2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
7 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * Kernel Filesystem interface
39 * NOTE! local ipdata pointers must be reloaded on any modifying operation
40 * to the inode as its underlying chain may have changed.
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/kernel.h>
46 #include <sys/fcntl.h>
49 #include <sys/namei.h>
50 #include <sys/mount.h>
51 #include <sys/vnode.h>
52 #include <sys/mountctl.h>
53 #include <sys/dirent.h>
55 #include <sys/objcache.h>
56 #include <sys/event.h>
58 #include <vfs/fifofs/fifo.h>
62 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
64 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
65 int ioflag, int seqcount);
66 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
67 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
69 struct objcache *cache_xops;
73 hammer2_knote(struct vnode *vp, int flags)
76 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
80 * Last reference to a vnode is going away but it is still cached.
84 hammer2_vop_inactive(struct vop_inactive_args *ap)
103 * Check for deleted inodes and recycle immediately on the last
104 * release. Be sure to destroy any left-over buffer cache buffers
105 * so we do not waste time trying to flush them.
107 * Note that deleting the file block chains under the inode chain
108 * would just be a waste of energy, so don't do it.
110 * WARNING: nvtruncbuf() can only be safely called without the inode
111 * lock held due to the way our write thread works.
113 if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
118 * Detect updates to the embedded data which may be
119 * synchronized by the strategy code. Simply mark the
120 * inode modified so it gets picked up by our normal flush.
122 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
123 nvtruncbuf(vp, 0, nblksize, 0, 0);
131 * Reclaim a vnode so that it can be reused; after the inode is
132 * disassociated, the filesystem must manage it alone.
136 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
152 * The final close of a deleted file or directory marks it for
153 * destruction. The DELETED flag allows the flusher to shortcut
154 * any modified blocks still unflushed (that is, just ignore them).
156 * HAMMER2 usually does not try to optimize the freemap by returning
157 * deleted blocks to it as it does not usually know how many snapshots
158 * might be referencing portions of the file/dir.
164 * NOTE! We do not attempt to flush chains here, flushing is
165 * really fragile and could also deadlock.
170 * An unlinked inode may have been relinked to the ihidden directory.
171 * This occurs if the inode was unlinked while open. Reclamation of
172 * these inodes requires processing we cannot safely do here so add
173 * the inode to the sideq in that situation.
175 * A modified inode may require chain synchronization which will no
176 * longer be driven by a sync or fsync without the vnode, also use
177 * the sideq for that.
179 * A reclaim can occur at any time so we cannot safely start a
180 * transaction to handle reclamation of unlinked files. Instead,
181 * the ip is left with a reference and placed on a linked list and
185 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
186 HAMMER2_INODE_MODIFIED |
187 HAMMER2_INODE_RESIZED)) &&
188 (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
189 hammer2_inode_sideq_t *ipul;
191 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
194 hammer2_spin_ex(&pmp->list_spin);
195 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
197 atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
198 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
199 hammer2_spin_unex(&pmp->list_spin);
201 hammer2_spin_unex(&pmp->list_spin);
202 kfree(ipul, pmp->minode);
203 hammer2_inode_drop(ip); /* vp ref */
205 /* retain ref from vp for ipul */
207 hammer2_inode_drop(ip); /* vp ref */
211 * XXX handle background sync when ip dirty, kernel will no longer
212 * notify us regarding this inode because there is no longer a
213 * vnode attached to it.
222 hammer2_vop_fsync(struct vop_fsync_args *ap)
232 /* XXX can't do this yet */
233 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_ISFLUSH);
234 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
236 hammer2_trans_init(ip->pmp, 0);
237 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
240 * Calling chain_flush here creates a lot of duplicative
241 * COW operations due to non-optimal vnode ordering.
243 * Only do it for an actual fsync() syscall. The other forms
244 * which call this function will eventually call chain_flush
245 * on the volume root as a catch-all, which is far more optimal.
247 hammer2_inode_lock(ip, 0);
248 if (ip->flags & HAMMER2_INODE_MODIFIED)
249 hammer2_inode_chain_sync(ip);
250 hammer2_inode_unlock(ip);
251 hammer2_trans_done(ip->pmp);
259 hammer2_vop_access(struct vop_access_args *ap)
261 hammer2_inode_t *ip = VTOI(ap->a_vp);
267 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
268 uid = hammer2_to_unix_xid(&ip->meta.uid);
269 gid = hammer2_to_unix_xid(&ip->meta.gid);
270 error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
271 hammer2_inode_unlock(ip);
279 hammer2_vop_getattr(struct vop_getattr_args *ap)
285 hammer2_chain_t *chain;
295 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
297 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
298 vap->va_fileid = ip->meta.inum;
299 vap->va_mode = ip->meta.mode;
300 vap->va_nlink = ip->meta.nlinks;
301 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
302 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
305 vap->va_size = ip->meta.size; /* protected by shared lock */
306 vap->va_blocksize = HAMMER2_PBUFSIZE;
307 vap->va_flags = ip->meta.uflags;
308 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
309 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
310 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
313 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
315 * Can't really calculate directory use sans the files under
316 * it, just assume one block for now.
