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 1.45 2008/05/03 05:28:55 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 if (ap->a_waitfor == MNT_WAIT)
177 hammer_wait_inode(ip);
182 * hammer_vop_read { vp, uio, ioflag, cred }
186 hammer_vop_read(struct vop_read_args *ap)
188 struct hammer_transaction trans;
197 if (ap->a_vp->v_type != VREG)
201 seqcount = ap->a_ioflag >> 16;
203 hammer_start_transaction(&trans, ip->hmp);
206 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
209 while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_rec.ino_size) {
210 offset = uio->uio_offset & HAMMER_BUFMASK;
212 error = cluster_read(ap->a_vp, ip->ino_rec.ino_size,
213 uio->uio_offset - offset, HAMMER_BUFSIZE,
214 MAXBSIZE, seqcount, &bp);
216 error = bread(ap->a_vp, uio->uio_offset - offset,
217 HAMMER_BUFSIZE, &bp);
222 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
223 n = HAMMER_BUFSIZE - offset;
224 if (n > uio->uio_resid)
226 if (n > ip->ino_rec.ino_size - uio->uio_offset)
227 n = (int)(ip->ino_rec.ino_size - uio->uio_offset);
228 error = uiomove((char *)bp->b_data + offset, n, uio);
235 if ((ip->flags & HAMMER_INODE_RO) == 0 &&
236 (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
237 ip->ino_rec.ino_atime = trans.time;
238 hammer_modify_inode(&trans, ip, HAMMER_INODE_ITIMES);
240 hammer_done_transaction(&trans);
245 * hammer_vop_write { vp, uio, ioflag, cred }
249 hammer_vop_write(struct vop_write_args *ap)
251 struct hammer_transaction trans;
252 struct hammer_inode *ip;
261 if (ap->a_vp->v_type != VREG)
266 if (ip->flags & HAMMER_INODE_RO)
270 * Create a transaction to cover the operations we perform.
272 hammer_start_transaction(&trans, ip->hmp);
278 if (ap->a_ioflag & IO_APPEND)
279 uio->uio_offset = ip->ino_rec.ino_size;
282 * Check for illegal write offsets. Valid range is 0...2^63-1
284 if (uio->uio_offset < 0 || uio->uio_offset + uio->uio_resid <= 0) {
285 hammer_done_transaction(&trans);
290 * Access the data in HAMMER_BUFSIZE blocks via the buffer cache.
293 while (uio->uio_resid > 0) {
297 * Do not allow huge writes to deadlock the buffer cache
299 if ((++count & 15) == 0) {
301 if ((ap->a_ioflag & IO_NOBWILL) == 0)
303 vn_lock(ap->a_vp, LK_EXCLUSIVE|LK_RETRY);
306 offset = uio->uio_offset & HAMMER_BUFMASK;
307 n = HAMMER_BUFSIZE - offset;
308 if (n > uio->uio_resid)
310 if (uio->uio_offset + n > ip->ino_rec.ino_size) {
311 vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
315 if (uio->uio_segflg == UIO_NOCOPY) {
317 * Issuing a write with the same data backing the
318 * buffer. Instantiate the buffer to collect the
319 * backing vm pages, then read-in any missing bits.
321 * This case is used by vop_stdputpages().
323 bp = getblk(ap->a_vp, uio->uio_offset - offset,
324 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
325 if ((bp->b_flags & B_CACHE) == 0) {
327 error = bread(ap->a_vp,
328 uio->uio_offset - offset,
329 HAMMER_BUFSIZE, &bp);
331 } else if (offset == 0 && uio->uio_resid >= HAMMER_BUFSIZE) {
333 * Even though we are entirely overwriting the buffer
334 * we may still have to zero it out to avoid a
335 * mmap/write visibility issue.
337 bp = getblk(ap->a_vp, uio->uio_offset - offset,
338 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
339 if ((bp->b_flags & B_CACHE) == 0)
341 } else if (uio->uio_offset - offset >= ip->ino_rec.ino_size) {
343 * If the base offset of the buffer is beyond the
344 * file EOF, we don't have to issue a read.
346 bp = getblk(ap->a_vp, uio->uio_offset - offset,
347 HAMMER_BUFSIZE, GETBLK_BHEAVY, 0);
351 * Partial overwrite, read in any missing bits then
352 * replace the portion being written.
354 error = bread(ap->a_vp, uio->uio_offset - offset,
355 HAMMER_BUFSIZE, &bp);
360 error = uiomove((char *)bp->b_data + offset, n, uio);
363 * If we screwed up we have to undo any VM size changes we
369 vtruncbuf(ap->a_vp, ip->ino_rec.ino_size,
374 /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
375 if (ip->ino_rec.ino_size < uio->uio_offset) {
376 ip->ino_rec.ino_size = uio->uio_offset;
377 flags = HAMMER_INODE_RDIRTY;
378 vnode_pager_setsize(ap->a_vp, ip->ino_rec.ino_size);
382 ip->ino_rec.ino_mtime = trans.time;
383 flags |= HAMMER_INODE_ITIMES | HAMMER_INODE_BUFS;
384 hammer_modify_inode(&trans, ip, flags);
386 if (ap->a_ioflag & IO_SYNC) {
388 } else if (ap->a_ioflag & IO_DIRECT) {
391 } else if ((ap->a_ioflag >> 16) == IO_SEQMAX &&
392 (uio->uio_offset & HAMMER_BUFMASK) == 0) {
394 * XXX HAMMER can only fsync the whole inode,
395 * doing it on every buffer would be a bad idea.
