| blob | 4a5a2939e6c1d2d39a0f5f2ff1a4ac21ede0b42d |
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 2018 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/vnode.h>
50 #include <sys/proc.h>
51 #include <sys/disp.h>
52 #include <sys/file.h>
53 #include <sys/fcntl.h>
54 #include <sys/flock.h>
55 #include <sys/atomic.h>
56 #include <sys/kmem.h>
57 #include <sys/uio.h>
58 #include <sys/dnlc.h>
59 #include <sys/conf.h>
60 #include <sys/mman.h>
61 #include <sys/pathname.h>
62 #include <sys/debug.h>
63 #include <sys/vmsystm.h>
64 #include <sys/cmn_err.h>
65 #include <sys/filio.h>
66 #include <sys/policy.h>
68 #include <sys/fs/ufs_fs.h>
69 #include <sys/fs/ufs_lockfs.h>
70 #include <sys/fs/ufs_filio.h>
71 #include <sys/fs/ufs_inode.h>
72 #include <sys/fs/ufs_fsdir.h>
73 #include <sys/fs/ufs_quota.h>
74 #include <sys/fs/ufs_log.h>
75 #include <sys/fs/ufs_snap.h>
76 #include <sys/fs/ufs_trans.h>
77 #include <sys/fs/ufs_panic.h>
78 #include <sys/fs/ufs_bio.h>
80 #include <sys/errno.h>
81 #include <sys/fssnap_if.h>
82 #include <sys/unistd.h>
83 #include <sys/sunddi.h>
87 #include <vm/hat.h>
88 #include <vm/page.h>
89 #include <vm/pvn.h>
90 #include <vm/as.h>
91 #include <vm/seg.h>
92 #include <vm/seg_map.h>
93 #include <vm/seg_vn.h>
94 #include <vm/seg_kmem.h>
95 #include <vm/rm.h>
96 #include <sys/swap.h>
98 #include <sys/fs_subr.h>
100 #include <sys/fs/decomp.h>
109 caller_context_t *);
117 caller_context_t *);
119 caller_context_t *);
121 caller_context_t *);
143 caller_context_t *);
152 caller_context_t *);
159 caller_context_t *);
168 caller_context_t *);
170 caller_context_t *);
172 caller_context_t *);
179 caller_context_t *);
181 caller_context_t *);
186 /*
187 * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
188 *
189 * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
190 *
191 * NOTE: "not blkd" below means that the operation isn't blocked by lockfs
192 */
234 };
236 #define MAX_BACKFILE_COUNT 9999
238 /*
239 * Created by ufs_dumpctl() to store a file's disk block info into memory.
240 * Used by ufs_dump() to dump data to disk directly.
241 */
247 };
251 /*
252 * Previously there was no special action required for ordinary files.
253 * (Devices are handled through the device file system.)
254 * Now we support Large Files and Large File API requires open to
255 * fail if file is large.
256 * We could take care to prevent data corruption
257 * by doing an atomic check of size and truncate if file is opened with
258 * FTRUNC flag set but traditionally this is being done by the vfs/vnode
259 * layers. So taking care of truncation here is a change in the existing
260 * semantics of fop_open and therefore we chose not to implement any thing
261 * here. The check for the size of the file > 2GB is being done at the
262 * vfs layer in routine vn_open().
263 */
265 /* ARGSUSED */
266 static int
268 {
270 }
272 /*ARGSUSED*/
273 static int
276 {
280 /*
281 * Push partially filled cluster at last close.
282 * ``last close'' is approximated because the dnlc
283 * may have a hold on the vnode.
284 * Checking for VBAD here will also act as a forced umount check.
285 */
293 }
294 }
297 }
299 /*ARGSUSED*/
300 static int
303 {
312 /*
313 * Mandatory locking needs to be done before ufs_lockfs_begin()
314 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
315 */
317 /*
318 * ufs_getattr ends up being called by chklock
319 */
324 }
331 /*
332 * In the case that a directory is opened for reading as a file
333 * (eg "cat .") with the O_RSYNC, O_SYNC and O_DSYNC flags set.
334 * The locking order had to be changed to avoid a deadlock with
335 * an update taking place on that directory at the same time.
336 */
347 }
356 TOP_READ_SIZE);
358 }
360 /*
361 * Only transact reads to files opened for sync-read and
362 * sync-write on a file system that is not write locked.
363 *
364 * The ``not write locked'' check prevents problems with
365 * enabling/disabling logging on a busy file system. E.g.,
366 * logging exists at the beginning of the read but does not
367 * at the end.
368 *
369 */
376 }
384 TOP_READ_SIZE);
385 }
386 }
390 }
391 out:
394 }
402 static int
404 {
407 /*
408 * If the FDSYNC flag is set then ignore the global
409 * ufs_allow_shared_writes in this case.
410 */
413 /*
414 * Filter to determine if this request is suitable as a
415 * concurrent rewrite. This write must not allocate blocks
416 * by extending the file or filling in holes. No use trying
417 * through FSYNC descriptors as the inode will be synchronously
418 * updated after the write. The uio structure has not yet been
419 * checked for sanity, so assume nothing.
420 */
426 shared_write);
427 }
429 /*ARGSUSED*/
430 static int
433 {
446 retry_mandlock:
447 /*
448 * Mandatory locking needs to be done before ufs_lockfs_begin()
449 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
450 * Check for forced unmounts normally done in ufs_lockfs_begin().
451 */
455 }
460 /*
461 * ufs_getattr ends up being called by chklock
462 */
467 }
469 /* i_rwlock can change in chklock */
473 /*
474 * Check for fast-path special case of directio re-writes.
475 */
487 uint_t i_flag_save;
491 /*
492 * Special treatment of access times for re-writes.
493 * If IMOD is not already set, then convert it
494 * to IMODACC for this operation. This defers
495 * entering a delta into the log until the inode
496 * is flushed. This mimics what is done for read
497 * operations and inode access time.
498 */
507 }
513 }
517 }
522 /*
523 * Mandatory locking could have been enabled
524 * after dropping the i_rwlock.
525 */
528 }
534 /*
535 * Amount of log space needed for this write
536 */
540 /*
541 * Throttle writes.
542 */
546 ufs_throttles++;
548 }
550 }
552 /*
553 * Enter Transaction
554 *
555 * If the write is a rewrite there is no need to open a transaction
556 * if the FDSYNC flag is set and not the FSYNC. In this case just
557 * set the IMODACC flag to modify do the update at a later time
558 * thus avoiding the overhead of the logging transaction that is
559 * not required.
560 */
564 uint_t i_flag_save;
575 }
583 }
584 }
588 }
590 /*
591 * Write the file
592 */
596 /*
597 * In append mode start at end of file.
598 */
600 }
602 /*
603 * Mild optimisation, don't call ufs_trans_write() unless we have to
604 * Also, suppress file system full messages if we will retry.
605 */
612 }
618 /*
619 * Leave Transaction
620 */
630 }
633 }
635 }
636 out:
638 /*
639 * Any blocks tied up in pending deletes?
640 */
644 }
650 }
652 /*
653 * Don't cache write blocks to files with the sticky bit set.
654 * Used to keep swap files from blowing the page cache on a server.
655 */
658 /*
659 * Free behind hacks. The pager is busted.
660 * XXX - need to pass the information down to writedone() in a flag like B_SEQ
661 * or B_FREE_IF_TIGHT_ON_MEMORY.
662 */
667 /*
668 * While we should, in most cases, cache the pages for write, we
669 * may also want to cache the pages for read as long as they are
670 * frequently re-usable.
671 *
672 * If cache_read_ahead = 1, the pages for read will go to the tail
673 * of the cache list when they are released, otherwise go to the head.
674 */
677 /*
678 * Freebehind exists so that as we read large files sequentially we
679 * don't consume most of memory with pages from a few files. It takes
680 * longer to re-read from disk multiple small files as it does reading
681 * one large one sequentially. As system memory grows customers need
682 * to retain bigger chunks of files in memory. The advent of the
683 * cachelist opens up of the possibility freeing pages to the head or
684 * tail of the list.
685 *
686 * Not freeing a page is a bet that the page will be read again before
687 * it's segmap slot is needed for something else. If we loose the bet,
688 * it means some other thread is burdened with the page free we did
689 * not do. If we win we save a free and reclaim.
690 *
691 * Freeing it at the tail vs the head of cachelist is a bet that the
692 * page will survive until the next read. It's also saying that this
693 * page is more likely to be re-used than a page freed some time ago
694 * and never reclaimed.
695 *
696 * Freebehind maintains a range of file offset [smallfile1; smallfile2]
697 *
698 * 0 < offset < smallfile1 : pages are not freed.
699 * smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
700 * smallfile2 < offset : pages freed to head of cachelist.
701 *
702 * The range is computed at most once per second and depends on
703 * freemem and ncpus_online. Both parameters are bounded to be
704 * >= smallfile && >= smallfile64.
705 *
706 * smallfile1 = (free memory / ncpu) / 1000
707 * smallfile2 = (free memory / ncpu) / 10
708 *
709 * A few examples values:
710 *
711 * Free Mem (in Bytes) [smallfile1; smallfile2] [smallfile1; smallfile2]
712 * ncpus_online = 4 ncpus_online = 64
713 * ------------------ ----------------------- -----------------------
714 * 1G [256K; 25M] [32K; 1.5M]
715 * 10G [2.5M; 250M] [156K; 15M]
716 * 100G [25M; 2.5G] [1.5M; 150M]
717 *
718 */
720 #define SMALLFILE1_D 1000
721 #define SMALLFILE2_D 10
728 /*
729 * wrip does the real work of write requests for ufs.