318 vap->va_bytes += HAMMER2_INODE_BYTES;
320 for (i = 0; i < ip->cluster.nchains; ++i) {
321 if ((chain = ip->cluster.array[i].chain) != NULL) {
322 if (vap->va_bytes < chain->bref.data_count)
323 vap->va_bytes = chain->bref.data_count;
327 vap->va_type = hammer2_get_vtype(ip->meta.type);
329 vap->va_uid_uuid = ip->meta.uid;
330 vap->va_gid_uuid = ip->meta.gid;
331 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
334 hammer2_inode_unlock(ip);
342 hammer2_vop_setattr(struct vop_setattr_args *ap)
354 hammer2_update_time(&ctime);
358 if (ip->pmp->ronly) {
363 hammer2_pfs_memory_wait(ip->pmp);
364 hammer2_trans_init(ip->pmp, 0);
365 hammer2_inode_lock(ip, 0);
368 if (vap->va_flags != VNOVAL) {
371 flags = ip->meta.uflags;
372 error = vop_helper_setattr_flags(&flags, vap->va_flags,
373 hammer2_to_unix_xid(&ip->meta.uid),
376 if (ip->meta.uflags != flags) {
377 hammer2_inode_modify(ip);
378 ip->meta.uflags = flags;
379 ip->meta.ctime = ctime;
380 kflags |= NOTE_ATTRIB;
382 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
389 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
393 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
394 mode_t cur_mode = ip->meta.mode;
395 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
396 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
400 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
402 &cur_uid, &cur_gid, &cur_mode);
404 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
405 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
406 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
407 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
408 ip->meta.mode != cur_mode
410 hammer2_inode_modify(ip);
411 ip->meta.uid = uuid_uid;
412 ip->meta.gid = uuid_gid;
413 ip->meta.mode = cur_mode;
414 ip->meta.ctime = ctime;
416 kflags |= NOTE_ATTRIB;
423 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
426 if (vap->va_size == ip->meta.size)
428 if (vap->va_size < ip->meta.size) {
429 hammer2_mtx_ex(&ip->truncate_lock);
430 hammer2_truncate_file(ip, vap->va_size);
431 hammer2_mtx_unlock(&ip->truncate_lock);
433 hammer2_extend_file(ip, vap->va_size);
435 hammer2_inode_modify(ip);
436 ip->meta.mtime = ctime;
444 /* atime not supported */
445 if (vap->va_atime.tv_sec != VNOVAL) {
446 hammer2_inode_modify(ip);
447 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
448 kflags |= NOTE_ATTRIB;
451 if (vap->va_mode != (mode_t)VNOVAL) {
452 mode_t cur_mode = ip->meta.mode;
453 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
454 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
456 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
457 cur_uid, cur_gid, &cur_mode);
458 if (error == 0 && ip->meta.mode != cur_mode) {
459 hammer2_inode_modify(ip);
460 ip->meta.mode = cur_mode;
461 ip->meta.ctime = ctime;
462 kflags |= NOTE_ATTRIB;
466 if (vap->va_mtime.tv_sec != VNOVAL) {
467 hammer2_inode_modify(ip);
468 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
469 kflags |= NOTE_ATTRIB;
474 * If a truncation occurred we must call inode_fsync() now in order
475 * to trim the related data chains, otherwise a later expansion can
478 * If an extend occured that changed the DIRECTDATA state, we must
479 * call inode_fsync now in order to prepare the inode's indirect
482 if (ip->flags & HAMMER2_INODE_RESIZED)
483 hammer2_inode_chain_sync(ip);
488 hammer2_inode_unlock(ip);
489 hammer2_trans_done(ip->pmp);
490 hammer2_knote(ip->vp, kflags);
498 hammer2_vop_readdir(struct vop_readdir_args *ap)
500 hammer2_xop_readdir_t *xop;
501 hammer2_blockref_t bref;
518 saveoff = uio->uio_offset;
523 * Setup cookies directory entry cookies if requested
525 if (ap->a_ncookies) {
526 ncookies = uio->uio_resid / 16 + 1;
529 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
536 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
539 * Handle artificial entries. To ensure that only positive 64 bit
540 * quantities are returned to userland we always strip off bit 63.
541 * The hash code is designed such that codes 0x0000-0x7FFF are not
542 * used, allowing us to use these codes for articial entries.
544 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
545 * allow '..' to cross the mount point into (e.g.) the super-root.
548 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
549 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
553 cookies[cookie_index] = saveoff;
556 if (cookie_index == ncookies)
562 * Be careful with lockorder when accessing ".."
564 * (ip is the current dir. xip is the parent dir).
566 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
567 if (ip != ip->pmp->iroot)
568 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
569 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
573 cookies[cookie_index] = saveoff;
576 if (cookie_index == ncookies)
580 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
581 if (hammer2_debug & 0x0020)
582 kprintf("readdir: lkey %016jx\n", lkey);
587 * Use XOP for cluster scan.
589 * parent is the inode cluster, already locked for us. Don't
590 * double lock shared locks as this will screw up upgrades.
592 xop = hammer2_xop_alloc(ip, 0);
594 hammer2_xop_start(&xop->head, hammer2_xop_readdir);
597 const hammer2_inode_data_t *ripdata;
599 error = hammer2_xop_collect(&xop->head, 0);
602 if (cookie_index == ncookies)
604 if (hammer2_debug & 0x0020)
605 kprintf("cluster chain %p %p\n",
606 xop->head.cluster.focus,
607 (xop->head.cluster.focus ?
608 xop->head.cluster.focus->data : (void *)-1));
609 ripdata = &hammer2_cluster_rdata(&xop->head.cluster)->ipdata;
610 hammer2_cluster_bref(&xop->head.cluster, &bref);
611 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
612 dtype = hammer2_get_dtype(ripdata);
613 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
614 r = vop_write_dirent(&error, uio,
616 HAMMER2_DIRHASH_USERMSK,
618 ripdata->meta.name_len,
623 cookies[cookie_index] = saveoff;
626 /* XXX chain error */
627 kprintf("bad chain type readdir %d\n", bref.type);
630 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
631 if (error == ENOENT) {
634 saveoff = (hammer2_key_t)-1;
636 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
639 hammer2_inode_unlock(ip);
641 *ap->a_eofflag = eofflag;
642 if (hammer2_debug & 0x0020)
643 kprintf("readdir: done at %016jx\n", saveoff);
644 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
645 if (error && cookie_index == 0) {
647 kfree(cookies, M_TEMP);
649 *ap->a_cookies = NULL;
653 *ap->a_ncookies = cookie_index;
654 *ap->a_cookies = cookies;
662 * hammer2_vop_readlink { vp, uio, cred }
666 hammer2_vop_readlink(struct vop_readlink_args *ap)
673 if (vp->v_type != VLNK)
677 error = hammer2_read_file(ip, ap->a_uio, 0);
683 hammer2_vop_read(struct vop_read_args *ap)
693 * Read operations supported on this vnode?
