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 2016, Joyent, Inc.
25 */
26
27 /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */
28 /* All Rights Reserved */
29
30 /*
31 * Portions of this source code were derived from Berkeley 4.3 BSD
32 * under license from the Regents of the University of California.
33 */
34
35 #include <sys/types.h>
36 #include <sys/t_lock.h>
37 #include <sys/ksynch.h>
38 #include <sys/param.h>
39 #include <sys/time.h>
40 #include <sys/systm.h>
41 #include <sys/sysmacros.h>
42 #include <sys/resource.h>
43 #include <sys/signal.h>
44 #include <sys/cred.h>
45 #include <sys/user.h>
46 #include <sys/buf.h>
47 #include <sys/vfs.h>
48 #include <sys/vfs_opreg.h>
49 #include <sys/vnode.h>
50 #include <sys/proc.h>
51 #include <sys/disp.h>
52 #include <sys/file.h>
53 #include <sys/fcntl.h>
54 #include <sys/flock.h>
55 #include <sys/atomic.h>
56 #include <sys/kmem.h>
57 #include <sys/uio.h>
58 #include <sys/dnlc.h>
59 #include <sys/conf.h>
60 #include <sys/mman.h>
61 #include <sys/pathname.h>
62 #include <sys/debug.h>
63 #include <sys/vmsystm.h>
64 #include <sys/cmn_err.h>
65 #include <sys/filio.h>
66 #include <sys/policy.h>
67
68 #include <sys/fs/ufs_fs.h>
69 #include <sys/fs/ufs_lockfs.h>
70 #include <sys/fs/ufs_filio.h>
71 #include <sys/fs/ufs_inode.h>
72 #include <sys/fs/ufs_fsdir.h>
73 #include <sys/fs/ufs_quota.h>
74 #include <sys/fs/ufs_log.h>
75 #include <sys/fs/ufs_snap.h>
76 #include <sys/fs/ufs_trans.h>
77 #include <sys/fs/ufs_panic.h>
78 #include <sys/fs/ufs_bio.h>
79 #include <sys/dirent.h> /* must be AFTER <sys/fs/fsdir.h>! */
80 #include <sys/errno.h>
81 #include <sys/fssnap_if.h>
82 #include <sys/unistd.h>
83 #include <sys/sunddi.h>
84
85 #include <sys/filio.h> /* _FIOIO */
86
87 #include <vm/hat.h>
88 #include <vm/page.h>
89 #include <vm/pvn.h>
90 #include <vm/as.h>
91 #include <vm/seg.h>
92 #include <vm/seg_map.h>
93 #include <vm/seg_vn.h>
94 #include <vm/seg_kmem.h>
95 #include <vm/rm.h>
96 #include <sys/swap.h>
97
98 #include <fs/fs_subr.h>
99
100 #include <sys/fs/decomp.h>
101
102 static struct instats ins;
103
104 static int ufs_getpage_ra(struct vnode *, u_offset_t, struct seg *, caddr_t);
105 static int ufs_getpage_miss(struct vnode *, u_offset_t, size_t, struct seg *,
106 caddr_t, struct page **, size_t, enum seg_rw, int);
107 static int ufs_open(struct vnode **, int, struct cred *, caller_context_t *);
108 static int ufs_close(struct vnode *, int, int, offset_t, struct cred *,
109 caller_context_t *);
110 static int ufs_read(struct vnode *, struct uio *, int, struct cred *,
111 struct caller_context *);
112 static int ufs_write(struct vnode *, struct uio *, int, struct cred *,
113 struct caller_context *);
114 static int ufs_ioctl(struct vnode *, int, intptr_t, int, struct cred *,
115 int *, caller_context_t *);
116 static int ufs_getattr(struct vnode *, struct vattr *, int, struct cred *,
117 caller_context_t *);
118 static int ufs_setattr(struct vnode *, struct vattr *, int, struct cred *,
119 caller_context_t *);
120 static int ufs_access(struct vnode *, int, int, struct cred *,
121 caller_context_t *);
122 static int ufs_lookup(struct vnode *, char *, struct vnode **,
123 struct pathname *, int, struct vnode *, struct cred *,
124 caller_context_t *, int *, pathname_t *);
125 static int ufs_create(struct vnode *, char *, struct vattr *, enum vcexcl,
126 int, struct vnode **, struct cred *, int,
127 caller_context_t *, vsecattr_t *);
128 static int ufs_remove(struct vnode *, char *, struct cred *,
129 caller_context_t *, int);
130 static int ufs_link(struct vnode *, struct vnode *, char *, struct cred *,
131 caller_context_t *, int);
132 static int ufs_rename(struct vnode *, char *, struct vnode *, char *,
133 struct cred *, caller_context_t *, int);
134 static int ufs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
135 struct cred *, caller_context_t *, int, vsecattr_t *);
136 static int ufs_rmdir(struct vnode *, char *, struct vnode *, struct cred *,
137 caller_context_t *, int);
138 static int ufs_readdir(struct vnode *, struct uio *, struct cred *, int *,
139 caller_context_t *, int);
140 static int ufs_symlink(struct vnode *, char *, struct vattr *, char *,
141 struct cred *, caller_context_t *, int);
142 static int ufs_readlink(struct vnode *, struct uio *, struct cred *,
143 caller_context_t *);
144 static int ufs_fsync(struct vnode *, int, struct cred *, caller_context_t *);
145 static void ufs_inactive(struct vnode *, struct cred *, caller_context_t *);
146 static int ufs_fid(struct vnode *, struct fid *, caller_context_t *);
147 static int ufs_rwlock(struct vnode *, int, caller_context_t *);
148 static void ufs_rwunlock(struct vnode *, int, caller_context_t *);
149 static int ufs_seek(struct vnode *, offset_t, offset_t *, caller_context_t *);
150 static int ufs_frlock(struct vnode *, int, struct flock64 *, int, offset_t,
151 struct flk_callback *, struct cred *,
152 caller_context_t *);
153 static int ufs_space(struct vnode *, int, struct flock64 *, int, offset_t,
154 cred_t *, caller_context_t *);
155 static int ufs_getpage(struct vnode *, offset_t, size_t, uint_t *,
156 struct page **, size_t, struct seg *, caddr_t,
157 enum seg_rw, struct cred *, caller_context_t *);
158 static int ufs_putpage(struct vnode *, offset_t, size_t, int, struct cred *,
159 caller_context_t *);
160 static int ufs_putpages(struct vnode *, offset_t, size_t, int, struct cred *);
161 static int ufs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
162 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
163 static int ufs_addmap(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_delmap(struct vnode *, offset_t, struct as *, caddr_t, size_t,
166 uint_t, uint_t, uint_t, struct cred *, caller_context_t *);
167 static int ufs_poll(vnode_t *, short, int, short *, struct pollhead **,
168 caller_context_t *);
169 static int ufs_dump(vnode_t *, caddr_t, offset_t, offset_t,
170 caller_context_t *);
171 static int ufs_l_pathconf(struct vnode *, int, ulong_t *, struct cred *,
172 caller_context_t *);
173 static int ufs_pageio(struct vnode *, struct page *, u_offset_t, size_t, int,
174 struct cred *, caller_context_t *);
175 static int ufs_dumpctl(vnode_t *, int, offset_t *, caller_context_t *);
176 static daddr32_t *save_dblks(struct inode *, struct ufsvfs *, daddr32_t *,
177 daddr32_t *, int, int);
178 static int ufs_getsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
179 caller_context_t *);
180 static int ufs_setsecattr(struct vnode *, vsecattr_t *, int, struct cred *,
181 caller_context_t *);
182 static int ufs_priv_access(void *, int, struct cred *);
183 static int ufs_eventlookup(struct vnode *, char *, struct cred *,
184 struct vnode **);
185 extern int as_map_locked(struct as *, caddr_t, size_t, int ((*)()), void *);
186
187 /*
188 * For lockfs: ulockfs begin/end is now inlined in the ufs_xxx functions.
189 *
190 * XXX - ULOCKFS in fs_pathconf and ufs_ioctl is not inlined yet.
191 */
192 struct vnodeops *ufs_vnodeops;
193
194 /* NOTE: "not blkd" below means that the operation isn't blocked by lockfs */
195 const fs_operation_def_t ufs_vnodeops_template[] = {
196 VOPNAME_OPEN, { .vop_open = ufs_open }, /* not blkd */
197 VOPNAME_CLOSE, { .vop_close = ufs_close }, /* not blkd */
198 VOPNAME_READ, { .vop_read = ufs_read },
199 VOPNAME_WRITE, { .vop_write = ufs_write },
200 VOPNAME_IOCTL, { .vop_ioctl = ufs_ioctl },
201 VOPNAME_GETATTR, { .vop_getattr = ufs_getattr },
202 VOPNAME_SETATTR, { .vop_setattr = ufs_setattr },
203 VOPNAME_ACCESS, { .vop_access = ufs_access },
204 VOPNAME_LOOKUP, { .vop_lookup = ufs_lookup },
205 VOPNAME_CREATE, { .vop_create = ufs_create },
206 VOPNAME_REMOVE, { .vop_remove = ufs_remove },
207 VOPNAME_LINK, { .vop_link = ufs_link },
208 VOPNAME_RENAME, { .vop_rename = ufs_rename },
209 VOPNAME_MKDIR, { .vop_mkdir = ufs_mkdir },
210 VOPNAME_RMDIR, { .vop_rmdir = ufs_rmdir },
211 VOPNAME_READDIR, { .vop_readdir = ufs_readdir },
212 VOPNAME_SYMLINK, { .vop_symlink = ufs_symlink },
213 VOPNAME_READLINK, { .vop_readlink = ufs_readlink },
214 VOPNAME_FSYNC, { .vop_fsync = ufs_fsync },
215 VOPNAME_INACTIVE, { .vop_inactive = ufs_inactive }, /* not blkd */
216 VOPNAME_FID, { .vop_fid = ufs_fid },
217 VOPNAME_RWLOCK, { .vop_rwlock = ufs_rwlock }, /* not blkd */
218 VOPNAME_RWUNLOCK, { .vop_rwunlock = ufs_rwunlock }, /* not blkd */
219 VOPNAME_SEEK, { .vop_seek = ufs_seek },
220 VOPNAME_FRLOCK, { .vop_frlock = ufs_frlock },
221 VOPNAME_SPACE, { .vop_space = ufs_space },
222 VOPNAME_GETPAGE, { .vop_getpage = ufs_getpage },
223 VOPNAME_PUTPAGE, { .vop_putpage = ufs_putpage },
224 VOPNAME_MAP, { .vop_map = ufs_map },
225 VOPNAME_ADDMAP, { .vop_addmap = ufs_addmap }, /* not blkd */
226 VOPNAME_DELMAP, { .vop_delmap = ufs_delmap }, /* not blkd */
227 VOPNAME_POLL, { .vop_poll = ufs_poll }, /* not blkd */
228 VOPNAME_DUMP, { .vop_dump = ufs_dump },
229 VOPNAME_PATHCONF, { .vop_pathconf = ufs_l_pathconf },
230 VOPNAME_PAGEIO, { .vop_pageio = ufs_pageio },
231 VOPNAME_DUMPCTL, { .vop_dumpctl = ufs_dumpctl },
232 VOPNAME_GETSECATTR, { .vop_getsecattr = ufs_getsecattr },
233 VOPNAME_SETSECATTR, { .vop_setsecattr = ufs_setsecattr },
234 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
235 NULL, NULL
236 };
237
238 #define MAX_BACKFILE_COUNT 9999
239
240 /*
241 * Created by ufs_dumpctl() to store a file's disk block info into memory.
242 * Used by ufs_dump() to dump data to disk directly.
243 */
244 struct dump {
245 struct inode *ip; /* the file we contain */
246 daddr_t fsbs; /* number of blocks stored */
247 struct timeval32 time; /* time stamp for the struct */
248 daddr32_t dblk[1]; /* place holder for block info */
249 };
250
251 static struct dump *dump_info = NULL;
252
253 /*
254 * Previously there was no special action required for ordinary files.
255 * (Devices are handled through the device file system.)
256 * Now we support Large Files and Large File API requires open to
257 * fail if file is large.
258 * We could take care to prevent data corruption
259 * by doing an atomic check of size and truncate if file is opened with
260 * FTRUNC flag set but traditionally this is being done by the vfs/vnode
261 * layers. So taking care of truncation here is a change in the existing
262 * semantics of VOP_OPEN and therefore we chose not to implement any thing
263 * here. The check for the size of the file > 2GB is being done at the
264 * vfs layer in routine vn_open().
