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_nex(ip, HAMMER2_RESOLVE_NEVER);
213 * Check for deleted inodes and recycle immediately.
215 * WARNING: nvtruncbuf() can only be safely called without the inode
216 * lock held due to the way our write thread works.
218 if (hammer2_cluster_isunlinked(cluster)) {
222 nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
223 hammer2_inode_unlock_ex(ip, cluster);
224 nvtruncbuf(vp, 0, nblksize, 0, 0);
227 hammer2_inode_unlock_ex(ip, cluster);
234 * Reclaim a vnode so that it can be reused; after the inode is
235 * disassociated, the filesystem must manage it alone.
239 hammer2_vop_reclaim(struct vop_reclaim_args *ap)
241 hammer2_cluster_t *cluster;
243 hammer2_pfsmount_t *pmp;
255 * Inode must be locked for reclaim.
258 cluster = hammer2_inode_lock_nex(ip, HAMMER2_RESOLVE_NEVER);
261 * The final close of a deleted file or directory marks it for
262 * destruction. The DELETED flag allows the flusher to shortcut
263 * any modified blocks still unflushed (that is, just ignore them).
265 * HAMMER2 usually does not try to optimize the freemap by returning
266 * deleted blocks to it as it does not usually know how many snapshots
267 * might be referencing portions of the file/dir.
273 * NOTE! We do not attempt to flush chains here, flushing is
274 * really fragile and could also deadlock.
279 * A reclaim can occur at any time so we cannot safely start a
280 * transaction to handle reclamation of unlinked files. Instead,
281 * the ip is left with a reference and placed on a linked list and
284 if (hammer2_cluster_isunlinked(cluster)) {
285 hammer2_inode_unlink_t *ipul;
287 ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
290 spin_lock(&pmp->list_spin);
291 TAILQ_INSERT_TAIL(&pmp->unlinkq, ipul, entry);
292 spin_unlock(&pmp->list_spin);
293 hammer2_inode_unlock_ex(ip, cluster); /* unlock */
294 /* retain ref from vp for ipul */
296 hammer2_inode_unlock_ex(ip, cluster); /* unlock */
297 hammer2_inode_drop(ip); /* vp ref */
299 /* cluster no longer referenced */
300 /* cluster = NULL; not needed */
303 * XXX handle background sync when ip dirty, kernel will no longer
304 * notify us regarding this inode because there is no longer a
305 * vnode attached to it.
314 hammer2_vop_fsync(struct vop_fsync_args *ap)
317 hammer2_trans_t trans;
318 hammer2_cluster_t *cluster;
326 /* XXX can't do this yet */
327 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_ISFLUSH);
328 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
330 hammer2_trans_init(&trans, ip->pmp, 0);
331 vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
334 * Calling chain_flush here creates a lot of duplicative
335 * COW operations due to non-optimal vnode ordering.
337 * Only do it for an actual fsync() syscall. The other forms
338 * which call this function will eventually call chain_flush
339 * on the volume root as a catch-all, which is far more optimal.
341 cluster = hammer2_inode_lock_ex(ip);
342 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
344 if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MTIME))
345 hammer2_inode_fsync(&trans, ip, cluster);
349 * XXX creates discontinuity w/modify_tid
351 if (ap->a_flags & VOP_FSYNC_SYSCALL) {
352 hammer2_flush(&trans, cluster);
355 hammer2_inode_unlock_ex(ip, cluster);
356 hammer2_trans_done(&trans);
364 hammer2_vop_access(struct vop_access_args *ap)
366 hammer2_inode_t *ip = VTOI(ap->a_vp);
367 const hammer2_inode_data_t *ripdata;
368 hammer2_cluster_t *cluster;
374 cluster = hammer2_inode_lock_sh(ip);
375 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
376 uid = hammer2_to_unix_xid(&ripdata->uid);
377 gid = hammer2_to_unix_xid(&ripdata->gid);
378 error = vop_helper_access(ap, uid, gid, ripdata->mode, ripdata->uflags);
379 hammer2_inode_unlock_sh(ip, cluster);
387 hammer2_vop_getattr(struct vop_getattr_args *ap)
389 const hammer2_inode_data_t *ripdata;
390 hammer2_cluster_t *cluster;
391 hammer2_pfsmount_t *pmp;
403 cluster = hammer2_inode_lock_sh(ip);
404 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
405 KKASSERT(hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
407 vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
408 vap->va_fileid = ripdata->inum;
409 vap->va_mode = ripdata->mode;
410 vap->va_nlink = ripdata->nlinks;
411 vap->va_uid = hammer2_to_unix_xid(&ripdata->uid);
412 vap->va_gid = hammer2_to_unix_xid(&ripdata->gid);
415 vap->va_size = ip->size; /* protected by shared lock */
416 vap->va_blocksize = HAMMER2_PBUFSIZE;
417 vap->va_flags = ripdata->uflags;
418 hammer2_time_to_timespec(ripdata->ctime, &vap->va_ctime);
419 hammer2_time_to_timespec(ripdata->mtime, &vap->va_mtime);
420 hammer2_time_to_timespec(ripdata->mtime, &vap->va_atime);
422 vap->va_bytes = vap->va_size; /* XXX */
423 vap->va_type = hammer2_get_vtype(ripdata);
425 vap->va_uid_uuid = ripdata->uid;
426 vap->va_gid_uuid = ripdata->gid;
427 vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
430 hammer2_inode_unlock_sh(ip, cluster);
438 hammer2_vop_setattr(struct vop_setattr_args *ap)
440 const hammer2_inode_data_t *ripdata;
441 hammer2_inode_data_t *wipdata;
443 hammer2_cluster_t *cluster;
444 hammer2_trans_t trans;
456 hammer2_update_time(&ctime);
460 if (ip->pmp->ronly) {
465 hammer2_pfs_memory_wait(ip->pmp);
466 hammer2_trans_init(&trans, ip->pmp, 0);
467 cluster = hammer2_inode_lock_ex(ip);
468 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
471 if (vap->va_flags != VNOVAL) {
474 flags = ripdata->uflags;
475 error = vop_helper_setattr_flags(&flags, vap->va_flags,
476 hammer2_to_unix_xid(&ripdata->uid),
479 if (ripdata->uflags != flags) {
480 wipdata = hammer2_cluster_modify_ip(&trans, ip,
482 wipdata->uflags = flags;
483 wipdata->ctime = ctime;
484 kflags |= NOTE_ATTRIB;
488 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
495 if (ripdata->uflags & (IMMUTABLE | APPEND)) {
499 if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
500 mode_t cur_mode = ripdata->mode;
501 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
502 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
506 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
508 &cur_uid, &cur_gid, &cur_mode);
510 hammer2_guid_to_uuid(&uuid_uid, cur_uid);
511 hammer2_guid_to_uuid(&uuid_gid, cur_gid);
512 if (bcmp(&uuid_uid, &ripdata->uid, sizeof(uuid_uid)) ||
513 bcmp(&uuid_gid, &ripdata->gid, sizeof(uuid_gid)) ||
514 ripdata->mode != cur_mode
516 wipdata = hammer2_cluster_modify_ip(&trans, ip,
518 wipdata->uid = uuid_uid;
519 wipdata->gid = uuid_gid;
520 wipdata->mode = cur_mode;
521 wipdata->ctime = ctime;
525 kflags |= NOTE_ATTRIB;
532 if (vap->va_size != VNOVAL && ip->size != vap->va_size) {
535 if (vap->va_size == ip->size)
537 hammer2_inode_unlock_ex(ip, cluster);
538 if (vap->va_size < ip->size) {
539 hammer2_truncate_file(ip, vap->va_size);
541 hammer2_extend_file(ip, vap->va_size);
543 cluster = hammer2_inode_lock_ex(ip);
545 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
554 /* atime not supported */
555 if (vap->va_atime.tv_sec != VNOVAL) {
556 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
557 wipdata->atime = hammer2_timespec_to_time(&vap->va_atime);
558 kflags |= NOTE_ATTRIB;
563 if (vap->va_mtime.tv_sec != VNOVAL) {
564 wipdata = hammer2_cluster_modify_ip(&trans, ip, cluster, 0);
565 wipdata->mtime = hammer2_timespec_to_time(&vap->va_mtime);
566 kflags |= NOTE_ATTRIB;
571 if (vap->va_mode != (mode_t)VNOVAL) {
572 mode_t cur_mode = ripdata->mode;
573 uid_t cur_uid = hammer2_to_unix_xid(&ripdata->uid);
574 gid_t cur_gid = hammer2_to_unix_xid(&ripdata->gid);
576 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
577 cur_uid, cur_gid, &cur_mode);
578 if (error == 0 && ripdata->mode != cur_mode) {
579 wipdata = hammer2_cluster_modify_ip(&trans, ip,
581 wipdata->mode = cur_mode;
582 wipdata->ctime = ctime;
583 kflags |= NOTE_ATTRIB;
590 * If a truncation occurred we must call inode_fsync() now in order
591 * to trim the related data chains, otherwise a later expansion can
595 hammer2_cluster_modsync(cluster);
598 hammer2_inode_fsync(&trans, ip, cluster);
601 * Cleanup. If domtime is set an additional inode modification
602 * must be flagged. All other modifications will have already
603 * set INODE_MODIFIED and called vsetisdirty().
