2 * Copyright (c) 2011-2018 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/mount.h>
50 #include <sys/vnode.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
54 #include <sys/objcache.h>
55 #include <sys/event.h>
57 #include <vfs/fifofs/fifo.h>
61 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
63 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
64 int ioflag, int seqcount);
65 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
66 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
68 struct objcache *cache_xops;
72 hammer2_knote(struct vnode *vp, int flags)
75 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
79 * Last reference to a vnode is going away but it is still cached.
83 hammer2_vop_inactive(struct vop_inactive_args *ap)
100 * Check for deleted inodes and recycle immediately on the last
101 * release. Be sure to destroy any left-over buffer cache buffers
102 * so we do not waste time trying to flush them.
104 * Note that deleting the file block chains under the inode chain
105 * would just be a waste of energy, so don't do it.
107 * WARNING: nvtruncbuf() can only be safely called without the inode
108 * lock held due to the way our write thread works.
110 if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
115 * Detect updates to the embedded data which may be
116 * synchronized by the strategy code. Simply mark the
117 * inode modified so it gets picked up by our normal flush.
119 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
120 nvtruncbuf(vp, 0, nblksize, 0, 0);
127 * Reclaim a vnode so that it can be reused; after the inode is
128 * disassociated, the filesystem must manage it alone.
132 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
146 * The final close of a deleted file or directory marks it for
147 * destruction. The DELETED flag allows the flusher to shortcut
148 * any modified blocks still unflushed (that is, just ignore them).
150 * HAMMER2 usually does not try to optimize the freemap by returning
151 * deleted blocks to it as it does not usually know how many snapshots
152 * might be referencing portions of the file/dir.
158 * NOTE! We do not attempt to flush chains here, flushing is
159 * really fragile and could also deadlock.
164 * Modified inodes will already be on SIDEQ or SYNCQ. However,
165 * unlinked-but-open inodes may already have been synced and might
166 * still require deletion-on-reclaim.
168 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
169 HAMMER2_INODE_DELETING)) ==
170 HAMMER2_INODE_ISUNLINKED) {
171 hammer2_inode_lock(ip, 0);
172 if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
173 HAMMER2_INODE_DELETING)) ==
174 HAMMER2_INODE_ISUNLINKED) {
175 atomic_set_int(&ip->flags, HAMMER2_INODE_DELETING);
176 hammer2_inode_delayed_sideq(ip);
178 hammer2_inode_unlock(ip);
182 * Modified inodes will already be on SIDEQ or SYNCQ, no further
185 * We cannot safely synchronize the inode from inside the reclaim
186 * due to potentially deep locks held as-of when the reclaim occurs.
187 * Interactions and potential deadlocks abound. We also can't do it
188 * here without desynchronizing from the related directory entrie(s).
190 hammer2_inode_drop(ip); /* vp ref */
193 * XXX handle background sync when ip dirty, kernel will no longer
194 * notify us regarding this inode because there is no longer a
195 * vnode attached to it.
202 * Currently this function synchronizes the front-end inode state to the
203 * backend chain topology, then flushes the inode's chain and sub-topology
204 * to backend media. This function does not flush the root topology down to
209 hammer2_vop_fsync(struct vop_fsync_args *ap)
220 hammer2_trans_init(ip->pmp, 0);
223 * Flush dirty buffers in the file's logical buffer cache.
224 * It is best to wait for the strategy code to commit the
225 * buffers to the device's backing buffer cache before
226 * then trying to flush the inode.
228 * This should be quick, but certain inode modifications cached
229 * entirely in the hammer2_inode structure may not trigger a
230 * buffer read until the flush so the fsync can wind up also
231 * doing scattered reads.
233 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
234 bio_track_wait(&vp->v_track_write, 0, 0);
237 * Flush any inode changes
239 hammer2_inode_lock(ip, 0);
240 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MODIFIED))
241 error1 = hammer2_inode_chain_sync(ip);
244 * Flush dirty chains related to the inode.
246 * NOTE! We are not in a flush transaction. The inode remains on
247 * the sideq so the filesystem syncer can synchronize it to
250 error2 = hammer2_inode_chain_flush(ip, HAMMER2_XOP_INODE_STOP);
255 * We may be able to clear the vnode dirty flag. The
256 * hammer2_pfs_moderate() code depends on this usually working.
258 if ((ip->flags & (HAMMER2_INODE_MODIFIED |
259 HAMMER2_INODE_RESIZED |
260 HAMMER2_INODE_DIRTYDATA)) == 0 &&
261 RB_EMPTY(&vp->v_rbdirty_tree) &&
262 !bio_track_active(&vp->v_track_write)) {
265 hammer2_inode_unlock(ip);
266 hammer2_trans_done(ip->pmp, 0);
272 * No lock needed, just handle ip->update
276 hammer2_vop_access(struct vop_access_args *ap)
278 hammer2_inode_t *ip = VTOI(ap->a_vp);
287 update = spin_access_start(&ip->cluster_spin);
289 /*hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);*/
290 uid = hammer2_to_unix_xid(&ip->meta.uid);
291 gid = hammer2_to_unix_xid(&ip->meta.gid);
292 mode = ip->meta.mode;
293 uflags = ip->meta.uflags;
294 /*hammer2_inode_unlock(ip);*/
296 if (__predict_false(spin_access_end(&ip->cluster_spin, update)))
299 error = vop_helper_access(ap, uid, gid, mode, uflags);
306 hammer2_vop_getattr(struct vop_getattr_args *ap)
312 hammer2_chain_t *chain;
323 update = spin_access_start(&ip->cluster_spin);
325 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
326 vap->va_fileid = ip->meta.inum;
327 vap->va_mode = ip->meta.mode;
328 vap->va_nlink = ip->meta.nlinks;
329 vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
330 vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
333 vap->va_size = ip->meta.size; /* protected by shared lock */
334 vap->va_blocksize = HAMMER2_PBUFSIZE;
335 vap->va_flags = ip->meta.uflags;
336 hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
337 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
338 hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
341 if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
343 * Can't really calculate directory use sans the files under
344 * it, just assume one block for now.
346 vap->va_bytes += HAMMER2_INODE_BYTES;
348 for (i = 0; i < ip->cluster.nchains; ++i) {
349 if ((chain = ip->cluster.array[i].chain) != NULL) {
351 chain->bref.embed.stats.data_count) {
353 chain->bref.embed.stats.data_count;
358 vap->va_type = hammer2_get_vtype(ip->meta.type);
360 vap->va_uid_uuid = ip->meta.uid;
361 vap->va_gid_uuid = ip->meta.gid;
362 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
365 if (__predict_false(spin_access_end(&ip->cluster_spin, update)))
373 hammer2_vop_getattr_lite(struct vop_getattr_lite_args *ap)
378 struct vattr_lite *lvap;
388 update = spin_access_start(&ip->cluster_spin);
391 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
392 vap->va_fileid = ip->meta.inum;
394 lvap->va_mode = ip->meta.mode;
395 lvap->va_nlink = ip->meta.nlinks;
396 lvap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
397 lvap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
402 lvap->va_size = ip->meta.size;
404 vap->va_blocksize = HAMMER2_PBUFSIZE;
406 lvap->va_flags = ip->meta.uflags;
407 lvap->va_type = hammer2_get_vtype(ip->meta.type);
410 vap->va_uid_uuid = ip->meta.uid;
411 vap->va_gid_uuid = ip->meta.gid;
412 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
416 if (__predict_false(spin_access_end(&ip->cluster_spin, update)))
424 hammer2_vop_setattr(struct vop_setattr_args *ap)
435 hammer2_update_time(&ctime);
443 * Normally disallow setattr if there is no space, unless we
444 * are in emergency mode (might be needed to chflags -R noschg
445 * files prior to removal).
