| blob | 898be4d7b231198082c53405503f13f6d387d03b |
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
22 /*
23 * Copyright (c) 1984, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2015, Joyent, Inc.
25 * Copyright (c) 2016 by Delphix. All rights reserved.
26 */
28 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
29 /* All Rights Reserved */
31 /*
32 * Portions of this source code were derived from Berkeley 4.3 BSD
33 * under license from the Regents of the University of California.
34 */
36 #include <sys/types.h>
37 #include <sys/t_lock.h>
38 #include <sys/ksynch.h>
39 #include <sys/param.h>
40 #include <sys/time.h>
41 #include <sys/systm.h>
42 #include <sys/sysmacros.h>
43 #include <sys/resource.h>
44 #include <sys/signal.h>
45 #include <sys/cred.h>
46 #include <sys/user.h>
47 #include <sys/buf.h>
48 #include <sys/vfs.h>
49 #include <sys/vfs_opreg.h>
50 #include <sys/vnode.h>
51 #include <sys/proc.h>
52 #include <sys/disp.h>
53 #include <sys/file.h>
54 #include <sys/fcntl.h>
55 #include <sys/flock.h>
56 #include <sys/atomic.h>
57 #include <sys/kmem.h>
58 #include <sys/uio.h>
59 #include <sys/dnlc.h>
60 #include <sys/conf.h>
61 #include <sys/mman.h>
62 #include <sys/pathname.h>
63 #include <sys/debug.h>
64 #include <sys/vmsystm.h>
65 #include <sys/cmn_err.h>
66 #include <sys/filio.h>
67 #include <sys/policy.h>
69 #include <sys/fs/ufs_fs.h>
70 #include <sys/fs/ufs_lockfs.h>
71 #include <sys/fs/ufs_filio.h>
72 #include <sys/fs/ufs_inode.h>
73 #include <sys/fs/ufs_fsdir.h>
74 #include <sys/fs/ufs_quota.h>
75 #include <sys/fs/ufs_log.h>
76 #include <sys/fs/ufs_snap.h>
77 #include <sys/fs/ufs_trans.h>
78 #include <sys/fs/ufs_panic.h>
79 #include <sys/fs/ufs_bio.h>
81 #include <sys/errno.h>
82 #include <sys/fssnap_if.h>
83 #include <sys/unistd.h>
84 #include <sys/sunddi.h>
88 #include <vm/hat.h>
89 #include <vm/page.h>
90 #include <vm/pvn.h>
91 #include <vm/as.h>
92 #include <vm/seg.h>
93 #include <vm/seg_map.h>
94 #include <vm/seg_vn.h>
95 #include <vm/seg_kmem.h>
96 #include <vm/rm.h>
97 #include <sys/swap.h>
99 #include <fs/fs_subr.h>
101 #include <sys/fs/decomp.h>
110 caller_context_t *);
118 caller_context_t *);
120 caller_context_t *);
122 caller_context_t *);
144 caller_context_t *);
153 caller_context_t *);
160 caller_context_t *);
169 caller_context_t *);
171 caller_context_t *);
173 caller_context_t *);
180 caller_context_t *);
182 caller_context_t *);
188 /*
189 * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
190 *
191 * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
192 */
195 /* NOTE: "not blkd" below means that the operation isn't blocked by lockfs */
236 NULL, NULL
237 };
239 #define MAX_BACKFILE_COUNT 9999
241 /*
242 * Created by ufs_dumpctl() to store a file's disk block info into memory.
243 * Used by ufs_dump() to dump data to disk directly.
244 */
250 };
254 /*
255 * Previously there was no special action required for ordinary files.
256 * (Devices are handled through the device file system.)
257 * Now we support Large Files and Large File API requires open to
258 * fail if file is large.
259 * We could take care to prevent data corruption
260 * by doing an atomic check of size and truncate if file is opened with
261 * FTRUNC flag set but traditionally this is being done by the vfs/vnode
262 * layers. So taking care of truncation here is a change in the existing
263 * semantics of VOP_OPEN and therefore we chose not to implement any thing
264 * here. The check for the size of the file > 2GB is being done at the
265 * vfs layer in routine vn_open().
266 */
268 /* ARGSUSED */
269 static int
271 {
273 }
275 /*ARGSUSED*/
276 static int
279 {
283 /*
284 * Push partially filled cluster at last close.
285 * ``last close'' is approximated because the dnlc
286 * may have a hold on the vnode.
287 * Checking for VBAD here will also act as a forced umount check.
288 */
296 }
297 }
300 }
302 /*ARGSUSED*/
303 static int
306 {
315 /*
316 * Mandatory locking needs to be done before ufs_lockfs_begin()
317 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
318 */
320 /*
321 * ufs_getattr ends up being called by chklock
322 */
327 }
334 /*
335 * In the case that a directory is opened for reading as a file
336 * (eg "cat .") with the O_RSYNC, O_SYNC and O_DSYNC flags set.
337 * The locking order had to be changed to avoid a deadlock with
338 * an update taking place on that directory at the same time.
339 */
350 }
356 error);
359 TOP_READ_SIZE);
361 }
363 /*
364 * Only transact reads to files opened for sync-read and
365 * sync-write on a file system that is not write locked.
366 *
367 * The ``not write locked'' check prevents problems with
368 * enabling/disabling logging on a busy file system. E.g.,
369 * logging exists at the beginning of the read but does not
370 * at the end.
371 *
372 */
376 error);
379 }
387 TOP_READ_SIZE);
388 }
389 }
393 }
394 out:
397 }
405 static int
407 {
410 /*
411 * If the FDSYNC flag is set then ignore the global
412 * ufs_allow_shared_writes in this case.
413 */
416 /*
417 * Filter to determine if this request is suitable as a
418 * concurrent rewrite. This write must not allocate blocks
419 * by extending the file or filling in holes. No use trying
420 * through FSYNC descriptors as the inode will be synchronously
421 * updated after the write. The uio structure has not yet been
422 * checked for sanity, so assume nothing.
423 */
429 shared_write);
430 }
432 /*ARGSUSED*/
433 static int
436 {
449 retry_mandlock:
450 /*
451 * Mandatory locking needs to be done before ufs_lockfs_begin()
452 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
453 * Check for forced unmounts normally done in ufs_lockfs_begin().
454 */
458 }
463 /*
464 * ufs_getattr ends up being called by chklock
465 */
470 }
472 /* i_rwlock can change in chklock */
476 /*
477 * Check for fast-path special case of directio re-writes.
478 */
490 uint_t i_flag_save;
494 /*
495 * Special treatment of access times for re-writes.
496 * If IMOD is not already set, then convert it
497 * to IMODACC for this operation. This defers
498 * entering a delta into the log until the inode
499 * is flushed. This mimics what is done for read
500 * operations and inode access time.
501 */
510 }
516 }
520 }
525 /*
526 * Mandatory locking could have been enabled
527 * after dropping the i_rwlock.
528 */
531 }
537 /*
538 * Amount of log space needed for this write
539 */
543 /*
544 * Throttle writes.
545 */
549 ufs_throttles++;
551 }
553 }
555 /*
556 * Enter Transaction
557 *
558 * If the write is a rewrite there is no need to open a transaction
559 * if the FDSYNC flag is set and not the FSYNC. In this case just
560 * set the IMODACC flag to modify do the update at a later time
561 * thus avoiding the overhead of the logging transaction that is
562 * not required.
563 */
567 uint_t i_flag_save;
578 }
584 terr);
586 }
587 }
591 }
593 /*
594 * Write the file
595 */
599 /*
600 * In append mode start at end of file.
601 */
603 }
605 /*
606 * Mild optimisation, don't call ufs_trans_write() unless we have to
607 * Also, suppress file system full messages if we will retry.
608 */
615 }
621 /*
622 * Leave Transaction
623 */
630 resv);
633 }
636 }
638 }
639 out:
641 /*
642 * Any blocks tied up in pending deletes?
643 */
647 }
653 }
655 /*
656 * Don't cache write blocks to files with the sticky bit set.
657 * Used to keep swap files from blowing the page cache on a server.
658 */
661 /*
662 * Free behind hacks. The pager is busted.
663 * XXX - need to pass the information down to writedone() in a flag like B_SEQ
664 * or B_FREE_IF_TIGHT_ON_MEMORY.
665 */
670 /*
671 * While we should, in most cases, cache the pages for write, we
672 * may also want to cache the pages for read as long as they are
673 * frequently re-usable.
674 *
675 * If cache_read_ahead = 1, the pages for read will go to the tail
676 * of the cache list when they are released, otherwise go to the head.
677 */
680 /*
681 * Freebehind exists so that as we read large files sequentially we
682 * don't consume most of memory with pages from a few files. It takes
683 * longer to re-read from disk multiple small files as it does reading
684 * one large one sequentially. As system memory grows customers need
685 * to retain bigger chunks of files in memory. The advent of the
686 * cachelist opens up of the possibility freeing pages to the head or
687 * tail of the list.
688 *
689 * Not freeing a page is a bet that the page will be read again before
690 * it's segmap slot is needed for something else. If we loose the bet,
691 * it means some other thread is burdened with the page free we did
692 * not do. If we win we save a free and reclaim.
693 *
694 * Freeing it at the tail vs the head of cachelist is a bet that the
695 * page will survive until the next read. It's also saying that this
696 * page is more likely to be re-used than a page freed some time ago
697 * and never reclaimed.
