2 * Copyright (c) 2011-2014 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>
61 #include "hammer2_lz4.h"
63 #include "zlib/hammer2_zlib.h"
65 #define ZFOFFSET (-2LL)
67 static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
69 static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
70 int ioflag, int seqcount);
71 static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
72 static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);
74 struct objcache *cache_buffer_read;
75 struct objcache *cache_buffer_write;
78 * Callback used in read path in case that a block is compressed with LZ4.
82 hammer2_decompress_LZ4_callback(const char *data, u_int bytes, struct bio *bio)
85 char *compressed_buffer;
92 if bio->bio_caller_info2.index &&
93 bio->bio_caller_info1.uvalue32 !=
94 crc32(bp->b_data, bp->b_bufsize) --- return error
97 KKASSERT(bp->b_bufsize <= HAMMER2_PBUFSIZE);
98 compressed_size = *(const int *)data;
99 KKASSERT(compressed_size <= bytes - sizeof(int));
101 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
102 result = LZ4_decompress_safe(__DECONST(char *, &data[sizeof(int)]),
107 kprintf("READ PATH: Error during decompression."
109 (intmax_t)bio->bio_offset, bytes);
110 /* make sure it isn't random garbage */
111 bzero(compressed_buffer, bp->b_bufsize);
113 KKASSERT(result <= bp->b_bufsize);
114 bcopy(compressed_buffer, bp->b_data, bp->b_bufsize);
115 if (result < bp->b_bufsize)
116 bzero(bp->b_data + result, bp->b_bufsize - result);
117 objcache_put(cache_buffer_read, compressed_buffer);
119 bp->b_flags |= B_AGE;
123 * Callback used in read path in case that a block is compressed with ZLIB.
124 * It is almost identical to LZ4 callback, so in theory they can be unified,
125 * but we didn't want to make changes in bio structure for that.
129 hammer2_decompress_ZLIB_callback(const char *data, u_int bytes, struct bio *bio)
132 char *compressed_buffer;
133 z_stream strm_decompress;
139 KKASSERT(bp->b_bufsize <= HAMMER2_PBUFSIZE);
140 strm_decompress.avail_in = 0;
141 strm_decompress.next_in = Z_NULL;
143 ret = inflateInit(&strm_decompress);
146 kprintf("HAMMER2 ZLIB: Fatal error in inflateInit.\n");
148 compressed_buffer = objcache_get(cache_buffer_read, M_INTWAIT);
149 strm_decompress.next_in = __DECONST(char *, data);
151 /* XXX supply proper size, subset of device bp */
152 strm_decompress.avail_in = bytes;
153 strm_decompress.next_out = compressed_buffer;
154 strm_decompress.avail_out = bp->b_bufsize;
156 ret = inflate(&strm_decompress, Z_FINISH);
157 if (ret != Z_STREAM_END) {
158 kprintf("HAMMER2 ZLIB: Fatar error during decompression.\n");
159 bzero(compressed_buffer, bp->b_bufsize);
161 bcopy(compressed_buffer, bp->b_data, bp->b_bufsize);
162 result = bp->b_bufsize - strm_decompress.avail_out;
163 if (result < bp->b_bufsize)
164 bzero(bp->b_data + result, strm_decompress.avail_out);
165 objcache_put(cache_buffer_read, compressed_buffer);
166 ret = inflateEnd(&strm_decompress);
169 bp->b_flags |= B_AGE;
174 hammer2_knote(struct vnode *vp, int flags)
177 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
181 * Last reference to a vnode is going away but it is still cached.
185 hammer2_vop_inactive(struct vop_inactive_args *ap)
188 hammer2_cluster_t *cluster;
205 * Detect updates to the embedded data which may be synchronized by
206 * the strategy code. Simply mark the inode modified so it gets
207 * picked up by our normal flush.
209 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_NEVER |
210 HAMMER2_RESOLVE_RDONLY);
214 * Check for deleted inodes and recycle immediately.
216 * WARNING: nvtruncbuf() can only be safely called without the inode
217 * lock held due to the way our write thread works.
219 if (hammer2_cluster_isunlinked(cluster)) {
223 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
224 hammer2_inode_unlock(ip, cluster);
225 nvtruncbuf(vp, 0, nblksize, 0, 0);
228 hammer2_inode_unlock(ip, cluster);
235 * Reclaim a vnode so that it can be reused; after the inode is
236 * disassociated, the filesystem must manage it alone.
240 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
242 hammer2_cluster_t *cluster;
256 * Inode must be locked for reclaim.
259 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_NEVER |
260 HAMMER2_RESOLVE_RDONLY);
263 * The final close of a deleted file or directory marks it for
264 * destruction. The DELETED flag allows the flusher to shortcut
265 * any modified blocks still unflushed (that is, just ignore them).
267 * HAMMER2 usually does not try to optimize the freemap by returning
268 * deleted blocks to it as it does not usually know how many snapshots
269 * might be referencing portions of the file/dir.
275 * NOTE! We do not attempt to flush chains here, flushing is
276 * really fragile and could also deadlock.
281 * A reclaim can occur at any time so we cannot safely start a
282 * transaction to handle reclamation of unlinked files. Instead,
283 * the ip is left with a reference and placed on a linked list and
286 if (hammer2_cluster_isunlinked(cluster)) {
287 hammer2_inode_unlink_t *ipul;
289 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
292 hammer2_spin_ex(&pmp->list_spin);
293 TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
294 hammer2_spin_unex(&pmp->list_spin);
295 hammer2_inode_unlock(ip, cluster); /* unlock */
296 /* retain ref from vp for ipul */
298 hammer2_inode_unlock(ip, cluster); /* unlock */
299 hammer2_inode_drop(ip); /* vp ref */
301 /* cluster no longer referenced */
302 /* cluster = NULL; not needed */
305 * XXX handle background sync when ip dirty, kernel will no longer
306 * notify us regarding this inode because there is no longer a
307 * vnode attached to it.
316 hammer2_vop_fsync(struct vop_fsync_args *ap)
319 hammer2_trans_t trans;
320 hammer2_cluster_t *cluster;
328 /* XXX can't do this yet */
329 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
330 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
332 hammer2_trans_init(&trans, ip->pmp, 0);
333 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
336 * Calling chain_flush here creates a lot of duplicative
337 * COW operations due to non-optimal vnode ordering.
339 * Only do it for an actual fsync() syscall. The other forms
340 * which call this function will eventually call chain_flush
341 * on the volume root as a catch-all, which is far more optimal.