607 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED |
608 HAMMER2_INODE_MTIME);
612 hammer2_cluster_modsync(cluster);
613 hammer2_inode_unlock_ex(ip, cluster);
614 hammer2_trans_done(&trans);
615 hammer2_knote(ip->vp, kflags);
623 hammer2_vop_readdir(struct vop_readdir_args *ap)
625 const hammer2_inode_data_t *ripdata;
627 hammer2_inode_t *xip;
628 hammer2_cluster_t *cparent;
629 hammer2_cluster_t *cluster;
630 hammer2_cluster_t *xcluster;
631 hammer2_blockref_t bref;
633 hammer2_key_t key_next;
648 saveoff = uio->uio_offset;
651 * Setup cookies directory entry cookies if requested
653 if (ap->a_ncookies) {
654 ncookies = uio->uio_resid / 16 + 1;
657 cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
664 cparent = hammer2_inode_lock_sh(ip);
665 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
668 * Handle artificial entries. To ensure that only positive 64 bit
669 * quantities are returned to userland we always strip off bit 63.
670 * The hash code is designed such that codes 0x0000-0x7FFF are not
671 * used, allowing us to use these codes for articial entries.
673 * Entry 0 is used for '.' and entry 1 is used for '..'. Do not
674 * allow '..' to cross the mount point into (e.g.) the super-root.
677 cluster = (void *)(intptr_t)-1; /* non-NULL for early goto done case */
680 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
681 r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
685 cookies[cookie_index] = saveoff;
688 if (cookie_index == ncookies)
694 * Be careful with lockorder when accessing ".."
696 * (ip is the current dir. xip is the parent dir).
698 inum = ripdata->inum & HAMMER2_DIRHASH_USERMSK;
699 while (ip->pip != NULL && ip != ip->pmp->iroot) {
701 hammer2_inode_ref(xip);
702 hammer2_inode_unlock_sh(ip, cparent);
703 xcluster = hammer2_inode_lock_sh(xip);
704 cparent = hammer2_inode_lock_sh(ip);
705 hammer2_inode_drop(xip);
706 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
707 if (xip == ip->pip) {
708 inum = hammer2_cluster_rdata(xcluster)->
709 ipdata.inum & HAMMER2_DIRHASH_USERMSK;
710 hammer2_inode_unlock_sh(xip, xcluster);
713 hammer2_inode_unlock_sh(xip, xcluster);
715 r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
719 cookies[cookie_index] = saveoff;
722 if (cookie_index == ncookies)
726 lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
727 if (hammer2_debug & 0x0020)
728 kprintf("readdir: lkey %016jx\n", lkey);
731 * parent is the inode cluster, already locked for us. Don't
732 * double lock shared locks as this will screw up upgrades.
737 cluster = hammer2_cluster_lookup(cparent, &key_next, lkey, lkey,
738 HAMMER2_LOOKUP_SHARED, &ddflag);
739 if (cluster == NULL) {
740 cluster = hammer2_cluster_lookup(cparent, &key_next,
741 lkey, (hammer2_key_t)-1,
742 HAMMER2_LOOKUP_SHARED, &ddflag);
745 hammer2_cluster_bref(cluster, &bref);
747 if (hammer2_debug & 0x0020)
748 kprintf("readdir: p=%p chain=%p %016jx (next %016jx)\n",
749 cparent->focus, cluster->focus,
752 if (bref.type == HAMMER2_BREF_TYPE_INODE) {
753 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
754 dtype = hammer2_get_dtype(ripdata);
755 saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
756 r = vop_write_dirent(&error, uio,
758 HAMMER2_DIRHASH_USERMSK,
765 cookies[cookie_index] = saveoff;
768 /* XXX chain error */
769 kprintf("bad chain type readdir %d\n", bref.type);
773 * Keys may not be returned in order so once we have a
774 * placemarker (cluster) the scan must allow the full range
775 * or some entries will be missed.
777 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
778 key_next, (hammer2_key_t)-1,
779 HAMMER2_LOOKUP_SHARED);
781 hammer2_cluster_bref(cluster, &bref);
782 saveoff = (bref.key & HAMMER2_DIRHASH_USERMSK) + 1;
784 saveoff = (hammer2_key_t)-1;
786 if (cookie_index == ncookies)
790 hammer2_cluster_unlock(cluster);
792 hammer2_inode_unlock_sh(ip, cparent);
794 *ap->a_eofflag = (cluster == NULL);
795 if (hammer2_debug & 0x0020)
796 kprintf("readdir: done at %016jx\n", saveoff);
797 uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
798 if (error && cookie_index == 0) {
800 kfree(cookies, M_TEMP);
802 *ap->a_cookies = NULL;
806 *ap->a_ncookies = cookie_index;
807 *ap->a_cookies = cookies;
815 * hammer2_vop_readlink { vp, uio, cred }
819 hammer2_vop_readlink(struct vop_readlink_args *ap)
826 if (vp->v_type != VLNK)
830 error = hammer2_read_file(ip, ap->a_uio, 0);
836 hammer2_vop_read(struct vop_read_args *ap)
846 * Read operations supported on this vnode?
849 if (vp->v_type != VREG)
859 seqcount = ap->a_ioflag >> 16;
860 bigread = (uio->uio_resid > 100 * 1024 * 1024);
862 error = hammer2_read_file(ip, uio, seqcount);
868 hammer2_vop_write(struct vop_write_args *ap)
871 hammer2_trans_t trans;
880 * Read operations supported on this vnode?
883 if (vp->v_type != VREG)
892 if (ip->pmp->ronly) {
896 seqcount = ap->a_ioflag >> 16;
897 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
900 * Check resource limit
902 if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
903 uio->uio_offset + uio->uio_resid >
904 td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
905 lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
909 bigwrite = (uio->uio_resid > 100 * 1024 * 1024);
912 * The transaction interlocks against flushes initiations
913 * (note: but will run concurrently with the actual flush).