447 if ((ip->pmp->flags & HAMMER2_PMPF_EMERG) == 0 &&
448 hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1) {
452 hammer2_trans_init(ip->pmp, 0);
453 hammer2_inode_lock(ip, 0);
456 if (vap->va_flags != VNOVAL) {
459 flags = ip->meta.uflags;
460 error = vop_helper_setattr_flags(&flags, vap->va_flags,
461 hammer2_to_unix_xid(&ip->meta.uid),
464 if (ip->meta.uflags != flags) {
465 hammer2_inode_modify(ip);
466 spin_lock_update(&ip->cluster_spin);
467 ip->meta.uflags = flags;
468 ip->meta.ctime = ctime;
469 spin_unlock_update(&ip->cluster_spin);
470 kflags |= NOTE_ATTRIB;
472 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
479 if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
483 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
484 mode_t cur_mode = ip->meta.mode;
485 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
486 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
490 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
492 &cur_uid, &cur_gid, &cur_mode);
494 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
495 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
496 if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
497 bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
498 ip->meta.mode != cur_mode
500 hammer2_inode_modify(ip);
501 spin_lock_update(&ip->cluster_spin);
502 ip->meta.uid = uuid_uid;
503 ip->meta.gid = uuid_gid;
504 ip->meta.mode = cur_mode;
505 ip->meta.ctime = ctime;
506 spin_unlock_update(&ip->cluster_spin);
508 kflags |= NOTE_ATTRIB;
515 if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
518 if (vap->va_size == ip->meta.size)
520 if (vap->va_size < ip->meta.size) {
521 hammer2_mtx_ex(&ip->truncate_lock);
522 hammer2_truncate_file(ip, vap->va_size);
523 hammer2_mtx_unlock(&ip->truncate_lock);
524 kflags |= NOTE_WRITE;
526 hammer2_extend_file(ip, vap->va_size);
527 kflags |= NOTE_WRITE | NOTE_EXTEND;
529 hammer2_inode_modify(ip);
530 ip->meta.mtime = ctime;
531 vclrflags(vp, VLASTWRITETS);
539 /* atime not supported */
540 if (vap->va_atime.tv_sec != VNOVAL) {
541 hammer2_inode_modify(ip);
542 ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
543 kflags |= NOTE_ATTRIB;
546 if (vap->va_mode != (mode_t)VNOVAL) {
547 mode_t cur_mode = ip->meta.mode;
548 uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
549 gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
551 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
552 cur_uid, cur_gid, &cur_mode);
553 if (error == 0 && ip->meta.mode != cur_mode) {
554 hammer2_inode_modify(ip);
555 spin_lock_update(&ip->cluster_spin);
556 ip->meta.mode = cur_mode;
557 ip->meta.ctime = ctime;
558 spin_unlock_update(&ip->cluster_spin);
559 kflags |= NOTE_ATTRIB;
563 if (vap->va_mtime.tv_sec != VNOVAL) {
564 hammer2_inode_modify(ip);
565 ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
566 kflags |= NOTE_ATTRIB;
567 vclrflags(vp, VLASTWRITETS);
572 * If a truncation occurred we must call chain_sync() now in order
573 * to trim the related data chains, otherwise a later expansion can
576 * If an extend occured that changed the DIRECTDATA state, we must
577 * call inode_chain_sync now in order to prepare the inode's indirect
580 * WARNING! This means we are making an adjustment to the inode's
581 * chain outside of sync/fsync, and not just to inode->meta, which
582 * may result in some consistency issues if a crash were to occur
583 * at just the wrong time.
585 if (ip->flags & HAMMER2_INODE_RESIZED)
586 hammer2_inode_chain_sync(ip);
591 hammer2_inode_unlock(ip);
592 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ);
593 hammer2_knote(ip->vp, kflags);
600 hammer2_vop_readdir(struct vop_readdir_args *ap)
602 hammer2_xop_readdir_t *xop;
603 hammer2_blockref_t bref;
618 saveoff = uio->uio_offset;
623 * Setup cookies directory entry cookies if requested
625 if (ap->a_ncookies) {
626 ncookies = uio->uio_resid / 16 + 1;
629 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
636 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
639 * Handle artificial entries. To ensure that only positive 64 bit
640 * quantities are returned to userland we always strip off bit 63.
641 * The hash code is designed such that codes 0x0000-0x7FFF are not
642 * used, allowing us to use these codes for articial entries.
644 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
645 * allow '..' to cross the mount point into (e.g.) the super-root.
648 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
649 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
653 cookies[cookie_index] = saveoff;
656 if (cookie_index == ncookies)
661 inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
662 if (ip != ip->pmp->iroot)
663 inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
664 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
668 cookies[cookie_index] = saveoff;
671 if (cookie_index == ncookies)
675 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
676 if (hammer2_debug & 0x0020)
677 kprintf("readdir: lkey %016jx\n", lkey);
682 * Use XOP for cluster scan.
684 * parent is the inode cluster, already locked for us. Don't
685 * double lock shared locks as this will screw up upgrades.
687 xop = hammer2_xop_alloc(ip, 0);
689 hammer2_xop_start(&xop->head, &hammer2_readdir_desc);
692 const hammer2_inode_data_t *ripdata;
696 error = hammer2_xop_collect(&xop->head, 0);
697 error = hammer2_error_to_errno(error);
701 if (cookie_index == ncookies)
703 if (hammer2_debug & 0x0020)
704 kprintf("cluster chain %p %p\n",
705 xop->head.cluster.focus,
706 (xop->head.cluster.focus ?
707 xop->head.cluster.focus->data : (void *)-1));
708 hammer2_cluster_bref(&xop->head.cluster, &bref);
710 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
711 ripdata = &hammer2_xop_gdata(&xop->head)->ipdata;
712 dtype = hammer2_get_dtype(ripdata->meta.type);
713 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
714 r = vop_write_dirent(&error, uio,
716 HAMMER2_DIRHASH_USERMSK,
718 ripdata->meta.name_len,
720 hammer2_xop_pdata(&xop->head);
724 cookies[cookie_index] = saveoff;
726 } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
729 dtype = hammer2_get_dtype(bref.embed.dirent.type);
730 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
731 namlen = bref.embed.dirent.namlen;
732 if (namlen <= sizeof(bref.check.buf)) {
733 dname = bref.check.buf;
735 dname = hammer2_xop_gdata(&xop->head)->buf;
737 r = vop_write_dirent(&error, uio,
738 bref.embed.dirent.inum, dtype,
740 if (namlen > sizeof(bref.check.buf))
741 hammer2_xop_pdata(&xop->head);
745 cookies[cookie_index] = saveoff;
748 /* XXX chain error */
749 kprintf("bad chain type readdir %d\n", bref.type);
752 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
753 if (error == ENOENT) {
756 saveoff = (hammer2_key_t)-1;
758 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
761 hammer2_inode_unlock(ip);
763 *ap->a_eofflag = eofflag;
764 if (hammer2_debug & 0x0020)
765 kprintf("readdir: done at %016jx\n", saveoff);
766 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
767 if (error && cookie_index == 0) {
769 kfree(cookies, M_TEMP);
771 *ap->a_cookies = NULL;
775 *ap->a_ncookies = cookie_index;
776 *ap->a_cookies = cookies;
783 * hammer2_vop_readlink { vp, uio, cred }
787 hammer2_vop_readlink(struct vop_readlink_args *ap)
794 if (vp->v_type != VLNK)
798 error = hammer2_read_file(ip, ap->a_uio, 0);
804 hammer2_vop_read(struct vop_read_args *ap)
814 * Read operations supported on this vnode?