343 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
344 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
346 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME))
347 hammer2_inode_fsync(&trans, ip, cluster);
349 hammer2_inode_unlock(ip, cluster);
350 hammer2_trans_done(&trans);
358 hammer2_vop_access(struct vop_access_args *ap)
360 hammer2_inode_t *ip = VTOI(ap->a_vp);
361 const hammer2_inode_data_t *ripdata;
362 hammer2_cluster_t *cluster;
368 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
369 HAMMER2_RESOLVE_SHARED);
370 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
371 uid = hammer2_to_unix_xid(&ripdata->uid);
372 gid = hammer2_to_unix_xid(&ripdata->gid);
373 error = vop_helper_access(ap, uid, gid, ripdata->mode, ripdata->uflags);
374 hammer2_inode_unlock(ip, cluster);
382 hammer2_vop_getattr(struct vop_getattr_args *ap)
384 const hammer2_inode_data_t *ripdata;
385 hammer2_cluster_t *cluster;
398 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
399 HAMMER2_RESOLVE_SHARED);
400 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
401 KKASSERT(hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
403 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
404 vap->va_fileid = ripdata->inum;
405 vap->va_mode = ripdata->mode;
406 vap->va_nlink = ripdata->nlinks;
407 vap->va_uid = hammer2_to_unix_xid(&ripdata->uid);
408 vap->va_gid = hammer2_to_unix_xid(&ripdata->gid);
411 vap->va_size = ip->size; /* protected by shared lock */
412 vap->va_blocksize = HAMMER2_PBUFSIZE;
413 vap->va_flags = ripdata->uflags;
414 hammer2_time_to_timespec(ripdata->ctime, &vap->va_ctime);
415 hammer2_time_to_timespec(ripdata->mtime, &vap->va_mtime);
416 hammer2_time_to_timespec(ripdata->mtime, &vap->va_atime);
418 vap->va_bytes = vap->va_size; /* XXX */
419 vap->va_type = hammer2_get_vtype(ripdata);
421 vap->va_uid_uuid = ripdata->uid;
422 vap->va_gid_uuid = ripdata->gid;
423 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
426 hammer2_inode_unlock(ip, cluster);
434 hammer2_vop_setattr(struct vop_setattr_args *ap)
436 const hammer2_inode_data_t *ripdata;
437 hammer2_inode_data_t *wipdata;
439 hammer2_cluster_t *cluster;
440 hammer2_trans_t trans;
452 hammer2_update_time(&ctime);
456 if (ip->pmp->ronly) {
461 hammer2_pfs_memory_wait(ip->pmp);
462 hammer2_trans_init(&trans, ip->pmp, 0);
463 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
464 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
467 if (vap->va_flags != VNOVAL) {
470 flags = ripdata->uflags;
471 error = vop_helper_setattr_flags(&flags, vap->va_flags,
472 hammer2_to_unix_xid(&ripdata->uid),
475 if (ripdata->uflags != flags) {
476 wipdata = hammer2_cluster_modify_ip(&trans, ip,
478 wipdata->uflags = flags;
479 wipdata->ctime = ctime;
480 kflags |= NOTE_ATTRIB;
484 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
491 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
495 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
496 mode_t cur_mode = ripdata->mode;
497 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
498 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
502 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
504 &cur_uid, &cur_gid, &cur_mode);
506 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
507 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
508 if (bcmp(&uuid_uid, &ripdata->uid, sizeof(uuid_uid)) ||
509 bcmp(&uuid_gid, &ripdata->gid, sizeof(uuid_gid)) ||
510 ripdata->mode != cur_mode
512 wipdata = hammer2_cluster_modify_ip(&trans, ip,
514 wipdata->uid = uuid_uid;
515 wipdata->gid = uuid_gid;
516 wipdata->mode = cur_mode;
517 wipdata->ctime = ctime;
521 kflags |= NOTE_ATTRIB;
528 if (vap->va_size != VNOVAL && ip->size != vap->va_size) {
531 if (vap->va_size == ip->size)
533 hammer2_inode_unlock(ip, cluster);
534 if (vap->va_size < ip->size) {
535 hammer2_truncate_file(ip, vap->va_size);
537 hammer2_extend_file(ip, vap->va_size);
539 cluster = hammer2_inode_lock(ip,
540 HAMMER2_RESOLVE_ALWAYS);
542 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
551 /* atime not supported */
552 if (vap->va_atime.tv_sec != VNOVAL) {
553 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
554 wipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
555 kflags |= NOTE_ATTRIB;
560 if (vap->va_mtime.tv_sec != VNOVAL) {
561 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
562 wipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
563 kflags |= NOTE_ATTRIB;
568 if (vap->va_mode != (mode_t)VNOVAL) {
569 mode_t cur_mode = ripdata->mode;
570 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
571 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
573 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
574 cur_uid, cur_gid, &cur_mode);
575 if (error == 0 && ripdata->mode != cur_mode) {
576 wipdata = hammer2_cluster_modify_ip(&trans, ip,
578 wipdata->mode = cur_mode;
579 wipdata->ctime = ctime;
580 kflags |= NOTE_ATTRIB;
587 * If a truncation occurred we must call inode_fsync() now in order
588 * to trim the related data chains, otherwise a later expansion can
592 hammer2_cluster_modsync(cluster);
595 hammer2_inode_fsync(&trans, ip, cluster);
598 * Cleanup. If domtime is set an additional inode modification
599 * must be flagged. All other modifications will have already
600 * set INODE_MODIFIED and called vsetisdirty().
604 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
605 HAMMER2_INODE_MTIME);
609 hammer2_cluster_modsync(cluster);
610 hammer2_inode_unlock(ip, cluster);
611 hammer2_trans_done(&trans);
612 hammer2_knote(ip->vp, kflags);
620 hammer2_vop_readdir(struct vop_readdir_args *ap)
622 const hammer2_inode_data_t *ripdata;
624 hammer2_inode_t *xip;
625 hammer2_cluster_t *cparent;
626 hammer2_cluster_t *cluster;
627 hammer2_cluster_t *xcluster;
628 hammer2_blockref_t bref;
630 hammer2_key_t key_next;
644 saveoff = uio->uio_offset;
647 * Setup cookies directory entry cookies if requested
649 if (ap->a_ncookies) {
650 ncookies = uio->uio_resid / 16 + 1;
653 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
660 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
661 HAMMER2_RESOLVE_SHARED);
663 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
666 * Handle artificial entries. To ensure that only positive 64 bit
667 * quantities are returned to userland we always strip off bit 63.
668 * The hash code is designed such that codes 0x0000-0x7FFF are not
669 * used, allowing us to use these codes for articial entries.
671 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
672 * allow '..' to cross the mount point into (e.g.) the super-root.
675 cluster = (void *)(intptr_t)-1; /* non-NULL for early goto done case */
678 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
679 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
683 cookies[cookie_index] = saveoff;
686 if (cookie_index == ncookies)
692 * Be careful with lockorder when accessing ".."
694 * (ip is the current dir. xip is the parent dir).
696 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
697 while (ip->pip != NULL && ip != ip->pmp->iroot) {
699 hammer2_inode_ref(xip);
700 hammer2_inode_unlock(ip, cparent);
701 xcluster = hammer2_inode_lock(xip,
702 HAMMER2_RESOLVE_ALWAYS |
703 HAMMER2_RESOLVE_SHARED);
705 cparent = hammer2_inode_lock(ip,
706 HAMMER2_RESOLVE_ALWAYS |
707 HAMMER2_RESOLVE_SHARED);
708 hammer2_inode_drop(xip);
709 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
710 if (xip == ip->pip) {
711 inum = hammer2_cluster_rdata(xcluster)->
712 ipdata.inum & HAMMER2_DIRHASH_USERMSK;
713 hammer2_inode_unlock(xip, xcluster);
716 hammer2_inode_unlock(xip, xcluster);
718 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
722 cookies[cookie_index] = saveoff;
725 if (cookie_index == ncookies)
729 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
730 if (hammer2_debug & 0x0020)
731 kprintf("readdir: lkey %016jx\n", lkey);
734 * parent is the inode cluster, already locked for us. Don't
735 * double lock shared locks as this will screw up upgrades.