730 */
731 int
733 {
735 uoff_t off;
736 uoff_t old_i_size;
740 caddr_t base;
743 uint_t flags;
752 /*
753 * ip->i_size is incremented before the uiomove
754 * is done on a write. If the move fails (bad user
755 * address) reset ip->i_size.
756 * The better way would be to increment ip->i_size
757 * only if the uiomove succeeds.
758 */
760 o_mode_t type;
765 /*
766 * check for forced unmount - should not happen as
767 * the request passed the lockfs checks.
768 */
776 /* check for valid filetype */
781 }
783 /*
784 * the actual limit of UFS file size
785 * is UFS_MAXOFFSET_T
786 */
798 }
800 /*
801 * if largefiles are disallowed, the limit is
802 * the pre-largefiles value of 2GB
803 */
806 else
811 }
814 }
824 }
825 /*
826 * Try to go direct
827 */
832 /*
833 * If ufs_directio wrote to the file or set the flags,
834 * we need to update i_seq, but it may be deferred.
835 */
840 }
843 }
845 /*
846 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
847 *
848 * o shadow inodes: vfs_dqrwlock is not held at all
849 * o quota updates: vfs_dqrwlock is read or write held
850 * o other updates: vfs_dqrwlock is read held
851 *
852 * The first case is the only one where we do not hold
853 * vfs_dqrwlock at all while entering wrip().
854 * We must make sure not to downgrade/drop vfs_dqrwlock if we
855 * have it as writer, i.e. if we are updating the quota inode.
856 * There is no potential deadlock scenario in this case as
857 * ufs_getpage() takes care of this and avoids reacquiring
858 * vfs_dqrwlock in that case.
859 *
860 * This check is done here since the above conditions do not change
861 * and we possibly loop below, so save a few cycles.
862 */
868 }
870 /*
871 * Large Files: We cast MAXBMASK to offset_t
872 * inorder to mask out the higher bits. Since offset_t
873 * is a signed value, the high order bit set in MAXBMASK
874 * value makes it do the right thing by having all bits 1
875 * in the higher word. May be removed for _SOLARIS64_.
876 */
891 }
892 /*
893 * since uoff + n >= limit,
894 * therefore n >= limit - uoff, and n is an int
895 * so it is safe to cast it to an int
896 */
898 }
900 /*
901 * We are extending the length of the file.
902 * bmap is used so that we are sure that
903 * if we need to allocate new blocks, that it
904 * is done here before we up the file size.
905 */
908 /*
909 * bmap_write never drops i_contents so if
910 * the flags are set it changed the file.
911 */
915 }
918 /*
919 * There is a window of vulnerability here.
920 * The sequence of operations: allocate file
921 * system blocks, uiomove the data into pages,
922 * and then update the size of the file in the
923 * inode, must happen atomically. However, due
924 * to current locking constraints, this can not
925 * be done.
926 */
930 /*
931 * If we are writing from the beginning of
932 * the mapping, we can just create the
933 * pages without having to read them.
934 */
937 /*
938 * Going to do a whole mappings worth,
939 * so we can just create the pages w/o
940 * having to read them in. But before
941 * we do that, we need to make sure any
942 * needed blocks are allocated first.
943 */
947 /*
948 * bmap_write never drops i_contents so if
949 * the flags are set it changed the file.
950 */
954 }
958 /*
959 * check if the new created page needed the
960 * allocation of new disk blocks.
961 */
966 /*
967 * In sync mode flush the indirect blocks which
968 * may have been allocated and not written on
969 * disk. In above cases bmap_write will allocate
970 * in sync mode.
971 */
976 }
977 }
979 /*
980 * At this point we can enter ufs_getpage() in one
981 * of two ways:
982 * 1) segmap_getmapflt() calls ufs_getpage() when the
983 * forcefault parameter is true (pagecreate == 0)
984 * 2) uiomove() causes a page fault.
985 *
986 * We have to drop the contents lock to prevent the VM
987 * system from trying to reacquire it in ufs_getpage()
988 * should the uiomove cause a pagefault.
989 *
990 * We have to drop the reader vfs_dqrwlock here as well.
991 */
997 }
1002 /*
1003 * Touch the page and fault it in if it is not in core
1004 * before segmap_getmapflt or vpm_data_copy can lock it.
1005 * This is to avoid the deadlock if the buffer is mapped
1006 * to the same file through mmap which we want to write.
1007 */
1011 /*
1012 * Copy data. If new pages are created, part of
1013 * the page that is not written will be initizliazed
1014 * with zeros.
1015 */
1023 /*
1024 * segmap_pagecreate() returns 1 if it calls
1025 * page_create_va() to allocate any pages.
1026 */
1033 }
1035 /*
1036 * If "newpage" is set, then a new page was created and it
1037 * does not contain valid data, so it needs to be initialized
1038 * at this point.
1039 * Otherwise the page contains old data, which was overwritten
1040 * partially or as a whole in uiomove.
1041 * If there is only one iovec structure within uio, then
1042 * on error uiomove will not be able to update uio->uio_loffset
1043 * and we would zero the whole page here!
1044 *
1045 * If uiomove fails because of an error, the old valid data
1046 * is kept instead of filling the rest of the page with zero's.
1047 */
1050 /*
1051 * We created pages w/o initializing them completely,
1052 * thus we need to zero the part that wasn't set up.
1053 * This happens on most EOF write cases and if
1054 * we had some sort of error during the uiomove.
1055 */
1063 }
1065 /*
1066 * Unlock the pages allocated by page_create_va()
1067 * in segmap_pagecreate()
1068 */
1072 /*
1073 * If the size of the file changed, then update the
1074 * size field in the inode now. This can't be done
1075 * before the call to segmap_pageunlock or there is
1076 * a potential deadlock with callers to ufs_putpage().
1077 * They will be holding i_contents and trying to lock
1078 * a page, while this thread is holding a page locked
1079 * and trying to acquire i_contents.
1080 */
1086 /*
1087 * file has grown larger than 2GB. Set flag
1088 * in superblock to indicate this, if it
1089 * is not already set.
1090 */
1098 }
1104 }
1107 /*
1108 * If we failed on a write, we may have already
1109 * allocated file blocks as well as pages. It's
1110 * hard to undo the block allocation, but we must
1111 * be sure to invalidate any pages that may have
1112 * been allocated.
1113 *
1114 * If the page was created without initialization
1115 * then we must check if it should be possible
1116 * to destroy the new page and to keep the old data
1117 * on the disk.
1118 *
1119 * It is possible to destroy the page without
1120 * having to write back its contents only when
1121 * - the size of the file keeps unchanged
1122 * - bmap_write() did not allocate new disk blocks
1123 * it is possible to create big files using "seek" and
1124 * write to the end of the file. A "write" to a
1125 * position before the end of the file would not
1126 * change the size of the file but it would allocate
1127 * new disk blocks.
1128 * - uiomove intended to overwrite the whole page.
1129 * - a new page was created (newpage == 1).
1130 */
1133 newpage) {
1135 /* unwind what uiomove eventually last did */
1138 /*
1139 * destroy the page, do not write ambiguous
1140 * data to the disk.
1141 */
1144 /*
1145 * write the page back to the disk, if dirty,
1146 * and remove the page from the cache.
1147 */
1149 }
1152 /*
1153 * Flush pages.
1154 */
1158 }
1161 /*
1162 * Force write back for synchronous write cases.
1163 */
1165 /*
1166 * If the sticky bit is set but the
1167 * execute bit is not set, we do a
1168 * synchronous write back and free
1169 * the page when done. We set up swap
1170 * files to be handled this way to
1171 * prevent servers from keeping around
1172 * the client's swap pages too long.
1173 * XXX - there ought to be a better way.
1174 */
1177 SM_DONTNEED;
1181 }
1183 /*
1184 * Have written a whole block.
1185 * Start an asynchronous write and
1186 * mark the buffer to indicate that
1187 * it won't be needed again soon.
1188 */
1190 }
1192 /*
1193 * Flush pages.
1194 */
1198 }
1199 /*
1200 * If the operation failed and is synchronous,
1201 * then we need to unwind what uiomove() last
1202 * did so we can potentially return an error to
1203 * the caller. If this write operation was
1204 * done in two pieces and the first succeeded,
1205 * then we won't return an error for the second
1206 * piece that failed. However, we only want to
1207 * return a resid value that reflects what was
1208 * really done.
1209 *
1210 * Failures for non-synchronous operations can
1211 * be ignored since the page subsystem will
1212 * retry the operation until it succeeds or the
1213 * file system is unmounted.
1214 */
1221 }
1222 }
1223 }
1225 /*
1226 * Re-acquire contents lock.
1227 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1228 */
1233 /*
1234 * If the uiomove() failed or if a synchronous
1235 * page push failed, fix up i_size.
1236 */
1239 /*
1240 * The uiomove failed, and we
1241 * allocated blocks,so get rid
1242 * of them.
1243 */
1245 }
1247 /*
1248 * XXX - Can this be out of the loop?
