 ip->ino_data.uid = uuid_uid;
1464 ip->ino_data.gid = uuid_gid;
1465 ip->ino_data.mode = cur_mode;
1467 modflags |= HAMMER_INODE_DDIRTY;
1470 while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
1471 switch(ap->a_vp->v_type) {
1473 if (vap->va_size == ip->ino_data.size)
1476 * XXX break atomicy, we can deadlock the backend
1477 * if we do not release the lock. Probably not a
1480 if (vap->va_size < ip->ino_data.size) {
1481 vtruncbuf(ap->a_vp, vap->va_size,
1485 vnode_pager_setsize(ap->a_vp, vap->va_size);
1488 ip->ino_data.size = vap->va_size;
1489 modflags |= HAMMER_INODE_DDIRTY;
1490 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1494 * on-media truncation is cached in the inode until
1495 * the inode is synchronized.
1498 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1499 ip->flags |= HAMMER_INODE_TRUNCATED;
1500 ip->trunc_off = vap->va_size;
1501 } else if (ip->trunc_off > vap->va_size) {
1502 ip->trunc_off = vap->va_size;
1507 * If truncating we have to clean out a portion of
1508 * the last block on-disk. We do this in the
1509 * front-end buffer cache.
1511 if (truncating && vap->va_size < aligned_size) {
1515 offset = vap->va_size & HAMMER_BUFMASK;
1516 error = bread(ap->a_vp,
1517 aligned_size - HAMMER_BUFSIZE,
1518 HAMMER_BUFSIZE, &bp);
1520 bzero(bp->b_data + offset,
1521 HAMMER_BUFSIZE - offset);
1529 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1530 ip->flags |= HAMMER_INODE_TRUNCATED;
1531 ip->trunc_off = vap->va_size;
1532 } else if (ip->trunc_off > vap->va_size) {
1533 ip->trunc_off = vap->va_size;
1535 ip->ino_data.size = vap->va_size;
1536 modflags |= HAMMER_INODE_DDIRTY;
1544 if (vap->va_atime.tv_sec != VNOVAL) {
1545 ip->ino_leaf.atime =
1546 hammer_timespec_to_transid(&vap->va_atime);
1547 modflags |= HAMMER_INODE_ITIMES;
1549 if (vap->va_mtime.tv_sec != VNOVAL) {
1550 ip->ino_data.mtime =
1551 hammer_timespec_to_transid(&vap->va_mtime);
1552 modflags |= HAMMER_INODE_ITIMES;
1553 modflags |= HAMMER_INODE_DDIRTY; /* XXX mtime */
1555 if (vap->va_mode != (mode_t)VNOVAL) {
1556 mode_t cur_mode = ip->ino_data.mode;
1557 uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
1558 gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);
1560 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
1561 cur_uid, cur_gid, &cur_mode);
1562 if (error == 0 && ip->ino_data.mode != cur_mode) {
1563 ip->ino_data.mode = cur_mode;
1564 modflags |= HAMMER_INODE_DDIRTY;
1569 hammer_modify_inode(&trans, ip, modflags);
1570 hammer_done_transaction(&trans);
1575 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1579 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1581 struct hammer_transaction trans;
1582 struct hammer_inode *dip;
1583 struct hammer_inode *nip;
1584 struct nchandle *nch;
1585 hammer_record_t record;
1589 ap->a_vap->va_type = VLNK;
1592 dip = VTOI(ap->a_dvp);
1594 if (dip->flags & HAMMER_INODE_RO)
1596 if ((error = hammer_checkspace(dip->hmp)) != 0)
1600 * Create a transaction to cover the operations we perform.
1602 hammer_start_transaction(&trans, dip->hmp);
1605 * Create a new filesystem object of the requested type. The
1606 * returned inode will be referenced but not locked.
1609 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1611 hammer_done_transaction(&trans);
1617 * Add a record representing the symlink. symlink stores the link
1618 * as pure data, not a string, and is no \0 terminated.
1621 bytes = strlen(ap->a_target);
1623 if (bytes <= HAMMER_INODE_BASESYMLEN) {
1624 bcopy(ap->a_target, nip->ino_data.ext.symlink, bytes);
1626 record = hammer_alloc_mem_record(nip, bytes);
1627 record->type = HAMMER_MEM_RECORD_GENERAL;
1629 record->leaf.base.localization = HAMMER_LOCALIZE_MISC;
1630 record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
1631 record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
1632 record->leaf.data_len = bytes;
1633 KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
1634 bcopy(ap->a_target, record->data->symlink.name, bytes);
1635 error = hammer_ip_add_record(&trans, record);
1639 * Set the file size to the length of the link.