740 cluster = hammer2_cluster_lookup(cparent, &key_next, lkey, lkey,
741 HAMMER2_LOOKUP_SHARED);
742 if (cluster == NULL) {
743 cluster = hammer2_cluster_lookup(cparent, &key_next,
744 lkey, (hammer2_key_t)-1,
745 HAMMER2_LOOKUP_SHARED);
748 hammer2_cluster_bref(cluster, &bref);
750 if (hammer2_debug & 0x0020)
751 kprintf("readdir: p=%p chain=%p %016jx (next %016jx)\n",
752 cparent->focus, cluster->focus,
755 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
756 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
757 dtype = hammer2_get_dtype(ripdata);
758 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
759 r = vop_write_dirent(&error, uio,
761 HAMMER2_DIRHASH_USERMSK,
768 cookies[cookie_index] = saveoff;
771 /* XXX chain error */
772 kprintf("bad chain type readdir %d\n", bref.type);
776 * Keys may not be returned in order so once we have a
777 * placemarker (cluster) the scan must allow the full range
778 * or some entries will be missed.
780 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
781 key_next, (hammer2_key_t)-1,
782 HAMMER2_LOOKUP_SHARED);
784 hammer2_cluster_bref(cluster, &bref);
785 saveoff = (bref.key & HAMMER2_DIRHASH_USERMSK) + 1;
787 saveoff = (hammer2_key_t)-1;
789 if (cookie_index == ncookies)
793 hammer2_cluster_unlock(cluster);
794 hammer2_cluster_drop(cluster);
797 hammer2_inode_unlock(ip, cparent);
799 *ap->a_eofflag = (cluster == NULL);
800 if (hammer2_debug & 0x0020)
801 kprintf("readdir: done at %016jx\n", saveoff);
802 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
803 if (error && cookie_index == 0) {
805 kfree(cookies, M_TEMP);
807 *ap->a_cookies = NULL;
811 *ap->a_ncookies = cookie_index;
812 *ap->a_cookies = cookies;
820 * hammer2_vop_readlink { vp, uio, cred }
824 hammer2_vop_readlink(struct vop_readlink_args *ap)
831 if (vp->v_type != VLNK)
835 error = hammer2_read_file(ip, ap->a_uio, 0);
841 hammer2_vop_read(struct vop_read_args *ap)
851 * Read operations supported on this vnode?
854 if (vp->v_type != VREG)
864 seqcount = ap->a_ioflag >> 16;
865 bigread = (uio->uio_resid > 100 * 1024 * 1024);
867 error = hammer2_read_file(ip, uio, seqcount);
873 hammer2_vop_write(struct vop_write_args *ap)
876 hammer2_trans_t trans;
885 * Read operations supported on this vnode?
888 if (vp->v_type != VREG)
897 if (ip->pmp->ronly) {
901 seqcount = ap->a_ioflag >> 16;
902 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
905 * Check resource limit
907 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
908 uio->uio_offset + uio->uio_resid >
909 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
910 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
914 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
917 * The transaction interlocks against flushes initiations
918 * (note: but will run concurrently with the actual flush).
920 hammer2_trans_init(&trans, ip->pmp, 0);
921 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
922 hammer2_trans_done(&trans);
928 * Perform read operations on a file or symlink given an UNLOCKED
931 * The passed ip is not locked.
935 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
946 * WARNING! Assumes that the kernel interlocks size changes at the
949 hammer2_mtx_sh(&ip->lock);
951 hammer2_mtx_unlock(&ip->lock);
953 while (uio->uio_resid > 0 && uio->uio_offset < size) {
960 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
963 error = cluster_read(ip->vp, leof, lbase, lblksize,
964 uio->uio_resid, seqcount * BKVASIZE,
969 loff = (int)(uio->uio_offset - lbase);
971 if (n > uio->uio_resid)
973 if (n > size - uio->uio_offset)
974 n = (int)(size - uio->uio_offset);
975 bp->b_flags |= B_AGE;
976 uiomove((char *)bp->b_data + loff, n, uio);
983 * Write to the file represented by the inode via the logical buffer cache.
984 * The inode may represent a regular file or a symlink.
986 * The inode must not be locked.
990 hammer2_write_file(hammer2_inode_t *ip,
991 struct uio *uio, int ioflag, int seqcount)
993 hammer2_key_t old_eof;
994 hammer2_key_t new_eof;
1003 * WARNING! Assumes that the kernel interlocks size changes at the
1006 hammer2_mtx_ex(&ip->lock);
1007 if (ioflag & IO_APPEND)
1008 uio->uio_offset = ip->size;
1010 hammer2_mtx_unlock(&ip->lock);
1013 * Extend the file if necessary. If the write fails at some point
1014 * we will truncate it back down to cover as much as we were able
1017 * Doing this now makes it easier to calculate buffer sizes in
1024 if (uio->uio_offset + uio->uio_resid > old_eof) {
1025 new_eof = uio->uio_offset + uio->uio_resid;
1027 hammer2_extend_file(ip, new_eof);
1028 kflags |= NOTE_EXTEND;
1036 while (uio->uio_resid > 0) {
1037 hammer2_key_t lbase;
1045 * Don't allow the buffer build to blow out the buffer
1048 if ((ioflag & IO_RECURSE) == 0)
1049 bwillwrite(HAMMER2_PBUFSIZE);
1052 * This nominally tells us how much we can cluster and
1053 * what the logical buffer size needs to be. Currently
1054 * we don't try to cluster the write and just handle one
1057 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1059 loff = (int)(uio->uio_offset - lbase);
1061 KKASSERT(lblksize <= 65536);
1064 * Calculate bytes to copy this transfer and whether the
1065 * copy completely covers the buffer or not.
1068 n = lblksize - loff;
1069 if (n > uio->uio_resid) {
1071 if (loff == lbase && uio->uio_offset + n == new_eof)
1083 if (uio->uio_segflg == UIO_NOCOPY) {
1085 * Issuing a write with the same data backing the
1086 * buffer. Instantiate the buffer to collect the
1087 * backing vm pages, then read-in any missing bits.
1089 * This case is used by vop_stdputpages().
1091 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1092 if ((bp->b_flags & B_CACHE) == 0) {
1094 error = bread(ip->vp, lbase, lblksize, &bp);
1096 } else if (trivial) {
1098 * Even though we are entirely overwriting the buffer
1099 * we may still have to zero it out to avoid a
1100 * mmap/write visibility issue.
1102 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1103 if ((bp->b_flags & B_CACHE) == 0)
1107 * Partial overwrite, read in any missing bits then
1108 * replace the portion being written.
1110 * (The strategy code will detect zero-fill physical
1111 * blocks for this case).
1113 error = bread(ip->vp, lbase, lblksize, &bp);
1124 * Ok, copy the data in
1126 error = uiomove(bp->b_data + loff, n, uio);
1127 kflags |= NOTE_WRITE;
1135 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1136 * with IO_SYNC or IO_ASYNC set. These writes
1137 * must be handled as the pageout daemon expects.
1139 if (ioflag & IO_SYNC) {
1141 } else if ((ioflag & IO_DIRECT) && endofblk) {
1143 } else if (ioflag & IO_ASYNC) {
1151 * Cleanup. If we extended the file EOF but failed to write through
1152 * the entire write is a failure and we have to back-up.
1154 if (error && new_eof != old_eof) {
1155 hammer2_truncate_file(ip, old_eof);
1156 } else if (modified) {
1157 hammer2_mtx_ex(&ip->lock);
1158 hammer2_update_time(&ip->mtime);
1159 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME);
1160 hammer2_mtx_unlock(&ip->lock);
1162 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1163 hammer2_knote(ip->vp, kflags);
1164 vsetisdirty(ip->vp);
1165 hammer2_trans_assert_strategy(ip->pmp);
1171 * Truncate the size of a file. The inode must not be locked.
1173 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1175 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1176 * held due to the way our write thread works.