265 */
266
267 /* ARGSUSED */
268 static int
269 ufs_open(struct vnode **vpp, int flag, struct cred *cr, caller_context_t *ct)
270 {
271 return (0);
272 }
273
274 /*ARGSUSED*/
275 static int
276 ufs_close(struct vnode *vp, int flag, int count, offset_t offset,
277 struct cred *cr, caller_context_t *ct)
278 {
279 cleanlocks(vp, ttoproc(curthread)->p_pid, 0);
280 cleanshares(vp, ttoproc(curthread)->p_pid);
281
282 /*
283 * Push partially filled cluster at last close.
284 * ``last close'' is approximated because the dnlc
285 * may have a hold on the vnode.
286 * Checking for VBAD here will also act as a forced umount check.
287 */
288 if (vp->v_count <= 2 && vp->v_type != VBAD) {
289 struct inode *ip = VTOI(vp);
290 if (ip->i_delaylen) {
291 ins.in_poc.value.ul++;
292 (void) ufs_putpages(vp, ip->i_delayoff, ip->i_delaylen,
293 B_ASYNC | B_FREE, cr);
294 ip->i_delaylen = 0;
295 }
296 }
297
298 return (0);
299 }
300
301 /*ARGSUSED*/
302 static int
303 ufs_read(struct vnode *vp, struct uio *uiop, int ioflag, struct cred *cr,
304 struct caller_context *ct)
305 {
306 struct inode *ip = VTOI(vp);
307 struct ufsvfs *ufsvfsp;
308 struct ulockfs *ulp = NULL;
309 int error = 0;
310 int intrans = 0;
311
312 ASSERT(RW_READ_HELD(&ip->i_rwlock));
313
314 /*
315 * Mandatory locking needs to be done before ufs_lockfs_begin()
316 * and TRANS_BEGIN_SYNC() calls since mandatory locks can sleep.
317 */
318 if (MANDLOCK(vp, ip->i_mode)) {
319 /*
320 * ufs_getattr ends up being called by chklock
321 */
322 error = chklock(vp, FREAD, uiop->uio_loffset,
323 uiop->uio_resid, uiop->uio_fmode, ct);
324 if (error)
325 goto out;
326 }
327
328 ufsvfsp = ip->i_ufsvfs;
329 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READ_MASK);
330 if (error)
331 goto out;
332
333 /*
334 * In the case that a directory is opened for reading as a file
335 * (eg "cat .") with the O_RSYNC, O_SYNC and O_DSYNC flags set.
336 * The locking order had to be changed to avoid a deadlock with
337 * an update taking place on that directory at the same time.
338 */
339 if ((ip->i_mode & IFMT) == IFDIR) {
340
341 rw_enter(&ip->i_contents, RW_READER);
342 error = rdip(ip, uiop, ioflag, cr);
343 rw_exit(&ip->i_contents);
344
345 if (error) {
346 if (ulp)
347 ufs_lockfs_end(ulp);
348 goto out;
349 }
350
351 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
352 TRANS_ISTRANS(ufsvfsp)) {
353 rw_exit(&ip->i_rwlock);
354 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
355 error);
356 ASSERT(!error);
357 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
358 TOP_READ_SIZE);
359 rw_enter(&ip->i_rwlock, RW_READER);
360 }
361 } else {
362 /*
363 * Only transact reads to files opened for sync-read and
364 * sync-write on a file system that is not write locked.
365 *
366 * The ``not write locked'' check prevents problems with
367 * enabling/disabling logging on a busy file system. E.g.,
368 * logging exists at the beginning of the read but does not
369 * at the end.
370 *
371 */
372 if (ulp && (ioflag & FRSYNC) && (ioflag & (FSYNC | FDSYNC)) &&
373 TRANS_ISTRANS(ufsvfsp)) {
374 TRANS_BEGIN_SYNC(ufsvfsp, TOP_READ_SYNC, TOP_READ_SIZE,
375 error);
376 ASSERT(!error);
377 intrans = 1;
378 }
379
380 rw_enter(&ip->i_contents, RW_READER);
381 error = rdip(ip, uiop, ioflag, cr);
382 rw_exit(&ip->i_contents);
383
384 if (intrans) {
385 TRANS_END_SYNC(ufsvfsp, error, TOP_READ_SYNC,
386 TOP_READ_SIZE);
387 }
388 }
389
390 if (ulp) {
391 ufs_lockfs_end(ulp);
392 }
393 out:
394
395 return (error);
396 }
397
398 extern int ufs_HW; /* high water mark */
399 extern int ufs_LW; /* low water mark */
400 int ufs_WRITES = 1; /* XXX - enable/disable */
401 int ufs_throttles = 0; /* throttling count */
402 int ufs_allow_shared_writes = 1; /* directio shared writes */
403
404 static int
405 ufs_check_rewrite(struct inode *ip, struct uio *uiop, int ioflag)
406 {
407 int shared_write;
408
409 /*
410 * If the FDSYNC flag is set then ignore the global
411 * ufs_allow_shared_writes in this case.
412 */
413 shared_write = (ioflag & FDSYNC) | ufs_allow_shared_writes;
414
415 /*
416 * Filter to determine if this request is suitable as a
417 * concurrent rewrite. This write must not allocate blocks
418 * by extending the file or filling in holes. No use trying
419 * through FSYNC descriptors as the inode will be synchronously
420 * updated after the write. The uio structure has not yet been
421 * checked for sanity, so assume nothing.
422 */
423 return (((ip->i_mode & IFMT) == IFREG) && !(ioflag & FAPPEND) &&
424 (uiop->uio_loffset >= (offset_t)0) &&
425 (uiop->uio_loffset < ip->i_size) && (uiop->uio_resid > 0) &&
426 ((ip->i_size - uiop->uio_loffset) >= uiop->uio_resid) &&
427 !(ioflag & FSYNC) && !bmap_has_holes(ip) &&
428 shared_write);
429 }
430
431 /*ARGSUSED*/
432 static int
433 ufs_write(struct vnode *vp, struct uio *uiop, int ioflag, cred_t *cr,
434 caller_context_t *ct)
435 {
436 struct inode *ip = VTOI(vp);
437 struct ufsvfs *ufsvfsp;
438 struct ulockfs *ulp;
439 int retry = 1;
440 int error, resv, resid = 0;
441 int directio_status;
442 int exclusive;
443 int rewriteflg;
444 long start_resid = uiop->uio_resid;
445
446 ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
447
448 retry_mandlock:
449 /*
450 * Mandatory locking needs to be done before ufs_lockfs_begin()
451 * and TRANS_BEGIN_[A]SYNC() calls since mandatory locks can sleep.
452 * Check for forced unmounts normally done in ufs_lockfs_begin().
453 */
454 if ((ufsvfsp = ip->i_ufsvfs) == NULL) {
455 error = EIO;
456 goto out;
457 }
458 if (MANDLOCK(vp, ip->i_mode)) {
459
460 ASSERT(RW_WRITE_HELD(&ip->i_rwlock));
461
462 /*
463 * ufs_getattr ends up being called by chklock
464 */
465 error = chklock(vp, FWRITE, uiop->uio_loffset,
466 uiop->uio_resid, uiop->uio_fmode, ct);
467 if (error)
468 goto out;
469 }
470
471 /* i_rwlock can change in chklock */
472 exclusive = rw_write_held(&ip->i_rwlock);
473 rewriteflg = ufs_check_rewrite(ip, uiop, ioflag);
474
475 /*
476 * Check for fast-path special case of directio re-writes.
477 */
478 if ((ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) &&
479 !exclusive && rewriteflg) {
480
481 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
482 if (error)
483 goto out;
484
485 rw_enter(&ip->i_contents, RW_READER);
486 error = ufs_directio_write(ip, uiop, ioflag, 1, cr,
487 &directio_status);
488 if (directio_status == DIRECTIO_SUCCESS) {
489 uint_t i_flag_save;
490
491 if (start_resid != uiop->uio_resid)
492 error = 0;
493 /*
494 * Special treatment of access times for re-writes.
495 * If IMOD is not already set, then convert it
496 * to IMODACC for this operation. This defers
497 * entering a delta into the log until the inode
498 * is flushed. This mimics what is done for read
499 * operations and inode access time.
500 */
501 mutex_enter(&ip->i_tlock);
502 i_flag_save = ip->i_flag;
503 ip->i_flag |= IUPD | ICHG;
504 ip->i_seq++;
505 ITIMES_NOLOCK(ip);
506 if ((i_flag_save & IMOD) == 0) {
507 ip->i_flag &= ~IMOD;
508 ip->i_flag |= IMODACC;
509 }
510 mutex_exit(&ip->i_tlock);
511 rw_exit(&ip->i_contents);
512 if (ulp)
513 ufs_lockfs_end(ulp);
514 goto out;
515 }
516 rw_exit(&ip->i_contents);
517 if (ulp)
518 ufs_lockfs_end(ulp);
519 }
520
521 if (!exclusive && !rw_tryupgrade(&ip->i_rwlock)) {
522 rw_exit(&ip->i_rwlock);
523 rw_enter(&ip->i_rwlock, RW_WRITER);
524 /*
525 * Mandatory locking could have been enabled
526 * after dropping the i_rwlock.
527 */
528 if (MANDLOCK(vp, ip->i_mode))
529 goto retry_mandlock;
530 }
531
532 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_WRITE_MASK);
533 if (error)
534 goto out;
535
536 /*
537 * Amount of log space needed for this write
538 */
539 if (!rewriteflg || !(ioflag & FDSYNC))
540 TRANS_WRITE_RESV(ip, uiop, ulp, &resv, &resid);
541
542 /*
543 * Throttle writes.
544 */
545 if (ufs_WRITES && (ip->i_writes > ufs_HW)) {
546 mutex_enter(&ip->i_tlock);
547 while (ip->i_writes > ufs_HW) {
548 ufs_throttles++;
549 cv_wait(&ip->i_wrcv, &ip->i_tlock);
550 }
551 mutex_exit(&ip->i_tlock);
552 }
553
554 /*
555 * Enter Transaction
556 *
557 * If the write is a rewrite there is no need to open a transaction
558 * if the FDSYNC flag is set and not the FSYNC. In this case just
559 * set the IMODACC flag to modify do the update at a later time
560 * thus avoiding the overhead of the logging transaction that is
561 * not required.
562 */
563 if (ioflag & (FSYNC|FDSYNC)) {
564 if (ulp) {
565 if (rewriteflg) {
566 uint_t i_flag_save;
567
568 rw_enter(&ip->i_contents, RW_READER);
569 mutex_enter(&ip->i_tlock);
570 i_flag_save = ip->i_flag;
571 ip->i_flag |= IUPD | ICHG;
572 ip->i_seq++;
573 ITIMES_NOLOCK(ip);
574 if ((i_flag_save & IMOD) == 0) {
575 ip->i_flag &= ~IMOD;
576 ip->i_flag |= IMODACC;
577 }
578 mutex_exit(&ip->i_tlock);
579 rw_exit(&ip->i_contents);
580 } else {
581 int terr = 0;
582 TRANS_BEGIN_SYNC(ufsvfsp, TOP_WRITE_SYNC, resv,
583 terr);
584 ASSERT(!terr);
585 }
586 }
587 } else {
588 if (ulp)
589 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_WRITE, resv);
590 }
591
592 /*
593 * Write the file
594 */
595 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
596 rw_enter(&ip->i_contents, RW_WRITER);
597 if ((ioflag & FAPPEND) != 0 && (ip->i_mode & IFMT) == IFREG) {
598 /*
599 * In append mode start at end of file.
600 */
601 uiop->uio_loffset = ip->i_size;
602 }
603
604 /*
605 * Mild optimisation, don't call ufs_trans_write() unless we have to
606 * Also, suppress file system full messages if we will retry.
607 */
608 if (retry)
609 ip->i_flag |= IQUIET;
610 if (resid) {
611 TRANS_WRITE(ip, uiop, ioflag, error, ulp, cr, resv, resid);
612 } else {
613 error = wrip(ip, uiop, ioflag, cr);
614 }
615 ip->i_flag &= ~IQUIET;
616
617 rw_exit(&ip->i_contents);
618 rw_exit(&ufsvfsp->vfs_dqrwlock);
619
620 /*
621 * Leave Transaction
622 */
623 if (ulp) {
624 if (ioflag & (FSYNC|FDSYNC)) {
625 if (!rewriteflg) {
626 int terr = 0;
627
628 TRANS_END_SYNC(ufsvfsp, terr, TOP_WRITE_SYNC,
629 resv);
630 if (error == 0)
631 error = terr;
632 }
633 } else {
634 TRANS_END_ASYNC(ufsvfsp, TOP_WRITE, resv);
635 }
636 ufs_lockfs_end(ulp);
637 }
638 out:
639 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
640 /*
641 * Any blocks tied up in pending deletes?
