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 */
21
22 /*
23 * Copyright (c) 1984, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2017 Joyent, Inc.
26 * Copyright (c) 2016 by Delphix. All rights reserved.
27 */
28
29 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
30 /* All Rights Reserved */
31
32 /*
33 * Portions of this source code were derived from Berkeley 4.3 BSD
34 * under license from the Regents of the University of California.
35 */
36
37 #include <sys/types.h>
38 #include <sys/t_lock.h>
39 #include <sys/ksynch.h>
40 #include <sys/param.h>
41 #include <sys/time.h>
42 #include <sys/systm.h>
43 #include <sys/sysmacros.h>
44 #include <sys/resource.h>
45 #include <sys/signal.h>
46 #include <sys/cred.h>
47 #include <sys/user.h>
48 #include <sys/buf.h>
49 #include <sys/vfs.h>
50 #include <sys/vfs_opreg.h>
51 #include <sys/vnode.h>
52 #include <sys/proc.h>
53 #include <sys/disp.h>
54 #include <sys/file.h>
55 #include <sys/fcntl.h>
56 #include <sys/flock.h>
57 #include <sys/atomic.h>
58 #include <sys/kmem.h>
59 #include <sys/uio.h>
60 #include <sys/dnlc.h>
61 #include <sys/conf.h>
62 #include <sys/mman.h>
63 #include <sys/pathname.h>
64 #include <sys/debug.h>
65 #include <sys/vmsystm.h>
66 #include <sys/cmn_err.h>
67 #include <sys/filio.h>
68 #include <sys/policy.h>
69
70 #include <sys/fs/ufs_fs.h>
71 #include <sys/fs/ufs_lockfs.h>
72 #include <sys/fs/ufs_filio.h>
73 #include <sys/fs/ufs_inode.h>
74 #include <sys/fs/ufs_fsdir.h>
75 #include <sys/fs/ufs_quota.h>
76 #include <sys/fs/ufs_log.h>
77 #include <sys/fs/ufs_snap.h>
78 #include <sys/fs/ufs_trans.h>
79 #include <sys/fs/ufs_panic.h>
80 #include <sys/fs/ufs_bio.h>
81 #include <sys/dirent.h> /* must be AFTER <sys/fs/fsdir.h>! */
82 #include <sys/errno.h>
83 #include <sys/fssnap_if.h>
84 #include <sys/unistd.h>
85 #include <sys/sunddi.h>
86
87 #include <sys/filio.h> /* _FIOIO */
88
89 #include <vm/hat.h>
90 #include <vm/page.h>
91 #include <vm/pvn.h>
92 #include <vm/as.h>
93 #include <vm/seg.h>
94 #include <vm/seg_map.h>
95 #include <vm/seg_vn.h>
96 #include <vm/seg_kmem.h>
97 #include <vm/rm.h>
98 #include <sys/swap.h>
99
100 #include <fs/fs_subr.h>
101
102 #include <sys/fs/decomp.h>
103
104 static struct instats ins;
105
106 static int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
107 static int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
108 caddr_t, struct page **, size_t, enum seg_rw, int);
109 static int ufs_open(struct vnode **, int, struct cred *, caller_context_t *);
110 static int ufs_close(struct vnode *, int, int, offset_t, struct cred *,
111 caller_context_t *);
112 static int ufs_read(struct vnode *, struct uio *, int, struct cred *,
113 struct caller_context *);
114 static int ufs_write(struct vnode *, struct uio *, int, struct cred *,
115 struct caller_context *);
116 static int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *,
117 int *, caller_context_t *);
118 static int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *,
119 caller_context_t *);
120 static int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
121 caller_context_t *);
122 static int ufs_access(struct vnode *, int, int, struct cred *,
123 caller_context_t *);
124 static int ufs_lookup(struct vnode *, char *, struct vnode **,
125 struct pathname *, int, struct vnode *, struct cred *,
126 caller_context_t *, int *, pathname_t *);
127 static int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
128 int, struct vnode **, struct cred *, int,
129 caller_context_t *, vsecattr_t *);
130 static int ufs_remove(struct vnode *, char *, struct cred *,
131 caller_context_t *, int);
132 static int ufs_link(struct vnode *, struct vnode *, char *, struct cred *,
133 caller_context_t *, int);
134 static int ufs_rename(struct vnode *, char *, struct vnode *, char *,
135 struct cred *, caller_context_t *, int);
136 static int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
137 struct cred *, caller_context_t *, int, vsecattr_t *);
138 static int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *,
139 caller_context_t *, int);
140 static int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *,
141 caller_context_t *, int);
142 static int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
143 struct cred *, caller_context_t *, int);
144 static int ufs_readlink(struct vnode *, struct uio *, struct cred *,
145 caller_context_t *);
146 static int ufs_fsync(struct vnode *, int, struct cred *, caller_context_t *);
147 static void ufs_inactive(struct vnode *, struct cred *, caller_context_t *);
148 static int ufs_fid(struct vnode *, struct fid *, caller_context_t *);
149 static int ufs_rwlock(struct vnode *, int, caller_context_t *);
150 static void ufs_rwunlock(struct vnode *, int, caller_context_t *);
151 static int ufs_seek(struct vnode *, offset_t, offset_t *, caller_context_t *);
152 static int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
153 struct flk_callback *, struct cred *,
154 caller_context_t *);
155 static int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
156 cred_t *, caller_context_t *);
157 static int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
158 struct page **, size_t, struct seg *, caddr_t,
159 enum seg_rw, struct cred *, caller_context_t *);
160 static int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *,
161 caller_context_t *);
162 static int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
163 static int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
164 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
165 static int ufs_addmap(struct vnode *, offset_t, struct as *, caddr_t, size_t,
166 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
167 static int ufs_delmap(struct vnode *, offset_t, struct as *, caddr_t, size_t,
168 uint_t, uint_t, uint_t, struct cred *, caller_context_t *);
169 static int ufs_poll(vnode_t *, short, int, short *, struct pollhead **,
170 caller_context_t *);
171 static int ufs_dump(vnode_t *, caddr_t, offset_t, offset_t,
172 caller_context_t *);
173 static int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *,
174 caller_context_t *);
175 static int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
176 struct cred *, caller_context_t *);
177 static int ufs_dumpctl(vnode_t *, int, offset_t *, caller_context_t *);
178 static daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
179 daddr32_t *, int, int);
180 static int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
181 caller_context_t *);
182 static int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
183 caller_context_t *);
184 static int ufs_priv_access(void *, int, struct cred *);
185 static int ufs_eventlookup(struct vnode *, char *, struct cred *,
186 struct vnode **);
187 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
188
189 /*
190 * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
191 *
192 * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
193 */
194 struct vnodeops *ufs_vnodeops;
195
196 /* NOTE: "not blkd" below means that the operation isn't blocked by lockfs */
197 const fs_operation_def_t ufs_vnodeops_template[] = {
198 VOPNAME_OPEN, { .vop_open = ufs_open }, /* not blkd */
199 VOPNAME_CLOSE, { .vop_close = ufs_close }, /* not blkd */
200 VOPNAME_READ, { .vop_read = ufs_read },
201 VOPNAME_WRITE, { .vop_write = ufs_write },
202 VOPNAME_IOCTL, { .vop_ioctl = ufs_ioctl },
203 VOPNAME_GETATTR, { .vop_getattr = ufs_getattr },
204 VOPNAME_SETATTR, { .vop_setattr = ufs_setattr },
205 VOPNAME_ACCESS, { .vop_access = ufs_access },
206 VOPNAME_LOOKUP, { .vop_lookup = ufs_lookup },
207 VOPNAME_CREATE, { .vop_create = ufs_create },
208 VOPNAME_REMOVE, { .vop_remove = ufs_remove },
209 VOPNAME_LINK, { .vop_link = ufs_link },
210 VOPNAME_RENAME, { .vop_rename = ufs_rename },
211 VOPNAME_MKDIR, { .vop_mkdir = ufs_mkdir },
212 VOPNAME_RMDIR, { .vop_rmdir = ufs_rmdir },
213 VOPNAME_READDIR, { .vop_readdir = ufs_readdir },
214 VOPNAME_SYMLINK, { .vop_symlink = ufs_symlink },
215 VOPNAME_READLINK, { .vop_readlink = ufs_readlink },
216 VOPNAME_FSYNC, { .vop_fsync = ufs_fsync },
217 VOPNAME_INACTIVE, { .vop_inactive = ufs_inactive }, /* not blkd */
218 VOPNAME_FID, { .vop_fid = ufs_fid },
219 VOPNAME_RWLOCK, { .vop_rwlock = ufs_rwlock }, /* not blkd */
220 VOPNAME_RWUNLOCK, { .vop_rwunlock = ufs_rwunlock }, /* not blkd */
221 VOPNAME_SEEK, { .vop_seek = ufs_seek },
222 VOPNAME_FRLOCK, { .vop_frlock = ufs_frlock },
223 VOPNAME_SPACE, { .vop_space = ufs_space },
224 VOPNAME_GETPAGE, { .vop_getpage = ufs_getpage },
225 VOPNAME_PUTPAGE, { .vop_putpage = ufs_putpage },
226 VOPNAME_MAP, { .vop_map = ufs_map },
227 VOPNAME_ADDMAP, { .vop_addmap = ufs_addmap }, /* not blkd */
228 VOPNAME_DELMAP, { .vop_delmap = ufs_delmap }, /* not blkd */
229 VOPNAME_POLL, { .vop_poll = ufs_poll }, /* not blkd */
230 VOPNAME_DUMP, { .vop_dump = ufs_dump },
231 VOPNAME_PATHCONF, { .vop_pathconf = ufs_l_pathconf },
232 VOPNAME_PAGEIO, { .vop_pageio = ufs_pageio },
233 VOPNAME_DUMPCTL, { .vop_dumpctl = ufs_dumpctl },
234 VOPNAME_GETSECATTR, { .vop_getsecattr = ufs_getsecattr },
235 VOPNAME_SETSECATTR, { .vop_setsecattr = ufs_setsecattr },
236 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
237 NULL, NULL
238 };
239
240 #define MAX_BACKFILE_COUNT 9999
241
242 /*
243 * Created by ufs_dumpctl() to store a file's disk block info into memory.
244 * Used by ufs_dump() to dump data to disk directly.
245 */
246 struct dump {
247 struct inode *ip; /* the file we contain */
248 daddr_t fsbs; /* number of blocks stored */
249 struct timeval32 time; /* time stamp for the struct */
250 daddr32_t dblk[1]; /* place holder for block info */
251 };
252
253 static struct dump *dump_info = NULL;
254
255 /*
256 * Previously there was no special action required for ordinary files.
257 * (Devices are handled through the device file system.)
258 * Now we support Large Files and Large File API requires open to
259 * fail if file is large.
260 * We could take care to prevent data corruption
261 * by doing an atomic check of size and truncate if file is opened with
262 * FTRUNC flag set but traditionally this is being done by the vfs/vnode
263 * layers. So taking care of truncation here is a change in the existing
264 * semantics of VOP_OPEN and therefore we chose not to implement any thing
265 * here. The check for the size of the file > 2GB is being done at the
266 * vfs layer in routine vn_open().
267 */
268
269 /* ARGSUSED */
270 static int
271 ufs_open(struct vnode **vpp, int flag, struct cred *cr, caller_context_t *ct)
272 {
273 return (0);
274 }
275
276 /*ARGSUSED*/
277 static int
278 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
279 struct cred *cr, caller_context_t *ct)
280 {
281 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
282 cleanshares(vp, ttoproc(curthread)->p_pid);
283
284 /*
285 * Push partially filled cluster at last close.
286 * ``last close'' is approximated because the dnlc
287 * may have a hold on the vnode.
288 * Checking for VBAD here will also act as a forced umount check.
289 */
290 if (vp->v_count <= 2 && vp->v_type != VBAD) {
291 struct inode *ip = VTOI(vp);
292 if (ip->i_delaylen) {
293 ins.in_poc.value.ul++;
294 (void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
295 B_ASYNC | B_FREE, cr);
296 ip->i_delaylen = 0;
297 }
298 }
299
300 return (0);
301 }
302
303 /*ARGSUSED*/
304 static int
305 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
306 struct caller_context *ct)
307 {
308 struct inode *ip = VTOI(vp);
309 struct ufsvfs *ufsvfsp;
310 struct ulockfs *ulp = NULL;
311 int error = 0;
312 int intrans = 0;
313
314 ASSERT(RW_READ_HELD(&ip->i_rwlock));
315
316 /*
317 * Mandatory locking needs to be done before ufs_lockfs_begin()
318 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
319 */
320 if (MANDLOCK(vp, ip->i_mode)) {
321 /*
322 * ufs_getattr ends up being called by chklock
323 */
324 error = chklock(vp, FREAD, uiop->uio_loffset,
325 uiop->uio_resid, uiop->uio_fmode, ct);
326 if (error)
327 goto out;
328 }
329
330 ufsvfsp = ip->i_ufsvfs;
331 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
332 if (error)
333 goto out;
334
335 /*
336 * In the case that a directory is opened for reading as a file
337 * (eg "cat .") with the O_RSYNC, O_SYNC and O_DSYNC flags set.
338 * The locking order had to be changed to avoid a deadlock with
339 * an update taking place on that directory at the same time.
340 */
341 if ((ip->i_mode & IFMT) == IFDIR) {
342
343 rw_enter(&ip->i_contents, RW_READER);
344 error = rdip(ip, uiop, ioflag, cr);
345 rw_exit(&ip->i_contents);
346
347 if (error) {
348 if (ulp)
349 ufs_lockfs_end(ulp);
350 goto out;
351 }
352
353 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
354 TRANS_ISTRANS(ufsvfsp)) {
355 rw_exit(&ip->i_rwlock);
356 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
357 error);
358 ASSERT(!error);
359 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
360 TOP_READ_SIZE);
361 rw_enter(&ip->i_rwlock, RW_READER);
362 }
363 } else {
364 /*
365 * Only transact reads to files opened for sync-read and
366 * sync-write on a file system that is not write locked.
367 *
368 * The ``not write locked'' check prevents problems with
369 * enabling/disabling logging on a busy file system. E.g.,
370 * logging exists at the beginning of the read but does not
371 * at the end.
372 *
373 */
374 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
375 TRANS_ISTRANS(ufsvfsp)) {
376 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
377 error);
378 ASSERT(!error);
379 intrans = 1;
380 }
381
382 rw_enter(&ip->i_contents, RW_READER);
383 error = rdip(ip, uiop, ioflag, cr);
384 rw_exit(&ip->i_contents);
385
386 if (intrans) {
387 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
388 TOP_READ_SIZE);
389 }
390 }
391
392 if (ulp) {
393 ufs_lockfs_end(ulp);
394 }
395 out:
396
397 return (error);
398 }
399
400 extern volatile int ufs_HW; /* high water mark */
401 extern volatile int ufs_LW; /* low water mark */
402 volatile int ufs_WRITES = 1; /* XXX - enable/disable */
403 int ufs_throttles = 0; /* throttling count */
404 int ufs_allow_shared_writes = 1; /* directio shared writes */
405
406 static int
407 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
408 {
409 int shared_write;
410
411 /*
412 * If the FDSYNC flag is set then ignore the global
413 * ufs_allow_shared_writes in this case.
414 */
415 shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
416
417 /*
418 * Filter to determine if this request is suitable as a
419 * concurrent rewrite. This write must not allocate blocks
420 * by extending the file or filling in holes. No use trying
421 * through FSYNC descriptors as the inode will be synchronously
422 * updated after the write. The uio structure has not yet been
423 * checked for sanity, so assume nothing.
424 */
425 return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
426 (uiop->uio_loffset >= (offset_t)0) &&
427 (uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
428 ((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
429 !(ioflag & FSYNC) && !bmap_has_holes(ip) &&
430 shared_write);
431 }
432
433 /*ARGSUSED*/
434 static int
435 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
436 caller_context_t *ct)
437 {
438 struct inode *ip = VTOI(vp);
439 struct ufsvfs *ufsvfsp;
440 struct ulockfs *ulp;
441 int retry = 1;
442 int error, resv, resid = 0;
443 int directio_status;
444 int exclusive;
445 int rewriteflg;
446 long start_resid = uiop->uio_resid;
447
448 ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
449
450 retry_mandlock:
451 /*
452 * Mandatory locking needs to be done before ufs_lockfs_begin()
453 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
454 * Check for forced unmounts normally done in ufs_lockfs_begin().
455 */
456 if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
457 error = EIO;
458 goto out;
459 }
460 if (MANDLOCK(vp, ip->i_mode)) {
461
462 ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
463
464 /*
465 * ufs_getattr ends up being called by chklock
466 */
467 error = chklock(vp, FWRITE, uiop->uio_loffset,
468 uiop->uio_resid, uiop->uio_fmode, ct);
469 if (error)
470 goto out;
471 }
472
473 /* i_rwlock can change in chklock */
474 exclusive = rw_write_held(&ip->i_rwlock);
475 rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
476
477 /*
478 * Check for fast-path special case of directio re-writes.
479 */
480 if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
481 !exclusive && rewriteflg) {
482
483 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
484 if (error)
485 goto out;
486
487 rw_enter(&ip->i_contents, RW_READER);
488 error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
489 &directio_status);
490 if (directio_status == DIRECTIO_SUCCESS) {
491 uint_t i_flag_save;
492
493 if (start_resid != uiop->uio_resid)
494 error = 0;
495 /*
496 * Special treatment of access times for re-writes.
497 * If IMOD is not already set, then convert it
498 * to IMODACC for this operation. This defers
499 * entering a delta into the log until the inode
500 * is flushed. This mimics what is done for read
501 * operations and inode access time.
502 */
503 mutex_enter(&ip->i_tlock);
504 i_flag_save = ip->i_flag;
505 ip->i_flag |= IUPD | ICHG;
506 ip->i_seq++;
507 ITIMES_NOLOCK(ip);
508 if ((i_flag_save & IMOD) == 0) {
509 ip->i_flag &= ~IMOD;
510 ip->i_flag |= IMODACC;
511 }
512 mutex_exit(&ip->i_tlock);
513 rw_exit(&ip->i_contents);
514 if (ulp)
515 ufs_lockfs_end(ulp);
516 goto out;
517 }
518 rw_exit(&ip->i_contents);
519 if (ulp)
520 ufs_lockfs_end(ulp);
521 }
522
523 if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
524 rw_exit(&ip->i_rwlock);
525 rw_enter(&ip->i_rwlock, RW_WRITER);
526 /*
527 * Mandatory locking could have been enabled
528 * after dropping the i_rwlock.
529 */
530 if (MANDLOCK(vp, ip->i_mode))
531 goto retry_mandlock;
532 }
533
534 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
535 if (error)
536 goto out;
537
538 /*
539 * Amount of log space needed for this write
540 */
541 if (!rewriteflg || !(ioflag & FDSYNC))
542 TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
543
544 /*
545 * Throttle writes.
546 */
547 if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
548 mutex_enter(&ip->i_tlock);
549 while (ip->i_writes > ufs_HW) {
550 ufs_throttles++;
551 cv_wait(&ip->i_wrcv, &ip->i_tlock);
552 }
553 mutex_exit(&ip->i_tlock);
554 }
555
556 /*
557 * Enter Transaction
558 *
559 * If the write is a rewrite there is no need to open a transaction
560 * if the FDSYNC flag is set and not the FSYNC. In this case just
561 * set the IMODACC flag to modify do the update at a later time
562 * thus avoiding the overhead of the logging transaction that is
563 * not required.
564 */
565 if (ioflag & (FSYNC|FDSYNC)) {
566 if (ulp) {
567 if (rewriteflg) {
568 uint_t i_flag_save;
569
570 rw_enter(&ip->i_contents, RW_READER);
571 mutex_enter(&ip->i_tlock);
572 i_flag_save = ip->i_flag;
573 ip->i_flag |= IUPD | ICHG;
574 ip->i_seq++;
575 ITIMES_NOLOCK(ip);
576 if ((i_flag_save & IMOD) == 0) {
577 ip->i_flag &= ~IMOD;
578 ip->i_flag |= IMODACC;
579 }
580 mutex_exit(&ip->i_tlock);
581 rw_exit(&ip->i_contents);
582 } else {
583 int terr = 0;
584 TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
585 terr);
586 ASSERT(!terr);
587 }
588 }
589 } else {
590 if (ulp)
591 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
592 }
593
594 /*
595 * Write the file
596 */
597 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
598 rw_enter(&ip->i_contents, RW_WRITER);
599 if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
600 /*
601 * In append mode start at end of file.
602 */
603 uiop->uio_loffset = ip->i_size;
604 }
605
606 /*
607 * Mild optimisation, don't call ufs_trans_write() unless we have to
608 * Also, suppress file system full messages if we will retry.
609 */
610 if (retry)
611 ip->i_flag |= IQUIET;
612 if (resid) {
613 TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
614 } else {
615 error = wrip(ip, uiop, ioflag, cr);
616 }
617 ip->i_flag &= ~IQUIET;
618
619 rw_exit(&ip->i_contents);
620 rw_exit(&ufsvfsp->vfs_dqrwlock);
621
622 /*
623 * Leave Transaction
624 */
625 if (ulp) {
626 if (ioflag & (FSYNC|FDSYNC)) {
627 if (!rewriteflg) {
628 int terr = 0;
629
630 TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
631 resv);
632 if (error == 0)
633 error = terr;
634 }
635 } else {
636 TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
637 }
638 ufs_lockfs_end(ulp);
639 }
640 out:
641 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
642 /*
643 * Any blocks tied up in pending deletes?
644 */
645 ufs_delete_drain_wait(ufsvfsp, 1);
646 retry = 0;
647 goto retry_mandlock;
648 }
649
650 if (error == ENOSPC && (start_resid != uiop->uio_resid))
651 error = 0;
652
653 return (error);
654 }
655
656 /*
657 * Don't cache write blocks to files with the sticky bit set.
658 * Used to keep swap files from blowing the page cache on a server.
659 */
660 int stickyhack = 1;
661
662 /*
663 * Free behind hacks. The pager is busted.
664 * XXX - need to pass the information down to writedone() in a flag like B_SEQ
665 * or B_FREE_IF_TIGHT_ON_MEMORY.
666 */
667 volatile int freebehind = 1;
668 volatile int smallfile = 0;
669 u_offset_t smallfile64 = 32 * 1024;
670
671 /*
672 * While we should, in most cases, cache the pages for write, we
673 * may also want to cache the pages for read as long as they are
674 * frequently re-usable.
675 *
676 * If cache_read_ahead = 1, the pages for read will go to the tail
677 * of the cache list when they are released, otherwise go to the head.
678 */
679 int cache_read_ahead = 0;
680
681 /*
682 * Freebehind exists so that as we read large files sequentially we
683 * don't consume most of memory with pages from a few files. It takes
684 * longer to re-read from disk multiple small files as it does reading
685 * one large one sequentially. As system memory grows customers need
686 * to retain bigger chunks of files in memory. The advent of the
687 * cachelist opens up of the possibility freeing pages to the head or
688 * tail of the list.
689 *
690 * Not freeing a page is a bet that the page will be read again before
691 * it's segmap slot is needed for something else. If we loose the bet,
692 * it means some other thread is burdened with the page free we did
693 * not do. If we win we save a free and reclaim.
694 *
695 * Freeing it at the tail vs the head of cachelist is a bet that the
696 * page will survive until the next read. It's also saying that this
697 * page is more likely to be re-used than a page freed some time ago
698 * and never reclaimed.
699 *
700 * Freebehind maintains a range of file offset [smallfile1; smallfile2]
701 *
702 * 0 < offset < smallfile1 : pages are not freed.
703 * smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
704 * smallfile2 < offset : pages freed to head of cachelist.
705 *
706 * The range is computed at most once per second and depends on
707 * freemem and ncpus_online. Both parameters are bounded to be
708 * >= smallfile && >= smallfile64.
709 *
710 * smallfile1 = (free memory / ncpu) / 1000
711 * smallfile2 = (free memory / ncpu) / 10
712 *
713 * A few examples values:
714 *
715 * Free Mem (in Bytes) [smallfile1; smallfile2] [smallfile1; smallfile2]
716 * ncpus_online = 4 ncpus_online = 64
717 * ------------------ ----------------------- -----------------------
718 * 1G [256K; 25M] [32K; 1.5M]
719 * 10G [2.5M; 250M] [156K; 15M]
720 * 100G [25M; 2.5G] [1.5M; 150M]
721 *
722 */
723
724 #define SMALLFILE1_D 1000
725 #define SMALLFILE2_D 10
726 static u_offset_t smallfile1 = 32 * 1024;
727 static u_offset_t smallfile2 = 32 * 1024;
728 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
729 uint_t smallfile1_d = SMALLFILE1_D;
730 uint_t smallfile2_d = SMALLFILE2_D;
731
732 /*
733 * wrip does the real work of write requests for ufs.
734 */
735 int
736 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
737 {
738 rlim64_t limit = uio->uio_llimit;
739 u_offset_t off;
740 u_offset_t old_i_size;
741 struct fs *fs;
742 struct vnode *vp;
743 struct ufsvfs *ufsvfsp;
744 caddr_t base;
745 long start_resid = uio->uio_resid; /* save starting resid */
746 long premove_resid; /* resid before uiomove() */
747 uint_t flags;
748 int newpage;
749 int iupdat_flag, directio_status;
750 int n, on, mapon;
751 int error, pagecreate;
752 int do_dqrwlock; /* drop/reacquire vfs_dqrwlock */
753 int32_t iblocks;
754 int new_iblocks;
755
756 /*
757 * ip->i_size is incremented before the uiomove
758 * is done on a write. If the move fails (bad user
759 * address) reset ip->i_size.