1178 * WARNING! Assumes that the kernel interlocks size changes at the
1183 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1185 hammer2_key_t lbase;
1190 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1191 nvtruncbuf(ip->vp, nsize,
1192 nblksize, (int)nsize & (nblksize - 1),
1195 hammer2_mtx_ex(&ip->lock);
1197 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1198 hammer2_mtx_unlock(&ip->lock);
1203 * Extend the size of a file. The inode must not be locked.
1205 * WARNING! Assumes that the kernel interlocks size changes at the
1208 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1212 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1214 hammer2_key_t lbase;
1215 hammer2_key_t osize;
1220 hammer2_mtx_ex(&ip->lock);
1223 hammer2_mtx_unlock(&ip->lock);
1226 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1227 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1233 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1239 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1241 hammer2_inode_t *ip;
1242 hammer2_inode_t *dip;
1243 hammer2_cluster_t *cparent;
1244 hammer2_cluster_t *cluster;
1245 const hammer2_inode_data_t *ripdata;
1246 hammer2_key_t key_next;
1248 struct namecache *ncp;
1249 const uint8_t *name;
1255 dip = VTOI(ap->a_dvp);
1256 ncp = ap->a_nch->ncp;
1257 name = ncp->nc_name;
1258 name_len = ncp->nc_nlen;
1259 lhc = hammer2_dirhash(name, name_len);
1262 * Note: In DragonFly the kernel handles '.' and '..'.
1264 cparent = hammer2_inode_lock(dip, HAMMER2_RESOLVE_ALWAYS |
1265 HAMMER2_RESOLVE_SHARED);
1267 cluster = hammer2_cluster_lookup(cparent, &key_next,
1268 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1269 HAMMER2_LOOKUP_SHARED);
1271 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE) {
1272 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1273 if (ripdata->name_len == name_len &&
1274 bcmp(ripdata->filename, name, name_len) == 0) {
1278 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1280 lhc + HAMMER2_DIRHASH_LOMASK,
1281 HAMMER2_LOOKUP_SHARED);
1283 hammer2_inode_unlock(dip, cparent);
1286 * Resolve hardlink entries before acquiring the inode.
1289 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1290 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1291 hammer2_tid_t inum = ripdata->inum;
1292 error = hammer2_hardlink_find(dip, NULL, &cluster);
1294 kprintf("hammer2: unable to find hardlink "
1295 "0x%016jx\n", inum);
1304 * nresolve needs to resolve hardlinks, the original cluster is not
1308 ip = hammer2_inode_get(dip->pmp, dip, cluster);
1309 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1310 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1311 kprintf("nresolve: fixup hardlink\n");
1312 hammer2_inode_ref(ip);
1313 hammer2_inode_unlock(ip, NULL);
1314 hammer2_cluster_unlock(cluster);
1315 hammer2_cluster_drop(cluster);
1316 cluster = hammer2_inode_lock(ip,
1317 HAMMER2_RESOLVE_ALWAYS);
1318 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1319 hammer2_inode_drop(ip);
1320 kprintf("nresolve: fixup to type %02x\n",
1329 * Deconsolidate any hardlink whos nlinks == 1. Ignore errors.
1330 * If an error occurs chain and ip are left alone.
1332 * XXX upgrade shared lock?
1334 if (ochain && chain &&
1335 chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1336 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1337 chain->data->ipdata.filename);
1338 /* XXX retain shared lock on dip? (currently not held) */
1339 hammer2_trans_init(&trans, dip->pmp, 0);
1340 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1341 hammer2_trans_done(&trans);
1346 * Acquire the related vnode
1348 * NOTE: For error processing, only ENOENT resolves the namecache
1349 * entry to NULL, otherwise we just return the error and
1350 * leave the namecache unresolved.
1352 * NOTE: multiple hammer2_inode structures can be aliased to the
1353 * same chain element, for example for hardlinks. This
1354 * use case does not 'reattach' inode associations that
1355 * might already exist, but always allocates a new one.
1357 * WARNING: inode structure is locked exclusively via inode_get
1358 * but chain was locked shared. inode_unlock()
1359 * will handle it properly.
1362 vp = hammer2_igetv(ip, cluster, &error);
1365 cache_setvp(ap->a_nch, vp);
1366 } else if (error == ENOENT) {
1367 cache_setvp(ap->a_nch, NULL);
1369 hammer2_inode_unlock(ip, cluster);
1372 * The vp should not be released until after we've disposed
1373 * of our locks, because it might cause vop_inactive() to
1380 cache_setvp(ap->a_nch, NULL);
1382 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1383 ("resolve error %d/%p ap %p\n",
1384 error, ap->a_nch->ncp->nc_vp, ap));
1391 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1393 hammer2_inode_t *dip;
1394 hammer2_inode_t *ip;
1395 hammer2_cluster_t *cparent;
1399 dip = VTOI(ap->a_dvp);
1401 if ((ip = dip->pip) == NULL) {
1406 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1407 *ap->a_vpp = hammer2_igetv(ip, cparent, &error);
1408 hammer2_inode_unlock(ip, cparent);
1416 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1418 hammer2_inode_t *dip;
1419 hammer2_inode_t *nip;
1420 hammer2_trans_t trans;
1421 hammer2_cluster_t *cluster;
1422 struct namecache *ncp;
1423 const uint8_t *name;
1428 dip = VTOI(ap->a_dvp);
1429 if (dip->pmp->ronly) {
1434 ncp = ap->a_nch->ncp;
1435 name = ncp->nc_name;
1436 name_len = ncp->nc_nlen;
1439 hammer2_pfs_memory_wait(dip->pmp);
1440 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1441 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1443 &cluster, 0, &error);
1445 KKASSERT(nip == NULL);
1448 *ap->a_vpp = hammer2_igetv(nip, cluster, &error);
1449 hammer2_inode_unlock(nip, cluster);
1451 hammer2_trans_done(&trans);
1454 cache_setunresolved(ap->a_nch);
1455 cache_setvp(ap->a_nch, *ap->a_vpp);
1462 * Return the largest contiguous physical disk range for the logical
1463 * request, in bytes.
1465 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1467 * Basically disabled, the logical buffer write thread has to deal with
1468 * buffers one-at-a-time.
1472 hammer2_vop_bmap(struct vop_bmap_args *ap)
1474 *ap->a_doffsetp = NOOFFSET;
1479 return (EOPNOTSUPP);
1484 hammer2_vop_open(struct vop_open_args *ap)
1486 return vop_stdopen(ap);
1490 * hammer2_vop_advlock { vp, id, op, fl, flags }
1494 hammer2_vop_advlock(struct vop_advlock_args *ap)
1496 hammer2_inode_t *ip = VTOI(ap->a_vp);
1497 const hammer2_inode_data_t *ripdata;
1498 hammer2_cluster_t *cparent;
1501 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
1502 HAMMER2_RESOLVE_SHARED);
1503 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
1504 size = ripdata->size;
1505 hammer2_inode_unlock(ip, cparent);
1506 return (lf_advlock(ap, &ip->advlock, size));
1512 hammer2_vop_close(struct vop_close_args *ap)
1514 return vop_stdclose(ap);
1518 * hammer2_vop_nlink { nch, dvp, vp, cred }
1520 * Create a hardlink from (vp) to {dvp, nch}.