642 */
643 ufs_delete_drain_wait(ufsvfsp, 1);
644 retry = 0;
645 goto retry_mandlock;
646 }
647
648 if (error == ENOSPC && (start_resid != uiop->uio_resid))
649 error = 0;
650
651 return (error);
652 }
653
654 /*
655 * Don't cache write blocks to files with the sticky bit set.
656 * Used to keep swap files from blowing the page cache on a server.
657 */
658 int stickyhack = 1;
659
660 /*
661 * Free behind hacks. The pager is busted.
662 * XXX - need to pass the information down to writedone() in a flag like B_SEQ
663 * or B_FREE_IF_TIGHT_ON_MEMORY.
664 */
665 int freebehind = 1;
666 int smallfile = 0;
667 u_offset_t smallfile64 = 32 * 1024;
668
669 /*
670 * While we should, in most cases, cache the pages for write, we
671 * may also want to cache the pages for read as long as they are
672 * frequently re-usable.
673 *
674 * If cache_read_ahead = 1, the pages for read will go to the tail
675 * of the cache list when they are released, otherwise go to the head.
676 */
677 int cache_read_ahead = 0;
678
679 /*
680 * Freebehind exists so that as we read large files sequentially we
681 * don't consume most of memory with pages from a few files. It takes
682 * longer to re-read from disk multiple small files as it does reading
683 * one large one sequentially. As system memory grows customers need
684 * to retain bigger chunks of files in memory. The advent of the
685 * cachelist opens up of the possibility freeing pages to the head or
686 * tail of the list.
687 *
688 * Not freeing a page is a bet that the page will be read again before
689 * it's segmap slot is needed for something else. If we loose the bet,
690 * it means some other thread is burdened with the page free we did
691 * not do. If we win we save a free and reclaim.
692 *
693 * Freeing it at the tail vs the head of cachelist is a bet that the
694 * page will survive until the next read. It's also saying that this
695 * page is more likely to be re-used than a page freed some time ago
696 * and never reclaimed.
697 *
698 * Freebehind maintains a range of file offset [smallfile1; smallfile2]
699 *
700 * 0 < offset < smallfile1 : pages are not freed.
701 * smallfile1 < offset < smallfile2 : pages freed to tail of cachelist.
702 * smallfile2 < offset : pages freed to head of cachelist.
703 *
704 * The range is computed at most once per second and depends on
705 * freemem and ncpus_online. Both parameters are bounded to be
706 * >= smallfile && >= smallfile64.
707 *
708 * smallfile1 = (free memory / ncpu) / 1000
709 * smallfile2 = (free memory / ncpu) / 10
710 *
711 * A few examples values:
712 *
713 * Free Mem (in Bytes) [smallfile1; smallfile2] [smallfile1; smallfile2]
714 * ncpus_online = 4 ncpus_online = 64
715 * ------------------ ----------------------- -----------------------
716 * 1G [256K; 25M] [32K; 1.5M]
717 * 10G [2.5M; 250M] [156K; 15M]
718 * 100G [25M; 2.5G] [1.5M; 150M]
719 *
720 */
721
722 #define SMALLFILE1_D 1000
723 #define SMALLFILE2_D 10
724 static u_offset_t smallfile1 = 32 * 1024;
725 static u_offset_t smallfile2 = 32 * 1024;
726 static clock_t smallfile_update = 0; /* lbolt value of when to recompute */
727 uint_t smallfile1_d = SMALLFILE1_D;
728 uint_t smallfile2_d = SMALLFILE2_D;
729
730 /*
731 * wrip does the real work of write requests for ufs.
732 */
733 int
734 wrip(struct inode *ip, struct uio *uio, int ioflag, struct cred *cr)
735 {
736 rlim64_t limit = uio->uio_llimit;
737 u_offset_t off;
738 u_offset_t old_i_size;
739 struct fs *fs;
740 struct vnode *vp;
741 struct ufsvfs *ufsvfsp;
742 caddr_t base;
743 long start_resid = uio->uio_resid; /* save starting resid */
744 long premove_resid; /* resid before uiomove() */
745 uint_t flags;
746 int newpage;
747 int iupdat_flag, directio_status;
748 int n, on, mapon;
749 int error, pagecreate;
750 int do_dqrwlock; /* drop/reacquire vfs_dqrwlock */
751 int32_t iblocks;
752 int new_iblocks;
753
754 /*
755 * ip->i_size is incremented before the uiomove
756 * is done on a write. If the move fails (bad user
757 * address) reset ip->i_size.
758 * The better way would be to increment ip->i_size
759 * only if the uiomove succeeds.
760 */
761 int i_size_changed = 0;
762 o_mode_t type;
763 int i_seq_needed = 0;
764
765 vp = ITOV(ip);
766
767 /*
768 * check for forced unmount - should not happen as
769 * the request passed the lockfs checks.
770 */
771 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
772 return (EIO);
773
774 fs = ip->i_fs;
775
776 ASSERT(RW_WRITE_HELD(&ip->i_contents));
777
778 /* check for valid filetype */
779 type = ip->i_mode & IFMT;
780 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
781 (type != IFLNK) && (type != IFSHAD)) {
782 return (EIO);
783 }
784
785 /*
786 * the actual limit of UFS file size
787 * is UFS_MAXOFFSET_T
788 */
789 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
790 limit = MAXOFFSET_T;
791
792 if (uio->uio_loffset >= limit) {
793 proc_t *p = ttoproc(curthread);
794
795 mutex_enter(&p->p_lock);
796 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
797 p, RCA_UNSAFE_SIGINFO);
798 mutex_exit(&p->p_lock);
799 return (EFBIG);
800 }
801
802 /*
803 * if largefiles are disallowed, the limit is
804 * the pre-largefiles value of 2GB
805 */
806 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
807 limit = MIN(UFS_MAXOFFSET_T, limit);
808 else
809 limit = MIN(MAXOFF32_T, limit);
810
811 if (uio->uio_loffset < (offset_t)0) {
812 return (EINVAL);
813 }
814 if (uio->uio_resid == 0) {
815 return (0);
816 }
817
818 if (uio->uio_loffset >= limit)
819 return (EFBIG);
820
821 ip->i_flag |= INOACC; /* don't update ref time in getpage */
822
823 if (ioflag & (FSYNC|FDSYNC)) {
824 ip->i_flag |= ISYNC;
825 iupdat_flag = 1;
826 }
827 /*
828 * Try to go direct
829 */
830 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
831 uio->uio_llimit = limit;
832 error = ufs_directio_write(ip, uio, ioflag, 0, cr,
833 &directio_status);
834 /*
835 * If ufs_directio wrote to the file or set the flags,
836 * we need to update i_seq, but it may be deferred.
837 */
838 if (start_resid != uio->uio_resid ||
839 (ip->i_flag & (ICHG|IUPD))) {
840 i_seq_needed = 1;
841 ip->i_flag |= ISEQ;
842 }
843 if (directio_status == DIRECTIO_SUCCESS)
844 goto out;
845 }
846
847 /*
848 * Behavior with respect to dropping/reacquiring vfs_dqrwlock:
849 *
850 * o shadow inodes: vfs_dqrwlock is not held at all
851 * o quota updates: vfs_dqrwlock is read or write held
852 * o other updates: vfs_dqrwlock is read held
853 *
854 * The first case is the only one where we do not hold
855 * vfs_dqrwlock at all while entering wrip().
856 * We must make sure not to downgrade/drop vfs_dqrwlock if we
857 * have it as writer, i.e. if we are updating the quota inode.
858 * There is no potential deadlock scenario in this case as
859 * ufs_getpage() takes care of this and avoids reacquiring
860 * vfs_dqrwlock in that case.
861 *
862 * This check is done here since the above conditions do not change
863 * and we possibly loop below, so save a few cycles.
864 */
865 if ((type == IFSHAD) ||
866 (rw_owner(&ufsvfsp->vfs_dqrwlock) == curthread)) {
867 do_dqrwlock = 0;
868 } else {
869 do_dqrwlock = 1;
870 }
871
872 /*
873 * Large Files: We cast MAXBMASK to offset_t
874 * inorder to mask out the higher bits. Since offset_t
875 * is a signed value, the high order bit set in MAXBMASK
876 * value makes it do the right thing by having all bits 1
877 * in the higher word. May be removed for _SOLARIS64_.
878 */
879
880 fs = ip->i_fs;
881 do {
882 u_offset_t uoff = uio->uio_loffset;
883 off = uoff & (offset_t)MAXBMASK;
884 mapon = (int)(uoff & (offset_t)MAXBOFFSET);
885 on = (int)blkoff(fs, uoff);
886 n = (int)MIN(fs->fs_bsize - on, uio->uio_resid);
887 new_iblocks = 1;
888
889 if (type == IFREG && uoff + n >= limit) {
890 if (uoff >= limit) {
891 error = EFBIG;
892 goto out;
893 }
894 /*
895 * since uoff + n >= limit,
896 * therefore n >= limit - uoff, and n is an int
897 * so it is safe to cast it to an int
898 */
899 n = (int)(limit - (rlim64_t)uoff);
900 }
901 if (uoff + n > ip->i_size) {
902 /*
903 * We are extending the length of the file.
904 * bmap is used so that we are sure that
905 * if we need to allocate new blocks, that it
906 * is done here before we up the file size.
907 */
908 error = bmap_write(ip, uoff, (int)(on + n),
909 mapon == 0, NULL, cr);
910 /*
911 * bmap_write never drops i_contents so if
912 * the flags are set it changed the file.
913 */
914 if (ip->i_flag & (ICHG|IUPD)) {
915 i_seq_needed = 1;
916 ip->i_flag |= ISEQ;
917 }
918 if (error)
919 break;
920 /*
921 * There is a window of vulnerability here.
922 * The sequence of operations: allocate file
923 * system blocks, uiomove the data into pages,
924 * and then update the size of the file in the
925 * inode, must happen atomically. However, due
926 * to current locking constraints, this can not
927 * be done.
928 */
929 ASSERT(ip->i_writer == NULL);
930 ip->i_writer = curthread;
931 i_size_changed = 1;
932 /*
933 * If we are writing from the beginning of
934 * the mapping, we can just create the
935 * pages without having to read them.
936 */
937 pagecreate = (mapon == 0);
938 } else if (n == MAXBSIZE) {
939 /*
940 * Going to do a whole mappings worth,
941 * so we can just create the pages w/o
942 * having to read them in. But before
943 * we do that, we need to make sure any
944 * needed blocks are allocated first.
945 */
946 iblocks = ip->i_blocks;
947 error = bmap_write(ip, uoff, (int)(on + n),
948 BI_ALLOC_ONLY, NULL, cr);
949 /*
950 * bmap_write never drops i_contents so if
951 * the flags are set it changed the file.
952 */
953 if (ip->i_flag & (ICHG|IUPD)) {
954 i_seq_needed = 1;
955 ip->i_flag |= ISEQ;
956 }
957 if (error)
958 break;
959 pagecreate = 1;
960 /*
961 * check if the new created page needed the
962 * allocation of new disk blocks.
963 */
964 if (iblocks == ip->i_blocks)
965 new_iblocks = 0; /* no new blocks allocated */
966 } else {
967 pagecreate = 0;
968 /*
969 * In sync mode flush the indirect blocks which
970 * may have been allocated and not written on
971 * disk. In above cases bmap_write will allocate
972 * in sync mode.
973 */
974 if (ioflag & (FSYNC|FDSYNC)) {
975 error = ufs_indirblk_sync(ip, uoff);
976 if (error)
977 break;
978 }
979 }
980
981 /*
982 * At this point we can enter ufs_getpage() in one
983 * of two ways:
984 * 1) segmap_getmapflt() calls ufs_getpage() when the
985 * forcefault parameter is true (pagecreate == 0)
986 * 2) uiomove() causes a page fault.
987 *
988 * We have to drop the contents lock to prevent the VM
989 * system from trying to reacquire it in ufs_getpage()
990 * should the uiomove cause a pagefault.
991 *
992 * We have to drop the reader vfs_dqrwlock here as well.
993 */
994 rw_exit(&ip->i_contents);
995 if (do_dqrwlock) {
996 ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock));
997 ASSERT(!(RW_WRITE_HELD(&ufsvfsp->vfs_dqrwlock)));
998 rw_exit(&ufsvfsp->vfs_dqrwlock);
999 }
1000
1001 newpage = 0;
1002 premove_resid = uio->uio_resid;
1003
1004 /*
1005 * Touch the page and fault it in if it is not in core
1006 * before segmap_getmapflt or vpm_data_copy can lock it.