760 * The better way would be to increment ip->i_size
761 * only if the uiomove succeeds.
762 */
763 int i_size_changed = 0;
764 o_mode_t type;
765 int i_seq_needed = 0;
766
767 vp = ITOV(ip);
768
769 /*
770 * check for forced unmount - should not happen as
771 * the request passed the lockfs checks.
772 */
773 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
774 return (EIO);
775
776 fs = ip->i_fs;
777
778 ASSERT(RW_WRITE_HELD(&ip->i_contents));
779
780 /* check for valid filetype */
781 type = ip->i_mode & IFMT;
782 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
783 (type != IFLNK) && (type != IFSHAD)) {
784 return (EIO);
785 }
786
787 /*
788 * the actual limit of UFS file size
789 * is UFS_MAXOFFSET_T
790 */
791 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
792 limit = MAXOFFSET_T;
793
794 if (uio->uio_loffset >= limit) {
795 proc_t *p = ttoproc(curthread);
796
797 mutex_enter(&p->p_lock);
798 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
799 p, RCA_UNSAFE_SIGINFO);
800 mutex_exit(&p->p_lock);
801 return (EFBIG);
802 }
803
804 /*
805 * if largefiles are disallowed, the limit is
806 * the pre-largefiles value of 2GB
807 */
808 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
809 limit = MIN(UFS_MAXOFFSET_T, limit);
810 else
811 limit = MIN(MAXOFF32_T, limit);
812
813 if (uio->uio_loffset < (offset_t)0) {
814 return (EINVAL);
815 }
816 if (uio->uio_resid == 0) {
817 return (0);
818 }
819
820 if (uio->uio_loffset >= limit)
821 return (EFBIG);
822
823 ip->i_flag |= INOACC; /* don't update ref time in getpage */
824
825 if (ioflag & (FSYNC|FDSYNC)) {
826 ip->i_flag |= ISYNC;
827 iupdat_flag = 1;
828 }
829 /*
830 * Try to go direct
831 */
832 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
833 uio->uio_llimit = limit;
834 error = ufs_directio_write(ip, uio, ioflag, 0, cr,
835 &directio_status);
836 /*
837 * If ufs_directio wrote to the file or set the flags,
838 * we need to update i_seq, but it may be deferred.
839 */
840 if (start_resid != uio->uio_resid ||
841 (ip->i_flag & (ICHG|IUPD))) {
842 i_seq_needed = 1;
843 ip->i_flag |= ISEQ;
844 }
845 if (directio_status == DIRECTIO_SUCCESS)
846 goto out;
847 }
848
849 /*
850 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
851 *
852 * o shadow inodes: vfs_dqrwlock is not held at all
853 * o quota updates: vfs_dqrwlock is read or write held
854 * o other updates: vfs_dqrwlock is read held
855 *
856 * The first case is the only one where we do not hold
857 * vfs_dqrwlock at all while entering wrip().
858 * We must make sure not to downgrade/drop vfs_dqrwlock if we
859 * have it as writer, i.e. if we are updating the quota inode.
860 * There is no potential deadlock scenario in this case as
861 * ufs_getpage() takes care of this and avoids reacquiring
862 * vfs_dqrwlock in that case.
863 *
864 * This check is done here since the above conditions do not change
865 * and we possibly loop below, so save a few cycles.
866 */
867 if ((type == IFSHAD) ||
868 (rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
869 do_dqrwlock = 0;
870 } else {
871 do_dqrwlock = 1;
872 }
873
874 /*
875 * Large Files: We cast MAXBMASK to offset_t
876 * inorder to mask out the higher bits. Since offset_t
877 * is a signed value, the high order bit set in MAXBMASK
878 * value makes it do the right thing by having all bits 1
879 * in the higher word. May be removed for _SOLARIS64_.
880 */
881
882 fs = ip->i_fs;
883 do {
884 u_offset_t uoff = uio->uio_loffset;
885 off = uoff & (offset_t)MAXBMASK;
886 mapon = (int)(uoff & (offset_t)MAXBOFFSET);
887 on = (int)blkoff(fs, uoff);
888 n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
889 new_iblocks = 1;
890
891 if (type == IFREG && uoff + n >= limit) {
892 if (uoff >= limit) {
893 error = EFBIG;
894 goto out;
895 }
896 /*
897 * since uoff + n >= limit,
898 * therefore n >= limit - uoff, and n is an int
899 * so it is safe to cast it to an int
900 */
901 n = (int)(limit - (rlim64_t)uoff);
902 }
903 if (uoff + n > ip->i_size) {
904 /*
905 * We are extending the length of the file.
906 * bmap is used so that we are sure that
907 * if we need to allocate new blocks, that it
908 * is done here before we up the file size.
909 */
910 error = bmap_write(ip, uoff, (int)(on + n),
911 mapon == 0, NULL, cr);
912 /*
913 * bmap_write never drops i_contents so if
914 * the flags are set it changed the file.
915 */
916 if (ip->i_flag & (ICHG|IUPD)) {
917 i_seq_needed = 1;
918 ip->i_flag |= ISEQ;
919 }
920 if (error)
921 break;
922 /*
923 * There is a window of vulnerability here.
924 * The sequence of operations: allocate file
925 * system blocks, uiomove the data into pages,
926 * and then update the size of the file in the
927 * inode, must happen atomically. However, due
928 * to current locking constraints, this can not
929 * be done.
930 */
931 ASSERT(ip->i_writer == NULL);
932 ip->i_writer = curthread;
933 i_size_changed = 1;
934 /*
935 * If we are writing from the beginning of
936 * the mapping, we can just create the
937 * pages without having to read them.
938 */
939 pagecreate = (mapon == 0);
940 } else if (n == MAXBSIZE) {
941 /*
942 * Going to do a whole mappings worth,
943 * so we can just create the pages w/o
944 * having to read them in. But before
945 * we do that, we need to make sure any
946 * needed blocks are allocated first.
947 */
948 iblocks = ip->i_blocks;
949 error = bmap_write(ip, uoff, (int)(on + n),
950 BI_ALLOC_ONLY, NULL, cr);
951 /*
952 * bmap_write never drops i_contents so if
953 * the flags are set it changed the file.
954 */
955 if (ip->i_flag & (ICHG|IUPD)) {
956 i_seq_needed = 1;
957 ip->i_flag |= ISEQ;
958 }
959 if (error)
960 break;
961 pagecreate = 1;
962 /*
963 * check if the new created page needed the
964 * allocation of new disk blocks.
965 */
966 if (iblocks == ip->i_blocks)
967 new_iblocks = 0; /* no new blocks allocated */
968 } else {
969 pagecreate = 0;
970 /*
971 * In sync mode flush the indirect blocks which
972 * may have been allocated and not written on
973 * disk. In above cases bmap_write will allocate
974 * in sync mode.
975 */
976 if (ioflag & (FSYNC|FDSYNC)) {
977 error = ufs_indirblk_sync(ip, uoff);
978 if (error)
979 break;
980 }
981 }
982
983 /*
984 * At this point we can enter ufs_getpage() in one
985 * of two ways:
986 * 1) segmap_getmapflt() calls ufs_getpage() when the
987 * forcefault parameter is true (pagecreate == 0)
988 * 2) uiomove() causes a page fault.
989 *
990 * We have to drop the contents lock to prevent the VM
991 * system from trying to reacquire it in ufs_getpage()
992 * should the uiomove cause a pagefault.
993 *
994 * We have to drop the reader vfs_dqrwlock here as well.
995 */
996 rw_exit(&ip->i_contents);
997 if (do_dqrwlock) {
998 ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
999 ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
1000 rw_exit(&ufsvfsp->vfs_dqrwlock);
1001 }
1002
1003 newpage = 0;
1004 premove_resid = uio->uio_resid;
1005
1006 /*
1007 * Touch the page and fault it in if it is not in core
1008 * before segmap_getmapflt or vpm_data_copy can lock it.
1009 * This is to avoid the deadlock if the buffer is mapped
1010 * to the same file through mmap which we want to write.
1011 */
1012 uio_prefaultpages((long)n, uio);
1013
1014 if (vpm_enable) {
1015 /*
1016 * Copy data. If new pages are created, part of
1017 * the page that is not written will be initizliazed
1018 * with zeros.
1019 */
1020 error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1021 uio, !pagecreate, &newpage, 0, S_WRITE);
1022 } else {
1023
1024 base = segmap_getmapflt(segkmap, vp, (off + mapon),
1025 (uint_t)n, !pagecreate, S_WRITE);
1026
1027 /*
1028 * segmap_pagecreate() returns 1 if it calls
1029 * page_create_va() to allocate any pages.
1030 */
1031
1032 if (pagecreate)
1033 newpage = segmap_pagecreate(segkmap, base,
1034 (size_t)n, 0);
1035
1036 error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1037 }
1038
1039 /*
1040 * If "newpage" is set, then a new page was created and it
1041 * does not contain valid data, so it needs to be initialized
1042 * at this point.
1043 * Otherwise the page contains old data, which was overwritten
1044 * partially or as a whole in uiomove.
1045 * If there is only one iovec structure within uio, then
1046 * on error uiomove will not be able to update uio->uio_loffset
1047 * and we would zero the whole page here!
1048 *
1049 * If uiomove fails because of an error, the old valid data
1050 * is kept instead of filling the rest of the page with zero's.
1051 */
1052 if (!vpm_enable && newpage &&
1053 uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1054 /*
1055 * We created pages w/o initializing them completely,
1056 * thus we need to zero the part that wasn't set up.
1057 * This happens on most EOF write cases and if
1058 * we had some sort of error during the uiomove.
1059 */
1060 int nzero, nmoved;
1061
1062 nmoved = (int)(uio->uio_loffset - (off + mapon));
1063 ASSERT(nmoved >= 0 && nmoved <= n);
1064 nzero = roundup(on + n, PAGESIZE) - nmoved;
1065 ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1066 (void) kzero(base + mapon + nmoved, (uint_t)nzero);
1067 }
1068
1069 /*
1070 * Unlock the pages allocated by page_create_va()
1071 * in segmap_pagecreate()
1072 */
1073 if (!vpm_enable && newpage)
1074 segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1075
1076 /*
1077 * If the size of the file changed, then update the
1078 * size field in the inode now. This can't be done
1079 * before the call to segmap_pageunlock or there is
1080 * a potential deadlock with callers to ufs_putpage().
1081 * They will be holding i_contents and trying to lock
1082 * a page, while this thread is holding a page locked
1083 * and trying to acquire i_contents.
1084 */
1085 if (i_size_changed) {
1086 rw_enter(&ip->i_contents, RW_WRITER);
1087 old_i_size = ip->i_size;
1088 UFS_SET_ISIZE(uoff + n, ip);
1089 TRANS_INODE(ufsvfsp, ip);
1090 /*
1091 * file has grown larger than 2GB. Set flag
1092 * in superblock to indicate this, if it
1093 * is not already set.
1094 */
1095 if ((ip->i_size > MAXOFF32_T) &&
1096 !(fs->fs_flags & FSLARGEFILES)) {
1097 ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1098 mutex_enter(&ufsvfsp->vfs_lock);
1099 fs->fs_flags |= FSLARGEFILES;
1100 ufs_sbwrite(ufsvfsp);
1101 mutex_exit(&ufsvfsp->vfs_lock);
1102 }
1103 mutex_enter(&ip->i_tlock);
1104 ip->i_writer = NULL;
1105 cv_broadcast(&ip->i_wrcv);
1106 mutex_exit(&ip->i_tlock);
1107 rw_exit(&ip->i_contents);
1108 }
1109
1110 if (error) {
1111 /*
1112 * If we failed on a write, we may have already
1113 * allocated file blocks as well as pages. It's
1114 * hard to undo the block allocation, but we must
1115 * be sure to invalidate any pages that may have
1116 * been allocated.
1117 *
1118 * If the page was created without initialization
1119 * then we must check if it should be possible
1120 * to destroy the new page and to keep the old data
1121 * on the disk.
1122 *
1123 * It is possible to destroy the page without
1124 * having to write back its contents only when
1125 * - the size of the file keeps unchanged
1126 * - bmap_write() did not allocate new disk blocks
1127 * it is possible to create big files using "seek" and
1128 * write to the end of the file. A "write" to a
1129 * position before the end of the file would not
1130 * change the size of the file but it would allocate
1131 * new disk blocks.
1132 * - uiomove intended to overwrite the whole page.
1133 * - a new page was created (newpage == 1).
1134 */
1135
1136 if (i_size_changed == 0 && new_iblocks == 0 &&
1137 newpage) {
1138
1139 /* unwind what uiomove eventually last did */
1140 uio->uio_resid = premove_resid;
1141
1142 /*
1143 * destroy the page, do not write ambiguous
1144 * data to the disk.
1145 */
1146 flags = SM_DESTROY;
1147 } else {
1148 /*
1149 * write the page back to the disk, if dirty,
1150 * and remove the page from the cache.
1151 */
1152 flags = SM_INVAL;
1153 }
1154
1155 if (vpm_enable) {
1156 /*
1157 * Flush pages.
1158 */
1159 (void) vpm_sync_pages(vp, off, n, flags);
1160 } else {
1161 (void) segmap_release(segkmap, base, flags);
1162 }
1163 } else {
1164 flags = 0;
1165 /*
1166 * Force write back for synchronous write cases.
1167 */
1168 if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1169 /*
1170 * If the sticky bit is set but the
1171 * execute bit is not set, we do a
1172 * synchronous write back and free
1173 * the page when done. We set up swap
1174 * files to be handled this way to
1175 * prevent servers from keeping around
1176 * the client's swap pages too long.
1177 * XXX - there ought to be a better way.
1178 */
1179 if (IS_SWAPVP(vp)) {
1180 flags = SM_WRITE | SM_FREE |
1181 SM_DONTNEED;
1182 iupdat_flag = 0;
1183 } else {
1184 flags = SM_WRITE;
1185 }
1186 } else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1187 /*
1188 * Have written a whole block.
1189 * Start an asynchronous write and
1190 * mark the buffer to indicate that
1191 * it won't be needed again soon.
1192 */
1193 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1194 }
1195 if (vpm_enable) {
1196 /*
1197 * Flush pages.
1198 */
1199 error = vpm_sync_pages(vp, off, n, flags);
1200 } else {
1201 error = segmap_release(segkmap, base, flags);
1202 }
1203 /*
1204 * If the operation failed and is synchronous,
1205 * then we need to unwind what uiomove() last
1206 * did so we can potentially return an error to
1207 * the caller. If this write operation was
1208 * done in two pieces and the first succeeded,
1209 * then we won't return an error for the second
1210 * piece that failed. However, we only want to
1211 * return a resid value that reflects what was
1212 * really done.
1213 *
1214 * Failures for non-synchronous operations can
1215 * be ignored since the page subsystem will
1216 * retry the operation until it succeeds or the
1217 * file system is unmounted.
1218 */
1219 if (error) {
1220 if ((ioflag & (FSYNC | FDSYNC)) ||
1221 type == IFDIR) {
1222 uio->uio_resid = premove_resid;
1223 } else {
1224 error = 0;
1225 }
1226 }
1227 }
1228
1229 /*
1230 * Re-acquire contents lock.
1231 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1232 */
1233 if (do_dqrwlock)
1234 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1235 rw_enter(&ip->i_contents, RW_WRITER);
1236
1237 /*
1238 * If the uiomove() failed or if a synchronous
1239 * page push failed, fix up i_size.
1240 */
1241 if (error) {
1242 if (i_size_changed) {
1243 /*
1244 * The uiomove failed, and we
1245 * allocated blocks,so get rid
1246 * of them.
1247 */
1248 (void) ufs_itrunc(ip, old_i_size, 0, cr);
1249 }
1250 } else {
1251 /*
1252 * XXX - Can this be out of the loop?
1253 */
1254 ip->i_flag |= IUPD | ICHG;
1255 /*
1256 * Only do one increase of i_seq for multiple
1257 * pieces. Because we drop locks, record
1258 * the fact that we changed the timestamp and
1259 * are deferring the increase in case another thread
1260 * pushes our timestamp update.
1261 */
1262 i_seq_needed = 1;
1263 ip->i_flag |= ISEQ;
1264 if (i_size_changed)
1265 ip->i_flag |= IATTCHG;
1266 if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1267 (IEXEC >> 6))) != 0 &&
1268 (ip->i_mode & (ISUID | ISGID)) != 0 &&
1269 secpolicy_vnode_setid_retain(cr,
1270 (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1271 /*
1272 * Clear Set-UID & Set-GID bits on
1273 * successful write if not privileged
1274 * and at least one of the execute bits
1275 * is set. If we always clear Set-GID,
1276 * mandatory file and record locking is
1277 * unuseable.
1278 */
1279 ip->i_mode &= ~(ISUID | ISGID);
1280 }
1281 }
1282 /*
1283 * In the case the FDSYNC flag is set and this is a
1284 * "rewrite" we won't log a delta.
1285 * The FSYNC flag overrides all cases.
1286 */
1287 if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1288 TRANS_INODE(ufsvfsp, ip);
1289 }
1290 } while (error == 0 && uio->uio_resid > 0 && n != 0);
1291
1292 out:
1293 /*
1294 * Make sure i_seq is increased at least once per write
1295 */
1296 if (i_seq_needed) {
1297 ip->i_seq++;
1298 ip->i_flag &= ~ISEQ; /* no longer deferred */
1299 }
1300
1301 /*
1302 * Inode is updated according to this table -
1303 *
1304 * FSYNC FDSYNC(posix.4)
1305 * --------------------------
1306 * always@ IATTCHG|IBDWRITE
1307 *
1308 * @ - If we are doing synchronous write the only time we should
1309 * not be sync'ing the ip here is if we have the stickyhack
1310 * activated, the file is marked with the sticky bit and
1311 * no exec bit, the file length has not been changed and
1312 * no new blocks have been allocated during this write.
1313 */
1314
1315 if ((ip->i_flag & ISYNC) != 0) {
1316 /*
1317 * we have eliminated nosync
1318 */
1319 if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1320 ((ioflag & FSYNC) && iupdat_flag)) {
1321 ufs_iupdat(ip, 1);
1322 }
1323 }
1324
1325 /*
1326 * If we've already done a partial-write, terminate
1327 * the write but return no error unless the error is ENOSPC
1328 * because the caller can detect this and free resources and
1329 * try again.
1330 */
1331 if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1332 error = 0;
1333
1334 ip->i_flag &= ~(INOACC | ISYNC);
1335 ITIMES_NOLOCK(ip);
1336 return (error);
1337 }
1338
1339 /*
1340 * rdip does the real work of read requests for ufs.
1341 */
1342 int
1343 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1344 {
1345 u_offset_t off;
1346 caddr_t base;
1347 struct fs *fs;
1348 struct ufsvfs *ufsvfsp;
1349 struct vnode *vp;
1350 long oresid = uio->uio_resid;
1351 u_offset_t n, on, mapon;
1352 int error = 0;
1353 int doupdate = 1;
1354 uint_t flags;
1355 int dofree, directio_status;
1356 krw_t rwtype;
1357 o_mode_t type;
1358 clock_t now;
1359
1360 vp = ITOV(ip);
1361
1362 ASSERT(RW_LOCK_HELD(&ip->i_contents));
1363
1364 ufsvfsp = ip->i_ufsvfs;
1365
1366 if (ufsvfsp == NULL)
1367 return (EIO);
1368
1369 fs = ufsvfsp->vfs_fs;
1370
1371 /* check for valid filetype */
1372 type = ip->i_mode & IFMT;
1373 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1374 (type != IFLNK) && (type != IFSHAD)) {
1375 return (EIO);
1376 }
1377
1378 if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1379 error = 0;
1380 goto out;
1381 }
1382 if (uio->uio_loffset < (offset_t)0) {
1383 return (EINVAL);
1384 }
1385 if (uio->uio_resid == 0) {
1386 return (0);
1387 }
1388
1389 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1390 (!ufsvfsp->vfs_noatime)) {
1391 mutex_enter(&ip->i_tlock);
1392 ip->i_flag |= IACC;
1393 mutex_exit(&ip->i_tlock);
1394 }
1395 /*
1396 * Try to go direct
1397 */
1398 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1399 error = ufs_directio_read(ip, uio, cr, &directio_status);
1400 if (directio_status == DIRECTIO_SUCCESS)
1401 goto out;
1402 }
1403
1404 rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1405
1406 do {
1407 offset_t diff;
1408 u_offset_t uoff = uio->uio_loffset;
1409 off = uoff & (offset_t)MAXBMASK;
1410 mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1411 on = (u_offset_t)blkoff(fs, uoff);
1412 n = MIN((u_offset_t)fs->fs_bsize - on,
1413 (u_offset_t)uio->uio_resid);
1414
1415 diff = ip->i_size - uoff;
1416
1417 if (diff <= (offset_t)0) {
1418 error = 0;
1419 goto out;
1420 }
1421 if (diff < (offset_t)n)
1422 n = (int)diff;
1423
1424 /*
1425 * We update smallfile2 and smallfile1 at most every second.
1426 */
1427 now = ddi_get_lbolt();
1428 if (now >= smallfile_update) {
1429 uint64_t percpufreeb;
1430 if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1431 if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1432 percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1433 smallfile1 = percpufreeb / smallfile1_d;
1434 smallfile2 = percpufreeb / smallfile2_d;
1435 smallfile1 = MAX(smallfile1, smallfile);
1436 smallfile1 = MAX(smallfile1, smallfile64);
1437 smallfile2 = MAX(smallfile1, smallfile2);
1438 smallfile_update = now + hz;
1439 }
1440
1441 dofree = freebehind &&
1442 ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1443
1444 /*
1445 * At this point we can enter ufs_getpage() in one of two
1446 * ways:
1447 * 1) segmap_getmapflt() calls ufs_getpage() when the
1448 * forcefault parameter is true (value of 1 is passed)
1449 * 2) uiomove() causes a page fault.
1450 *
1451 * We cannot hold onto an i_contents reader lock without
1452 * risking deadlock in ufs_getpage() so drop a reader lock.
1453 * The ufs_getpage() dolock logic already allows for a
1454 * thread holding i_contents as writer to work properly
1455 * so we keep a writer lock.
1456 */
1457 if (rwtype == RW_READER)
1458 rw_exit(&ip->i_contents);
1459
1460 if (vpm_enable) {
1461 /*
1462 * Copy data.
1463 */
1464 error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1465 uio, 1, NULL, 0, S_READ);
1466 } else {
1467 base = segmap_getmapflt(segkmap, vp, (off + mapon),
1468 (uint_t)n, 1, S_READ);
1469 error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1470 }
1471
1472 flags = 0;
1473 if (!error) {
1474 /*
1475 * If reading sequential we won't need this
1476 * buffer again soon. For offsets in range
1477 * [smallfile1, smallfile2] release the pages
1478 * at the tail of the cache list, larger
1479 * offsets are released at the head.
1480 */
1481 if (dofree) {
1482 flags = SM_FREE | SM_ASYNC;
1483 if ((cache_read_ahead == 0) &&
1484 (off > smallfile2))
1485 flags |= SM_DONTNEED;
1486 }
1487 /*
1488 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1489 * we want to make sure that the page which has
1490 * been read, is written on disk if it is dirty.
1491 * And corresponding indirect blocks should also
1492 * be flushed out.
1493 */
1494 if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1495 flags &= ~SM_ASYNC;
1496 flags |= SM_WRITE;
1497 }
1498 if (vpm_enable) {
1499 error = vpm_sync_pages(vp, off, n, flags);
1500 } else {
1501 error = segmap_release(segkmap, base, flags);
1502 }
1503 } else {
1504 if (vpm_enable) {
1505 (void) vpm_sync_pages(vp, off, n, flags);
1506 } else {
1507 (void) segmap_release(segkmap, base, flags);
1508 }
1509 }
1510
1511 if (rwtype == RW_READER)
1512 rw_enter(&ip->i_contents, rwtype);
1513 } while (error == 0 && uio->uio_resid > 0 && n != 0);
1514 out:
1515 /*
1516 * Inode is updated according to this table if FRSYNC is set.
1517 *
1518 * FSYNC FDSYNC(posix.4)
1519 * --------------------------
1520 * always IATTCHG|IBDWRITE
1521 */
1522 /*
1523 * The inode is not updated if we're logging and the inode is a
1524 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1525 */
1526 if (ioflag & FRSYNC) {
1527 if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1528 doupdate = 0;
1529 }
1530 if (doupdate) {
1531 if ((ioflag & FSYNC) ||
1532 ((ioflag & FDSYNC) &&
1533 (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1534 ufs_iupdat(ip, 1);
1535 }
1536 }
1537 }
1538 /*
1539 * If we've already done a partial read, terminate
1540 * the read but return no error.