1642 nip->ino_data.size = bytes;
1643 hammer_modify_inode(&trans, nip, HAMMER_INODE_DDIRTY);
1647 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1653 hammer_rel_inode(nip, 0);
1656 error = hammer_get_vnode(nip, ap->a_vpp);
1657 hammer_rel_inode(nip, 0);
1659 cache_setunresolved(ap->a_nch);
1660 cache_setvp(ap->a_nch, *ap->a_vpp);
1663 hammer_done_transaction(&trans);
1668 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1672 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1674 struct hammer_transaction trans;
1675 struct hammer_inode *dip;
1678 dip = VTOI(ap->a_dvp);
1680 if (hammer_nohistory(dip) == 0 &&
1681 (error = hammer_checkspace(dip->hmp)) != 0) {
1685 hammer_start_transaction(&trans, dip->hmp);
1686 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1687 ap->a_cred, ap->a_flags);
1688 hammer_done_transaction(&trans);
1694 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1698 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1700 struct hammer_inode *ip = ap->a_vp->v_data;
1702 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1703 ap->a_fflag, ap->a_cred));
1708 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1713 mp = ap->a_head.a_ops->head.vv_mount;
1716 case MOUNTCTL_SET_EXPORT:
1717 if (ap->a_ctllen != sizeof(struct export_args))
1719 error = hammer_vfs_export(mp, ap->a_op,
1720 (const struct export_args *)ap->a_ctl);
1723 error = journal_mountctl(ap);
1730 * hammer_vop_strategy { vp, bio }
1732 * Strategy call, used for regular file read & write only. Note that the
1733 * bp may represent a cluster.
1735 * To simplify operation and allow better optimizations in the future,
1736 * this code does not make any assumptions with regards to buffer alignment
1741 hammer_vop_strategy(struct vop_strategy_args *ap)
1746 bp = ap->a_bio->bio_buf;
1750 error = hammer_vop_strategy_read(ap);
1753 error = hammer_vop_strategy_write(ap);
1756 bp->b_error = error = EINVAL;
1757 bp->b_flags |= B_ERROR;
1765 * Read from a regular file. Iterate the related records and fill in the
1766 * BIO/BUF. Gaps are zero-filled.
1768 * The support code in hammer_object.c should be used to deal with mixed
1769 * in-memory and on-disk records.
1775 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1777 struct hammer_transaction trans;
1778 struct hammer_inode *ip;
1779 struct hammer_cursor cursor;
1780 hammer_base_elm_t base;
1793 ip = ap->a_vp->v_data;
1795 hammer_simple_transaction(&trans, ip->hmp);
1796 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1799 * Key range (begin and end inclusive) to scan. Note that the key's
1800 * stored in the actual records represent BASE+LEN, not BASE. The
1801 * first record containing bio_offset will have a key > bio_offset.
1803 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
1804 cursor.key_beg.obj_id = ip->obj_id;
1805 cursor.key_beg.create_tid = 0;
1806 cursor.key_beg.delete_tid = 0;
1807 cursor.key_beg.obj_type = 0;
1808 cursor.key_beg.key = bio->bio_offset + 1;
1809 cursor.asof = ip->obj_asof;
1810 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1812 cursor.key_end = cursor.key_beg;
1813 KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
1815 if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
1816 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1817 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1818 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1822 ran_end = bio->bio_offset + bp->b_bufsize;
1823 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1824 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1825 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1826 if (tmp64 < ran_end)
1827 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1829 cursor.key_end.key = ran_end + MAXPHYS + 1;
1831 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1833 error = hammer_ip_first(&cursor);
1836 while (error == 0) {
1837 error = hammer_ip_resolve_data(&cursor);
1840 base = &cursor.leaf->base;
1842 rec_offset = base->key - cursor.leaf->data_len;
1845 * Calculate the gap, if any, and zero-fill it.
1847 * n is the offset of the start of the record verses our
1848 * current seek offset in the bio.
1850 n = (int)(rec_offset - (bio->bio_offset + boff));
1852 if (n > bp->b_bufsize - boff)
1853 n = bp->b_bufsize - boff;
1854 bzero((char *)bp->b_data + boff, n);
1860 * Calculate the data offset in the record and the number
1861 * of bytes we can copy.