1249 */
1251 /*
1252 * Only do one increase of i_seq for multiple
1253 * pieces. Because we drop locks, record
1254 * the fact that we changed the timestamp and
1255 * are deferring the increase in case another thread
1256 * pushes our timestamp update.
1257 */
1267 /*
1268 * Clear Set-UID & Set-GID bits on
1269 * successful write if not privileged
1270 * and at least one of the execute bits
1271 * is set. If we always clear Set-GID,
1272 * mandatory file and record locking is
1273 * unuseable.
1274 */
1276 }
1277 }
1278 /*
1279 * In the case the FDSYNC flag is set and this is a
1280 * "rewrite" we won't log a delta.
1281 * The FSYNC flag overrides all cases.
1282 */
1285 }
1288 out:
1289 /*
1290 * Make sure i_seq is increased at least once per write
1291 */
1295 }
1297 /*
1298 * Inode is updated according to this table -
1299 *
1300 * FSYNC FDSYNC(posix.4)
1301 * --------------------------
1302 * always@ IATTCHG|IBDWRITE
1303 *
1304 * @ - If we are doing synchronous write the only time we should
1305 * not be sync'ing the ip here is if we have the stickyhack
1306 * activated, the file is marked with the sticky bit and
1307 * no exec bit, the file length has not been changed and
1308 * no new blocks have been allocated during this write.
1309 */
1312 /*
1313 * we have eliminated nosync
1314 */
1318 }
1319 }
1321 /*
1322 * If we've already done a partial-write, terminate
1323 * the write but return no error unless the error is ENOSPC
1324 * because the caller can detect this and free resources and
1325 * try again.
1326 */
1333 }
1335 /*
1336 * rdip does the real work of read requests for ufs.
1337 */
1338 int
1340 {
1341 uoff_t off;
1342 caddr_t base;
1350 uint_t flags;
1352 krw_t rwtype;
1353 o_mode_t type;
1367 /* check for valid filetype */
1372 }
1377 }
1380 }
1383 }
1390 }
1391 /*
1392 * Try to go direct
1393 */
1398 }
1403 offset_t diff;
1416 }
1420 /*
1421 * We update smallfile2 and smallfile1 at most every second.
1422 */
1435 }
1440 /*
1441 * At this point we can enter ufs_getpage() in one of two
1442 * ways:
1443 * 1) segmap_getmapflt() calls ufs_getpage() when the
1444 * forcefault parameter is true (value of 1 is passed)
1445 * 2) uiomove() causes a page fault.
1446 *
1447 * We cannot hold onto an i_contents reader lock without
1448 * risking deadlock in ufs_getpage() so drop a reader lock.
1449 * The ufs_getpage() dolock logic already allows for a
1450 * thread holding i_contents as writer to work properly
1451 * so we keep a writer lock.
1452 */
1457 /*
1458 * Copy data.
1459 */
1466 }
1470 /*
1471 * If reading sequential we won't need this
1472 * buffer again soon. For offsets in range
1473 * [smallfile1, smallfile2] release the pages
1474 * at the tail of the cache list, larger
1475 * offsets are released at the head.
1476 */
1482 }
1483 /*
1484 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1485 * we want to make sure that the page which has
1486 * been read, is written on disk if it is dirty.
1487 * And corresponding indirect blocks should also
1488 * be flushed out.
1489 */
1493 }
1498 }
1504 }
1505 }
1510 out:
1511 /*
1512 * Inode is updated according to this table if FRSYNC is set.
1513 *
1514 * FSYNC FDSYNC(posix.4)
1515 * --------------------------
1516 * always IATTCHG|IBDWRITE
1517 */
1518 /*
1519 * The inode is not updated if we're logging and the inode is a
1520 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1521 */
1525 }
1531 }
1532 }
1533 }
1534 /*
1535 * If we've already done a partial read, terminate
1536 * the read but return no error.
1537 */
1543 }
1545 /* ARGSUSED */
1546 static int
1555 {
1561 offset_t off;
1568 /*
1569 * forcibly unmounted
1570 */
1584 }
1592 }
1594 }
1598 /*
1599 * file system locking
1600 */
1608 }
1609 #ifdef _SYSCALL32_IMPL
1612 /* Translate ILP32 lockfs to LP64 lockfs */
1622 }
1628 comment =
1634 }
1637 }
1642 DATAMODEL_NATIVE) {
1645 }
1646 #ifdef _SYSCALL32_IMPL
1649 /* Translate LP64 to ILP32 lockfs */
1663 }
1669 }
1673 /*
1674 * get file system locking status
1675 */
1681 }
1682 #ifdef _SYSCALL32_IMPL
1685 /* Translate ILP32 lockfs to LP64 lockfs */
1695 }
1710 }
1711 #ifdef _SYSCALL32_IMPL
1713 /* Translate LP64 to ILP32 lockfs */
1724 }
1735 /*
1736 * set access time
1737 */
1739 /*
1740 * if mounted w/o atime, return quietly.
1741 * I briefly thought about returning ENOSYS, but
1742 * figured that most apps would consider this fatal
1743 * but the idea is to make this as seamless as poss.
1744 */
1749 ULOCKFS_SETATTR_MASK);
1757 }
1766 }
1770 /*
1771 * set delayed-io
1772 */
1776 /*
1777 * get delayed-io
1778 */
1782 /*
1783 * inode open
1784 */
1786 ULOCKFS_VGET_MASK);
1794 }
1798 /*
1799 * file system flush (push w/invalidate)
1800 */
1806 /*
1807 * Contract-private interface for Legato
1808 * Purge this vnode from the DNLC and decide
1809 * if this vnode is busy (*arg == 1) or not
1810 * (*arg == 0)
1811 */
1821 /*
1822 * Tune the file system (aka setting fs attributes)
1823 */
1825 ULOCKFS_SETATTR_MASK);
1849 {
1864 }
1866 /*
1867 * Do copyout even if there is an error because
1868 * the details of error is stored in fcp.
1869 */
1874 }
1877 {
1886 }
1898 /*
1899 * The following 3 ioctls are for TSufs support
1900 * although could potentially be used elsewhere
1901 */
1915 {
1921 /* Copy stucture if statistics are being kept */
1924 }
1930 }
1936 /* offset paramater is in/out */
1945 {
1946 /*
1947 * This is a project private ufs ioctl() to mark
1948 * the inode as that belonging to a compressed
1949 * file. This is used to mark individual
1950 * compressed files in a miniroot archive.
1951 * The files compressed in this manner are
1952 * automatically decompressed by the dcfs filesystem
1953 * (via an interception in ufs_lookup - see decompvp())
1954 * which is layered on top of ufs on a system running
1955 * from the archive. See uts/common/fs/dcfs for details.
1956 * This ioctl only marks the file as compressed - the
1957 * actual compression is done by fiocompress (a
1958 * userland utility) which invokes this ioctl().
1959 */
1963 ULOCKFS_SETATTR_MASK);
1970 }
1978 }
1982 }
1986 }
1987 }
1990 /* ARGSUSED */
1991 static int
1994 {
2000 /*
2001 * for performance, if only the size is requested don't bother
2002 * with anything else.
2003 */
2006 }
2008 /*
2009 * inlined lockfs checks
2010 */
2015 }
2018 /*
2019 * Return all the attributes. This should be refined so
2020 * that it only returns what's asked for.
2021 */
2023 /*
2024 * Copy from inode table.
2025 */
2028 /*
2029 * If there is an ACL and there is a mask entry, then do the
2030 * extra work that completes the equivalent of an acltomode(3)
2031 * call. According to POSIX P1003.1e, the acl mask should be
2032 * returned in the group permissions field.
2033 *
2034 * - start with the original permission and mode bits (from above)
2035 * - clear the group owner bits
2036 * - add in the mask bits.
2037 */
2042 }
2051 else
2077 }
2082 out:
2084 }
2086 /*
2087 * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2088 * The i_contents lock is already held by the caller and we need to
2089 * declare the inode as 'void *' argument.
2090 */
2091 static int
2093 {
2097 }
2099 /*ARGSUSED4*/
2100 static int
2103 {
2120 timestruc_t now;
2121 vattr_t oldva;
2125 /*
2126 * Cannot set these attributes.
2127 */
2131 /*
2132 * check for forced unmount
2133 */
2141 again:
2152 /*
2153 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2154 * This follows the protocol for read()/write().
2155 */
2157 /*
2158 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2159 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2160 * possible, retries the operation.
2161 */
2167 }
2169 }
2171 /*
2172 * Truncate file. Must have write permission and not be a directory.
2173 */
2179 }
2188 }
2192 }
2198 }
2200 /*
2201 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2202 * This follows the protocol established by
2203 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2204 */
2209 trans_size);
2213 }
2215 /*
2216 * Grab quota lock if we are changing the file's owner.
2217 */
2221 }
2237 /*
2238 * Change file access modes.
2239 */
2248 else
2251 }
2252 }
2255 /*
2256 * Don't change ownership of the quota inode.
2257 */
2262 }
2264 /*
2265 * No real ownership change.
2266 */
2270 }
2271 /*
2272 * Remove the blocks and the file, from the old user's
2273 * quota.
2274 */
2285 }
2289 /*
2290 * There is a real ownership change.
2291 */
2293 /*
2294 * Add the blocks and the file to the new
2295 * user's quota.
2296 */
2303 }
2304 }
2307 }
2310 }
2311 /*
2312 * Change file access or modified times.