696 if (vp->v_type != VREG)
706 seqcount = ap->a_ioflag >> 16;
707 bigread = (uio->uio_resid > 100 * 1024 * 1024);
709 error = hammer2_read_file(ip, uio, seqcount);
715 hammer2_vop_write(struct vop_write_args *ap)
725 * Read operations supported on this vnode?
728 if (vp->v_type != VREG)
737 if (ip->pmp->ronly) {
741 seqcount = ap->a_ioflag >> 16;
744 * Check resource limit
746 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
747 uio->uio_offset + uio->uio_resid >
748 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
749 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
754 * The transaction interlocks against flushes initiations
755 * (note: but will run concurrently with the actual flush).
757 hammer2_trans_init(ip->pmp, 0);
758 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
759 hammer2_trans_done(ip->pmp);
765 * Perform read operations on a file or symlink given an UNLOCKED
768 * The passed ip is not locked.
772 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
783 * WARNING! Assumes that the kernel interlocks size changes at the
786 hammer2_mtx_sh(&ip->lock);
787 hammer2_mtx_sh(&ip->truncate_lock);
788 size = ip->meta.size;
789 hammer2_mtx_unlock(&ip->lock);
791 while (uio->uio_resid > 0 && uio->uio_offset < size) {
798 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
802 error = cluster_read(ip->vp, leof, lbase, lblksize,
803 uio->uio_resid, seqcount * BKVASIZE,
806 if (uio->uio_segflg == UIO_NOCOPY) {
807 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
808 if (bp->b_flags & B_CACHE) {
811 if (bp->b_xio.xio_npages != 16)
812 kprintf("NPAGES BAD\n");
813 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
815 m = bp->b_xio.xio_pages[i];
816 if (m == NULL || m->valid == 0) {
817 kprintf("bp %016jx %016jx pg %d inv",
820 kprintf("m->object %p/%p", m->object, ip->vp->v_object);
826 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
830 error = bread(ip->vp, lbase, lblksize, &bp);
836 loff = (int)(uio->uio_offset - lbase);
838 if (n > uio->uio_resid)
840 if (n > size - uio->uio_offset)
841 n = (int)(size - uio->uio_offset);
842 bp->b_flags |= B_AGE;
843 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
846 hammer2_mtx_unlock(&ip->truncate_lock);
852 * Write to the file represented by the inode via the logical buffer cache.
853 * The inode may represent a regular file or a symlink.
855 * The inode must not be locked.
859 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
860 int ioflag, int seqcount)
862 hammer2_key_t old_eof;
863 hammer2_key_t new_eof;
872 * WARNING! Assumes that the kernel interlocks size changes at the
875 hammer2_mtx_ex(&ip->lock);
876 hammer2_mtx_sh(&ip->truncate_lock);
877 if (ioflag & IO_APPEND)
878 uio->uio_offset = ip->meta.size;
879 old_eof = ip->meta.size;
882 * Extend the file if necessary. If the write fails at some point
883 * we will truncate it back down to cover as much as we were able
886 * Doing this now makes it easier to calculate buffer sizes in
893 if (uio->uio_offset + uio->uio_resid > old_eof) {
894 new_eof = uio->uio_offset + uio->uio_resid;
896 hammer2_extend_file(ip, new_eof);
897 kflags |= NOTE_EXTEND;
901 hammer2_mtx_unlock(&ip->lock);
906 while (uio->uio_resid > 0) {
915 * Don't allow the buffer build to blow out the buffer
918 if ((ioflag & IO_RECURSE) == 0)
919 bwillwrite(HAMMER2_PBUFSIZE);
922 * This nominally tells us how much we can cluster and
923 * what the logical buffer size needs to be. Currently
924 * we don't try to cluster the write and just handle one
927 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
929 loff = (int)(uio->uio_offset - lbase);
931 KKASSERT(lblksize <= 65536);
934 * Calculate bytes to copy this transfer and whether the
935 * copy completely covers the buffer or not.
939 if (n > uio->uio_resid) {
941 if (loff == lbase && uio->uio_offset + n == new_eof)
949 if (lbase >= new_eof)
955 if (uio->uio_segflg == UIO_NOCOPY) {
957 * Issuing a write with the same data backing the
958 * buffer. Instantiate the buffer to collect the
959 * backing vm pages, then read-in any missing bits.
961 * This case is used by vop_stdputpages().
963 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
964 if ((bp->b_flags & B_CACHE) == 0) {
966 error = bread(ip->vp, lbase, lblksize, &bp);
968 } else if (trivial) {
970 * Even though we are entirely overwriting the buffer
971 * we may still have to zero it out to avoid a
972 * mmap/write visibility issue.
974 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
975 if ((bp->b_flags & B_CACHE) == 0)
979 * Partial overwrite, read in any missing bits then
980 * replace the portion being written.
982 * (The strategy code will detect zero-fill physical
983 * blocks for this case).
985 error = bread(ip->vp, lbase, lblksize, &bp);
996 * Ok, copy the data in
998 error = uiomovebp(bp, bp->b_data + loff, n, uio);
999 kflags |= NOTE_WRITE;
1007 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1008 * with IO_SYNC or IO_ASYNC set. These writes
1009 * must be handled as the pageout daemon expects.
1011 * NOTE! H2 relies on cluster_write() here because it
1012 * cannot preallocate disk blocks at the logical
1013 * level due to not knowing what the compression
1014 * size will be at this time.