1863 * There are two degenerate cases. First, boff may already
1864 * be at bp->b_bufsize. Secondly, the data offset within
1865 * the record may exceed the record's size.
1869 n = cursor.leaf->data_len - roff;
1871 kprintf("strategy_read: bad n=%d roff=%d\n", n, roff);
1873 } else if (n > bp->b_bufsize - boff) {
1874 n = bp->b_bufsize - boff;
1878 * If we cached a truncation point on our front-end the
1879 * on-disk version may still have physical records beyond
1880 * that point. Truncate visibility.
1882 if (ip->trunc_off <= rec_offset)
1884 else if (ip->trunc_off < rec_offset + n)
1885 n = (int)(ip->trunc_off - rec_offset);
1891 bcopy((char *)cursor.data + roff,
1892 (char *)bp->b_data + boff, n);
1895 if (boff == bp->b_bufsize)
1897 error = hammer_ip_next(&cursor);
1899 hammer_done_cursor(&cursor);
1900 hammer_done_transaction(&trans);
1903 * There may have been a gap after the last record
1905 if (error == ENOENT)
1907 if (error == 0 && boff != bp->b_bufsize) {
1908 KKASSERT(boff < bp->b_bufsize);
1909 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1910 /* boff = bp->b_bufsize; */
1913 bp->b_error = error;
1915 bp->b_flags |= B_ERROR;
1921 * Write to a regular file. Because this is a strategy call the OS is
1922 * trying to actually sync data to the media. HAMMER can only flush
1923 * the entire inode (so the TID remains properly synchronized).
1925 * Basically all we do here is place the bio on the inode's flush queue
1926 * and activate the flusher.
1930 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1938 ip = ap->a_vp->v_data;
1940 if (ip->flags & HAMMER_INODE_RO) {
1941 bp->b_error = EROFS;
1942 bp->b_flags |= B_ERROR;
1944 hammer_cleanup_write_io(ip);
1949 * Interlock with inode destruction (no in-kernel or directory
1950 * topology visibility). If we queue new IO while trying to
1951 * destroy the inode we can deadlock the vtrunc call in
1952 * hammer_inode_unloadable_check().
1954 if (ip->flags & (HAMMER_INODE_DELETING|HAMMER_INODE_DELETED)) {
1957 hammer_cleanup_write_io(ip);
1962 * If the inode is being flushed we cannot re-queue buffers
1963 * it may have already flushed, or it could result in duplicate
1964 * records in the database.
1967 if (ip->flags & HAMMER_INODE_WRITE_ALT)
1968 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1970 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1972 hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1974 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1979 * If the write was not part of an integrated flush operation then
1982 if (ip->flush_state != HAMMER_FST_FLUSH ||
1983 (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1984 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1991 * Backend code which actually performs the write to the media. This
1992 * routine is typically called from the flusher. The bio will be disposed
1993 * of (biodone'd) by this routine.
1995 * Iterate the related records and mark for deletion. If existing edge
1996 * records (left and right side) overlap our write they have to be marked
1997 * deleted and new records created, usually referencing a portion of the
1998 * original data. Then add a record to represent the buffer.
2001 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
2003 struct buf *bp = bio->bio_buf;
2006 KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
2009 * If the inode is going or gone, just throw away any frontend
2012 if (ip->flags & HAMMER_INODE_DELETED) {
2016 hammer_cleanup_write_io(ip);
2021 * Delete any records overlapping our range. This function will
2022 * (eventually) properly truncate partial overlaps.
2024 if (ip->sync_ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
2025 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
2028 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
2034 * Add a single record to cover the write. We can write a record
2035 * with only the actual file data - for example, a small 200 byte
2036 * file does not have to write out a 16K record.
2038 * While the data size does not have to be aligned, we still do it
2039 * to reduce fragmentation in a future allocation model.