698 *
699 * Freebehind maintains a range of file offset [smallfile1; smallfile2]
700 *
701 * 0 < offset < smallfile1 : pages are not freed.
702 * smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
703 * smallfile2 < offset : pages freed to head of cachelist.
704 *
705 * The range is computed at most once per second and depends on
706 * freemem and ncpus_online. Both parameters are bounded to be
707 * >= smallfile && >= smallfile64.
708 *
709 * smallfile1 = (free memory / ncpu) / 1000
710 * smallfile2 = (free memory / ncpu) / 10
711 *
712 * A few examples values:
713 *
714 * Free Mem (in Bytes) [smallfile1; smallfile2] [smallfile1; smallfile2]
715 * ncpus_online = 4 ncpus_online = 64
716 * ------------------ ----------------------- -----------------------
717 * 1G [256K; 25M] [32K; 1.5M]
718 * 10G [2.5M; 250M] [156K; 15M]
719 * 100G [25M; 2.5G] [1.5M; 150M]
720 *
721 */
723 #define SMALLFILE1_D 1000
724 #define SMALLFILE2_D 10
731 /*
732 * wrip does the real work of write requests for ufs.
733 */
734 int
736 {
738 u_offset_t off;
739 u_offset_t old_i_size;
743 caddr_t base;
746 uint_t flags;
755 /*
756 * ip->i_size is incremented before the uiomove
757 * is done on a write. If the move fails (bad user
758 * address) reset ip->i_size.
759 * The better way would be to increment ip->i_size
760 * only if the uiomove succeeds.
761 */
763 o_mode_t type;
768 /*
769 * check for forced unmount - should not happen as
770 * the request passed the lockfs checks.
771 */
779 /* check for valid filetype */
784 }
786 /*
787 * the actual limit of UFS file size
788 * is UFS_MAXOFFSET_T
789 */
801 }
803 /*
804 * if largefiles are disallowed, the limit is
805 * the pre-largefiles value of 2GB
806 */
809 else
814 }
817 }
827 }
828 /*
829 * Try to go direct
830 */
835 /*
836 * If ufs_directio wrote to the file or set the flags,
837 * we need to update i_seq, but it may be deferred.
838 */
843 }
846 }
848 /*
849 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
850 *
851 * o shadow inodes: vfs_dqrwlock is not held at all
852 * o quota updates: vfs_dqrwlock is read or write held
853 * o other updates: vfs_dqrwlock is read held
854 *
855 * The first case is the only one where we do not hold
856 * vfs_dqrwlock at all while entering wrip().
857 * We must make sure not to downgrade/drop vfs_dqrwlock if we
858 * have it as writer, i.e. if we are updating the quota inode.
859 * There is no potential deadlock scenario in this case as
860 * ufs_getpage() takes care of this and avoids reacquiring
861 * vfs_dqrwlock in that case.
862 *
863 * This check is done here since the above conditions do not change
864 * and we possibly loop below, so save a few cycles.
865 */
871 }
873 /*
874 * Large Files: We cast MAXBMASK to offset_t
875 * inorder to mask out the higher bits. Since offset_t
876 * is a signed value, the high order bit set in MAXBMASK
877 * value makes it do the right thing by having all bits 1
878 * in the higher word. May be removed for _SOLARIS64_.
879 */
894 }
895 /*
896 * since uoff + n >= limit,
897 * therefore n >= limit - uoff, and n is an int
898 * so it is safe to cast it to an int
899 */
901 }
903 /*
904 * We are extending the length of the file.
905 * bmap is used so that we are sure that
906 * if we need to allocate new blocks, that it
907 * is done here before we up the file size.
908 */
911 /*
912 * bmap_write never drops i_contents so if
913 * the flags are set it changed the file.
914 */
918 }
921 /*
922 * There is a window of vulnerability here.
923 * The sequence of operations: allocate file
924 * system blocks, uiomove the data into pages,
925 * and then update the size of the file in the
926 * inode, must happen atomically. However, due
927 * to current locking constraints, this can not
928 * be done.
929 */
933 /*
934 * If we are writing from the beginning of
935 * the mapping, we can just create the
936 * pages without having to read them.
937 */
940 /*
941 * Going to do a whole mappings worth,
942 * so we can just create the pages w/o
943 * having to read them in. But before
944 * we do that, we need to make sure any
945 * needed blocks are allocated first.
946 */
950 /*
951 * bmap_write never drops i_contents so if
952 * the flags are set it changed the file.
953 */
957 }
961 /*
962 * check if the new created page needed the
963 * allocation of new disk blocks.
964 */
969 /*
970 * In sync mode flush the indirect blocks which
971 * may have been allocated and not written on
972 * disk. In above cases bmap_write will allocate
973 * in sync mode.
974 */
979 }
980 }
982 /*
983 * At this point we can enter ufs_getpage() in one
984 * of two ways:
985 * 1) segmap_getmapflt() calls ufs_getpage() when the
986 * forcefault parameter is true (pagecreate == 0)
987 * 2) uiomove() causes a page fault.
988 *
989 * We have to drop the contents lock to prevent the VM
990 * system from trying to reacquire it in ufs_getpage()
991 * should the uiomove cause a pagefault.
992 *
993 * We have to drop the reader vfs_dqrwlock here as well.
994 */
1000 }
1005 /*
1006 * Touch the page and fault it in if it is not in core
1007 * before segmap_getmapflt or vpm_data_copy can lock it.
1008 * This is to avoid the deadlock if the buffer is mapped
1009 * to the same file through mmap which we want to write.
1010 */
1014 /*
1015 * Copy data. If new pages are created, part of
1016 * the page that is not written will be initizliazed
1017 * with zeros.
1018 */
1026 /*
1027 * segmap_pagecreate() returns 1 if it calls
1028 * page_create_va() to allocate any pages.
1029 */
1036 }
1038 /*
1039 * If "newpage" is set, then a new page was created and it
1040 * does not contain valid data, so it needs to be initialized
1041 * at this point.
1042 * Otherwise the page contains old data, which was overwritten
1043 * partially or as a whole in uiomove.
1044 * If there is only one iovec structure within uio, then
1045 * on error uiomove will not be able to update uio->uio_loffset
1046 * and we would zero the whole page here!
1047 *
1048 * If uiomove fails because of an error, the old valid data
1049 * is kept instead of filling the rest of the page with zero's.
1050 */
1053 /*
1054 * We created pages w/o initializing them completely,
1055 * thus we need to zero the part that wasn't set up.
1056 * This happens on most EOF write cases and if
1057 * we had some sort of error during the uiomove.
1058 */
1066 }
1068 /*
1069 * Unlock the pages allocated by page_create_va()
1070 * in segmap_pagecreate()
1071 */
1075 /*
1076 * If the size of the file changed, then update the
1077 * size field in the inode now. This can't be done
1078 * before the call to segmap_pageunlock or there is
1079 * a potential deadlock with callers to ufs_putpage().
1080 * They will be holding i_contents and trying to lock
1081 * a page, while this thread is holding a page locked
1082 * and trying to acquire i_contents.
1083 */
1089 /*
1090 * file has grown larger than 2GB. Set flag
1091 * in superblock to indicate this, if it
1092 * is not already set.
1093 */
1101 }
1107 }
1110 /*
1111 * If we failed on a write, we may have already
1112 * allocated file blocks as well as pages. It's
1113 * hard to undo the block allocation, but we must
1114 * be sure to invalidate any pages that may have
1115 * been allocated.
1116 *
1117 * If the page was created without initialization
1118 * then we must check if it should be possible
1119 * to destroy the new page and to keep the old data
1120 * on the disk.
1121 *
1122 * It is possible to destroy the page without
1123 * having to write back its contents only when
1124 * - the size of the file keeps unchanged
1125 * - bmap_write() did not allocate new disk blocks
1126 * it is possible to create big files using "seek" and
1127 * write to the end of the file. A "write" to a
1128 * position before the end of the file would not
1129 * change the size of the file but it would allocate
1130 * new disk blocks.
1131 * - uiomove intended to overwrite the whole page.
1132 * - a new page was created (newpage == 1).
1133 */
1136 newpage) {
1138 /* unwind what uiomove eventually last did */
1141 /*
1142 * destroy the page, do not write ambiguous
1143 * data to the disk.
1144 */
1147 /*
1148 * write the page back to the disk, if dirty,
1149 * and remove the page from the cache.
1150 */
1152 }
1155 /*
1156 * Flush pages.
1157 */
1161 }
1164 /*
1165 * Force write back for synchronous write cases.
1166 */
1168 /*
1169 * If the sticky bit is set but the
1170 * execute bit is not set, we do a
1171 * synchronous write back and free
1172 * the page when done. We set up swap
1173 * files to be handled this way to
1174 * prevent servers from keeping around
1175 * the client's swap pages too long.
1176 * XXX - there ought to be a better way.
1177 */
1180 SM_DONTNEED;
1184 }
1186 /*
1187 * Have written a whole block.
1188 * Start an asynchronous write and
1189 * mark the buffer to indicate that
1190 * it won't be needed again soon.
1191 */
1193 }
1195 /*
1196 * Flush pages.
1197 */
1201 }
1202 /*
1203 * If the operation failed and is synchronous,
1204 * then we need to unwind what uiomove() last
1205 * did so we can potentially return an error to
1206 * the caller. If this write operation was
1207 * done in two pieces and the first succeeded,
1208 * then we won't return an error for the second
1209 * piece that failed. However, we only want to
1210 * return a resid value that reflects what was
1211 * really done.