915 hammer2_trans_init(&trans, ip->pmp, 0);
916 error = hammer2_write_file(ip, uio, ap->a_ioflag, seqcount);
917 hammer2_trans_done(&trans);
923 * Perform read operations on a file or symlink given an UNLOCKED
926 * The passed ip is not locked.
930 hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
941 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
943 ccms_thread_unlock(&ip->topo_cst);
945 while (uio->uio_resid > 0 && uio->uio_offset < size) {
952 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
955 error = cluster_read(ip->vp, leof, lbase, lblksize,
956 uio->uio_resid, seqcount * BKVASIZE,
961 loff = (int)(uio->uio_offset - lbase);
963 if (n > uio->uio_resid)
965 if (n > size - uio->uio_offset)
966 n = (int)(size - uio->uio_offset);
967 bp->b_flags |= B_AGE;
968 uiomove((char *)bp->b_data + loff, n, uio);
975 * Write to the file represented by the inode via the logical buffer cache.
976 * The inode may represent a regular file or a symlink.
978 * The inode must not be locked.
982 hammer2_write_file(hammer2_inode_t *ip,
983 struct uio *uio, int ioflag, int seqcount)
985 hammer2_key_t old_eof;
986 hammer2_key_t new_eof;
995 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
996 if (ioflag & IO_APPEND)
997 uio->uio_offset = ip->size;
999 ccms_thread_unlock(&ip->topo_cst);
1002 * Extend the file if necessary. If the write fails at some point
1003 * we will truncate it back down to cover as much as we were able
1006 * Doing this now makes it easier to calculate buffer sizes in
1013 if (uio->uio_offset + uio->uio_resid > old_eof) {
1014 new_eof = uio->uio_offset + uio->uio_resid;
1016 hammer2_extend_file(ip, new_eof);
1017 kflags |= NOTE_EXTEND;
1025 while (uio->uio_resid > 0) {
1026 hammer2_key_t lbase;
1034 * Don't allow the buffer build to blow out the buffer
1037 if ((ioflag & IO_RECURSE) == 0)
1038 bwillwrite(HAMMER2_PBUFSIZE);
1041 * This nominally tells us how much we can cluster and
1042 * what the logical buffer size needs to be. Currently
1043 * we don't try to cluster the write and just handle one
1046 lblksize = hammer2_calc_logical(ip, uio->uio_offset,
1048 loff = (int)(uio->uio_offset - lbase);
1050 KKASSERT(lblksize <= 65536);
1053 * Calculate bytes to copy this transfer and whether the
1054 * copy completely covers the buffer or not.
1057 n = lblksize - loff;
1058 if (n > uio->uio_resid) {
1060 if (loff == lbase && uio->uio_offset + n == new_eof)
1072 if (uio->uio_segflg == UIO_NOCOPY) {
1074 * Issuing a write with the same data backing the
1075 * buffer. Instantiate the buffer to collect the
1076 * backing vm pages, then read-in any missing bits.
1078 * This case is used by vop_stdputpages().
1080 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1081 if ((bp->b_flags & B_CACHE) == 0) {
1083 error = bread(ip->vp, lbase, lblksize, &bp);
1085 } else if (trivial) {
1087 * Even though we are entirely overwriting the buffer
1088 * we may still have to zero it out to avoid a
1089 * mmap/write visibility issue.
1091 bp = getblk(ip->vp, lbase, lblksize, GETBLK_BHEAVY, 0);
1092 if ((bp->b_flags & B_CACHE) == 0)
1096 * Partial overwrite, read in any missing bits then
1097 * replace the portion being written.
1099 * (The strategy code will detect zero-fill physical
1100 * blocks for this case).
1102 error = bread(ip->vp, lbase, lblksize, &bp);
1113 * Ok, copy the data in
1115 error = uiomove(bp->b_data + loff, n, uio);
1116 kflags |= NOTE_WRITE;
1124 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
1125 * with IO_SYNC or IO_ASYNC set. These writes
1126 * must be handled as the pageout daemon expects.
1128 if (ioflag & IO_SYNC) {
1130 } else if ((ioflag & IO_DIRECT) && endofblk) {
1132 } else if (ioflag & IO_ASYNC) {
1140 * Cleanup. If we extended the file EOF but failed to write through
1141 * the entire write is a failure and we have to back-up.
1143 if (error && new_eof != old_eof) {
1144 hammer2_truncate_file(ip, old_eof);
1145 } else if (modified) {
1146 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1147 hammer2_update_time(&ip->mtime);
1148 atomic_set_int(&ip->flags, HAMMER2_INODE_MTIME);
1149 ccms_thread_unlock(&ip->topo_cst);
1151 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1152 hammer2_knote(ip->vp, kflags);
1153 vsetisdirty(ip->vp);
1159 * Truncate the size of a file. The inode must not be locked.
1161 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1163 * WARNING: nvtruncbuf() can only be safely called without the inode lock
1164 * held due to the way our write thread works.
1168 hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1170 hammer2_key_t lbase;
1175 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1176 nvtruncbuf(ip->vp, nsize,
1177 nblksize, (int)nsize & (nblksize - 1),
1180 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1182 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1183 ccms_thread_unlock(&ip->topo_cst);
1188 * Extend the size of a file. The inode must not be locked.
1190 * NOTE: Caller handles setting HAMMER2_INODE_MODIFIED
1194 hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
1196 hammer2_key_t lbase;
1197 hammer2_key_t osize;
1202 ccms_thread_lock(&ip->topo_cst, CCMS_STATE_EXCLUSIVE);
1205 ccms_thread_unlock(&ip->topo_cst);
1208 oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
1209 nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
1215 atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
1221 hammer2_vop_nresolve(struct vop_nresolve_args *ap)
1223 hammer2_inode_t *ip;
1224 hammer2_inode_t *dip;
1225 hammer2_cluster_t *cparent;
1226 hammer2_cluster_t *cluster;
1227 const hammer2_inode_data_t *ripdata;
1228 hammer2_key_t key_next;
1230 struct namecache *ncp;
1231 const uint8_t *name;
1238 dip = VTOI(ap->a_dvp);
1239 ncp = ap->a_nch->ncp;
1240 name = ncp->nc_name;
1241 name_len = ncp->nc_nlen;
1242 lhc = hammer2_dirhash(name, name_len);
1245 * Note: In DragonFly the kernel handles '.' and '..'.
1247 cparent = hammer2_inode_lock_sh(dip);
1248 cluster = hammer2_cluster_lookup(cparent, &key_next,
1249 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1250 HAMMER2_LOOKUP_SHARED, &ddflag);
1252 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE) {
1253 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1254 if (ripdata->name_len == name_len &&
1255 bcmp(ripdata->filename, name, name_len) == 0) {
1259 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1261 lhc + HAMMER2_DIRHASH_LOMASK,
1262 HAMMER2_LOOKUP_SHARED);
1264 hammer2_inode_unlock_sh(dip, cparent);
1267 * Resolve hardlink entries before acquiring the inode.