817 if (vp->v_type == VDIR)
819 if (vp->v_type != VREG)
829 seqcount = ap->a_ioflag >> IO_SEQSHIFT;
830 bigread = (uio->uio_resid > 100 * 1024 * 1024);
832 error = hammer2_read_file(ip, uio, seqcount);
838 hammer2_vop_write(struct vop_write_args *ap)
849 * Read operations supported on this vnode?
852 if (vp->v_type != VREG)
859 ioflag = ap->a_ioflag;
862 if (ip->pmp->ronly || (ip->pmp->flags & HAMMER2_PMPF_EMERG))
864 switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) {
868 ioflag |= IO_DIRECT; /* semi-synchronous */
874 seqcount = ioflag >> IO_SEQSHIFT;
877 * Check resource limit
879 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
880 uio->uio_offset + uio->uio_resid >
881 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
882 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
887 * The transaction interlocks against flush initiations
888 * (note: but will run concurrently with the actual flush).
890 * To avoid deadlocking against the VM system, we must flag any
891 * transaction related to the buffer cache or other direct
892 * VM page manipulation.
894 if (uio->uio_segflg == UIO_NOCOPY) {
895 hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
897 hammer2_trans_init(ip->pmp, 0);
899 error = hammer2_write_file(ip, uio, ioflag, seqcount);
900 if (uio->uio_segflg == UIO_NOCOPY)
901 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_BUFCACHE |
902 HAMMER2_TRANS_SIDEQ);
904 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ);
910 * Perform read operations on a file or symlink given an UNLOCKED
913 * The passed ip is not locked.
917 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
928 * WARNING! Assumes that the kernel interlocks size changes at the
931 hammer2_mtx_sh(&ip->lock);
932 hammer2_mtx_sh(&ip->truncate_lock);
933 size = ip->meta.size;
934 hammer2_mtx_unlock(&ip->lock);
936 while (uio->uio_resid > 0 && uio->uio_offset < size) {
943 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
948 error = cluster_readx(ip->vp, leof, lbase, lblksize,
949 B_NOTMETA | B_KVABIO,
954 if (uio->uio_segflg == UIO_NOCOPY) {
955 bp = getblk(ip->vp, lbase, lblksize,
956 GETBLK_BHEAVY | GETBLK_KVABIO, 0);
957 if (bp->b_flags & B_CACHE) {
960 if (bp->b_xio.xio_npages != 16)
961 kprintf("NPAGES BAD\n");
962 for (i = 0; i < bp->b_xio.xio_npages; ++i) {
964 m = bp->b_xio.xio_pages[i];
965 if (m == NULL || m->valid == 0) {
966 kprintf("bp %016jx %016jx pg %d inv",
969 kprintf("m->object %p/%p", m->object, ip->vp->v_object);
975 kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
979 error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
986 loff = (int)(uio->uio_offset - lbase);
988 if (n > uio->uio_resid)
990 if (n > size - uio->uio_offset)
991 n = (int)(size - uio->uio_offset);
992 bp->b_flags |= B_AGE;
993 uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
996 hammer2_mtx_unlock(&ip->truncate_lock);
1002 * Write to the file represented by the inode via the logical buffer cache.
1003 * The inode may represent a regular file or a symlink.
1005 * The inode must not be locked.
1009 hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
1010 int ioflag, int seqcount)
1012 hammer2_key_t old_eof;
1013 hammer2_key_t new_eof;
1022 * WARNING! Assumes that the kernel interlocks size changes at the
1025 hammer2_mtx_ex(&ip->lock);
1026 hammer2_mtx_sh(&ip->truncate_lock);
1027 if (ioflag & IO_APPEND)
1028 uio->uio_offset = ip->meta.size;
1029 old_eof = ip->meta.size;
1032 * Extend the file if necessary. If the write fails at some point
1033 * we will truncate it back down to cover as much as we were able
1036 * Doing this now makes it easier to calculate buffer sizes in
1043 if (uio->uio_offset + uio->uio_resid > old_eof) {
1044 new_eof = uio->uio_offset + uio->uio_resid;
1046 hammer2_extend_file(ip, new_eof);
1047 kflags |= NOTE_EXTEND;
1051 hammer2_mtx_unlock(&ip->lock);
1056 while (uio->uio_resid > 0) {
1057 hammer2_key_t lbase;
1065 * Don't allow the buffer build to blow out the buffer
1068 if ((ioflag & IO_RECURSE) == 0)
1069 bwillwrite(HAMMER2_PBUFSIZE);
1072 * This nominally tells us how much we can cluster and
1073 * what the logical buffer size needs to be. Currently
1074 * we don't try to cluster the write and just handle one
1077 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1079 loff = (int)(uio->uio_offset - lbase);
1081 KKASSERT(lblksize <= MAXBSIZE);
1084 * Calculate bytes to copy this transfer and whether the
1085 * copy completely covers the buffer or not.
1088 n = lblksize - loff;
1089 if (n > uio->uio_resid) {
1091 if (loff == lbase && uio->uio_offset + n == new_eof)
1099 if (lbase >= new_eof)
1105 if (uio->uio_segflg == UIO_NOCOPY) {
1107 * Issuing a write with the same data backing the
1108 * buffer. Instantiate the buffer to collect the
1109 * backing vm pages, then read-in any missing bits.
1111 * This case is used by vop_stdputpages().
1113 bp = getblk(ip->vp, lbase, lblksize,
1114 GETBLK_BHEAVY | GETBLK_KVABIO, 0);
1115 if ((bp->b_flags & B_CACHE) == 0) {
1117 error = bread_kvabio(ip->vp, lbase,
1120 } else if (trivial) {
1122 * Even though we are entirely overwriting the buffer
1123 * we may still have to zero it out to avoid a
1124 * mmap/write visibility issue.
1126 bp = getblk(ip->vp, lbase, lblksize,
1127 GETBLK_BHEAVY | GETBLK_KVABIO, 0);
1128 if ((bp->b_flags & B_CACHE) == 0)
1132 * Partial overwrite, read in any missing bits then
1133 * replace the portion being written.
1135 * (The strategy code will detect zero-fill physical
1136 * blocks for this case).