398 * If seqcount indicates sequential operation and
399 * we just finished filling a buffer, push it out
400 * now to prevent the buffer cache from becoming
401 * too full, which would trigger non-optimal
410 hammer_done_transaction(&trans);
415 * hammer_vop_access { vp, mode, cred }
419 hammer_vop_access(struct vop_access_args *ap)
421 struct hammer_inode *ip = VTOI(ap->a_vp);
426 uid = hammer_to_unix_xid(&ip->ino_data.uid);
427 gid = hammer_to_unix_xid(&ip->ino_data.gid);
429 error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
430 ip->ino_data.uflags);
435 * hammer_vop_advlock { vp, id, op, fl, flags }
439 hammer_vop_advlock(struct vop_advlock_args *ap)
441 struct hammer_inode *ip = VTOI(ap->a_vp);
443 return (lf_advlock(ap, &ip->advlock, ip->ino_rec.ino_size));
447 * hammer_vop_close { vp, fflag }
451 hammer_vop_close(struct vop_close_args *ap)
453 return (vop_stdclose(ap));
457 * hammer_vop_ncreate { nch, dvp, vpp, cred, vap }
459 * The operating system has already ensured that the directory entry
460 * does not exist and done all appropriate namespace locking.
464 hammer_vop_ncreate(struct vop_ncreate_args *ap)
466 struct hammer_transaction trans;
467 struct hammer_inode *dip;
468 struct hammer_inode *nip;
469 struct nchandle *nch;
473 dip = VTOI(ap->a_dvp);
475 if (dip->flags & HAMMER_INODE_RO)
479 * Create a transaction to cover the operations we perform.
481 hammer_start_transaction(&trans, dip->hmp);
484 * Create a new filesystem object of the requested type. The
485 * returned inode will be referenced and shared-locked to prevent
486 * it from being moved to the flusher.
489 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
491 kprintf("hammer_create_inode error %d\n", error);
492 hammer_done_transaction(&trans);
498 * Add the new filesystem object to the directory. This will also
499 * bump the inode's link count.
501 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
503 kprintf("hammer_ip_add_directory error %d\n", error);
509 hammer_rel_inode(nip, 0);
510 hammer_done_transaction(&trans);
513 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
514 hammer_done_transaction(&trans);
515 hammer_rel_inode(nip, 0);
517 cache_setunresolved(ap->a_nch);
518 cache_setvp(ap->a_nch, *ap->a_vpp);
525 * hammer_vop_getattr { vp, vap }
529 hammer_vop_getattr(struct vop_getattr_args *ap)
531 struct hammer_inode *ip = VTOI(ap->a_vp);
532 struct vattr *vap = ap->a_vap;
535 if (cache_check_fsmid_vp(ap->a_vp, &ip->fsmid) &&
536 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0 &&
541 hammer_itimes(ap->a_vp);
544 vap->va_fsid = ip->hmp->fsid_udev;
545 vap->va_fileid = ip->ino_rec.base.base.obj_id;
546 vap->va_mode = ip->ino_data.mode;
547 vap->va_nlink = ip->ino_rec.ino_nlinks;
548 vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
549 vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
552 vap->va_size = ip->ino_rec.ino_size;
553 hammer_to_timespec(ip->ino_rec.ino_atime, &vap->va_atime);
554 hammer_to_timespec(ip->ino_rec.ino_mtime, &vap->va_mtime);
555 hammer_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
556 vap->va_flags = ip->ino_data.uflags;
557 vap->va_gen = 1; /* hammer inums are unique for all time */
558 vap->va_blocksize = HAMMER_BUFSIZE;
559 vap->va_bytes = (ip->ino_rec.ino_size + 63) & ~63;
560 vap->va_type = hammer_get_vnode_type(ip->ino_rec.base.base.obj_type);
561 vap->va_filerev = 0; /* XXX */
562 /* mtime uniquely identifies any adjustments made to the file */
563 vap->va_fsmid = ip->ino_rec.ino_mtime;
564 vap->va_uid_uuid = ip->ino_data.uid;
565 vap->va_gid_uuid = ip->ino_data.gid;
566 vap->va_fsid_uuid = ip->hmp->fsid;
567 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
570 switch (ip->ino_rec.base.base.obj_type) {
571 case HAMMER_OBJTYPE_CDEV:
572 case HAMMER_OBJTYPE_BDEV:
573 vap->va_rmajor = ip->ino_data.rmajor;
574 vap->va_rminor = ip->ino_data.rminor;
584 * hammer_vop_nresolve { nch, dvp, cred }
586 * Locate the requested directory entry.
590 hammer_vop_nresolve(struct vop_nresolve_args *ap)
592 struct hammer_transaction trans;
593 struct namecache *ncp;
597 struct hammer_cursor cursor;
598 union hammer_record_ondisk *rec;
608 * Misc initialization, plus handle as-of name extensions. Look for
609 * the '@@' extension. Note that as-of files and directories cannot
612 dip = VTOI(ap->a_dvp);
613 ncp = ap->a_nch->ncp;
614 asof = dip->obj_asof;
618 hammer_simple_transaction(&trans, dip->hmp);
620 for (i = 0; i < nlen; ++i) {
621 if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
622 asof = hammer_str_to_tid(ncp->nc_name + i + 2);
623 flags |= HAMMER_INODE_RO;
630 * If there is no path component the time extension is relative to
634 ip = hammer_get_inode(&trans, &dip->cache[1], dip->obj_id,
635 asof, flags, &error);
637 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
638 hammer_rel_inode(ip, 0);
644 cache_setvp(ap->a_nch, vp);
651 * Calculate the namekey and setup the key range for the scan. This
652 * works kinda like a chained hash table where the lower 32 bits
653 * of the namekey synthesize the chain.
655 * The key range is inclusive of both key_beg and key_end.