1212 *
1213 * Failures for non-synchronous operations can
1214 * be ignored since the page subsystem will
1215 * retry the operation until it succeeds or the
1216 * file system is unmounted.
1217 */
1224 }
1225 }
1226 }
1228 /*
1229 * Re-acquire contents lock.
1230 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1231 */
1236 /*
1237 * If the uiomove() failed or if a synchronous
1238 * page push failed, fix up i_size.
1239 */
1242 /*
1243 * The uiomove failed, and we
1244 * allocated blocks,so get rid
1245 * of them.
1246 */
1248 }
1250 /*
1251 * XXX - Can this be out of the loop?
1252 */
1254 /*
1255 * Only do one increase of i_seq for multiple
1256 * pieces. Because we drop locks, record
1257 * the fact that we changed the timestamp and
1258 * are deferring the increase in case another thread
1259 * pushes our timestamp update.
1260 */
1270 /*
1271 * Clear Set-UID & Set-GID bits on
1272 * successful write if not privileged
1273 * and at least one of the execute bits
1274 * is set. If we always clear Set-GID,
1275 * mandatory file and record locking is
1276 * unuseable.
1277 */
1279 }
1280 }
1281 /*
1282 * In the case the FDSYNC flag is set and this is a
1283 * "rewrite" we won't log a delta.
1284 * The FSYNC flag overrides all cases.
1285 */
1288 }
1291 out:
1292 /*
1293 * Make sure i_seq is increased at least once per write
1294 */
1298 }
1300 /*
1301 * Inode is updated according to this table -
1302 *
1303 * FSYNC FDSYNC(posix.4)
1304 * --------------------------
1305 * always@ IATTCHG|IBDWRITE
1306 *
1307 * @ - If we are doing synchronous write the only time we should
1308 * not be sync'ing the ip here is if we have the stickyhack
1309 * activated, the file is marked with the sticky bit and
1310 * no exec bit, the file length has not been changed and
1311 * no new blocks have been allocated during this write.
1312 */
1315 /*
1316 * we have eliminated nosync
1317 */
1321 }
1322 }
1324 /*
1325 * If we've already done a partial-write, terminate
1326 * the write but return no error unless the error is ENOSPC
1327 * because the caller can detect this and free resources and
1328 * try again.
1329 */
1336 }
1338 /*
1339 * rdip does the real work of read requests for ufs.
1340 */
1341 int
1343 {
1344 u_offset_t off;
1345 caddr_t base;
1353 uint_t flags;
1355 krw_t rwtype;
1356 o_mode_t type;
1370 /* check for valid filetype */
1375 }
1380 }
1383 }
1386 }
1393 }
1394 /*
1395 * Try to go direct
1396 */
1401 }
1406 offset_t diff;
1419 }
1423 /*
1424 * We update smallfile2 and smallfile1 at most every second.
1425 */
1438 }
1443 /*
1444 * At this point we can enter ufs_getpage() in one of two
1445 * ways:
1446 * 1) segmap_getmapflt() calls ufs_getpage() when the
1447 * forcefault parameter is true (value of 1 is passed)
1448 * 2) uiomove() causes a page fault.
1449 *
1450 * We cannot hold onto an i_contents reader lock without
1451 * risking deadlock in ufs_getpage() so drop a reader lock.
1452 * The ufs_getpage() dolock logic already allows for a
1453 * thread holding i_contents as writer to work properly
1454 * so we keep a writer lock.
1455 */
1460 /*
1461 * Copy data.
1462 */
1469 }
1473 /*
1474 * If reading sequential we won't need this
1475 * buffer again soon. For offsets in range
1476 * [smallfile1, smallfile2] release the pages
1477 * at the tail of the cache list, larger
1478 * offsets are released at the head.
1479 */
1485 }
1486 /*
1487 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1488 * we want to make sure that the page which has
1489 * been read, is written on disk if it is dirty.
1490 * And corresponding indirect blocks should also
1491 * be flushed out.
1492 */
1496 }
1501 }
1507 }
1508 }
1513 out:
1514 /*
1515 * Inode is updated according to this table if FRSYNC is set.
1516 *
1517 * FSYNC FDSYNC(posix.4)
1518 * --------------------------
1519 * always IATTCHG|IBDWRITE
1520 */
1521 /*
1522 * The inode is not updated if we're logging and the inode is a
1523 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1524 */
1528 }
1534 }
1535 }
1536 }
1537 /*
1538 * If we've already done a partial read, terminate
1539 * the read but return no error.
1540 */
1546 }
1548 /* ARGSUSED */
1549 static int
1558 {
1564 offset_t off;
1571 /*
1572 * forcibly unmounted
1573 */
1587 }
1595 }
1597 }
1601 /*
1602 * file system locking
1603 */
1611 }
1612 #ifdef _SYSCALL32_IMPL
1615 /* Translate ILP32 lockfs to LP64 lockfs */
1625 }
1631 comment =
1637 }
1640 }
1645 DATAMODEL_NATIVE) {
1648 }
1649 #ifdef _SYSCALL32_IMPL
1652 /* Translate LP64 to ILP32 lockfs */
1666 }
1672 }
1676 /*
1677 * get file system locking status
1678 */
1684 }
1685 #ifdef _SYSCALL32_IMPL
1688 /* Translate ILP32 lockfs to LP64 lockfs */
1698 }
1713 }
1714 #ifdef _SYSCALL32_IMPL
1716 /* Translate LP64 to ILP32 lockfs */
1727 }
1738 /*
1739 * set access time
1740 */
1742 /*
1743 * if mounted w/o atime, return quietly.
1744 * I briefly thought about returning ENOSYS, but
1745 * figured that most apps would consider this fatal
1746 * but the idea is to make this as seamless as poss.
1747 */
1752 ULOCKFS_SETATTR_MASK);
1760 }
1769 }
1773 /*
1774 * set delayed-io
1775 */
1779 /*
1780 * get delayed-io
1781 */
1785 /*
1786 * inode open
1787 */
1789 ULOCKFS_VGET_MASK);
1797 }
1801 /*
1802 * file system flush (push w/invalidate)
1803 */
1809 /*
1810 * Contract-private interface for Legato
1811 * Purge this vnode from the DNLC and decide
1812 * if this vnode is busy (*arg == 1) or not
1813 * (*arg == 0)
1814 */
1824 /*
1825 * Tune the file system (aka setting fs attributes)
1826 */
1828 ULOCKFS_SETATTR_MASK);
1852 {
1867 }
1869 /*
1870 * Do copyout even if there is an error because
1871 * the details of error is stored in fcp.
1872 */
1877 }
1880 {
1889 }
1901 /*
1902 * The following 3 ioctls are for TSufs support
1903 * although could potentially be used elsewhere
1904 */
1918 {
1924 /* Copy stucture if statistics are being kept */
1927 }
1933 }
1939 /* offset paramater is in/out */
1948 {
1949 /*
1950 * This is a project private ufs ioctl() to mark
1951 * the inode as that belonging to a compressed
1952 * file. This is used to mark individual
1953 * compressed files in a miniroot archive.
1954 * The files compressed in this manner are
1955 * automatically decompressed by the dcfs filesystem
1956 * (via an interception in ufs_lookup - see decompvp())
1957 * which is layered on top of ufs on a system running
1958 * from the archive. See uts/common/fs/dcfs for details.
1959 * This ioctl only marks the file as compressed - the
1960 * actual compression is done by fiocompress (a
1961 * userland utility) which invokes this ioctl().
1962 */
1966 ULOCKFS_SETATTR_MASK);
1973 }
1981 }
1985 }
1989 }
1990 }
1993 /* ARGSUSED */
1994 static int
1997 {
2003 /*
2004 * for performance, if only the size is requested don't bother
2005 * with anything else.
2006 */
2009 }
2011 /*
2012 * inlined lockfs checks
2013 */
2018 }
2021 /*
2022 * Return all the attributes. This should be refined so
2023 * that it only returns what's asked for.
2024 */
2026 /*
2027 * Copy from inode table.
2028 */
2031 /*
2032 * If there is an ACL and there is a mask entry, then do the
2033 * extra work that completes the equivalent of an acltomode(3)
2034 * call. According to POSIX P1003.1e, the acl mask should be
2035 * returned in the group permissions field.
2036 *
2037 * - start with the original permission and mode bits (from above)
2038 * - clear the group owner bits
2039 * - add in the mask bits.
2040 */
2045 }
2054 else
2080 }
2085 out:
2087 }
2089 /*
2090 * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2091 * The i_contents lock is already held by the caller and we need to
2092 * declare the inode as 'void *' argument.
2093 */
2094 static int
2096 {
2100 }
2102 /*ARGSUSED4*/
2103 static int
2110 {
2127 timestruc_t now;
2128 vattr_t oldva;
2132 /*
2133 * Cannot set these attributes.
2134 */
2138 /*
2139 * check for forced unmount
2140 */
2148 again:
2159 /*
2160 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2161 * This follows the protocol for read()/write().
2162 */
2164 /*
2165 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2166 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2167 * possible, retries the operation.
2168 */
2174 }
2176 }
2178 /*
2179 * Truncate file. Must have write permission and not be a directory.
2180 */
2186 }
2195 }
2199 }
2205 }
2207 /*
2208 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2209 * This follows the protocol established by
2210 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2211 */
2214 retry_dir);
2218 }
2220 /*
2221 * Grab quota lock if we are changing the file's owner.
2222 */
2226 }
2242 /*
2243 * Change file access modes.