1138 error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
1149 * Ok, copy the data in
1152 error = uiomovebp(bp, bp->b_data + loff, n, uio);
1153 kflags |= NOTE_WRITE;
1161 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1162 * with IO_SYNC or IO_ASYNC set. These writes
1163 * must be handled as the pageout daemon expects.
1165 * NOTE! H2 relies on cluster_write() here because it
1166 * cannot preallocate disk blocks at the logical
1167 * level due to not knowing what the compression
1168 * size will be at this time.
1170 * We must use cluster_write() here and we depend
1171 * on the write-behind feature to flush buffers
1172 * appropriately. If we let the buffer daemons do
1173 * it the block allocations will be all over the
1176 if (ioflag & IO_SYNC) {
1178 } else if ((ioflag & IO_DIRECT) && endofblk) {
1180 } else if (ioflag & IO_ASYNC) {
1182 } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
1186 bp->b_flags |= B_CLUSTEROK;
1187 cluster_write(bp, new_eof, lblksize, seqcount);
1189 bp->b_flags |= B_CLUSTEROK;
1196 * Cleanup. If we extended the file EOF but failed to write through
1197 * the entire write is a failure and we have to back-up.
1199 if (error && new_eof != old_eof) {
1200 hammer2_mtx_unlock(&ip->truncate_lock);
1201 hammer2_mtx_ex(&ip->lock); /* note lock order */
1202 hammer2_mtx_ex(&ip->truncate_lock); /* note lock order */
1203 hammer2_truncate_file(ip, old_eof);
1204 if (ip->flags & HAMMER2_INODE_MODIFIED)
1205 hammer2_inode_chain_sync(ip);
1206 hammer2_mtx_unlock(&ip->lock);
1207 } else if (modified) {
1208 struct vnode *vp = ip->vp;
1210 hammer2_mtx_ex(&ip->lock);
1211 hammer2_inode_modify(ip);
1212 if (uio->uio_segflg == UIO_NOCOPY) {
1213 if (vp->v_flag & VLASTWRITETS) {
1215 (unsigned long)vp->v_lastwrite_ts.tv_sec *
1217 vp->v_lastwrite_ts.tv_nsec / 1000;
1220 hammer2_update_time(&ip->meta.mtime);
1221 vclrflags(vp, VLASTWRITETS);
1226 * REMOVED - handled by hammer2_extend_file(). Do not issue
1227 * a chain_sync() outside of a sync/fsync except for DIRECTDATA
1230 * Under normal conditions we only issue a chain_sync if
1231 * the inode's DIRECTDATA state changed.
1233 if (ip->flags & HAMMER2_INODE_RESIZED)
1234 hammer2_inode_chain_sync(ip);
1236 hammer2_mtx_unlock(&ip->lock);
1237 hammer2_knote(ip->vp, kflags);
1239 hammer2_trans_assert_strategy(ip->pmp);
1240 hammer2_mtx_unlock(&ip->truncate_lock);
1246 * Truncate the size of a file. The inode must be locked.
1248 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
1249 * ensure that any on-media data beyond the new file EOF has been destroyed.
1251 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1252 * held due to the way our write thread works. If the truncation
1253 * occurs in the middle of a buffer, nvtruncbuf() is responsible
1254 * for dirtying that buffer and zeroing out trailing bytes.
1256 * WARNING! Assumes that the kernel interlocks size changes at the
1259 * WARNING! Caller assumes responsibility for removing dead blocks
1260 * if INODE_RESIZED is set.
1264 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1266 hammer2_key_t lbase;
1269 hammer2_mtx_unlock(&ip->lock);
1271 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1272 nvtruncbuf(ip->vp, nsize,
1273 nblksize, (int)nsize & (nblksize - 1),
1276 hammer2_mtx_ex(&ip->lock);
1277 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1278 ip->osize = ip->meta.size;
1279 ip->meta.size = nsize;
1280 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1281 hammer2_inode_modify(ip);
1285 * Extend the size of a file. The inode must be locked.
1287 * Even though the file size is changing, we do not have to set the
1288 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
1289 * boundary. When this occurs a hammer2_inode_chain_sync() is required
1290 * to prepare the inode cluster's indirect block table, otherwise
1291 * async execution of the strategy code will implode on us.
1293 * WARNING! Assumes that the kernel interlocks size changes at the
1296 * WARNING! Caller assumes responsibility for transitioning out
1297 * of the inode DIRECTDATA mode if INODE_RESIZED is set.
1301 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1303 hammer2_key_t lbase;
1304 hammer2_key_t osize;
1309 KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
1310 hammer2_inode_modify(ip);
1311 osize = ip->meta.size;
1313 ip->meta.size = nsize;
1316 * We must issue a chain_sync() when the DIRECTDATA state changes
1317 * to prevent confusion between the flush code and the in-memory
1318 * state. This is not perfect because we are doing it outside of
1319 * a sync/fsync operation, so it might not be fully synchronized
1320 * with the meta-data topology flush.
1322 * We must retain and re-dirty the buffer cache buffer containing
1323 * the direct data so it can be written to a real block. It should
1324 * not be possible for a bread error to occur since the original data
1325 * is extracted from the inode structure directly.
1327 if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
1331 oblksize = hammer2_calc_logical(ip, 0, NULL, NULL);
1332 error = bread_kvabio(ip->vp, 0, oblksize, &bp);
1333 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1334 hammer2_inode_chain_sync(ip);
1342 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1343 hammer2_inode_chain_sync(ip);
1346 hammer2_mtx_unlock(&ip->lock);
1348 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1349 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1355 hammer2_mtx_ex(&ip->lock);
1360 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1362 hammer2_xop_nresolve_t *xop;
1363 hammer2_inode_t *ip;
1364 hammer2_inode_t *dip;
1365 struct namecache *ncp;
1369 dip = VTOI(ap->a_dvp);
1370 xop = hammer2_xop_alloc(dip, 0);
1372 ncp = ap->a_nch->ncp;
1373 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1376 * Note: In DragonFly the kernel handles '.' and '..'.
1378 hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
1379 hammer2_xop_start(&xop->head, &hammer2_nresolve_desc);
1381 error = hammer2_xop_collect(&xop->head, 0);
1382 error = hammer2_error_to_errno(error);
1386 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1);
1388 hammer2_inode_unlock(dip);
1391 * Acquire the related vnode
1393 * NOTE: For error processing, only ENOENT resolves the namecache
1394 * entry to NULL, otherwise we just return the error and
1395 * leave the namecache unresolved.
1397 * NOTE: multiple hammer2_inode structures can be aliased to the
1398 * same chain element, for example for hardlinks. This
1399 * use case does not 'reattach' inode associations that
1400 * might already exist, but always allocates a new one.
1402 * WARNING: inode structure is locked exclusively via inode_get
1403 * but chain was locked shared. inode_unlock()
1404 * will handle it properly.