1007 * This is to avoid the deadlock if the buffer is mapped
1008 * to the same file through mmap which we want to write.
1009 */
1010 uio_prefaultpages((long)n, uio);
1011
1012 if (vpm_enable) {
1013 /*
1014 * Copy data. If new pages are created, part of
1015 * the page that is not written will be initizliazed
1016 * with zeros.
1017 */
1018 error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1019 uio, !pagecreate, &newpage, 0, S_WRITE);
1020 } else {
1021
1022 base = segmap_getmapflt(segkmap, vp, (off + mapon),
1023 (uint_t)n, !pagecreate, S_WRITE);
1024
1025 /*
1026 * segmap_pagecreate() returns 1 if it calls
1027 * page_create_va() to allocate any pages.
1028 */
1029
1030 if (pagecreate)
1031 newpage = segmap_pagecreate(segkmap, base,
1032 (size_t)n, 0);
1033
1034 error = uiomove(base + mapon, (long)n, UIO_WRITE, uio);
1035 }
1036
1037 /*
1038 * If "newpage" is set, then a new page was created and it
1039 * does not contain valid data, so it needs to be initialized
1040 * at this point.
1041 * Otherwise the page contains old data, which was overwritten
1042 * partially or as a whole in uiomove.
1043 * If there is only one iovec structure within uio, then
1044 * on error uiomove will not be able to update uio->uio_loffset
1045 * and we would zero the whole page here!
1046 *
1047 * If uiomove fails because of an error, the old valid data
1048 * is kept instead of filling the rest of the page with zero's.
1049 */
1050 if (!vpm_enable && newpage &&
1051 uio->uio_loffset < roundup(off + mapon + n, PAGESIZE)) {
1052 /*
1053 * We created pages w/o initializing them completely,
1054 * thus we need to zero the part that wasn't set up.
1055 * This happens on most EOF write cases and if
1056 * we had some sort of error during the uiomove.
1057 */
1058 int nzero, nmoved;
1059
1060 nmoved = (int)(uio->uio_loffset - (off + mapon));
1061 ASSERT(nmoved >= 0 && nmoved <= n);
1062 nzero = roundup(on + n, PAGESIZE) - nmoved;
1063 ASSERT(nzero > 0 && mapon + nmoved + nzero <= MAXBSIZE);
1064 (void) kzero(base + mapon + nmoved, (uint_t)nzero);
1065 }
1066
1067 /*
1068 * Unlock the pages allocated by page_create_va()
1069 * in segmap_pagecreate()
1070 */
1071 if (!vpm_enable && newpage)
1072 segmap_pageunlock(segkmap, base, (size_t)n, S_WRITE);
1073
1074 /*
1075 * If the size of the file changed, then update the
1076 * size field in the inode now. This can't be done
1077 * before the call to segmap_pageunlock or there is
1078 * a potential deadlock with callers to ufs_putpage().
1079 * They will be holding i_contents and trying to lock
1080 * a page, while this thread is holding a page locked
1081 * and trying to acquire i_contents.
1082 */
1083 if (i_size_changed) {
1084 rw_enter(&ip->i_contents, RW_WRITER);
1085 old_i_size = ip->i_size;
1086 UFS_SET_ISIZE(uoff + n, ip);
1087 TRANS_INODE(ufsvfsp, ip);
1088 /*
1089 * file has grown larger than 2GB. Set flag
1090 * in superblock to indicate this, if it
1091 * is not already set.
1092 */
1093 if ((ip->i_size > MAXOFF32_T) &&
1094 !(fs->fs_flags & FSLARGEFILES)) {
1095 ASSERT(ufsvfsp->vfs_lfflags & UFS_LARGEFILES);
1096 mutex_enter(&ufsvfsp->vfs_lock);
1097 fs->fs_flags |= FSLARGEFILES;
1098 ufs_sbwrite(ufsvfsp);
1099 mutex_exit(&ufsvfsp->vfs_lock);
1100 }
1101 mutex_enter(&ip->i_tlock);
1102 ip->i_writer = NULL;
1103 cv_broadcast(&ip->i_wrcv);
1104 mutex_exit(&ip->i_tlock);
1105 rw_exit(&ip->i_contents);
1106 }
1107
1108 if (error) {
1109 /*
1110 * If we failed on a write, we may have already
1111 * allocated file blocks as well as pages. It's
1112 * hard to undo the block allocation, but we must
1113 * be sure to invalidate any pages that may have
1114 * been allocated.
1115 *
1116 * If the page was created without initialization
1117 * then we must check if it should be possible
1118 * to destroy the new page and to keep the old data
1119 * on the disk.
1120 *
1121 * It is possible to destroy the page without
1122 * having to write back its contents only when
1123 * - the size of the file keeps unchanged
1124 * - bmap_write() did not allocate new disk blocks
1125 * it is possible to create big files using "seek" and
1126 * write to the end of the file. A "write" to a
1127 * position before the end of the file would not
1128 * change the size of the file but it would allocate
1129 * new disk blocks.
1130 * - uiomove intended to overwrite the whole page.
1131 * - a new page was created (newpage == 1).
1132 */
1133
1134 if (i_size_changed == 0 && new_iblocks == 0 &&
1135 newpage) {
1136
1137 /* unwind what uiomove eventually last did */
1138 uio->uio_resid = premove_resid;
1139
1140 /*
1141 * destroy the page, do not write ambiguous
1142 * data to the disk.
1143 */
1144 flags = SM_DESTROY;
1145 } else {
1146 /*
1147 * write the page back to the disk, if dirty,
1148 * and remove the page from the cache.
1149 */
1150 flags = SM_INVAL;
1151 }
1152
1153 if (vpm_enable) {
1154 /*
1155 * Flush pages.
1156 */
1157 (void) vpm_sync_pages(vp, off, n, flags);
1158 } else {
1159 (void) segmap_release(segkmap, base, flags);
1160 }
1161 } else {
1162 flags = 0;
1163 /*
1164 * Force write back for synchronous write cases.
1165 */
1166 if ((ioflag & (FSYNC|FDSYNC)) || type == IFDIR) {
1167 /*
1168 * If the sticky bit is set but the
1169 * execute bit is not set, we do a
1170 * synchronous write back and free
1171 * the page when done. We set up swap
1172 * files to be handled this way to
1173 * prevent servers from keeping around
1174 * the client's swap pages too long.
1175 * XXX - there ought to be a better way.
1176 */
1177 if (IS_SWAPVP(vp)) {
1178 flags = SM_WRITE | SM_FREE |
1179 SM_DONTNEED;
1180 iupdat_flag = 0;
1181 } else {
1182 flags = SM_WRITE;
1183 }
1184 } else if (n + on == MAXBSIZE || IS_SWAPVP(vp)) {
1185 /*
1186 * Have written a whole block.
1187 * Start an asynchronous write and
1188 * mark the buffer to indicate that
1189 * it won't be needed again soon.
1190 */
1191 flags = SM_WRITE | SM_ASYNC | SM_DONTNEED;
1192 }
1193 if (vpm_enable) {
1194 /*
1195 * Flush pages.
1196 */
1197 error = vpm_sync_pages(vp, off, n, flags);
1198 } else {
1199 error = segmap_release(segkmap, base, flags);
1200 }
1201 /*
1202 * If the operation failed and is synchronous,
1203 * then we need to unwind what uiomove() last
1204 * did so we can potentially return an error to
1205 * the caller. If this write operation was
1206 * done in two pieces and the first succeeded,
1207 * then we won't return an error for the second
1208 * piece that failed. However, we only want to
1209 * return a resid value that reflects what was
1210 * really done.
1211 *
1212 * Failures for non-synchronous operations can
1213 * be ignored since the page subsystem will
1214 * retry the operation until it succeeds or the
1215 * file system is unmounted.
1216 */
1217 if (error) {
1218 if ((ioflag & (FSYNC | FDSYNC)) ||
1219 type == IFDIR) {
1220 uio->uio_resid = premove_resid;
1221 } else {
1222 error = 0;
1223 }
1224 }
1225 }
1226
1227 /*
1228 * Re-acquire contents lock.
1229 * If it was dropped, reacquire reader vfs_dqrwlock as well.
1230 */
1231 if (do_dqrwlock)
1232 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1233 rw_enter(&ip->i_contents, RW_WRITER);
1234
1235 /*
1236 * If the uiomove() failed or if a synchronous
1237 * page push failed, fix up i_size.
1238 */
1239 if (error) {
1240 if (i_size_changed) {
1241 /*
1242 * The uiomove failed, and we
1243 * allocated blocks,so get rid
1244 * of them.
1245 */
1246 (void) ufs_itrunc(ip, old_i_size, 0, cr);
1247 }
1248 } else {
1249 /*
1250 * XXX - Can this be out of the loop?
1251 */
1252 ip->i_flag |= IUPD | ICHG;
1253 /*
1254 * Only do one increase of i_seq for multiple
1255 * pieces. Because we drop locks, record
1256 * the fact that we changed the timestamp and
1257 * are deferring the increase in case another thread
1258 * pushes our timestamp update.
1259 */
1260 i_seq_needed = 1;
1261 ip->i_flag |= ISEQ;
1262 if (i_size_changed)
1263 ip->i_flag |= IATTCHG;
1264 if ((ip->i_mode & (IEXEC | (IEXEC >> 3) |
1265 (IEXEC >> 6))) != 0 &&
1266 (ip->i_mode & (ISUID | ISGID)) != 0 &&
1267 secpolicy_vnode_setid_retain(cr,
1268 (ip->i_mode & ISUID) != 0 && ip->i_uid == 0) != 0) {
1269 /*
1270 * Clear Set-UID & Set-GID bits on
1271 * successful write if not privileged
1272 * and at least one of the execute bits
1273 * is set. If we always clear Set-GID,
1274 * mandatory file and record locking is
1275 * unuseable.
1276 */
1277 ip->i_mode &= ~(ISUID | ISGID);
1278 }
1279 }
1280 /*
1281 * In the case the FDSYNC flag is set and this is a
1282 * "rewrite" we won't log a delta.
1283 * The FSYNC flag overrides all cases.
1284 */
1285 if (!ufs_check_rewrite(ip, uio, ioflag) || !(ioflag & FDSYNC)) {
1286 TRANS_INODE(ufsvfsp, ip);
1287 }
1288 } while (error == 0 && uio->uio_resid > 0 && n != 0);
1289
1290 out:
1291 /*
1292 * Make sure i_seq is increased at least once per write
1293 */
1294 if (i_seq_needed) {
1295 ip->i_seq++;
1296 ip->i_flag &= ~ISEQ; /* no longer deferred */
1297 }
1298
1299 /*
1300 * Inode is updated according to this table -
1301 *
1302 * FSYNC FDSYNC(posix.4)
1303 * --------------------------
1304 * always@ IATTCHG|IBDWRITE
1305 *
1306 * @ - If we are doing synchronous write the only time we should
1307 * not be sync'ing the ip here is if we have the stickyhack
1308 * activated, the file is marked with the sticky bit and
1309 * no exec bit, the file length has not been changed and
1310 * no new blocks have been allocated during this write.
1311 */
1312
1313 if ((ip->i_flag & ISYNC) != 0) {
1314 /*
1315 * we have eliminated nosync
1316 */
1317 if ((ip->i_flag & (IATTCHG|IBDWRITE)) ||
1318 ((ioflag & FSYNC) && iupdat_flag)) {
1319 ufs_iupdat(ip, 1);
1320 }
1321 }
1322
1323 /*
1324 * If we've already done a partial-write, terminate
1325 * the write but return no error unless the error is ENOSPC
1326 * because the caller can detect this and free resources and
1327 * try again.
1328 */
1329 if ((start_resid != uio->uio_resid) && (error != ENOSPC))
1330 error = 0;
1331
1332 ip->i_flag &= ~(INOACC | ISYNC);
1333 ITIMES_NOLOCK(ip);
1334 return (error);
1335 }
1336
1337 /*
1338 * rdip does the real work of read requests for ufs.