1541 */
1542 if (oresid != uio->uio_resid)
1543 error = 0;
1544 ITIMES(ip);
1545
1546 return (error);
1547 }
1548
1549 /* ARGSUSED */
1550 static int
1551 ufs_ioctl(
1552 struct vnode *vp,
1553 int cmd,
1554 intptr_t arg,
1555 int flag,
1556 struct cred *cr,
1557 int *rvalp,
1558 caller_context_t *ct)
1559 {
1560 struct lockfs lockfs, lockfs_out;
1561 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
1562 char *comment, *original_comment;
1563 struct fs *fs;
1564 struct ulockfs *ulp;
1565 offset_t off;
1566 int error;
1567 int issync;
1568 int trans_size;
1569
1570
1571 /*
1572 * forcibly unmounted
1573 */
1574 if (ufsvfsp == NULL || vp->v_vfsp == NULL ||
1575 vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1576 return (EIO);
1577 fs = ufsvfsp->vfs_fs;
1578
1579 if (cmd == Q_QUOTACTL) {
1580 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1581 if (error)
1582 return (error);
1583
1584 if (ulp) {
1585 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1586 TOP_SETQUOTA_SIZE(fs));
1587 }
1588
1589 error = quotactl(vp, arg, flag, cr);
1590
1591 if (ulp) {
1592 TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1593 TOP_SETQUOTA_SIZE(fs));
1594 ufs_lockfs_end(ulp);
1595 }
1596 return (error);
1597 }
1598
1599 switch (cmd) {
1600 case _FIOLFS:
1601 /*
1602 * file system locking
1603 */
1604 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1605 return (EPERM);
1606
1607 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1608 if (copyin((caddr_t)arg, &lockfs,
1609 sizeof (struct lockfs)))
1610 return (EFAULT);
1611 }
1612 #ifdef _SYSCALL32_IMPL
1613 else {
1614 struct lockfs32 lockfs32;
1615 /* Translate ILP32 lockfs to LP64 lockfs */
1616 if (copyin((caddr_t)arg, &lockfs32,
1617 sizeof (struct lockfs32)))
1618 return (EFAULT);
1619 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1620 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1621 lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1622 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1623 lockfs.lf_comment =
1624 (caddr_t)(uintptr_t)lockfs32.lf_comment;
1625 }
1626 #endif /* _SYSCALL32_IMPL */
1627
1628 if (lockfs.lf_comlen) {
1629 if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1630 return (ENAMETOOLONG);
1631 comment =
1632 kmem_alloc(lockfs.lf_comlen, KM_SLEEP);
1633 if (copyin(lockfs.lf_comment, comment,
1634 lockfs.lf_comlen)) {
1635 kmem_free(comment, lockfs.lf_comlen);
1636 return (EFAULT);
1637 }
1638 original_comment = lockfs.lf_comment;
1639 lockfs.lf_comment = comment;
1640 }
1641 if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1642 lockfs.lf_comment = original_comment;
1643
1644 if ((flag & DATAMODEL_MASK) ==
1645 DATAMODEL_NATIVE) {
1646 (void) copyout(&lockfs, (caddr_t)arg,
1647 sizeof (struct lockfs));
1648 }
1649 #ifdef _SYSCALL32_IMPL
1650 else {
1651 struct lockfs32 lockfs32;
1652 /* Translate LP64 to ILP32 lockfs */
1653 lockfs32.lf_lock =
1654 (uint32_t)lockfs.lf_lock;
1655 lockfs32.lf_flags =
1656 (uint32_t)lockfs.lf_flags;
1657 lockfs32.lf_key =
1658 (uint32_t)lockfs.lf_key;
1659 lockfs32.lf_comlen =
1660 (uint32_t)lockfs.lf_comlen;
1661 lockfs32.lf_comment =
1662 (uint32_t)(uintptr_t)
1663 lockfs.lf_comment;
1664 (void) copyout(&lockfs32, (caddr_t)arg,
1665 sizeof (struct lockfs32));
1666 }
1667 #endif /* _SYSCALL32_IMPL */
1668
1669 } else {
1670 if (lockfs.lf_comlen)
1671 kmem_free(comment, lockfs.lf_comlen);
1672 }
1673 return (error);
1674
1675 case _FIOLFSS:
1676 /*
1677 * get file system locking status
1678 */
1679
1680 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1681 if (copyin((caddr_t)arg, &lockfs,
1682 sizeof (struct lockfs)))
1683 return (EFAULT);
1684 }
1685 #ifdef _SYSCALL32_IMPL
1686 else {
1687 struct lockfs32 lockfs32;
1688 /* Translate ILP32 lockfs to LP64 lockfs */
1689 if (copyin((caddr_t)arg, &lockfs32,
1690 sizeof (struct lockfs32)))
1691 return (EFAULT);
1692 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1693 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1694 lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1695 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1696 lockfs.lf_comment =
1697 (caddr_t)(uintptr_t)lockfs32.lf_comment;
1698 }
1699 #endif /* _SYSCALL32_IMPL */
1700
1701 if (error = ufs_fiolfss(vp, &lockfs_out))
1702 return (error);
1703 lockfs.lf_lock = lockfs_out.lf_lock;
1704 lockfs.lf_key = lockfs_out.lf_key;
1705 lockfs.lf_flags = lockfs_out.lf_flags;
1706 lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1707 lockfs_out.lf_comlen);
1708
1709 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1710 if (copyout(&lockfs, (caddr_t)arg,
1711 sizeof (struct lockfs)))
1712 return (EFAULT);
1713 }
1714 #ifdef _SYSCALL32_IMPL
1715 else {
1716 /* Translate LP64 to ILP32 lockfs */
1717 struct lockfs32 lockfs32;
1718 lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1719 lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1720 lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1721 lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1722 lockfs32.lf_comment =
1723 (uint32_t)(uintptr_t)lockfs.lf_comment;
1724 if (copyout(&lockfs32, (caddr_t)arg,
1725 sizeof (struct lockfs32)))
1726 return (EFAULT);
1727 }
1728 #endif /* _SYSCALL32_IMPL */
1729
1730 if (lockfs.lf_comlen &&
1731 lockfs.lf_comment && lockfs_out.lf_comment)
1732 if (copyout(lockfs_out.lf_comment,
1733 lockfs.lf_comment, lockfs.lf_comlen))
1734 return (EFAULT);
1735 return (0);
1736
1737 case _FIOSATIME:
1738 /*
1739 * set access time
1740 */
1741
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 */
1748 if (ufsvfsp->vfs_noatime)
1749 return (0);
1750
1751 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1752 ULOCKFS_SETATTR_MASK);
1753 if (error)
1754 return (error);
1755
1756 if (ulp) {
1757 trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1758 TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1759 TOP_SETATTR, trans_size);
1760 }
1761
1762 error = ufs_fiosatime(vp, (struct timeval *)arg,
1763 flag, cr);
1764
1765 if (ulp) {
1766 TRANS_END_CSYNC(ufsvfsp, error, issync,
1767 TOP_SETATTR, trans_size);
1768 ufs_lockfs_end(ulp);
1769 }
1770 return (error);
1771
1772 case _FIOSDIO:
1773 /*
1774 * set delayed-io
1775 */
1776 return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1777
1778 case _FIOGDIO:
1779 /*
1780 * get delayed-io
1781 */
1782 return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1783
1784 case _FIOIO:
1785 /*
1786 * inode open
1787 */
1788 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1789 ULOCKFS_VGET_MASK);
1790 if (error)
1791 return (error);
1792
1793 error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1794
1795 if (ulp) {
1796 ufs_lockfs_end(ulp);
1797 }
1798 return (error);
1799
1800 case _FIOFFS:
1801 /*
1802 * file system flush (push w/invalidate)
1803 */
1804 if ((caddr_t)arg != NULL)
1805 return (EINVAL);
1806 return (ufs_fioffs(vp, NULL, cr));
1807
1808 case _FIOISBUSY:
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 */
1815 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1816 return (EPERM);
1817 error = ufs_fioisbusy(vp, (int *)arg, cr);
1818 return (error);
1819
1820 case _FIODIRECTIO:
1821 return (ufs_fiodirectio(vp, (int)arg, cr));
1822
1823 case _FIOTUNE:
1824 /*
1825 * Tune the file system (aka setting fs attributes)
1826 */
1827 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1828 ULOCKFS_SETATTR_MASK);
1829 if (error)
1830 return (error);
1831
1832 error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1833
1834 if (ulp)
1835 ufs_lockfs_end(ulp);
1836 return (error);
1837
1838 case _FIOLOGENABLE:
1839 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1840 return (EPERM);
1841 return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1842
1843 case _FIOLOGDISABLE:
1844 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1845 return (EPERM);
1846 return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1847
1848 case _FIOISLOG:
1849 return (ufs_fioislog(vp, (void *)arg, cr, flag));
1850
1851 case _FIOSNAPSHOTCREATE_MULTI:
1852 {
1853 struct fiosnapcreate_multi fc, *fcp;
1854 size_t fcm_size;
1855
1856 if (copyin((void *)arg, &fc, sizeof (fc)))
1857 return (EFAULT);
1858 if (fc.backfilecount > MAX_BACKFILE_COUNT)
1859 return (EINVAL);
1860 fcm_size = sizeof (struct fiosnapcreate_multi) +
1861 (fc.backfilecount - 1) * sizeof (int);
1862 fcp = (struct fiosnapcreate_multi *)
1863 kmem_alloc(fcm_size, KM_SLEEP);
1864 if (copyin((void *)arg, fcp, fcm_size)) {
1865 kmem_free(fcp, fcm_size);
1866 return (EFAULT);
1867 }
1868 error = ufs_snap_create(vp, fcp, cr);
1869 /*
1870 * Do copyout even if there is an error because
1871 * the details of error is stored in fcp.
1872 */
1873 if (copyout(fcp, (void *)arg, fcm_size))
1874 error = EFAULT;
1875 kmem_free(fcp, fcm_size);
1876 return (error);
1877 }
1878
1879 case _FIOSNAPSHOTDELETE:
1880 {
1881 struct fiosnapdelete fc;
1882
1883 if (copyin((void *)arg, &fc, sizeof (fc)))
1884 return (EFAULT);
1885 error = ufs_snap_delete(vp, &fc, cr);
1886 if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1887 error = EFAULT;
1888 return (error);
1889 }
1890
1891 case _FIOGETSUPERBLOCK:
1892 if (copyout(fs, (void *)arg, SBSIZE))
1893 return (EFAULT);
1894 return (0);
1895
1896 case _FIOGETMAXPHYS:
1897 if (copyout((void *)&maxphys, (void *)arg,
1898 sizeof (maxphys)))
1899 return (EFAULT);
1900 return (0);
1901
1902 /*
1903 * The following 3 ioctls are for TSufs support
1904 * although could potentially be used elsewhere
1905 */
1906 case _FIO_SET_LUFS_DEBUG:
1907 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1908 return (EPERM);
1909 lufs_debug = (uint32_t)arg;
1910 return (0);
1911
1912 case _FIO_SET_LUFS_ERROR:
1913 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1914 return (EPERM);
1915 TRANS_SETERROR(ufsvfsp);
1916 return (0);
1917
1918 case _FIO_GET_TOP_STATS:
1919 {
1920 fio_lufs_stats_t *ls;
1921 ml_unit_t *ul = ufsvfsp->vfs_log;
1922
1923 ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1924 ls->ls_debug = ul->un_debug; /* return debug value */
1925 /* Copy stucture if statistics are being kept */
1926 if (ul->un_logmap->mtm_tops) {
1927 ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1928 }
1929 error = 0;
1930 if (copyout(ls, (void *)arg, sizeof (*ls)))
1931 error = EFAULT;
1932 kmem_free(ls, sizeof (*ls));
1933 return (error);
1934 }
1935
1936 case _FIO_SEEK_DATA:
1937 case _FIO_SEEK_HOLE:
1938 if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1939 return (EFAULT);
1940 /* offset paramater is in/out */
1941 error = ufs_fio_holey(vp, cmd, &off);
1942 if (error)
1943 return (error);
1944 if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1945 return (EFAULT);
1946 return (0);
1947
1948 case _FIO_COMPRESSED:
1949 {
1950 /*
1951 * This is a project private ufs ioctl() to mark
1952 * the inode as that belonging to a compressed
1953 * file. This is used to mark individual
1954 * compressed files in a miniroot archive.
1955 * The files compressed in this manner are
1956 * automatically decompressed by the dcfs filesystem
1957 * (via an interception in ufs_lookup - see decompvp())
1958 * which is layered on top of ufs on a system running
1959 * from the archive. See uts/common/fs/dcfs for details.
1960 * This ioctl only marks the file as compressed - the
1961 * actual compression is done by fiocompress (a
1962 * userland utility) which invokes this ioctl().
1963 */
1964 struct inode *ip = VTOI(vp);
1965
1966 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1967 ULOCKFS_SETATTR_MASK);
1968 if (error)
1969 return (error);
1970
1971 if (ulp) {
1972 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_IUPDAT,
1973 TOP_IUPDAT_SIZE(ip));
1974 }
1975
1976 error = ufs_mark_compressed(vp);
1977
1978 if (ulp) {
1979 TRANS_END_ASYNC(ufsvfsp, TOP_IUPDAT,
1980 TOP_IUPDAT_SIZE(ip));
1981 ufs_lockfs_end(ulp);
1982 }
1983
1984 return (error);
1985
1986 }
1987
1988 default:
1989 return (ENOTTY);
1990 }
1991 }
1992
1993
1994 /* ARGSUSED */
1995 static int
1996 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1997 struct cred *cr, caller_context_t *ct)
1998 {
1999 struct inode *ip = VTOI(vp);
2000 struct ufsvfs *ufsvfsp;
2001 int err;
2002
2003 if (vap->va_mask == AT_SIZE) {
2004 /*
2005 * for performance, if only the size is requested don't bother
2006 * with anything else.
2007 */
2008 UFS_GET_ISIZE(&vap->va_size, ip);
2009 return (0);
2010 }
2011
2012 /*
2013 * inlined lockfs checks
2014 */
2015 ufsvfsp = ip->i_ufsvfs;
2016 if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
2017 err = EIO;
2018 goto out;
2019 }
2020
2021 rw_enter(&ip->i_contents, RW_READER);
2022 /*
2023 * Return all the attributes. This should be refined so
2024 * that it only returns what's asked for.
2025 */
2026
2027 /*
2028 * Copy from inode table.
2029 */
2030 vap->va_type = vp->v_type;
2031 vap->va_mode = ip->i_mode & MODEMASK;
2032 /*
2033 * If there is an ACL and there is a mask entry, then do the
2034 * extra work that completes the equivalent of an acltomode(3)
2035 * call. According to POSIX P1003.1e, the acl mask should be
2036 * returned in the group permissions field.
2037 *
2038 * - start with the original permission and mode bits (from above)
2039 * - clear the group owner bits
2040 * - add in the mask bits.
2041 */
2042 if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2043 vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2044 vap->va_mode |=
2045 (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2046 }
2047 vap->va_uid = ip->i_uid;
2048 vap->va_gid = ip->i_gid;
2049 vap->va_fsid = ip->i_dev;
2050 vap->va_nodeid = (ino64_t)ip->i_number;
2051 vap->va_nlink = ip->i_nlink;
2052 vap->va_size = ip->i_size;
2053 if (vp->v_type == VCHR || vp->v_type == VBLK)
2054 vap->va_rdev = ip->i_rdev;
2055 else
2056 vap->va_rdev = 0; /* not a b/c spec. */
2057 mutex_enter(&ip->i_tlock);
2058 ITIMES_NOLOCK(ip); /* mark correct time in inode */
2059 vap->va_seq = ip->i_seq;
2060 vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2061 vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2062 vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2063 vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2064 vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2065 vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2066 mutex_exit(&ip->i_tlock);
2067
2068 switch (ip->i_mode & IFMT) {
2069
2070 case IFBLK:
2071 vap->va_blksize = MAXBSIZE; /* was BLKDEV_IOSIZE */
2072 break;
2073
2074 case IFCHR:
2075 vap->va_blksize = MAXBSIZE;
2076 break;
2077
2078 default:
2079 vap->va_blksize = ip->i_fs->fs_bsize;
2080 break;
2081 }
2082 vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2083 rw_exit(&ip->i_contents);
2084 err = 0;
2085
2086 out:
2087 return (err);
2088 }
2089
2090 /*
2091 * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2092 * The i_contents lock is already held by the caller and we need to
2093 * declare the inode as 'void *' argument.
2094 */
2095 static int
2096 ufs_priv_access(void *vip, int mode, struct cred *cr)
2097 {
2098 struct inode *ip = vip;
2099
2100 return (ufs_iaccess(ip, mode, cr, 0));
2101 }
2102
2103 /*ARGSUSED4*/
2104 static int
2105 ufs_setattr(struct vnode *vp, struct vattr *vap, int flags, struct cred *cr,
2106 caller_context_t *ct)
2107 {
2108 struct inode *ip = VTOI(vp);
2109 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2110 struct fs *fs;
2111 struct ulockfs *ulp;
2112 char *errmsg1;
2113 char *errmsg2;
2114 long blocks;
2115 long int mask = vap->va_mask;
2116 size_t len1, len2;
2117 int issync;
2118 int trans_size;
2119 int dotrans;
2120 int dorwlock;
2121 int error;
2122 int owner_change;
2123 int dodqlock;
2124 timestruc_t now;
2125 vattr_t oldva;
2126 int retry = 1;
2127 int indeadlock;
2128
2129 /*
2130 * Cannot set these attributes.
2131 */
2132 if ((mask & AT_NOSET) || (mask & AT_XVATTR))
2133 return (EINVAL);
2134
2135 /*
2136 * check for forced unmount
2137 */
2138 if (ufsvfsp == NULL)
2139 return (EIO);
2140
2141 fs = ufsvfsp->vfs_fs;
2142 if (fs->fs_ronly != 0)
2143 return (EROFS);
2144
2145 again:
2146 errmsg1 = NULL;
2147 errmsg2 = NULL;
2148 dotrans = 0;
2149 dorwlock = 0;
2150 dodqlock = 0;
2151
2152 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2153 if (error)
2154 goto out;
2155
2156 /*
2157 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2158 * This follows the protocol for read()/write().
2159 */
2160 if (vp->v_type != VDIR) {
2161 /*
2162 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2163 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2164 * possible, retries the operation.
2165 */
2166 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2167 if (indeadlock) {
2168 if (ulp)
2169 ufs_lockfs_end(ulp);
2170 goto again;
2171 }
2172 dorwlock = 1;
2173 }
2174
2175 /*
2176 * Truncate file. Must have write permission and not be a directory.
2177 */
2178 if (mask & AT_SIZE) {
2179 rw_enter(&ip->i_contents, RW_WRITER);
2180 if (vp->v_type == VDIR) {
2181 error = EISDIR;
2182 goto update_inode;
2183 }
2184 if (error = ufs_iaccess(ip, IWRITE, cr, 0))
2185 goto update_inode;
2186
2187 rw_exit(&ip->i_contents);
2188 error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2189 if (error) {
2190 rw_enter(&ip->i_contents, RW_WRITER);
2191 goto update_inode;
2192 }
2193
2194 if (error == 0 && vap->va_size)
2195 vnevent_truncate(vp, ct);
2196 }
2197
2198 if (ulp) {
2199 trans_size = (int)TOP_SETATTR_SIZE(ip);
2200 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2201 ++dotrans;
2202 }
2203
2204 /*
2205 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2206 * This follows the protocol established by
2207 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2208 */
2209 if (vp->v_type == VDIR) {
2210 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2211 retry_dir);
2212 if (indeadlock)
2213 goto again;
2214 dorwlock = 1;
2215 }
2216
2217 /*
2218 * Grab quota lock if we are changing the file's owner.
2219 */
2220 if (mask & AT_UID) {
2221 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2222 dodqlock = 1;
2223 }
2224 rw_enter(&ip->i_contents, RW_WRITER);
2225
2226 oldva.va_mode = ip->i_mode;
2227 oldva.va_uid = ip->i_uid;
2228 oldva.va_gid = ip->i_gid;
2229
2230 vap->va_mask &= ~AT_SIZE;
2231
2232 error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2233 ufs_priv_access, ip);
2234 if (error)
2235 goto update_inode;
2236
2237 mask = vap->va_mask;
2238
2239 /*
2240 * Change file access modes.
2241 */
2242 if (mask & AT_MODE) {
2243 ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2244 TRANS_INODE(ufsvfsp, ip);
2245 ip->i_flag |= ICHG;
2246 if (stickyhack) {
2247 mutex_enter(&vp->v_lock);
2248 if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2249 vp->v_flag |= VSWAPLIKE;
2250 else
2251 vp->v_flag &= ~VSWAPLIKE;
2252 mutex_exit(&vp->v_lock);
2253 }
2254 }
2255 if (mask & (AT_UID|AT_GID)) {
2256 if (mask & AT_UID) {
2257 /*
2258 * Don't change ownership of the quota inode.
2259 */
2260 if (ufsvfsp->vfs_qinod == ip) {
2261 ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2262 error = EINVAL;
2263 goto update_inode;
2264 }
2265
2266 /*
2267 * No real ownership change.
2268 */
2269 if (ip->i_uid == vap->va_uid) {
2270 blocks = 0;
2271 owner_change = 0;
2272 }
2273 /*
2274 * Remove the blocks and the file, from the old user's
2275 * quota.
2276 */
2277 else {
2278 blocks = ip->i_blocks;
2279 owner_change = 1;
2280
2281 (void) chkdq(ip, -blocks, /* force */ 1, cr,
2282 (char **)NULL, (size_t *)NULL);
2283 (void) chkiq(ufsvfsp, /* change */ -1, ip,
2284 (uid_t)ip->i_uid, /* force */ 1, cr,
2285 (char **)NULL, (size_t *)NULL);
2286 dqrele(ip->i_dquot);
2287 }
2288
2289 ip->i_uid = vap->va_uid;
2290
2291 /*
2292 * There is a real ownership change.
2293 */
2294 if (owner_change) {
2295 /*
2296 * Add the blocks and the file to the new
2297 * user's quota.
2298 */
2299 ip->i_dquot = getinoquota(ip);
2300 (void) chkdq(ip, blocks, /* force */ 1, cr,
2301 &errmsg1, &len1);
2302 (void) chkiq(ufsvfsp, /* change */ 1,
2303 (struct inode *)NULL, (uid_t)ip->i_uid,
2304 /* force */ 1, cr, &errmsg2, &len2);
2305 }
2306 }
2307 if (mask & AT_GID) {
2308 ip->i_gid = vap->va_gid;
2309 }
2310 TRANS_INODE(ufsvfsp, ip);
2311 ip->i_flag |= ICHG;
2312 }
2313 /*
2314 * Change file access or modified times.
2315 */
2316 if (mask & (AT_ATIME|AT_MTIME)) {
2317 /* Check that the time value is within ufs range */
2318 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2319 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2320 error = EOVERFLOW;
2321 goto update_inode;
2322 }
2323
2324 /*
2325 * if the "noaccess" mount option is set and only atime
2326 * update is requested, do nothing. No error is returned.
2327 */
2328 if ((ufsvfsp->vfs_noatime) &&
2329 ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2330 goto skip_atime;
2331
2332 if (mask & AT_ATIME) {
2333 ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2334 ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2335 ip->i_flag &= ~IACC;
2336 }
2337 if (mask & AT_MTIME) {
2338 ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2339 ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2340 gethrestime(&now);
2341 if (now.tv_sec > TIME32_MAX) {
2342 /*
2343 * In 2038, ctime sticks forever..
2344 */
2345 ip->i_ctime.tv_sec = TIME32_MAX;
2346 ip->i_ctime.tv_usec = 0;
2347 } else {
2348 ip->i_ctime.tv_sec = now.tv_sec;
2349 ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2350 }
2351 ip->i_flag &= ~(IUPD|ICHG);
2352 ip->i_flag |= IMODTIME;
2353 }
2354 TRANS_INODE(ufsvfsp, ip);
2355 ip->i_flag |= IMOD;
2356 }
2357
2358 skip_atime:
2359 /*
2360 * The presence of a shadow inode may indicate an ACL, but does
2361 * not imply an ACL. Future FSD types should be handled here too
2362 * and check for the presence of the attribute-specific data
2363 * before referencing it.
2364 */
2365 if (ip->i_shadow) {
2366 /*
2367 * XXX if ufs_iupdat is changed to sandbagged write fix
2368 * ufs_acl_setattr to push ip to keep acls consistent
2369 *
2370 * Suppress out of inodes messages if we will retry.
2371 */
2372 if (retry)
2373 ip->i_flag |= IQUIET;
2374 error = ufs_acl_setattr(ip, vap, cr);
2375 ip->i_flag &= ~IQUIET;
2376 }
2377
2378 update_inode:
2379 /*
2380 * Setattr always increases the sequence number
2381 */
2382 ip->i_seq++;
2383
2384 /*
2385 * if nfsd and not logging; push synchronously
2386 */
2387 if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2388 ufs_iupdat(ip, 1);
2389 } else {
2390 ITIMES_NOLOCK(ip);
2391 }
2392
2393 rw_exit(&ip->i_contents);
2394 if (dodqlock) {
2395 rw_exit(&ufsvfsp->vfs_dqrwlock);
2396 }
2397 if (dorwlock)
2398 rw_exit(&ip->i_rwlock);
2399
2400 if (ulp) {
2401 if (dotrans) {
2402 int terr = 0;
2403 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2404 trans_size);
2405 if (error == 0)
2406 error = terr;
2407 }
2408 ufs_lockfs_end(ulp);
2409 }
2410 out:
2411 /*
2412 * If out of inodes or blocks, see if we can free something
2413 * up from the delete queue.
2414 */
2415 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2416 ufs_delete_drain_wait(ufsvfsp, 1);
2417 retry = 0;
2418 if (errmsg1 != NULL)
2419 kmem_free(errmsg1, len1);
2420 if (errmsg2 != NULL)
2421 kmem_free(errmsg2, len2);
2422 goto again;
2423 }
2424 if (errmsg1 != NULL) {
2425 uprintf(errmsg1);
2426 kmem_free(errmsg1, len1);
2427 }
2428 if (errmsg2 != NULL) {
2429 uprintf(errmsg2);
2430 kmem_free(errmsg2, len2);
2431 }
2432 return (error);
2433 }
2434
2435 /*ARGSUSED*/
2436 static int
2437 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr,
2438 caller_context_t *ct)
2439 {
2440 struct inode *ip = VTOI(vp);
2441
2442 if (ip->i_ufsvfs == NULL)
2443 return (EIO);
2444
2445 /*
2446 * The ufs_iaccess function wants to be called with
2447 * mode bits expressed as "ufs specific" bits.