2313 */
2315 /* Check that the time value is within ufs range */
2320 }
2322 /*
2323 * if the "noaccess" mount option is set and only atime
2324 * update is requested, do nothing. No error is returned.
2325 */
2334 }
2340 /*
2341 * In 2038, ctime sticks forever..
2342 */
2348 }
2351 }
2354 }
2356 skip_atime:
2357 /*
2358 * The presence of a shadow inode may indicate an ACL, but does
2359 * not imply an ACL. Future FSD types should be handled here too
2360 * and check for the presence of the attribute-specific data
2361 * before referencing it.
2362 */
2364 /*
2365 * XXX if ufs_iupdat is changed to sandbagged write fix
2366 * ufs_acl_setattr to push ip to keep acls consistent
2367 *
2368 * Suppress out of inodes messages if we will retry.
2369 */
2374 }
2376 update_inode:
2377 /*
2378 * Setattr always increases the sequence number
2379 */
2382 /*
2383 * if nfsd and not logging; push synchronously
2384 */
2389 }
2394 }
2402 trans_size);
2405 }
2407 }
2408 out:
2409 /*
2410 * If out of inodes or blocks, see if we can free something
2411 * up from the delete queue.
2412 */
2421 }
2425 }
2429 }
2431 }
2433 /*ARGSUSED*/
2434 static int
2437 {
2443 /*
2444 * The ufs_iaccess function wants to be called with
2445 * mode bits expressed as "ufs specific" bits.
2446 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2447 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2448 * But since they're the same we just pass the vnode mode
2449 * bit but just verify that assumption at compile time.
2450 */
2451 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2453 #endif
2455 }
2457 /* ARGSUSED */
2458 static int
2461 {
2471 }
2473 /*
2474 * If the symbolic link is empty there is nothing to read.
2475 * Fast-track these empty symbolic links
2476 */
2480 }
2486 /*
2487 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2488 */
2489 again:
2501 }
2507 }
2510 }
2514 ino_t ino;
2534 }
2536 /* can this be a fast symlink and is it a user buffer? */
2542 /*
2543 * setup a kernel buffer to read link into. this
2544 * is to fix a race condition where the user buffer
2545 * got corrupted before copying it into the inode.
2546 */
2562 }
2569 }
2580 /*
2581 * free page
2582 */
2589 /* error, clear garbage left behind */
2594 }
2595 }
2597 /* now, copy it into the user buffer */
2601 }
2604 }
2605 out:
2608 }
2609 nolockout:
2611 }
2613 /* ARGSUSED */
2614 static int
2616 {
2627 /*
2628 * First push out any data pages
2629 */
2636 }
2638 /*
2639 * Delta any delayed inode times updates
2640 * and push inode to log.
2641 * All other inode deltas will have already been delta'd
2642 * and will be pushed during the commit.
2643 */
2648 TOP_SYNCIP_SIZE);
2649 }
2658 TOP_SYNCIP_SIZE);
2659 }
2660 }
2662 /*
2663 * Commit the Moby transaction
2664 *
2665 * Deltas have already been made so we just need to
2666 * commit them with a synchronous transaction.
2667 * TRANS_BEGIN_SYNC() will return an error
2668 * if there are no deltas to commit, for an
2669 * empty transaction.
2670 */
2677 }
2679 TOP_COMMIT_SIZE);
2680 }
2684 /* Just update the inode only */
2688 /* Do data-synchronous writes */
2690 else
2691 /* Do synchronous writes */
2698 }
2699 out:
2702 }
2704 }
2706 /*ARGSUSED*/
2707 static void
2709 {
2711 }
2713 /*
2714 * Unix file system operations having to do with directory manipulation.
2715 */
2717 /* ARGSUSED */
2718 static int
2722 {
2731 /*
2732 * Check flags for type of lookup (regular file or attribute file)
2733 */
2739 /*
2740 * If not mounted with XATTR support then return EINVAL
2741 */
2745 /*
2746 * We don't allow recursive attributes...
2747 * Maybe someday we will.
2748 */
2751 }
2758 }
2762 }
2764 /*
2765 * Check accessibility of directory.
2766 */
2771 }
2775 }
2779 }
2781 /*
2782 * Check for a null component, which we should treat as
2783 * looking at dvp from within it's parent, so we don't
2784 * need a call to ufs_iaccess(), as it has already been
2785 * done.
2786 */
2792 }
2794 /*
2795 * Check for "." ie itself. this is a quick check and
2796 * avoids adding "." into the dnlc (which have been seen
2797 * to occupy >10% of the cache).
2798 */
2800 /*
2801 * Don't return without checking accessibility
2802 * of the directory. We only need the lock if
2803 * we are going to return it.
2804 */
2808 }
2810 }
2812 /*
2813 * Fast path: Check the directory name lookup cache.
2814 */
2816 /*
2817 * Check accessibility of directory.
2818 */
2822 }
2827 }
2831 }
2833 /*
2834 * Keep the idle queue from getting too long by
2835 * idling two inodes before attempting to allocate another.
2836 * This operation must be performed before entering
2837 * lockfs or a transaction.
2838 */
2843 }
2845 retry_lookup:
2846 /*
2847 * Check accessibility of directory.
2848 */
2859 fastpath:
2864 /*
2865 * If vnode is a device return special vnode instead.
2866 */
2871 cr);
2875 else
2880 /*
2881 * Compressed file, substitute dcfs vnode
2882 */
2887 else
2889 }
2890 }
2893 }
2898 out:
2900 }
2902 /*ARGSUSED*/
2903 static int
2907 {
2923 again:
2935 }
2941 /*
2942 * Null component name refers to the directory itself.
2943 */
2945 /*
2946 * Even though this is an error case, we need to grab the
2947 * quota lock since the error handling code below is common.
2948 */
2955 /*
2956 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2957 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2958 * possible, retries the operation.
2959 */
2963 trans_size);
2972 }
2980 }
2982 /*
2983 * Suppress file system full message if we will retry
2984 */
2991 trans_size);
2993 }
2995 }
2997 }
3002 }
3003 }
3005 /*
3006 * If the file already exists and this is a non-exclusive create,
3007 * check permissions and allow access for non-directories.
3008 * Read-only create of an existing directory is also allowed.
3009 * We fail an exclusive create of anything which already exists.
3010 */
3020 else
3022 }
3028 }
3029 /*
3030 * If the error EEXIST was set, then i_seq can not
3031 * have been updated. The sequence number interface
3032 * is defined such that a non-error fop_create must
3033 * increase the dir va_seq it by at least one. If we
3034 * have cleared the error, increase i_seq. Note that
3035 * we are increasing the dir i_seq and in rare cases
3036 * ip may actually be from the dvp, so we already have
3037 * the locks and it will not be subject to truncation.
3038 * In case we have to update i_seq of the parent
3039 * directory dip, we have to defer it till we have
3040 * released our locks on ip due to lock ordering requirements.
3041 */
3044 else
3049 /*
3050 * Truncate regular files, if requested by caller.
3051 * Grab i_rwlock to make sure no one else is
3052 * currently writing to the file (we promised
3053 * bmap we would do this).
3054 * Must get the locks in the correct order.
3055 */
3061 /*
3062 * Large Files: Why this check here?
3063 * Though we do it in vn_create() we really
3064 * want to guarantee that we do not destroy
3065 * Large file data by atomically checking
3066 * the size while holding the contents
3067 * lock.
3068 */
3076 }
3078 truncflag++;
3084 RW_WRITER,
3085 TOP_CREATE,
3086 ufsvfsp,
3088 issync,
3089 trans_size);
3093 }
3095 RW_READER);
3098 cr);
3100 }
3102 }
3105 }
3106 }
3107 }
3113 }
3115 }
3122 /*
3123 * If vnode is a device return special vnode instead.
3124 */
3133 }
3136 }
3137 unlock:
3139 /*
3140 * Do the deferred update of the parent directory's sequence
3141 * number now.
3142 */
3149 }
3155 trans_size);
3157 /*
3158 * If we haven't had a more interesting failure
3159 * already, then anything that might've happened
3160 * here should be reported.
3161 */
3164 }
3173 }
3176 }
3181 /*
3182 * If no inodes available, try to free one up out of the
3183 * pending delete queue.
3184 */
3189 }
3191 out:
3193 }
3196 /*ARGSUSED*/
3197 static int
3200 {
3210 /*
3211 * don't let the delete queue get too long
3212 */
3216 }
3222 /* Only send the event if there were no errors */
3226 }
3228 retry_remove:
3237 /*
3238 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3239 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3240 * possible, retries the operation.
3241 */
3252 trans_size);
3254 }
3256 out:
3258 }
3260 /*
3261 * Link a file or a directory. Only privileged processes are allowed to
3262 * make links to directories.
3263 */
3264 /*ARGSUSED*/
3265 static int
3268 {
3280 retry_link:
3292 /*
3293 * Make sure link for extended attributes is valid
3294 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3295 *
3296 * Make certain we don't attempt to look at a device node as
3297 * a ufs inode.
3298 */
3307 }
3315 }
3317 /*
3318 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3319 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3320 * possible, retries the operation.
3321 */
3324 trans_size);
3330 unlock:
3334 }
3338 }
3339 out:
3341 }
3348 /*
3349 * Rename a file or directory.