1407 vp = hammer2_igetv(ip, &error); /* error set to UNIX error */
1410 cache_setvp(ap->a_nch, vp);
1411 } else if (error == ENOENT) {
1412 cache_setvp(ap->a_nch, NULL);
1414 hammer2_inode_unlock(ip);
1417 * The vp should not be released until after we've disposed
1418 * of our locks, because it might cause vop_inactive() to
1425 cache_setvp(ap->a_nch, NULL);
1427 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1428 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1429 ("resolve error %d/%p ap %p\n",
1430 error, ap->a_nch->ncp->nc_vp, ap));
1437 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1439 hammer2_inode_t *dip;
1443 dip = VTOI(ap->a_dvp);
1444 inum = dip->meta.iparent;
1448 error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
1458 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1460 hammer2_inode_t *dip;
1461 hammer2_inode_t *nip;
1462 struct namecache *ncp;
1463 const uint8_t *name;
1468 dip = VTOI(ap->a_dvp);
1469 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG))
1471 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1474 ncp = ap->a_nch->ncp;
1475 name = ncp->nc_name;
1476 name_len = ncp->nc_nlen;
1478 hammer2_trans_init(dip->pmp, 0);
1480 inum = hammer2_trans_newinum(dip->pmp);
1483 * Create the actual inode as a hidden file in the iroot, then
1484 * create the directory entry. The creation of the actual inode
1485 * sets its nlinks to 1 which is the value we desire.
1487 * dip must be locked before nip to avoid deadlock.
1489 hammer2_inode_lock(dip, 0);
1490 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred,
1493 error = hammer2_error_to_errno(error);
1495 error = hammer2_dirent_create(dip, name, name_len,
1496 nip->meta.inum, nip->meta.type);
1497 /* returns UNIX error code */
1501 hammer2_inode_unlink_finisher(nip, 0);
1502 hammer2_inode_unlock(nip);
1508 * inode_depend() must occur before the igetv() because
1509 * the igetv() can temporarily release the inode lock.
1511 hammer2_inode_depend(dip, nip); /* before igetv */
1512 *ap->a_vpp = hammer2_igetv(nip, &error);
1513 hammer2_inode_unlock(nip);
1517 * Update dip's mtime
1519 * We can use a shared inode lock and allow the meta.mtime update
1520 * SMP race. hammer2_inode_modify() is MPSAFE w/a shared lock.
1525 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
1526 hammer2_update_time(&mtime);
1527 hammer2_inode_modify(dip);
1528 dip->meta.mtime = mtime;
1529 /*hammer2_inode_unlock(dip);*/
1531 hammer2_inode_unlock(dip);
1533 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
1536 cache_setunresolved(ap->a_nch);
1537 cache_setvp(ap->a_nch, *ap->a_vpp);
1538 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
1545 hammer2_vop_open(struct vop_open_args *ap)
1547 return vop_stdopen(ap);
1551 * hammer2_vop_advlock { vp, id, op, fl, flags }
1555 hammer2_vop_advlock(struct vop_advlock_args *ap)
1557 hammer2_inode_t *ip = VTOI(ap->a_vp);
1560 size = ip->meta.size;
1561 return (lf_advlock(ap, &ip->advlock, size));
1566 hammer2_vop_close(struct vop_close_args *ap)
1568 return vop_stdclose(ap);
1572 * hammer2_vop_nlink { nch, dvp, vp, cred }
1574 * Create a hardlink from (vp) to {dvp, nch}.
1578 hammer2_vop_nlink(struct vop_nlink_args *ap)
1580 hammer2_inode_t *tdip; /* target directory to create link in */
1581 hammer2_inode_t *ip; /* inode we are hardlinking to */
1582 struct namecache *ncp;
1583 const uint8_t *name;
1587 if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
1590 tdip = VTOI(ap->a_dvp);
1591 if (tdip->pmp->ronly || (tdip->pmp->flags & HAMMER2_PMPF_EMERG))
1593 if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1)
1596 ncp = ap->a_nch->ncp;
1597 name = ncp->nc_name;
1598 name_len = ncp->nc_nlen;
1601 * ip represents the file being hardlinked. The file could be a
1602 * normal file or a hardlink target if it has already been hardlinked.
1603 * (with the new semantics, it will almost always be a hardlink
1606 * Bump nlinks and potentially also create or move the hardlink
1607 * target in the parent directory common to (ip) and (tdip). The
1608 * consolidation code can modify ip->cluster. The returned cluster
1611 ip = VTOI(ap->a_vp);
1612 KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
1613 hammer2_trans_init(ip->pmp, 0);
1616 * Target should be an indexed inode or there's no way we will ever
1617 * be able to find it!
1619 KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);
1624 * Can return NULL and error == EXDEV if the common parent
1625 * crosses a directory with the xlink flag set.
1627 hammer2_inode_lock4(tdip, ip, NULL, NULL);
1630 * Create the directory entry and bump nlinks.
1633 error = hammer2_dirent_create(tdip, name, name_len,
1634 ip->meta.inum, ip->meta.type);
1635 hammer2_inode_modify(ip);
1640 * Update dip's mtime
1644 hammer2_update_time(&mtime);
1645 hammer2_inode_modify(tdip);
1646 tdip->meta.mtime = mtime;
1648 cache_setunresolved(ap->a_nch);
1649 cache_setvp(ap->a_nch, ap->a_vp);
1651 hammer2_inode_unlock(ip);
1652 hammer2_inode_unlock(tdip);
1654 hammer2_trans_done(ip->pmp, HAMMER2_TRANS_SIDEQ);
1655 hammer2_knote(ap->a_vp, NOTE_LINK);
1656 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1662 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1664 * The operating system has already ensured that the directory entry
1665 * does not exist and done all appropriate namespace locking.
1669 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1671 hammer2_inode_t *dip;
1672 hammer2_inode_t *nip;
1673 struct namecache *ncp;
1674 const uint8_t *name;
1679 dip = VTOI(ap->a_dvp);
1680 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG))
1682 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1685 ncp = ap->a_nch->ncp;
1686 name = ncp->nc_name;
1687 name_len = ncp->nc_nlen;
1688 hammer2_trans_init(dip->pmp, 0);
1690 inum = hammer2_trans_newinum(dip->pmp);
1693 * Create the actual inode as a hidden file in the iroot, then
1694 * create the directory entry. The creation of the actual inode
1695 * sets its nlinks to 1 which is the value we desire.
1697 * dip must be locked before nip to avoid deadlock.
1699 hammer2_inode_lock(dip, 0);
1700 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred,
1704 error = hammer2_error_to_errno(error);
1706 error = hammer2_dirent_create(dip, name, name_len,
1707 nip->meta.inum, nip->meta.type);
1711 hammer2_inode_unlink_finisher(nip, 0);
1712 hammer2_inode_unlock(nip);
1717 hammer2_inode_depend(dip, nip); /* before igetv */
1718 *ap->a_vpp = hammer2_igetv(nip, &error);
1719 hammer2_inode_unlock(nip);
1723 * Update dip's mtime
1728 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
1729 hammer2_update_time(&mtime);
1730 hammer2_inode_modify(dip);
1731 dip->meta.mtime = mtime;
1732 /*hammer2_inode_unlock(dip);*/
1734 hammer2_inode_unlock(dip);
1736 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
1739 cache_setunresolved(ap->a_nch);
1740 cache_setvp(ap->a_nch, *ap->a_vpp);
1741 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1747 * Make a device node (typically a fifo)
1751 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1753 hammer2_inode_t *dip;
1754 hammer2_inode_t *nip;
1755 struct namecache *ncp;
1756 const uint8_t *name;
1761 dip = VTOI(ap->a_dvp);
1762 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG))
1764 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1767 ncp = ap->a_nch->ncp;
1768 name = ncp->nc_name;
1769 name_len = ncp->nc_nlen;
1770 hammer2_trans_init(dip->pmp, 0);
1773 * Create the device inode and then create the directory entry.