2448 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2449 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2450 * But since they're the same we just pass the vnode mode
2451 * bit but just verify that assumption at compile time.
2452 */
2453 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2454 #error "ufs_access needs to map Vmodes to Imodes"
2455 #endif
2456 return (ufs_iaccess(ip, mode, cr, 1));
2457 }
2458
2459 /* ARGSUSED */
2460 static int
2461 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr,
2462 caller_context_t *ct)
2463 {
2464 struct inode *ip = VTOI(vp);
2465 struct ufsvfs *ufsvfsp;
2466 struct ulockfs *ulp;
2467 int error;
2468 int fastsymlink;
2469
2470 if (vp->v_type != VLNK) {
2471 error = EINVAL;
2472 goto nolockout;
2473 }
2474
2475 /*
2476 * If the symbolic link is empty there is nothing to read.
2477 * Fast-track these empty symbolic links
2478 */
2479 if (ip->i_size == 0) {
2480 error = 0;
2481 goto nolockout;
2482 }
2483
2484 ufsvfsp = ip->i_ufsvfs;
2485 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2486 if (error)
2487 goto nolockout;
2488 /*
2489 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2490 */
2491 again:
2492 fastsymlink = 0;
2493 if (ip->i_flag & IFASTSYMLNK) {
2494 rw_enter(&ip->i_rwlock, RW_READER);
2495 rw_enter(&ip->i_contents, RW_READER);
2496 if (ip->i_flag & IFASTSYMLNK) {
2497 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2498 (ip->i_fs->fs_ronly == 0) &&
2499 (!ufsvfsp->vfs_noatime)) {
2500 mutex_enter(&ip->i_tlock);
2501 ip->i_flag |= IACC;
2502 mutex_exit(&ip->i_tlock);
2503 }
2504 error = uiomove((caddr_t)&ip->i_db[1],
2505 MIN(ip->i_size, uiop->uio_resid),
2506 UIO_READ, uiop);
2507 ITIMES(ip);
2508 ++fastsymlink;
2509 }
2510 rw_exit(&ip->i_contents);
2511 rw_exit(&ip->i_rwlock);
2512 }
2513 if (!fastsymlink) {
2514 ssize_t size; /* number of bytes read */
2515 caddr_t basep; /* pointer to input data */
2516 ino_t ino;
2517 long igen;
2518 struct uio tuio; /* temp uio struct */
2519 struct uio *tuiop;
2520 iovec_t tiov; /* temp iovec struct */
2521 char kbuf[FSL_SIZE]; /* buffer to hold fast symlink */
2522 int tflag = 0; /* flag to indicate temp vars used */
2523
2524 ino = ip->i_number;
2525 igen = ip->i_gen;
2526 size = uiop->uio_resid;
2527 basep = uiop->uio_iov->iov_base;
2528 tuiop = uiop;
2529
2530 rw_enter(&ip->i_rwlock, RW_WRITER);
2531 rw_enter(&ip->i_contents, RW_WRITER);
2532 if (ip->i_flag & IFASTSYMLNK) {
2533 rw_exit(&ip->i_contents);
2534 rw_exit(&ip->i_rwlock);
2535 goto again;
2536 }
2537
2538 /* can this be a fast symlink and is it a user buffer? */
2539 if (ip->i_size <= FSL_SIZE &&
2540 (uiop->uio_segflg == UIO_USERSPACE ||
2541 uiop->uio_segflg == UIO_USERISPACE)) {
2542
2543 bzero(&tuio, sizeof (struct uio));
2544 /*
2545 * setup a kernel buffer to read link into. this
2546 * is to fix a race condition where the user buffer
2547 * got corrupted before copying it into the inode.
2548 */
2549 size = ip->i_size;
2550 tiov.iov_len = size;
2551 tiov.iov_base = kbuf;
2552 tuio.uio_iov = &tiov;
2553 tuio.uio_iovcnt = 1;
2554 tuio.uio_offset = uiop->uio_offset;
2555 tuio.uio_segflg = UIO_SYSSPACE;
2556 tuio.uio_fmode = uiop->uio_fmode;
2557 tuio.uio_extflg = uiop->uio_extflg;
2558 tuio.uio_limit = uiop->uio_limit;
2559 tuio.uio_resid = size;
2560
2561 basep = tuio.uio_iov->iov_base;
2562 tuiop = &tuio;
2563 tflag = 1;
2564 }
2565
2566 error = rdip(ip, tuiop, 0, cr);
2567 if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2568 rw_exit(&ip->i_contents);
2569 rw_exit(&ip->i_rwlock);
2570 goto out;
2571 }
2572
2573 if (tflag == 0)
2574 size -= uiop->uio_resid;
2575
2576 if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2577 ip->i_size == size) || (tflag == 1 &&
2578 tuio.uio_resid == 0)) {
2579 error = kcopy(basep, &ip->i_db[1], ip->i_size);
2580 if (error == 0) {
2581 ip->i_flag |= IFASTSYMLNK;
2582 /*
2583 * free page
2584 */
2585 (void) VOP_PUTPAGE(ITOV(ip),
2586 (offset_t)0, PAGESIZE,
2587 (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2588 cr, ct);
2589 } else {
2590 int i;
2591 /* error, clear garbage left behind */
2592 for (i = 1; i < NDADDR; i++)
2593 ip->i_db[i] = 0;
2594 for (i = 0; i < NIADDR; i++)
2595 ip->i_ib[i] = 0;
2596 }
2597 }
2598 if (tflag == 1) {
2599 /* now, copy it into the user buffer */
2600 error = uiomove((caddr_t)kbuf,
2601 MIN(size, uiop->uio_resid),
2602 UIO_READ, uiop);
2603 }
2604 rw_exit(&ip->i_contents);
2605 rw_exit(&ip->i_rwlock);
2606 }
2607 out:
2608 if (ulp) {
2609 ufs_lockfs_end(ulp);
2610 }
2611 nolockout:
2612 return (error);
2613 }
2614
2615 /* ARGSUSED */
2616 static int
2617 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr, caller_context_t *ct)
2618 {
2619 struct inode *ip = VTOI(vp);
2620 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2621 struct ulockfs *ulp;
2622 int error;
2623
2624 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2625 if (error)
2626 return (error);
2627
2628 if (TRANS_ISTRANS(ufsvfsp)) {
2629 /*
2630 * First push out any data pages
2631 */
2632 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2633 (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2634 error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2635 0, CRED(), ct);
2636 if (error)
2637 goto out;
2638 }
2639
2640 /*
2641 * Delta any delayed inode times updates
2642 * and push inode to log.
2643 * All other inode deltas will have already been delta'd
2644 * and will be pushed during the commit.
2645 */
2646 if (!(syncflag & FDSYNC) &&
2647 ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2648 if (ulp) {
2649 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2650 TOP_SYNCIP_SIZE);
2651 }
2652 rw_enter(&ip->i_contents, RW_READER);
2653 mutex_enter(&ip->i_tlock);
2654 ip->i_flag &= ~IMODTIME;
2655 mutex_exit(&ip->i_tlock);
2656 ufs_iupdat(ip, I_SYNC);
2657 rw_exit(&ip->i_contents);
2658 if (ulp) {
2659 TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2660 TOP_SYNCIP_SIZE);
2661 }
2662 }
2663
2664 /*
2665 * Commit the Moby transaction
2666 *
2667 * Deltas have already been made so we just need to
2668 * commit them with a synchronous transaction.
2669 * TRANS_BEGIN_SYNC() will return an error
2670 * if there are no deltas to commit, for an
2671 * empty transaction.
2672 */
2673 if (ulp) {
2674 TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2675 error);
2676 if (error) {
2677 error = 0; /* commit wasn't needed */
2678 goto out;
2679 }
2680 TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2681 TOP_COMMIT_SIZE);
2682 }
2683 } else { /* not logging */
2684 if (!(IS_SWAPVP(vp)))
2685 if (syncflag & FNODSYNC) {
2686 /* Just update the inode only */
2687 TRANS_IUPDAT(ip, 1);
2688 error = 0;
2689 } else if (syncflag & FDSYNC)
2690 /* Do data-synchronous writes */
2691 error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2692 else
2693 /* Do synchronous writes */
2694 error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2695
2696 rw_enter(&ip->i_contents, RW_WRITER);
2697 if (!error)
2698 error = ufs_sync_indir(ip);
2699 rw_exit(&ip->i_contents);
2700 }
2701 out:
2702 if (ulp) {
2703 ufs_lockfs_end(ulp);
2704 }
2705 return (error);
2706 }
2707
2708 /*ARGSUSED*/
2709 static void
2710 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
2711 {
2712 ufs_iinactive(VTOI(vp));
2713 }
2714
2715 /*
2716 * Unix file system operations having to do with directory manipulation.
2717 */
2718 int ufs_lookup_idle_count = 2; /* Number of inodes to idle each time */
2719 /* ARGSUSED */
2720 static int
2721 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2722 struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr,
2723 caller_context_t *ct, int *direntflags, pathname_t *realpnp)
2724 {
2725 struct inode *ip;
2726 struct inode *sip;
2727 struct inode *xip;
2728 struct ufsvfs *ufsvfsp;
2729 struct ulockfs *ulp;
2730 struct vnode *vp;
2731 int error;
2732
2733 /*
2734 * Check flags for type of lookup (regular file or attribute file)
2735 */
2736
2737 ip = VTOI(dvp);
2738
2739 if (flags & LOOKUP_XATTR) {
2740
2741 /*
2742 * If not mounted with XATTR support then return EINVAL
2743 */
2744
2745 if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR))
2746 return (EINVAL);
2747 /*
2748 * We don't allow recursive attributes...
2749 * Maybe someday we will.
2750 */
2751 if ((ip->i_cflags & IXATTR)) {
2752 return (EINVAL);
2753 }
2754
2755 if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2756 error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2757 if (error) {
2758 *vpp = NULL;
2759 goto out;
2760 }
2761
2762 vp = ITOV(sip);
2763 dnlc_update(dvp, XATTR_DIR_NAME, vp);
2764 }
2765
2766 /*
2767 * Check accessibility of directory.
2768 */
2769 if (vp == DNLC_NO_VNODE) {
2770 VN_RELE(vp);
2771 error = ENOENT;
2772 goto out;
2773 }
2774 if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr, 1)) != 0) {
2775 VN_RELE(vp);
2776 goto out;
2777 }
2778
2779 *vpp = vp;
2780 return (0);
2781 }
2782
2783 /*
2784 * Check for a null component, which we should treat as
2785 * looking at dvp from within it's parent, so we don't
2786 * need a call to ufs_iaccess(), as it has already been
2787 * done.
2788 */
2789 if (nm[0] == 0) {
2790 VN_HOLD(dvp);
2791 error = 0;
2792 *vpp = dvp;
2793 goto out;
2794 }
2795
2796 /*
2797 * Check for "." ie itself. this is a quick check and
2798 * avoids adding "." into the dnlc (which have been seen
2799 * to occupy >10% of the cache).
2800 */
2801 if ((nm[0] == '.') && (nm[1] == 0)) {
2802 /*
2803 * Don't return without checking accessibility
2804 * of the directory. We only need the lock if
2805 * we are going to return it.
2806 */
2807 if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) == 0) {
2808 VN_HOLD(dvp);
2809 *vpp = dvp;
2810 }
2811 goto out;
2812 }
2813
2814 /*
2815 * Fast path: Check the directory name lookup cache.
2816 */
2817 if (vp = dnlc_lookup(dvp, nm)) {
2818 /*
2819 * Check accessibility of directory.
2820 */
2821 if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) != 0) {
2822 VN_RELE(vp);
2823 goto out;
2824 }
2825 if (vp == DNLC_NO_VNODE) {
2826 VN_RELE(vp);
2827 error = ENOENT;
2828 goto out;
2829 }
2830 xip = VTOI(vp);
2831 ulp = NULL;
2832 goto fastpath;
2833 }
2834
2835 /*
2836 * Keep the idle queue from getting too long by
2837 * idling two inodes before attempting to allocate another.
2838 * This operation must be performed before entering
2839 * lockfs or a transaction.
2840 */
2841 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2842 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2843 ins.in_lidles.value.ul += ufs_lookup_idle_count;
2844 ufs_idle_some(ufs_lookup_idle_count);
2845 }
2846
2847 retry_lookup:
2848 /*
2849 * Check accessibility of directory.
2850 */
2851 if (error = ufs_diraccess(ip, IEXEC, cr))
2852 goto out;
2853
2854 ufsvfsp = ip->i_ufsvfs;
2855 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2856 if (error)
2857 goto out;
2858
2859 error = ufs_dirlook(ip, nm, &xip, cr, 1, 0);
2860
2861 fastpath:
2862 if (error == 0) {
2863 ip = xip;
2864 *vpp = ITOV(ip);
2865
2866 /*
2867 * If vnode is a device return special vnode instead.
2868 */
2869 if (IS_DEVVP(*vpp)) {
2870 struct vnode *newvp;
2871
2872 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2873 cr);
2874 VN_RELE(*vpp);
2875 if (newvp == NULL)
2876 error = ENOSYS;
2877 else
2878 *vpp = newvp;
2879 } else if (ip->i_cflags & ICOMPRESS) {
2880 struct vnode *newvp;
2881
2882 /*
2883 * Compressed file, substitute dcfs vnode
2884 */
2885 newvp = decompvp(*vpp, cr, ct);
2886 VN_RELE(*vpp);
2887 if (newvp == NULL)
2888 error = ENOSYS;
2889 else
2890 *vpp = newvp;
2891 }
2892 }
2893 if (ulp) {
2894 ufs_lockfs_end(ulp);
2895 }
2896
2897 if (error == EAGAIN)
2898 goto retry_lookup;
2899
2900 out:
2901 return (error);
2902 }
2903
2904 /*ARGSUSED*/
2905 static int
2906 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2907 int mode, struct vnode **vpp, struct cred *cr, int flag,
2908 caller_context_t *ct, vsecattr_t *vsecp)
2909 {
2910 struct inode *ip;
2911 struct inode *xip;
2912 struct inode *dip;
2913 struct vnode *xvp;
2914 struct ufsvfs *ufsvfsp;
2915 struct ulockfs *ulp;
2916 int error;
2917 int issync;
2918 int truncflag;
2919 int trans_size;
2920 int noentry;
2921 int defer_dip_seq_update = 0; /* need to defer update of dip->i_seq */
2922 int retry = 1;
2923 int indeadlock;
2924
2925 again:
2926 ip = VTOI(dvp);
2927 ufsvfsp = ip->i_ufsvfs;
2928 truncflag = 0;
2929
2930 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2931 if (error)
2932 goto out;
2933
2934 if (ulp) {
2935 trans_size = (int)TOP_CREATE_SIZE(ip);
2936 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2937 }
2938
2939 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2940 vap->va_mode &= ~VSVTX;
2941
2942 if (*name == '\0') {
2943 /*
2944 * Null component name refers to the directory itself.
2945 */
2946 VN_HOLD(dvp);
2947 /*
2948 * Even though this is an error case, we need to grab the
2949 * quota lock since the error handling code below is common.
2950 */
2951 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2952 rw_enter(&ip->i_contents, RW_WRITER);
2953 error = EEXIST;
2954 } else {
2955 xip = NULL;
2956 noentry = 0;
2957 /*
2958 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2959 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2960 * possible, retries the operation.
2961 */
2962 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2963 retry_dir);
2964 if (indeadlock)
2965 goto again;
2966
2967 xvp = dnlc_lookup(dvp, name);
2968 if (xvp == DNLC_NO_VNODE) {
2969 noentry = 1;
2970 VN_RELE(xvp);
2971 xvp = NULL;
2972 }
2973 if (xvp) {
2974 rw_exit(&ip->i_rwlock);
2975 if (error = ufs_iaccess(ip, IEXEC, cr, 1)) {
2976 VN_RELE(xvp);
2977 } else {
2978 error = EEXIST;
2979 xip = VTOI(xvp);
2980 }
2981 } else {
2982 /*
2983 * Suppress file system full message if we will retry
2984 */
2985 error = ufs_direnter_cm(ip, name, DE_CREATE,
2986 vap, &xip, cr, (noentry | (retry ? IQUIET : 0)));
2987 if (error == EAGAIN) {
2988 if (ulp) {
2989 TRANS_END_CSYNC(ufsvfsp, error, issync,
2990 TOP_CREATE, trans_size);
2991 ufs_lockfs_end(ulp);
2992 }
2993 goto again;
2994 }
2995 rw_exit(&ip->i_rwlock);
2996 }
2997 ip = xip;
2998 if (ip != NULL) {
2999 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
3000 rw_enter(&ip->i_contents, RW_WRITER);
3001 }
3002 }
3003
3004 /*
3005 * If the file already exists and this is a non-exclusive create,
3006 * check permissions and allow access for non-directories.
3007 * Read-only create of an existing directory is also allowed.
3008 * We fail an exclusive create of anything which already exists.
3009 */
3010 if (error == EEXIST) {
3011 dip = VTOI(dvp);
3012 if (excl == NONEXCL) {
3013 if ((((ip->i_mode & IFMT) == IFDIR) ||
3014 ((ip->i_mode & IFMT) == IFATTRDIR)) &&
3015 (mode & IWRITE))
3016 error = EISDIR;
3017 else if (mode)
3018 error = ufs_iaccess(ip, mode, cr, 0);
3019 else
3020 error = 0;
3021 }
3022 if (error) {
3023 rw_exit(&ip->i_contents);
3024 rw_exit(&ufsvfsp->vfs_dqrwlock);
3025 VN_RELE(ITOV(ip));
3026 goto unlock;
3027 }
3028 /*
3029 * If the error EEXIST was set, then i_seq can not
3030 * have been updated. The sequence number interface
3031 * is defined such that a non-error VOP_CREATE must
3032 * increase the dir va_seq it by at least one. If we
3033 * have cleared the error, increase i_seq. Note that
3034 * we are increasing the dir i_seq and in rare cases
3035 * ip may actually be from the dvp, so we already have
3036 * the locks and it will not be subject to truncation.
3037 * In case we have to update i_seq of the parent
3038 * directory dip, we have to defer it till we have
3039 * released our locks on ip due to lock ordering requirements.
3040 */
3041 if (ip != dip)
3042 defer_dip_seq_update = 1;
3043 else
3044 ip->i_seq++;
3045
3046 if (((ip->i_mode & IFMT) == IFREG) &&
3047 (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3048 /*
3049 * Truncate regular files, if requested by caller.
3050 * Grab i_rwlock to make sure no one else is
3051 * currently writing to the file (we promised
3052 * bmap we would do this).
3053 * Must get the locks in the correct order.
3054 */
3055 if (ip->i_size == 0) {
3056 ip->i_flag |= ICHG | IUPD;
3057 ip->i_seq++;
3058 TRANS_INODE(ufsvfsp, ip);
3059 } else {
3060 /*
3061 * Large Files: Why this check here?
3062 * Though we do it in vn_create() we really
3063 * want to guarantee that we do not destroy
3064 * Large file data by atomically checking
3065 * the size while holding the contents
3066 * lock.
3067 */
3068 if (flag && !(flag & FOFFMAX) &&
3069 ((ip->i_mode & IFMT) == IFREG) &&
3070 (ip->i_size > (offset_t)MAXOFF32_T)) {
3071 rw_exit(&ip->i_contents);
3072 rw_exit(&ufsvfsp->vfs_dqrwlock);
3073 error = EOVERFLOW;
3074 goto unlock;
3075 }
3076 if (TRANS_ISTRANS(ufsvfsp))
3077 truncflag++;
3078 else {
3079 rw_exit(&ip->i_contents);
3080 rw_exit(&ufsvfsp->vfs_dqrwlock);
3081 ufs_tryirwlock_trans(&ip->i_rwlock,
3082 RW_WRITER, TOP_CREATE,
3083 retry_file);
3084 if (indeadlock) {
3085 VN_RELE(ITOV(ip));
3086 goto again;
3087 }
3088 rw_enter(&ufsvfsp->vfs_dqrwlock,
3089 RW_READER);
3090 rw_enter(&ip->i_contents, RW_WRITER);
3091 (void) ufs_itrunc(ip, (u_offset_t)0, 0,
3092 cr);
3093 rw_exit(&ip->i_rwlock);
3094 }
3095
3096 }
3097 if (error == 0) {
3098 vnevent_create(ITOV(ip), ct);
3099 }
3100 }
3101 }
3102
3103 if (error) {
3104 if (ip != NULL) {
3105 rw_exit(&ufsvfsp->vfs_dqrwlock);
3106 rw_exit(&ip->i_contents);
3107 }
3108 goto unlock;
3109 }
3110
3111 *vpp = ITOV(ip);
3112 ITIMES(ip);
3113 rw_exit(&ip->i_contents);
3114 rw_exit(&ufsvfsp->vfs_dqrwlock);
3115
3116 /*
3117 * If vnode is a device return special vnode instead.
3118 */
3119 if (!error && IS_DEVVP(*vpp)) {
3120 struct vnode *newvp;
3121
3122 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3123 VN_RELE(*vpp);
3124 if (newvp == NULL) {
3125 error = ENOSYS;
3126 goto unlock;
3127 }
3128 truncflag = 0;
3129 *vpp = newvp;
3130 }
3131 unlock:
3132
3133 /*
3134 * Do the deferred update of the parent directory's sequence
3135 * number now.
3136 */
3137 if (defer_dip_seq_update == 1) {
3138 rw_enter(&dip->i_contents, RW_READER);
3139 mutex_enter(&dip->i_tlock);
3140 dip->i_seq++;
3141 mutex_exit(&dip->i_tlock);
3142 rw_exit(&dip->i_contents);
3143 }
3144
3145 if (ulp) {
3146 int terr = 0;
3147
3148 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3149 trans_size);
3150
3151 /*
3152 * If we haven't had a more interesting failure
3153 * already, then anything that might've happened
3154 * here should be reported.
3155 */
3156 if (error == 0)
3157 error = terr;
3158 }
3159
3160 if (!error && truncflag) {
3161 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3162 if (indeadlock) {
3163 if (ulp)
3164 ufs_lockfs_end(ulp);
3165 VN_RELE(ITOV(ip));
3166 goto again;
3167 }
3168 (void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3169 rw_exit(&ip->i_rwlock);
3170 }
3171
3172 if (ulp)
3173 ufs_lockfs_end(ulp);
3174
3175 /*
3176 * If no inodes available, try to free one up out of the
3177 * pending delete queue.
3178 */
3179 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3180 ufs_delete_drain_wait(ufsvfsp, 1);
3181 retry = 0;
3182 goto again;
3183 }
3184
3185 out:
3186 return (error);
3187 }
3188
3189 extern int ufs_idle_max;
3190 /*ARGSUSED*/
3191 static int
3192 ufs_remove(struct vnode *vp, char *nm, struct cred *cr, caller_context_t *ct,
3193 int flags)
3194 {
3195 struct inode *ip = VTOI(vp);
3196 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3197 struct ulockfs *ulp;
3198 vnode_t *rmvp = NULL; /* Vnode corresponding to name being removed */
3199 int indeadlock;
3200 int error;
3201 int issync;
3202 int trans_size;
3203
3204 /*
3205 * don't let the delete queue get too long
3206 */
3207 if (ufsvfsp == NULL) {
3208 error = EIO;
3209 goto out;
3210 }
3211 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3212 ufs_delete_drain(vp->v_vfsp, 1, 1);
3213
3214 error = ufs_eventlookup(vp, nm, cr, &rmvp);
3215 if (rmvp != NULL) {
3216 /* Only send the event if there were no errors */
3217 if (error == 0)
3218 vnevent_remove(rmvp, vp, nm, ct);
3219 VN_RELE(rmvp);
3220 }
3221
3222 retry_remove:
3223 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3224 if (error)
3225 goto out;
3226
3227 if (ulp)
3228 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3229 trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3230
3231 /*
3232 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3233 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3234 * possible, retries the operation.
3235 */
3236 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3237 if (indeadlock)
3238 goto retry_remove;
3239 error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3240 DR_REMOVE, cr);
3241 rw_exit(&ip->i_rwlock);
3242
3243 if (ulp) {
3244 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3245 ufs_lockfs_end(ulp);
3246 }
3247
3248 out:
3249 return (error);
3250 }
3251
3252 /*
3253 * Link a file or a directory. Only privileged processes are allowed to
3254 * make links to directories.
3255 */
3256 /*ARGSUSED*/
3257 static int
3258 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr,
3259 caller_context_t *ct, int flags)
3260 {
3261 struct inode *sip;
3262 struct inode *tdp = VTOI(tdvp);
3263 struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3264 struct ulockfs *ulp;
3265 struct vnode *realvp;
3266 int error;
3267 int issync;
3268 int trans_size;
3269 int isdev;
3270 int indeadlock;
3271
3272 retry_link:
3273 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3274 if (error)
3275 goto out;
3276
3277 if (ulp)
3278 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3279 trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3280
3281 if (VOP_REALVP(svp, &realvp, ct) == 0)
3282 svp = realvp;
3283
3284 /*
3285 * Make sure link for extended attributes is valid
3286 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3287 *
3288 * Make certain we don't attempt to look at a device node as
3289 * a ufs inode.