2244 */
2253 else
2256 }
2257 }
2260 /*
2261 * Don't change ownership of the quota inode.
2262 */
2267 }
2269 /*
2270 * No real ownership change.
2271 */
2275 }
2276 /*
2277 * Remove the blocks and the file, from the old user's
2278 * quota.
2279 */
2290 }
2294 /*
2295 * There is a real ownership change.
2296 */
2298 /*
2299 * Add the blocks and the file to the new
2300 * user's quota.
2301 */
2308 }
2309 }
2312 }
2315 }
2316 /*
2317 * Change file access or modified times.
2318 */
2320 /* Check that the time value is within ufs range */
2325 }
2327 /*
2328 * if the "noaccess" mount option is set and only atime
2329 * update is requested, do nothing. No error is returned.
2330 */
2339 }
2345 /*
2346 * In 2038, ctime sticks forever..
2347 */
2353 }
2356 }
2359 }
2361 skip_atime:
2362 /*
2363 * The presence of a shadow inode may indicate an ACL, but does
2364 * not imply an ACL. Future FSD types should be handled here too
2365 * and check for the presence of the attribute-specific data
2366 * before referencing it.
2367 */
2369 /*
2370 * XXX if ufs_iupdat is changed to sandbagged write fix
2371 * ufs_acl_setattr to push ip to keep acls consistent
2372 *
2373 * Suppress out of inodes messages if we will retry.
2374 */
2379 }
2381 update_inode:
2382 /*
2383 * Setattr always increases the sequence number
2384 */
2387 /*
2388 * if nfsd and not logging; push synchronously
2389 */
2394 }
2399 }
2407 trans_size);
2410 }
2412 }
2413 out:
2414 /*
2415 * If out of inodes or blocks, see if we can free something
2416 * up from the delete queue.
2417 */
2426 }
2430 }
2434 }
2436 }
2438 /*ARGSUSED*/
2439 static int
2442 {
2448 /*
2449 * The ufs_iaccess function wants to be called with
2450 * mode bits expressed as "ufs specific" bits.
2451 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2452 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2453 * But since they're the same we just pass the vnode mode
2454 * bit but just verify that assumption at compile time.
2455 */
2456 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2458 #endif
2460 }
2462 /* ARGSUSED */
2463 static int
2466 {
2476 }
2478 /*
2479 * If the symbolic link is empty there is nothing to read.
2480 * Fast-track these empty symbolic links
2481 */
2485 }
2491 /*
2492 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2493 */
2494 again:
2506 }
2512 }
2515 }
2519 ino_t ino;
2539 }
2541 /* can this be a fast symlink and is it a user buffer? */
2547 /*
2548 * setup a kernel buffer to read link into. this
2549 * is to fix a race condition where the user buffer
2550 * got corrupted before copying it into the inode.
2551 */
2567 }
2574 }
2585 /*
2586 * free page
2587 */
2594 /* error, clear garbage left behind */
2599 }
2600 }
2602 /* now, copy it into the user buffer */
2606 }
2609 }
2610 out:
2613 }
2614 nolockout:
2616 }
2618 /* ARGSUSED */
2619 static int
2622 {
2633 /*
2634 * First push out any data pages
2635 */
2642 }
2644 /*
2645 * Delta any delayed inode times updates
2646 * and push inode to log.
2647 * All other inode deltas will have already been delta'd
2648 * and will be pushed during the commit.
2649 */
2654 TOP_SYNCIP_SIZE);
2655 }
2664 TOP_SYNCIP_SIZE);
2665 }
2666 }
2668 /*
2669 * Commit the Moby transaction
2670 *
2671 * Deltas have already been made so we just need to
2672 * commit them with a synchronous transaction.
2673 * TRANS_BEGIN_SYNC() will return an error
2674 * if there are no deltas to commit, for an
2675 * empty transaction.
2676 */
2679 error);
2683 }
2685 TOP_COMMIT_SIZE);
2686 }
2690 /* Just update the inode only */
2694 /* Do data-synchronous writes */
2696 else
2697 /* Do synchronous writes */
2704 }
2705 out:
2708 }
2710 }
2712 /*ARGSUSED*/
2713 static void
2715 {
2717 }
2719 /*
2720 * Unix file system operations having to do with directory manipulation.
2721 */
2723 /* ARGSUSED */
2724 static int
2728 {
2737 /*
2738 * Check flags for type of lookup (regular file or attribute file)
2739 */
2745 /*
2746 * If not mounted with XATTR support then return EINVAL
2747 */
2751 /*
2752 * We don't allow recursive attributes...
2753 * Maybe someday we will.
2754 */
2757 }
2764 }
2768 }
2770 /*
2771 * Check accessibility of directory.
2772 */
2777 }
2781 }
2785 }
2787 /*
2788 * Check for a null component, which we should treat as
2789 * looking at dvp from within it's parent, so we don't
2790 * need a call to ufs_iaccess(), as it has already been
2791 * done.
2792 */
2798 }
2800 /*
2801 * Check for "." ie itself. this is a quick check and
2802 * avoids adding "." into the dnlc (which have been seen
2803 * to occupy >10% of the cache).
2804 */
2806 /*
2807 * Don't return without checking accessibility
2808 * of the directory. We only need the lock if
2809 * we are going to return it.
2810 */
2814 }
2816 }
2818 /*
2819 * Fast path: Check the directory name lookup cache.
2820 */
2822 /*
2823 * Check accessibility of directory.
2824 */
2828 }
2833 }
2837 }
2839 /*
2840 * Keep the idle queue from getting too long by
2841 * idling two inodes before attempting to allocate another.
2842 * This operation must be performed before entering
2843 * lockfs or a transaction.
2844 */
2849 }
2851 retry_lookup:
2852 /*
2853 * Check accessibility of directory.
2854 */
2865 fastpath:
2870 /*
2871 * If vnode is a device return special vnode instead.
2872 */
2877 cr);
2881 else
2886 /*
2887 * Compressed file, substitute dcfs vnode
2888 */
2893 else
2895 }
2896 }
2899 }
2904 out:
2906 }
2908 /*ARGSUSED*/
2909 static int
2913 {
2929 again:
2941 }
2947 /*
2948 * Null component name refers to the directory itself.
2949 */
2951 /*
2952 * Even though this is an error case, we need to grab the
2953 * quota lock since the error handling code below is common.
2954 */
2961 /*
2962 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2963 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2964 * possible, retries the operation.
2965 */
2967 retry_dir);
2976 }
2984 }
2986 /*
2987 * Suppress file system full message if we will retry
2988 */
2996 }
2998 }
3000 }
3005 }
3006 }
3008 /*
3009 * If the file already exists and this is a non-exclusive create,
3010 * check permissions and allow access for non-directories.
3011 * Read-only create of an existing directory is also allowed.
3012 * We fail an exclusive create of anything which already exists.
3013 */
3023 else
3025 }
3031 }
3032 /*
3033 * If the error EEXIST was set, then i_seq can not
3034 * have been updated. The sequence number interface
3035 * is defined such that a non-error VOP_CREATE must
3036 * increase the dir va_seq it by at least one. If we
3037 * have cleared the error, increase i_seq. Note that
3038 * we are increasing the dir i_seq and in rare cases
3039 * ip may actually be from the dvp, so we already have
3040 * the locks and it will not be subject to truncation.
3041 * In case we have to update i_seq of the parent
3042 * directory dip, we have to defer it till we have
3043 * released our locks on ip due to lock ordering requirements.
3044 */
3047 else
3052 /*
3053 * Truncate regular files, if requested by caller.
3054 * Grab i_rwlock to make sure no one else is
3055 * currently writing to the file (we promised
3056 * bmap we would do this).
3057 * Must get the locks in the correct order.
3058 */
3064 /*
3065 * Large Files: Why this check here?
3066 * Though we do it in vn_create() we really
3067 * want to guarantee that we do not destroy
3068 * Large file data by atomically checking
3069 * the size while holding the contents
3070 * lock.
3071 */
3079 }
3081 truncflag++;
3087 retry_file);
3091 }
3093 RW_READER);
3096 cr);
3098 }
3100 }
3103 }
3104 }
3105 }
3111 }
3113 }
3120 /*
3121 * If vnode is a device return special vnode instead.
3122 */
3131 }
3134 }
3135 unlock:
3137 /*
3138 * Do the deferred update of the parent directory's sequence
3139 * number now.
3140 */
3147 }
3153 trans_size);
3155 /*
3156 * If we haven't had a more interesting failure
3157 * already, then anything that might've happened
3158 * here should be reported.
3159 */
3162 }
3171 }
3174 }
3179 /*
3180 * If no inodes available, try to free one up out of the
3181 * pending delete queue.
3182 */
3187 }
3189 out:
3191 }
3194 /*ARGSUSED*/
3195 static int
3198 {
3208 /*
3209 * don't let the delete queue get too long
3210 */
3214 }
3220 /* Only send the event if there were no errors */
3224 }
3226 retry_remove:
3235 /*
3236 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3237 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3238 * possible, retries the operation.
3239 */
3250 }
3252 out:
3254 }
3256 /*
3257 * Link a file or a directory. Only privileged processes are allowed to
3258 * make links to directories.
3259 */
3260 /*ARGSUSED*/
3261 static int
3264 {
3276 retry_link:
3288 /*
3289 * Make sure link for extended attributes is valid
3290 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3291 *
3292 * Make certain we don't attempt to look at a device node as
3293 * a ufs inode.
3294 */
3303 }
3311 }
3313 /*
3314 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3315 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3316 * possible, retries the operation.