1016 * We must use cluster_write() here and we depend
1017 * on the write-behind feature to flush buffers
1018 * appropriately. If we let the buffer daemons do
1019 * it the block allocations will be all over the
1022 if (ioflag & IO_SYNC) {
1024 } else if ((ioflag & IO_DIRECT) && endofblk) {
1026 } else if (ioflag & IO_ASYNC) {
1028 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1032 bp->b_flags |= B_CLUSTEROK;
1033 cluster_write(bp, new_eof, lblksize, seqcount);
1035 bp->b_flags |= B_CLUSTEROK;
1042 * Cleanup. If we extended the file EOF but failed to write through
1043 * the entire write is a failure and we have to back-up.
1045 if (error && new_eof != old_eof) {
1046 hammer2_mtx_unlock(&ip->truncate_lock);
1047 hammer2_mtx_ex(&ip->lock);
1048 hammer2_mtx_ex(&ip->truncate_lock);
1049 hammer2_truncate_file(ip, old_eof);
1050 if (ip->flags & HAMMER2_INODE_MODIFIED)
1051 hammer2_inode_chain_sync(ip);
1052 hammer2_mtx_unlock(&ip->lock);
1053 } else if (modified) {
1054 hammer2_mtx_ex(&ip->lock);
1055 hammer2_inode_modify(ip);
1056 hammer2_update_time(&ip->meta.mtime);
1057 if (ip->flags & HAMMER2_INODE_MODIFIED)
1058 hammer2_inode_chain_sync(ip);
1059 hammer2_mtx_unlock(&ip->lock);
1060 hammer2_knote(ip->vp, kflags);
1062 hammer2_trans_assert_strategy(ip->pmp);
1063 hammer2_mtx_unlock(&ip->truncate_lock);
1069 * Truncate the size of a file. The inode must not be locked.
1071 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1072 * ensure that any on-media data beyond the new file EOF has been destroyed.
1074 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1075 * held due to the way our write thread works. If the truncation
1076 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1077 * for dirtying that buffer and zeroing out trailing bytes.
1079 * WARNING! Assumes that the kernel interlocks size changes at the
1082 * WARNING! Caller assumes responsibility for removing dead blocks
1083 * if INODE_RESIZED is set.
1087 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1089 hammer2_key_t lbase;
1093 hammer2_mtx_unlock(&ip->lock);
1095 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1096 nvtruncbuf(ip->vp, nsize,
1097 nblksize, (int)nsize & (nblksize - 1),
1100 hammer2_mtx_ex(&ip->lock);
1101 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1102 ip->osize = ip->meta.size;
1103 ip->meta.size = nsize;
1104 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1105 hammer2_inode_modify(ip);
1110 * Extend the size of a file. The inode must not be locked.
1112 * Even though the file size is changing, we do not have to set the
1113 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1114 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1115 * to prepare the inode cluster's indirect block table, otherwise
1116 * async execution of the strategy code will implode on us.
1118 * WARNING! Assumes that the kernel interlocks size changes at the
1121 * WARNING! Caller assumes responsibility for transitioning out
1122 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1126 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1128 hammer2_key_t lbase;
1129 hammer2_key_t osize;
1135 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1136 hammer2_inode_modify(ip);
1137 osize = ip->meta.size;
1139 ip->meta.size = nsize;
1141 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1142 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1143 hammer2_inode_chain_sync(ip);
1146 hammer2_mtx_unlock(&ip->lock);
1148 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1149 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1155 hammer2_mtx_ex(&ip->lock);
1162 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1164 hammer2_xop_nresolve_t *xop;
1165 hammer2_inode_t *ip;
1166 hammer2_inode_t *dip;
1167 struct namecache *ncp;
1172 dip = VTOI(ap->a_dvp);
1173 xop = hammer2_xop_alloc(dip, 0);
1175 ncp = ap->a_nch->ncp;
1176 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1179 * Note: In DragonFly the kernel handles '.' and '..'.
1181 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1182 hammer2_xop_start(&xop->head, hammer2_xop_nresolve);
1184 error = hammer2_xop_collect(&xop->head, 0);
1188 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1190 hammer2_inode_unlock(dip);
1193 * Acquire the related vnode
1195 * NOTE: For error processing, only ENOENT resolves the namecache
1196 * entry to NULL, otherwise we just return the error and
1197 * leave the namecache unresolved.
1199 * NOTE: multiple hammer2_inode structures can be aliased to the
1200 * same chain element, for example for hardlinks. This
1201 * use case does not 'reattach' inode associations that
1202 * might already exist, but always allocates a new one.
1204 * WARNING: inode structure is locked exclusively via inode_get
1205 * but chain was locked shared. inode_unlock()
1206 * will handle it properly.
1209 vp = hammer2_igetv(ip, &error);
1212 cache_setvp(ap->a_nch, vp);
1213 } else if (error == ENOENT) {
1214 cache_setvp(ap->a_nch, NULL);
1216 hammer2_inode_unlock(ip);
1219 * The vp should not be released until after we've disposed
1220 * of our locks, because it might cause vop_inactive() to
1227 cache_setvp(ap->a_nch, NULL);
1229 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1230 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1231 ("resolve error %d/%p ap %p\n",
1232 error, ap->a_nch->ncp->nc_vp, ap));
1240 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1242 hammer2_inode_t *dip;
1247 dip = VTOI(ap->a_dvp);
1248 inum = dip->meta.iparent;
1252 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1263 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1265 hammer2_inode_t *dip;
1266 hammer2_inode_t *nip;
1267 struct namecache *ncp;
1268 const uint8_t *name;
1274 dip = VTOI(ap->a_dvp);
1275 if (dip->pmp->ronly) {
1280 ncp = ap->a_nch->ncp;
1281 name = ncp->nc_name;
1282 name_len = ncp->nc_nlen;
1284 hammer2_pfs_memory_wait(dip->pmp);
1285 hammer2_trans_init(dip->pmp, 0);
1287 inum = hammer2_trans_newinum(dip->pmp);
1290 * Create the actual inode as a hidden file in the iroot, then
1291 * create the directory entry as a hardlink to it. The creation
1292 * of the actual inode sets its nlinks to 1 which is the value
1295 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1300 hammer2_inode_create(dip, dip, NULL, NULL,
1303 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1308 KKASSERT(nip == NULL);
1311 *ap->a_vpp = hammer2_igetv(nip, &error);
1312 hammer2_inode_unlock(nip);
1316 * Update dip's mtime
1321 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1322 hammer2_update_time(&mtime);
1323 hammer2_inode_modify(dip);
1324 dip->meta.mtime = mtime;
1325 hammer2_inode_unlock(dip);
1328 hammer2_trans_done(dip->pmp);
1331 cache_setunresolved(ap->a_nch);
1332 cache_setvp(ap->a_nch, *ap->a_vpp);
1340 hammer2_vop_open(struct vop_open_args *ap)
1342 return vop_stdopen(ap);
1346 * hammer2_vop_advlock { vp, id, op, fl, flags }
1350 hammer2_vop_advlock(struct vop_advlock_args *ap)
1352 hammer2_inode_t *ip = VTOI(ap->a_vp);
1355 size = ip->meta.size;
1356 return (lf_advlock(ap, &ip->advlock, size));
1361 hammer2_vop_close(struct vop_close_args *ap)
1363 return vop_stdclose(ap);
1367 * hammer2_vop_nlink { nch, dvp, vp, cred }
1369 * Create a hardlink from (vp) to {dvp, nch}.