657 namekey = hammer_directory_namekey(ncp->nc_name, nlen);
659 error = hammer_init_cursor(&trans, &cursor, &dip->cache[0], dip);
660 cursor.key_beg.obj_id = dip->obj_id;
661 cursor.key_beg.key = namekey;
662 cursor.key_beg.create_tid = 0;
663 cursor.key_beg.delete_tid = 0;
664 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
665 cursor.key_beg.obj_type = 0;
667 cursor.key_end = cursor.key_beg;
668 cursor.key_end.key |= 0xFFFFFFFFULL;
670 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
673 * Scan all matching records (the chain), locate the one matching
674 * the requested path component.
676 * The hammer_ip_*() functions merge in-memory records with on-disk
677 * records for the purposes of the search.
683 error = hammer_ip_first(&cursor);
685 error = hammer_ip_resolve_data(&cursor);
689 if (nlen == rec->entry.base.data_len &&
690 bcmp(ncp->nc_name, cursor.data, nlen) == 0) {
691 obj_id = rec->entry.obj_id;
694 error = hammer_ip_next(&cursor);
697 hammer_done_cursor(&cursor);
699 ip = hammer_get_inode(&trans, &dip->cache[1],
700 obj_id, asof, flags, &error);
702 error = hammer_get_vnode(ip, LK_EXCLUSIVE, &vp);
703 hammer_rel_inode(ip, 0);
705 kprintf("nresolve: lookup %s failed dip %p (%016llx) on"
706 " inode %016llx error %d\n",
708 dip, dip->obj_id, obj_id, error);
714 cache_setvp(ap->a_nch, vp);
717 } else if (error == ENOENT) {
718 cache_setvp(ap->a_nch, NULL);
721 hammer_done_transaction(&trans);
726 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
728 * Locate the parent directory of a directory vnode.
730 * dvp is referenced but not locked. *vpp must be returned referenced and
731 * locked. A parent_obj_id of 0 does not necessarily indicate that we are
732 * at the root, instead it could indicate that the directory we were in was
735 * NOTE: as-of sequences are not linked into the directory structure. If
736 * we are at the root with a different asof then the mount point, reload
737 * the same directory with the mount point's asof. I'm not sure what this
738 * will do to NFS. We encode ASOF stamps in NFS file handles so it might not
739 * get confused, but it hasn't been tested.
743 hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
745 struct hammer_transaction trans;
746 struct hammer_inode *dip;
747 struct hammer_inode *ip;
748 int64_t parent_obj_id;
752 dip = VTOI(ap->a_dvp);
753 asof = dip->obj_asof;
754 parent_obj_id = dip->ino_data.parent_obj_id;
756 if (parent_obj_id == 0) {
757 if (dip->obj_id == HAMMER_OBJID_ROOT &&
758 asof != dip->hmp->asof) {
759 parent_obj_id = dip->obj_id;
760 asof = dip->hmp->asof;
761 *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
762 ksnprintf(*ap->a_fakename, 19, "0x%016llx",
770 hammer_simple_transaction(&trans, dip->hmp);
772 ip = hammer_get_inode(&trans, &dip->cache[1], parent_obj_id,
773 asof, dip->flags, &error);
775 error = hammer_get_vnode(ip, LK_EXCLUSIVE, ap->a_vpp);
776 hammer_rel_inode(ip, 0);
780 hammer_done_transaction(&trans);
785 * hammer_vop_nlink { nch, dvp, vp, cred }
789 hammer_vop_nlink(struct vop_nlink_args *ap)
791 struct hammer_transaction trans;
792 struct hammer_inode *dip;
793 struct hammer_inode *ip;
794 struct nchandle *nch;
798 dip = VTOI(ap->a_dvp);
801 if (dip->flags & HAMMER_INODE_RO)
803 if (ip->flags & HAMMER_INODE_RO)
807 * Create a transaction to cover the operations we perform.
809 hammer_start_transaction(&trans, dip->hmp);
812 * Add the filesystem object to the directory. Note that neither
813 * dip nor ip are referenced or locked, but their vnodes are
814 * referenced. This function will bump the inode's link count.
816 error = hammer_ip_add_directory(&trans, dip, nch->ncp, ip);
822 cache_setunresolved(nch);
823 cache_setvp(nch, ap->a_vp);
825 hammer_done_transaction(&trans);
830 * hammer_vop_nmkdir { nch, dvp, vpp, cred, vap }
832 * The operating system has already ensured that the directory entry
833 * does not exist and done all appropriate namespace locking.
837 hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
839 struct hammer_transaction trans;
840 struct hammer_inode *dip;
841 struct hammer_inode *nip;
842 struct nchandle *nch;
846 dip = VTOI(ap->a_dvp);
848 if (dip->flags & HAMMER_INODE_RO)
852 * Create a transaction to cover the operations we perform.
854 hammer_start_transaction(&trans, dip->hmp);
857 * Create a new filesystem object of the requested type. The
858 * returned inode will be referenced but not locked.
860 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
862 kprintf("hammer_mkdir error %d\n", error);
863 hammer_done_transaction(&trans);
868 * Add the new filesystem object to the directory. This will also
869 * bump the inode's link count.
871 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
873 kprintf("hammer_mkdir (add) error %d\n", error);
879 hammer_rel_inode(nip, 0);
882 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
883 hammer_rel_inode(nip, 0);
885 cache_setunresolved(ap->a_nch);
886 cache_setvp(ap->a_nch, *ap->a_vpp);
889 hammer_done_transaction(&trans);
894 * hammer_vop_nmknod { nch, dvp, vpp, cred, vap }
896 * The operating system has already ensured that the directory entry
897 * does not exist and done all appropriate namespace locking.
901 hammer_vop_nmknod(struct vop_nmknod_args *ap)
903 struct hammer_transaction trans;
904 struct hammer_inode *dip;
905 struct hammer_inode *nip;
906 struct nchandle *nch;
910 dip = VTOI(ap->a_dvp);
912 if (dip->flags & HAMMER_INODE_RO)
916 * Create a transaction to cover the operations we perform.