1339 */
1340 int
1341 rdip(struct inode *ip, struct uio *uio, int ioflag, cred_t *cr)
1342 {
1343 u_offset_t off;
1344 caddr_t base;
1345 struct fs *fs;
1346 struct ufsvfs *ufsvfsp;
1347 struct vnode *vp;
1348 long oresid = uio->uio_resid;
1349 u_offset_t n, on, mapon;
1350 int error = 0;
1351 int doupdate = 1;
1352 uint_t flags;
1353 int dofree, directio_status;
1354 krw_t rwtype;
1355 o_mode_t type;
1356 clock_t now;
1357
1358 vp = ITOV(ip);
1359
1360 ASSERT(RW_LOCK_HELD(&ip->i_contents));
1361
1362 ufsvfsp = ip->i_ufsvfs;
1363
1364 if (ufsvfsp == NULL)
1365 return (EIO);
1366
1367 fs = ufsvfsp->vfs_fs;
1368
1369 /* check for valid filetype */
1370 type = ip->i_mode & IFMT;
1371 if ((type != IFREG) && (type != IFDIR) && (type != IFATTRDIR) &&
1372 (type != IFLNK) && (type != IFSHAD)) {
1373 return (EIO);
1374 }
1375
1376 if (uio->uio_loffset > UFS_MAXOFFSET_T) {
1377 error = 0;
1378 goto out;
1379 }
1380 if (uio->uio_loffset < (offset_t)0) {
1381 return (EINVAL);
1382 }
1383 if (uio->uio_resid == 0) {
1384 return (0);
1385 }
1386
1387 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (fs->fs_ronly == 0) &&
1388 (!ufsvfsp->vfs_noatime)) {
1389 mutex_enter(&ip->i_tlock);
1390 ip->i_flag |= IACC;
1391 mutex_exit(&ip->i_tlock);
1392 }
1393 /*
1394 * Try to go direct
1395 */
1396 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio) {
1397 error = ufs_directio_read(ip, uio, cr, &directio_status);
1398 if (directio_status == DIRECTIO_SUCCESS)
1399 goto out;
1400 }
1401
1402 rwtype = (rw_write_held(&ip->i_contents)?RW_WRITER:RW_READER);
1403
1404 do {
1405 offset_t diff;
1406 u_offset_t uoff = uio->uio_loffset;
1407 off = uoff & (offset_t)MAXBMASK;
1408 mapon = (u_offset_t)(uoff & (offset_t)MAXBOFFSET);
1409 on = (u_offset_t)blkoff(fs, uoff);
1410 n = MIN((u_offset_t)fs->fs_bsize - on,
1411 (u_offset_t)uio->uio_resid);
1412
1413 diff = ip->i_size - uoff;
1414
1415 if (diff <= (offset_t)0) {
1416 error = 0;
1417 goto out;
1418 }
1419 if (diff < (offset_t)n)
1420 n = (int)diff;
1421
1422 /*
1423 * We update smallfile2 and smallfile1 at most every second.
1424 */
1425 now = ddi_get_lbolt();
1426 if (now >= smallfile_update) {
1427 uint64_t percpufreeb;
1428 if (smallfile1_d == 0) smallfile1_d = SMALLFILE1_D;
1429 if (smallfile2_d == 0) smallfile2_d = SMALLFILE2_D;
1430 percpufreeb = ptob((uint64_t)freemem) / ncpus_online;
1431 smallfile1 = percpufreeb / smallfile1_d;
1432 smallfile2 = percpufreeb / smallfile2_d;
1433 smallfile1 = MAX(smallfile1, smallfile);
1434 smallfile1 = MAX(smallfile1, smallfile64);
1435 smallfile2 = MAX(smallfile1, smallfile2);
1436 smallfile_update = now + hz;
1437 }
1438
1439 dofree = freebehind &&
1440 ip->i_nextr == (off & PAGEMASK) && off > smallfile1;
1441
1442 /*
1443 * At this point we can enter ufs_getpage() in one of two
1444 * ways:
1445 * 1) segmap_getmapflt() calls ufs_getpage() when the
1446 * forcefault parameter is true (value of 1 is passed)
1447 * 2) uiomove() causes a page fault.
1448 *
1449 * We cannot hold onto an i_contents reader lock without
1450 * risking deadlock in ufs_getpage() so drop a reader lock.
1451 * The ufs_getpage() dolock logic already allows for a
1452 * thread holding i_contents as writer to work properly
1453 * so we keep a writer lock.
1454 */
1455 if (rwtype == RW_READER)
1456 rw_exit(&ip->i_contents);
1457
1458 if (vpm_enable) {
1459 /*
1460 * Copy data.
1461 */
1462 error = vpm_data_copy(vp, (off + mapon), (uint_t)n,
1463 uio, 1, NULL, 0, S_READ);
1464 } else {
1465 base = segmap_getmapflt(segkmap, vp, (off + mapon),
1466 (uint_t)n, 1, S_READ);
1467 error = uiomove(base + mapon, (long)n, UIO_READ, uio);
1468 }
1469
1470 flags = 0;
1471 if (!error) {
1472 /*
1473 * If reading sequential we won't need this
1474 * buffer again soon. For offsets in range
1475 * [smallfile1, smallfile2] release the pages
1476 * at the tail of the cache list, larger
1477 * offsets are released at the head.
1478 */
1479 if (dofree) {
1480 flags = SM_FREE | SM_ASYNC;
1481 if ((cache_read_ahead == 0) &&
1482 (off > smallfile2))
1483 flags |= SM_DONTNEED;
1484 }
1485 /*
1486 * In POSIX SYNC (FSYNC and FDSYNC) read mode,
1487 * we want to make sure that the page which has
1488 * been read, is written on disk if it is dirty.
1489 * And corresponding indirect blocks should also
1490 * be flushed out.
1491 */
1492 if ((ioflag & FRSYNC) && (ioflag & (FSYNC|FDSYNC))) {
1493 flags &= ~SM_ASYNC;
1494 flags |= SM_WRITE;
1495 }
1496 if (vpm_enable) {
1497 error = vpm_sync_pages(vp, off, n, flags);
1498 } else {
1499 error = segmap_release(segkmap, base, flags);
1500 }
1501 } else {
1502 if (vpm_enable) {
1503 (void) vpm_sync_pages(vp, off, n, flags);
1504 } else {
1505 (void) segmap_release(segkmap, base, flags);
1506 }
1507 }
1508
1509 if (rwtype == RW_READER)
1510 rw_enter(&ip->i_contents, rwtype);
1511 } while (error == 0 && uio->uio_resid > 0 && n != 0);
1512 out:
1513 /*
1514 * Inode is updated according to this table if FRSYNC is set.
1515 *
1516 * FSYNC FDSYNC(posix.4)
1517 * --------------------------
1518 * always IATTCHG|IBDWRITE
1519 */
1520 /*
1521 * The inode is not updated if we're logging and the inode is a
1522 * directory with FRSYNC, FSYNC and FDSYNC flags set.
1523 */
1524 if (ioflag & FRSYNC) {
1525 if (TRANS_ISTRANS(ufsvfsp) && ((ip->i_mode & IFMT) == IFDIR)) {
1526 doupdate = 0;
1527 }
1528 if (doupdate) {
1529 if ((ioflag & FSYNC) ||
1530 ((ioflag & FDSYNC) &&
1531 (ip->i_flag & (IATTCHG|IBDWRITE)))) {
1532 ufs_iupdat(ip, 1);
1533 }
1534 }
1535 }
1536 /*
1537 * If we've already done a partial read, terminate
1538 * the read but return no error.
1539 */
1540 if (oresid != uio->uio_resid)
1541 error = 0;
1542 ITIMES(ip);
1543
1544 return (error);
1545 }
1546
1547 /* ARGSUSED */
1548 static int
1549 ufs_ioctl(
1550 struct vnode *vp,
1551 int cmd,
1552 intptr_t arg,
1553 int flag,
1554 struct cred *cr,
1555 int *rvalp,
1556 caller_context_t *ct)
1557 {
1558 struct lockfs lockfs, lockfs_out;
1559 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
1560 char *comment, *original_comment;
1561 struct fs *fs;
1562 struct ulockfs *ulp;
1563 offset_t off;
1564 extern int maxphys;
1565 int error;
1566 int issync;
1567 int trans_size;
1568
1569
1570 /*
1571 * forcibly unmounted
1572 */
1573 if (ufsvfsp == NULL || vp->v_vfsp == NULL ||
1574 vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1575 return (EIO);
1576 fs = ufsvfsp->vfs_fs;
1577
1578 if (cmd == Q_QUOTACTL) {
1579 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_QUOTA_MASK);
1580 if (error)
1581 return (error);
1582
1583 if (ulp) {
1584 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_QUOTA,
1585 TOP_SETQUOTA_SIZE(fs));
1586 }
1587
1588 error = quotactl(vp, arg, flag, cr);
1589
1590 if (ulp) {
1591 TRANS_END_ASYNC(ufsvfsp, TOP_QUOTA,
1592 TOP_SETQUOTA_SIZE(fs));
1593 ufs_lockfs_end(ulp);
1594 }
1595 return (error);
1596 }
1597
1598 switch (cmd) {
1599 case _FIOLFS:
1600 /*
1601 * file system locking
1602 */
1603 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1604 return (EPERM);
1605
1606 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1607 if (copyin((caddr_t)arg, &lockfs,
1608 sizeof (struct lockfs)))
1609 return (EFAULT);
1610 }
1611 #ifdef _SYSCALL32_IMPL
1612 else {
1613 struct lockfs32 lockfs32;
1614 /* Translate ILP32 lockfs to LP64 lockfs */
1615 if (copyin((caddr_t)arg, &lockfs32,
1616 sizeof (struct lockfs32)))
1617 return (EFAULT);
1618 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1619 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1620 lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1621 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1622 lockfs.lf_comment =
1623 (caddr_t)(uintptr_t)lockfs32.lf_comment;
1624 }
1625 #endif /* _SYSCALL32_IMPL */
1626
1627 if (lockfs.lf_comlen) {
1628 if (lockfs.lf_comlen > LOCKFS_MAXCOMMENTLEN)
1629 return (ENAMETOOLONG);
1630 comment =
1631 kmem_alloc(lockfs.lf_comlen, KM_SLEEP);
1632 if (copyin(lockfs.lf_comment, comment,
1633 lockfs.lf_comlen)) {
1634 kmem_free(comment, lockfs.lf_comlen);
1635 return (EFAULT);
1636 }
1637 original_comment = lockfs.lf_comment;
1638 lockfs.lf_comment = comment;
1639 }
1640 if ((error = ufs_fiolfs(vp, &lockfs, 0)) == 0) {
1641 lockfs.lf_comment = original_comment;
1642
1643 if ((flag & DATAMODEL_MASK) ==
1644 DATAMODEL_NATIVE) {
1645 (void) copyout(&lockfs, (caddr_t)arg,
1646 sizeof (struct lockfs));
1647 }
1648 #ifdef _SYSCALL32_IMPL
1649 else {
1650 struct lockfs32 lockfs32;
1651 /* Translate LP64 to ILP32 lockfs */
1652 lockfs32.lf_lock =
1653 (uint32_t)lockfs.lf_lock;
1654 lockfs32.lf_flags =
1655 (uint32_t)lockfs.lf_flags;
1656 lockfs32.lf_key =
1657 (uint32_t)lockfs.lf_key;
1658 lockfs32.lf_comlen =
1659 (uint32_t)lockfs.lf_comlen;
1660 lockfs32.lf_comment =
1661 (uint32_t)(uintptr_t)
1662 lockfs.lf_comment;
1663 (void) copyout(&lockfs32, (caddr_t)arg,
1664 sizeof (struct lockfs32));
1665 }
1666 #endif /* _SYSCALL32_IMPL */
1667
1668 } else {
1669 if (lockfs.lf_comlen)
1670 kmem_free(comment, lockfs.lf_comlen);
1671 }
1672 return (error);
1673
1674 case _FIOLFSS:
1675 /*
1676 * get file system locking status
1677 */
1678
1679 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1680 if (copyin((caddr_t)arg, &lockfs,
1681 sizeof (struct lockfs)))
1682 return (EFAULT);
1683 }
1684 #ifdef _SYSCALL32_IMPL
1685 else {
1686 struct lockfs32 lockfs32;
1687 /* Translate ILP32 lockfs to LP64 lockfs */
1688 if (copyin((caddr_t)arg, &lockfs32,
1689 sizeof (struct lockfs32)))
1690 return (EFAULT);
1691 lockfs.lf_lock = (ulong_t)lockfs32.lf_lock;
1692 lockfs.lf_flags = (ulong_t)lockfs32.lf_flags;
1693 lockfs.lf_key = (ulong_t)lockfs32.lf_key;
1694 lockfs.lf_comlen = (ulong_t)lockfs32.lf_comlen;
1695 lockfs.lf_comment =
1696 (caddr_t)(uintptr_t)lockfs32.lf_comment;
1697 }
1698 #endif /* _SYSCALL32_IMPL */
1699
1700 if (error = ufs_fiolfss(vp, &lockfs_out))
1701 return (error);
1702 lockfs.lf_lock = lockfs_out.lf_lock;
1703 lockfs.lf_key = lockfs_out.lf_key;
1704 lockfs.lf_flags = lockfs_out.lf_flags;
1705 lockfs.lf_comlen = MIN(lockfs.lf_comlen,
1706 lockfs_out.lf_comlen);
1707
1708 if ((flag & DATAMODEL_MASK) == DATAMODEL_NATIVE) {
1709 if (copyout(&lockfs, (caddr_t)arg,
1710 sizeof (struct lockfs)))
1711 return (EFAULT);
1712 }
1713 #ifdef _SYSCALL32_IMPL
1714 else {
1715 /* Translate LP64 to ILP32 lockfs */
1716 struct lockfs32 lockfs32;
1717 lockfs32.lf_lock = (uint32_t)lockfs.lf_lock;
1718 lockfs32.lf_flags = (uint32_t)lockfs.lf_flags;
1719 lockfs32.lf_key = (uint32_t)lockfs.lf_key;
1720 lockfs32.lf_comlen = (uint32_t)lockfs.lf_comlen;
1721 lockfs32.lf_comment =
1722 (uint32_t)(uintptr_t)lockfs.lf_comment;
1723 if (copyout(&lockfs32, (caddr_t)arg,
1724 sizeof (struct lockfs32)))
1725 return (EFAULT);
1726 }
1727 #endif /* _SYSCALL32_IMPL */
1728
1729 if (lockfs.lf_comlen &&
1730 lockfs.lf_comment && lockfs_out.lf_comment)
1731 if (copyout(lockfs_out.lf_comment,
1732 lockfs.lf_comment, lockfs.lf_comlen))
1733 return (EFAULT);
1734 return (0);
1735
1736 case _FIOSATIME:
1737 /*
1738 * set access time
1739 */
1740
1741 /*
1742 * if mounted w/o atime, return quietly.