3317 */
3325 unlock:
3329 }
3333 }
3334 out:
3336 }
3343 /*
3344 * Rename a file or directory.
3345 * We are given the vnode and entry string of the source and the
3346 * vnode and entry string of the place we want to move the source
3347 * to (the target). The essential operation is:
3348 * unlink(target);
3349 * link(source, target);
3350 * unlink(source);
3351 * but "atomically". Can't do full commit without saving state in
3352 * the inode on disk, which isn't feasible at this time. Best we
3353 * can do is always guarantee that the TARGET exists.
3354 */
3356 /*ARGSUSED*/
3357 static int
3366 {
3377 timestruc_t now;
3393 /* Must do this before taking locks in case of DNLC miss */
3401 /*
3402 * Notify the target directory of the rename event
3403 * if source and target directories are not the same.
3404 */
3410 }
3415 retry_rename:
3429 /*
3430 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3431 */
3435 }
3437 /*
3438 * Check accessibility of directory.
3439 */
3443 /*
3444 * Look up inode of file we're supposed to rename.
3445 */
3453 }
3455 }
3458 }
3460 /*
3461 * Lock both the source and target directories (they may be
3462 * the same) to provide the atomicity semantics that was
3463 * previously provided by the per file system vfs_rename_lock
3464 *
3465 * with vfs_rename_lock removed to allow simultaneous renames
3466 * within a file system, ufs_dircheckpath can deadlock while
3467 * traversing back to ensure that source is not a parent directory
3468 * of target parent directory. This is because we get into
3469 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3470 * If the tdp and sdp of the simultaneous renames happen to be
3471 * in the path of each other, it can lead to a deadlock. This
3472 * can be avoided by getting the locks as RW_READER here and then
3473 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3474 *
3475 * We hold the target directory's i_rwlock after calling
3476 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3477 * VOP_RWLOCK is explicitly called by the filesystem independent code
3478 * before calling the file system operation. In these cases the order
3479 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3480 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3481 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3482 * synchronizing object which might lead to a deadlock. So we use
3483 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3484 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3485 * operation.
3486 */
3487 retry:
3490 retry_firstlock:
3492 /*
3493 * We didn't get the lock. Check if the SLOCK is set in the
3494 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3495 * and wait for SLOCK to be cleared.
3496 */
3500 trans_size);
3505 /*
3506 * SLOCK isn't set so this is a genuine synchronization
3507 * case. Let's try again after giving them a breather.
3508 */
3511 }
3512 }
3513 /*
3514 * Need to check if the tdp and sdp are same !!!
3515 */
3517 /*
3518 * We didn't get the lock. Check if the SLOCK is set in the
3519 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3520 * and wait for SLOCK to be cleared.
3521 */
3526 trans_size);
3531 /*
3532 * So we couldn't get the second level peer lock *and*
3533 * the SLOCK bit isn't set. Too bad we can be
3534 * contentding with someone wanting these locks otherway
3535 * round. Reverse the locks in case there is a heavy
3536 * contention for the second level lock.
3537 */
3541 ufs_rename_retry_cnt++;
3543 }
3544 }
3549 }
3550 /*
3551 * Make sure we can delete the source entry. This requires
3552 * write permission on the containing directory.
3553 * Check for sticky directories.
3554 */
3562 }
3564 /*
3565 * If this is a rename of a directory and the parent is
3566 * different (".." must be changed), then the source
3567 * directory must not be in the directory hierarchy
3568 * above the target, as this would orphan everything
3569 * below the source directory. Also the user must have
3570 * write permission in the source so as to be able to
3571 * change "..".
3572 */
3575 ino_t inum;
3581 }
3586 /*
3587 * If we got EAGAIN ufs_dircheckpath detected a
3588 * potential deadlock and backed out. We need
3589 * to retry the operation since sdp and tdp have
3590 * to be released to avoid the deadlock.
3591 */
3597 ufs_rename_dircheck_retry_cnt++;
3599 }
3601 }
3605 }
3608 /*
3609 * Check for renaming '.' or '..' or alias of '.'
3610 */
3614 }
3616 /*
3617 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3618 * tries to traverse back the file tree with both tdp and sdp held
3619 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3620 * as RW_READERS till ufs_dircheckpath is done.
3621 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3622 * to RW_WRITER.
3623 */
3625 /*
3626 * The upgrade failed. We got to give away the lock
3627 * as to avoid deadlocking with someone else who is
3628 * waiting for writer lock. With the lock gone, we
3629 * cannot be sure the checks done above will hold
3630 * good when we eventually get them back as writer.
3631 * So if we can't upgrade we drop the locks and retry
3632 * everything again.
3633 */
3638 ufs_rename_upgrade_retry_cnt++;
3640 }
3643 /*
3644 * The upgrade failed. We got to give away the lock
3645 * as to avoid deadlocking with someone else who is
3646 * waiting for writer lock. With the lock gone, we
3647 * cannot be sure the checks done above will hold
3648 * good when we eventually get them back as writer.
3649 * So if we can't upgrade we drop the locks and retry
3650 * everything again.
3651 */
3655 ufs_rename_upgrade_retry_cnt++;
3657 }
3658 }
3660 /*
3661 * Now that all the locks are held check to make sure another thread
3662 * didn't slip in and take out the sip.
3663 */
3675 }
3680 /*
3681 * If the inode was found need to drop the v_count
3682 * so as not to keep the filesystem from being
3683 * unmounted at a later time.
3684 */
3686 }
3688 /*
3689 * Release the slot.fbp that has the page mapped and
3690 * locked SE_SHARED, and could be used in in
3691 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3692 * on said page.
3693 */
3697 }
3698 }
3700 /*
3701 * Link source to the target.
3702 */
3704 /*
3705 * ESAME isn't really an error; it indicates that the
3706 * operation should not be done because the source and target
3707 * are the same file, but that no error should be reported.
3708 */
3712 }
3717 /*
3718 * Unlink the source.
3719 * Remove the source entry. ufs_dirremove() checks that the entry
3720 * still reflects sip, and returns an error if it doesn't.
3721 * If the entry has changed just forget about it. Release
3722 * the source inode.
3723 */
3730 /*
3731 * Notify the target directory of the rename event
3732 * if source and target directories are not the same.
3733 */
3736 }
3738 errout:
3745 }
3747 unlock:
3756 }
3759 }
3761 /*ARGSUSED*/
3762 static int
3766 {
3779 /*
3780 * Can't make directory in attr hidden dir
3781 */
3785 again:
3795 /*
3796 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3797 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3798 * possible, retries the operation.
3799 */
3809 trans_size);
3811 }
3813 }
3828 }
3829 out:
3834 }
3837 }
3839 /*ARGSUSED*/
3840 static int
3843 {
3853 /*
3854 * don't let the delete queue get too long
3855 */
3859 }
3865 /* Only send the event if there were no errors */
3869 }
3871 retry_rmdir:
3880 /*
3881 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3882 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3883 * possible, retries the operation.
3884 */
3894 trans_size);
3896 }
3898 out:
3900 }
3902 /* ARGSUSED */
3903 static int
3911 {
3919 caddr_t outbuf;
3921 uint_t offset;
3934 }
3936 /*
3937 * Check if we have been called with a valid iov_len
3938 * and bail out if not, otherwise we may potentially loop
3939 * forever further down.
3940 */
3944 }
3946 /*
3947 * Large Files: When we come here we are guaranteed that
3948 * uio_offset can be used safely. The high word is zero.
3949 */
3959 /* Large Files: directory files should not be "large" */
3963 /* Force offset to be valid (to guard against bogus lseek() values) */
3966 /* Quit if at end of file or link count of zero (posix) */
3972 }
3974 /*
3975 * Get space to change directory entries into fs independent format.
3976 * Do fast alloc for the most commonly used-request size (filesystem
3977 * block size).
3978 */
3986 }
3988 nextblk:
3991 /* Truncate request to file size */
3995 /* Comply with MAXBSIZE boundary restrictions of fbread() */
3999 /*
4000 * Read in the next chunk.
4001 * We are still holding the i_rwlock.
4002 */
4010 }
4020 }
4021 /* Transform to file-system independent format */
4023 /*
4024 * If the current directory entry is mangled, then skip
4025 * to the next block. It would be nice to set the FSBAD
4026 * flag in the super-block so that a fsck is forced on
4027 * next reboot, but locking is a problem.
4028 */
4032 }
4034 /* Skip to requested offset and skip empty entries */
4036 ushort_t this_reclen =
4038 /* Buffer too small for any entries */
4043 }
4044 /* If would overrun the buffer, quit */
4047 }
4048 /* Take this entry */
4053 /* use strncpy(9f) to zero out uninitialized bytes */
4063 }
4071 }
4072 }
4073 /* Release the chunk */
4076 /* Read whole block, but got no entries, read another if not eof */
4078 /*
4079 * Large Files: casting i_size to int here is not a problem
4080 * because directory sizes are always less than MAXOFF32_T.
4081 * See assertion above.
4082 */
4087 /* Copy out the entry data */
4096 update_inode:
4103 unlock:
4106 }
4107 out:
4109 }
4111 /*ARGSUSED*/
4112 static int
4121 {
4132 /*
4133 * No symlinks in attrdirs at this time
4134 */
4138 again:
4151 /*
4152 * We must create the inode before the directory entry, to avoid
4153 * racing with readlink(). ufs_dirmakeinode requires that we
4154 * hold the quota lock as reader, and directory locks as writer.