918 hammer_start_transaction(&trans, dip->hmp);
921 * Create a new filesystem object of the requested type. The
922 * returned inode will be referenced but not locked.
924 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
926 hammer_done_transaction(&trans);
932 * Add the new filesystem object to the directory. This will also
933 * bump the inode's link count.
935 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
941 hammer_rel_inode(nip, 0);
944 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
945 hammer_rel_inode(nip, 0);
947 cache_setunresolved(ap->a_nch);
948 cache_setvp(ap->a_nch, *ap->a_vpp);
951 hammer_done_transaction(&trans);
956 * hammer_vop_open { vp, mode, cred, fp }
960 hammer_vop_open(struct vop_open_args *ap)
962 if ((ap->a_mode & FWRITE) && (VTOI(ap->a_vp)->flags & HAMMER_INODE_RO))
965 return(vop_stdopen(ap));
969 * hammer_vop_pathconf { vp, name, retval }
973 hammer_vop_pathconf(struct vop_pathconf_args *ap)
979 * hammer_vop_print { vp }
983 hammer_vop_print(struct vop_print_args *ap)
989 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
993 hammer_vop_readdir(struct vop_readdir_args *ap)
995 struct hammer_transaction trans;
996 struct hammer_cursor cursor;
997 struct hammer_inode *ip;
999 hammer_record_ondisk_t rec;
1000 hammer_base_elm_t base;
1008 ip = VTOI(ap->a_vp);
1010 saveoff = uio->uio_offset;
1012 if (ap->a_ncookies) {
1013 ncookies = uio->uio_resid / 16 + 1;
1014 if (ncookies > 1024)
1016 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
1024 hammer_simple_transaction(&trans, ip->hmp);
1027 * Handle artificial entries
1031 r = vop_write_dirent(&error, uio, ip->obj_id, DT_DIR, 1, ".");
1035 cookies[cookie_index] = saveoff;
1038 if (cookie_index == ncookies)
1042 if (ip->ino_data.parent_obj_id) {
1043 r = vop_write_dirent(&error, uio,
1044 ip->ino_data.parent_obj_id,
1047 r = vop_write_dirent(&error, uio,
1048 ip->obj_id, DT_DIR, 2, "..");
1053 cookies[cookie_index] = saveoff;
1056 if (cookie_index == ncookies)
1061 * Key range (begin and end inclusive) to scan. Directory keys
1062 * directly translate to a 64 bit 'seek' position.
1064 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1065 cursor.key_beg.obj_id = ip->obj_id;
1066 cursor.key_beg.create_tid = 0;
1067 cursor.key_beg.delete_tid = 0;
1068 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1069 cursor.key_beg.obj_type = 0;
1070 cursor.key_beg.key = saveoff;
1072 cursor.key_end = cursor.key_beg;
1073 cursor.key_end.key = HAMMER_MAX_KEY;
1074 cursor.asof = ip->obj_asof;
1075 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1077 error = hammer_ip_first(&cursor);
1079 while (error == 0) {
1080 error = hammer_ip_resolve_record_and_data(&cursor);
1083 rec = cursor.record;
1084 base = &rec->base.base;
1085 saveoff = base->key;
1087 if (base->obj_id != ip->obj_id)
1088 panic("readdir: bad record at %p", cursor.node);
1090 r = vop_write_dirent(
1091 &error, uio, rec->entry.obj_id,
1092 hammer_get_dtype(rec->entry.base.base.obj_type),
1093 rec->entry.base.data_len,
1094 (void *)cursor.data);
1099 cookies[cookie_index] = base->key;
1101 if (cookie_index == ncookies)
1103 error = hammer_ip_next(&cursor);
1105 hammer_done_cursor(&cursor);
1108 hammer_done_transaction(&trans);
1111 *ap->a_eofflag = (error == ENOENT);
1112 uio->uio_offset = saveoff;
1113 if (error && cookie_index == 0) {
1114 if (error == ENOENT)
1117 kfree(cookies, M_TEMP);
1118 *ap->a_ncookies = 0;
1119 *ap->a_cookies = NULL;
1122 if (error == ENOENT)
1125 *ap->a_ncookies = cookie_index;
1126 *ap->a_cookies = cookies;
1133 * hammer_vop_readlink { vp, uio, cred }
1137 hammer_vop_readlink(struct vop_readlink_args *ap)
1139 struct hammer_transaction trans;
1140 struct hammer_cursor cursor;
1141 struct hammer_inode *ip;
1144 ip = VTOI(ap->a_vp);
1146 hammer_simple_transaction(&trans, ip->hmp);
1148 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1151 * Key range (begin and end inclusive) to scan. Directory keys
1152 * directly translate to a 64 bit 'seek' position.
1154 cursor.key_beg.obj_id = ip->obj_id;
1155 cursor.key_beg.create_tid = 0;
1156 cursor.key_beg.delete_tid = 0;
1157 cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
1158 cursor.key_beg.obj_type = 0;
1159 cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
1160 cursor.asof = ip->obj_asof;
1161 cursor.flags |= HAMMER_CURSOR_ASOF;
1163 error = hammer_ip_lookup(&cursor, ip);
1165 error = hammer_ip_resolve_data(&cursor);
1167 error = uiomove((char *)cursor.data,
1168 cursor.record->base.data_len,
1172 hammer_done_cursor(&cursor);
1173 hammer_done_transaction(&trans);
1178 * hammer_vop_nremove { nch, dvp, cred }
1182 hammer_vop_nremove(struct vop_nremove_args *ap)
1184 struct hammer_transaction trans;
1187 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1188 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1189 hammer_done_transaction(&trans);
1195 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1199 hammer_vop_nrename(struct vop_nrename_args *ap)
1201 struct hammer_transaction trans;
1202 struct namecache *fncp;
1203 struct namecache *tncp;
1204 struct hammer_inode *fdip;
1205 struct hammer_inode *tdip;
1206 struct hammer_inode *ip;
1207 struct hammer_cursor cursor;
1208 union hammer_record_ondisk *rec;
1212 fdip = VTOI(ap->a_fdvp);
1213 tdip = VTOI(ap->a_tdvp);
1214 fncp = ap->a_fnch->ncp;
1215 tncp = ap->a_tnch->ncp;
1216 ip = VTOI(fncp->nc_vp);
1217 KKASSERT(ip != NULL);
1219 if (fdip->flags & HAMMER_INODE_RO)
1221 if (tdip->flags & HAMMER_INODE_RO)
1223 if (ip->flags & HAMMER_INODE_RO)
1226 hammer_start_transaction(&trans, fdip->hmp);
1229 * Remove tncp from the target directory and then link ip as
1230 * tncp. XXX pass trans to dounlink
1232 * Force the inode sync-time to match the transaction so it is
1233 * in-sync with the creation of the target directory entry.