1373 hammer2_vop_nlink(struct vop_nlink_args *ap)
1375 hammer2_inode_t *tdip; /* target directory to create link in */
1376 hammer2_inode_t *ip; /* inode we are hardlinking to */
1377 struct namecache *ncp;
1378 const uint8_t *name;
1383 tdip = VTOI(ap->a_dvp);
1384 if (tdip->pmp->ronly) {
1389 ncp = ap->a_nch->ncp;
1390 name = ncp->nc_name;
1391 name_len = ncp->nc_nlen;
1394 * ip represents the file being hardlinked. The file could be a
1395 * normal file or a hardlink target if it has already been hardlinked.
1396 * (with the new semantics, it will almost always be a hardlink
1399 * Bump nlinks and potentially also create or move the hardlink
1400 * target in the parent directory common to (ip) and (tdip). The
1401 * consolidation code can modify ip->cluster. The returned cluster
1404 ip = VTOI(ap->a_vp);
1405 hammer2_pfs_memory_wait(ip->pmp);
1406 hammer2_trans_init(ip->pmp, 0);
1409 * Target should be an indexed inode or there's no way we will ever
1410 * be able to find it!
1412 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1417 * Can return NULL and error == EXDEV if the common parent
1418 * crosses a directory with the xlink flag set.
1420 hammer2_inode_lock(tdip, 0);
1421 hammer2_inode_lock(ip, 0);
1424 * Create the hardlink target and bump nlinks.
1427 hammer2_inode_create(tdip, tdip, NULL, NULL,
1430 HAMMER2_OBJTYPE_HARDLINK, ip->meta.type,
1432 hammer2_inode_modify(ip);
1437 * Update dip's mtime
1441 hammer2_update_time(&mtime);
1442 hammer2_inode_modify(tdip);
1443 tdip->meta.mtime = mtime;
1445 cache_setunresolved(ap->a_nch);
1446 cache_setvp(ap->a_nch, ap->a_vp);
1448 hammer2_inode_unlock(ip);
1449 hammer2_inode_unlock(tdip);
1451 hammer2_trans_done(ip->pmp);
1458 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1460 * The operating system has already ensured that the directory entry
1461 * does not exist and done all appropriate namespace locking.
1465 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1467 hammer2_inode_t *dip;
1468 hammer2_inode_t *nip;
1469 struct namecache *ncp;
1470 const uint8_t *name;
1476 dip = VTOI(ap->a_dvp);
1477 if (dip->pmp->ronly) {
1482 ncp = ap->a_nch->ncp;
1483 name = ncp->nc_name;
1484 name_len = ncp->nc_nlen;
1485 hammer2_pfs_memory_wait(dip->pmp);
1486 hammer2_trans_init(dip->pmp, 0);
1488 inum = hammer2_trans_newinum(dip->pmp);
1491 * Create the actual inode as a hidden file in the iroot, then
1492 * create the directory entry as a hardlink to it. The creation
1493 * of the actual inode sets its nlinks to 1 which is the value
1496 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1502 hammer2_inode_create(dip, dip, NULL, NULL,
1505 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1509 KKASSERT(nip == NULL);
1512 *ap->a_vpp = hammer2_igetv(nip, &error);
1513 hammer2_inode_unlock(nip);
1517 * Update dip's mtime
1522 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1523 hammer2_update_time(&mtime);
1524 hammer2_inode_modify(dip);
1525 dip->meta.mtime = mtime;
1526 hammer2_inode_unlock(dip);
1529 hammer2_trans_done(dip->pmp);
1532 cache_setunresolved(ap->a_nch);
1533 cache_setvp(ap->a_nch, *ap->a_vpp);
1540 * Make a device node (typically a fifo)
1544 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1546 hammer2_inode_t *dip;
1547 hammer2_inode_t *nip;
1548 struct namecache *ncp;
1549 const uint8_t *name;
1555 dip = VTOI(ap->a_dvp);
1556 if (dip->pmp->ronly) {
1561 ncp = ap->a_nch->ncp;
1562 name = ncp->nc_name;
1563 name_len = ncp->nc_nlen;
1564 hammer2_pfs_memory_wait(dip->pmp);
1565 hammer2_trans_init(dip->pmp, 0);
1568 * The device node is entered as the directory entry itself and not
1569 * as a hardlink to an inode. Since one cannot obtain a
1570 * file handle on the filesystem entry representing the device, we
1571 * do not have to worry about indexing its inode.