2044 if (ip->sync_ino_data.size - bio->bio_offset >
2046 limit_size = bp->b_bufsize;
2048 limit_size = (int)(ip->sync_ino_data.size -
2050 KKASSERT(limit_size >= 0);
2051 limit_size = (limit_size + 63) & ~63;
2054 error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
2055 bp->b_data, limit_size);
2059 Debugger("hammer_dowrite: error");
2062 bp->b_resid = bp->b_bufsize;
2063 bp->b_error = error;
2064 bp->b_flags |= B_ERROR;
2070 hammer_cleanup_write_io(ip);
2075 hammer_cleanup_write_io(hammer_inode_t ip)
2077 if (ip->rsv_databufs) {
2079 --ip->hmp->rsv_databufs;
2084 * dounlink - disconnect a directory entry
2086 * XXX whiteout support not really in yet
2089 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
2090 struct vnode *dvp, struct ucred *cred, int flags)
2092 struct namecache *ncp;
2095 struct hammer_cursor cursor;
2100 * Calculate the namekey and setup the key range for the scan. This
2101 * works kinda like a chained hash table where the lower 32 bits
2102 * of the namekey synthesize the chain.
2104 * The key range is inclusive of both key_beg and key_end.
2109 if (dip->flags & HAMMER_INODE_RO)
2112 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
2114 hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
2115 cursor.key_beg.localization = HAMMER_LOCALIZE_MISC;
2116 cursor.key_beg.obj_id = dip->obj_id;
2117 cursor.key_beg.key = namekey;
2118 cursor.key_beg.create_tid = 0;
2119 cursor.key_beg.delete_tid = 0;
2120 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
2121 cursor.key_beg.obj_type = 0;
2123 cursor.key_end = cursor.key_beg;
2124 cursor.key_end.key |= 0xFFFFFFFFULL;
2125 cursor.asof = dip->obj_asof;
2126 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2129 * Scan all matching records (the chain), locate the one matching
2130 * the requested path component. info->last_error contains the
2131 * error code on search termination and could be 0, ENOENT, or
2134 * The hammer_ip_*() functions merge in-memory records with on-disk
2135 * records for the purposes of the search.
2137 error = hammer_ip_first(&cursor);
2139 while (error == 0) {
2140 error = hammer_ip_resolve_data(&cursor);
2143 nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
2145 if (ncp->nc_nlen == nlen &&
2146 bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
2149 error = hammer_ip_next(&cursor);
2153 * If all is ok we have to get the inode so we can adjust nlinks.
2155 * If the target is a directory, it must be empty.
2158 ip = hammer_get_inode(trans, &dip->cache[1],
2159 cursor.data->entry.obj_id,
2160 dip->hmp->asof, 0, &error);
2161 if (error == ENOENT) {
2162 kprintf("obj_id %016llx\n", cursor.data->entry.obj_id);
2163 Debugger("ENOENT unlinking object that should exist");
2167 * If we are trying to remove a directory the directory must
2170 * WARNING: hammer_ip_check_directory_empty() may have to
2171 * terminate the cursor to avoid a deadlock. It is ok to
2172 * call hammer_done_cursor() twice.
2174 if (error == 0 && ip->ino_data.obj_type ==
2175 HAMMER_OBJTYPE_DIRECTORY) {
2176 error = hammer_ip_check_directory_empty(trans, ip);
2180 * Delete the directory entry.
2182 * WARNING: hammer_ip_del_directory() may have to terminate
2183 * the cursor to avoid a deadlock. It is ok to call
2184 * hammer_done_cursor() twice.
2187 error = hammer_ip_del_directory(trans, &cursor,
2191 cache_setunresolved(nch);
2192 cache_setvp(nch, NULL);
2195 cache_inval_vp(ip->vp, CINV_DESTROY);
2197 hammer_rel_inode(ip, 0);
2199 hammer_done_cursor(&cursor);
2200 if (error == EDEADLK)
2206 /************************************************************************
2207 * FIFO AND SPECFS OPS *
2208 ************************************************************************
2213 hammer_vop_fifoclose (struct vop_close_args *ap)
2215 /* XXX update itimes */
2216 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2220 hammer_vop_fiforead (struct vop_read_args *ap)
2224 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2225 /* XXX update access time */
2230 hammer_vop_fifowrite (struct vop_write_args *ap)
2234 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2235 /* XXX update access time */
2240 hammer_vop_specclose (struct vop_close_args *ap)
2242 /* XXX update itimes */
2243 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2247 hammer_vop_specread (struct vop_read_args *ap)
2249 /* XXX update access time */
2250 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2254 hammer_vop_specwrite (struct vop_write_args *ap)
2256 /* XXX update last change time */
2257 return (VOCALL(&spec_vnode_vops, &ap->a_head));