1235 error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp, ap->a_cred, 0);
1236 if (error == 0 || error == ENOENT) {
1237 error = hammer_ip_add_directory(&trans, tdip, tncp, ip);
1239 ip->ino_data.parent_obj_id = tdip->obj_id;
1240 hammer_modify_inode(&trans, ip, HAMMER_INODE_DDIRTY);
1244 goto failed; /* XXX */
1247 * Locate the record in the originating directory and remove it.
1249 * Calculate the namekey and setup the key range for the scan. This
1250 * works kinda like a chained hash table where the lower 32 bits
1251 * of the namekey synthesize the chain.
1253 * The key range is inclusive of both key_beg and key_end.
1255 namekey = hammer_directory_namekey(fncp->nc_name, fncp->nc_nlen);
1257 hammer_init_cursor(&trans, &cursor, &fdip->cache[0], fdip);
1258 cursor.key_beg.obj_id = fdip->obj_id;
1259 cursor.key_beg.key = namekey;
1260 cursor.key_beg.create_tid = 0;
1261 cursor.key_beg.delete_tid = 0;
1262 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1263 cursor.key_beg.obj_type = 0;
1265 cursor.key_end = cursor.key_beg;
1266 cursor.key_end.key |= 0xFFFFFFFFULL;
1267 cursor.asof = fdip->obj_asof;
1268 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
1271 * Scan all matching records (the chain), locate the one matching
1272 * the requested path component.
1274 * The hammer_ip_*() functions merge in-memory records with on-disk
1275 * records for the purposes of the search.
1277 error = hammer_ip_first(&cursor);
1278 while (error == 0) {
1279 if (hammer_ip_resolve_data(&cursor) != 0)
1281 rec = cursor.record;
1282 if (fncp->nc_nlen == rec->entry.base.data_len &&
1283 bcmp(fncp->nc_name, cursor.data, fncp->nc_nlen) == 0) {
1286 error = hammer_ip_next(&cursor);
1290 * If all is ok we have to get the inode so we can adjust nlinks.
1292 * WARNING: hammer_ip_del_directory() may have to terminate the
1293 * cursor to avoid a recursion. It's ok to call hammer_done_cursor()
1297 error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);
1300 * XXX A deadlock here will break rename's atomicy for the purposes
1301 * of crash recovery.
1303 if (error == EDEADLK) {
1304 hammer_done_cursor(&cursor);
1309 * Cleanup and tell the kernel that the rename succeeded.
1311 hammer_done_cursor(&cursor);
1313 cache_rename(ap->a_fnch, ap->a_tnch);
1316 hammer_done_transaction(&trans);
1321 * hammer_vop_nrmdir { nch, dvp, cred }
1325 hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
1327 struct hammer_transaction trans;
1330 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1331 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0);
1332 hammer_done_transaction(&trans);
1338 * hammer_vop_setattr { vp, vap, cred }
1342 hammer_vop_setattr(struct vop_setattr_args *ap)
1344 struct hammer_transaction trans;
1346 struct hammer_inode *ip;
1355 ip = ap->a_vp->v_data;
1358 if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
1360 if (ip->flags & HAMMER_INODE_RO)
1363 hammer_start_transaction(&trans, ip->hmp);
1366 if (vap->va_flags != VNOVAL) {
1367 flags = ip->ino_data.uflags;
1368 error = vop_helper_setattr_flags(&flags, vap->va_flags,
1369 hammer_to_unix_xid(&ip->ino_data.uid),
1372 if (ip->ino_data.uflags != flags) {
1373 ip->ino_data.uflags = flags;
1374 modflags |= HAMMER_INODE_DDIRTY;
1376 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1383 if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
1387 if (vap->va_uid != (uid_t)VNOVAL) {
1388 hammer_guid_to_uuid(&uuid, vap->va_uid);
1389 if (bcmp(&uuid, &ip->ino_data.uid, sizeof(uuid)) != 0) {
1390 ip->ino_data.uid = uuid;
1391 modflags |= HAMMER_INODE_DDIRTY;
1394 if (vap->va_gid != (uid_t)VNOVAL) {
1395 hammer_guid_to_uuid(&uuid, vap->va_gid);
1396 if (bcmp(&uuid, &ip->ino_data.gid, sizeof(uuid)) != 0) {
1397 ip->ino_data.gid = uuid;
1398 modflags |= HAMMER_INODE_DDIRTY;
1401 while (vap->va_size != VNOVAL && ip->ino_rec.ino_size != vap->va_size) {
1402 switch(ap->a_vp->v_type) {
1404 if (vap->va_size == ip->ino_rec.ino_size)
1407 * XXX break atomicy, we can deadlock the backend
1408 * if we do not release the lock. Probably not a
1411 if (vap->va_size < ip->ino_rec.ino_size) {
1412 vtruncbuf(ap->a_vp, vap->va_size,
1416 vnode_pager_setsize(ap->a_vp, vap->va_size);
1419 ip->ino_rec.ino_size = vap->va_size;
1420 modflags |= HAMMER_INODE_RDIRTY;
1421 aligned_size = (vap->va_size + HAMMER_BUFMASK) &
1425 * on-media truncation is cached in the inode until
1426 * the inode is synchronized.