1524 hammer2_vop_nlink(struct vop_nlink_args *ap)
1526 hammer2_inode_t *fdip; /* target directory to create link in */
1527 hammer2_inode_t *tdip; /* target directory to create link in */
1528 hammer2_inode_t *cdip; /* common parent directory */
1529 hammer2_inode_t *ip; /* inode we are hardlinking to */
1530 hammer2_cluster_t *cluster;
1531 hammer2_cluster_t *fdcluster;
1532 hammer2_cluster_t *tdcluster;
1533 hammer2_cluster_t *cdcluster;
1534 hammer2_trans_t trans;
1535 struct namecache *ncp;
1536 const uint8_t *name;
1541 tdip = VTOI(ap->a_dvp);
1542 if (tdip->pmp->ronly) {
1547 ncp = ap->a_nch->ncp;
1548 name = ncp->nc_name;
1549 name_len = ncp->nc_nlen;
1552 * ip represents the file being hardlinked. The file could be a
1553 * normal file or a hardlink target if it has already been hardlinked.
1554 * If ip is a hardlinked target then ip->pip represents the location
1555 * of the hardlinked target, NOT the location of the hardlink pointer.
1557 * Bump nlinks and potentially also create or move the hardlink
1558 * target in the parent directory common to (ip) and (tdip). The
1559 * consolidation code can modify ip->cluster and ip->pip. The
1560 * returned cluster is locked.
1562 ip = VTOI(ap->a_vp);
1563 hammer2_pfs_memory_wait(ip->pmp);
1564 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_NEWINODE);
1567 * The common parent directory must be locked first to avoid deadlocks.
1568 * Also note that fdip and/or tdip might match cdip.
1571 cdip = hammer2_inode_common_parent(fdip, tdip);
1572 cdcluster = hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1573 fdcluster = hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1574 tdcluster = hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1575 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1576 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1577 cdip, cdcluster, 1);
1582 * Create a directory entry connected to the specified cluster.
1584 * WARNING! chain can get moved by the connect (indirectly due to
1585 * potential indirect block creation).
1587 error = hammer2_inode_connect(&trans, &cluster, 1,
1591 cache_setunresolved(ap->a_nch);
1592 cache_setvp(ap->a_nch, ap->a_vp);
1595 hammer2_inode_unlock(ip, cluster);
1596 hammer2_inode_unlock(tdip, tdcluster);
1597 hammer2_inode_unlock(fdip, fdcluster);
1598 hammer2_inode_unlock(cdip, cdcluster);
1599 hammer2_inode_drop(cdip);
1600 hammer2_trans_done(&trans);
1607 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1609 * The operating system has already ensured that the directory entry
1610 * does not exist and done all appropriate namespace locking.
1614 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1616 hammer2_inode_t *dip;
1617 hammer2_inode_t *nip;
1618 hammer2_trans_t trans;
1619 hammer2_cluster_t *ncluster;
1620 struct namecache *ncp;
1621 const uint8_t *name;
1626 dip = VTOI(ap->a_dvp);
1627 if (dip->pmp->ronly) {
1632 ncp = ap->a_nch->ncp;
1633 name = ncp->nc_name;
1634 name_len = ncp->nc_nlen;
1635 hammer2_pfs_memory_wait(dip->pmp);
1636 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1639 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1641 &ncluster, 0, &error);
1643 KKASSERT(nip == NULL);
1646 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1647 hammer2_inode_unlock(nip, ncluster);
1649 hammer2_trans_done(&trans);
1652 cache_setunresolved(ap->a_nch);
1653 cache_setvp(ap->a_nch, *ap->a_vpp);
1660 * Make a device node (typically a fifo)
1664 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1666 hammer2_inode_t *dip;
1667 hammer2_inode_t *nip;
1668 hammer2_trans_t trans;
1669 hammer2_cluster_t *ncluster;
1670 struct namecache *ncp;
1671 const uint8_t *name;
1676 dip = VTOI(ap->a_dvp);
1677 if (dip->pmp->ronly) {
1682 ncp = ap->a_nch->ncp;
1683 name = ncp->nc_name;
1684 name_len = ncp->nc_nlen;
1685 hammer2_pfs_memory_wait(dip->pmp);
1686 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1689 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1691 &ncluster, 0, &error);
1693 KKASSERT(nip == NULL);
1696 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1697 hammer2_inode_unlock(nip, ncluster);
1699 hammer2_trans_done(&trans);
1702 cache_setunresolved(ap->a_nch);
1703 cache_setvp(ap->a_nch, *ap->a_vpp);
1710 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1714 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1716 hammer2_inode_t *dip;
1717 hammer2_inode_t *nip;
1718 hammer2_cluster_t *ncparent;
1719 hammer2_trans_t trans;
1720 struct namecache *ncp;
1721 const uint8_t *name;
1725 dip = VTOI(ap->a_dvp);
1726 if (dip->pmp->ronly)
1729 ncp = ap->a_nch->ncp;
1730 name = ncp->nc_name;
1731 name_len = ncp->nc_nlen;
1732 hammer2_pfs_memory_wait(dip->pmp);
1733 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1736 ap->a_vap->va_type = VLNK; /* enforce type */
1738 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1740 &ncparent, 0, &error);
1742 KKASSERT(nip == NULL);
1744 hammer2_trans_done(&trans);
1747 *ap->a_vpp = hammer2_igetv(nip, ncparent, &error);
1750 * Build the softlink (~like file data) and finalize the namecache.
1756 hammer2_inode_data_t *nipdata;
1758 nipdata = &hammer2_cluster_wdata(ncparent)->ipdata;
1759 /* nipdata = &nip->chain->data->ipdata;XXX */
1760 bytes = strlen(ap->a_target);
1762 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1763 KKASSERT(nipdata->op_flags &
1764 HAMMER2_OPFLAG_DIRECTDATA);
1765 bcopy(ap->a_target, nipdata->u.data, bytes);
1766 nipdata->size = bytes;
1768 hammer2_cluster_modsync(ncparent);
1769 hammer2_inode_unlock(nip, ncparent);
1770 /* nipdata = NULL; not needed */
1772 hammer2_inode_unlock(nip, ncparent);
1773 /* nipdata = NULL; not needed */
1774 bzero(&auio, sizeof(auio));
1775 bzero(&aiov, sizeof(aiov));
1776 auio.uio_iov = &aiov;
1777 auio.uio_segflg = UIO_SYSSPACE;
1778 auio.uio_rw = UIO_WRITE;
1779 auio.uio_resid = bytes;
1780 auio.uio_iovcnt = 1;
1781 auio.uio_td = curthread;
1782 aiov.iov_base = ap->a_target;
1783 aiov.iov_len = bytes;
1784 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1785 /* XXX handle error */
1789 hammer2_inode_unlock(nip, ncparent);
1791 hammer2_trans_done(&trans);
1794 * Finalize namecache
1797 cache_setunresolved(ap->a_nch);
1798 cache_setvp(ap->a_nch, *ap->a_vpp);
1799 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1805 * hammer2_vop_nremove { nch, dvp, cred }
1809 hammer2_vop_nremove(struct vop_nremove_args *ap)
1811 hammer2_inode_t *dip;
1812 hammer2_trans_t trans;
1813 struct namecache *ncp;
1814 const uint8_t *name;
1819 dip = VTOI(ap->a_dvp);
1820 if (dip->pmp->ronly) {
1825 ncp = ap->a_nch->ncp;
1826 name = ncp->nc_name;
1827 name_len = ncp->nc_nlen;
1829 hammer2_pfs_memory_wait(dip->pmp);
1830 hammer2_trans_init(&trans, dip->pmp, 0);