4155 */
4161 /*
4162 * Suppress any out of inodes messages if we will retry on
4163 * ENOSP
4164 */
4179 /*
4180 * OK. The inode has been created. Write out the data of the
4181 * symbolic link. Since symbolic links are metadata, and should
4182 * remain consistent across a system crash, we need to force the
4183 * data out synchronously.
4184 *
4185 * (This is a change from the semantics in earlier releases, which
4186 * only created symbolic links synchronously if the semi-documented
4187 * 'syncdir' option was set, or if we were being invoked by the NFS
4188 * server, which requires symbolic links to be created synchronously.)
4189 *
4190 * We need to pass in a pointer for the residual length; otherwise
4191 * ufs_rdwri() will always return EIO if it can't write the data,
4192 * even if the error was really ENOSPC or EDQUOT.
4193 */
4201 /*
4202 * Suppress file system full messages if we will retry
4203 */
4216 }
4218 /*
4219 * If the link's data is small enough, we can cache it in the inode.
4220 * This is a "fast symbolic link". We don't use the first direct
4221 * block because that's actually used to point at the symbolic link's
4222 * contents on disk; but we know that none of the other direct or
4223 * indirect blocks can be used because symbolic links are restricted
4224 * to be smaller than a file system block.
4225 */
4234 /* error, clear garbage left behind */
4239 }
4240 }
4245 /*
4246 * OK. We've successfully created the symbolic link. All that
4247 * remains is to insert it into the appropriate directory.
4248 */
4254 /*
4255 * Fall through into remove-on-error code. We're either done, or we
4256 * need to remove the inode (if we couldn't insert it).
4257 */
4259 remove:
4267 }
4269 unlock:
4277 trans_size);
4281 }
4283 /*
4284 * We may have failed due to lack of an inode or of a block to
4285 * store the target in. Try flushing the delete queue to free
4286 * logically-available things up and try again.
4287 */
4292 }
4294 out:
4296 }
4298 /*
4299 * Ufs specific routine used to do ufs io.
4300 */
4301 int
4305 {
4333 }
4339 }
4341 }
4343 /*ARGSUSED*/
4344 static int
4346 {
4356 }
4365 }
4367 /* ARGSUSED2 */
4368 static int
4370 {
4375 /*
4376 * Read case is easy.
4377 */
4381 }
4383 /*
4384 * Caller has requested a writer lock, but that inhibits any
4385 * concurrency in the VOPs that follow. Acquire the lock shared
4386 * and defer exclusive access until it is known to be needed in
4387 * other VOP handlers. Some cases can be determined here.
4388 */
4390 /*
4391 * If directio is not set, there is no chance of concurrency,
4392 * so just acquire the lock exclusive. Beware of a forced
4393 * unmount before looking at the mount option.
4394 */
4401 }
4403 /*
4404 * Mandatory locking forces acquiring i_rwlock exclusive.
4405 */
4409 }
4411 /*
4412 * Acquire the lock shared in case a concurrent write follows.
4413 * Mandatory locking could have become enabled before the lock
4414 * was acquired. Re-check and upgrade if needed.
4415 */
4421 }
4423 }
4425 /*ARGSUSED*/
4426 static void
4428 {
4432 }
4434 /* ARGSUSED */
4435 static int
4438 {
4440 }
4442 /* ARGSUSED */
4443 static int
4447 {
4453 /*
4454 * If file is being mapped, disallow frlock.
4455 * XXX I am not holding tlock while checking i_mapcnt because the
4456 * current locking strategy drops all locks before calling fs_frlock.
4457 * So, mapcnt could change before we enter fs_frlock making is
4458 * meaningless to have held tlock in the first place.
4459 */
4463 }
4465 /* ARGSUSED */
4466 static int
4469 {
4477 ULOCKFS_SPACE_MASK);
4486 ULOCKFS_FALLOCATE_MASK);
4496 }
4498 }
4500 /*
4501 * Used to determine if read ahead should be done. Also used to
4502 * to determine when write back occurs.
4503 */
4504 #define CLUSTSZ(ip) ((ip)->i_ufsvfs->vfs_ioclustsz)
4506 /*
4507 * A faster version of ufs_getpage.
4508 *
4509 * We optimize by inlining the pvn_getpages iterator, eliminating
4510 * calls to bmap_read if file doesn't have UFS holes, and avoiding
4511 * the overhead of page_exists().
4512 *
4513 * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4514 * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4515 * victimizes performance when a file with UFS holes is faulted
4516 * first in the S_READ mode, and then in the S_WRITE mode. We will get
4517 * two MMU faults in this case.
4518 *
4519 * XXX - the inode fields which control the sequential mode are not
4520 * protected by any mutex. The read ahead will act wild if
4521 * multiple processes will access the file concurrently and
4522 * some of them in sequential mode. One particulary bad case
4523 * is if another thread will change the value of i_nextrio between
4524 * the time this thread tests the i_nextrio value and then reads it
4525 * again to use it as the offset for the read ahead.
4526 */
4527 /*ARGSUSED*/
4528 static int
4532 {
4534 u_offset_t pgoff;
4535 u_offset_t eoff;
4541 caddr_t pgaddr;
4542 krw_t rwtype;
4557 /*
4558 * Obey the lockfs protocol
4559 */
4569 /*
4570 * Try to start a transaction, will return if blocking is
4571 * expected to occur and the address space is not the
4572 * kernel address space.
4573 */
4579 /*
4580 * Use EDEADLK here because the VM code
4581 * can normally never see this error.
4582 */
4586 }
4589 }
4590 }
4595 }
4601 /*
4602 * If this thread owns the lock, i.e., this thread grabbed it
4603 * as writer somewhere above, then we don't need to grab the
4604 * lock as reader in this routine.
4605 */
4608 retrylock:
4610 /*
4611 * Grab the quota lock if we need to call
4612 * bmap_write() below (with i_contents as writer).
4613 */
4617 }
4619 /*
4620 * We may be getting called as a side effect of a bmap using
4621 * fbread() when the blocks might be being allocated and the
4622 * size has not yet been up'ed. In this case we want to be
4623 * able to return zero pages if we get back UFS_HOLE from
4624 * calling bmap for a non write case here. We also might have
4625 * to read some frags from the disk into a page if we are
4626 * extending the number of frags for a given lbn in bmap().
4627 * Large Files: The read of i_size here is atomic because
4628 * i_contents is held here. If dolock is zero, the lock
4629 * is held in bmap routines.
4630 */
4638 }
4641 }
4643 /*
4644 * Must hold i_contents lock throughout the call to pvn_getpages
4645 * since locked pages are returned from each call to ufs_getapage.
4646 * Must *not* return locked pages and then try for contents lock
4647 * due to lock ordering requirements (inode > page)
4648 */
4654 u_offset_t offset;
4656 /*
4657 * We must acquire the RW_WRITER lock in order to
4658 * call bmap_write().
4659 */
4663 /*
4664 * Grab the quota lock before
4665 * upgrading i_contents, but if we can't grab it
4666 * don't wait here due to lock order:
4667 * vfs_dqrwlock > i_contents.
4668 */
4674 }
4680 }
4681 }
4683 /*
4684 * May be allocating disk blocks for holes here as
4685 * a result of mmap faults. write(2) does the bmap_write
4686 * in rdip/wrip, not here. We are not dealing with frags
4687 * in this case.
4688 */
4689 /*
4690 * Large Files: We cast fs_bmask field to offset_t
4691 * just as we do for MAXBMASK because uoff is a 64-bit
4692 * data type. fs_bmask will still be a 32-bit type
4693 * as we cannot change any ondisk data structures.
4694 */
4706 }
4707 }
4709 /*
4710 * Can be a reader from now on.
4711 */
4714 /*
4715 * We can release vfs_dqrwlock early so do it, but make
4716 * sure we don't try to release it again at the bottom.
4717 */
4721 }
4722 }
4724 /*
4725 * We remove PROT_WRITE in cases when the file has UFS holes
4726 * because we don't want to call bmap_read() to check each
4727 * page if it is backed with a disk block.
4728 */
4734 /*
4735 * The loop looks up pages in the range [off, off + len).
4736 * For each page, we first check if we should initiate an asynchronous
4737 * read ahead before we call page_lookup (we may sleep in page_lookup
4738 * for a previously initiated disk read).
4739 */
4744 u_offset_t nextrio;
4745 se_t se;
4750 /* Handle async getpage (faultahead) */
4757 }
4758 /*
4759 * Check if we should initiate read ahead of next cluster.
4760 * We call page_exists only when we need to confirm that
4761 * we have the current page before we initiate the read ahead.
4762 */
4768 /*
4769 * We always read ahead the next cluster of data
4770 * starting from i_nextrio. If the page (vp,nextrio)
4771 * is actually in core at this point, the routine
4772 * ufs_getpage_ra() will stop pre-fetching data
4773 * until we read that page in a synchronized manner
4774 * through ufs_getpage_miss(). So, we should increase
4775 * i_nextrio if the page (vp, nextrio) exists.
4776 */
4779 }
4780 }
4783 /*
4784 * We found the page in the page cache.
4785 */
4792 /*
4793 * We have to create the page, or read it from disk.
4794 */
4800 pl++;
4805 }
4806 }
4807 }
4809 /*
4810 * Return pages up to plsz if they are in the page cache.
4811 * We cannot return pages if there is a chance that they are
4812 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4813 */
4830 }
4831 }
4837 error:
4839 /*
4840 * Release any pages we have locked.