1429 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1430 ip->flags |= HAMMER_INODE_TRUNCATED;
1431 ip->trunc_off = vap->va_size;
1432 } else if (ip->trunc_off > vap->va_size) {
1433 ip->trunc_off = vap->va_size;
1438 * If truncating we have to clean out a portion of
1439 * the last block on-disk. We do this in the
1440 * front-end buffer cache.
1442 if (truncating && vap->va_size < aligned_size) {
1446 offset = vap->va_size & HAMMER_BUFMASK;
1447 error = bread(ap->a_vp,
1448 aligned_size - HAMMER_BUFSIZE,
1449 HAMMER_BUFSIZE, &bp);
1451 bzero(bp->b_data + offset,
1452 HAMMER_BUFSIZE - offset);
1460 if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
1461 ip->flags |= HAMMER_INODE_TRUNCATED;
1462 ip->trunc_off = vap->va_size;
1463 } else if (ip->trunc_off > vap->va_size) {
1464 ip->trunc_off = vap->va_size;
1466 ip->ino_rec.ino_size = vap->va_size;
1467 modflags |= HAMMER_INODE_RDIRTY;
1475 if (vap->va_atime.tv_sec != VNOVAL) {
1476 ip->ino_rec.ino_atime =
1477 hammer_timespec_to_transid(&vap->va_atime);
1478 modflags |= HAMMER_INODE_ITIMES;
1480 if (vap->va_mtime.tv_sec != VNOVAL) {
1481 ip->ino_rec.ino_mtime =
1482 hammer_timespec_to_transid(&vap->va_mtime);
1483 modflags |= HAMMER_INODE_ITIMES;
1485 if (vap->va_mode != (mode_t)VNOVAL) {
1486 if (ip->ino_data.mode != vap->va_mode) {
1487 ip->ino_data.mode = vap->va_mode;
1488 modflags |= HAMMER_INODE_DDIRTY;
1493 hammer_modify_inode(&trans, ip, modflags);
1494 hammer_done_transaction(&trans);
1499 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1503 hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
1505 struct hammer_transaction trans;
1506 struct hammer_inode *dip;
1507 struct hammer_inode *nip;
1508 struct nchandle *nch;
1509 hammer_record_t record;
1513 ap->a_vap->va_type = VLNK;
1516 dip = VTOI(ap->a_dvp);
1518 if (dip->flags & HAMMER_INODE_RO)
1522 * Create a transaction to cover the operations we perform.
1524 hammer_start_transaction(&trans, dip->hmp);
1527 * Create a new filesystem object of the requested type. The
1528 * returned inode will be referenced but not locked.
1531 error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred, dip, &nip);
1533 hammer_done_transaction(&trans);
1539 * Add a record representing the symlink. symlink stores the link
1540 * as pure data, not a string, and is no \0 terminated.
1543 record = hammer_alloc_mem_record(nip);
1544 bytes = strlen(ap->a_target);
1546 record->rec.base.base.key = HAMMER_FIXKEY_SYMLINK;
1547 record->rec.base.base.rec_type = HAMMER_RECTYPE_FIX;
1548 record->rec.base.data_len = bytes;
1549 record->data = (void *)ap->a_target;
1550 /* will be reallocated by routine below */
1551 error = hammer_ip_add_record(&trans, record);
1554 * Set the file size to the length of the link.
1557 nip->ino_rec.ino_size = bytes;
1558 hammer_modify_inode(&trans, nip, HAMMER_INODE_RDIRTY);
1562 error = hammer_ip_add_directory(&trans, dip, nch->ncp, nip);
1568 hammer_rel_inode(nip, 0);
1571 error = hammer_get_vnode(nip, LK_EXCLUSIVE, ap->a_vpp);
1572 hammer_rel_inode(nip, 0);
1574 cache_setunresolved(ap->a_nch);
1575 cache_setvp(ap->a_nch, *ap->a_vpp);
1578 hammer_done_transaction(&trans);
1583 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
1587 hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
1589 struct hammer_transaction trans;
1592 hammer_start_transaction(&trans, VTOI(ap->a_dvp)->hmp);
1593 error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
1594 ap->a_cred, ap->a_flags);
1595 hammer_done_transaction(&trans);
1601 * hammer_vop_ioctl { vp, command, data, fflag, cred }
1605 hammer_vop_ioctl(struct vop_ioctl_args *ap)
1607 struct hammer_inode *ip = ap->a_vp->v_data;
1609 return(hammer_ioctl(ip, ap->a_command, ap->a_data,
1610 ap->a_fflag, ap->a_cred));
1615 hammer_vop_mountctl(struct vop_mountctl_args *ap)
1620 mp = ap->a_head.a_ops->head.vv_mount;
1623 case MOUNTCTL_SET_EXPORT:
1624 if (ap->a_ctllen != sizeof(struct export_args))
1626 error = hammer_vfs_export(mp, ap->a_op,
1627 (const struct export_args *)ap->a_ctl);
1630 error = journal_mountctl(ap);
1637 * hammer_vop_strategy { vp, bio }
1639 * Strategy call, used for regular file read & write only. Note that the
1640 * bp may represent a cluster.