1270 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1271 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1272 hammer2_tid_t inum = ripdata->inum;
1273 error = hammer2_hardlink_find(dip, NULL, cluster);
1275 kprintf("hammer2: unable to find hardlink "
1276 "0x%016jx\n", inum);
1277 hammer2_cluster_unlock(cluster);
1285 * nresolve needs to resolve hardlinks, the original cluster is not
1289 ip = hammer2_inode_get(dip->pmp, dip, cluster);
1290 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1291 if (ripdata->type == HAMMER2_OBJTYPE_HARDLINK) {
1292 kprintf("nresolve: fixup hardlink\n");
1293 hammer2_inode_ref(ip);
1294 hammer2_inode_unlock_ex(ip, NULL);
1295 hammer2_cluster_unlock(cluster);
1296 cluster = hammer2_inode_lock_ex(ip);
1297 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
1298 hammer2_inode_drop(ip);
1299 kprintf("nresolve: fixup to type %02x\n",
1308 * Deconsolidate any hardlink whos nlinks == 1. Ignore errors.
1309 * If an error occurs chain and ip are left alone.
1311 * XXX upgrade shared lock?
1313 if (ochain && chain &&
1314 chain->data->ipdata.nlinks == 1 && !dip->pmp->ronly) {
1315 kprintf("hammer2: need to unconsolidate hardlink for %s\n",
1316 chain->data->ipdata.filename);
1317 /* XXX retain shared lock on dip? (currently not held) */
1318 hammer2_trans_init(&trans, dip->pmp, 0);
1319 hammer2_hardlink_deconsolidate(&trans, dip, &chain, &ochain);
1320 hammer2_trans_done(&trans);
1325 * Acquire the related vnode
1327 * NOTE: For error processing, only ENOENT resolves the namecache
1328 * entry to NULL, otherwise we just return the error and
1329 * leave the namecache unresolved.
1331 * NOTE: multiple hammer2_inode structures can be aliased to the
1332 * same chain element, for example for hardlinks. This
1333 * use case does not 'reattach' inode associations that
1334 * might already exist, but always allocates a new one.
1336 * WARNING: inode structure is locked exclusively via inode_get
1337 * but chain was locked shared. inode_unlock_ex()
1338 * will handle it properly.
1341 vp = hammer2_igetv(ip, cluster, &error);
1344 cache_setvp(ap->a_nch, vp);
1345 } else if (error == ENOENT) {
1346 cache_setvp(ap->a_nch, NULL);
1348 hammer2_inode_unlock_ex(ip, cluster);
1351 * The vp should not be released until after we've disposed
1352 * of our locks, because it might cause vop_inactive() to
1359 cache_setvp(ap->a_nch, NULL);
1361 KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
1362 ("resolve error %d/%p ap %p\n",
1363 error, ap->a_nch->ncp->nc_vp, ap));
1370 hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
1372 hammer2_inode_t *dip;
1373 hammer2_inode_t *ip;
1374 hammer2_cluster_t *cparent;
1378 dip = VTOI(ap->a_dvp);
1380 if ((ip = dip->pip) == NULL) {
1385 cparent = hammer2_inode_lock_ex(ip);
1386 *ap->a_vpp = hammer2_igetv(ip, cparent, &error);
1387 hammer2_inode_unlock_ex(ip, cparent);
1395 hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
1397 hammer2_inode_t *dip;
1398 hammer2_inode_t *nip;
1399 hammer2_trans_t trans;
1400 hammer2_cluster_t *cluster;
1401 struct namecache *ncp;
1402 const uint8_t *name;
1407 dip = VTOI(ap->a_dvp);
1408 if (dip->pmp->ronly) {
1413 ncp = ap->a_nch->ncp;
1414 name = ncp->nc_name;
1415 name_len = ncp->nc_nlen;
1418 hammer2_pfs_memory_wait(dip->pmp);
1419 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1420 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1421 name, name_len, &cluster, &error);
1423 KKASSERT(nip == NULL);
1426 *ap->a_vpp = hammer2_igetv(nip, cluster, &error);
1427 hammer2_inode_unlock_ex(nip, cluster);
1429 hammer2_trans_done(&trans);
1432 cache_setunresolved(ap->a_nch);
1433 cache_setvp(ap->a_nch, *ap->a_vpp);
1440 * Return the largest contiguous physical disk range for the logical
1441 * request, in bytes.
1443 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
1445 * Basically disabled, the logical buffer write thread has to deal with
1446 * buffers one-at-a-time.
1450 hammer2_vop_bmap(struct vop_bmap_args *ap)
1452 *ap->a_doffsetp = NOOFFSET;
1457 return (EOPNOTSUPP);
1462 hammer2_vop_open(struct vop_open_args *ap)
1464 return vop_stdopen(ap);
1468 * hammer2_vop_advlock { vp, id, op, fl, flags }
1472 hammer2_vop_advlock(struct vop_advlock_args *ap)
1474 hammer2_inode_t *ip = VTOI(ap->a_vp);
1475 const hammer2_inode_data_t *ripdata;
1476 hammer2_cluster_t *cparent;
1479 cparent = hammer2_inode_lock_sh(ip);
1480 ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
1481 size = ripdata->size;
1482 hammer2_inode_unlock_sh(ip, cparent);
1483 return (lf_advlock(ap, &ip->advlock, size));
1489 hammer2_vop_close(struct vop_close_args *ap)
1491 return vop_stdclose(ap);
1495 * hammer2_vop_nlink { nch, dvp, vp, cred }
1497 * Create a hardlink from (vp) to {dvp, nch}.
1501 hammer2_vop_nlink(struct vop_nlink_args *ap)
1503 hammer2_inode_t *fdip; /* target directory to create link in */
1504 hammer2_inode_t *tdip; /* target directory to create link in */
1505 hammer2_inode_t *cdip; /* common parent directory */
1506 hammer2_inode_t *ip; /* inode we are hardlinking to */
1507 hammer2_cluster_t *cluster;
1508 hammer2_cluster_t *fdcluster;
1509 hammer2_cluster_t *tdcluster;
1510 hammer2_cluster_t *cdcluster;
1511 hammer2_trans_t trans;
1512 struct namecache *ncp;
1513 const uint8_t *name;
1518 tdip = VTOI(ap->a_dvp);
1519 if (tdip->pmp->ronly) {
1524 ncp = ap->a_nch->ncp;
1525 name = ncp->nc_name;
1526 name_len = ncp->nc_nlen;
1529 * ip represents the file being hardlinked. The file could be a
1530 * normal file or a hardlink target if it has already been hardlinked.
1531 * If ip is a hardlinked target then ip->pip represents the location
1532 * of the hardlinked target, NOT the location of the hardlink pointer.
1534 * Bump nlinks and potentially also create or move the hardlink
1535 * target in the parent directory common to (ip) and (tdip). The
1536 * consolidation code can modify ip->cluster and ip->pip. The
1537 * returned cluster is locked.
1539 ip = VTOI(ap->a_vp);
1540 hammer2_pfs_memory_wait(ip->pmp);
1541 hammer2_trans_init(&trans, ip->pmp, HAMMER2_TRANS_NEWINODE);
1544 * The common parent directory must be locked first to avoid deadlocks.
1545 * Also note that fdip and/or tdip might match cdip.
1548 cdip = hammer2_inode_common_parent(fdip, tdip);
1549 cdcluster = hammer2_inode_lock_ex(cdip);
1550 fdcluster = hammer2_inode_lock_ex(fdip);
1551 tdcluster = hammer2_inode_lock_ex(tdip);
1552 cluster = hammer2_inode_lock_ex(ip);
1553 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1554 cdip, cdcluster, 1);
1559 * Create a directory entry connected to the specified cluster.
1561 * WARNING! chain can get moved by the connect (indirectly due to
1562 * potential indirect block creation).