1775 * dip must be locked before nip to avoid deadlock.
1777 inum = hammer2_trans_newinum(dip->pmp);
1779 hammer2_inode_lock(dip, 0);
1780 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred,
1783 error = hammer2_dirent_create(dip, name, name_len,
1784 nip->meta.inum, nip->meta.type);
1788 hammer2_inode_unlink_finisher(nip, 0);
1789 hammer2_inode_unlock(nip);
1794 hammer2_inode_depend(dip, nip); /* before igetv */
1795 *ap->a_vpp = hammer2_igetv(nip, &error);
1796 hammer2_inode_unlock(nip);
1800 * Update dip's mtime
1805 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
1806 hammer2_update_time(&mtime);
1807 hammer2_inode_modify(dip);
1808 dip->meta.mtime = mtime;
1809 /*hammer2_inode_unlock(dip);*/
1811 hammer2_inode_unlock(dip);
1813 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
1816 cache_setunresolved(ap->a_nch);
1817 cache_setvp(ap->a_nch, *ap->a_vpp);
1818 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1824 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1828 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1830 hammer2_inode_t *dip;
1831 hammer2_inode_t *nip;
1832 struct namecache *ncp;
1833 const uint8_t *name;
1838 dip = VTOI(ap->a_dvp);
1839 if (dip->pmp->ronly || (dip->pmp->flags & HAMMER2_PMPF_EMERG))
1841 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1844 ncp = ap->a_nch->ncp;
1845 name = ncp->nc_name;
1846 name_len = ncp->nc_nlen;
1847 hammer2_trans_init(dip->pmp, 0);
1849 ap->a_vap->va_type = VLNK; /* enforce type */
1852 * Create the softlink as an inode and then create the directory
1855 * dip must be locked before nip to avoid deadlock.
1857 inum = hammer2_trans_newinum(dip->pmp);
1859 hammer2_inode_lock(dip, 0);
1860 nip = hammer2_inode_create_normal(dip, ap->a_vap, ap->a_cred,
1863 error = hammer2_dirent_create(dip, name, name_len,
1864 nip->meta.inum, nip->meta.type);
1868 hammer2_inode_unlink_finisher(nip, 0);
1869 hammer2_inode_unlock(nip);
1873 hammer2_inode_unlock(dip);
1874 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
1877 hammer2_inode_depend(dip, nip); /* before igetv */
1878 *ap->a_vpp = hammer2_igetv(nip, &error);
1881 * Build the softlink (~like file data) and finalize the namecache.
1888 bytes = strlen(ap->a_target);
1890 hammer2_inode_unlock(nip);
1891 bzero(&auio, sizeof(auio));
1892 bzero(&aiov, sizeof(aiov));
1893 auio.uio_iov = &aiov;
1894 auio.uio_segflg = UIO_SYSSPACE;
1895 auio.uio_rw = UIO_WRITE;
1896 auio.uio_resid = bytes;
1897 auio.uio_iovcnt = 1;
1898 auio.uio_td = curthread;
1899 aiov.iov_base = ap->a_target;
1900 aiov.iov_len = bytes;
1901 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1902 /* XXX handle error */
1905 hammer2_inode_unlock(nip);
1909 * Update dip's mtime
1914 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
1915 hammer2_update_time(&mtime);
1916 hammer2_inode_modify(dip);
1917 dip->meta.mtime = mtime;
1918 /*hammer2_inode_unlock(dip);*/
1920 hammer2_inode_unlock(dip);
1922 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
1925 * Finalize namecache
1928 cache_setunresolved(ap->a_nch);
1929 cache_setvp(ap->a_nch, *ap->a_vpp);
1930 hammer2_knote(ap->a_dvp, NOTE_WRITE);
1936 * hammer2_vop_nremove { nch, dvp, cred }
1940 hammer2_vop_nremove(struct vop_nremove_args *ap)
1942 hammer2_xop_unlink_t *xop;
1943 hammer2_inode_t *dip;
1944 hammer2_inode_t *ip;
1945 struct namecache *ncp;
1949 dip = VTOI(ap->a_dvp);
1950 if (dip->pmp->ronly)
1953 /* allow removals, except user to also bulkfree */
1954 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
1958 ncp = ap->a_nch->ncp;
1960 if (hammer2_debug_inode && dip->meta.inum == hammer2_debug_inode) {
1961 kprintf("hammer2: attempt to delete inside debug inode: %s\n",
1963 while (hammer2_debug_inode &&
1964 dip->meta.inum == hammer2_debug_inode) {
1965 tsleep(&hammer2_debug_inode, 0, "h2debug", hz*5);
1969 hammer2_trans_init(dip->pmp, 0);
1970 hammer2_inode_lock(dip, 0);
1973 * The unlink XOP unlinks the path from the directory and
1974 * locates and returns the cluster associated with the real inode.
1975 * We have to handle nlinks here on the frontend.
1977 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
1978 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
1981 * The namecache entry is locked so nobody can use this namespace.
1982 * Calculate isopen to determine if this namespace has an open vp
1983 * associated with it and resolve the vp only if it does.
1985 * We try to avoid resolving the vnode if nobody has it open, but
1986 * note that the test is via this namespace only.
1988 isopen = cache_isopen(ap->a_nch);
1990 xop->dopermanent = 0;
1991 hammer2_xop_start(&xop->head, &hammer2_unlink_desc);
1994 * Collect the real inode and adjust nlinks, destroy the real
1995 * inode if nlinks transitions to 0 and it was the real inode
1996 * (else it has already been removed).
1998 error = hammer2_xop_collect(&xop->head, 0);
1999 error = hammer2_error_to_errno(error);
2002 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1);
2003 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2005 if (hammer2_debug_inode &&
2006 ip->meta.inum == hammer2_debug_inode) {
2007 kprintf("hammer2: attempt to delete debug "
2009 while (hammer2_debug_inode &&
2010 ip->meta.inum == hammer2_debug_inode) {
2011 tsleep(&hammer2_debug_inode, 0,
2015 hammer2_inode_unlink_finisher(ip, isopen);
2016 hammer2_inode_depend(dip, ip); /* after modified */
2017 hammer2_inode_unlock(ip);
2020 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2024 * Update dip's mtime
2029 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
2030 hammer2_update_time(&mtime);
2031 hammer2_inode_modify(dip);
2032 dip->meta.mtime = mtime;
2033 /*hammer2_inode_unlock(dip);*/
2035 hammer2_inode_unlock(dip);
2037 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
2039 cache_unlink(ap->a_nch);
2040 hammer2_knote(ap->a_dvp, NOTE_WRITE);
2046 * hammer2_vop_nrmdir { nch, dvp, cred }
2050 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
2052 hammer2_xop_unlink_t *xop;
2053 hammer2_inode_t *dip;
2054 hammer2_inode_t *ip;
2055 struct namecache *ncp;
2059 dip = VTOI(ap->a_dvp);
2060 if (dip->pmp->ronly)
2063 /* allow removals, except user to also bulkfree */
2064 if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
2068 hammer2_trans_init(dip->pmp, 0);
2069 hammer2_inode_lock(dip, 0);
2071 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
2073 ncp = ap->a_nch->ncp;
2074 hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
2075 isopen = cache_isopen(ap->a_nch);
2077 xop->dopermanent = 0;
2078 hammer2_xop_start(&xop->head, &hammer2_unlink_desc);
2081 * Collect the real inode and adjust nlinks, destroy the real
2082 * inode if nlinks transitions to 0 and it was the real inode
2083 * (else it has already been removed).