1642 * To simplify operation and allow better optimizations in the future,
1643 * this code does not make any assumptions with regards to buffer alignment
1648 hammer_vop_strategy(struct vop_strategy_args *ap)
1653 bp = ap->a_bio->bio_buf;
1657 error = hammer_vop_strategy_read(ap);
1660 error = hammer_vop_strategy_write(ap);
1663 bp->b_error = error = EINVAL;
1664 bp->b_flags |= B_ERROR;
1672 * Read from a regular file. Iterate the related records and fill in the
1673 * BIO/BUF. Gaps are zero-filled.
1675 * The support code in hammer_object.c should be used to deal with mixed
1676 * in-memory and on-disk records.
1682 hammer_vop_strategy_read(struct vop_strategy_args *ap)
1684 struct hammer_transaction trans;
1685 struct hammer_inode *ip;
1686 struct hammer_cursor cursor;
1687 hammer_record_ondisk_t rec;
1688 hammer_base_elm_t base;
1701 ip = ap->a_vp->v_data;
1703 hammer_simple_transaction(&trans, ip->hmp);
1704 hammer_init_cursor(&trans, &cursor, &ip->cache[0], ip);
1707 * Key range (begin and end inclusive) to scan. Note that the key's
1708 * stored in the actual records represent BASE+LEN, not BASE. The
1709 * first record containing bio_offset will have a key > bio_offset.
1711 cursor.key_beg.obj_id = ip->obj_id;
1712 cursor.key_beg.create_tid = 0;
1713 cursor.key_beg.delete_tid = 0;
1714 cursor.key_beg.obj_type = 0;
1715 cursor.key_beg.key = bio->bio_offset + 1;
1716 cursor.asof = ip->obj_asof;
1717 cursor.flags |= HAMMER_CURSOR_ASOF | HAMMER_CURSOR_DATAEXTOK;
1719 cursor.key_end = cursor.key_beg;
1720 KKASSERT(ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_REGFILE);
1722 if (ip->ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1723 cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
1724 cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
1725 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1729 ran_end = bio->bio_offset + bp->b_bufsize;
1730 cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
1731 cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
1732 tmp64 = ran_end + MAXPHYS + 1; /* work-around GCC-4 bug */
1733 if (tmp64 < ran_end)
1734 cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
1736 cursor.key_end.key = ran_end + MAXPHYS + 1;
1738 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;
1740 error = hammer_ip_first(&cursor);
1743 while (error == 0) {
1744 error = hammer_ip_resolve_data(&cursor);
1747 rec = cursor.record;
1748 base = &rec->base.base;
1750 rec_offset = base->key - rec->data.base.data_len;
1753 * Calculate the gap, if any, and zero-fill it.
1755 n = (int)(rec_offset - (bio->bio_offset + boff));
1757 if (n > bp->b_bufsize - boff)
1758 n = bp->b_bufsize - boff;
1759 bzero((char *)bp->b_data + boff, n);
1765 * Calculate the data offset in the record and the number
1766 * of bytes we can copy.
1768 * Note there is a degenerate case here where boff may
1769 * already be at bp->b_bufsize.
1773 n = rec->data.base.data_len - roff;
1775 if (n > bp->b_bufsize - boff)
1776 n = bp->b_bufsize - boff;
1779 * If we cached a truncation point on our front-end the
1780 * on-disk version may still have physical records beyond
1781 * that point. Truncate visibility.
1783 if (ip->trunc_off <= rec_offset)
1785 else if (ip->trunc_off < rec_offset + n)
1786 n = (int)(ip->trunc_off - rec_offset);
1792 bcopy((char *)cursor.data + roff,
1793 (char *)bp->b_data + boff, n);
1796 if (boff == bp->b_bufsize)
1798 error = hammer_ip_next(&cursor);
1800 hammer_done_cursor(&cursor);
1801 hammer_done_transaction(&trans);
1804 * There may have been a gap after the last record
1806 if (error == ENOENT)
1808 if (error == 0 && boff != bp->b_bufsize) {
1809 KKASSERT(boff < bp->b_bufsize);
1810 bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
1811 /* boff = bp->b_bufsize; */
1814 bp->b_error = error;
1816 bp->b_flags |= B_ERROR;
1822 * Write to a regular file. Because this is a strategy call the OS is
1823 * trying to actually sync data to the media. HAMMER can only flush
1824 * the entire inode (so the TID remains properly synchronized).
1826 * Basically all we do here is place the bio on the inode's flush queue
1827 * and activate the flusher.
1831 hammer_vop_strategy_write(struct vop_strategy_args *ap)
1839 ip = ap->a_vp->v_data;
1841 if (ip->flags & HAMMER_INODE_RO) {
1842 bp->b_error = EROFS;
1843 bp->b_flags |= B_ERROR;
1849 * If the inode is being flushed we cannot re-queue buffers
1850 * it may have already flushed, or it could result in duplicate
1851 * records in the database.
1854 if (ip->flags & HAMMER_INODE_WRITE_ALT)
1855 TAILQ_INSERT_TAIL(&ip->bio_alt_list, bio, bio_act);
1857 TAILQ_INSERT_TAIL(&ip->bio_list, bio, bio_act);
1859 hammer_modify_inode(NULL, ip, HAMMER_INODE_BUFS);
1861 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1866 * If the write was not part of an integrated flush operation then
1869 if (ip->flush_state != HAMMER_FST_FLUSH ||
1870 (ip->flags & HAMMER_INODE_WRITE_ALT)) {
1871 hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
1878 * Backend code which actually performs the write to the media. This
1879 * routine is typically called from the flusher. The bio will be disposed
1880 * of (biodone'd) by this routine.
1882 * Iterate the related records and mark for deletion. If existing edge
1883 * records (left and right side) overlap our write they have to be marked
1884 * deleted and new records created, usually referencing a portion of the
1885 * original data. Then add a record to represent the buffer.