1573 inum = hammer2_trans_newinum(dip->pmp);
1574 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1579 hammer2_inode_create(dip, dip, NULL, NULL,
1582 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1588 KKASSERT(nip == NULL);
1591 *ap->a_vpp = hammer2_igetv(nip, &error);
1592 hammer2_inode_unlock(nip);
1596 * Update dip's mtime
1601 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1602 hammer2_update_time(&mtime);
1603 hammer2_inode_modify(dip);
1604 dip->meta.mtime = mtime;
1605 hammer2_inode_unlock(dip);
1608 hammer2_trans_done(dip->pmp);
1611 cache_setunresolved(ap->a_nch);
1612 cache_setvp(ap->a_nch, *ap->a_vpp);
1619 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1623 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1625 hammer2_inode_t *dip;
1626 hammer2_inode_t *nip;
1627 struct namecache *ncp;
1628 const uint8_t *name;
1633 dip = VTOI(ap->a_dvp);
1634 if (dip->pmp->ronly)
1637 ncp = ap->a_nch->ncp;
1638 name = ncp->nc_name;
1639 name_len = ncp->nc_nlen;
1640 hammer2_pfs_memory_wait(dip->pmp);
1641 hammer2_trans_init(dip->pmp, 0);
1643 ap->a_vap->va_type = VLNK; /* enforce type */
1646 * The softlink is entered into the directory itself and not
1647 * as a hardlink to an inode. Since one cannot obtain a
1648 * file handle on the softlink itself we do not have to worry
1649 * about indexing its inode.
1651 inum = hammer2_trans_newinum(dip->pmp);
1653 nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
1658 hammer2_inode_create(dip, dip, NULL, NULL,
1661 HAMMER2_OBJTYPE_HARDLINK, nip->meta.type,
1667 KKASSERT(nip == NULL);
1669 hammer2_trans_done(dip->pmp);
1672 *ap->a_vpp = hammer2_igetv(nip, &error);
1675 * Build the softlink (~like file data) and finalize the namecache.
1682 bytes = strlen(ap->a_target);
1684 hammer2_inode_unlock(nip);
1685 bzero(&auio, sizeof(auio));
1686 bzero(&aiov, sizeof(aiov));
1687 auio.uio_iov = &aiov;
1688 auio.uio_segflg = UIO_SYSSPACE;
1689 auio.uio_rw = UIO_WRITE;
1690 auio.uio_resid = bytes;
1691 auio.uio_iovcnt = 1;
1692 auio.uio_td = curthread;
1693 aiov.iov_base = ap->a_target;
1694 aiov.iov_len = bytes;
1695 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1696 /* XXX handle error */
1699 hammer2_inode_unlock(nip);
1703 * Update dip's mtime
1708 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1709 hammer2_update_time(&mtime);
1710 hammer2_inode_modify(dip);
1711 dip->meta.mtime = mtime;
1712 hammer2_inode_unlock(dip);
1715 hammer2_trans_done(dip->pmp);
1718 * Finalize namecache
1721 cache_setunresolved(ap->a_nch);
1722 cache_setvp(ap->a_nch, *ap->a_vpp);
1723 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1729 * hammer2_vop_nremove { nch, dvp, cred }
1733 hammer2_vop_nremove(struct vop_nremove_args *ap)
1735 hammer2_xop_unlink_t *xop;
1736 hammer2_inode_t *dip;
1737 hammer2_inode_t *ip;
1738 struct namecache *ncp;
1743 dip = VTOI(ap->a_dvp);
1744 if (dip->pmp->ronly) {
1749 ncp = ap->a_nch->ncp;
1751 hammer2_pfs_memory_wait(dip->pmp);
1752 hammer2_trans_init(dip->pmp, 0);
1753 hammer2_inode_lock(dip, 0);
1756 * The unlink XOP unlinks the path from the directory and
1757 * locates and returns the cluster associated with the real inode.
1758 * We have to handle nlinks here on the frontend.
1760 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1761 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1762 isopen = cache_isopen(ap->a_nch);
1764 xop->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT;
1765 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1768 * Collect the real inode and adjust nlinks, destroy the real
1769 * inode if nlinks transitions to 0 and it was the real inode
1770 * (else it has already been removed).
1772 error = hammer2_xop_collect(&xop->head, 0);
1773 hammer2_inode_unlock(dip);
1776 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1777 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1779 hammer2_inode_unlink_finisher(ip, isopen);
1780 hammer2_inode_unlock(ip);
1783 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1787 * Update dip's mtime
1792 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1793 hammer2_update_time(&mtime);
1794 hammer2_inode_modify(dip);
1795 dip->meta.mtime = mtime;
1796 hammer2_inode_unlock(dip);
1799 hammer2_inode_run_sideq(dip->pmp);
1800 hammer2_trans_done(dip->pmp);
1802 cache_unlink(ap->a_nch);
1808 * hammer2_vop_nrmdir { nch, dvp, cred }
1812 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1814 hammer2_xop_unlink_t *xop;
1815 hammer2_inode_t *dip;
1816 hammer2_inode_t *ip;
1817 struct namecache *ncp;
1822 dip = VTOI(ap->a_dvp);
1823 if (dip->pmp->ronly) {
1828 hammer2_pfs_memory_wait(dip->pmp);
1829 hammer2_trans_init(dip->pmp, 0);
1830 hammer2_inode_lock(dip, 0);
1832 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1834 ncp = ap->a_nch->ncp;
1835 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1836 isopen = cache_isopen(ap->a_nch);
1838 xop->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT;
1839 hammer2_xop_start(&xop->head, hammer2_xop_unlink);
1842 * Collect the real inode and adjust nlinks, destroy the real
1843 * inode if nlinks transitions to 0 and it was the real inode
1844 * (else it has already been removed).