1831 error = hammer2_unlink_file(&trans, dip, name, name_len,
1832 0, NULL, ap->a_nch, -1);
1833 hammer2_run_unlinkq(&trans, dip->pmp);
1834 hammer2_trans_done(&trans);
1836 cache_unlink(ap->a_nch);
1842 * hammer2_vop_nrmdir { nch, dvp, cred }
1846 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1848 hammer2_inode_t *dip;
1849 hammer2_trans_t trans;
1850 struct namecache *ncp;
1851 const uint8_t *name;
1856 dip = VTOI(ap->a_dvp);
1857 if (dip->pmp->ronly) {
1862 ncp = ap->a_nch->ncp;
1863 name = ncp->nc_name;
1864 name_len = ncp->nc_nlen;
1866 hammer2_pfs_memory_wait(dip->pmp);
1867 hammer2_trans_init(&trans, dip->pmp, 0);
1868 hammer2_run_unlinkq(&trans, dip->pmp);
1869 error = hammer2_unlink_file(&trans, dip, name, name_len,
1870 1, NULL, ap->a_nch, -1);
1871 hammer2_trans_done(&trans);
1873 cache_unlink(ap->a_nch);
1879 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1883 hammer2_vop_nrename(struct vop_nrename_args *ap)
1885 struct namecache *fncp;
1886 struct namecache *tncp;
1887 hammer2_inode_t *cdip;
1888 hammer2_inode_t *fdip;
1889 hammer2_inode_t *tdip;
1890 hammer2_inode_t *ip;
1891 hammer2_cluster_t *cluster;
1892 hammer2_cluster_t *fdcluster;
1893 hammer2_cluster_t *tdcluster;
1894 hammer2_cluster_t *cdcluster;
1895 hammer2_trans_t trans;
1896 const uint8_t *fname;
1898 const uint8_t *tname;
1904 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1906 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1909 fdip = VTOI(ap->a_fdvp); /* source directory */
1910 tdip = VTOI(ap->a_tdvp); /* target directory */
1912 if (fdip->pmp->ronly)
1916 fncp = ap->a_fnch->ncp; /* entry name in source */
1917 fname = fncp->nc_name;
1918 fname_len = fncp->nc_nlen;
1920 tncp = ap->a_tnch->ncp; /* entry name in target */
1921 tname = tncp->nc_name;
1922 tname_len = tncp->nc_nlen;
1924 hammer2_pfs_memory_wait(tdip->pmp);
1925 hammer2_trans_init(&trans, tdip->pmp, 0);
1928 * ip is the inode being renamed. If this is a hardlink then
1929 * ip represents the actual file and not the hardlink marker.
1931 ip = VTOI(fncp->nc_vp);
1936 * The common parent directory must be locked first to avoid deadlocks.
1937 * Also note that fdip and/or tdip might match cdip.
1939 * WARNING! fdip may not match ip->pip. That is, if the source file
1940 * is already a hardlink then what we are renaming is the
1941 * hardlink pointer, not the hardlink itself. The hardlink
1942 * directory (ip->pip) will already be at a common parent
1945 * Be sure to use ip->pip when finding the common parent
1946 * against tdip or we might accidently move the hardlink
1947 * target into a subdirectory that makes it inaccessible to
1950 cdip = hammer2_inode_common_parent(ip->pip, tdip);
1951 cdcluster = hammer2_inode_lock(cdip, HAMMER2_RESOLVE_ALWAYS);
1952 fdcluster = hammer2_inode_lock(fdip, HAMMER2_RESOLVE_ALWAYS);
1953 tdcluster = hammer2_inode_lock(tdip, HAMMER2_RESOLVE_ALWAYS);
1956 * Keep a tight grip on the inode so the temporary unlinking from
1957 * the source location prior to linking to the target location
1958 * does not cause the cluster to be destroyed.
1960 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1961 * unlinking elements from their directories. Locking
1962 * the nlinks field does not lock the whole inode.
1964 hammer2_inode_ref(ip);
1967 * Remove target if it exists.
1969 error = hammer2_unlink_file(&trans, tdip, tname, tname_len,
1970 -1, NULL, ap->a_tnch, -1);
1972 if (error && error != ENOENT)
1976 * When renaming a hardlinked file we may have to re-consolidate
1977 * the location of the hardlink target.
1979 * If ip represents a regular file the consolidation code essentially
1980 * does nothing other than return the same locked cluster that was
1983 * The returned cluster will be locked.
1985 * WARNING! We do not currently have a local copy of ipdata but
1986 * we do use one later remember that it must be reloaded
1987 * on any modification to the inode, including connects.
1989 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
1990 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1991 cdip, cdcluster, 0);
1996 * Disconnect (fdip, fname) from the source directory. This will
1997 * disconnect (ip) if it represents a direct file. If (ip) represents
1998 * a hardlink the HARDLINK pointer object will be removed but the
1999 * hardlink will stay intact.
2001 * Always pass nch as NULL because we intend to reconnect the inode,
2002 * so we don't want hammer2_unlink_file() to rename it to the hidden
2003 * open-but-unlinked directory.
2005 * The target cluster may be marked DELETED but will not be destroyed
2006 * since we retain our hold on ip and cluster.
2008 * NOTE: We pass nlinks as 0 (not -1) in order to retain the file's
2011 error = hammer2_unlink_file(&trans, fdip, fname, fname_len,
2012 -1, &hlink, NULL, 0);
2013 KKASSERT(error != EAGAIN);
2018 * Reconnect ip to target directory using cluster. Chains cannot
2019 * actually be moved, so this will duplicate the cluster in the new
2020 * spot and assign it to the ip, replacing the old cluster.
2022 * WARNING: Because recursive locks are allowed and we unlinked the
2023 * file that we have a cluster-in-hand for just above, the
2024 * cluster might have been delete-duplicated. We must
2025 * refactor the cluster.
2027 * WARNING: Chain locks can lock buffer cache buffers, to avoid
2028 * deadlocks we want to unlock before issuing a cache_*()
2029 * op (that might have to lock a vnode).
2031 * NOTE: Pass nlinks as 0 because we retained the link count from
2032 * the unlink, so we do not have to modify it.
2034 error = hammer2_inode_connect(&trans, &cluster, hlink,
2036 tname, tname_len, 0);
2038 KKASSERT(cluster != NULL);
2039 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), cluster);
2042 hammer2_inode_unlock(ip, cluster);
2043 hammer2_inode_unlock(tdip, tdcluster);
2044 hammer2_inode_unlock(fdip, fdcluster);
2045 hammer2_inode_unlock(cdip, cdcluster);
2046 hammer2_inode_drop(ip);
2047 hammer2_inode_drop(cdip);
2048 hammer2_run_unlinkq(&trans, fdip->pmp);
2049 hammer2_trans_done(&trans);
2052 * Issue the namecache update after unlocking all the internal
2053 * hammer structures, otherwise we might deadlock.
2055 if (tnch_error == 0) {
2056 cache_unlink(ap->a_tnch);
2057 cache_setunresolved(ap->a_tnch);
2060 cache_rename(ap->a_fnch, ap->a_tnch);
2067 * Strategy code (async logical file buffer I/O from system)
2069 * WARNING: The strategy code cannot safely use hammer2 transactions
2070 * as this can deadlock against vfs_sync's vfsync() call
2071 * if multiple flushes are queued. All H2 structures must
2072 * already be present and ready for the DIO.
2074 * Reads can be initiated asynchronously, writes have to be
2075 * spooled to a separate thread for action to avoid deadlocks.