1743 * I briefly thought about returning ENOSYS, but
1744 * figured that most apps would consider this fatal
1745 * but the idea is to make this as seamless as poss.
1746 */
1747 if (ufsvfsp->vfs_noatime)
1748 return (0);
1749
1750 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1751 ULOCKFS_SETATTR_MASK);
1752 if (error)
1753 return (error);
1754
1755 if (ulp) {
1756 trans_size = (int)TOP_SETATTR_SIZE(VTOI(vp));
1757 TRANS_BEGIN_CSYNC(ufsvfsp, issync,
1758 TOP_SETATTR, trans_size);
1759 }
1760
1761 error = ufs_fiosatime(vp, (struct timeval *)arg,
1762 flag, cr);
1763
1764 if (ulp) {
1765 TRANS_END_CSYNC(ufsvfsp, error, issync,
1766 TOP_SETATTR, trans_size);
1767 ufs_lockfs_end(ulp);
1768 }
1769 return (error);
1770
1771 case _FIOSDIO:
1772 /*
1773 * set delayed-io
1774 */
1775 return (ufs_fiosdio(vp, (uint_t *)arg, flag, cr));
1776
1777 case _FIOGDIO:
1778 /*
1779 * get delayed-io
1780 */
1781 return (ufs_fiogdio(vp, (uint_t *)arg, flag, cr));
1782
1783 case _FIOIO:
1784 /*
1785 * inode open
1786 */
1787 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1788 ULOCKFS_VGET_MASK);
1789 if (error)
1790 return (error);
1791
1792 error = ufs_fioio(vp, (struct fioio *)arg, flag, cr);
1793
1794 if (ulp) {
1795 ufs_lockfs_end(ulp);
1796 }
1797 return (error);
1798
1799 case _FIOFFS:
1800 /*
1801 * file system flush (push w/invalidate)
1802 */
1803 if ((caddr_t)arg != NULL)
1804 return (EINVAL);
1805 return (ufs_fioffs(vp, NULL, cr));
1806
1807 case _FIOISBUSY:
1808 /*
1809 * Contract-private interface for Legato
1810 * Purge this vnode from the DNLC and decide
1811 * if this vnode is busy (*arg == 1) or not
1812 * (*arg == 0)
1813 */
1814 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1815 return (EPERM);
1816 error = ufs_fioisbusy(vp, (int *)arg, cr);
1817 return (error);
1818
1819 case _FIODIRECTIO:
1820 return (ufs_fiodirectio(vp, (int)arg, cr));
1821
1822 case _FIOTUNE:
1823 /*
1824 * Tune the file system (aka setting fs attributes)
1825 */
1826 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1827 ULOCKFS_SETATTR_MASK);
1828 if (error)
1829 return (error);
1830
1831 error = ufs_fiotune(vp, (struct fiotune *)arg, cr);
1832
1833 if (ulp)
1834 ufs_lockfs_end(ulp);
1835 return (error);
1836
1837 case _FIOLOGENABLE:
1838 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1839 return (EPERM);
1840 return (ufs_fiologenable(vp, (void *)arg, cr, flag));
1841
1842 case _FIOLOGDISABLE:
1843 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1844 return (EPERM);
1845 return (ufs_fiologdisable(vp, (void *)arg, cr, flag));
1846
1847 case _FIOISLOG:
1848 return (ufs_fioislog(vp, (void *)arg, cr, flag));
1849
1850 case _FIOSNAPSHOTCREATE_MULTI:
1851 {
1852 struct fiosnapcreate_multi fc, *fcp;
1853 size_t fcm_size;
1854
1855 if (copyin((void *)arg, &fc, sizeof (fc)))
1856 return (EFAULT);
1857 if (fc.backfilecount > MAX_BACKFILE_COUNT)
1858 return (EINVAL);
1859 fcm_size = sizeof (struct fiosnapcreate_multi) +
1860 (fc.backfilecount - 1) * sizeof (int);
1861 fcp = (struct fiosnapcreate_multi *)
1862 kmem_alloc(fcm_size, KM_SLEEP);
1863 if (copyin((void *)arg, fcp, fcm_size)) {
1864 kmem_free(fcp, fcm_size);
1865 return (EFAULT);
1866 }
1867 error = ufs_snap_create(vp, fcp, cr);
1868 /*
1869 * Do copyout even if there is an error because
1870 * the details of error is stored in fcp.
1871 */
1872 if (copyout(fcp, (void *)arg, fcm_size))
1873 error = EFAULT;
1874 kmem_free(fcp, fcm_size);
1875 return (error);
1876 }
1877
1878 case _FIOSNAPSHOTDELETE:
1879 {
1880 struct fiosnapdelete fc;
1881
1882 if (copyin((void *)arg, &fc, sizeof (fc)))
1883 return (EFAULT);
1884 error = ufs_snap_delete(vp, &fc, cr);
1885 if (!error && copyout(&fc, (void *)arg, sizeof (fc)))
1886 error = EFAULT;
1887 return (error);
1888 }
1889
1890 case _FIOGETSUPERBLOCK:
1891 if (copyout(fs, (void *)arg, SBSIZE))
1892 return (EFAULT);
1893 return (0);
1894
1895 case _FIOGETMAXPHYS:
1896 if (copyout(&maxphys, (void *)arg, sizeof (maxphys)))
1897 return (EFAULT);
1898 return (0);
1899
1900 /*
1901 * The following 3 ioctls are for TSufs support
1902 * although could potentially be used elsewhere
1903 */
1904 case _FIO_SET_LUFS_DEBUG:
1905 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1906 return (EPERM);
1907 lufs_debug = (uint32_t)arg;
1908 return (0);
1909
1910 case _FIO_SET_LUFS_ERROR:
1911 if (secpolicy_fs_config(cr, ufsvfsp->vfs_vfs) != 0)
1912 return (EPERM);
1913 TRANS_SETERROR(ufsvfsp);
1914 return (0);
1915
1916 case _FIO_GET_TOP_STATS:
1917 {
1918 fio_lufs_stats_t *ls;
1919 ml_unit_t *ul = ufsvfsp->vfs_log;
1920
1921 ls = kmem_zalloc(sizeof (*ls), KM_SLEEP);
1922 ls->ls_debug = ul->un_debug; /* return debug value */
1923 /* Copy stucture if statistics are being kept */
1924 if (ul->un_logmap->mtm_tops) {
1925 ls->ls_topstats = *(ul->un_logmap->mtm_tops);
1926 }
1927 error = 0;
1928 if (copyout(ls, (void *)arg, sizeof (*ls)))
1929 error = EFAULT;
1930 kmem_free(ls, sizeof (*ls));
1931 return (error);
1932 }
1933
1934 case _FIO_SEEK_DATA:
1935 case _FIO_SEEK_HOLE:
1936 if (ddi_copyin((void *)arg, &off, sizeof (off), flag))
1937 return (EFAULT);
1938 /* offset paramater is in/out */
1939 error = ufs_fio_holey(vp, cmd, &off);
1940 if (error)
1941 return (error);
1942 if (ddi_copyout(&off, (void *)arg, sizeof (off), flag))
1943 return (EFAULT);
1944 return (0);
1945
1946 case _FIO_COMPRESSED:
1947 {
1948 /*
1949 * This is a project private ufs ioctl() to mark
1950 * the inode as that belonging to a compressed
1951 * file. This is used to mark individual
1952 * compressed files in a miniroot archive.
1953 * The files compressed in this manner are
1954 * automatically decompressed by the dcfs filesystem
1955 * (via an interception in ufs_lookup - see decompvp())
1956 * which is layered on top of ufs on a system running
1957 * from the archive. See uts/common/fs/dcfs for details.
1958 * This ioctl only marks the file as compressed - the
1959 * actual compression is done by fiocompress (a
1960 * userland utility) which invokes this ioctl().
1961 */
1962 struct inode *ip = VTOI(vp);
1963
1964 error = ufs_lockfs_begin(ufsvfsp, &ulp,
1965 ULOCKFS_SETATTR_MASK);
1966 if (error)
1967 return (error);
1968
1969 if (ulp) {
1970 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_IUPDAT,
1971 TOP_IUPDAT_SIZE(ip));
1972 }
1973
1974 error = ufs_mark_compressed(vp);
1975
1976 if (ulp) {
1977 TRANS_END_ASYNC(ufsvfsp, TOP_IUPDAT,
1978 TOP_IUPDAT_SIZE(ip));
1979 ufs_lockfs_end(ulp);
1980 }
1981
1982 return (error);
1983
1984 }
1985
1986 default:
1987 return (ENOTTY);
1988 }
1989 }
1990
1991
1992 /* ARGSUSED */
1993 static int
1994 ufs_getattr(struct vnode *vp, struct vattr *vap, int flags,
1995 struct cred *cr, caller_context_t *ct)
1996 {
1997 struct inode *ip = VTOI(vp);
1998 struct ufsvfs *ufsvfsp;
1999 int err;
2000
2001 if (vap->va_mask == AT_SIZE) {
2002 /*
2003 * for performance, if only the size is requested don't bother
2004 * with anything else.
2005 */
2006 UFS_GET_ISIZE(&vap->va_size, ip);
2007 return (0);
2008 }
2009
2010 /*
2011 * inlined lockfs checks
2012 */
2013 ufsvfsp = ip->i_ufsvfs;
2014 if ((ufsvfsp == NULL) || ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs)) {
2015 err = EIO;
2016 goto out;
2017 }
2018
2019 rw_enter(&ip->i_contents, RW_READER);
2020 /*
2021 * Return all the attributes. This should be refined so
2022 * that it only returns what's asked for.
2023 */
2024
2025 /*
2026 * Copy from inode table.
2027 */
2028 vap->va_type = vp->v_type;
2029 vap->va_mode = ip->i_mode & MODEMASK;
2030 /*
2031 * If there is an ACL and there is a mask entry, then do the
2032 * extra work that completes the equivalent of an acltomode(3)
2033 * call. According to POSIX P1003.1e, the acl mask should be
2034 * returned in the group permissions field.
2035 *
2036 * - start with the original permission and mode bits (from above)
2037 * - clear the group owner bits
2038 * - add in the mask bits.