1846 error = hammer2_xop_collect(&xop->head, 0);
1847 hammer2_inode_unlock(dip);
1850 ip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
1851 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1853 hammer2_inode_unlink_finisher(ip, isopen);
1854 hammer2_inode_unlock(ip);
1857 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1861 * Update dip's mtime
1866 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1867 hammer2_update_time(&mtime);
1868 hammer2_inode_modify(dip);
1869 dip->meta.mtime = mtime;
1870 hammer2_inode_unlock(dip);
1873 hammer2_inode_run_sideq(dip->pmp);
1874 hammer2_trans_done(dip->pmp);
1876 cache_unlink(ap->a_nch);
1882 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1886 hammer2_vop_nrename(struct vop_nrename_args *ap)
1888 struct namecache *fncp;
1889 struct namecache *tncp;
1890 hammer2_inode_t *fdip;
1891 hammer2_inode_t *tdip;
1892 hammer2_inode_t *ip;
1893 const uint8_t *fname;
1895 const uint8_t *tname;
1903 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1905 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1908 fdip = VTOI(ap->a_fdvp); /* source directory */
1909 tdip = VTOI(ap->a_tdvp); /* target directory */
1911 if (fdip->pmp->ronly)
1915 fncp = ap->a_fnch->ncp; /* entry name in source */
1916 fname = fncp->nc_name;
1917 fname_len = fncp->nc_nlen;
1919 tncp = ap->a_tnch->ncp; /* entry name in target */
1920 tname = tncp->nc_name;
1921 tname_len = tncp->nc_nlen;
1923 hammer2_pfs_memory_wait(tdip->pmp);
1924 hammer2_trans_init(tdip->pmp, 0);
1930 * ip is the inode being renamed. If this is a hardlink then
1931 * ip represents the actual file and not the hardlink marker.
1933 ip = VTOI(fncp->nc_vp);
1935 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1938 * Can return NULL and error == EXDEV if the common parent
1939 * crosses a directory with the xlink flag set.
1942 hammer2_inode_lock(fdip, 0);
1943 hammer2_inode_lock(tdip, 0);
1944 hammer2_inode_ref(ip); /* extra ref */
1946 hammer2_inode_lock(ip, 0);
1949 * Delete the target namespace.
1952 hammer2_xop_unlink_t *xop2;
1953 hammer2_inode_t *tip;
1957 * The unlink XOP unlinks the path from the directory and
1958 * locates and returns the cluster associated with the real
1959 * inode. We have to handle nlinks here on the frontend.
1961 xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
1962 hammer2_xop_setname(&xop2->head, tname, tname_len);
1963 isopen = cache_isopen(ap->a_tnch);
1965 xop2->dopermanent = isopen ? 0 : HAMMER2_DELETE_PERMANENT;
1966 hammer2_xop_start(&xop2->head, hammer2_xop_unlink);
1969 * Collect the real inode and adjust nlinks, destroy the real
1970 * inode if nlinks transitions to 0 and it was the real inode
1971 * (else it has already been removed).
1973 tnch_error = hammer2_xop_collect(&xop2->head, 0);
1974 /* hammer2_inode_unlock(tdip); */
1976 if (tnch_error == 0) {
1977 tip = hammer2_inode_get(tdip->pmp, NULL,
1978 &xop2->head.cluster, -1);
1979 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1981 hammer2_inode_unlink_finisher(tip, isopen);
1982 hammer2_inode_unlock(tip);
1985 hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
1987 /* hammer2_inode_lock(tdip, 0); */
1989 if (tnch_error && tnch_error != ENOENT) {
1997 * Resolve the collision space for (tdip, tname, tname_len)
1999 * tdip must be held exclusively locked to prevent races.
2002 hammer2_xop_scanlhc_t *sxop;
2003 hammer2_tid_t lhcbase;
2005 tlhc = hammer2_dirhash(tname, tname_len);
2007 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2009 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
2010 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2011 if (tlhc != sxop->head.cluster.focus->bref.key)
2015 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2018 if (error != ENOENT)
2023 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2030 * Everything is setup, do the rename.
2032 * We have to synchronize ip->meta to the underlying operation.
2034 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
2035 * unlinking elements from their directories. Locking
2036 * the nlinks field does not lock the whole inode.
2038 /* hammer2_inode_lock(ip, 0); */
2040 hammer2_xop_nrename_t *xop4;
2042 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2044 xop4->ip_key = ip->meta.name_key;
2045 hammer2_xop_setip2(&xop4->head, ip);
2046 hammer2_xop_setip3(&xop4->head, tdip);
2047 hammer2_xop_setname(&xop4->head, fname, fname_len);
2048 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2049 hammer2_xop_start(&xop4->head, hammer2_xop_nrename);
2051 error = hammer2_xop_collect(&xop4->head, 0);
2052 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2054 if (error == ENOENT)
2057 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2058 hammer2_inode_modify(ip);
2059 ip->meta.name_len = tname_len;
2060 ip->meta.name_key = tlhc;
2069 * Update directory mtimes to represent the something changed.
2071 if (update_fdip || update_tdip) {
2074 hammer2_update_time(&mtime);
2076 hammer2_inode_modify(fdip);
2077 fdip->meta.mtime = mtime;
2080 hammer2_inode_modify(tdip);
2081 tdip->meta.mtime = mtime;
2084 hammer2_inode_unlock(ip);
2085 hammer2_inode_unlock(tdip);
2086 hammer2_inode_unlock(fdip);
2087 hammer2_inode_drop(ip);
2088 hammer2_inode_run_sideq(fdip->pmp);
2090 hammer2_trans_done(tdip->pmp);
2093 * Issue the namecache update after unlocking all the internal
2094 * hammer structures, otherwise we might deadlock.