4841 */
4846 }
4848 update_inode:
4849 /*
4850 * If the inode is not already marked for IACC (in rdip() for read)
4851 * and the inode is not marked for no access time update (in wrip()
4852 * for write) then update the inode access time and mod time now.
4853 */
4863 }
4864 }
4865 }
4871 }
4873 unlock:
4878 }
4880 }
4881 out:
4883 }
4885 /*
4886 * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4887 * cache. The page is either read from the disk, or it's created.
4888 * A page is created (without disk read) if rw == S_CREATE, or if
4889 * the page is not backed with a real disk block (UFS hole).
4890 */
4891 /* ARGSUSED */
4892 static int
4895 {
4898 daddr_t bn;
4905 /*
4906 * Figure out whether the page can be created, or must be
4907 * must be read from the disk.
4908 */
4918 /*
4919 * If its also a fallocated block that hasn't been written to
4920 * yet, we will treat it just like a UFS_HOLE and create
4921 * a zero page for it
4922 */
4925 }
4932 }
4939 u_offset_t io_off;
4940 uint_t xlen;
4944 /*
4945 * If access is not in sequential order, we read from disk
4946 * in bsize units.
4947 *
4948 * We limit the size of the transfer to bsize if we are reading
4949 * from the beginning of the file. Note in this situation we
4950 * will hedge our bets and initiate an async read ahead of
4951 * the second block.
4952 */
4959 /*
4960 * Some other thread has entered the page.
4961 * ufs_getpage will retry page_lookup.
4962 */
4966 }
4968 /*
4969 * Zero part of the page which we are not
4970 * going to read from the disk.
4971 */
4993 }
4997 /*
4998 * If the file access is sequential, initiate read ahead
4999 * of the next cluster.
5000 */
5009 }
5010 }
5014 }
5016 /*
5017 * Read ahead a cluster from the disk. Returns the length in bytes.
5018 */
5019 static int
5021 {
5028 daddr_t bn;
5035 /*
5036 * If the directio advisory is in effect on this file,
5037 * then do not do buffered read ahead. Read ahead makes
5038 * it more difficult on threads using directio as they
5039 * will be forced to flush the pages from this vnode.
5040 */
5046 /*
5047 * Is this test needed?
5048 */
5054 /*
5055 * If its a UFS_HOLE or a fallocated block, do not perform
5056 * any read ahead's since there probably is nothing to read ahead
5057 */
5061 /*
5062 * Limit the transfer size to bsize if this is the 2nd block.
5063 */
5071 /*
5072 * Zero part of page which we are not going to read from disk
5073 */
5096 }
5099 }
5102 /*
5103 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5104 *
5105 * LMXXX - the inode really ought to contain a pointer to one of these
5106 * async args. Stuff gunk in there and just hand the whole mess off.
5107 * This would replace i_delaylen, i_delayoff.
5108 */
5109 /*ARGSUSED*/
5110 static int
5113 {
5119 }
5121 /*
5122 * XXX - Why should this check be made here?
5123 */
5127 }
5132 }
5138 /*
5139 * If nobody stalled, start a new cluster.
5140 */
5146 }
5147 /*
5148 * If we have a full cluster or they are not contig,
5149 * then push last cluster and start over.
5150 */
5153 u_offset_t doff;
5163 /* LMXXX - flags are new val, not old */
5165 }
5166 /*
5167 * There is something there, it's not full, and
5168 * it is contig.
5169 */
5173 }
5174 /*
5175 * Must have weird flags or we are not clustering.
5176 */
5177 }
5181 errout:
5183 }
5185 /*
5186 * If len == 0, do from off to EOF.
5187 *
5188 * The normal cases should be len == 0 & off == 0 (entire vp list),
5189 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5190 * (from pageout).
5191 */
5192 /*ARGSUSED*/
5193 static int
5196 offset_t off,
5200 {
5201 u_offset_t io_off;
5202 u_offset_t eoff;
5211 /*
5212 * Acquire the readers/write inode lock before locking
5213 * any pages in this inode.
5214 * The inode lock is held during i/o.
5215 */
5220 }
5223 /*
5224 * Must synchronize this thread and any possible thread
5225 * operating in the window of vulnerability in wrip().
5226 * It is dangerous to allow both a thread doing a putpage
5227 * and a thread writing, so serialize them. The exception
5228 * is when the thread in wrip() does something which causes
5229 * a putpage operation. Then, the thread must be allowed
5230 * to continue. It may encounter a bmap_read problem in
5231 * ufs_putapage, but that is handled in ufs_putapage.
5232 * Allow async writers to proceed, we don't want to block
5233 * the pageout daemon.
5234 */
5241 /*
5242 * If there is no thread in the critical
5243 * section of wrip(), then proceed.
5244 * Otherwise, wait until there isn't one.
5245 */
5249 }
5251 /*
5252 * Bounce async writers when we have a writer
5253 * working on this file so we don't deadlock
5254 * the pageout daemon.
5255 */
5259 }
5262 }
5263 }
5264 }
5270 }
5273 /*
5274 * Search the entire vp list for pages >= off.
5275 */
5279 /*
5280 * Loop over all offsets in the range looking for
5281 * pages to deal with.
5282 */
5285 else
5289 /*
5290 * If we are not invalidating, synchronously
5291 * freeing or writing pages, use the routine
5292 * page_lookup_nowait() to prevent reclaiming
5293 * them from the free list.
5294 */
5302 }
5313 /*
5314 * "io_off" and "io_len" are returned as
5315 * the range of pages we actually wrote.
5316 * This allows us to skip ahead more quickly
5317 * since several pages may've been dealt
5318 * with by this iteration of the loop.
5319 */
5320 }
5321 }
5322 }
5324 /*
5325 * We have just sync'ed back all the pages on
5326 * the inode, turn off the IMODTIME flag.
5327 */
5331 }
5335 }
5337 static void
5339 {
5355 }
5359 }
5361 /*
5362 * Write out a single page, possibly klustering adjacent
5363 * dirty pages. The inode lock must be held.
5364 *
5365 * LMXXX - bsize < pagesize not done.
5366 */
5367 /*ARGSUSED*/
5368 int
5376 {
5377 u_offset_t io_off;
5378 u_offset_t off;
5384 daddr_t bn;
5394 }
5399 /*
5400 * If the modified time on the inode has not already been
5401 * set elsewhere (e.g. for write/setattr) we set the time now.
5402 * This gives us approximate modified times for mmap'ed files
5403 * which are modified via stores in the user address space.
5404 */
5411 }
5413 /*
5414 * Align the request to a block boundry (for old file systems),
5415 * and go ask bmap() how contiguous things are for this file.
5416 */
5423 /*
5424 * logging device is in error mode; simply return EIO
5425 */
5429 }
5430 /*
5431 * Oops, the thread in the window in wrip() did some
5432 * sort of operation which caused a putpage in the bad
5433 * range. In this case, just return an error which will
5434 * cause the software modified bit on the page to set
5435 * and the page will get written out again later.
5436 */
5440 }
5441 /*
5442 * If the pager is trying to push a page in the bad range
5443 * just tell it to try again later when things are better.
5444 */
5448 }
5451 }
5453 /*
5454 * If it is an fallocate'd block, reverse the negativity since
5455 * we are now writing to it
5456 */
5463 }
5465 /*
5466 * Take the length (of contiguous bytes) passed back from bmap()
5467 * and _try_ and get a set of pages covering that extent.
5468 */
5471 /*
5472 * May have run out of memory and not clustered backwards.
5473 * off p_offset
5474 * [ pp - 1 ][ pp ]
5475 * [ block ]
5476 * We told bmap off, so we have to adjust the bn accordingly.
5477 */
5481 }
5483 /*
5484 * bmap was carefull to tell us the right size so use that.
5485 * There might be unallocated frags at the end.
5486 * LMXXX - bzero the end of the page? We must be writing after EOF.
5487 */
5491 }
5493 /*
5494 * Handle the case where we are writing the last page after EOF.
5495 *
5496 * XXX - just a patch for i-mt3.
5497 */
5502 }
5514 /*
5515 * File contents of shadow or quota inodes are metadata, and updates
5516 * to these need to be put into a logging transaction. All direct
5517 * callers in UFS do that, but fsflush can come here _before_ the
5518 * normal codepath. An example would be updating ACL information, for
5519 * which the normal codepath would be:
5520 * ufs_si_store()
5521 * ufs_rdwri()
5522 * wrip()
5523 * segmap_release()
5524 * VOP_PUTPAGE()
5525 * Here, fsflush can pick up the dirty page before segmap_release()
5526 * forces it out. If that happens, there's no transaction.
5527 * We therefore need to test whether a transaction exists, and if not
5528 * create one - for fsflush.
5529 */
5530 dotrans =
5538 }
5545 }
5546 }
5550 }
5552 /* write throttle */
5570 }
5581 }
5585 }
5589 out:
5597 out_trace:
5599 }
5604 /* ARGSUSED */
5605 static int
5607 offset_t off,
5611 uchar_t prot,
5612 uchar_t maxprot,
5613 uint_t flags,
5616 {
5621 k_sigset_t smask;
5627 }
5632 }
5637 }
5639 retry_map:
5641 /*
5642 * If file is being locked, disallow mapping.
5643 */
5647 }
5650 /*
5651 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5652 * the previous attempt), some other thread could have grabbed
5653 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5654 * would unmap the valid VA range, that is ok.