1564 error = hammer2_inode_connect(&trans, &cluster, 1,
1568 cache_setunresolved(ap->a_nch);
1569 cache_setvp(ap->a_nch, ap->a_vp);
1572 hammer2_inode_unlock_ex(ip, cluster);
1573 hammer2_inode_unlock_ex(tdip, tdcluster);
1574 hammer2_inode_unlock_ex(fdip, fdcluster);
1575 hammer2_inode_unlock_ex(cdip, cdcluster);
1576 hammer2_inode_drop(cdip);
1577 hammer2_trans_done(&trans);
1584 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
1586 * The operating system has already ensured that the directory entry
1587 * does not exist and done all appropriate namespace locking.
1591 hammer2_vop_ncreate(struct vop_ncreate_args *ap)
1593 hammer2_inode_t *dip;
1594 hammer2_inode_t *nip;
1595 hammer2_trans_t trans;
1596 hammer2_cluster_t *ncluster;
1597 struct namecache *ncp;
1598 const uint8_t *name;
1603 dip = VTOI(ap->a_dvp);
1604 if (dip->pmp->ronly) {
1609 ncp = ap->a_nch->ncp;
1610 name = ncp->nc_name;
1611 name_len = ncp->nc_nlen;
1612 hammer2_pfs_memory_wait(dip->pmp);
1613 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1616 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1617 name, name_len, &ncluster, &error);
1619 KKASSERT(nip == NULL);
1622 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1623 hammer2_inode_unlock_ex(nip, ncluster);
1625 hammer2_trans_done(&trans);
1628 cache_setunresolved(ap->a_nch);
1629 cache_setvp(ap->a_nch, *ap->a_vpp);
1636 * Make a device node (typically a fifo)
1640 hammer2_vop_nmknod(struct vop_nmknod_args *ap)
1642 hammer2_inode_t *dip;
1643 hammer2_inode_t *nip;
1644 hammer2_trans_t trans;
1645 hammer2_cluster_t *ncluster;
1646 struct namecache *ncp;
1647 const uint8_t *name;
1652 dip = VTOI(ap->a_dvp);
1653 if (dip->pmp->ronly) {
1658 ncp = ap->a_nch->ncp;
1659 name = ncp->nc_name;
1660 name_len = ncp->nc_nlen;
1661 hammer2_pfs_memory_wait(dip->pmp);
1662 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1665 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1666 name, name_len, &ncluster, &error);
1668 KKASSERT(nip == NULL);
1671 *ap->a_vpp = hammer2_igetv(nip, ncluster, &error);
1672 hammer2_inode_unlock_ex(nip, ncluster);
1674 hammer2_trans_done(&trans);
1677 cache_setunresolved(ap->a_nch);
1678 cache_setvp(ap->a_nch, *ap->a_vpp);
1685 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
1689 hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
1691 hammer2_inode_t *dip;
1692 hammer2_inode_t *nip;
1693 hammer2_cluster_t *ncparent;
1694 hammer2_trans_t trans;
1695 struct namecache *ncp;
1696 const uint8_t *name;
1700 dip = VTOI(ap->a_dvp);
1701 if (dip->pmp->ronly)
1704 ncp = ap->a_nch->ncp;
1705 name = ncp->nc_name;
1706 name_len = ncp->nc_nlen;
1707 hammer2_pfs_memory_wait(dip->pmp);
1708 hammer2_trans_init(&trans, dip->pmp, HAMMER2_TRANS_NEWINODE);
1711 ap->a_vap->va_type = VLNK; /* enforce type */
1713 nip = hammer2_inode_create(&trans, dip, ap->a_vap, ap->a_cred,
1714 name, name_len, &ncparent, &error);
1716 KKASSERT(nip == NULL);
1718 hammer2_trans_done(&trans);
1721 *ap->a_vpp = hammer2_igetv(nip, ncparent, &error);
1724 * Build the softlink (~like file data) and finalize the namecache.
1730 hammer2_inode_data_t *nipdata;
1732 nipdata = &hammer2_cluster_wdata(ncparent)->ipdata;
1733 /* nipdata = &nip->chain->data->ipdata;XXX */
1734 bytes = strlen(ap->a_target);
1736 if (bytes <= HAMMER2_EMBEDDED_BYTES) {
1737 KKASSERT(nipdata->op_flags &
1738 HAMMER2_OPFLAG_DIRECTDATA);
1739 bcopy(ap->a_target, nipdata->u.data, bytes);
1740 nipdata->size = bytes;
1742 hammer2_cluster_modsync(ncparent);
1743 hammer2_inode_unlock_ex(nip, ncparent);
1744 /* nipdata = NULL; not needed */
1746 hammer2_inode_unlock_ex(nip, ncparent);
1747 /* nipdata = NULL; not needed */
1748 bzero(&auio, sizeof(auio));
1749 bzero(&aiov, sizeof(aiov));
1750 auio.uio_iov = &aiov;
1751 auio.uio_segflg = UIO_SYSSPACE;
1752 auio.uio_rw = UIO_WRITE;
1753 auio.uio_resid = bytes;
1754 auio.uio_iovcnt = 1;
1755 auio.uio_td = curthread;
1756 aiov.iov_base = ap->a_target;
1757 aiov.iov_len = bytes;
1758 error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
1759 /* XXX handle error */
1763 hammer2_inode_unlock_ex(nip, ncparent);
1765 hammer2_trans_done(&trans);
1768 * Finalize namecache
1771 cache_setunresolved(ap->a_nch);
1772 cache_setvp(ap->a_nch, *ap->a_vpp);
1773 /* hammer2_knote(ap->a_dvp, NOTE_WRITE); */
1779 * hammer2_vop_nremove { nch, dvp, cred }
1783 hammer2_vop_nremove(struct vop_nremove_args *ap)
1785 hammer2_inode_t *dip;
1786 hammer2_trans_t trans;
1787 struct namecache *ncp;
1788 const uint8_t *name;
1793 dip = VTOI(ap->a_dvp);
1794 if (dip->pmp->ronly) {
1799 ncp = ap->a_nch->ncp;
1800 name = ncp->nc_name;
1801 name_len = ncp->nc_nlen;
1803 hammer2_pfs_memory_wait(dip->pmp);
1804 hammer2_trans_init(&trans, dip->pmp, 0);
1805 error = hammer2_unlink_file(&trans, dip, name, name_len,
1806 0, NULL, ap->a_nch, -1);
1807 hammer2_run_unlinkq(&trans, dip->pmp);
1808 hammer2_trans_done(&trans);
1810 cache_unlink(ap->a_nch);
1816 * hammer2_vop_nrmdir { nch, dvp, cred }
1820 hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
1822 hammer2_inode_t *dip;
1823 hammer2_trans_t trans;
1824 struct namecache *ncp;
1825 const uint8_t *name;
1830 dip = VTOI(ap->a_dvp);
1831 if (dip->pmp->ronly) {
1836 ncp = ap->a_nch->ncp;
1837 name = ncp->nc_name;
1838 name_len = ncp->nc_nlen;
1840 hammer2_pfs_memory_wait(dip->pmp);
1841 hammer2_trans_init(&trans, dip->pmp, 0);
1842 hammer2_run_unlinkq(&trans, dip->pmp);
1843 error = hammer2_unlink_file(&trans, dip, name, name_len,
1844 1, NULL, ap->a_nch, -1);
1845 hammer2_trans_done(&trans);
1847 cache_unlink(ap->a_nch);
1853 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
1857 hammer2_vop_nrename(struct vop_nrename_args *ap)
1859 struct namecache *fncp;
1860 struct namecache *tncp;
1861 hammer2_inode_t *cdip;
1862 hammer2_inode_t *fdip;
1863 hammer2_inode_t *tdip;
1864 hammer2_inode_t *ip;
1865 hammer2_cluster_t *cluster;
1866 hammer2_cluster_t *fdcluster;
1867 hammer2_cluster_t *tdcluster;
1868 hammer2_cluster_t *cdcluster;
1869 hammer2_trans_t trans;
1870 const uint8_t *fname;
1872 const uint8_t *tname;
1878 if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
1880 if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
1883 fdip = VTOI(ap->a_fdvp); /* source directory */
1884 tdip = VTOI(ap->a_tdvp); /* target directory */
1886 if (fdip->pmp->ronly)
1890 fncp = ap->a_fnch->ncp; /* entry name in source */
1891 fname = fncp->nc_name;
1892 fname_len = fncp->nc_nlen;
1894 tncp = ap->a_tnch->ncp; /* entry name in target */
1895 tname = tncp->nc_name;
1896 tname_len = tncp->nc_nlen;
1898 hammer2_pfs_memory_wait(tdip->pmp);
1899 hammer2_trans_init(&trans, tdip->pmp, 0);
1902 * ip is the inode being renamed. If this is a hardlink then
1903 * ip represents the actual file and not the hardlink marker.