2085 error = hammer2_xop_collect(&xop->head, 0);
2086 error = hammer2_error_to_errno(error);
2089 ip = hammer2_inode_get(dip->pmp, &xop->head, -1, -1);
2090 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2092 hammer2_inode_unlink_finisher(ip, isopen);
2093 hammer2_inode_depend(dip, ip); /* after modified */
2094 hammer2_inode_unlock(ip);
2097 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2101 * Update dip's mtime
2106 /*hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);*/
2107 hammer2_update_time(&mtime);
2108 hammer2_inode_modify(dip);
2109 dip->meta.mtime = mtime;
2110 /*hammer2_inode_unlock(dip);*/
2112 hammer2_inode_unlock(dip);
2114 hammer2_trans_done(dip->pmp, HAMMER2_TRANS_SIDEQ);
2116 cache_unlink(ap->a_nch);
2117 hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
2123 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
2127 hammer2_vop_nrename(struct vop_nrename_args *ap)
2129 struct namecache *fncp;
2130 struct namecache *tncp;
2131 hammer2_inode_t *fdip; /* source directory */
2132 hammer2_inode_t *tdip; /* target directory */
2133 hammer2_inode_t *ip; /* file being renamed */
2134 hammer2_inode_t *tip; /* replaced target during rename or NULL */
2135 const uint8_t *fname;
2137 const uint8_t *tname;
2144 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
2146 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
2149 fdip = VTOI(ap->a_fdvp); /* source directory */
2150 tdip = VTOI(ap->a_tdvp); /* target directory */
2152 if (fdip->pmp->ronly || (fdip->pmp->flags & HAMMER2_PMPF_EMERG))
2154 if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1)
2157 fncp = ap->a_fnch->ncp; /* entry name in source */
2158 fname = fncp->nc_name;
2159 fname_len = fncp->nc_nlen;
2161 tncp = ap->a_tnch->ncp; /* entry name in target */
2162 tname = tncp->nc_name;
2163 tname_len = tncp->nc_nlen;
2165 hammer2_trans_init(tdip->pmp, 0);
2170 ip = VTOI(fncp->nc_vp);
2171 hammer2_inode_ref(ip); /* extra ref */
2174 * Lookup the target name to determine if a directory entry
2175 * is being overwritten. We only hold related inode locks
2176 * temporarily, the operating system is expected to protect
2177 * against rename races.
2179 tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL;
2181 hammer2_inode_ref(tip); /* extra ref */
2184 * Can return NULL and error == EXDEV if the common parent
2185 * crosses a directory with the xlink flag set.
2187 * For now try to avoid deadlocks with a simple pointer address
2188 * test. (tip) can be NULL.
2192 hammer2_inode_t *ip1 = fdip;
2193 hammer2_inode_t *ip2 = tdip;
2194 hammer2_inode_t *ip3 = ip;
2195 hammer2_inode_t *ip4 = tip; /* may be NULL */
2201 if (tip && ip > tip) {
2205 hammer2_inode_lock4(ip1, ip2, ip3, ip4);
2209 * Resolve the collision space for (tdip, tname, tname_len)
2211 * tdip must be held exclusively locked to prevent races since
2212 * multiple filenames can end up in the same collision space.
2215 hammer2_xop_scanlhc_t *sxop;
2216 hammer2_tid_t lhcbase;
2218 tlhc = hammer2_dirhash(tname, tname_len);
2220 sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
2222 hammer2_xop_start(&sxop->head, &hammer2_scanlhc_desc);
2223 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
2224 if (tlhc != sxop->head.cluster.focus->bref.key)
2228 error = hammer2_error_to_errno(error);
2229 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
2232 if (error != ENOENT)
2237 if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
2244 * Ready to go, issue the rename to the backend. Note that meta-data
2245 * updates to the related inodes occur separately from the rename
2248 * NOTE: While it is not necessary to update ip->meta.name*, doing
2249 * so aids catastrophic recovery and debugging.
2252 hammer2_xop_nrename_t *xop4;
2254 xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
2256 xop4->ip_key = ip->meta.name_key;
2257 hammer2_xop_setip2(&xop4->head, ip);
2258 hammer2_xop_setip3(&xop4->head, tdip);
2259 hammer2_xop_setname(&xop4->head, fname, fname_len);
2260 hammer2_xop_setname2(&xop4->head, tname, tname_len);
2261 hammer2_xop_start(&xop4->head, &hammer2_nrename_desc);
2263 error = hammer2_xop_collect(&xop4->head, 0);
2264 error = hammer2_error_to_errno(error);
2265 hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);
2267 if (error == ENOENT)
2271 * Update inode meta-data.
2273 * WARNING! The in-memory inode (ip) structure does not
2274 * maintain a copy of the inode's filename buffer.
2277 (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
2278 hammer2_inode_modify(ip);
2279 ip->meta.name_len = tname_len;
2280 ip->meta.name_key = tlhc;
2283 hammer2_inode_modify(ip);
2284 ip->meta.iparent = tdip->meta.inum;
2292 * If no error, the backend has replaced the target directory entry.
2293 * We must adjust nlinks on the original replace target if it exists.
2295 if (error == 0 && tip) {
2298 isopen = cache_isopen(ap->a_tnch);
2299 hammer2_inode_unlink_finisher(tip, isopen);
2303 * Update directory mtimes to represent the something changed.
2305 if (update_fdip || update_tdip) {
2308 hammer2_update_time(&mtime);
2310 hammer2_inode_modify(fdip);
2311 fdip->meta.mtime = mtime;
2314 hammer2_inode_modify(tdip);
2315 tdip->meta.mtime = mtime;
2319 hammer2_inode_unlock(tip);
2320 hammer2_inode_drop(tip);
2322 hammer2_inode_unlock(ip);
2323 hammer2_inode_unlock(tdip);
2324 hammer2_inode_unlock(fdip);
2325 hammer2_inode_drop(ip);
2326 hammer2_trans_done(tdip->pmp, HAMMER2_TRANS_SIDEQ);
2329 * Issue the namecache update after unlocking all the internal
2330 * hammer2 structures, otherwise we might deadlock.
2332 * WARNING! The target namespace must be updated atomically,
2333 * and we depend on cache_rename() to handle that for
2334 * us. Do not do a separate cache_unlink() because
2335 * that leaves a small window of opportunity for other
2336 * threads to allocate the target namespace before we
2337 * manage to complete our rename.