3290 */
3291
3292 isdev = IS_DEVVP(svp);
3293 if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3294 ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3295 ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3296 ((tdp->i_mode & IFMT) == IFDIR))) {
3297 error = EINVAL;
3298 goto unlock;
3299 }
3300
3301 sip = VTOI(svp);
3302 if ((svp->v_type == VDIR &&
3303 secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3304 (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3305 error = EPERM;
3306 goto unlock;
3307 }
3308
3309 /*
3310 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3311 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3312 * possible, retries the operation.
3313 */
3314 ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3315 if (indeadlock)
3316 goto retry_link;
3317 error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3318 sip, cr);
3319 rw_exit(&tdp->i_rwlock);
3320
3321 unlock:
3322 if (ulp) {
3323 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3324 ufs_lockfs_end(ulp);
3325 }
3326
3327 if (!error) {
3328 vnevent_link(svp, ct);
3329 }
3330 out:
3331 return (error);
3332 }
3333
3334 uint64_t ufs_rename_retry_cnt;
3335 uint64_t ufs_rename_upgrade_retry_cnt;
3336 uint64_t ufs_rename_dircheck_retry_cnt;
3337 clock_t ufs_rename_backoff_delay = 1;
3338
3339 /*
3340 * Rename a file or directory.
3341 * We are given the vnode and entry string of the source and the
3342 * vnode and entry string of the place we want to move the source
3343 * to (the target). The essential operation is:
3344 * unlink(target);
3345 * link(source, target);
3346 * unlink(source);
3347 * but "atomically". Can't do full commit without saving state in
3348 * the inode on disk, which isn't feasible at this time. Best we
3349 * can do is always guarantee that the TARGET exists.
3350 */
3351
3352 /*ARGSUSED*/
3353 static int
3354 ufs_rename(struct vnode *sdvp, char *snm, struct vnode *tdvp, char *tnm,
3355 struct cred *cr, caller_context_t *ct, int flags)
3356 {
3357 struct inode *sip = NULL; /* source inode */
3358 struct inode *ip = NULL; /* check inode */
3359 struct inode *sdp; /* old (source) parent inode */
3360 struct inode *tdp; /* new (target) parent inode */
3361 struct vnode *svp = NULL; /* source vnode */
3362 struct vnode *tvp = NULL; /* target vnode, if it exists */
3363 struct vnode *realvp;
3364 struct ufsvfs *ufsvfsp;
3365 struct ulockfs *ulp = NULL;
3366 struct ufs_slot slot;
3367 timestruc_t now;
3368 int error;
3369 int issync;
3370 int trans_size;
3371 krwlock_t *first_lock;
3372 krwlock_t *second_lock;
3373 krwlock_t *reverse_lock;
3374 int serr, terr;
3375
3376 sdp = VTOI(sdvp);
3377 slot.fbp = NULL;
3378 ufsvfsp = sdp->i_ufsvfs;
3379
3380 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3381 tdvp = realvp;
3382
3383 /* Must do this before taking locks in case of DNLC miss */
3384 terr = ufs_eventlookup(tdvp, tnm, cr, &tvp);
3385 serr = ufs_eventlookup(sdvp, snm, cr, &svp);
3386
3387 if ((serr == 0) && ((terr == 0) || (terr == ENOENT))) {
3388 if (tvp != NULL)
3389 vnevent_pre_rename_dest(tvp, tdvp, tnm, ct);
3390
3391 /*
3392 * Notify the target directory of the rename event
3393 * if source and target directories are not the same.
3394 */
3395 if (sdvp != tdvp)
3396 vnevent_pre_rename_dest_dir(tdvp, svp, tnm, ct);
3397
3398 if (svp != NULL)
3399 vnevent_pre_rename_src(svp, sdvp, snm, ct);
3400 }
3401
3402 if (svp != NULL)
3403 VN_RELE(svp);
3404
3405 retry_rename:
3406 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3407 if (error)
3408 goto unlock;
3409
3410 if (ulp)
3411 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3412 trans_size = (int)TOP_RENAME_SIZE(sdp));
3413
3414 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3415 tdvp = realvp;
3416
3417 tdp = VTOI(tdvp);
3418
3419 /*
3420 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3421 */
3422 if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3423 error = EINVAL;
3424 goto unlock;
3425 }
3426
3427 /*
3428 * Check accessibility of directory.
3429 */
3430 if (error = ufs_diraccess(sdp, IEXEC, cr))
3431 goto unlock;
3432
3433 /*
3434 * Look up inode of file we're supposed to rename.
3435 */
3436 gethrestime(&now);
3437 if (error = ufs_dirlook(sdp, snm, &sip, cr, 0, 0)) {
3438 if (error == EAGAIN) {
3439 if (ulp) {
3440 TRANS_END_CSYNC(ufsvfsp, error, issync,
3441 TOP_RENAME, trans_size);
3442 ufs_lockfs_end(ulp);
3443 }
3444 goto retry_rename;
3445 }
3446
3447 goto unlock;
3448 }
3449
3450 /*
3451 * Lock both the source and target directories (they may be
3452 * the same) to provide the atomicity semantics that was
3453 * previously provided by the per file system vfs_rename_lock
3454 *
3455 * with vfs_rename_lock removed to allow simultaneous renames
3456 * within a file system, ufs_dircheckpath can deadlock while
3457 * traversing back to ensure that source is not a parent directory
3458 * of target parent directory. This is because we get into
3459 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3460 * If the tdp and sdp of the simultaneous renames happen to be
3461 * in the path of each other, it can lead to a deadlock. This
3462 * can be avoided by getting the locks as RW_READER here and then
3463 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3464 *
3465 * We hold the target directory's i_rwlock after calling
3466 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3467 * VOP_RWLOCK is explicitly called by the filesystem independent code
3468 * before calling the file system operation. In these cases the order
3469 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3470 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3471 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3472 * synchronizing object which might lead to a deadlock. So we use
3473 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3474 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3475 * operation.
3476 */
3477 retry:
3478 first_lock = &tdp->i_rwlock;
3479 second_lock = &sdp->i_rwlock;
3480 retry_firstlock:
3481 if (!rw_tryenter(first_lock, RW_READER)) {
3482 /*
3483 * We didn't get the lock. Check if the SLOCK is set in the
3484 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3485 * and wait for SLOCK to be cleared.
3486 */
3487
3488 if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3489 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3490 trans_size);
3491 ufs_lockfs_end(ulp);
3492 goto retry_rename;
3493
3494 } else {
3495 /*
3496 * SLOCK isn't set so this is a genuine synchronization
3497 * case. Let's try again after giving them a breather.
3498 */
3499 delay(RETRY_LOCK_DELAY);
3500 goto retry_firstlock;
3501 }
3502 }
3503 /*
3504 * Need to check if the tdp and sdp are same !!!
3505 */
3506 if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3507 /*
3508 * We didn't get the lock. Check if the SLOCK is set in the
3509 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3510 * and wait for SLOCK to be cleared.
3511 */
3512
3513 rw_exit(first_lock);
3514 if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3515 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3516 trans_size);
3517 ufs_lockfs_end(ulp);
3518 goto retry_rename;
3519
3520 } else {
3521 /*
3522 * So we couldn't get the second level peer lock *and*
3523 * the SLOCK bit isn't set. Too bad we can be
3524 * contentding with someone wanting these locks otherway
3525 * round. Reverse the locks in case there is a heavy
3526 * contention for the second level lock.
3527 */
3528 reverse_lock = first_lock;
3529 first_lock = second_lock;
3530 second_lock = reverse_lock;
3531 ufs_rename_retry_cnt++;
3532 goto retry_firstlock;
3533 }
3534 }
3535
3536 if (sip == tdp) {
3537 error = EINVAL;
3538 goto errout;
3539 }
3540 /*
3541 * Make sure we can delete the source entry. This requires
3542 * write permission on the containing directory.
3543 * Check for sticky directories.
3544 */
3545 rw_enter(&sdp->i_contents, RW_READER);
3546 rw_enter(&sip->i_contents, RW_READER);
3547 if ((error = ufs_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
3548 (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3549 rw_exit(&sip->i_contents);
3550 rw_exit(&sdp->i_contents);
3551 goto errout;
3552 }
3553
3554 /*
3555 * If this is a rename of a directory and the parent is
3556 * different (".." must be changed), then the source
3557 * directory must not be in the directory hierarchy
3558 * above the target, as this would orphan everything
3559 * below the source directory. Also the user must have
3560 * write permission in the source so as to be able to
3561 * change "..".
3562 */
3563 if ((((sip->i_mode & IFMT) == IFDIR) ||
3564 ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3565 ino_t inum;
3566
3567 if (error = ufs_iaccess(sip, IWRITE, cr, 0)) {
3568 rw_exit(&sip->i_contents);
3569 rw_exit(&sdp->i_contents);
3570 goto errout;
3571 }
3572 inum = sip->i_number;
3573 rw_exit(&sip->i_contents);
3574 rw_exit(&sdp->i_contents);
3575 if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3576 /*
3577 * If we got EAGAIN ufs_dircheckpath detected a
3578 * potential deadlock and backed out. We need
3579 * to retry the operation since sdp and tdp have
3580 * to be released to avoid the deadlock.
3581 */
3582 if (error == EAGAIN) {
3583 rw_exit(&tdp->i_rwlock);
3584 if (tdp != sdp)
3585 rw_exit(&sdp->i_rwlock);
3586 delay(ufs_rename_backoff_delay);
3587 ufs_rename_dircheck_retry_cnt++;
3588 goto retry;
3589 }
3590 goto errout;
3591 }
3592 } else {
3593 rw_exit(&sip->i_contents);
3594 rw_exit(&sdp->i_contents);
3595 }
3596
3597
3598 /*
3599 * Check for renaming '.' or '..' or alias of '.'
3600 */
3601 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3602 error = EINVAL;
3603 goto errout;
3604 }
3605
3606 /*
3607 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3608 * tries to traverse back the file tree with both tdp and sdp held
3609 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3610 * as RW_READERS till ufs_dircheckpath is done.
3611 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3612 * to RW_WRITER.
3613 */
3614 if (!rw_tryupgrade(&tdp->i_rwlock)) {
3615 /*
3616 * The upgrade failed. We got to give away the lock
3617 * as to avoid deadlocking with someone else who is
3618 * waiting for writer lock. With the lock gone, we
3619 * cannot be sure the checks done above will hold
3620 * good when we eventually get them back as writer.
3621 * So if we can't upgrade we drop the locks and retry
3622 * everything again.
3623 */
3624 rw_exit(&tdp->i_rwlock);
3625 if (tdp != sdp)
3626 rw_exit(&sdp->i_rwlock);
3627 delay(ufs_rename_backoff_delay);
3628 ufs_rename_upgrade_retry_cnt++;
3629 goto retry;
3630 }
3631 if (tdp != sdp) {
3632 if (!rw_tryupgrade(&sdp->i_rwlock)) {
3633 /*
3634 * The upgrade failed. We got to give away the lock
3635 * as to avoid deadlocking with someone else who is
3636 * waiting for writer lock. With the lock gone, we
3637 * cannot be sure the checks done above will hold
3638 * good when we eventually get them back as writer.
3639 * So if we can't upgrade we drop the locks and retry
3640 * everything again.
3641 */
3642 rw_exit(&tdp->i_rwlock);
3643 rw_exit(&sdp->i_rwlock);
3644 delay(ufs_rename_backoff_delay);
3645 ufs_rename_upgrade_retry_cnt++;
3646 goto retry;
3647 }
3648 }
3649
3650 /*
3651 * Now that all the locks are held check to make sure another thread
3652 * didn't slip in and take out the sip.
3653 */
3654 slot.status = NONE;
3655 if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3656 sip->i_ctime.tv_sec > now.tv_sec) {
3657 rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3658 rw_enter(&sdp->i_contents, RW_WRITER);
3659 error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3660 &ip, cr, 0);
3661 rw_exit(&sdp->i_contents);
3662 rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3663 if (error) {
3664 goto errout;
3665 }
3666 if (ip == NULL) {
3667 error = ENOENT;
3668 goto errout;
3669 } else {
3670 /*
3671 * If the inode was found need to drop the v_count
3672 * so as not to keep the filesystem from being
3673 * unmounted at a later time.
3674 */
3675 VN_RELE(ITOV(ip));
3676 }
3677
3678 /*
3679 * Release the slot.fbp that has the page mapped and
3680 * locked SE_SHARED, and could be used in in
3681 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3682 * on said page.
3683 */
3684 if (slot.fbp) {
3685 fbrelse(slot.fbp, S_OTHER);
3686 slot.fbp = NULL;
3687 }
3688 }
3689
3690 /*
3691 * Link source to the target.
3692 */
3693 if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr)) {
3694 /*
3695 * ESAME isn't really an error; it indicates that the
3696 * operation should not be done because the source and target
3697 * are the same file, but that no error should be reported.
3698 */
3699 if (error == ESAME)
3700 error = 0;
3701 goto errout;
3702 }
3703
3704 if (error == 0 && tvp != NULL)
3705 vnevent_rename_dest(tvp, tdvp, tnm, ct);
3706
3707 /*
3708 * Unlink the source.
3709 * Remove the source entry. ufs_dirremove() checks that the entry
3710 * still reflects sip, and returns an error if it doesn't.
3711 * If the entry has changed just forget about it. Release
3712 * the source inode.
3713 */
3714 if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3715 DR_RENAME, cr)) == ENOENT)
3716 error = 0;
3717
3718 if (error == 0) {
3719 vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
3720 /*
3721 * Notify the target directory of the rename event
3722 * if source and target directories are not the same.
3723 */
3724 if (sdvp != tdvp)
3725 vnevent_rename_dest_dir(tdvp, ct);
3726 }
3727
3728 errout:
3729 if (slot.fbp)
3730 fbrelse(slot.fbp, S_OTHER);
3731
3732 rw_exit(&tdp->i_rwlock);
3733 if (sdp != tdp) {
3734 rw_exit(&sdp->i_rwlock);
3735 }
3736
3737 unlock:
3738 if (tvp != NULL)
3739 VN_RELE(tvp);
3740 if (sip != NULL)
3741 VN_RELE(ITOV(sip));
3742
3743 if (ulp) {
3744 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3745 ufs_lockfs_end(ulp);
3746 }
3747
3748 return (error);
3749 }
3750
3751 /*ARGSUSED*/
3752 static int
3753 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3754 struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags,
3755 vsecattr_t *vsecp)
3756 {
3757 struct inode *ip;
3758 struct inode *xip;
3759 struct ufsvfs *ufsvfsp;
3760 struct ulockfs *ulp;
3761 int error;
3762 int issync;
3763 int trans_size;
3764 int indeadlock;
3765 int retry = 1;
3766
3767 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3768
3769 /*
3770 * Can't make directory in attr hidden dir
3771 */
3772 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3773 return (EINVAL);
3774
3775 again:
3776 ip = VTOI(dvp);
3777 ufsvfsp = ip->i_ufsvfs;
3778 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3779 if (error)
3780 goto out;
3781 if (ulp)
3782 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3783 trans_size = (int)TOP_MKDIR_SIZE(ip));
3784
3785 /*
3786 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3787 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3788 * possible, retries the operation.
3789 */
3790 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3791 if (indeadlock)
3792 goto again;
3793
3794 error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3795 (retry ? IQUIET : 0));
3796 if (error == EAGAIN) {
3797 if (ulp) {
3798 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3799 trans_size);
3800 ufs_lockfs_end(ulp);
3801 }
3802 goto again;
3803 }
3804
3805 rw_exit(&ip->i_rwlock);
3806 if (error == 0) {
3807 ip = xip;
3808 *vpp = ITOV(ip);
3809 } else if (error == EEXIST)
3810 VN_RELE(ITOV(xip));
3811
3812 if (ulp) {
3813 int terr = 0;
3814 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3815 ufs_lockfs_end(ulp);
3816 if (error == 0)
3817 error = terr;
3818 }
3819 out:
3820 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3821 ufs_delete_drain_wait(ufsvfsp, 1);
3822 retry = 0;
3823 goto again;
3824 }
3825
3826 return (error);
3827 }
3828
3829 /*ARGSUSED*/
3830 static int
3831 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr,
3832 caller_context_t *ct, int flags)
3833 {
3834 struct inode *ip = VTOI(vp);
3835 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3836 struct ulockfs *ulp;
3837 vnode_t *rmvp = NULL; /* Vnode of removed directory */
3838 int error;
3839 int issync;
3840 int trans_size;
3841 int indeadlock;
3842
3843 /*
3844 * don't let the delete queue get too long
3845 */
3846 if (ufsvfsp == NULL) {
3847 error = EIO;
3848 goto out;
3849 }
3850 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3851 ufs_delete_drain(vp->v_vfsp, 1, 1);
3852
3853 error = ufs_eventlookup(vp, nm, cr, &rmvp);
3854 if (rmvp != NULL) {
3855 /* Only send the event if there were no errors */
3856 if (error == 0)
3857 vnevent_rmdir(rmvp, vp, nm, ct);
3858 VN_RELE(rmvp);
3859 }
3860
3861 retry_rmdir:
3862 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3863 if (error)
3864 goto out;
3865
3866 if (ulp)
3867 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3868 trans_size = TOP_RMDIR_SIZE);
3869
3870 /*
3871 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3872 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3873 * possible, retries the operation.
3874 */
3875 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3876 if (indeadlock)
3877 goto retry_rmdir;
3878 error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr);
3879
3880 rw_exit(&ip->i_rwlock);
3881
3882 if (ulp) {
3883 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3884 trans_size);
3885 ufs_lockfs_end(ulp);
3886 }
3887
3888 out:
3889 return (error);
3890 }
3891
3892 /* ARGSUSED */
3893 static int
3894 ufs_readdir(struct vnode *vp, struct uio *uiop, struct cred *cr, int *eofp,
3895 caller_context_t *ct, int flags)
3896 {
3897 struct iovec *iovp;
3898 struct inode *ip;
3899 struct direct *idp;
3900 struct dirent64 *odp;
3901 struct fbuf *fbp;
3902 struct ufsvfs *ufsvfsp;
3903 struct ulockfs *ulp;
3904 caddr_t outbuf;
3905 size_t bufsize;
3906 uint_t offset;
3907 uint_t bytes_wanted, total_bytes_wanted;
3908 int incount = 0;
3909 int outcount = 0;
3910 int error;
3911
3912 ip = VTOI(vp);
3913 ASSERT(RW_READ_HELD(&ip->i_rwlock));
3914
3915 if (uiop->uio_loffset >= MAXOFF32_T) {
3916 if (eofp)
3917 *eofp = 1;
3918 return (0);
3919 }
3920
3921 /*
3922 * Check if we have been called with a valid iov_len
3923 * and bail out if not, otherwise we may potentially loop
3924 * forever further down.
3925 */
3926 if (uiop->uio_iov->iov_len <= 0) {
3927 error = EINVAL;
3928 goto out;
3929 }
3930
3931 /*
3932 * Large Files: When we come here we are guaranteed that
3933 * uio_offset can be used safely. The high word is zero.
3934 */
3935
3936 ufsvfsp = ip->i_ufsvfs;
3937 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3938 if (error)
3939 goto out;
3940
3941 iovp = uiop->uio_iov;
3942 total_bytes_wanted = iovp->iov_len;
3943
3944 /* Large Files: directory files should not be "large" */
3945
3946 ASSERT(ip->i_size <= MAXOFF32_T);
3947
3948 /* Force offset to be valid (to guard against bogus lseek() values) */
3949 offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3950
3951 /* Quit if at end of file or link count of zero (posix) */
3952 if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3953 if (eofp)
3954 *eofp = 1;
3955 error = 0;
3956 goto unlock;
3957 }
3958
3959 /*
3960 * Get space to change directory entries into fs independent format.
3961 * Do fast alloc for the most commonly used-request size (filesystem
3962 * block size).
3963 */
3964 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3965 bufsize = total_bytes_wanted;
3966 outbuf = kmem_alloc(bufsize, KM_SLEEP);
3967 odp = (struct dirent64 *)outbuf;
3968 } else {
3969 bufsize = total_bytes_wanted;
3970 odp = (struct dirent64 *)iovp->iov_base;
3971 }
3972
3973 nextblk:
3974 bytes_wanted = total_bytes_wanted;
3975
3976 /* Truncate request to file size */
3977 if (offset + bytes_wanted > (int)ip->i_size)
3978 bytes_wanted = (int)(ip->i_size - offset);
3979
3980 /* Comply with MAXBSIZE boundary restrictions of fbread() */
3981 if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3982 bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3983
3984 /*
3985 * Read in the next chunk.
3986 * We are still holding the i_rwlock.
3987 */
3988 error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
3989
3990 if (error)
3991 goto update_inode;
3992 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
3993 (!ufsvfsp->vfs_noatime)) {
3994 ip->i_flag |= IACC;
3995 }
3996 incount = 0;
3997 idp = (struct direct *)fbp->fb_addr;
3998 if (idp->d_ino == 0 && idp->d_reclen == 0 && idp->d_namlen == 0) {
3999 cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
4000 "fs = %s\n",
4001 (u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
4002 fbrelse(fbp, S_OTHER);
4003 error = ENXIO;
4004 goto update_inode;
4005 }
4006 /* Transform to file-system independent format */
4007 while (incount < bytes_wanted) {
4008 /*
4009 * If the current directory entry is mangled, then skip
4010 * to the next block. It would be nice to set the FSBAD
4011 * flag in the super-block so that a fsck is forced on
4012 * next reboot, but locking is a problem.
4013 */
4014 if (idp->d_reclen & 0x3) {
4015 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4016 break;
4017 }
4018
4019 /* Skip to requested offset and skip empty entries */
4020 if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
4021 ushort_t this_reclen =
4022 DIRENT64_RECLEN(idp->d_namlen);
4023 /* Buffer too small for any entries */
4024 if (!outcount && this_reclen > bufsize) {
4025 fbrelse(fbp, S_OTHER);
4026 error = EINVAL;
4027 goto update_inode;
4028 }
4029 /* If would overrun the buffer, quit */
4030 if (outcount + this_reclen > bufsize) {
4031 break;
4032 }
4033 /* Take this entry */
4034 odp->d_ino = (ino64_t)idp->d_ino;
4035 odp->d_reclen = (ushort_t)this_reclen;
4036 odp->d_off = (offset_t)(offset + idp->d_reclen);
4037
4038 /* use strncpy(9f) to zero out uninitialized bytes */
4039
4040 ASSERT(strlen(idp->d_name) + 1 <=
4041 DIRENT64_NAMELEN(this_reclen));
4042 (void) strncpy(odp->d_name, idp->d_name,
4043 DIRENT64_NAMELEN(this_reclen));
4044 outcount += odp->d_reclen;
4045 odp = (struct dirent64 *)
4046 ((intptr_t)odp + odp->d_reclen);
4047 ASSERT(outcount <= bufsize);
4048 }
4049 if (idp->d_reclen) {
4050 incount += idp->d_reclen;
4051 offset += idp->d_reclen;
4052 idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4053 } else {
4054 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4055 break;
4056 }
4057 }
4058 /* Release the chunk */
4059 fbrelse(fbp, S_OTHER);
4060
4061 /* Read whole block, but got no entries, read another if not eof */
4062
4063 /*
4064 * Large Files: casting i_size to int here is not a problem
4065 * because directory sizes are always less than MAXOFF32_T.
4066 * See assertion above.
4067 */
4068
4069 if (offset < (int)ip->i_size && !outcount)
4070 goto nextblk;
4071
4072 /* Copy out the entry data */
4073 if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4074 iovp->iov_base += outcount;
4075 iovp->iov_len -= outcount;
4076 uiop->uio_resid -= outcount;
4077 uiop->uio_offset = offset;
4078 } else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4079 uiop)) == 0)
4080 uiop->uio_offset = offset;
4081 update_inode:
4082 ITIMES(ip);
4083 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4084 kmem_free(outbuf, bufsize);
4085
4086 if (eofp && error == 0)
4087 *eofp = (uiop->uio_offset >= (int)ip->i_size);
4088 unlock:
4089 if (ulp) {
4090 ufs_lockfs_end(ulp);
4091 }
4092 out:
4093 return (error);
4094 }
4095
4096 /*ARGSUSED*/
4097 static int
4098 ufs_symlink(struct vnode *dvp, char *linkname, struct vattr *vap, char *target,
4099 struct cred *cr, caller_context_t *ct, int flags)
4100 {
4101 struct inode *ip, *dip = VTOI(dvp);
4102 struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4103 struct ulockfs *ulp;
4104 int error;
4105 int issync;
4106 int trans_size;
4107 int residual;
4108 int ioflag;
4109 int retry = 1;
4110
4111 /*
4112 * No symlinks in attrdirs at this time
4113 */
4114 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4115 return (EINVAL);
4116
4117 again:
4118 ip = (struct inode *)NULL;
4119 vap->va_type = VLNK;
4120 vap->va_rdev = 0;
4121
4122 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4123 if (error)
4124 goto out;
4125
4126 if (ulp)
4127 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4128 trans_size = (int)TOP_SYMLINK_SIZE(dip));
4129
4130 /*
4131 * We must create the inode before the directory entry, to avoid
4132 * racing with readlink(). ufs_dirmakeinode requires that we
4133 * hold the quota lock as reader, and directory locks as writer.