1905 ip = VTOI(fncp->nc_vp);
1910 * The common parent directory must be locked first to avoid deadlocks.
1911 * Also note that fdip and/or tdip might match cdip.
1913 * WARNING! fdip may not match ip->pip. That is, if the source file
1914 * is already a hardlink then what we are renaming is the
1915 * hardlink pointer, not the hardlink itself. The hardlink
1916 * directory (ip->pip) will already be at a common parent
1919 * Be sure to use ip->pip when finding the common parent
1920 * against tdip or we might accidently move the hardlink
1921 * target into a subdirectory that makes it inaccessible to
1924 cdip = hammer2_inode_common_parent(ip->pip, tdip);
1925 cdcluster = hammer2_inode_lock_ex(cdip);
1926 fdcluster = hammer2_inode_lock_ex(fdip);
1927 tdcluster = hammer2_inode_lock_ex(tdip);
1930 * Keep a tight grip on the inode so the temporary unlinking from
1931 * the source location prior to linking to the target location
1932 * does not cause the cluster to be destroyed.
1934 * NOTE: To avoid deadlocks we cannot lock (ip) while we are
1935 * unlinking elements from their directories. Locking
1936 * the nlinks field does not lock the whole inode.
1938 hammer2_inode_ref(ip);
1941 * Remove target if it exists.
1943 error = hammer2_unlink_file(&trans, tdip, tname, tname_len,
1944 -1, NULL, ap->a_tnch, -1);
1946 if (error && error != ENOENT)
1950 * When renaming a hardlinked file we may have to re-consolidate
1951 * the location of the hardlink target.
1953 * If ip represents a regular file the consolidation code essentially
1954 * does nothing other than return the same locked cluster that was
1957 * The returned cluster will be locked.
1959 * WARNING! We do not currently have a local copy of ipdata but
1960 * we do use one later remember that it must be reloaded
1961 * on any modification to the inode, including connects.
1963 cluster = hammer2_inode_lock_ex(ip);
1964 error = hammer2_hardlink_consolidate(&trans, ip, &cluster,
1965 cdip, cdcluster, 0);
1970 * Disconnect (fdip, fname) from the source directory. This will
1971 * disconnect (ip) if it represents a direct file. If (ip) represents
1972 * a hardlink the HARDLINK pointer object will be removed but the
1973 * hardlink will stay intact.
1975 * Always pass nch as NULL because we intend to reconnect the inode,
1976 * so we don't want hammer2_unlink_file() to rename it to the hidden
1977 * open-but-unlinked directory.
1979 * The target cluster may be marked DELETED but will not be destroyed
1980 * since we retain our hold on ip and cluster.
1982 * NOTE: We pass nlinks as 0 (not -1) in order to retain the file's
1985 error = hammer2_unlink_file(&trans, fdip, fname, fname_len,
1986 -1, &hlink, NULL, 0);
1987 KKASSERT(error != EAGAIN);
1992 * Reconnect ip to target directory using cluster. Chains cannot
1993 * actually be moved, so this will duplicate the cluster in the new
1994 * spot and assign it to the ip, replacing the old cluster.
1996 * WARNING: Because recursive locks are allowed and we unlinked the
1997 * file that we have a cluster-in-hand for just above, the
1998 * cluster might have been delete-duplicated. We must
1999 * refactor the cluster.
2001 * WARNING: Chain locks can lock buffer cache buffers, to avoid
2002 * deadlocks we want to unlock before issuing a cache_*()
2003 * op (that might have to lock a vnode).
2005 * NOTE: Pass nlinks as 0 because we retained the link count from
2006 * the unlink, so we do not have to modify it.
2008 error = hammer2_inode_connect(&trans, &cluster, hlink,
2010 tname, tname_len, 0);
2012 KKASSERT(cluster != NULL);
2013 hammer2_inode_repoint(ip, (hlink ? ip->pip : tdip), cluster);
2016 hammer2_inode_unlock_ex(ip, cluster);
2017 hammer2_inode_unlock_ex(tdip, tdcluster);
2018 hammer2_inode_unlock_ex(fdip, fdcluster);
2019 hammer2_inode_unlock_ex(cdip, cdcluster);
2020 hammer2_inode_drop(ip);
2021 hammer2_inode_drop(cdip);
2022 hammer2_run_unlinkq(&trans, fdip->pmp);
2023 hammer2_trans_done(&trans);
2026 * Issue the namecache update after unlocking all the internal
2027 * hammer structures, otherwise we might deadlock.
2029 if (tnch_error == 0) {
2030 cache_unlink(ap->a_tnch);
2031 cache_setunresolved(ap->a_tnch);
2034 cache_rename(ap->a_fnch, ap->a_tnch);
2041 * Strategy code (async logical file buffer I/O from system)
2043 * WARNING: The strategy code cannot safely use hammer2 transactions
2044 * as this can deadlock against vfs_sync's vfsync() call
2045 * if multiple flushes are queued. All H2 structures must
2046 * already be present and ready for the DIO.
2048 * Reads can be initiated asynchronously, writes have to be
2049 * spooled to a separate thread for action to avoid deadlocks.
2051 static int hammer2_strategy_read(struct vop_strategy_args *ap);
2052 static int hammer2_strategy_write(struct vop_strategy_args *ap);
2053 static void hammer2_strategy_read_callback(hammer2_iocb_t *iocb);
2057 hammer2_vop_strategy(struct vop_strategy_args *ap)
2068 error = hammer2_strategy_read(ap);
2069 ++hammer2_iod_file_read;
2072 error = hammer2_strategy_write(ap);
2073 ++hammer2_iod_file_write;
2076 bp->b_error = error = EINVAL;
2077 bp->b_flags |= B_ERROR;
2085 * Logical buffer I/O, async read.
2089 hammer2_strategy_read(struct vop_strategy_args *ap)
2094 hammer2_inode_t *ip;
2095 hammer2_cluster_t *cparent;
2096 hammer2_cluster_t *cluster;
2097 hammer2_key_t key_dummy;
2098 hammer2_key_t lbase;
2104 ip = VTOI(ap->a_vp);
2105 nbio = push_bio(bio);
2107 lbase = bio->bio_offset;
2108 KKASSERT(((int)lbase & HAMMER2_PBUFMASK) == 0);
2111 * Lookup the file offset.