2077 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2078 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2079 static void hammer2_strategy_read_callback(hammer2_iocb_t *iocb);
2083 hammer2_vop_strategy(struct vop_strategy_args *ap)
2094 error = hammer2_strategy_read(ap);
2095 ++hammer2_iod_file_read;
2098 error = hammer2_strategy_write(ap);
2099 ++hammer2_iod_file_write;
2102 bp->b_error = error = EINVAL;
2103 bp->b_flags |= B_ERROR;
2111 * Logical buffer I/O, async read.
2115 hammer2_strategy_read(struct vop_strategy_args *ap)
2120 hammer2_inode_t *ip;
2121 hammer2_cluster_t *cparent;
2122 hammer2_cluster_t *cluster;
2123 hammer2_key_t key_dummy;
2124 hammer2_key_t lbase;
2129 ip = VTOI(ap->a_vp);
2130 nbio = push_bio(bio);
2132 lbase = bio->bio_offset;
2133 KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2136 * Lookup the file offset.
2138 cparent = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS |
2139 HAMMER2_RESOLVE_SHARED);
2140 cluster = hammer2_cluster_lookup(cparent, &key_dummy,
2142 HAMMER2_LOOKUP_NODATA |
2143 HAMMER2_LOOKUP_SHARED);
2144 hammer2_inode_unlock(ip, cparent);
2147 * Data is zero-fill if no cluster could be found
2148 * (XXX or EIO on a cluster failure).
2150 if (cluster == NULL) {
2153 bzero(bp->b_data, bp->b_bcount);
2159 * Cluster elements must be type INODE or type DATA, but the
2160 * compression mode (or not) for DATA chains can be different for
2161 * each chain. This will be handled by the callback.
2163 * If the cluster already has valid data the callback will be made
2164 * immediately/synchronously.
2166 btype = hammer2_cluster_type(cluster);
2167 if (btype != HAMMER2_BREF_TYPE_INODE &&
2168 btype != HAMMER2_BREF_TYPE_DATA) {
2169 panic("READ PATH: hammer2_strategy_read: unknown bref type");
2171 hammer2_cluster_load_async(cluster, hammer2_strategy_read_callback,
2177 * Read callback for hammer2_cluster_load_async(). The load function may
2178 * start several actual I/Os but will only make one callback, typically with
2179 * the first valid I/O XXX
2183 hammer2_strategy_read_callback(hammer2_iocb_t *iocb)
2185 struct bio *bio = iocb->ptr; /* original logical buffer */
2186 struct buf *bp = bio->bio_buf; /* original logical buffer */
2187 hammer2_chain_t *chain;
2188 hammer2_cluster_t *cluster;
2194 * Extract data and handle iteration on I/O failure. iocb->off
2195 * is the cluster index for iteration.
2197 cluster = iocb->cluster;
2198 dio = iocb->dio; /* can be NULL if iocb not in progress */
2201 * Work to do if INPROG set, else dio is already good or dio is
2202 * NULL (which is the shortcut case if chain->data is already good).
2204 if (iocb->flags & HAMMER2_IOCB_INPROG) {
2206 * Read attempt not yet made. Issue an asynchronous read
2207 * if necessary and return, operation will chain back to
2210 if ((iocb->flags & HAMMER2_IOCB_READ) == 0) {
2211 if (dio->bp == NULL ||
2212 (dio->bp->b_flags & B_CACHE) == 0) {
2217 iocb->flags |= HAMMER2_IOCB_READ;
2218 breadcb(dio->hmp->devvp,
2219 dio->pbase, dio->psize,
2220 hammer2_io_callback, iocb);
2227 * If we have a DIO it is now done, check for an error and
2228 * calculate the data.
2230 * If there is no DIO it is an optimization by
2231 * hammer2_cluster_load_async(), the data is available in
2235 if (dio->bp->b_flags & B_ERROR) {
2236 i = (int)iocb->lbase + 1;
2237 if (i >= cluster->nchains) {
2238 bp->b_flags |= B_ERROR;
2239 bp->b_error = dio->bp->b_error;
2240 hammer2_io_complete(iocb);
2242 hammer2_cluster_unlock(cluster);
2243 hammer2_cluster_drop(cluster);
2245 hammer2_io_complete(iocb); /* XXX */
2246 chain = cluster->array[i].chain;
2247 kprintf("hammer2: IO CHAIN-%d %p\n", i, chain);
2248 hammer2_adjreadcounter(&chain->bref,
2250 iocb->chain = chain;
2251 iocb->lbase = (off_t)i;
2254 hammer2_io_getblk(chain->hmp,
2255 chain->bref.data_off,
2261 chain = iocb->chain;
2262 data = hammer2_io_data(dio, chain->bref.data_off);
2265 * Special synchronous case, data present in chain->data.
2267 chain = iocb->chain;
2268 data = (void *)chain->data;
2271 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2273 * Data is embedded in the inode (copy from inode).
2275 bcopy(((hammer2_inode_data_t *)data)->u.data,
2276 bp->b_data, HAMMER2_EMBEDDED_BYTES);
2277 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2278 bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2281 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2283 * Data is on-media, issue device I/O and copy.
2285 * XXX direct-IO shortcut could go here XXX.
2287 switch (HAMMER2_DEC_COMP(chain->bref.methods)) {
2288 case HAMMER2_COMP_LZ4:
2289 hammer2_decompress_LZ4_callback(data, chain->bytes,
2292 case HAMMER2_COMP_ZLIB:
2293 hammer2_decompress_ZLIB_callback(data, chain->bytes,
2296 case HAMMER2_COMP_NONE:
2297 KKASSERT(chain->bytes <= bp->b_bcount);
2298 bcopy(data, bp->b_data, chain->bytes);
2299 if (chain->bytes < bp->b_bcount) {
2300 bzero(bp->b_data + chain->bytes,
2301 bp->b_bcount - chain->bytes);
2303 bp->b_flags |= B_NOTMETA;
2308 panic("hammer2_strategy_read: "
2309 "unknown compression type");
2312 /* bqrelse the dio to help stabilize the call to panic() */
2314 hammer2_io_bqrelse(&dio);
2315 panic("hammer2_strategy_read: unknown bref type");
2319 * Once the iocb is cleaned up the DIO (if any) will no longer be
2320 * in-progress but will still have a ref. Be sure to release
2323 hammer2_io_complete(iocb); /* physical management */
2324 if (dio) /* physical dio & buffer */
2325 hammer2_io_bqrelse(&dio);
2326 hammer2_cluster_unlock(cluster); /* cluster management */
2327 hammer2_cluster_drop(cluster); /* cluster management */
2328 biodone(bio); /* logical buffer */
2333 hammer2_strategy_write(struct vop_strategy_args *ap)
2338 hammer2_inode_t *ip;
2342 ip = VTOI(ap->a_vp);
2345 hammer2_lwinprog_ref(pmp);
2346 hammer2_trans_assert_strategy(pmp);
2347 hammer2_mtx_ex(&pmp->wthread_mtx);
2348 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
2349 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2350 hammer2_mtx_unlock(&pmp->wthread_mtx);
2351 wakeup(&pmp->wthread_bioq);
2353 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2354 hammer2_mtx_unlock(&pmp->wthread_mtx);
2356 hammer2_lwinprog_wait(pmp);
2362 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2366 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2368 hammer2_inode_t *ip;
2372 ip = VTOI(ap->a_vp);
2374 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2375 ap->a_fflag, ap->a_cred);
2382 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2390 case (MOUNTCTL_SET_EXPORT):
2391 mp = ap->a_head.a_ops->head.vv_mount;
2394 if (ap->a_ctllen != sizeof(struct export_args))
2397 rc = vfs_export(mp, &pmp->export,
2398 (const struct export_args *)ap->a_ctl);
2401 rc = vop_stdmountctl(ap);
2409 * This handles unlinked open files after the vnode is finally dereferenced.