3350 * We are given the vnode and entry string of the source and the
3351 * vnode and entry string of the place we want to move the source
3352 * to (the target). The essential operation is:
3353 * unlink(target);
3354 * link(source, target);
3355 * unlink(source);
3356 * but "atomically". Can't do full commit without saving state in
3357 * the inode on disk, which isn't feasible at this time. Best we
3358 * can do is always guarantee that the TARGET exists.
3359 */
3361 /*ARGSUSED*/
3362 static int
3365 {
3376 timestruc_t now;
3392 /* Must do this before taking locks in case of DNLC miss */
3400 /*
3401 * Notify the target directory of the rename event
3402 * if source and target directories are not the same.
3403 */
3409 }
3414 retry_rename:
3428 /*
3429 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3430 */
3434 }
3436 /*
3437 * Check accessibility of directory.
3438 */
3442 /*
3443 * Look up inode of file we're supposed to rename.
3444 */
3452 }
3454 }
3457 }
3459 /*
3460 * Lock both the source and target directories (they may be
3461 * the same) to provide the atomicity semantics that was
3462 * previously provided by the per file system vfs_rename_lock
3463 *
3464 * with vfs_rename_lock removed to allow simultaneous renames
3465 * within a file system, ufs_dircheckpath can deadlock while
3466 * traversing back to ensure that source is not a parent directory
3467 * of target parent directory. This is because we get into
3468 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3469 * If the tdp and sdp of the simultaneous renames happen to be
3470 * in the path of each other, it can lead to a deadlock. This
3471 * can be avoided by getting the locks as RW_READER here and then
3472 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3473 *
3474 * We hold the target directory's i_rwlock after calling
3475 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3476 * fop_rwlock is explicitly called by the filesystem independent code
3477 * before calling the file system operation. In these cases the order
3478 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3479 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3480 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3481 * synchronizing object which might lead to a deadlock. So we use
3482 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3483 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3484 * operation.
3485 */
3486 retry:
3489 retry_firstlock:
3491 /*
3492 * We didn't get the lock. Check if the SLOCK is set in the
3493 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3494 * and wait for SLOCK to be cleared.
3495 */
3499 trans_size);
3504 /*
3505 * SLOCK isn't set so this is a genuine synchronization
3506 * case. Let's try again after giving them a breather.
3507 */
3510 }
3511 }
3512 /*
3513 * Need to check if the tdp and sdp are same !!!
3514 */
3516 /*
3517 * We didn't get the lock. Check if the SLOCK is set in the
3518 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3519 * and wait for SLOCK to be cleared.
3520 */
3525 trans_size);
3530 /*
3531 * So we couldn't get the second level peer lock *and*
3532 * the SLOCK bit isn't set. Too bad we can be
3533 * contentding with someone wanting these locks otherway
3534 * round. Reverse the locks in case there is a heavy
3535 * contention for the second level lock.
3536 */
3540 ufs_rename_retry_cnt++;
3542 }
3543 }
3548 }
3549 /*
3550 * Make sure we can delete the source entry. This requires
3551 * write permission on the containing directory.
3552 * Check for sticky directories.
3553 */
3561 }
3563 /*
3564 * If this is a rename of a directory and the parent is
3565 * different (".." must be changed), then the source
3566 * directory must not be in the directory hierarchy
3567 * above the target, as this would orphan everything
3568 * below the source directory. Also the user must have
3569 * write permission in the source so as to be able to
3570 * change "..".
3571 */
3574 ino_t inum;
3580 }
3585 /*
3586 * If we got EAGAIN ufs_dircheckpath detected a
3587 * potential deadlock and backed out. We need
3588 * to retry the operation since sdp and tdp have
3589 * to be released to avoid the deadlock.
3590 */
3596 ufs_rename_dircheck_retry_cnt++;
3598 }
3600 }
3604 }
3607 /*
3608 * Check for renaming '.' or '..' or alias of '.'
3609 */
3613 }
3615 /*
3616 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3617 * tries to traverse back the file tree with both tdp and sdp held
3618 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3619 * as RW_READERS till ufs_dircheckpath is done.
3620 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3621 * to RW_WRITER.
3622 */
3624 /*
3625 * The upgrade failed. We got to give away the lock
3626 * as to avoid deadlocking with someone else who is
3627 * waiting for writer lock. With the lock gone, we
3628 * cannot be sure the checks done above will hold
3629 * good when we eventually get them back as writer.
3630 * So if we can't upgrade we drop the locks and retry
3631 * everything again.
3632 */
3637 ufs_rename_upgrade_retry_cnt++;
3639 }
3642 /*
3643 * The upgrade failed. We got to give away the lock
3644 * as to avoid deadlocking with someone else who is
3645 * waiting for writer lock. With the lock gone, we
3646 * cannot be sure the checks done above will hold
3647 * good when we eventually get them back as writer.
3648 * So if we can't upgrade we drop the locks and retry
3649 * everything again.
3650 */
3654 ufs_rename_upgrade_retry_cnt++;
3656 }
3657 }
3659 /*
3660 * Now that all the locks are held check to make sure another thread
3661 * didn't slip in and take out the sip.
3662 */
3674 }
3679 /*
3680 * If the inode was found need to drop the v_count
3681 * so as not to keep the filesystem from being
3682 * unmounted at a later time.
3683 */
3685 }
3687 /*
3688 * Release the slot.fbp that has the page mapped and
3689 * locked SE_SHARED, and could be used in in
3690 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3691 * on said page.
3692 */
3696 }
3697 }
3699 /*
3700 * Link source to the target.
3701 */
3703 /*
3704 * ESAME isn't really an error; it indicates that the
3705 * operation should not be done because the source and target
3706 * are the same file, but that no error should be reported.
3707 */
3711 }
3716 /*
3717 * Unlink the source.
3718 * Remove the source entry. ufs_dirremove() checks that the entry
3719 * still reflects sip, and returns an error if it doesn't.
3720 * If the entry has changed just forget about it. Release
3721 * the source inode.
3722 */
3729 /*
3730 * Notify the target directory of the rename event
3731 * if source and target directories are not the same.
3732 */
3735 }
3737 errout:
3744 }
3746 unlock:
3754 trans_size);
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 */
3802 trans_size);
3811 trans_size);
3813 }
3815 }
3830 }
3831 out:
3836 }
3839 }
3841 /*ARGSUSED*/
3842 static int
3845 {
3855 /*
3856 * don't let the delete queue get too long
3857 */
3861 }
3867 /* Only send the event if there were no errors */
3871 }
3873 retry_rmdir:
3882 /*
3883 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3884 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3885 * possible, retries the operation.
3886 */
3889 trans_size);
3898 trans_size);
3900 }
3902 out:
3904 }
3906 /* ARGSUSED */
3907 static int
3910 {
3918 caddr_t outbuf;
3920 uint_t offset;
3933 }
3935 /*
3936 * Check if we have been called with a valid iov_len
3937 * and bail out if not, otherwise we may potentially loop
3938 * forever further down.
3939 */
3943 }
3945 /*
3946 * Large Files: When we come here we are guaranteed that
3947 * uio_offset can be used safely. The high word is zero.
3948 */
3958 /* Large Files: directory files should not be "large" */
3962 /* Force offset to be valid (to guard against bogus lseek() values) */
3965 /* Quit if at end of file or link count of zero (posix) */
3971 }
3973 /*
3974 * Get space to change directory entries into fs independent format.
3975 * Do fast alloc for the most commonly used-request size (filesystem
3976 * block size).
3977 */
3985 }
3987 nextblk:
3990 /* Truncate request to file size */
3994 /* Comply with MAXBSIZE boundary restrictions of fbread() */
3998 /*
3999 * Read in the next chunk.
4000 * We are still holding the i_rwlock.
4001 */
4009 }
4019 }
4020 /* Transform to file-system independent format */
4022 /*
4023 * If the current directory entry is mangled, then skip
4024 * to the next block. It would be nice to set the FSBAD
4025 * flag in the super-block so that a fsck is forced on
4026 * next reboot, but locking is a problem.
4027 */
4031 }
4033 /* Skip to requested offset and skip empty entries */
4035 ushort_t this_reclen =
4037 /* Buffer too small for any entries */
4042 }
4043 /* If would overrun the buffer, quit */
4046 }
4047 /* Take this entry */
4052 /* use strncpy(9f) to zero out uninitialized bytes */
4062 }
4070 }
4071 }
4072 /* Release the chunk */
4075 /* Read whole block, but got no entries, read another if not eof */
4077 /*
4078 * Large Files: casting i_size to int here is not a problem
4079 * because directory sizes are always less than MAXOFF32_T.
4080 * See assertion above.
4081 */
4086 /* Copy out the entry data */
4095 update_inode:
4102 unlock:
4105 }
4106 out:
4108 }
4110 /*ARGSUSED*/
4111 static int
4114 {
4125 /*
4126 * No symlinks in attrdirs at this time
4127 */
4131 again:
4144 /*
4145 * We must create the inode before the directory entry, to avoid
4146 * racing with readlink(). ufs_dirmakeinode requires that we
4147 * hold the quota lock as reader, and directory locks as writer.
4148 */
4154 /*
4155 * Suppress any out of inodes messages if we will retry on
4156 * ENOSP
4157 */
4172 /*
4173 * OK. The inode has been created. Write out the data of the
4174 * symbolic link. Since symbolic links are metadata, and should
4175 * remain consistent across a system crash, we need to force the
4176 * data out synchronously.