1888 hammer_dowrite(hammer_cursor_t cursor, hammer_inode_t ip, struct bio *bio)
1890 struct buf *bp = bio->bio_buf;
1893 KKASSERT(ip->flush_state == HAMMER_FST_FLUSH);
1896 * If the inode is going or gone, just throw away any frontend
1899 if (ip->flags & HAMMER_INODE_DELETED) {
1905 * Delete any records overlapping our range. This function will
1906 * (eventually) properly truncate partial overlaps.
1908 if (ip->sync_ino_rec.base.base.obj_type == HAMMER_OBJTYPE_DBFILE) {
1909 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1912 error = hammer_ip_delete_range(cursor, ip, bio->bio_offset,
1918 * Add a single record to cover the write. We can write a record
1919 * with only the actual file data - for example, a small 200 byte
1920 * file does not have to write out a 16K record.
1922 * While the data size does not have to be aligned, we still do it
1923 * to reduce fragmentation in a future allocation model.
1928 if (ip->sync_ino_rec.ino_size - bio->bio_offset >
1930 limit_size = bp->b_bufsize;
1932 limit_size = (int)(ip->sync_ino_rec.ino_size -
1934 KKASSERT(limit_size >= 0);
1935 limit_size = (limit_size + 63) & ~63;
1938 error = hammer_ip_sync_data(cursor, ip, bio->bio_offset,
1939 bp->b_data, limit_size);
1943 Debugger("hammer_dowrite: error");
1946 bp->b_resid = bp->b_bufsize;
1947 bp->b_error = error;
1948 bp->b_flags |= B_ERROR;
1958 * dounlink - disconnect a directory entry
1960 * XXX whiteout support not really in yet
1963 hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
1964 struct vnode *dvp, struct ucred *cred, int flags)
1966 struct namecache *ncp;
1969 hammer_record_ondisk_t rec;
1970 struct hammer_cursor cursor;
1975 * Calculate the namekey and setup the key range for the scan. This
1976 * works kinda like a chained hash table where the lower 32 bits
1977 * of the namekey synthesize the chain.
1979 * The key range is inclusive of both key_beg and key_end.
1984 if (dip->flags & HAMMER_INODE_RO)
1987 namekey = hammer_directory_namekey(ncp->nc_name, ncp->nc_nlen);
1989 hammer_init_cursor(trans, &cursor, &dip->cache[0], dip);
1990 cursor.key_beg.obj_id = dip->obj_id;
1991 cursor.key_beg.key = namekey;
1992 cursor.key_beg.create_tid = 0;
1993 cursor.key_beg.delete_tid = 0;
1994 cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
1995 cursor.key_beg.obj_type = 0;
1997 cursor.key_end = cursor.key_beg;
1998 cursor.key_end.key |= 0xFFFFFFFFULL;
1999 cursor.asof = dip->obj_asof;
2000 cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;
2003 * Scan all matching records (the chain), locate the one matching
2004 * the requested path component. info->last_error contains the
2005 * error code on search termination and could be 0, ENOENT, or
2008 * The hammer_ip_*() functions merge in-memory records with on-disk
2009 * records for the purposes of the search.
2012 error = hammer_ip_first(&cursor);
2014 while (error == 0) {
2015 error = hammer_ip_resolve_data(&cursor);
2018 rec = cursor.record;
2019 if (ncp->nc_nlen == rec->entry.base.data_len &&
2020 bcmp(ncp->nc_name, cursor.data, ncp->nc_nlen) == 0) {
2023 error = hammer_ip_next(&cursor);
2027 * If all is ok we have to get the inode so we can adjust nlinks.
2029 * If the target is a directory, it must be empty.
2032 ip = hammer_get_inode(trans, &dip->cache[1],
2034 dip->hmp->asof, 0, &error);
2035 if (error == ENOENT) {
2036 kprintf("obj_id %016llx\n", rec->entry.obj_id);
2037 Debugger("ENOENT unlinking object that should exist");
2041 * If we are trying to remove a directory the directory must
2044 * WARNING: hammer_ip_check_directory_empty() may have to
2045 * terminate the cursor to avoid a deadlock. It is ok to
2046 * call hammer_done_cursor() twice.
2048 if (error == 0 && ip->ino_rec.base.base.obj_type ==
2049 HAMMER_OBJTYPE_DIRECTORY) {
2050 error = hammer_ip_check_directory_empty(trans, &cursor,
2055 * Delete the directory entry.
2057 * WARNING: hammer_ip_del_directory() may have to terminate
2058 * the cursor to avoid a deadlock. It is ok to call
2059 * hammer_done_cursor() twice.
2062 error = hammer_ip_del_directory(trans, &cursor,
2066 cache_setunresolved(nch);
2067 cache_setvp(nch, NULL);
2070 cache_inval_vp(ip->vp, CINV_DESTROY);
2072 hammer_rel_inode(ip, 0);
2074 hammer_done_cursor(&cursor);
2075 if (error == EDEADLK)
2081 /************************************************************************
2082 * FIFO AND SPECFS OPS *
2083 ************************************************************************
2088 hammer_vop_fifoclose (struct vop_close_args *ap)
2090 /* XXX update itimes */
2091 return (VOCALL(&fifo_vnode_vops, &ap->a_head));
2095 hammer_vop_fiforead (struct vop_read_args *ap)
2099 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2100 /* XXX update access time */
2105 hammer_vop_fifowrite (struct vop_write_args *ap)
2109 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2110 /* XXX update access time */
2115 hammer_vop_specclose (struct vop_close_args *ap)
2117 /* XXX update itimes */
2118 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2122 hammer_vop_specread (struct vop_read_args *ap)
2124 /* XXX update access time */
2125 return (VOCALL(&spec_vnode_vops, &ap->a_head));
2129 hammer_vop_specwrite (struct vop_write_args *ap)
2131 /* XXX update last change time */
2132 return (VOCALL(&spec_vnode_vops, &ap->a_head));