2410 * To avoid deadlocks it cannot be called from the normal vnode recycling
2411 * path, so we call it (1) after a unlink, rmdir, or rename, (2) on every
2412 * flush, and (3) on umount.
2415 hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfs_t *pmp)
2417 const hammer2_inode_data_t *ripdata;
2418 hammer2_inode_unlink_t *ipul;
2419 hammer2_inode_t *ip;
2420 hammer2_cluster_t *cluster;
2421 hammer2_cluster_t *cparent;
2423 if (TAILQ_EMPTY(&pmp->unlinkq))
2427 hammer2_spin_ex(&pmp->list_spin);
2428 while ((ipul = TAILQ_FIRST(&pmp->unlinkq)) != NULL) {
2429 TAILQ_REMOVE(&pmp->unlinkq, ipul, entry);
2430 hammer2_spin_unex(&pmp->list_spin);
2432 kfree(ipul, pmp->minode);
2434 cluster = hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
2435 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
2436 if (hammer2_debug & 0x400) {
2437 kprintf("hammer2: unlink on reclaim: %s refs=%d\n",
2438 ripdata->filename, ip->refs);
2440 KKASSERT(ripdata->nlinks == 0);
2442 cparent = hammer2_cluster_parent(cluster);
2443 hammer2_cluster_delete(trans, cparent, cluster,
2444 HAMMER2_DELETE_PERMANENT);
2445 hammer2_cluster_unlock(cparent);
2446 hammer2_cluster_drop(cparent);
2447 hammer2_inode_unlock(ip, cluster); /* inode lock */
2448 hammer2_inode_drop(ip); /* ipul ref */
2450 hammer2_spin_ex(&pmp->list_spin);
2452 hammer2_spin_unex(&pmp->list_spin);
2460 static void filt_hammer2detach(struct knote *kn);
2461 static int filt_hammer2read(struct knote *kn, long hint);
2462 static int filt_hammer2write(struct knote *kn, long hint);
2463 static int filt_hammer2vnode(struct knote *kn, long hint);
2465 static struct filterops hammer2read_filtops =
2466 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2467 NULL, filt_hammer2detach, filt_hammer2read };
2468 static struct filterops hammer2write_filtops =
2469 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2470 NULL, filt_hammer2detach, filt_hammer2write };
2471 static struct filterops hammer2vnode_filtops =
2472 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2473 NULL, filt_hammer2detach, filt_hammer2vnode };
2477 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2479 struct vnode *vp = ap->a_vp;
2480 struct knote *kn = ap->a_kn;
2482 switch (kn->kn_filter) {
2484 kn->kn_fop = &hammer2read_filtops;
2487 kn->kn_fop = &hammer2write_filtops;
2490 kn->kn_fop = &hammer2vnode_filtops;
2493 return (EOPNOTSUPP);
2496 kn->kn_hook = (caddr_t)vp;
2498 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2504 filt_hammer2detach(struct knote *kn)
2506 struct vnode *vp = (void *)kn->kn_hook;
2508 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2512 filt_hammer2read(struct knote *kn, long hint)
2514 struct vnode *vp = (void *)kn->kn_hook;
2515 hammer2_inode_t *ip = VTOI(vp);
2518 if (hint == NOTE_REVOKE) {
2519 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2522 off = ip->size - kn->kn_fp->f_offset;
2523 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2524 if (kn->kn_sfflags & NOTE_OLDAPI)
2526 return (kn->kn_data != 0);
2531 filt_hammer2write(struct knote *kn, long hint)
2533 if (hint == NOTE_REVOKE)
2534 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2540 filt_hammer2vnode(struct knote *kn, long hint)
2542 if (kn->kn_sfflags & hint)
2543 kn->kn_fflags |= hint;
2544 if (hint == NOTE_REVOKE) {
2545 kn->kn_flags |= (EV_EOF | EV_NODATA);
2548 return (kn->kn_fflags != 0);
2556 hammer2_vop_markatime(struct vop_markatime_args *ap)
2558 hammer2_inode_t *ip;
2571 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2575 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2577 error = hammer2_vop_kqfilter(ap);
2584 struct vop_ops hammer2_vnode_vops = {
2585 .vop_default = vop_defaultop,
2586 .vop_fsync = hammer2_vop_fsync,
2587 .vop_getpages = vop_stdgetpages,
2588 .vop_putpages = vop_stdputpages,
2589 .vop_access = hammer2_vop_access,
2590 .vop_advlock = hammer2_vop_advlock,
2591 .vop_close = hammer2_vop_close,
2592 .vop_nlink = hammer2_vop_nlink,
2593 .vop_ncreate = hammer2_vop_ncreate,
2594 .vop_nsymlink = hammer2_vop_nsymlink,
2595 .vop_nremove = hammer2_vop_nremove,
2596 .vop_nrmdir = hammer2_vop_nrmdir,
2597 .vop_nrename = hammer2_vop_nrename,
2598 .vop_getattr = hammer2_vop_getattr,
2599 .vop_setattr = hammer2_vop_setattr,
2600 .vop_readdir = hammer2_vop_readdir,
2601 .vop_readlink = hammer2_vop_readlink,
2602 .vop_getpages = vop_stdgetpages,
2603 .vop_putpages = vop_stdputpages,
2604 .vop_read = hammer2_vop_read,
2605 .vop_write = hammer2_vop_write,
2606 .vop_open = hammer2_vop_open,
2607 .vop_inactive = hammer2_vop_inactive,
2608 .vop_reclaim = hammer2_vop_reclaim,
2609 .vop_nresolve = hammer2_vop_nresolve,
2610 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2611 .vop_nmkdir = hammer2_vop_nmkdir,
2612 .vop_nmknod = hammer2_vop_nmknod,
2613 .vop_ioctl = hammer2_vop_ioctl,
2614 .vop_mountctl = hammer2_vop_mountctl,
2615 .vop_bmap = hammer2_vop_bmap,
2616 .vop_strategy = hammer2_vop_strategy,
2617 .vop_kqfilter = hammer2_vop_kqfilter
2620 struct vop_ops hammer2_spec_vops = {
2621 .vop_default = vop_defaultop,
2622 .vop_fsync = hammer2_vop_fsync,
2623 .vop_read = vop_stdnoread,
2624 .vop_write = vop_stdnowrite,
2625 .vop_access = hammer2_vop_access,
2626 .vop_close = hammer2_vop_close,
2627 .vop_markatime = hammer2_vop_markatime,
2628 .vop_getattr = hammer2_vop_getattr,
2629 .vop_inactive = hammer2_vop_inactive,
2630 .vop_reclaim = hammer2_vop_reclaim,
2631 .vop_setattr = hammer2_vop_setattr
2634 struct vop_ops hammer2_fifo_vops = {
2635 .vop_default = fifo_vnoperate,
2636 .vop_fsync = hammer2_vop_fsync,
2638 .vop_read = hammer2_vop_fiforead,
2639 .vop_write = hammer2_vop_fifowrite,
2641 .vop_access = hammer2_vop_access,
2643 .vop_close = hammer2_vop_fifoclose,
2645 .vop_markatime = hammer2_vop_markatime,
2646 .vop_getattr = hammer2_vop_getattr,
2647 .vop_inactive = hammer2_vop_inactive,
2648 .vop_reclaim = hammer2_vop_reclaim,
2649 .vop_setattr = hammer2_vop_setattr,
2650 .vop_kqfilter = hammer2_vop_fifokqfilter