4177 *
4178 * (This is a change from the semantics in earlier releases, which
4179 * only created symbolic links synchronously if the semi-documented
4180 * 'syncdir' option was set, or if we were being invoked by the NFS
4181 * server, which requires symbolic links to be created synchronously.)
4182 *
4183 * We need to pass in a pointer for the residual length; otherwise
4184 * ufs_rdwri() will always return EIO if it can't write the data,
4185 * even if the error was really ENOSPC or EDQUOT.
4186 */
4194 /*
4195 * Suppress file system full messages if we will retry
4196 */
4209 }
4211 /*
4212 * If the link's data is small enough, we can cache it in the inode.
4213 * This is a "fast symbolic link". We don't use the first direct
4214 * block because that's actually used to point at the symbolic link's
4215 * contents on disk; but we know that none of the other direct or
4216 * indirect blocks can be used because symbolic links are restricted
4217 * to be smaller than a file system block.
4218 */
4227 /* error, clear garbage left behind */
4232 }
4233 }
4238 /*
4239 * OK. We've successfully created the symbolic link. All that
4240 * remains is to insert it into the appropriate directory.
4241 */
4247 /*
4248 * Fall through into remove-on-error code. We're either done, or we
4249 * need to remove the inode (if we couldn't insert it).
4250 */
4252 remove:
4260 }
4262 unlock:
4270 trans_size);
4274 }
4276 /*
4277 * We may have failed due to lack of an inode or of a block to
4278 * store the target in. Try flushing the delete queue to free
4279 * logically-available things up and try again.
4280 */
4285 }
4287 out:
4289 }
4291 /*
4292 * Ufs specific routine used to do ufs io.
4293 */
4294 int
4298 {
4326 }
4332 }
4334 }
4336 /*ARGSUSED*/
4337 static int
4339 {
4349 }
4358 }
4360 /* ARGSUSED2 */
4361 static int
4363 {
4368 /*
4369 * Read case is easy.
4370 */
4374 }
4376 /*
4377 * Caller has requested a writer lock, but that inhibits any
4378 * concurrency in the VOPs that follow. Acquire the lock shared
4379 * and defer exclusive access until it is known to be needed in
4380 * other VOP handlers. Some cases can be determined here.
4381 */
4383 /*
4384 * If directio is not set, there is no chance of concurrency,
4385 * so just acquire the lock exclusive. Beware of a forced
4386 * unmount before looking at the mount option.
4387 */
4394 }
4396 /*
4397 * Mandatory locking forces acquiring i_rwlock exclusive.
4398 */
4402 }
4404 /*
4405 * Acquire the lock shared in case a concurrent write follows.
4406 * Mandatory locking could have become enabled before the lock
4407 * was acquired. Re-check and upgrade if needed.
4408 */
4414 }
4416 }
4418 /*ARGSUSED*/
4419 static void
4421 {
4425 }
4427 /* ARGSUSED */
4428 static int
4430 {
4432 }
4434 /* ARGSUSED */
4435 static int
4439 {
4445 /*
4446 * If file is being mapped, disallow frlock.
4447 * XXX I am not holding tlock while checking i_mapcnt because the
4448 * current locking strategy drops all locks before calling fs_frlock.
4449 * So, mapcnt could change before we enter fs_frlock making is
4450 * meaningless to have held tlock in the first place.
4451 */
4455 }
4457 /* ARGSUSED */
4458 static int
4461 {
4469 ULOCKFS_SPACE_MASK);
4478 ULOCKFS_FALLOCATE_MASK);
4488 }
4490 }
4492 /*
4493 * Used to determine if read ahead should be done. Also used to
4494 * to determine when write back occurs.
4495 */
4496 #define CLUSTSZ(ip) ((ip)->i_ufsvfs->vfs_ioclustsz)
4498 /*
4499 * A faster version of ufs_getpage.
4500 *
4501 * We optimize by inlining the pvn_getpages iterator, eliminating
4502 * calls to bmap_read if file doesn't have UFS holes, and avoiding
4503 * the overhead of page_exists().
4504 *
4505 * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4506 * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4507 * victimizes performance when a file with UFS holes is faulted
4508 * first in the S_READ mode, and then in the S_WRITE mode. We will get
4509 * two MMU faults in this case.
4510 *
4511 * XXX - the inode fields which control the sequential mode are not
4512 * protected by any mutex. The read ahead will act wild if
4513 * multiple processes will access the file concurrently and
4514 * some of them in sequential mode. One particulary bad case
4515 * is if another thread will change the value of i_nextrio between
4516 * the time this thread tests the i_nextrio value and then reads it
4517 * again to use it as the offset for the read ahead.
4518 */
4519 /*ARGSUSED*/
4520 static int
4524 {
4526 uoff_t pgoff;
4527 uoff_t eoff;
4533 caddr_t pgaddr;
4534 krw_t rwtype;
4549 /*
4550 * Obey the lockfs protocol
4551 */
4561 /*
4562 * Try to start a transaction, will return if blocking is
4563 * expected to occur and the address space is not the
4564 * kernel address space.
4565 */
4571 /*
4572 * Use EDEADLK here because the VM code
4573 * can normally never see this error.
4574 */
4578 }
4581 }
4582 }
4587 }
4593 /*
4594 * If this thread owns the lock, i.e., this thread grabbed it
4595 * as writer somewhere above, then we don't need to grab the
4596 * lock as reader in this routine.
4597 */
4600 retrylock:
4602 /*
4603 * Grab the quota lock if we need to call
4604 * bmap_write() below (with i_contents as writer).
4605 */
4609 }
4611 /*
4612 * We may be getting called as a side effect of a bmap using
4613 * fbread() when the blocks might be being allocated and the
4614 * size has not yet been up'ed. In this case we want to be
4615 * able to return zero pages if we get back UFS_HOLE from
4616 * calling bmap for a non write case here. We also might have
4617 * to read some frags from the disk into a page if we are
4618 * extending the number of frags for a given lbn in bmap().
4619 * Large Files: The read of i_size here is atomic because
4620 * i_contents is held here. If dolock is zero, the lock
4621 * is held in bmap routines.
4622 */
4630 }
4633 }
4635 /*
4636 * Must hold i_contents lock throughout the call to pvn_getpages
4637 * since locked pages are returned from each call to ufs_getapage.
4638 * Must *not* return locked pages and then try for contents lock
4639 * due to lock ordering requirements (inode > page)
4640 */
4646 uoff_t offset;
4648 /*
4649 * We must acquire the RW_WRITER lock in order to
4650 * call bmap_write().
4651 */
4655 /*
4656 * Grab the quota lock before
4657 * upgrading i_contents, but if we can't grab it
4658 * don't wait here due to lock order:
4659 * vfs_dqrwlock > i_contents.
4660 */
4666 }
4672 }
4673 }
4675 /*
4676 * May be allocating disk blocks for holes here as
4677 * a result of mmap faults. write(2) does the bmap_write
4678 * in rdip/wrip, not here. We are not dealing with frags
4679 * in this case.
4680 */
4681 /*
4682 * Large Files: We cast fs_bmask field to offset_t
4683 * just as we do for MAXBMASK because uoff is a 64-bit
4684 * data type. fs_bmask will still be a 32-bit type
4685 * as we cannot change any ondisk data structures.
4686 */
4698 }
4699 }
4701 /*
4702 * Can be a reader from now on.
4703 */
4706 /*
4707 * We can release vfs_dqrwlock early so do it, but make
4708 * sure we don't try to release it again at the bottom.
4709 */
4713 }
4714 }
4716 /*
4717 * We remove PROT_WRITE in cases when the file has UFS holes
4718 * because we don't want to call bmap_read() to check each
4719 * page if it is backed with a disk block.
4720 */
4726 /*
4727 * The loop looks up pages in the range [off, off + len).
4728 * For each page, we first check if we should initiate an asynchronous
4729 * read ahead before we call page_lookup (we may sleep in page_lookup
4730 * for a previously initiated disk read).
4731 */
4736 uoff_t nextrio;
4737 se_t se;
4742 /* Handle async getpage (faultahead) */
4749 }
4750 /*
4751 * Check if we should initiate read ahead of next cluster.
4752 * We call page_exists only when we need to confirm that
4753 * we have the current page before we initiate the read ahead.
4754 */
4760 /*
4761 * We always read ahead the next cluster of data
4762 * starting from i_nextrio. If the page (vp,nextrio)
4763 * is actually in core at this point, the routine
4764 * ufs_getpage_ra() will stop pre-fetching data
4765 * until we read that page in a synchronized manner
4766 * through ufs_getpage_miss(). So, we should increase
4767 * i_nextrio if the page (vp, nextrio) exists.
4768 */
4771 }
4772 }
4775 /*
4776 * We found the page in the page cache.
4777 */
4784 /*
4785 * We have to create the page, or read it from disk.
4786 */
4792 pl++;
4797 }
4798 }
4799 }
4801 /*
4802 * Return pages up to plsz if they are in the page cache.
4803 * We cannot return pages if there is a chance that they are
4804 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4805 */
4821 }
4822 }
4828 error:
4830 /*
4831 * Release any pages we have locked.
4832 */
4837 }
4839 update_inode:
4840 /*
4841 * If the inode is not already marked for IACC (in rdip() for read)
4842 * and the inode is not marked for no access time update (in wrip()
4843 * for write) then update the inode access time and mod time now.