4134 */
4135
4136 rw_enter(&dip->i_rwlock, RW_WRITER);
4137 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4138 rw_enter(&dip->i_contents, RW_WRITER);
4139
4140 /*
4141 * Suppress any out of inodes messages if we will retry on
4142 * ENOSP
4143 */
4144 if (retry)
4145 dip->i_flag |= IQUIET;
4146
4147 error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4148
4149 dip->i_flag &= ~IQUIET;
4150
4151 rw_exit(&dip->i_contents);
4152 rw_exit(&ufsvfsp->vfs_dqrwlock);
4153 rw_exit(&dip->i_rwlock);
4154
4155 if (error)
4156 goto unlock;
4157
4158 /*
4159 * OK. The inode has been created. Write out the data of the
4160 * symbolic link. Since symbolic links are metadata, and should
4161 * remain consistent across a system crash, we need to force the
4162 * data out synchronously.
4163 *
4164 * (This is a change from the semantics in earlier releases, which
4165 * only created symbolic links synchronously if the semi-documented
4166 * 'syncdir' option was set, or if we were being invoked by the NFS
4167 * server, which requires symbolic links to be created synchronously.)
4168 *
4169 * We need to pass in a pointer for the residual length; otherwise
4170 * ufs_rdwri() will always return EIO if it can't write the data,
4171 * even if the error was really ENOSPC or EDQUOT.
4172 */
4173
4174 ioflag = FWRITE | FDSYNC;
4175 residual = 0;
4176
4177 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4178 rw_enter(&ip->i_contents, RW_WRITER);
4179
4180 /*
4181 * Suppress file system full messages if we will retry
4182 */
4183 if (retry)
4184 ip->i_flag |= IQUIET;
4185
4186 error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4187 (offset_t)0, UIO_SYSSPACE, &residual, cr);
4188
4189 ip->i_flag &= ~IQUIET;
4190
4191 if (error) {
4192 rw_exit(&ip->i_contents);
4193 rw_exit(&ufsvfsp->vfs_dqrwlock);
4194 goto remove;
4195 }
4196
4197 /*
4198 * If the link's data is small enough, we can cache it in the inode.
4199 * This is a "fast symbolic link". We don't use the first direct
4200 * block because that's actually used to point at the symbolic link's
4201 * contents on disk; but we know that none of the other direct or
4202 * indirect blocks can be used because symbolic links are restricted
4203 * to be smaller than a file system block.
4204 */
4205
4206 ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4207
4208 if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4209 if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4210 ip->i_flag |= IFASTSYMLNK;
4211 } else {
4212 int i;
4213 /* error, clear garbage left behind */
4214 for (i = 1; i < NDADDR; i++)
4215 ip->i_db[i] = 0;
4216 for (i = 0; i < NIADDR; i++)
4217 ip->i_ib[i] = 0;
4218 }
4219 }
4220
4221 rw_exit(&ip->i_contents);
4222 rw_exit(&ufsvfsp->vfs_dqrwlock);
4223
4224 /*
4225 * OK. We've successfully created the symbolic link. All that
4226 * remains is to insert it into the appropriate directory.
4227 */
4228
4229 rw_enter(&dip->i_rwlock, RW_WRITER);
4230 error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr);
4231 rw_exit(&dip->i_rwlock);
4232
4233 /*
4234 * Fall through into remove-on-error code. We're either done, or we
4235 * need to remove the inode (if we couldn't insert it).
4236 */
4237
4238 remove:
4239 if (error && (ip != NULL)) {
4240 rw_enter(&ip->i_contents, RW_WRITER);
4241 ip->i_nlink--;
4242 ip->i_flag |= ICHG;
4243 ip->i_seq++;
4244 ufs_setreclaim(ip);
4245 rw_exit(&ip->i_contents);
4246 }
4247
4248 unlock:
4249 if (ip != NULL)
4250 VN_RELE(ITOV(ip));
4251
4252 if (ulp) {
4253 int terr = 0;
4254
4255 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4256 trans_size);
4257 ufs_lockfs_end(ulp);
4258 if (error == 0)
4259 error = terr;
4260 }
4261
4262 /*
4263 * We may have failed due to lack of an inode or of a block to
4264 * store the target in. Try flushing the delete queue to free
4265 * logically-available things up and try again.
4266 */
4267 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4268 ufs_delete_drain_wait(ufsvfsp, 1);
4269 retry = 0;
4270 goto again;
4271 }
4272
4273 out:
4274 return (error);
4275 }
4276
4277 /*
4278 * Ufs specific routine used to do ufs io.
4279 */
4280 int
4281 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4282 ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4283 struct cred *cr)
4284 {
4285 struct uio auio;
4286 struct iovec aiov;
4287 int error;
4288
4289 ASSERT(RW_LOCK_HELD(&ip->i_contents));
4290
4291 bzero((caddr_t)&auio, sizeof (uio_t));
4292 bzero((caddr_t)&aiov, sizeof (iovec_t));
4293
4294 aiov.iov_base = base;
4295 aiov.iov_len = len;
4296 auio.uio_iov = &aiov;
4297 auio.uio_iovcnt = 1;
4298 auio.uio_loffset = offset;
4299 auio.uio_segflg = (short)seg;
4300 auio.uio_resid = len;
4301
4302 if (rw == UIO_WRITE) {
4303 auio.uio_fmode = FWRITE;
4304 auio.uio_extflg = UIO_COPY_DEFAULT;
4305 auio.uio_llimit = curproc->p_fsz_ctl;
4306 error = wrip(ip, &auio, ioflag, cr);
4307 } else {
4308 auio.uio_fmode = FREAD;
4309 auio.uio_extflg = UIO_COPY_CACHED;
4310 auio.uio_llimit = MAXOFFSET_T;
4311 error = rdip(ip, &auio, ioflag, cr);
4312 }
4313
4314 if (aresid) {
4315 *aresid = auio.uio_resid;
4316 } else if (auio.uio_resid) {
4317 error = EIO;
4318 }
4319 return (error);
4320 }
4321
4322 /*ARGSUSED*/
4323 static int
4324 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
4325 {
4326 struct ufid *ufid;
4327 struct inode *ip = VTOI(vp);
4328
4329 if (ip->i_ufsvfs == NULL)
4330 return (EIO);
4331
4332 if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4333 fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4334 return (ENOSPC);
4335 }
4336
4337 ufid = (struct ufid *)fidp;
4338 bzero((char *)ufid, sizeof (struct ufid));
4339 ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4340 ufid->ufid_ino = ip->i_number;
4341 ufid->ufid_gen = ip->i_gen;
4342
4343 return (0);
4344 }
4345
4346 /* ARGSUSED2 */
4347 static int
4348 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4349 {
4350 struct inode *ip = VTOI(vp);
4351 struct ufsvfs *ufsvfsp;
4352 int forcedirectio;
4353
4354 /*
4355 * Read case is easy.
4356 */
4357 if (!write_lock) {
4358 rw_enter(&ip->i_rwlock, RW_READER);
4359 return (V_WRITELOCK_FALSE);
4360 }
4361
4362 /*
4363 * Caller has requested a writer lock, but that inhibits any
4364 * concurrency in the VOPs that follow. Acquire the lock shared
4365 * and defer exclusive access until it is known to be needed in
4366 * other VOP handlers. Some cases can be determined here.
4367 */
4368
4369 /*
4370 * If directio is not set, there is no chance of concurrency,
4371 * so just acquire the lock exclusive. Beware of a forced
4372 * unmount before looking at the mount option.
4373 */
4374 ufsvfsp = ip->i_ufsvfs;
4375 forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4376 if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4377 !ufs_allow_shared_writes) {
4378 rw_enter(&ip->i_rwlock, RW_WRITER);
4379 return (V_WRITELOCK_TRUE);
4380 }
4381
4382 /*
4383 * Mandatory locking forces acquiring i_rwlock exclusive.
4384 */
4385 if (MANDLOCK(vp, ip->i_mode)) {
4386 rw_enter(&ip->i_rwlock, RW_WRITER);
4387 return (V_WRITELOCK_TRUE);
4388 }
4389
4390 /*
4391 * Acquire the lock shared in case a concurrent write follows.
4392 * Mandatory locking could have become enabled before the lock
4393 * was acquired. Re-check and upgrade if needed.
4394 */
4395 rw_enter(&ip->i_rwlock, RW_READER);
4396 if (MANDLOCK(vp, ip->i_mode)) {
4397 rw_exit(&ip->i_rwlock);
4398 rw_enter(&ip->i_rwlock, RW_WRITER);
4399 return (V_WRITELOCK_TRUE);
4400 }
4401 return (V_WRITELOCK_FALSE);
4402 }
4403
4404 /*ARGSUSED*/
4405 static void
4406 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4407 {
4408 struct inode *ip = VTOI(vp);
4409
4410 rw_exit(&ip->i_rwlock);
4411 }
4412
4413 /* ARGSUSED */
4414 static int
4415 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp, caller_context_t *ct)
4416 {
4417 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4418 }
4419
4420 /* ARGSUSED */
4421 static int
4422 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4423 offset_t offset, struct flk_callback *flk_cbp, struct cred *cr,
4424 caller_context_t *ct)
4425 {
4426 struct inode *ip = VTOI(vp);
4427
4428 if (ip->i_ufsvfs == NULL)
4429 return (EIO);
4430
4431 /*
4432 * If file is being mapped, disallow frlock.
4433 * XXX I am not holding tlock while checking i_mapcnt because the
4434 * current locking strategy drops all locks before calling fs_frlock.
4435 * So, mapcnt could change before we enter fs_frlock making is
4436 * meaningless to have held tlock in the first place.
4437 */
4438 if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4439 return (EAGAIN);
4440 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4441 }
4442
4443 /* ARGSUSED */
4444 static int
4445 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4446 offset_t offset, cred_t *cr, caller_context_t *ct)
4447 {
4448 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4449 struct ulockfs *ulp;
4450 int error;
4451
4452 if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4453 if (cmd == F_FREESP) {
4454 error = ufs_lockfs_begin(ufsvfsp, &ulp,
4455 ULOCKFS_SPACE_MASK);
4456 if (error)
4457 return (error);
4458 error = ufs_freesp(vp, bfp, flag, cr);
4459
4460 if (error == 0 && bfp->l_start == 0)
4461 vnevent_truncate(vp, ct);
4462 } else if (cmd == F_ALLOCSP) {
4463 error = ufs_lockfs_begin(ufsvfsp, &ulp,
4464 ULOCKFS_FALLOCATE_MASK);
4465 if (error)
4466 return (error);
4467 error = ufs_allocsp(vp, bfp, cr);
4468 } else
4469 return (EINVAL); /* Command not handled here */
4470
4471 if (ulp)
4472 ufs_lockfs_end(ulp);
4473
4474 }
4475 return (error);
4476 }
4477
4478 /*
4479 * Used to determine if read ahead should be done. Also used to
4480 * to determine when write back occurs.
4481 */
4482 #define CLUSTSZ(ip) ((ip)->i_ufsvfs->vfs_ioclustsz)
4483
4484 /*
4485 * A faster version of ufs_getpage.
4486 *
4487 * We optimize by inlining the pvn_getpages iterator, eliminating
4488 * calls to bmap_read if file doesn't have UFS holes, and avoiding
4489 * the overhead of page_exists().
4490 *
4491 * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4492 * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4493 * victimizes performance when a file with UFS holes is faulted
4494 * first in the S_READ mode, and then in the S_WRITE mode. We will get
4495 * two MMU faults in this case.
4496 *
4497 * XXX - the inode fields which control the sequential mode are not
4498 * protected by any mutex. The read ahead will act wild if
4499 * multiple processes will access the file concurrently and
4500 * some of them in sequential mode. One particulary bad case
4501 * is if another thread will change the value of i_nextrio between
4502 * the time this thread tests the i_nextrio value and then reads it
4503 * again to use it as the offset for the read ahead.
4504 */
4505 /*ARGSUSED*/
4506 static int
4507 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4508 page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4509 enum seg_rw rw, struct cred *cr, caller_context_t *ct)
4510 {
4511 u_offset_t uoff = (u_offset_t)off; /* type conversion */
4512 u_offset_t pgoff;
4513 u_offset_t eoff;
4514 struct inode *ip = VTOI(vp);
4515 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
4516 struct fs *fs;
4517 struct ulockfs *ulp;
4518 page_t **pl;
4519 caddr_t pgaddr;
4520 krw_t rwtype;
4521 int err;
4522 int has_holes;
4523 int beyond_eof;
4524 int seqmode;
4525 int pgsize = PAGESIZE;
4526 int dolock;
4527 int do_qlock;
4528 int trans_size;
4529
4530 ASSERT((uoff & PAGEOFFSET) == 0);
4531
4532 if (protp)
4533 *protp = PROT_ALL;
4534
4535 /*
4536 * Obey the lockfs protocol
4537 */
4538 err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4539 rw == S_READ || rw == S_EXEC, protp);
4540 if (err)
4541 goto out;
4542
4543 fs = ufsvfsp->vfs_fs;
4544
4545 if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4546 !(vp->v_flag & VISSWAP)) {
4547 /*
4548 * Try to start a transaction, will return if blocking is
4549 * expected to occur and the address space is not the
4550 * kernel address space.
4551 */
4552 trans_size = TOP_GETPAGE_SIZE(ip);
4553 if (seg->s_as != &kas) {
4554 TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4555 trans_size, err)
4556 if (err == EWOULDBLOCK) {
4557 /*
4558 * Use EDEADLK here because the VM code
4559 * can normally never see this error.
4560 */
4561 err = EDEADLK;
4562 ufs_lockfs_end(ulp);
4563 goto out;
4564 }
4565 } else {
4566 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4567 }
4568 }
4569
4570 if (vp->v_flag & VNOMAP) {
4571 err = ENOSYS;
4572 goto unlock;
4573 }
4574
4575 seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4576
4577 rwtype = RW_READER; /* start as a reader */
4578 dolock = (rw_owner(&ip->i_contents) != curthread);
4579 /*
4580 * If this thread owns the lock, i.e., this thread grabbed it
4581 * as writer somewhere above, then we don't need to grab the
4582 * lock as reader in this routine.
4583 */
4584 do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4585
4586 retrylock:
4587 if (dolock) {
4588 /*
4589 * Grab the quota lock if we need to call
4590 * bmap_write() below (with i_contents as writer).
4591 */
4592 if (do_qlock && rwtype == RW_WRITER)
4593 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4594 rw_enter(&ip->i_contents, rwtype);
4595 }
4596
4597 /*
4598 * We may be getting called as a side effect of a bmap using
4599 * fbread() when the blocks might be being allocated and the
4600 * size has not yet been up'ed. In this case we want to be
4601 * able to return zero pages if we get back UFS_HOLE from
4602 * calling bmap for a non write case here. We also might have
4603 * to read some frags from the disk into a page if we are
4604 * extending the number of frags for a given lbn in bmap().
4605 * Large Files: The read of i_size here is atomic because
4606 * i_contents is held here. If dolock is zero, the lock
4607 * is held in bmap routines.
4608 */
4609 beyond_eof = uoff + len >
4610 P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t);
4611 if (beyond_eof && seg != segkmap) {
4612 if (dolock) {
4613 rw_exit(&ip->i_contents);
4614 if (do_qlock && rwtype == RW_WRITER)
4615 rw_exit(&ufsvfsp->vfs_dqrwlock);
4616 }
4617 err = EFAULT;
4618 goto unlock;
4619 }
4620
4621 /*
4622 * Must hold i_contents lock throughout the call to pvn_getpages
4623 * since locked pages are returned from each call to ufs_getapage.
4624 * Must *not* return locked pages and then try for contents lock
4625 * due to lock ordering requirements (inode > page)
4626 */
4627
4628 has_holes = bmap_has_holes(ip);
4629
4630 if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4631 int blk_size;
4632 u_offset_t offset;
4633
4634 /*
4635 * We must acquire the RW_WRITER lock in order to
4636 * call bmap_write().
4637 */
4638 if (dolock && rwtype == RW_READER) {
4639 rwtype = RW_WRITER;
4640
4641 /*
4642 * Grab the quota lock before
4643 * upgrading i_contents, but if we can't grab it
4644 * don't wait here due to lock order:
4645 * vfs_dqrwlock > i_contents.
4646 */
4647 if (do_qlock &&
4648 rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER)
4649 == 0) {
4650 rw_exit(&ip->i_contents);
4651 goto retrylock;
4652 }
4653 if (!rw_tryupgrade(&ip->i_contents)) {
4654 rw_exit(&ip->i_contents);
4655 if (do_qlock)
4656 rw_exit(&ufsvfsp->vfs_dqrwlock);
4657 goto retrylock;
4658 }
4659 }
4660
4661 /*
4662 * May be allocating disk blocks for holes here as
4663 * a result of mmap faults. write(2) does the bmap_write
4664 * in rdip/wrip, not here. We are not dealing with frags
4665 * in this case.
4666 */
4667 /*
4668 * Large Files: We cast fs_bmask field to offset_t
4669 * just as we do for MAXBMASK because uoff is a 64-bit
4670 * data type. fs_bmask will still be a 32-bit type
4671 * as we cannot change any ondisk data structures.
4672 */
4673
4674 offset = uoff & (offset_t)fs->fs_bmask;
4675 while (offset < uoff + len) {
4676 blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4677 err = bmap_write(ip, offset, blk_size,
4678 BI_NORMAL, NULL, cr);
4679 if (ip->i_flag & (ICHG|IUPD))
4680 ip->i_seq++;
4681 if (err)
4682 goto update_inode;
4683 offset += blk_size; /* XXX - make this contig */
4684 }
4685 }
4686
4687 /*
4688 * Can be a reader from now on.
4689 */
4690 if (dolock && rwtype == RW_WRITER) {
4691 rw_downgrade(&ip->i_contents);
4692 /*
4693 * We can release vfs_dqrwlock early so do it, but make
4694 * sure we don't try to release it again at the bottom.
4695 */
4696 if (do_qlock) {
4697 rw_exit(&ufsvfsp->vfs_dqrwlock);
4698 do_qlock = 0;
4699 }
4700 }
4701
4702 /*
4703 * We remove PROT_WRITE in cases when the file has UFS holes
4704 * because we don't want to call bmap_read() to check each
4705 * page if it is backed with a disk block.
4706 */
4707 if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4708 *protp &= ~PROT_WRITE;
4709
4710 err = 0;
4711
4712 /*
4713 * The loop looks up pages in the range [off, off + len).
4714 * For each page, we first check if we should initiate an asynchronous
4715 * read ahead before we call page_lookup (we may sleep in page_lookup
4716 * for a previously initiated disk read).
4717 */
4718 eoff = (uoff + len);
4719 for (pgoff = uoff, pgaddr = addr, pl = plarr;
4720 pgoff < eoff; /* empty */) {
4721 page_t *pp;
4722 u_offset_t nextrio;
4723 se_t se;
4724 int retval;
4725
4726 se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4727
4728 /* Handle async getpage (faultahead) */
4729 if (plarr == NULL) {
4730 ip->i_nextrio = pgoff;
4731 (void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4732 pgoff += pgsize;
4733 pgaddr += pgsize;
4734 continue;
4735 }
4736 /*
4737 * Check if we should initiate read ahead of next cluster.
4738 * We call page_exists only when we need to confirm that
4739 * we have the current page before we initiate the read ahead.
4740 */
4741 nextrio = ip->i_nextrio;
4742 if (seqmode &&
4743 pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4744 nextrio < ip->i_size && page_exists(vp, pgoff)) {
4745 retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4746 /*
4747 * We always read ahead the next cluster of data
4748 * starting from i_nextrio. If the page (vp,nextrio)
4749 * is actually in core at this point, the routine
4750 * ufs_getpage_ra() will stop pre-fetching data
4751 * until we read that page in a synchronized manner
4752 * through ufs_getpage_miss(). So, we should increase
4753 * i_nextrio if the page (vp, nextrio) exists.
4754 */
4755 if ((retval == 0) && page_exists(vp, nextrio)) {
4756 ip->i_nextrio = nextrio + pgsize;
4757 }
4758 }
4759
4760 if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4761 /*
4762 * We found the page in the page cache.
4763 */
4764 *pl++ = pp;
4765 pgoff += pgsize;
4766 pgaddr += pgsize;
4767 len -= pgsize;
4768 plsz -= pgsize;
4769 } else {
4770 /*
4771 * We have to create the page, or read it from disk.
4772 */
4773 if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4774 pl, plsz, rw, seqmode))
4775 goto error;
4776
4777 while (*pl != NULL) {
4778 pl++;
4779 pgoff += pgsize;
4780 pgaddr += pgsize;
4781 len -= pgsize;
4782 plsz -= pgsize;
4783 }
4784 }
4785 }
4786
4787 /*
4788 * Return pages up to plsz if they are in the page cache.
4789 * We cannot return pages if there is a chance that they are
4790 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4791 */
4792 if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4793
4794 ASSERT((protp == NULL) ||
4795 !(has_holes && (*protp & PROT_WRITE)));
4796
4797 eoff = pgoff + plsz;
4798 while (pgoff < eoff) {
4799 page_t *pp;
4800
4801 if ((pp = page_lookup_nowait(vp, pgoff,
4802 SE_SHARED)) == NULL)
4803 break;
4804
4805 *pl++ = pp;
4806 pgoff += pgsize;
4807 plsz -= pgsize;
4808 }
4809 }
4810
4811 if (plarr)
4812 *pl = NULL; /* Terminate page list */
4813 ip->i_nextr = pgoff;
4814
4815 error:
4816 if (err && plarr) {
4817 /*
4818 * Release any pages we have locked.
4819 */
4820 while (pl > &plarr[0])
4821 page_unlock(*--pl);
4822
4823 plarr[0] = NULL;
4824 }
4825
4826 update_inode:
4827 /*
4828 * If the inode is not already marked for IACC (in rdip() for read)
4829 * and the inode is not marked for no access time update (in wrip()
4830 * for write) then update the inode access time and mod time now.
4831 */
4832 if ((ip->i_flag & (IACC | INOACC)) == 0) {
4833 if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4834 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4835 (fs->fs_ronly == 0) &&
4836 (!ufsvfsp->vfs_noatime)) {
4837 mutex_enter(&ip->i_tlock);
4838 ip->i_flag |= IACC;
4839 ITIMES_NOLOCK(ip);
4840 mutex_exit(&ip->i_tlock);
4841 }
4842 }
4843 }
4844
4845 if (dolock) {
4846 rw_exit(&ip->i_contents);
4847 if (do_qlock && rwtype == RW_WRITER)
4848 rw_exit(&ufsvfsp->vfs_dqrwlock);
4849 }
4850
4851 unlock:
4852 if (ulp) {
4853 if ((rw == S_CREATE || rw == S_WRITE) &&
4854 !(vp->v_flag & VISSWAP)) {
4855 TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4856 }
4857 ufs_lockfs_end(ulp);
4858 }
4859 out:
4860 return (err);
4861 }
4862
4863 /*
4864 * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4865 * cache. The page is either read from the disk, or it's created.
4866 * A page is created (without disk read) if rw == S_CREATE, or if
4867 * the page is not backed with a real disk block (UFS hole).
4868 */
4869 /* ARGSUSED */
4870 static int
4871 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4872 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4873 {
4874 struct inode *ip = VTOI(vp);
4875 page_t *pp;
4876 daddr_t bn;
4877 size_t io_len;
4878 int crpage = 0;
4879 int err;
4880 int contig;
4881 int bsize = ip->i_fs->fs_bsize;
4882
4883 /*
4884 * Figure out whether the page can be created, or must be
4885 * must be read from the disk.
4886 */
4887 if (rw == S_CREATE)
4888 crpage = 1;
4889 else {
4890 contig = 0;
4891 if (err = bmap_read(ip, off, &bn, &contig))
4892 return (err);
4893
4894 crpage = (bn == UFS_HOLE);
4895
4896 /*
4897 * If its also a fallocated block that hasn't been written to
4898 * yet, we will treat it just like a UFS_HOLE and create
4899 * a zero page for it
4900 */
4901 if (ISFALLOCBLK(ip, bn))
4902 crpage = 1;
4903 }
4904
4905 if (crpage) {
4906 if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4907 addr)) == NULL) {
4908 return (ufs_fault(vp,
4909 "ufs_getpage_miss: page_create == NULL"));
4910 }
4911
4912 if (rw != S_CREATE)
4913 pagezero(pp, 0, PAGESIZE);
4914
4915 io_len = PAGESIZE;
4916 } else {
4917 u_offset_t io_off;
4918 uint_t xlen;
4919 struct buf *bp;
4920 ufsvfs_t *ufsvfsp = ip->i_ufsvfs;
4921
4922 /*
4923 * If access is not in sequential order, we read from disk
4924 * in bsize units.
4925 *
4926 * We limit the size of the transfer to bsize if we are reading
4927 * from the beginning of the file. Note in this situation we
4928 * will hedge our bets and initiate an async read ahead of
4929 * the second block.
4930 */
4931 if (!seq || off == 0)
4932 contig = MIN(contig, bsize);
4933
4934 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4935 &io_len, off, contig, 0);
4936
4937 /*
4938 * Some other thread has entered the page.
4939 * ufs_getpage will retry page_lookup.
4940 */
4941 if (pp == NULL) {
4942 pl[0] = NULL;
4943 return (0);
4944 }
4945
4946 /*
4947 * Zero part of the page which we are not
4948 * going to read from the disk.