2039 */
2040 if (ip->i_ufs_acl && ip->i_ufs_acl->aclass.acl_ismask) {
2041 vap->va_mode &= ~((VREAD | VWRITE | VEXEC) >> 3);
2042 vap->va_mode |=
2043 (ip->i_ufs_acl->aclass.acl_maskbits & PERMMASK) << 3;
2044 }
2045 vap->va_uid = ip->i_uid;
2046 vap->va_gid = ip->i_gid;
2047 vap->va_fsid = ip->i_dev;
2048 vap->va_nodeid = (ino64_t)ip->i_number;
2049 vap->va_nlink = ip->i_nlink;
2050 vap->va_size = ip->i_size;
2051 if (vp->v_type == VCHR || vp->v_type == VBLK)
2052 vap->va_rdev = ip->i_rdev;
2053 else
2054 vap->va_rdev = 0; /* not a b/c spec. */
2055 mutex_enter(&ip->i_tlock);
2056 ITIMES_NOLOCK(ip); /* mark correct time in inode */
2057 vap->va_seq = ip->i_seq;
2058 vap->va_atime.tv_sec = (time_t)ip->i_atime.tv_sec;
2059 vap->va_atime.tv_nsec = ip->i_atime.tv_usec*1000;
2060 vap->va_mtime.tv_sec = (time_t)ip->i_mtime.tv_sec;
2061 vap->va_mtime.tv_nsec = ip->i_mtime.tv_usec*1000;
2062 vap->va_ctime.tv_sec = (time_t)ip->i_ctime.tv_sec;
2063 vap->va_ctime.tv_nsec = ip->i_ctime.tv_usec*1000;
2064 mutex_exit(&ip->i_tlock);
2065
2066 switch (ip->i_mode & IFMT) {
2067
2068 case IFBLK:
2069 vap->va_blksize = MAXBSIZE; /* was BLKDEV_IOSIZE */
2070 break;
2071
2072 case IFCHR:
2073 vap->va_blksize = MAXBSIZE;
2074 break;
2075
2076 default:
2077 vap->va_blksize = ip->i_fs->fs_bsize;
2078 break;
2079 }
2080 vap->va_nblocks = (fsblkcnt64_t)ip->i_blocks;
2081 rw_exit(&ip->i_contents);
2082 err = 0;
2083
2084 out:
2085 return (err);
2086 }
2087
2088 /*
2089 * Special wrapper to provide a callback for secpolicy_vnode_setattr().
2090 * The i_contents lock is already held by the caller and we need to
2091 * declare the inode as 'void *' argument.
2092 */
2093 static int
2094 ufs_priv_access(void *vip, int mode, struct cred *cr)
2095 {
2096 struct inode *ip = vip;
2097
2098 return (ufs_iaccess(ip, mode, cr, 0));
2099 }
2100
2101 /*ARGSUSED4*/
2102 static int
2103 ufs_setattr(
2104 struct vnode *vp,
2105 struct vattr *vap,
2106 int flags,
2107 struct cred *cr,
2108 caller_context_t *ct)
2109 {
2110 struct inode *ip = VTOI(vp);
2111 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2112 struct fs *fs;
2113 struct ulockfs *ulp;
2114 char *errmsg1;
2115 char *errmsg2;
2116 long blocks;
2117 long int mask = vap->va_mask;
2118 size_t len1, len2;
2119 int issync;
2120 int trans_size;
2121 int dotrans;
2122 int dorwlock;
2123 int error;
2124 int owner_change;
2125 int dodqlock;
2126 timestruc_t now;
2127 vattr_t oldva;
2128 int retry = 1;
2129 int indeadlock;
2130
2131 /*
2132 * Cannot set these attributes.
2133 */
2134 if ((mask & AT_NOSET) || (mask & AT_XVATTR))
2135 return (EINVAL);
2136
2137 /*
2138 * check for forced unmount
2139 */
2140 if (ufsvfsp == NULL)
2141 return (EIO);
2142
2143 fs = ufsvfsp->vfs_fs;
2144 if (fs->fs_ronly != 0)
2145 return (EROFS);
2146
2147 again:
2148 errmsg1 = NULL;
2149 errmsg2 = NULL;
2150 dotrans = 0;
2151 dorwlock = 0;
2152 dodqlock = 0;
2153
2154 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK);
2155 if (error)
2156 goto out;
2157
2158 /*
2159 * Acquire i_rwlock before TRANS_BEGIN_CSYNC() if this is a file.
2160 * This follows the protocol for read()/write().
2161 */
2162 if (vp->v_type != VDIR) {
2163 /*
2164 * ufs_tryirwlock uses rw_tryenter and checks for SLOCK to
2165 * avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2166 * possible, retries the operation.
2167 */
2168 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_file);
2169 if (indeadlock) {
2170 if (ulp)
2171 ufs_lockfs_end(ulp);
2172 goto again;
2173 }
2174 dorwlock = 1;
2175 }
2176
2177 /*
2178 * Truncate file. Must have write permission and not be a directory.
2179 */
2180 if (mask & AT_SIZE) {
2181 rw_enter(&ip->i_contents, RW_WRITER);
2182 if (vp->v_type == VDIR) {
2183 error = EISDIR;
2184 goto update_inode;
2185 }
2186 if (error = ufs_iaccess(ip, IWRITE, cr, 0))
2187 goto update_inode;
2188
2189 rw_exit(&ip->i_contents);
2190 error = TRANS_ITRUNC(ip, vap->va_size, 0, cr);
2191 if (error) {
2192 rw_enter(&ip->i_contents, RW_WRITER);
2193 goto update_inode;
2194 }
2195
2196 if (error == 0) {
2197 if (vap->va_size) {
2198 vnevent_truncate(vp, ct);
2199 } else {
2200 vnevent_resize(vp, ct);
2201 }
2202 }
2203 }
2204
2205 if (ulp) {
2206 trans_size = (int)TOP_SETATTR_SIZE(ip);
2207 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SETATTR, trans_size);
2208 ++dotrans;
2209 }
2210
2211 /*
2212 * Acquire i_rwlock after TRANS_BEGIN_CSYNC() if this is a directory.
2213 * This follows the protocol established by
2214 * ufs_link/create/remove/rename/mkdir/rmdir/symlink.
2215 */
2216 if (vp->v_type == VDIR) {
2217 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_SETATTR,
2218 retry_dir);
2219 if (indeadlock)
2220 goto again;
2221 dorwlock = 1;
2222 }
2223
2224 /*
2225 * Grab quota lock if we are changing the file's owner.
2226 */
2227 if (mask & AT_UID) {
2228 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2229 dodqlock = 1;
2230 }
2231 rw_enter(&ip->i_contents, RW_WRITER);
2232
2233 oldva.va_mode = ip->i_mode;
2234 oldva.va_uid = ip->i_uid;
2235 oldva.va_gid = ip->i_gid;
2236
2237 vap->va_mask &= ~AT_SIZE;
2238
2239 error = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags,
2240 ufs_priv_access, ip);
2241 if (error)
2242 goto update_inode;
2243
2244 mask = vap->va_mask;
2245
2246 /*
2247 * Change file access modes.
2248 */
2249 if (mask & AT_MODE) {
2250 ip->i_mode = (ip->i_mode & IFMT) | (vap->va_mode & ~IFMT);
2251 TRANS_INODE(ufsvfsp, ip);
2252 ip->i_flag |= ICHG;
2253 if (stickyhack) {
2254 mutex_enter(&vp->v_lock);
2255 if ((ip->i_mode & (ISVTX | IEXEC | IFDIR)) == ISVTX)
2256 vp->v_flag |= VSWAPLIKE;
2257 else
2258 vp->v_flag &= ~VSWAPLIKE;
2259 mutex_exit(&vp->v_lock);
2260 }
2261 }
2262 if (mask & (AT_UID|AT_GID)) {
2263 if (mask & AT_UID) {
2264 /*
2265 * Don't change ownership of the quota inode.
2266 */
2267 if (ufsvfsp->vfs_qinod == ip) {
2268 ASSERT(ufsvfsp->vfs_qflags & MQ_ENABLED);
2269 error = EINVAL;
2270 goto update_inode;
2271 }
2272
2273 /*
2274 * No real ownership change.
2275 */
2276 if (ip->i_uid == vap->va_uid) {
2277 blocks = 0;
2278 owner_change = 0;
2279 }
2280 /*
2281 * Remove the blocks and the file, from the old user's
2282 * quota.
2283 */
2284 else {
2285 blocks = ip->i_blocks;
2286 owner_change = 1;
2287
2288 (void) chkdq(ip, -blocks, /* force */ 1, cr,
2289 (char **)NULL, (size_t *)NULL);
2290 (void) chkiq(ufsvfsp, /* change */ -1, ip,
2291 (uid_t)ip->i_uid, /* force */ 1, cr,
2292 (char **)NULL, (size_t *)NULL);
2293 dqrele(ip->i_dquot);
2294 }
2295
2296 ip->i_uid = vap->va_uid;
2297
2298 /*
2299 * There is a real ownership change.
2300 */
2301 if (owner_change) {
2302 /*
2303 * Add the blocks and the file to the new
2304 * user's quota.
2305 */
2306 ip->i_dquot = getinoquota(ip);
2307 (void) chkdq(ip, blocks, /* force */ 1, cr,
2308 &errmsg1, &len1);
2309 (void) chkiq(ufsvfsp, /* change */ 1,
2310 (struct inode *)NULL, (uid_t)ip->i_uid,
2311 /* force */ 1, cr, &errmsg2, &len2);
2312 }
2313 }
2314 if (mask & AT_GID) {
2315 ip->i_gid = vap->va_gid;
2316 }
2317 TRANS_INODE(ufsvfsp, ip);
2318 ip->i_flag |= ICHG;
2319 }
2320 /*
2321 * Change file access or modified times.
2322 */
2323 if (mask & (AT_ATIME|AT_MTIME)) {
2324 /* Check that the time value is within ufs range */
2325 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) ||
2326 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) {
2327 error = EOVERFLOW;
2328 goto update_inode;
2329 }
2330
2331 /*
2332 * if the "noaccess" mount option is set and only atime
2333 * update is requested, do nothing. No error is returned.
2334 */
2335 if ((ufsvfsp->vfs_noatime) &&
2336 ((mask & (AT_ATIME|AT_MTIME)) == AT_ATIME))
2337 goto skip_atime;
2338
2339 if (mask & AT_ATIME) {
2340 ip->i_atime.tv_sec = vap->va_atime.tv_sec;
2341 ip->i_atime.tv_usec = vap->va_atime.tv_nsec / 1000;
2342 ip->i_flag &= ~IACC;
2343 }
2344 if (mask & AT_MTIME) {
2345 ip->i_mtime.tv_sec = vap->va_mtime.tv_sec;
2346 ip->i_mtime.tv_usec = vap->va_mtime.tv_nsec / 1000;
2347 gethrestime(&now);
2348 if (now.tv_sec > TIME32_MAX) {
2349 /*
2350 * In 2038, ctime sticks forever..
2351 */
2352 ip->i_ctime.tv_sec = TIME32_MAX;
2353 ip->i_ctime.tv_usec = 0;
2354 } else {
2355 ip->i_ctime.tv_sec = now.tv_sec;
2356 ip->i_ctime.tv_usec = now.tv_nsec / 1000;
2357 }
2358 ip->i_flag &= ~(IUPD|ICHG);
2359 ip->i_flag |= IMODTIME;
2360 }
2361 TRANS_INODE(ufsvfsp, ip);
2362 ip->i_flag |= IMOD;
2363 }
2364
2365 skip_atime:
2366 /*
2367 * The presence of a shadow inode may indicate an ACL, but does
2368 * not imply an ACL. Future FSD types should be handled here too
2369 * and check for the presence of the attribute-specific data
2370 * before referencing it.
2371 */
2372 if (ip->i_shadow) {
2373 /*
2374 * XXX if ufs_iupdat is changed to sandbagged write fix
2375 * ufs_acl_setattr to push ip to keep acls consistent
2376 *
2377 * Suppress out of inodes messages if we will retry.
2378 */
2379 if (retry)
2380 ip->i_flag |= IQUIET;
2381 error = ufs_acl_setattr(ip, vap, cr);
2382 ip->i_flag &= ~IQUIET;
2383 }
2384
2385 update_inode:
2386 /*
2387 * Setattr always increases the sequence number
2388 */
2389 ip->i_seq++;
2390
2391 /*
2392 * if nfsd and not logging; push synchronously
2393 */
2394 if ((curthread->t_flag & T_DONTPEND) && !TRANS_ISTRANS(ufsvfsp)) {
2395 ufs_iupdat(ip, 1);
2396 } else {
2397 ITIMES_NOLOCK(ip);
2398 }
2399
2400 rw_exit(&ip->i_contents);
2401 if (dodqlock) {
2402 rw_exit(&ufsvfsp->vfs_dqrwlock);
2403 }
2404 if (dorwlock)
2405 rw_exit(&ip->i_rwlock);
2406
2407 if (ulp) {
2408 if (dotrans) {
2409 int terr = 0;
2410 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SETATTR,
2411 trans_size);
2412 if (error == 0)
2413 error = terr;
2414 }
2415 ufs_lockfs_end(ulp);
2416 }
2417 out:
2418 /*
2419 * If out of inodes or blocks, see if we can free something
2420 * up from the delete queue.