2339 * WARNING! cache_rename() (and cache_unlink()) will properly
2340 * set VREF_FINALIZE on any attached vnode. Do not
2341 * call cache_setunresolved() manually before-hand as
2342 * this will prevent the flag from being set later via
2343 * cache_rename(). If VREF_FINALIZE is not properly set
2344 * and the inode is no longer in the topology, related
2345 * chains can remain dirty indefinitely.
2347 if (error == 0 && tip) {
2348 /*cache_unlink(ap->a_tnch); see above */
2349 /*cache_setunresolved(ap->a_tnch); see above */
2352 cache_rename(ap->a_fnch, ap->a_tnch);
2353 hammer2_knote(ap->a_fdvp, NOTE_WRITE);
2354 hammer2_knote(ap->a_tdvp, NOTE_WRITE);
2355 hammer2_knote(fncp->nc_vp, NOTE_RENAME);
2362 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2366 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2368 hammer2_inode_t *ip;
2371 ip = VTOI(ap->a_vp);
2373 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2374 ap->a_fflag, ap->a_cred);
2380 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2387 case (MOUNTCTL_SET_EXPORT):
2388 mp = ap->a_head.a_ops->head.vv_mount;
2391 if (ap->a_ctllen != sizeof(struct export_args))
2394 rc = vfs_export(mp, &pmp->export,
2395 (const struct export_args *)ap->a_ctl);
2398 rc = vop_stdmountctl(ap);
2407 static void filt_hammer2detach(struct knote *kn);
2408 static int filt_hammer2read(struct knote *kn, long hint);
2409 static int filt_hammer2write(struct knote *kn, long hint);
2410 static int filt_hammer2vnode(struct knote *kn, long hint);
2412 static struct filterops hammer2read_filtops =
2413 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2414 NULL, filt_hammer2detach, filt_hammer2read };
2415 static struct filterops hammer2write_filtops =
2416 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2417 NULL, filt_hammer2detach, filt_hammer2write };
2418 static struct filterops hammer2vnode_filtops =
2419 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2420 NULL, filt_hammer2detach, filt_hammer2vnode };
2424 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2426 struct vnode *vp = ap->a_vp;
2427 struct knote *kn = ap->a_kn;
2429 switch (kn->kn_filter) {
2431 kn->kn_fop = &hammer2read_filtops;
2434 kn->kn_fop = &hammer2write_filtops;
2437 kn->kn_fop = &hammer2vnode_filtops;
2440 return (EOPNOTSUPP);
2443 kn->kn_hook = (caddr_t)vp;
2445 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2451 filt_hammer2detach(struct knote *kn)
2453 struct vnode *vp = (void *)kn->kn_hook;
2455 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2459 filt_hammer2read(struct knote *kn, long hint)
2461 struct vnode *vp = (void *)kn->kn_hook;
2462 hammer2_inode_t *ip = VTOI(vp);
2465 if (hint == NOTE_REVOKE) {
2466 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2469 off = ip->meta.size - kn->kn_fp->f_offset;
2470 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2471 if (kn->kn_sfflags & NOTE_OLDAPI)
2473 return (kn->kn_data != 0);
2478 filt_hammer2write(struct knote *kn, long hint)
2480 if (hint == NOTE_REVOKE)
2481 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2487 filt_hammer2vnode(struct knote *kn, long hint)
2489 if (kn->kn_sfflags & hint)
2490 kn->kn_fflags |= hint;
2491 if (hint == NOTE_REVOKE) {
2492 kn->kn_flags |= (EV_EOF | EV_NODATA);
2495 return (kn->kn_fflags != 0);
2503 hammer2_vop_markatime(struct vop_markatime_args *ap)
2505 hammer2_inode_t *ip;
2511 if (ip->pmp->ronly || (ip->pmp->flags & HAMMER2_PMPF_EMERG))
2518 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2522 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2524 error = hammer2_vop_kqfilter(ap);
2531 struct vop_ops hammer2_vnode_vops = {
2532 .vop_default = vop_defaultop,
2533 .vop_fsync = hammer2_vop_fsync,
2534 .vop_getpages = vop_stdgetpages,
2535 .vop_putpages = vop_stdputpages,
2536 .vop_access = hammer2_vop_access,
2537 .vop_advlock = hammer2_vop_advlock,
2538 .vop_close = hammer2_vop_close,
2539 .vop_nlink = hammer2_vop_nlink,
2540 .vop_ncreate = hammer2_vop_ncreate,
2541 .vop_nsymlink = hammer2_vop_nsymlink,
2542 .vop_nremove = hammer2_vop_nremove,
2543 .vop_nrmdir = hammer2_vop_nrmdir,
2544 .vop_nrename = hammer2_vop_nrename,
2545 .vop_getattr = hammer2_vop_getattr,
2546 .vop_getattr_lite = hammer2_vop_getattr_lite,
2547 .vop_setattr = hammer2_vop_setattr,
2548 .vop_readdir = hammer2_vop_readdir,
2549 .vop_readlink = hammer2_vop_readlink,
2550 .vop_read = hammer2_vop_read,
2551 .vop_write = hammer2_vop_write,
2552 .vop_open = hammer2_vop_open,
2553 .vop_inactive = hammer2_vop_inactive,
2554 .vop_reclaim = hammer2_vop_reclaim,
2555 .vop_nresolve = hammer2_vop_nresolve,
2556 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2557 .vop_nmkdir = hammer2_vop_nmkdir,
2558 .vop_nmknod = hammer2_vop_nmknod,
2559 .vop_ioctl = hammer2_vop_ioctl,
2560 .vop_mountctl = hammer2_vop_mountctl,
2561 .vop_bmap = hammer2_vop_bmap,
2562 .vop_strategy = hammer2_vop_strategy,
2563 .vop_kqfilter = hammer2_vop_kqfilter
2566 struct vop_ops hammer2_spec_vops = {
2567 .vop_default = vop_defaultop,
2568 .vop_fsync = hammer2_vop_fsync,
2569 .vop_read = vop_stdnoread,
2570 .vop_write = vop_stdnowrite,
2571 .vop_access = hammer2_vop_access,
2572 .vop_close = hammer2_vop_close,
2573 .vop_markatime = hammer2_vop_markatime,
2574 .vop_getattr = hammer2_vop_getattr,
2575 .vop_inactive = hammer2_vop_inactive,
2576 .vop_reclaim = hammer2_vop_reclaim,
2577 .vop_setattr = hammer2_vop_setattr
2580 struct vop_ops hammer2_fifo_vops = {
2581 .vop_default = fifo_vnoperate,
2582 .vop_fsync = hammer2_vop_fsync,
2584 .vop_read = hammer2_vop_fiforead,
2585 .vop_write = hammer2_vop_fifowrite,
2587 .vop_access = hammer2_vop_access,
2589 .vop_close = hammer2_vop_fifoclose,
2591 .vop_markatime = hammer2_vop_markatime,
2592 .vop_getattr = hammer2_vop_getattr,
2593 .vop_inactive = hammer2_vop_inactive,
2594 .vop_reclaim = hammer2_vop_reclaim,
2595 .vop_setattr = hammer2_vop_setattr,
2596 .vop_kqfilter = hammer2_vop_fifokqfilter