4949 */
4950 xlen = io_len & PAGEOFFSET;
4951 if (xlen != 0)
4952 pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4953
4954 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4955 bp->b_edev = ip->i_dev;
4956 bp->b_dev = cmpdev(ip->i_dev);
4957 bp->b_blkno = bn;
4958 bp->b_un.b_addr = (caddr_t)0;
4959 bp->b_file = ip->i_vnode;
4960 bp->b_offset = off;
4961
4962 if (ufsvfsp->vfs_log) {
4963 lufs_read_strategy(ufsvfsp->vfs_log, bp);
4964 } else if (ufsvfsp->vfs_snapshot) {
4965 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4966 } else {
4967 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
4968 ub.ub_getpages.value.ul++;
4969 (void) bdev_strategy(bp);
4970 lwp_stat_update(LWP_STAT_INBLK, 1);
4971 }
4972
4973 ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
4974
4975 /*
4976 * If the file access is sequential, initiate read ahead
4977 * of the next cluster.
4978 */
4979 if (seq && ip->i_nextrio < ip->i_size)
4980 (void) ufs_getpage_ra(vp, off, seg, addr);
4981 err = biowait(bp);
4982 pageio_done(bp);
4983
4984 if (err) {
4985 pvn_read_done(pp, B_ERROR);
4986 return (err);
4987 }
4988 }
4989
4990 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
4991 return (0);
4992 }
4993
4994 /*
4995 * Read ahead a cluster from the disk. Returns the length in bytes.
4996 */
4997 static int
4998 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
4999 {
5000 struct inode *ip = VTOI(vp);
5001 page_t *pp;
5002 u_offset_t io_off = ip->i_nextrio;
5003 ufsvfs_t *ufsvfsp;
5004 caddr_t addr2 = addr + (io_off - off);
5005 struct buf *bp;
5006 daddr_t bn;
5007 size_t io_len;
5008 int err;
5009 int contig;
5010 int xlen;
5011 int bsize = ip->i_fs->fs_bsize;
5012
5013 /*
5014 * If the directio advisory is in effect on this file,
5015 * then do not do buffered read ahead. Read ahead makes
5016 * it more difficult on threads using directio as they
5017 * will be forced to flush the pages from this vnode.
5018 */
5019 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5020 return (0);
5021 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5022 return (0);
5023
5024 /*
5025 * Is this test needed?
5026 */
5027 if (addr2 >= seg->s_base + seg->s_size)
5028 return (0);
5029
5030 contig = 0;
5031 err = bmap_read(ip, io_off, &bn, &contig);
5032 /*
5033 * If its a UFS_HOLE or a fallocated block, do not perform
5034 * any read ahead's since there probably is nothing to read ahead
5035 */
5036 if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5037 return (0);
5038
5039 /*
5040 * Limit the transfer size to bsize if this is the 2nd block.
5041 */
5042 if (io_off == (u_offset_t)bsize)
5043 contig = MIN(contig, bsize);
5044
5045 if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5046 &io_len, io_off, contig, 1)) == NULL)
5047 return (0);
5048
5049 /*
5050 * Zero part of page which we are not going to read from disk
5051 */
5052 if ((xlen = (io_len & PAGEOFFSET)) > 0)
5053 pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5054
5055 ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5056
5057 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5058 bp->b_edev = ip->i_dev;
5059 bp->b_dev = cmpdev(ip->i_dev);
5060 bp->b_blkno = bn;
5061 bp->b_un.b_addr = (caddr_t)0;
5062 bp->b_file = ip->i_vnode;
5063 bp->b_offset = off;
5064
5065 if (ufsvfsp->vfs_log) {
5066 lufs_read_strategy(ufsvfsp->vfs_log, bp);
5067 } else if (ufsvfsp->vfs_snapshot) {
5068 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5069 } else {
5070 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5071 ub.ub_getras.value.ul++;
5072 (void) bdev_strategy(bp);
5073 lwp_stat_update(LWP_STAT_INBLK, 1);
5074 }
5075
5076 return (io_len);
5077 }
5078
5079 int ufs_delay = 1;
5080 /*
5081 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5082 *
5083 * LMXXX - the inode really ought to contain a pointer to one of these
5084 * async args. Stuff gunk in there and just hand the whole mess off.
5085 * This would replace i_delaylen, i_delayoff.
5086 */
5087 /*ARGSUSED*/
5088 static int
5089 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5090 struct cred *cr, caller_context_t *ct)
5091 {
5092 struct inode *ip = VTOI(vp);
5093 int err = 0;
5094
5095 if (vp->v_count == 0) {
5096 return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5097 }
5098
5099 /*
5100 * XXX - Why should this check be made here?
5101 */
5102 if (vp->v_flag & VNOMAP) {
5103 err = ENOSYS;
5104 goto errout;
5105 }
5106
5107 if (ip->i_ufsvfs == NULL) {
5108 err = EIO;
5109 goto errout;
5110 }
5111
5112 if (flags & B_ASYNC) {
5113 if (ufs_delay && len &&
5114 (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5115 mutex_enter(&ip->i_tlock);
5116 /*
5117 * If nobody stalled, start a new cluster.
5118 */
5119 if (ip->i_delaylen == 0) {
5120 ip->i_delayoff = off;
5121 ip->i_delaylen = len;
5122 mutex_exit(&ip->i_tlock);
5123 goto errout;
5124 }
5125 /*
5126 * If we have a full cluster or they are not contig,
5127 * then push last cluster and start over.
5128 */
5129 if (ip->i_delaylen >= CLUSTSZ(ip) ||
5130 ip->i_delayoff + ip->i_delaylen != off) {
5131 u_offset_t doff;
5132 size_t dlen;
5133
5134 doff = ip->i_delayoff;
5135 dlen = ip->i_delaylen;
5136 ip->i_delayoff = off;
5137 ip->i_delaylen = len;
5138 mutex_exit(&ip->i_tlock);
5139 err = ufs_putpages(vp, doff, dlen,
5140 flags, cr);
5141 /* LMXXX - flags are new val, not old */
5142 goto errout;
5143 }
5144 /*
5145 * There is something there, it's not full, and
5146 * it is contig.
5147 */
5148 ip->i_delaylen += len;
5149 mutex_exit(&ip->i_tlock);
5150 goto errout;
5151 }
5152 /*
5153 * Must have weird flags or we are not clustering.
5154 */
5155 }
5156
5157 err = ufs_putpages(vp, off, len, flags, cr);
5158
5159 errout:
5160 return (err);
5161 }
5162
5163 /*
5164 * If len == 0, do from off to EOF.
5165 *
5166 * The normal cases should be len == 0 & off == 0 (entire vp list),
5167 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5168 * (from pageout).
5169 */
5170 /*ARGSUSED*/
5171 static int
5172 ufs_putpages(struct vnode *vp, offset_t off, size_t len, int flags,
5173 struct cred *cr)
5174 {
5175 u_offset_t io_off;
5176 u_offset_t eoff;
5177 struct inode *ip = VTOI(vp);
5178 page_t *pp;
5179 size_t io_len;
5180 int err = 0;
5181 int dolock;
5182
5183 if (vp->v_count == 0)
5184 return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5185 /*
5186 * Acquire the readers/write inode lock before locking
5187 * any pages in this inode.
5188 * The inode lock is held during i/o.
5189 */
5190 if (len == 0) {
5191 mutex_enter(&ip->i_tlock);
5192 ip->i_delayoff = ip->i_delaylen = 0;
5193 mutex_exit(&ip->i_tlock);
5194 }
5195 dolock = (rw_owner(&ip->i_contents) != curthread);
5196 if (dolock) {
5197 /*
5198 * Must synchronize this thread and any possible thread
5199 * operating in the window of vulnerability in wrip().
5200 * It is dangerous to allow both a thread doing a putpage
5201 * and a thread writing, so serialize them. The exception
5202 * is when the thread in wrip() does something which causes
5203 * a putpage operation. Then, the thread must be allowed
5204 * to continue. It may encounter a bmap_read problem in
5205 * ufs_putapage, but that is handled in ufs_putapage.
5206 * Allow async writers to proceed, we don't want to block
5207 * the pageout daemon.
5208 */
5209 if (ip->i_writer == curthread)
5210 rw_enter(&ip->i_contents, RW_READER);
5211 else {
5212 for (;;) {
5213 rw_enter(&ip->i_contents, RW_READER);
5214 mutex_enter(&ip->i_tlock);
5215 /*
5216 * If there is no thread in the critical
5217 * section of wrip(), then proceed.
5218 * Otherwise, wait until there isn't one.
5219 */
5220 if (ip->i_writer == NULL) {
5221 mutex_exit(&ip->i_tlock);
5222 break;
5223 }
5224 rw_exit(&ip->i_contents);
5225 /*
5226 * Bounce async writers when we have a writer
5227 * working on this file so we don't deadlock
5228 * the pageout daemon.
5229 */
5230 if (flags & B_ASYNC) {
5231 mutex_exit(&ip->i_tlock);
5232 return (0);
5233 }
5234 cv_wait(&ip->i_wrcv, &ip->i_tlock);
5235 mutex_exit(&ip->i_tlock);
5236 }
5237 }
5238 }
5239
5240 if (!vn_has_cached_data(vp)) {
5241 if (dolock)
5242 rw_exit(&ip->i_contents);
5243 return (0);
5244 }
5245
5246 if (len == 0) {
5247 /*
5248 * Search the entire vp list for pages >= off.
5249 */
5250 err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5251 flags, cr);
5252 } else {
5253 /*
5254 * Loop over all offsets in the range looking for
5255 * pages to deal with.
5256 */
5257 if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5258 eoff = MIN(off + len, eoff);
5259 else
5260 eoff = off + len;
5261
5262 for (io_off = off; io_off < eoff; io_off += io_len) {
5263 /*
5264 * If we are not invalidating, synchronously
5265 * freeing or writing pages, use the routine
5266 * page_lookup_nowait() to prevent reclaiming
5267 * them from the free list.
5268 */
5269 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5270 pp = page_lookup(vp, io_off,
5271 (flags & (B_INVAL | B_FREE)) ?
5272 SE_EXCL : SE_SHARED);
5273 } else {
5274 pp = page_lookup_nowait(vp, io_off,
5275 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
5276 }
5277
5278 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5279 io_len = PAGESIZE;
5280 else {
5281 u_offset_t *io_offp = &io_off;
5282
5283 err = ufs_putapage(vp, pp, io_offp, &io_len,
5284 flags, cr);
5285 if (err != 0)
5286 break;
5287 /*
5288 * "io_off" and "io_len" are returned as
5289 * the range of pages we actually wrote.
5290 * This allows us to skip ahead more quickly
5291 * since several pages may've been dealt
5292 * with by this iteration of the loop.
5293 */
5294 }
5295 }
5296 }
5297 if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5298 /*
5299 * We have just sync'ed back all the pages on
5300 * the inode, turn off the IMODTIME flag.
5301 */
5302 mutex_enter(&ip->i_tlock);
5303 ip->i_flag &= ~IMODTIME;
5304 mutex_exit(&ip->i_tlock);
5305 }
5306 if (dolock)
5307 rw_exit(&ip->i_contents);
5308 return (err);
5309 }
5310
5311 static void
5312 ufs_iodone(buf_t *bp)
5313 {
5314 struct inode *ip;
5315
5316 ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5317
5318 bp->b_iodone = NULL;
5319
5320 ip = VTOI(bp->b_pages->p_vnode);
5321
5322 mutex_enter(&ip->i_tlock);
5323 if (ip->i_writes >= ufs_LW) {
5324 if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5325 if (ufs_WRITES)
5326 cv_broadcast(&ip->i_wrcv); /* wake all up */
5327 } else {
5328 ip->i_writes -= bp->b_bcount;
5329 }
5330
5331 mutex_exit(&ip->i_tlock);
5332 iodone(bp);
5333 }
5334
5335 /*
5336 * Write out a single page, possibly klustering adjacent
5337 * dirty pages. The inode lock must be held.
5338 *
5339 * LMXXX - bsize < pagesize not done.
5340 */
5341 /*ARGSUSED*/
5342 int
5343 ufs_putapage(struct vnode *vp, page_t *pp, u_offset_t *offp, size_t *lenp,
5344 int flags, struct cred *cr)
5345 {
5346 u_offset_t io_off;
5347 u_offset_t off;
5348 struct inode *ip = VTOI(vp);
5349 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5350 struct fs *fs;
5351 struct buf *bp;
5352 size_t io_len;
5353 daddr_t bn;
5354 int err;
5355 int contig;
5356 int dotrans;
5357
5358 ASSERT(RW_LOCK_HELD(&ip->i_contents));
5359
5360 if (ufsvfsp == NULL) {
5361 err = EIO;
5362 goto out_trace;
5363 }
5364
5365 fs = ip->i_fs;
5366 ASSERT(fs->fs_ronly == 0);
5367
5368 /*
5369 * If the modified time on the inode has not already been
5370 * set elsewhere (e.g. for write/setattr) we set the time now.
5371 * This gives us approximate modified times for mmap'ed files
5372 * which are modified via stores in the user address space.
5373 */
5374 if ((ip->i_flag & IMODTIME) == 0) {
5375 mutex_enter(&ip->i_tlock);
5376 ip->i_flag |= IUPD;
5377 ip->i_seq++;
5378 ITIMES_NOLOCK(ip);
5379 mutex_exit(&ip->i_tlock);
5380 }
5381
5382 /*
5383 * Align the request to a block boundry (for old file systems),
5384 * and go ask bmap() how contiguous things are for this file.
5385 */
5386 off = pp->p_offset & (offset_t)fs->fs_bmask; /* block align it */
5387 contig = 0;
5388 err = bmap_read(ip, off, &bn, &contig);
5389 if (err)
5390 goto out;
5391 if (bn == UFS_HOLE) { /* putpage never allocates */
5392 /*
5393 * logging device is in error mode; simply return EIO
5394 */
5395 if (TRANS_ISERROR(ufsvfsp)) {
5396 err = EIO;
5397 goto out;
5398 }
5399 /*
5400 * Oops, the thread in the window in wrip() did some
5401 * sort of operation which caused a putpage in the bad
5402 * range. In this case, just return an error which will
5403 * cause the software modified bit on the page to set
5404 * and the page will get written out again later.
5405 */
5406 if (ip->i_writer == curthread) {
5407 err = EIO;
5408 goto out;
5409 }
5410 /*
5411 * If the pager is trying to push a page in the bad range
5412 * just tell it to try again later when things are better.
5413 */
5414 if (flags & B_ASYNC) {
5415 err = EAGAIN;
5416 goto out;
5417 }
5418 err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5419 goto out;
5420 }
5421
5422 /*
5423 * If it is an fallocate'd block, reverse the negativity since
5424 * we are now writing to it
5425 */
5426 if (ISFALLOCBLK(ip, bn)) {
5427 err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5428 if (err)
5429 goto out;
5430
5431 bn = -bn;
5432 }
5433
5434 /*
5435 * Take the length (of contiguous bytes) passed back from bmap()
5436 * and _try_ and get a set of pages covering that extent.
5437 */
5438 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5439
5440 /*
5441 * May have run out of memory and not clustered backwards.
5442 * off p_offset
5443 * [ pp - 1 ][ pp ]
5444 * [ block ]
5445 * We told bmap off, so we have to adjust the bn accordingly.
5446 */
5447 if (io_off > off) {
5448 bn += btod(io_off - off);
5449 contig -= (io_off - off);
5450 }
5451
5452 /*
5453 * bmap was carefull to tell us the right size so use that.
5454 * There might be unallocated frags at the end.
5455 * LMXXX - bzero the end of the page? We must be writing after EOF.
5456 */
5457 if (io_len > contig) {
5458 ASSERT(io_len - contig < fs->fs_bsize);
5459 io_len -= (io_len - contig);
5460 }
5461
5462 /*
5463 * Handle the case where we are writing the last page after EOF.
5464 *
5465 * XXX - just a patch for i-mt3.
5466 */
5467 if (io_len == 0) {
5468 ASSERT(pp->p_offset >=
5469 (u_offset_t)(roundup(ip->i_size, PAGESIZE)));
5470 io_len = PAGESIZE;
5471 }
5472
5473 bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5474
5475 ULOCKFS_SET_MOD(ITOUL(ip));
5476
5477 bp->b_edev = ip->i_dev;
5478 bp->b_dev = cmpdev(ip->i_dev);
5479 bp->b_blkno = bn;
5480 bp->b_un.b_addr = (caddr_t)0;
5481 bp->b_file = ip->i_vnode;
5482
5483 /*
5484 * File contents of shadow or quota inodes are metadata, and updates
5485 * to these need to be put into a logging transaction. All direct
5486 * callers in UFS do that, but fsflush can come here _before_ the
5487 * normal codepath. An example would be updating ACL information, for
5488 * which the normal codepath would be:
5489 * ufs_si_store()
5490 * ufs_rdwri()
5491 * wrip()
5492 * segmap_release()
5493 * VOP_PUTPAGE()
5494 * Here, fsflush can pick up the dirty page before segmap_release()
5495 * forces it out. If that happens, there's no transaction.
5496 * We therefore need to test whether a transaction exists, and if not
5497 * create one - for fsflush.
5498 */
5499 dotrans =
5500 (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) &&
5501 ((curthread->t_flag & T_DONTBLOCK) == 0) &&
5502 (TRANS_ISTRANS(ufsvfsp)));
5503
5504 if (dotrans) {
5505 curthread->t_flag |= T_DONTBLOCK;
5506 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5507 }
5508 if (TRANS_ISTRANS(ufsvfsp)) {
5509 if ((ip->i_mode & IFMT) == IFSHAD) {
5510 TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5511 } else if (ufsvfsp->vfs_qinod == ip) {
5512 TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5513 0, 0);
5514 }
5515 }
5516 if (dotrans) {
5517 TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5518 curthread->t_flag &= ~T_DONTBLOCK;
5519 }
5520
5521 /* write throttle */
5522
5523 ASSERT(bp->b_iodone == NULL);
5524 bp->b_iodone = (int (*)())ufs_iodone;
5525 mutex_enter(&ip->i_tlock);
5526 ip->i_writes += bp->b_bcount;
5527 mutex_exit(&ip->i_tlock);
5528
5529 if (bp->b_flags & B_ASYNC) {
5530 if (ufsvfsp->vfs_log) {
5531 lufs_write_strategy(ufsvfsp->vfs_log, bp);
5532 } else if (ufsvfsp->vfs_snapshot) {
5533 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5534 } else {
5535 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5536 ub.ub_putasyncs.value.ul++;
5537 (void) bdev_strategy(bp);
5538 lwp_stat_update(LWP_STAT_OUBLK, 1);
5539 }
5540 } else {
5541 if (ufsvfsp->vfs_log) {
5542 lufs_write_strategy(ufsvfsp->vfs_log, bp);
5543 } else if (ufsvfsp->vfs_snapshot) {
5544 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5545 } else {
5546 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5547 ub.ub_putsyncs.value.ul++;
5548 (void) bdev_strategy(bp);
5549 lwp_stat_update(LWP_STAT_OUBLK, 1);
5550 }
5551 err = biowait(bp);
5552 pageio_done(bp);
5553 pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5554 }
5555
5556 pp = NULL;
5557
5558 out:
5559 if (err != 0 && pp != NULL)
5560 pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5561
5562 if (offp)
5563 *offp = io_off;
5564 if (lenp)
5565 *lenp = io_len;
5566 out_trace:
5567 return (err);
5568 }
5569
5570 uint64_t ufs_map_alock_retry_cnt;
5571 uint64_t ufs_map_lockfs_retry_cnt;
5572
5573 /* ARGSUSED */
5574 static int
5575 ufs_map(struct vnode *vp, offset_t off, struct as *as, caddr_t *addrp,
5576 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, struct cred *cr,
5577 caller_context_t *ct)
5578 {
5579 struct segvn_crargs vn_a;
5580 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5581 struct ulockfs *ulp;
5582 int error, sig;
5583 k_sigset_t smask;
5584 caddr_t hint = *addrp;
5585
5586 if (vp->v_flag & VNOMAP) {
5587 error = ENOSYS;
5588 goto out;
5589 }
5590
5591 if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5592 error = ENXIO;
5593 goto out;
5594 }
5595
5596 if (vp->v_type != VREG) {
5597 error = ENODEV;
5598 goto out;
5599 }
5600
5601 retry_map:
5602 *addrp = hint;
5603 /*
5604 * If file is being locked, disallow mapping.
5605 */
5606 if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5607 error = EAGAIN;
5608 goto out;
5609 }
5610
5611 as_rangelock(as);
5612 /*
5613 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5614 * the previous attempt), some other thread could have grabbed
5615 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5616 * would unmap the valid VA range, that is ok.
5617 */
5618 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5619 if (error != 0) {
5620 as_rangeunlock(as);
5621 goto out;
5622 }
5623
5624 /*
5625 * a_lock has to be acquired before entering the lockfs protocol
5626 * because that is the order in which pagefault works. Also we cannot
5627 * block on a_lock here because this waiting writer will prevent
5628 * further readers like ufs_read from progressing and could cause
5629 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5630 * pending.
5631 */
5632 while (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
5633 ufs_map_alock_retry_cnt++;
5634 delay(RETRY_LOCK_DELAY);
5635 }
5636
5637 /*
5638 * We can't hold as->a_lock and wait for lockfs to succeed because
5639 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5640 * instead.
5641 */
5642 if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) {
5643 /*
5644 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5645 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5646 */
5647 ufs_map_lockfs_retry_cnt++;
5648 AS_LOCK_EXIT(as);
5649 as_rangeunlock(as);
5650 if (error == EIO)
5651 goto out;
5652
5653 mutex_enter(&ulp->ul_lock);
5654 while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) {
5655 if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) {
5656 cv_wait(&ulp->ul_cv, &ulp->ul_lock);
5657 } else {
5658 sigintr(&smask, 1);
5659 sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock);
5660 sigunintr(&smask);
5661 if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) &&
5662 !sig) || ufsvfsp->vfs_dontblock) {
5663 mutex_exit(&ulp->ul_lock);
5664 return (EINTR);
5665 }
5666 }
5667 }
5668 mutex_exit(&ulp->ul_lock);
5669 goto retry_map;
5670 }
5671
5672 vn_a.vp = vp;
5673 vn_a.offset = (u_offset_t)off;
5674 vn_a.type = flags & MAP_TYPE;
5675 vn_a.prot = prot;
5676 vn_a.maxprot = maxprot;
5677 vn_a.cred = cr;
5678 vn_a.amp = NULL;
5679 vn_a.flags = flags & ~MAP_TYPE;
5680 vn_a.szc = 0;
5681 vn_a.lgrp_mem_policy_flags = 0;
5682
5683 error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5684 if (ulp)
5685 ufs_lockfs_end(ulp);
5686 as_rangeunlock(as);
5687 out:
5688 return (error);
5689 }
5690
5691 /* ARGSUSED */
5692 static int
5693 ufs_addmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5694 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags,
5695 struct cred *cr, caller_context_t *ct)
5696 {
5697 struct inode *ip = VTOI(vp);
5698
5699 if (vp->v_flag & VNOMAP) {
5700 return (ENOSYS);
5701 }
5702
5703 mutex_enter(&ip->i_tlock);
5704 ip->i_mapcnt += btopr(len);
5705 mutex_exit(&ip->i_tlock);
5706 return (0);
5707 }
5708
5709 /*ARGSUSED*/
5710 static int
5711 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5712 size_t len, uint_t prot, uint_t maxprot, uint_t flags, struct cred *cr,
5713 caller_context_t *ct)
5714 {
5715 struct inode *ip = VTOI(vp);
5716
5717 if (vp->v_flag & VNOMAP) {
5718 return (ENOSYS);
5719 }
5720
5721 mutex_enter(&ip->i_tlock);
5722 ip->i_mapcnt -= btopr(len); /* Count released mappings */
5723 ASSERT(ip->i_mapcnt >= 0);
5724 mutex_exit(&ip->i_tlock);
5725 return (0);
5726 }
5727 /*
5728 * Return the answer requested to poll() for non-device files
5729 */
5730 struct pollhead ufs_pollhd;
5731
5732 /* ARGSUSED */
5733 int
5734 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp,
5735 caller_context_t *ct)
5736 {
5737 struct ufsvfs *ufsvfsp;
5738
5739 /*
5740 * Regular files reject edge-triggered pollers.
5741 * See the comment in fs_poll() for a more detailed explanation.
5742 */
5743 if (ev & POLLET) {
5744 return (EPERM);
5745 }
5746
5747 *revp = 0;
5748 ufsvfsp = VTOI(vp)->i_ufsvfs;
5749
5750 if (!ufsvfsp) {
5751 *revp = POLLHUP;
5752 goto out;
5753 }
5754
5755 if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5756 ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5757 *revp |= POLLERR;
5758
5759 } else {
5760 if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5761 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5762 *revp |= POLLOUT;
5763
5764 if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5765 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5766 *revp |= POLLWRBAND;
5767
5768 if (ev & POLLIN)
5769 *revp |= POLLIN;
5770
5771 if (ev & POLLRDNORM)
5772 *revp |= POLLRDNORM;
5773
5774 if (ev & POLLRDBAND)
5775 *revp |= POLLRDBAND;
5776 }
5777
5778 if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5779 *revp |= POLLPRI;
5780 out:
5781 if (*revp == 0 && ! any) {
5782 *phpp = &ufs_pollhd;
5783 }
5784
5785 return (0);
5786 }
5787
5788 /* ARGSUSED */
5789 static int
5790 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
5791 caller_context_t *ct)
5792 {
5793 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5794 struct ulockfs *ulp = NULL;
5795 struct inode *sip = NULL;
5796 int error;
5797 struct inode *ip = VTOI(vp);
5798 int issync;
5799
5800 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5801 if (error)
5802 return (error);
5803
5804 switch (cmd) {
5805 /*
5806 * Have to handle _PC_NAME_MAX here, because the normal way
5807 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5808 * results in a lock ordering reversal between
5809 * ufs_lockfs_{begin,end}() and
5810 * ufs_thread_{suspend,continue}().