4844 */
4854 }
4855 }
4856 }
4862 }
4864 unlock:
4869 }
4871 }
4872 out:
4874 }
4876 /*
4877 * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4878 * cache. The page is either read from the disk, or it's created.
4879 * A page is created (without disk read) if rw == S_CREATE, or if
4880 * the page is not backed with a real disk block (UFS hole).
4881 */
4882 /* ARGSUSED */
4883 static int
4886 {
4889 daddr_t bn;
4896 /*
4897 * Figure out whether the page can be created, or must be
4898 * must be read from the disk.
4899 */
4909 /*
4910 * If its also a fallocated block that hasn't been written to
4911 * yet, we will treat it just like a UFS_HOLE and create
4912 * a zero page for it
4913 */
4916 }
4923 }
4930 uoff_t io_off;
4931 uint_t xlen;
4935 /*
4936 * If access is not in sequential order, we read from disk
4937 * in bsize units.
4938 *
4939 * We limit the size of the transfer to bsize if we are reading
4940 * from the beginning of the file. Note in this situation we
4941 * will hedge our bets and initiate an async read ahead of
4942 * the second block.
4943 */
4950 /*
4951 * Some other thread has entered the page.
4952 * ufs_getpage will retry page_lookup.
4953 */
4957 }
4959 /*
4960 * Zero part of the page which we are not
4961 * going to read from the disk.
4962 */
4984 }
4988 /*
4989 * If the file access is sequential, initiate read ahead
4990 * of the next cluster.
4991 */
5000 }
5001 }
5005 }
5007 /*
5008 * Read ahead a cluster from the disk. Returns the length in bytes.
5009 */
5010 static int
5012 {
5019 daddr_t bn;
5026 /*
5027 * If the directio advisory is in effect on this file,
5028 * then do not do buffered read ahead. Read ahead makes
5029 * it more difficult on threads using directio as they
5030 * will be forced to flush the pages from this vnode.
5031 */
5037 /*
5038 * Is this test needed?
5039 */
5045 /*
5046 * If its a UFS_HOLE or a fallocated block, do not perform
5047 * any read ahead's since there probably is nothing to read ahead
5048 */
5052 /*
5053 * Limit the transfer size to bsize if this is the 2nd block.
5054 */
5062 /*
5063 * Zero part of page which we are not going to read from disk
5064 */
5087 }
5090 }
5093 /*
5094 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5095 *
5096 * LMXXX - the inode really ought to contain a pointer to one of these
5097 * async args. Stuff gunk in there and just hand the whole mess off.
5098 * This would replace i_delaylen, i_delayoff.
5099 */
5100 /*ARGSUSED*/
5101 static int
5104 {
5110 }
5112 /*
5113 * XXX - Why should this check be made here?
5114 */
5118 }
5123 }
5129 /*
5130 * If nobody stalled, start a new cluster.
5131 */
5137 }
5138 /*
5139 * If we have a full cluster or they are not contig,
5140 * then push last cluster and start over.
5141 */
5144 uoff_t doff;
5154 /* LMXXX - flags are new val, not old */
5156 }
5157 /*
5158 * There is something there, it's not full, and
5159 * it is contig.
5160 */
5164 }
5165 /*
5166 * Must have weird flags or we are not clustering.
5167 */
5168 }
5172 errout:
5174 }
5176 /*
5177 * If len == 0, do from off to EOF.
5178 *
5179 * The normal cases should be len == 0 & off == 0 (entire vp list),
5180 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5181 * (from pageout).
5182 */
5183 /*ARGSUSED*/
5184 static int
5187 {
5188 uoff_t io_off;
5189 uoff_t eoff;
5198 /*
5199 * Acquire the readers/write inode lock before locking
5200 * any pages in this inode.
5201 * The inode lock is held during i/o.
5202 */
5207 }
5210 /*
5211 * Must synchronize this thread and any possible thread
5212 * operating in the window of vulnerability in wrip().
5213 * It is dangerous to allow both a thread doing a putpage
5214 * and a thread writing, so serialize them. The exception
5215 * is when the thread in wrip() does something which causes
5216 * a putpage operation. Then, the thread must be allowed
5217 * to continue. It may encounter a bmap_read problem in
5218 * ufs_putapage, but that is handled in ufs_putapage.
5219 * Allow async writers to proceed, we don't want to block
5220 * the pageout daemon.
5221 */
5228 /*
5229 * If there is no thread in the critical
5230 * section of wrip(), then proceed.
5231 * Otherwise, wait until there isn't one.
5232 */
5236 }
5238 /*
5239 * Bounce async writers when we have a writer
5240 * working on this file so we don't deadlock
5241 * the pageout daemon.
5242 */
5246 }
5249 }
5250 }
5251 }
5257 }
5260 /*
5261 * Search the entire vp list for pages >= off.
5262 */
5266 /*
5267 * Loop over all offsets in the range looking for
5268 * pages to deal with.
5269 */
5272 else
5276 /*
5277 * If we are not invalidating, synchronously
5278 * freeing or writing pages, use the routine
5279 * page_lookup_nowait() to prevent reclaiming
5280 * them from the free list.
5281 */
5287 io_off,
5289 }
5300 /*
5301 * "io_off" and "io_len" are returned as
5302 * the range of pages we actually wrote.
5303 * This allows us to skip ahead more quickly
5304 * since several pages may've been dealt
5305 * with by this iteration of the loop.
5306 */
5307 }
5308 }
5309 }
5311 /*
5312 * We have just sync'ed back all the pages on
5313 * the inode, turn off the IMODTIME flag.
5314 */
5318 }
5322 }
5324 static void
5326 {
5344 }
5348 }
5350 /*
5351 * Write out a single page, possibly klustering adjacent
5352 * dirty pages. The inode lock must be held.
5353 *
5354 * LMXXX - bsize < pagesize not done.
5355 */
5356 /*ARGSUSED*/
5357 int
5360 {
5361 uoff_t io_off;
5362 uoff_t off;
5368 daddr_t bn;
5378 }
5383 /*
5384 * If the modified time on the inode has not already been
5385 * set elsewhere (e.g. for write/setattr) we set the time now.
5386 * This gives us approximate modified times for mmap'ed files
5387 * which are modified via stores in the user address space.
5388 */
5395 }
5397 /*
5398 * Align the request to a block boundry (for old file systems),
5399 * and go ask bmap() how contiguous things are for this file.
5400 */
5407 /*
5408 * logging device is in error mode; simply return EIO
5409 */
5413 }
5414 /*
5415 * Oops, the thread in the window in wrip() did some
5416 * sort of operation which caused a putpage in the bad
5417 * range. In this case, just return an error which will
5418 * cause the software modified bit on the page to set
5419 * and the page will get written out again later.
5420 */
5424 }
5425 /*
5426 * If the pager is trying to push a page in the bad range
5427 * just tell it to try again later when things are better.
5428 */
5432 }
5435 }
5437 /*
5438 * If it is an fallocate'd block, reverse the negativity since
5439 * we are now writing to it
5440 */
5447 }
5449 /*
5450 * Take the length (of contiguous bytes) passed back from bmap()
5451 * and _try_ and get a set of pages covering that extent.
5452 */
5455 /*
5456 * May have run out of memory and not clustered backwards.
5457 * off p_offset
5458 * [ pp - 1 ][ pp ]
5459 * [ block ]
5460 * We told bmap off, so we have to adjust the bn accordingly.
5461 */
5465 }
5467 /*
5468 * bmap was carefull to tell us the right size so use that.
5469 * There might be unallocated frags at the end.
5470 * LMXXX - bzero the end of the page? We must be writing after EOF.
5471 */
5475 }
5477 /*
5478 * Handle the case where we are writing the last page after EOF.
5479 *
5480 * XXX - just a patch for i-mt3.
5481 */
5486 }
5498 /*
5499 * File contents of shadow or quota inodes are metadata, and updates
5500 * to these need to be put into a logging transaction. All direct
5501 * callers in UFS do that, but fsflush can come here _before_ the
5502 * normal codepath. An example would be updating ACL information, for
5503 * which the normal codepath would be:
5504 * ufs_si_store()
5505 * ufs_rdwri()
5506 * wrip()
5507 * segmap_release()
5508 * fop_putpage()
5509 * Here, fsflush can pick up the dirty page before segmap_release()
5510 * forces it out. If that happens, there's no transaction.
5511 * We therefore need to test whether a transaction exists, and if not
5512 * create one - for fsflush.
5513 */
5514 dotrans =
5522 }
5529 }
5530 }
5534 }
5536 /* write throttle */
5554 }
5565 }
5569 }
5573 out:
5581 out_trace:
5583 }
5588 /* ARGSUSED */
5589 static int
5593 {
5598 k_sigset_t smask;
5604 }
5609 }
5614 }
5616 retry_map:
5618 /*
5619 * If file is being locked, disallow mapping.
5620 */
5624 }
5627 /*
5628 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5629 * the previous attempt), some other thread could have grabbed
5630 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5631 * would unmap the valid VA range, that is ok.
5632 */
5637 }
5639 /*
5640 * a_lock has to be acquired before entering the lockfs protocol
5641 * because that is the order in which pagefault works. Also we cannot
5642 * block on a_lock here because this waiting writer will prevent
5643 * further readers like ufs_read from progressing and could cause
5644 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5645 * pending.
5646 */
5648 ufs_map_alock_retry_cnt++;
5650 }
5652 /*
5653 * We can't hold as->a_lock and wait for lockfs to succeed because
5654 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5655 * instead.