2096 if (tnch_error == 0) {
2097 cache_unlink(ap->a_tnch);
2098 cache_setunresolved(ap->a_tnch);
2101 cache_rename(ap->a_fnch, ap->a_tnch);
2108 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2112 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2114 hammer2_inode_t *ip;
2118 ip = VTOI(ap->a_vp);
2120 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2121 ap->a_fflag, ap->a_cred);
2128 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2136 case (MOUNTCTL_SET_EXPORT):
2137 mp = ap->a_head.a_ops->head.vv_mount;
2140 if (ap->a_ctllen != sizeof(struct export_args))
2143 rc = vfs_export(mp, &pmp->export,
2144 (const struct export_args *)ap->a_ctl);
2147 rc = vop_stdmountctl(ap);
2157 static void filt_hammer2detach(struct knote *kn);
2158 static int filt_hammer2read(struct knote *kn, long hint);
2159 static int filt_hammer2write(struct knote *kn, long hint);
2160 static int filt_hammer2vnode(struct knote *kn, long hint);
2162 static struct filterops hammer2read_filtops =
2163 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2164 NULL, filt_hammer2detach, filt_hammer2read };
2165 static struct filterops hammer2write_filtops =
2166 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2167 NULL, filt_hammer2detach, filt_hammer2write };
2168 static struct filterops hammer2vnode_filtops =
2169 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2170 NULL, filt_hammer2detach, filt_hammer2vnode };
2174 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2176 struct vnode *vp = ap->a_vp;
2177 struct knote *kn = ap->a_kn;
2179 switch (kn->kn_filter) {
2181 kn->kn_fop = &hammer2read_filtops;
2184 kn->kn_fop = &hammer2write_filtops;
2187 kn->kn_fop = &hammer2vnode_filtops;
2190 return (EOPNOTSUPP);
2193 kn->kn_hook = (caddr_t)vp;
2195 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2201 filt_hammer2detach(struct knote *kn)
2203 struct vnode *vp = (void *)kn->kn_hook;
2205 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2209 filt_hammer2read(struct knote *kn, long hint)
2211 struct vnode *vp = (void *)kn->kn_hook;
2212 hammer2_inode_t *ip = VTOI(vp);
2215 if (hint == NOTE_REVOKE) {
2216 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2219 off = ip->meta.size - kn->kn_fp->f_offset;
2220 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2221 if (kn->kn_sfflags & NOTE_OLDAPI)
2223 return (kn->kn_data != 0);
2228 filt_hammer2write(struct knote *kn, long hint)
2230 if (hint == NOTE_REVOKE)
2231 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2237 filt_hammer2vnode(struct knote *kn, long hint)
2239 if (kn->kn_sfflags & hint)
2240 kn->kn_fflags |= hint;
2241 if (hint == NOTE_REVOKE) {
2242 kn->kn_flags |= (EV_EOF | EV_NODATA);
2245 return (kn->kn_fflags != 0);
2253 hammer2_vop_markatime(struct vop_markatime_args *ap)
2255 hammer2_inode_t *ip;
2268 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2272 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2274 error = hammer2_vop_kqfilter(ap);
2281 struct vop_ops hammer2_vnode_vops = {
2282 .vop_default = vop_defaultop,
2283 .vop_fsync = hammer2_vop_fsync,
2284 .vop_getpages = vop_stdgetpages,
2285 .vop_putpages = vop_stdputpages,
2286 .vop_access = hammer2_vop_access,
2287 .vop_advlock = hammer2_vop_advlock,
2288 .vop_close = hammer2_vop_close,
2289 .vop_nlink = hammer2_vop_nlink,
2290 .vop_ncreate = hammer2_vop_ncreate,
2291 .vop_nsymlink = hammer2_vop_nsymlink,
2292 .vop_nremove = hammer2_vop_nremove,
2293 .vop_nrmdir = hammer2_vop_nrmdir,
2294 .vop_nrename = hammer2_vop_nrename,
2295 .vop_getattr = hammer2_vop_getattr,
2296 .vop_setattr = hammer2_vop_setattr,
2297 .vop_readdir = hammer2_vop_readdir,
2298 .vop_readlink = hammer2_vop_readlink,
2299 .vop_getpages = vop_stdgetpages,
2300 .vop_putpages = vop_stdputpages,
2301 .vop_read = hammer2_vop_read,
2302 .vop_write = hammer2_vop_write,
2303 .vop_open = hammer2_vop_open,
2304 .vop_inactive = hammer2_vop_inactive,
2305 .vop_reclaim = hammer2_vop_reclaim,
2306 .vop_nresolve = hammer2_vop_nresolve,
2307 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2308 .vop_nmkdir = hammer2_vop_nmkdir,
2309 .vop_nmknod = hammer2_vop_nmknod,
2310 .vop_ioctl = hammer2_vop_ioctl,
2311 .vop_mountctl = hammer2_vop_mountctl,
2312 .vop_bmap = hammer2_vop_bmap,
2313 .vop_strategy = hammer2_vop_strategy,
2314 .vop_kqfilter = hammer2_vop_kqfilter
2317 struct vop_ops hammer2_spec_vops = {
2318 .vop_default = vop_defaultop,
2319 .vop_fsync = hammer2_vop_fsync,
2320 .vop_read = vop_stdnoread,
2321 .vop_write = vop_stdnowrite,
2322 .vop_access = hammer2_vop_access,
2323 .vop_close = hammer2_vop_close,
2324 .vop_markatime = hammer2_vop_markatime,
2325 .vop_getattr = hammer2_vop_getattr,
2326 .vop_inactive = hammer2_vop_inactive,
2327 .vop_reclaim = hammer2_vop_reclaim,
2328 .vop_setattr = hammer2_vop_setattr
2331 struct vop_ops hammer2_fifo_vops = {
2332 .vop_default = fifo_vnoperate,
2333 .vop_fsync = hammer2_vop_fsync,
2335 .vop_read = hammer2_vop_fiforead,
2336 .vop_write = hammer2_vop_fifowrite,
2338 .vop_access = hammer2_vop_access,
2340 .vop_close = hammer2_vop_fifoclose,
2342 .vop_markatime = hammer2_vop_markatime,
2343 .vop_getattr = hammer2_vop_getattr,
2344 .vop_inactive = hammer2_vop_inactive,
2345 .vop_reclaim = hammer2_vop_reclaim,
2346 .vop_setattr = hammer2_vop_setattr,
2347 .vop_kqfilter = hammer2_vop_fifokqfilter