5811 *
5812 * Keep in sync with ufs_statvfs().
5813 */
5814 case _PC_NAME_MAX:
5815 *valp = MAXNAMLEN;
5816 break;
5817
5818 case _PC_FILESIZEBITS:
5819 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5820 *valp = UFS_FILESIZE_BITS;
5821 else
5822 *valp = 32;
5823 break;
5824
5825 case _PC_XATTR_EXISTS:
5826 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5827
5828 error =
5829 ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr);
5830 if (error == 0 && sip != NULL) {
5831 /* Start transaction */
5832 if (ulp) {
5833 TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5834 TOP_RMDIR, TOP_RMDIR_SIZE);
5835 }
5836 /*
5837 * Is directory empty
5838 */
5839 rw_enter(&sip->i_rwlock, RW_WRITER);
5840 rw_enter(&sip->i_contents, RW_WRITER);
5841 if (ufs_xattrdirempty(sip,
5842 sip->i_number, CRED())) {
5843 rw_enter(&ip->i_contents, RW_WRITER);
5844 ufs_unhook_shadow(ip, sip);
5845 rw_exit(&ip->i_contents);
5846
5847 *valp = 0;
5848
5849 } else
5850 *valp = 1;
5851 rw_exit(&sip->i_contents);
5852 rw_exit(&sip->i_rwlock);
5853 if (ulp) {
5854 TRANS_END_CSYNC(ufsvfsp, error, issync,
5855 TOP_RMDIR, TOP_RMDIR_SIZE);
5856 }
5857 VN_RELE(ITOV(sip));
5858 } else if (error == ENOENT) {
5859 *valp = 0;
5860 error = 0;
5861 }
5862 } else {
5863 error = fs_pathconf(vp, cmd, valp, cr, ct);
5864 }
5865 break;
5866
5867 case _PC_ACL_ENABLED:
5868 *valp = _ACL_ACLENT_ENABLED;
5869 break;
5870
5871 case _PC_MIN_HOLE_SIZE:
5872 *valp = (ulong_t)ip->i_fs->fs_bsize;
5873 break;
5874
5875 case _PC_SATTR_ENABLED:
5876 case _PC_SATTR_EXISTS:
5877 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5878 (vp->v_type == VREG || vp->v_type == VDIR);
5879 break;
5880
5881 case _PC_TIMESTAMP_RESOLUTION:
5882 /*
5883 * UFS keeps only microsecond timestamp resolution.
5884 * This is historical and will probably never change.
5885 */
5886 *valp = 1000L;
5887 break;
5888
5889 default:
5890 error = fs_pathconf(vp, cmd, valp, cr, ct);
5891 break;
5892 }
5893
5894 if (ulp != NULL) {
5895 ufs_lockfs_end(ulp);
5896 }
5897 return (error);
5898 }
5899
5900 int ufs_pageio_writes, ufs_pageio_reads;
5901
5902 /*ARGSUSED*/
5903 static int
5904 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5905 int flags, struct cred *cr, caller_context_t *ct)
5906 {
5907 struct inode *ip = VTOI(vp);
5908 struct ufsvfs *ufsvfsp;
5909 page_t *npp = NULL, *opp = NULL, *cpp = pp;
5910 struct buf *bp;
5911 daddr_t bn;
5912 size_t done_len = 0, cur_len = 0;
5913 int err = 0;
5914 int contig = 0;
5915 int dolock;
5916 int vmpss = 0;
5917 struct ulockfs *ulp;
5918
5919 if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5920 vp->v_mpssdata != NULL) {
5921 vmpss = 1;
5922 }
5923
5924 dolock = (rw_owner(&ip->i_contents) != curthread);
5925 /*
5926 * We need a better check. Ideally, we would use another
5927 * vnodeops so that hlocked and forcibly unmounted file
5928 * systems would return EIO where appropriate and w/o the
5929 * need for these checks.
5930 */
5931 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5932 return (EIO);
5933
5934 /*
5935 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5936 * ul_lock must be taken before locking pages so we can't use it here
5937 * if pp is non NULL because segvn already locked pages
5938 * SE_EXCL. Instead we rely on the fact that a forced umount or
5939 * applying a filesystem lock via ufs_fiolfs() will block in the
5940 * implicit call to ufs_flush() until we unlock the pages after the
5941 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5942 * above 0 until they are done. We have to be careful not to increment
5943 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5944 *
5945 * If pp is NULL use ul_lock to make sure we don't increment
5946 * ul_vnops_cnt after forceful unmount hlocks the file system.
5947 */
5948 if (vmpss || pp == NULL) {
5949 ulp = &ufsvfsp->vfs_ulockfs;
5950 if (pp == NULL)
5951 mutex_enter(&ulp->ul_lock);
5952 if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5953 if (pp == NULL) {
5954 mutex_exit(&ulp->ul_lock);
5955 }
5956 return (vmpss ? EIO : EINVAL);
5957 }
5958 atomic_inc_ulong(&ulp->ul_vnops_cnt);
5959 if (pp == NULL)
5960 mutex_exit(&ulp->ul_lock);
5961 if (ufs_quiesce_pend) {
5962 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5963 cv_broadcast(&ulp->ul_cv);
5964 return (vmpss ? EIO : EINVAL);
5965 }
5966 }
5967
5968 if (dolock) {
5969 /*
5970 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
5971 * handle a fault against a segment that maps vnode pages with
5972 * large mappings. Segvn creates pages and holds them locked
5973 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
5974 * use rw_tryenter() to avoid a potential deadlock since in
5975 * lock order i_contents needs to be taken first.
5976 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
5977 */
5978 if (!vmpss) {
5979 rw_enter(&ip->i_contents, RW_READER);
5980 } else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
5981 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5982 cv_broadcast(&ulp->ul_cv);
5983 return (EDEADLK);
5984 }
5985 }
5986
5987 /*
5988 * Return an error to segvn because the pagefault request is beyond
5989 * PAGESIZE rounded EOF.
5990 */
5991 if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
5992 if (dolock)
5993 rw_exit(&ip->i_contents);
5994 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
5995 cv_broadcast(&ulp->ul_cv);
5996 return (EFAULT);
5997 }
5998
5999 if (pp == NULL) {
6000 if (bmap_has_holes(ip)) {
6001 err = ENOSYS;
6002 } else {
6003 err = EINVAL;
6004 }
6005 if (dolock)
6006 rw_exit(&ip->i_contents);
6007 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6008 cv_broadcast(&ulp->ul_cv);
6009 return (err);
6010 }
6011
6012 /*
6013 * Break the io request into chunks, one for each contiguous
6014 * stretch of disk blocks in the target file.
6015 */
6016 while (done_len < io_len) {
6017 ASSERT(cpp);
6018 contig = 0;
6019 if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
6020 &bn, &contig))
6021 break;
6022
6023 if (bn == UFS_HOLE) { /* No holey swapfiles */
6024 if (vmpss) {
6025 err = EFAULT;
6026 break;
6027 }
6028 err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
6029 break;
6030 }
6031
6032 cur_len = MIN(io_len - done_len, contig);
6033 /*
6034 * Zero out a page beyond EOF, when the last block of
6035 * a file is a UFS fragment so that ufs_pageio() can be used
6036 * instead of ufs_getpage() to handle faults against
6037 * segvn segments that use large pages.
6038 */
6039 page_list_break(&cpp, &npp, btopr(cur_len));
6040 if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
6041 size_t xlen = cur_len & PAGEOFFSET;
6042 pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
6043 }
6044
6045 bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
6046 ASSERT(bp != NULL);
6047
6048 bp->b_edev = ip->i_dev;
6049 bp->b_dev = cmpdev(ip->i_dev);
6050 bp->b_blkno = bn;
6051 bp->b_un.b_addr = (caddr_t)0;
6052 bp->b_file = ip->i_vnode;
6053
6054 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
6055 ub.ub_pageios.value.ul++;
6056 if (ufsvfsp->vfs_snapshot)
6057 fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6058 else
6059 (void) bdev_strategy(bp);
6060
6061 if (flags & B_READ)
6062 ufs_pageio_reads++;
6063 else
6064 ufs_pageio_writes++;
6065 if (flags & B_READ)
6066 lwp_stat_update(LWP_STAT_INBLK, 1);
6067 else
6068 lwp_stat_update(LWP_STAT_OUBLK, 1);
6069 /*
6070 * If the request is not B_ASYNC, wait for i/o to complete
6071 * and re-assemble the page list to return to the caller.
6072 * If it is B_ASYNC we leave the page list in pieces and
6073 * cleanup() will dispose of them.
6074 */
6075 if ((flags & B_ASYNC) == 0) {
6076 err = biowait(bp);
6077 pageio_done(bp);
6078 if (err)
6079 break;
6080 page_list_concat(&opp, &cpp);
6081 }
6082 cpp = npp;
6083 npp = NULL;
6084 if (flags & B_READ)
6085 cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6086 done_len += cur_len;
6087 }
6088 ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6089 if (err) {
6090 if (flags & B_ASYNC) {
6091 /* Cleanup unprocessed parts of list */
6092 page_list_concat(&cpp, &npp);
6093 if (flags & B_READ)
6094 pvn_read_done(cpp, B_ERROR);
6095 else
6096 pvn_write_done(cpp, B_ERROR);
6097 } else {
6098 /* Re-assemble list and let caller clean up */
6099 page_list_concat(&opp, &cpp);
6100 page_list_concat(&opp, &npp);
6101 }
6102 }
6103
6104 if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6105 ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6106 mutex_enter(&ip->i_tlock);
6107 ip->i_flag |= IACC;
6108 ITIMES_NOLOCK(ip);
6109 mutex_exit(&ip->i_tlock);
6110 }
6111
6112 if (dolock)
6113 rw_exit(&ip->i_contents);
6114 if (vmpss && !atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6115 cv_broadcast(&ulp->ul_cv);
6116 return (err);
6117 }
6118
6119 /*
6120 * Called when the kernel is in a frozen state to dump data
6121 * directly to the device. It uses a private dump data structure,
6122 * set up by dump_ctl, to locate the correct disk block to which to dump.
6123 */
6124 /*ARGSUSED*/
6125 static int
6126 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks,
6127 caller_context_t *ct)
6128 {
6129 u_offset_t file_size;
6130 struct inode *ip = VTOI(vp);
6131 struct fs *fs = ip->i_fs;
6132 daddr_t dbn, lfsbn;
6133 int disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6134 int error = 0;
6135 int ndbs, nfsbs;
6136
6137 /*
6138 * forced unmount case
6139 */
6140 if (ip->i_ufsvfs == NULL)
6141 return (EIO);
6142 /*
6143 * Validate the inode that it has not been modified since
6144 * the dump structure is allocated.
6145 */
6146 mutex_enter(&ip->i_tlock);
6147 if ((dump_info == NULL) ||
6148 (dump_info->ip != ip) ||
6149 (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6150 (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6151 mutex_exit(&ip->i_tlock);
6152 return (-1);
6153 }
6154 mutex_exit(&ip->i_tlock);
6155
6156 /*
6157 * See that the file has room for this write
6158 */
6159 UFS_GET_ISIZE(&file_size, ip);
6160
6161 if (ldbtob(ldbn + dblks) > file_size)
6162 return (ENOSPC);
6163
6164 /*
6165 * Find the physical disk block numbers from the dump
6166 * private data structure directly and write out the data
6167 * in contiguous block lumps
6168 */
6169 while (dblks > 0 && !error) {
6170 lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn));
6171 dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6172 nfsbs = 1;
6173 ndbs = disk_blks - ldbn % disk_blks;
6174 while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6175 nfsbs]) == dbn + ndbs) {
6176 nfsbs++;
6177 ndbs += disk_blks;
6178 }
6179 if (ndbs > dblks)
6180 ndbs = dblks;
6181 error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6182 addr += ldbtob((offset_t)ndbs);
6183 dblks -= ndbs;
6184 ldbn += ndbs;
6185 }
6186 return (error);
6187
6188 }
6189
6190 /*
6191 * Prepare the file system before and after the dump operation.
6192 *
6193 * action = DUMP_ALLOC:
6194 * Preparation before dump, allocate dump private data structure
6195 * to hold all the direct and indirect block info for dump.
6196 *
6197 * action = DUMP_FREE:
6198 * Clean up after dump, deallocate the dump private data structure.
6199 *
6200 * action = DUMP_SCAN:
6201 * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6202 * if found, the starting file-relative DEV_BSIZE lbn is written
6203 * to *bklp; that lbn is intended for use with VOP_DUMP()
6204 */
6205 /*ARGSUSED*/
6206 static int
6207 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct)
6208 {
6209 struct inode *ip = VTOI(vp);
6210 ufsvfs_t *ufsvfsp = ip->i_ufsvfs;
6211 struct fs *fs;
6212 daddr32_t *dblk, *storeblk;
6213 daddr32_t *nextblk, *endblk;
6214 struct buf *bp;
6215 int i, entry, entries;
6216 int n, ncontig;
6217
6218 /*
6219 * check for forced unmount
6220 */
6221 if (ufsvfsp == NULL)
6222 return (EIO);
6223
6224 if (action == DUMP_ALLOC) {
6225 /*
6226 * alloc and record dump_info
6227 */
6228 if (dump_info != NULL)
6229 return (EINVAL);
6230
6231 ASSERT(vp->v_type == VREG);
6232 fs = ufsvfsp->vfs_fs;
6233
6234 rw_enter(&ip->i_contents, RW_READER);
6235
6236 if (bmap_has_holes(ip)) {
6237 rw_exit(&ip->i_contents);
6238 return (EFAULT);
6239 }
6240
6241 /*
6242 * calculate and allocate space needed according to i_size
6243 */
6244 entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6245 dump_info = kmem_alloc(sizeof (struct dump) +
6246 (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP);
6247 if (dump_info == NULL) {
6248 rw_exit(&ip->i_contents);
6249 return (ENOMEM);
6250 }
6251
6252 /* Start saving the info */
6253 dump_info->fsbs = entries;
6254 dump_info->ip = ip;
6255 storeblk = &dump_info->dblk[0];
6256
6257 /* Direct Blocks */
6258 for (entry = 0; entry < NDADDR && entry < entries; entry++)
6259 *storeblk++ = ip->i_db[entry];
6260
6261 /* Indirect Blocks */
6262 for (i = 0; i < NIADDR; i++) {
6263 int error = 0;
6264
6265 bp = UFS_BREAD(ufsvfsp,
6266 ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize);
6267 if (bp->b_flags & B_ERROR)
6268 error = EIO;
6269 else {
6270 dblk = bp->b_un.b_daddr;
6271 if ((storeblk = save_dblks(ip, ufsvfsp,
6272 storeblk, dblk, i, entries)) == NULL)
6273 error = EIO;
6274 }
6275
6276 brelse(bp);
6277
6278 if (error != 0) {
6279 kmem_free(dump_info, sizeof (struct dump) +
6280 (entries - 1) * sizeof (daddr32_t));
6281 rw_exit(&ip->i_contents);
6282 dump_info = NULL;
6283 return (error);
6284 }
6285 }
6286 /* and time stamp the information */
6287 mutex_enter(&ip->i_tlock);
6288 dump_info->time = ip->i_mtime;
6289 mutex_exit(&ip->i_tlock);
6290
6291 rw_exit(&ip->i_contents);
6292 } else if (action == DUMP_FREE) {
6293 /*
6294 * free dump_info
6295 */
6296 if (dump_info == NULL)
6297 return (EINVAL);
6298 entries = dump_info->fsbs - 1;
6299 kmem_free(dump_info, sizeof (struct dump) +
6300 entries * sizeof (daddr32_t));
6301 dump_info = NULL;
6302 } else if (action == DUMP_SCAN) {
6303 /*
6304 * scan dump_info
6305 */
6306 if (dump_info == NULL)
6307 return (EINVAL);
6308
6309 dblk = dump_info->dblk;
6310 nextblk = dblk + 1;
6311 endblk = dblk + dump_info->fsbs - 1;
6312 fs = ufsvfsp->vfs_fs;
6313 ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6314
6315 /*
6316 * scan dblk[] entries; contig fs space is found when:
6317 * ((current blkno + frags per block) == next blkno)
6318 */
6319 n = 0;
6320 while (n < ncontig && dblk < endblk) {
6321 if ((*dblk + fs->fs_frag) == *nextblk)
6322 n++;
6323 else
6324 n = 0;
6325 dblk++;
6326 nextblk++;
6327 }
6328
6329 /*
6330 * index is where size bytes of contig space begins;
6331 * conversion from index to the file's DEV_BSIZE lbn
6332 * is equivalent to: (index * fs_bsize) / DEV_BSIZE
6333 */
6334 if (n == ncontig) {
6335 i = (dblk - dump_info->dblk) - ncontig;
6336 *blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6337 } else
6338 return (EFAULT);
6339 }
6340 return (0);
6341 }
6342
6343 /*
6344 * Recursive helper function for ufs_dumpctl(). It follows the indirect file
6345 * system blocks until it reaches the the disk block addresses, which are
6346 * then stored into the given buffer, storeblk.
6347 */
6348 static daddr32_t *
6349 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp, daddr32_t *storeblk,
6350 daddr32_t *dblk, int level, int entries)
6351 {
6352 struct fs *fs = ufsvfsp->vfs_fs;
6353 struct buf *bp;
6354 int i;
6355
6356 if (level == 0) {
6357 for (i = 0; i < NINDIR(fs); i++) {
6358 if (storeblk - dump_info->dblk >= entries)
6359 break;
6360 *storeblk++ = dblk[i];
6361 }
6362 return (storeblk);
6363 }
6364 for (i = 0; i < NINDIR(fs); i++) {
6365 if (storeblk - dump_info->dblk >= entries)
6366 break;
6367 bp = UFS_BREAD(ufsvfsp,
6368 ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6369 if (bp->b_flags & B_ERROR) {
6370 brelse(bp);
6371 return (NULL);
6372 }
6373 storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6374 level - 1, entries);
6375 brelse(bp);
6376
6377 if (storeblk == NULL)
6378 return (NULL);
6379 }
6380 return (storeblk);
6381 }
6382
6383 /* ARGSUSED */
6384 static int
6385 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6386 struct cred *cr, caller_context_t *ct)
6387 {
6388 struct inode *ip = VTOI(vp);
6389 struct ulockfs *ulp;
6390 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
6391 ulong_t vsa_mask = vsap->vsa_mask;
6392 int err = EINVAL;
6393
6394 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6395
6396 /*
6397 * Only grab locks if needed - they're not needed to check vsa_mask
6398 * or if the mask contains no acl flags.
6399 */
6400 if (vsa_mask != 0) {
6401 if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6402 ULOCKFS_GETATTR_MASK))
6403 return (err);
6404
6405 rw_enter(&ip->i_contents, RW_READER);
6406 err = ufs_acl_get(ip, vsap, flag, cr);
6407 rw_exit(&ip->i_contents);
6408
6409 if (ulp)
6410 ufs_lockfs_end(ulp);
6411 }
6412 return (err);
6413 }
6414
6415 /* ARGSUSED */
6416 static int
6417 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr,
6418 caller_context_t *ct)
6419 {
6420 struct inode *ip = VTOI(vp);
6421 struct ulockfs *ulp = NULL;
6422 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
6423 ulong_t vsa_mask = vsap->vsa_mask;
6424 int err;
6425 int haverwlock = 1;
6426 int trans_size;
6427 int donetrans = 0;
6428 int retry = 1;
6429
6430 ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6431
6432 /* Abort now if the request is either empty or invalid. */
6433 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6434 if ((vsa_mask == 0) ||
6435 ((vsap->vsa_aclentp == NULL) &&
6436 (vsap->vsa_dfaclentp == NULL))) {
6437 err = EINVAL;
6438 goto out;
6439 }
6440
6441 /*
6442 * Following convention, if this is a directory then we acquire the
6443 * inode's i_rwlock after starting a UFS logging transaction;
6444 * otherwise, we acquire it beforehand. Since we were called (and
6445 * must therefore return) with the lock held, we will have to drop it,
6446 * and later reacquire it, if operating on a directory.
6447 */
6448 if (vp->v_type == VDIR) {
6449 rw_exit(&ip->i_rwlock);
6450 haverwlock = 0;
6451 } else {
6452 /* Upgrade the lock if required. */
6453 if (!rw_write_held(&ip->i_rwlock)) {
6454 rw_exit(&ip->i_rwlock);
6455 rw_enter(&ip->i_rwlock, RW_WRITER);
6456 }
6457 }
6458
6459 again:
6460 ASSERT(!(vp->v_type == VDIR && haverwlock));
6461 if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6462 ulp = NULL;
6463 retry = 0;
6464 goto out;
6465 }
6466
6467 /*
6468 * Check that the file system supports this operation. Note that
6469 * ufs_lockfs_begin() will have checked that the file system had
6470 * not been forcibly unmounted.
6471 */
6472 if (ufsvfsp->vfs_fs->fs_ronly) {
6473 err = EROFS;
6474 goto out;
6475 }
6476 if (ufsvfsp->vfs_nosetsec) {
6477 err = ENOSYS;
6478 goto out;
6479 }
6480
6481 if (ulp) {
6482 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6483 trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6484 donetrans = 1;
6485 }
6486
6487 if (vp->v_type == VDIR) {
6488 rw_enter(&ip->i_rwlock, RW_WRITER);
6489 haverwlock = 1;
6490 }
6491
6492 ASSERT(haverwlock);
6493
6494 /* Do the actual work. */
6495 rw_enter(&ip->i_contents, RW_WRITER);
6496 /*
6497 * Suppress out of inodes messages if we will retry.
6498 */
6499 if (retry)
6500 ip->i_flag |= IQUIET;
6501 err = ufs_acl_set(ip, vsap, flag, cr);
6502 ip->i_flag &= ~IQUIET;
6503 rw_exit(&ip->i_contents);
6504
6505 out:
6506 if (ulp) {
6507 if (donetrans) {
6508 /*
6509 * top_end_async() can eventually call
6510 * top_end_sync(), which can block. We must
6511 * therefore observe the lock-ordering protocol
6512 * here as well.
6513 */
6514 if (vp->v_type == VDIR) {
6515 rw_exit(&ip->i_rwlock);
6516 haverwlock = 0;
6517 }
6518 TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6519 }
6520 ufs_lockfs_end(ulp);
6521 }
6522 /*
6523 * If no inodes available, try scaring a logically-
6524 * free one out of the delete queue to someplace
6525 * that we can find it.
6526 */
6527 if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6528 ufs_delete_drain_wait(ufsvfsp, 1);
6529 retry = 0;
6530 if (vp->v_type == VDIR && haverwlock) {
6531 rw_exit(&ip->i_rwlock);
6532 haverwlock = 0;
6533 }
6534 goto again;
6535 }
6536 /*
6537 * If we need to reacquire the lock then it is safe to do so
6538 * as a reader. This is because ufs_rwunlock(), which will be
6539 * called by our caller after we return, does not differentiate
6540 * between shared and exclusive locks.
6541 */
6542 if (!haverwlock) {
6543 ASSERT(vp->v_type == VDIR);
6544 rw_enter(&ip->i_rwlock, RW_READER);
6545 }
6546
6547 return (err);
6548 }
6549
6550 /*
6551 * Locate the vnode to be used for an event notification. As this will
6552 * be called prior to the name space change perform basic verification
6553 * that the change will be allowed.
6554 */
6555
6556 static int
6557 ufs_eventlookup(struct vnode *dvp, char *nm, struct cred *cr,
6558 struct vnode **vpp)
6559 {
6560 int namlen;
6561 int error;
6562 struct vnode *vp;
6563 struct inode *ip;
6564 struct inode *xip;
6565 struct ufsvfs *ufsvfsp;
6566 struct ulockfs *ulp;
6567
6568 ip = VTOI(dvp);
6569 *vpp = NULL;
6570
6571 if ((namlen = strlen(nm)) == 0)
6572 return (EINVAL);
6573
6574 if (nm[0] == '.') {
6575 if (namlen == 1)
6576 return (EINVAL);
6577 else if ((namlen == 2) && nm[1] == '.') {
6578 return (EEXIST);
6579 }
6580 }
6581
6582 /*
6583 * Check accessibility and write access of parent directory as we
6584 * only want to post the event if we're able to make a change.
6585 */
6586 if (error = ufs_diraccess(ip, IEXEC|IWRITE, cr))
6587 return (error);
6588
6589 if (vp = dnlc_lookup(dvp, nm)) {
6590 if (vp == DNLC_NO_VNODE) {
6591 VN_RELE(vp);
6592 return (ENOENT);
6593 }
6594
6595 *vpp = vp;
6596 return (0);
6597 }
6598
6599 /*
6600 * Keep the idle queue from getting too long by idling two
6601 * inodes before attempting to allocate another.
6602 * This operation must be performed before entering lockfs
6603 * or a transaction.
6604 */
6605 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
6606 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
6607 ins.in_lidles.value.ul += ufs_lookup_idle_count;
6608 ufs_idle_some(ufs_lookup_idle_count);
6609 }
6610
6611 ufsvfsp = ip->i_ufsvfs;
6612
6613 retry_lookup:
6614 if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK))
6615 return (error);
6616
6617 if ((error = ufs_dirlook(ip, nm, &xip, cr, 1, 1)) == 0) {
6618 vp = ITOV(xip);
6619 *vpp = vp;
6620 }
6621
6622 if (ulp) {
6623 ufs_lockfs_end(ulp);
6624 }
6625
6626 if (error == EAGAIN)
6627 goto retry_lookup;
6628
6629 return (error);
6630 }