5656 */
5658 /*
5659 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5660 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5661 */
5662 ufs_map_lockfs_retry_cnt++;
5680 }
5681 }
5682 }
5685 }
5702 out:
5704 }
5706 /* ARGSUSED */
5707 static int
5711 {
5716 }
5722 }
5724 /*ARGSUSED*/
5725 static int
5729 {
5734 }
5741 }
5742 /*
5743 * Return the answer requested to poll() for non-device files
5744 */
5747 /* ARGSUSED */
5748 int
5751 {
5754 /*
5755 * Regular files reject edge-triggered pollers.
5756 * See the comment in fs_poll() for a more detailed explanation.
5757 */
5767 }
5790 }
5794 out:
5799 }
5801 /* ARGSUSED */
5802 static int
5805 {
5818 /*
5819 * Have to handle _PC_NAME_MAX here, because the normal way
5820 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5821 * results in a lock ordering reversal between
5822 * ufs_lockfs_{begin,end}() and
5823 * ufs_thread_{suspend,continue}().
5824 *
5825 * Keep in sync with ufs_statvfs().
5826 */
5834 else
5841 error =
5844 /* Start transaction */
5847 TOP_RMDIR,
5848 TOP_RMDIR_SIZE);
5849 }
5850 /*
5851 * Is directory empty
5852 */
5870 TOP_RMDIR_SIZE);
5871 }
5876 }
5879 }
5897 /*
5898 * UFS keeps only microsecond timestamp resolution.
5899 * This is historical and will probably never change.
5900 */
5907 }
5911 }
5913 }
5917 /*ARGSUSED*/
5918 static int
5921 {
5926 daddr_t bn;
5937 }
5940 /*
5941 * We need a better check. Ideally, we would use another
5942 * vnodeops so that hlocked and forcibly unmounted file
5943 * systems would return EIO where appropriate and w/o the
5944 * need for these checks.
5945 */
5949 /*
5950 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5951 * ul_lock must be taken before locking pages so we can't use it here
5952 * if pp is non NULL because segvn already locked pages
5953 * SE_EXCL. Instead we rely on the fact that a forced umount or
5954 * applying a filesystem lock via ufs_fiolfs() will block in the
5955 * implicit call to ufs_flush() until we unlock the pages after the
5956 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5957 * above 0 until they are done. We have to be careful not to increment
5958 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5959 *
5960 * If pp is NULL use ul_lock to make sure we don't increment
5961 * ul_vnops_cnt after forceful unmount hlocks the file system.
5962 */
5970 }
5972 }
5980 }
5981 }
5984 /*
5985 * segvn may call fop_pageio() instead of fop_getpage() to
5986 * handle a fault against a segment that maps vnode pages with
5987 * large mappings. Segvn creates pages and holds them locked
5988 * SE_EXCL during fop_pageio() call. In this case we have to
5989 * use rw_tryenter() to avoid a potential deadlock since in
5990 * lock order i_contents needs to be taken first.
5991 * Segvn will retry via fop_getpage() if fop_pageio() fails.
5992 */
5999 }
6000 }
6002 /*
6003 * Return an error to segvn because the pagefault request is beyond
6004 * PAGESIZE rounded EOF.
6005 */
6012 }
6019 }
6025 }
6027 /*
6028 * Break the io request into chunks, one for each contiguous
6029 * stretch of disk blocks in the target file.
6030 */
6042 }
6045 }
6048 /*
6049 * Zero out a page beyond EOF, when the last block of
6050 * a file is a UFS fragment so that ufs_pageio() can be used
6051 * instead of ufs_getpage() to handle faults against
6052 * segvn segments that use large pages.
6053 */
6058 }
6073 else
6077 ufs_pageio_reads++;
6078 else
6079 ufs_pageio_writes++;
6082 else
6084 /*
6085 * If the request is not B_ASYNC, wait for i/o to complete
6086 * and re-assemble the page list to return to the caller.
6087 * If it is B_ASYNC we leave the page list in pieces and
6088 * cleanup() will dispose of them.
6089 */
6096 }
6102 }
6106 /* Cleanup unprocessed parts of list */
6110 else
6113 /* Re-assemble list and let caller clean up */
6116 }
6117 }
6125 }
6132 }
6134 /*
6135 * Called when the kernel is in a frozen state to dump data
6136 * directly to the device. It uses a private dump data structure,
6137 * set up by dump_ctl, to locate the correct disk block to which to dump.
6138 */
6139 /*ARGSUSED*/
6140 static int
6143 {
6144 uoff_t file_size;
6152 /*
6153 * forced unmount case
6154 */
6157 /*
6158 * Validate the inode that it has not been modified since
6159 * the dump structure is allocated.
6160 */
6168 }
6171 /*
6172 * See that the file has room for this write
6173 */
6179 /*
6180 * Find the physical disk block numbers from the dump
6181 * private data structure directly and write out the data
6182 * in contiguous block lumps
6183 */
6191 nfsbs++;
6193 }
6200 }
6203 }
6205 /*
6206 * Prepare the file system before and after the dump operation.
6207 *
6208 * action = DUMP_ALLOC:
6209 * Preparation before dump, allocate dump private data structure
6210 * to hold all the direct and indirect block info for dump.
6211 *
6212 * action = DUMP_FREE:
6213 * Clean up after dump, deallocate the dump private data structure.
6214 *
6215 * action = DUMP_SCAN:
6216 * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6217 * if found, the starting file-relative DEV_BSIZE lbn is written
6218 * to *bklp; that lbn is intended for use with fop_dump()
6219 */
6220 /*ARGSUSED*/
6221 static int
6223 {
6233 /*
6234 * check for forced unmount
6235 */
6240 /*
6241 * alloc and record dump_info
6242 */
6254 }
6256 /*
6257 * calculate and allocate space needed according to i_size
6258 */
6265 }
6267 /* Start saving the info */
6272 /* Direct Blocks */
6276 /* Indirect Blocks */
6289 }
6299 }
6300 }
6301 /* and time stamp the information */
6308 /*
6309 * free dump_info
6310 */
6318 /*
6319 * scan dump_info
6320 */
6330 /*
6331 * scan dblk[] entries; contig fs space is found when:
6332 * ((current blkno + frags per block) == next blkno)
6333 */
6337 n++;
6338 else
6340 dblk++;
6341 nextblk++;
6342 }
6344 /*
6345 * index is where size bytes of contig space begins;
6346 * conversion from index to the file's DEV_BSIZE lbn
6347 * is equivalent to: (index * fs_bsize) / DEV_BSIZE
6348 */
6354 }
6356 }
6358 /*
6359 * Recursive helper function for ufs_dumpctl(). It follows the indirect file
6360 * system blocks until it reaches the the disk block addresses, which are
6361 * then stored into the given buffer, storeblk.
6362 */
6366 {
6376 }
6378 }
6387 }
6394 }
6396 }
6398 /* ARGSUSED */
6399 static int
6402 {
6411 /*
6412 * Only grab locks if needed - they're not needed to check vsa_mask
6413 * or if the mask contains no acl flags.
6414 */
6417 ULOCKFS_GETATTR_MASK))
6426 }
6428 }
6430 /* ARGSUSED */
6431 static int
6434 {
6447 /* Abort now if the request is either empty or invalid. */
6454 }
6456 /*
6457 * Following convention, if this is a directory then we acquire the
6458 * inode's i_rwlock after starting a UFS logging transaction;
6459 * otherwise, we acquire it beforehand. Since we were called (and
6460 * must therefore return) with the lock held, we will have to drop it,
6461 * and later reacquire it, if operating on a directory.
6462 */
6467 /* Upgrade the lock if required. */
6471 }
6472 }
6474 again:
6480 }
6482 /*
6483 * Check that the file system supports this operation. Note that
6484 * ufs_lockfs_begin() will have checked that the file system had
6485 * not been forcibly unmounted.
6486 */
6490 }
6494 }
6500 }
6505 }
6509 /* Do the actual work. */
6511 /*
6512 * Suppress out of inodes messages if we will retry.
6513 */
6520 out:
6523 /*
6524 * top_end_async() can eventually call
6525 * top_end_sync(), which can block. We must
6526 * therefore observe the lock-ordering protocol
6527 * here as well.
6528 */
6532 }
6534 }
6536 }
6537 /*
6538 * If no inodes available, try scaring a logically-
6539 * free one out of the delete queue to someplace
6540 * that we can find it.
6541 */
6548 }
6550 }
6551 /*
6552 * If we need to reacquire the lock then it is safe to do so
6553 * as a reader. This is because ufs_rwunlock(), which will be
6554 * called by our caller after we return, does not differentiate
6555 * between shared and exclusive locks.
6556 */
6560 }
6563 }
6565 /*
6566 * Locate the vnode to be used for an event notification. As this will
6567 * be called prior to the name space change perform basic verification
6568 * that the change will be allowed.
6569 */
6571 static int
6574 {
6594 }
6595 }
6597 /*
6598 * Check accessibility and write access of parent directory as we
6599 * only want to post the event if we're able to make a change.
6600 */
6608 }
6612 }
6614 /*
6615 * Keep the idle queue from getting too long by idling two
6616 * inodes before attempting to allocate another.
6617 * This operation must be performed before entering lockfs
6618 * or a transaction.
6619 */
6624 }
6628 retry_lookup:
6635 }
6639 }
6645 }