5655 */
5660 }
5662 /*
5663 * a_lock has to be acquired before entering the lockfs protocol
5664 * because that is the order in which pagefault works. Also we cannot
5665 * block on a_lock here because this waiting writer will prevent
5666 * further readers like ufs_read from progressing and could cause
5667 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5668 * pending.
5669 */
5671 ufs_map_alock_retry_cnt++;
5673 }
5675 /*
5676 * We can't hold as->a_lock and wait for lockfs to succeed because
5677 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5678 * instead.
5679 */
5681 /*
5682 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5683 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5684 */
5685 ufs_map_lockfs_retry_cnt++;
5703 }
5704 }
5705 }
5708 }
5725 out:
5727 }
5729 /* ARGSUSED */
5730 static int
5732 offset_t off,
5734 caddr_t addr,
5736 uchar_t prot,
5737 uchar_t maxprot,
5738 uint_t flags,
5741 {
5746 }
5752 }
5754 /*ARGSUSED*/
5755 static int
5759 {
5764 }
5771 }
5772 /*
5773 * Return the answer requested to poll() for non-device files
5774 */
5777 /* ARGSUSED */
5778 int
5781 {
5790 }
5813 }
5817 out:
5821 }
5823 /* ARGSUSED */
5824 static int
5827 {
5840 /*
5841 * Have to handle _PC_NAME_MAX here, because the normal way
5842 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5843 * results in a lock ordering reversal between
5844 * ufs_lockfs_{begin,end}() and
5845 * ufs_thread_{suspend,continue}().
5846 *
5847 * Keep in sync with ufs_statvfs().
5848 */
5856 else
5863 error =
5866 /* Start transaction */
5870 }
5871 /*
5872 * Is directory empty
5873 */
5891 }
5896 }
5899 }
5917 /*
5918 * UFS keeps only microsecond timestamp resolution.
5919 * This is historical and will probably never change.
5920 */
5927 }
5931 }
5933 }
5937 /*ARGSUSED*/
5938 static int
5941 {
5946 daddr_t bn;
5957 }
5960 /*
5961 * We need a better check. Ideally, we would use another
5962 * vnodeops so that hlocked and forcibly unmounted file
5963 * systems would return EIO where appropriate and w/o the
5964 * need for these checks.
5965 */
5969 /*
5970 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5971 * ul_lock must be taken before locking pages so we can't use it here
5972 * if pp is non NULL because segvn already locked pages
5973 * SE_EXCL. Instead we rely on the fact that a forced umount or
5974 * applying a filesystem lock via ufs_fiolfs() will block in the
5975 * implicit call to ufs_flush() until we unlock the pages after the
5976 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5977 * above 0 until they are done. We have to be careful not to increment
5978 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5979 *
5980 * If pp is NULL use ul_lock to make sure we don't increment
5981 * ul_vnops_cnt after forceful unmount hlocks the file system.
5982 */
5990 }
5992 }
6000 }
6001 }
6004 /*
6005 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
6006 * handle a fault against a segment that maps vnode pages with
6007 * large mappings. Segvn creates pages and holds them locked
6008 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
6009 * use rw_tryenter() to avoid a potential deadlock since in
6010 * lock order i_contents needs to be taken first.
6011 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
6012 */
6019 }
6020 }
6022 /*
6023 * Return an error to segvn because the pagefault request is beyond
6024 * PAGESIZE rounded EOF.
6025 */
6032 }
6039 }
6045 }
6047 /*
6048 * Break the io request into chunks, one for each contiguous
6049 * stretch of disk blocks in the target file.
6050 */
6062 }
6065 }
6068 /*
6069 * Zero out a page beyond EOF, when the last block of
6070 * a file is a UFS fragment so that ufs_pageio() can be used
6071 * instead of ufs_getpage() to handle faults against
6072 * segvn segments that use large pages.
6073 */
6078 }
6093 else
6097 ufs_pageio_reads++;
6098 else
6099 ufs_pageio_writes++;
6102 else
6104 /*
6105 * If the request is not B_ASYNC, wait for i/o to complete
6106 * and re-assemble the page list to return to the caller.
6107 * If it is B_ASYNC we leave the page list in pieces and
6108 * cleanup() will dispose of them.
6109 */
6116 }
6122 }
6126 /* Cleanup unprocessed parts of list */
6130 else
6133 /* Re-assemble list and let caller clean up */
6136 }
6137 }
6145 }
6152 }
6154 /*
6155 * Called when the kernel is in a frozen state to dump data
6156 * directly to the device. It uses a private dump data structure,
6157 * set up by dump_ctl, to locate the correct disk block to which to dump.
6158 */
6159 /*ARGSUSED*/
6160 static int
6163 {
6164 u_offset_t file_size;
6172 /*
6173 * forced unmount case
6174 */
6177 /*
6178 * Validate the inode that it has not been modified since
6179 * the dump structure is allocated.
6180 */
6188 }
6191 /*
6192 * See that the file has room for this write
6193 */
6199 /*
6200 * Find the physical disk block numbers from the dump
6201 * private data structure directly and write out the data
6202 * in contiguous block lumps
6203 */
6211 nfsbs++;
6213 }
6220 }
6223 }
6225 /*
6226 * Prepare the file system before and after the dump operation.
6227 *
6228 * action = DUMP_ALLOC:
6229 * Preparation before dump, allocate dump private data structure
6230 * to hold all the direct and indirect block info for dump.
6231 *
6232 * action = DUMP_FREE:
6233 * Clean up after dump, deallocate the dump private data structure.
6234 *
6235 * action = DUMP_SCAN:
6236 * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6237 * if found, the starting file-relative DEV_BSIZE lbn is written
6238 * to *bklp; that lbn is intended for use with VOP_DUMP()
6239 */
6240 /*ARGSUSED*/
6241 static int
6243 {
6253 /*
6254 * check for forced unmount
6255 */
6260 /*
6261 * alloc and record dump_info
6262 */
6274 }
6276 /*
6277 * calculate and allocate space needed according to i_size
6278 */
6285 }
6287 /* Start saving the info */
6292 /* Direct Blocks */
6296 /* Indirect Blocks */
6309 }
6319 }
6320 }
6321 /* and time stamp the information */
6328 /*
6329 * free dump_info
6330 */
6338 /*
6339 * scan dump_info
6340 */
6350 /*
6351 * scan dblk[] entries; contig fs space is found when:
6352 * ((current blkno + frags per block) == next blkno)
6353 */
6357 n++;
6358 else
6360 dblk++;
6361 nextblk++;
6362 }
6364 /*
6365 * index is where size bytes of contig space begins;
6366 * conversion from index to the file's DEV_BSIZE lbn
6367 * is equivalent to: (index * fs_bsize) / DEV_BSIZE
6368 */
6374 }
6376 }
6378 /*
6379 * Recursive helper function for ufs_dumpctl(). It follows the indirect file
6380 * system blocks until it reaches the the disk block addresses, which are
6381 * then stored into the given buffer, storeblk.
6382 */
6386 {
6396 }
6398 }
6407 }
6414 }
6416 }
6418 /* ARGSUSED */
6419 static int
6422 {
6431 /*
6432 * Only grab locks if needed - they're not needed to check vsa_mask
6433 * or if the mask contains no acl flags.
6434 */
6437 ULOCKFS_GETATTR_MASK))
6446 }
6448 }
6450 /* ARGSUSED */
6451 static int
6454 {
6467 /* Abort now if the request is either empty or invalid. */
6474 }
6476 /*
6477 * Following convention, if this is a directory then we acquire the
6478 * inode's i_rwlock after starting a UFS logging transaction;
6479 * otherwise, we acquire it beforehand. Since we were called (and
6480 * must therefore return) with the lock held, we will have to drop it,
6481 * and later reacquire it, if operating on a directory.
6482 */
6487 /* Upgrade the lock if required. */
6491 }
6492 }
6494 again:
6500 }
6502 /*
6503 * Check that the file system supports this operation. Note that
6504 * ufs_lockfs_begin() will have checked that the file system had
6505 * not been forcibly unmounted.
6506 */
6510 }
6514 }
6520 }
6525 }
6529 /* Do the actual work. */
6531 /*
6532 * Suppress out of inodes messages if we will retry.
6533 */
6540 out:
6543 /*
6544 * top_end_async() can eventually call
6545 * top_end_sync(), which can block. We must
6546 * therefore observe the lock-ordering protocol
6547 * here as well.
6548 */
6552 }
6554 }
6556 }
6557 /*
6558 * If no inodes available, try scaring a logically-
6559 * free one out of the delete queue to someplace
6560 * that we can find it.
6561 */
6568 }
6570 }
6571 /*
6572 * If we need to reacquire the lock then it is safe to do so
6573 * as a reader. This is because ufs_rwunlock(), which will be
6574 * called by our caller after we return, does not differentiate
6575 * between shared and exclusive locks.
6576 */
6580 }
6583 }
6585 /*
6586 * Locate the vnode to be used for an event notification. As this will
6587 * be called prior to the name space change perform basic verification
6588 * that the change will be allowed.
6589 */
6591 static int
6594 {
6614 }
6615 }
6617 /*
6618 * Check accessibility and write access of parent directory as we
6619 * only want to post the event if we're able to make a change.
6620 */
6628 }
6632 }
6634 /*
6635 * Keep the idle queue from getting too long by idling two
6636 * inodes before attempting to allocate another.
6637 * This operation must be performed before entering lockfs
6638 * or a transaction.
6639 */
6644 }
6648 retry_lookup:
6655 }
6659 }
6665 }