2113 cparent = hammer2_inode_lock_sh(ip);
2114 cluster = hammer2_cluster_lookup(cparent, &key_dummy,
2116 HAMMER2_LOOKUP_NODATA |
2117 HAMMER2_LOOKUP_SHARED,
2119 hammer2_inode_unlock_sh(ip, cparent);
2122 * Data is zero-fill if no cluster could be found
2123 * (XXX or EIO on a cluster failure).
2125 if (cluster == NULL) {
2128 bzero(bp->b_data, bp->b_bcount);
2134 * Cluster elements must be type INODE or type DATA, but the
2135 * compression mode (or not) for DATA chains can be different for
2136 * each chain. This will be handled by the callback.
2138 * If the cluster already has valid data the callback will be made
2139 * immediately/synchronously.
2141 btype = hammer2_cluster_type(cluster);
2142 if (btype != HAMMER2_BREF_TYPE_INODE &&
2143 btype != HAMMER2_BREF_TYPE_DATA) {
2144 panic("READ PATH: hammer2_strategy_read: unknown bref type");
2146 hammer2_cluster_load_async(cluster, hammer2_strategy_read_callback,
2152 * Read callback for hammer2_cluster_load_async(). The load function may
2153 * start several actual I/Os but will only make one callback, typically with
2154 * the first valid I/O XXX
2158 hammer2_strategy_read_callback(hammer2_iocb_t *iocb)
2160 struct bio *bio = iocb->ptr; /* original logical buffer */
2161 struct buf *bp = bio->bio_buf; /* original logical buffer */
2162 hammer2_chain_t *chain;
2163 hammer2_cluster_t *cluster;
2169 * Extract data and handle iteration on I/O failure. iocb->off
2170 * is the cluster index for iteration.
2172 cluster = iocb->cluster;
2173 dio = iocb->dio; /* can be NULL if iocb not in progress */
2176 * Work to do if INPROG set, else dio is already good or dio is
2177 * NULL (which is the shortcut case if chain->data is already good).
2179 if (iocb->flags & HAMMER2_IOCB_INPROG) {
2181 * Read attempt not yet made. Issue an asynchronous read
2182 * if necessary and return, operation will chain back to
2185 if ((iocb->flags & HAMMER2_IOCB_READ) == 0) {
2186 if (dio->bp == NULL ||
2187 (dio->bp->b_flags & B_CACHE) == 0) {
2192 iocb->flags |= HAMMER2_IOCB_READ;
2193 breadcb(dio->hmp->devvp,
2194 dio->pbase, dio->psize,
2195 hammer2_io_callback, iocb);
2202 * If we have a DIO it is now done, check for an error and
2203 * calculate the data.
2205 * If there is no DIO it is an optimization by
2206 * hammer2_cluster_load_async(), the data is available in
2210 if (dio->bp->b_flags & B_ERROR) {
2211 i = (int)iocb->lbase + 1;
2212 if (i >= cluster->nchains) {
2213 bp->b_flags |= B_ERROR;
2214 bp->b_error = dio->bp->b_error;
2215 hammer2_io_complete(iocb);
2217 hammer2_cluster_unlock(cluster);
2219 hammer2_io_complete(iocb); /* XXX */
2220 chain = cluster->array[i];
2221 kprintf("hammer2: IO CHAIN-%d %p\n", i, chain);
2222 hammer2_adjreadcounter(&chain->bref,
2224 iocb->chain = chain;
2225 iocb->lbase = (off_t)i;
2228 hammer2_io_getblk(chain->hmp,
2229 chain->bref.data_off,
2235 chain = iocb->chain;
2236 data = hammer2_io_data(dio, chain->bref.data_off);
2239 * Special synchronous case, data present in chain->data.
2241 chain = iocb->chain;
2242 data = (void *)chain->data;
2245 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
2247 * Data is embedded in the inode (copy from inode).
2249 bcopy(((hammer2_inode_data_t *)data)->u.data,
2250 bp->b_data, HAMMER2_EMBEDDED_BYTES);
2251 bzero(bp->b_data + HAMMER2_EMBEDDED_BYTES,
2252 bp->b_bcount - HAMMER2_EMBEDDED_BYTES);
2255 } else if (chain->bref.type == HAMMER2_BREF_TYPE_DATA) {
2257 * Data is on-media, issue device I/O and copy.
2259 * XXX direct-IO shortcut could go here XXX.
2261 switch (HAMMER2_DEC_COMP(chain->bref.methods)) {
2262 case HAMMER2_COMP_LZ4:
2263 hammer2_decompress_LZ4_callback(data, chain->bytes,
2266 case HAMMER2_COMP_ZLIB:
2267 hammer2_decompress_ZLIB_callback(data, chain->bytes,
2270 case HAMMER2_COMP_NONE:
2271 KKASSERT(chain->bytes <= bp->b_bcount);
2272 bcopy(data, bp->b_data, chain->bytes);
2273 if (chain->bytes < bp->b_bcount) {
2274 bzero(bp->b_data + chain->bytes,
2275 bp->b_bcount - chain->bytes);
2277 bp->b_flags |= B_NOTMETA;
2282 panic("hammer2_strategy_read: "
2283 "unknown compression type");
2286 /* bqrelse the dio to help stabilize the call to panic() */
2288 hammer2_io_bqrelse(&dio);
2289 panic("hammer2_strategy_read: unknown bref type");
2293 * Once the iocb is cleaned up the DIO (if any) will no longer be
2294 * in-progress but will still have a ref. Be sure to release
2297 hammer2_io_complete(iocb); /* physical management */
2298 if (dio) /* physical dio & buffer */
2299 hammer2_io_bqrelse(&dio);
2300 hammer2_cluster_unlock(cluster); /* cluster management */
2301 biodone(bio); /* logical buffer */
2306 hammer2_strategy_write(struct vop_strategy_args *ap)
2308 hammer2_pfsmount_t *pmp;
2311 hammer2_inode_t *ip;
2315 ip = VTOI(ap->a_vp);
2318 hammer2_lwinprog_ref(pmp);
2319 mtx_lock(&pmp->wthread_mtx);
2320 if (TAILQ_EMPTY(&pmp->wthread_bioq.queue)) {
2321 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2322 mtx_unlock(&pmp->wthread_mtx);
2323 wakeup(&pmp->wthread_bioq);
2325 bioq_insert_tail(&pmp->wthread_bioq, ap->a_bio);
2326 mtx_unlock(&pmp->wthread_mtx);
2328 hammer2_lwinprog_wait(pmp);
2334 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
2338 hammer2_vop_ioctl(struct vop_ioctl_args *ap)
2340 hammer2_inode_t *ip;
2344 ip = VTOI(ap->a_vp);
2346 error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
2347 ap->a_fflag, ap->a_cred);
2354 hammer2_vop_mountctl(struct vop_mountctl_args *ap)
2357 hammer2_pfsmount_t *pmp;
2362 case (MOUNTCTL_SET_EXPORT):
2363 mp = ap->a_head.a_ops->head.vv_mount;
2366 if (ap->a_ctllen != sizeof(struct export_args))
2369 rc = vfs_export(mp, &pmp->export,
2370 (const struct export_args *)ap->a_ctl);
2373 rc = vop_stdmountctl(ap);
2381 * This handles unlinked open files after the vnode is finally dereferenced.
2382 * To avoid deadlocks it cannot be called from the normal vnode recycling
2383 * path, so we call it (1) after a unlink, rmdir, or rename, (2) on every
2384 * flush, and (3) on umount.
2387 hammer2_run_unlinkq(hammer2_trans_t *trans, hammer2_pfsmount_t *pmp)
2389 const hammer2_inode_data_t *ripdata;
2390 hammer2_inode_unlink_t *ipul;
2391 hammer2_inode_t *ip;
2392 hammer2_cluster_t *cluster;
2393 hammer2_cluster_t *cparent;
2395 if (TAILQ_EMPTY(&pmp->unlinkq))
2399 spin_lock(&pmp->list_spin);
2400 while ((ipul = TAILQ_FIRST(&pmp->unlinkq)) != NULL) {
2401 TAILQ_REMOVE(&pmp->unlinkq, ipul, entry);
2402 spin_unlock(&pmp->list_spin);
2404 kfree(ipul, pmp->minode);
2406 cluster = hammer2_inode_lock_ex(ip);
2407 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
2408 if (hammer2_debug & 0x400) {
2409 kprintf("hammer2: unlink on reclaim: %s refs=%d\n",
2410 ripdata->filename, ip->refs);
2412 KKASSERT(ripdata->nlinks == 0);
2414 cparent = hammer2_cluster_parent(cluster);
2415 hammer2_cluster_delete(trans, cparent, cluster,
2416 HAMMER2_DELETE_PERMANENT);
2417 hammer2_cluster_unlock(cparent);
2418 hammer2_inode_unlock_ex(ip, cluster); /* inode lock */
2419 hammer2_inode_drop(ip); /* ipul ref */
2421 spin_lock(&pmp->list_spin);
2423 spin_unlock(&pmp->list_spin);
2431 static void filt_hammer2detach(struct knote *kn);
2432 static int filt_hammer2read(struct knote *kn, long hint);
2433 static int filt_hammer2write(struct knote *kn, long hint);
2434 static int filt_hammer2vnode(struct knote *kn, long hint);
2436 static struct filterops hammer2read_filtops =
2437 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2438 NULL, filt_hammer2detach, filt_hammer2read };
2439 static struct filterops hammer2write_filtops =
2440 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2441 NULL, filt_hammer2detach, filt_hammer2write };
2442 static struct filterops hammer2vnode_filtops =
2443 { FILTEROP_ISFD | FILTEROP_MPSAFE,
2444 NULL, filt_hammer2detach, filt_hammer2vnode };
2448 hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
2450 struct vnode *vp = ap->a_vp;
2451 struct knote *kn = ap->a_kn;
2453 switch (kn->kn_filter) {
2455 kn->kn_fop = &hammer2read_filtops;
2458 kn->kn_fop = &hammer2write_filtops;
2461 kn->kn_fop = &hammer2vnode_filtops;
2464 return (EOPNOTSUPP);
2467 kn->kn_hook = (caddr_t)vp;
2469 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2475 filt_hammer2detach(struct knote *kn)
2477 struct vnode *vp = (void *)kn->kn_hook;
2479 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
2483 filt_hammer2read(struct knote *kn, long hint)
2485 struct vnode *vp = (void *)kn->kn_hook;
2486 hammer2_inode_t *ip = VTOI(vp);
2489 if (hint == NOTE_REVOKE) {
2490 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2493 off = ip->size - kn->kn_fp->f_offset;
2494 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
2495 if (kn->kn_sfflags & NOTE_OLDAPI)
2497 return (kn->kn_data != 0);
2502 filt_hammer2write(struct knote *kn, long hint)
2504 if (hint == NOTE_REVOKE)
2505 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
2511 filt_hammer2vnode(struct knote *kn, long hint)
2513 if (kn->kn_sfflags & hint)
2514 kn->kn_fflags |= hint;
2515 if (hint == NOTE_REVOKE) {
2516 kn->kn_flags |= (EV_EOF | EV_NODATA);
2519 return (kn->kn_fflags != 0);
2527 hammer2_vop_markatime(struct vop_markatime_args *ap)
2529 hammer2_inode_t *ip;
2542 hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
2546 error = VOCALL(&fifo_vnode_vops, &ap->a_head);
2548 error = hammer2_vop_kqfilter(ap);
2555 struct vop_ops hammer2_vnode_vops = {
2556 .vop_default = vop_defaultop,
2557 .vop_fsync = hammer2_vop_fsync,
2558 .vop_getpages = vop_stdgetpages,
2559 .vop_putpages = vop_stdputpages,
2560 .vop_access = hammer2_vop_access,
2561 .vop_advlock = hammer2_vop_advlock,
2562 .vop_close = hammer2_vop_close,
2563 .vop_nlink = hammer2_vop_nlink,
2564 .vop_ncreate = hammer2_vop_ncreate,
2565 .vop_nsymlink = hammer2_vop_nsymlink,
2566 .vop_nremove = hammer2_vop_nremove,
2567 .vop_nrmdir = hammer2_vop_nrmdir,
2568 .vop_nrename = hammer2_vop_nrename,
2569 .vop_getattr = hammer2_vop_getattr,
2570 .vop_setattr = hammer2_vop_setattr,
2571 .vop_readdir = hammer2_vop_readdir,
2572 .vop_readlink = hammer2_vop_readlink,
2573 .vop_getpages = vop_stdgetpages,
2574 .vop_putpages = vop_stdputpages,
2575 .vop_read = hammer2_vop_read,
2576 .vop_write = hammer2_vop_write,
2577 .vop_open = hammer2_vop_open,
2578 .vop_inactive = hammer2_vop_inactive,
2579 .vop_reclaim = hammer2_vop_reclaim,
2580 .vop_nresolve = hammer2_vop_nresolve,
2581 .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
2582 .vop_nmkdir = hammer2_vop_nmkdir,
2583 .vop_nmknod = hammer2_vop_nmknod,
2584 .vop_ioctl = hammer2_vop_ioctl,
2585 .vop_mountctl = hammer2_vop_mountctl,
2586 .vop_bmap = hammer2_vop_bmap,
2587 .vop_strategy = hammer2_vop_strategy,
2588 .vop_kqfilter = hammer2_vop_kqfilter
2591 struct vop_ops hammer2_spec_vops = {
2592 .vop_default = vop_defaultop,
2593 .vop_fsync = hammer2_vop_fsync,
2594 .vop_read = vop_stdnoread,
2595 .vop_write = vop_stdnowrite,
2596 .vop_access = hammer2_vop_access,
2597 .vop_close = hammer2_vop_close,
2598 .vop_markatime = hammer2_vop_markatime,
2599 .vop_getattr = hammer2_vop_getattr,
2600 .vop_inactive = hammer2_vop_inactive,
2601 .vop_reclaim = hammer2_vop_reclaim,
2602 .vop_setattr = hammer2_vop_setattr
2605 struct vop_ops hammer2_fifo_vops = {
2606 .vop_default = fifo_vnoperate,
2607 .vop_fsync = hammer2_vop_fsync,
2609 .vop_read = hammer2_vop_fiforead,
2610 .vop_write = hammer2_vop_fifowrite,
2612 .vop_access = hammer2_vop_access,
2614 .vop_close = hammer2_vop_fifoclose,
2616 .vop_markatime = hammer2_vop_markatime,
2617 .vop_getattr = hammer2_vop_getattr,
2618 .vop_inactive = hammer2_vop_inactive,
2619 .vop_reclaim = hammer2_vop_reclaim,
2620 .vop_setattr = hammer2_vop_setattr,
2621 .vop_kqfilter = hammer2_vop_fifokqfilter