2421 */
2422 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
2423 ufs_delete_drain_wait(ufsvfsp, 1);
2424 retry = 0;
2425 if (errmsg1 != NULL)
2426 kmem_free(errmsg1, len1);
2427 if (errmsg2 != NULL)
2428 kmem_free(errmsg2, len2);
2429 goto again;
2430 }
2431 if (errmsg1 != NULL) {
2432 uprintf(errmsg1);
2433 kmem_free(errmsg1, len1);
2434 }
2435 if (errmsg2 != NULL) {
2436 uprintf(errmsg2);
2437 kmem_free(errmsg2, len2);
2438 }
2439 return (error);
2440 }
2441
2442 /*ARGSUSED*/
2443 static int
2444 ufs_access(struct vnode *vp, int mode, int flags, struct cred *cr,
2445 caller_context_t *ct)
2446 {
2447 struct inode *ip = VTOI(vp);
2448
2449 if (ip->i_ufsvfs == NULL)
2450 return (EIO);
2451
2452 /*
2453 * The ufs_iaccess function wants to be called with
2454 * mode bits expressed as "ufs specific" bits.
2455 * I.e., VWRITE|VREAD|VEXEC do not make sense to
2456 * ufs_iaccess() but IWRITE|IREAD|IEXEC do.
2457 * But since they're the same we just pass the vnode mode
2458 * bit but just verify that assumption at compile time.
2459 */
2460 #if IWRITE != VWRITE || IREAD != VREAD || IEXEC != VEXEC
2461 #error "ufs_access needs to map Vmodes to Imodes"
2462 #endif
2463 return (ufs_iaccess(ip, mode, cr, 1));
2464 }
2465
2466 /* ARGSUSED */
2467 static int
2468 ufs_readlink(struct vnode *vp, struct uio *uiop, struct cred *cr,
2469 caller_context_t *ct)
2470 {
2471 struct inode *ip = VTOI(vp);
2472 struct ufsvfs *ufsvfsp;
2473 struct ulockfs *ulp;
2474 int error;
2475 int fastsymlink;
2476
2477 if (vp->v_type != VLNK) {
2478 error = EINVAL;
2479 goto nolockout;
2480 }
2481
2482 /*
2483 * If the symbolic link is empty there is nothing to read.
2484 * Fast-track these empty symbolic links
2485 */
2486 if (ip->i_size == 0) {
2487 error = 0;
2488 goto nolockout;
2489 }
2490
2491 ufsvfsp = ip->i_ufsvfs;
2492 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READLINK_MASK);
2493 if (error)
2494 goto nolockout;
2495 /*
2496 * The ip->i_rwlock protects the data blocks used for FASTSYMLINK
2497 */
2498 again:
2499 fastsymlink = 0;
2500 if (ip->i_flag & IFASTSYMLNK) {
2501 rw_enter(&ip->i_rwlock, RW_READER);
2502 rw_enter(&ip->i_contents, RW_READER);
2503 if (ip->i_flag & IFASTSYMLNK) {
2504 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
2505 (ip->i_fs->fs_ronly == 0) &&
2506 (!ufsvfsp->vfs_noatime)) {
2507 mutex_enter(&ip->i_tlock);
2508 ip->i_flag |= IACC;
2509 mutex_exit(&ip->i_tlock);
2510 }
2511 error = uiomove((caddr_t)&ip->i_db[1],
2512 MIN(ip->i_size, uiop->uio_resid),
2513 UIO_READ, uiop);
2514 ITIMES(ip);
2515 ++fastsymlink;
2516 }
2517 rw_exit(&ip->i_contents);
2518 rw_exit(&ip->i_rwlock);
2519 }
2520 if (!fastsymlink) {
2521 ssize_t size; /* number of bytes read */
2522 caddr_t basep; /* pointer to input data */
2523 ino_t ino;
2524 long igen;
2525 struct uio tuio; /* temp uio struct */
2526 struct uio *tuiop;
2527 iovec_t tiov; /* temp iovec struct */
2528 char kbuf[FSL_SIZE]; /* buffer to hold fast symlink */
2529 int tflag = 0; /* flag to indicate temp vars used */
2530
2531 ino = ip->i_number;
2532 igen = ip->i_gen;
2533 size = uiop->uio_resid;
2534 basep = uiop->uio_iov->iov_base;
2535 tuiop = uiop;
2536
2537 rw_enter(&ip->i_rwlock, RW_WRITER);
2538 rw_enter(&ip->i_contents, RW_WRITER);
2539 if (ip->i_flag & IFASTSYMLNK) {
2540 rw_exit(&ip->i_contents);
2541 rw_exit(&ip->i_rwlock);
2542 goto again;
2543 }
2544
2545 /* can this be a fast symlink and is it a user buffer? */
2546 if (ip->i_size <= FSL_SIZE &&
2547 (uiop->uio_segflg == UIO_USERSPACE ||
2548 uiop->uio_segflg == UIO_USERISPACE)) {
2549
2550 bzero(&tuio, sizeof (struct uio));
2551 /*
2552 * setup a kernel buffer to read link into. this
2553 * is to fix a race condition where the user buffer
2554 * got corrupted before copying it into the inode.
2555 */
2556 size = ip->i_size;
2557 tiov.iov_len = size;
2558 tiov.iov_base = kbuf;
2559 tuio.uio_iov = &tiov;
2560 tuio.uio_iovcnt = 1;
2561 tuio.uio_offset = uiop->uio_offset;
2562 tuio.uio_segflg = UIO_SYSSPACE;
2563 tuio.uio_fmode = uiop->uio_fmode;
2564 tuio.uio_extflg = uiop->uio_extflg;
2565 tuio.uio_limit = uiop->uio_limit;
2566 tuio.uio_resid = size;
2567
2568 basep = tuio.uio_iov->iov_base;
2569 tuiop = &tuio;
2570 tflag = 1;
2571 }
2572
2573 error = rdip(ip, tuiop, 0, cr);
2574 if (!(error == 0 && ip->i_number == ino && ip->i_gen == igen)) {
2575 rw_exit(&ip->i_contents);
2576 rw_exit(&ip->i_rwlock);
2577 goto out;
2578 }
2579
2580 if (tflag == 0)
2581 size -= uiop->uio_resid;
2582
2583 if ((tflag == 0 && ip->i_size <= FSL_SIZE &&
2584 ip->i_size == size) || (tflag == 1 &&
2585 tuio.uio_resid == 0)) {
2586 error = kcopy(basep, &ip->i_db[1], ip->i_size);
2587 if (error == 0) {
2588 ip->i_flag |= IFASTSYMLNK;
2589 /*
2590 * free page
2591 */
2592 (void) VOP_PUTPAGE(ITOV(ip),
2593 (offset_t)0, PAGESIZE,
2594 (B_DONTNEED | B_FREE | B_FORCE | B_ASYNC),
2595 cr, ct);
2596 } else {
2597 int i;
2598 /* error, clear garbage left behind */
2599 for (i = 1; i < NDADDR; i++)
2600 ip->i_db[i] = 0;
2601 for (i = 0; i < NIADDR; i++)
2602 ip->i_ib[i] = 0;
2603 }
2604 }
2605 if (tflag == 1) {
2606 /* now, copy it into the user buffer */
2607 error = uiomove((caddr_t)kbuf,
2608 MIN(size, uiop->uio_resid),
2609 UIO_READ, uiop);
2610 }
2611 rw_exit(&ip->i_contents);
2612 rw_exit(&ip->i_rwlock);
2613 }
2614 out:
2615 if (ulp) {
2616 ufs_lockfs_end(ulp);
2617 }
2618 nolockout:
2619 return (error);
2620 }
2621
2622 /* ARGSUSED */
2623 static int
2624 ufs_fsync(struct vnode *vp, int syncflag, struct cred *cr,
2625 caller_context_t *ct)
2626 {
2627 struct inode *ip = VTOI(vp);
2628 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
2629 struct ulockfs *ulp;
2630 int error;
2631
2632 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_FSYNC_MASK);
2633 if (error)
2634 return (error);
2635
2636 if (TRANS_ISTRANS(ufsvfsp)) {
2637 /*
2638 * First push out any data pages
2639 */
2640 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) &&
2641 (vp->v_type != VCHR) && !(IS_SWAPVP(vp))) {
2642 error = VOP_PUTPAGE(vp, (offset_t)0, (size_t)0,
2643 0, CRED(), ct);
2644 if (error)
2645 goto out;
2646 }
2647
2648 /*
2649 * Delta any delayed inode times updates
2650 * and push inode to log.
2651 * All other inode deltas will have already been delta'd
2652 * and will be pushed during the commit.
2653 */
2654 if (!(syncflag & FDSYNC) &&
2655 ((ip->i_flag & (IMOD|IMODACC)) == IMODACC)) {
2656 if (ulp) {
2657 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_FSYNC,
2658 TOP_SYNCIP_SIZE);
2659 }
2660 rw_enter(&ip->i_contents, RW_READER);
2661 mutex_enter(&ip->i_tlock);
2662 ip->i_flag &= ~IMODTIME;
2663 mutex_exit(&ip->i_tlock);
2664 ufs_iupdat(ip, I_SYNC);
2665 rw_exit(&ip->i_contents);
2666 if (ulp) {
2667 TRANS_END_ASYNC(ufsvfsp, TOP_FSYNC,
2668 TOP_SYNCIP_SIZE);
2669 }
2670 }
2671
2672 /*
2673 * Commit the Moby transaction
2674 *
2675 * Deltas have already been made so we just need to
2676 * commit them with a synchronous transaction.
2677 * TRANS_BEGIN_SYNC() will return an error
2678 * if there are no deltas to commit, for an
2679 * empty transaction.
2680 */
2681 if (ulp) {
2682 TRANS_BEGIN_SYNC(ufsvfsp, TOP_FSYNC, TOP_COMMIT_SIZE,
2683 error);
2684 if (error) {
2685 error = 0; /* commit wasn't needed */
2686 goto out;
2687 }
2688 TRANS_END_SYNC(ufsvfsp, error, TOP_FSYNC,
2689 TOP_COMMIT_SIZE);
2690 }
2691 } else { /* not logging */
2692 if (!(IS_SWAPVP(vp)))
2693 if (syncflag & FNODSYNC) {
2694 /* Just update the inode only */
2695 TRANS_IUPDAT(ip, 1);
2696 error = 0;
2697 } else if (syncflag & FDSYNC)
2698 /* Do data-synchronous writes */
2699 error = TRANS_SYNCIP(ip, 0, I_DSYNC, TOP_FSYNC);
2700 else
2701 /* Do synchronous writes */
2702 error = TRANS_SYNCIP(ip, 0, I_SYNC, TOP_FSYNC);
2703
2704 rw_enter(&ip->i_contents, RW_WRITER);
2705 if (!error)
2706 error = ufs_sync_indir(ip);
2707 rw_exit(&ip->i_contents);
2708 }
2709 out:
2710 if (ulp) {
2711 ufs_lockfs_end(ulp);
2712 }
2713 return (error);
2714 }
2715
2716 /*ARGSUSED*/
2717 static void
2718 ufs_inactive(struct vnode *vp, struct cred *cr, caller_context_t *ct)
2719 {
2720 ufs_iinactive(VTOI(vp));
2721 }
2722
2723 /*
2724 * Unix file system operations having to do with directory manipulation.
2725 */
2726 int ufs_lookup_idle_count = 2; /* Number of inodes to idle each time */
2727 /* ARGSUSED */
2728 static int
2729 ufs_lookup(struct vnode *dvp, char *nm, struct vnode **vpp,
2730 struct pathname *pnp, int flags, struct vnode *rdir, struct cred *cr,
2731 caller_context_t *ct, int *direntflags, pathname_t *realpnp)
2732 {
2733 struct inode *ip;
2734 struct inode *sip;
2735 struct inode *xip;
2736 struct ufsvfs *ufsvfsp;
2737 struct ulockfs *ulp;
2738 struct vnode *vp;
2739 int error;
2740
2741 /*
2742 * Check flags for type of lookup (regular file or attribute file)
2743 */
2744
2745 ip = VTOI(dvp);
2746
2747 if (flags & LOOKUP_XATTR) {
2748
2749 /*
2750 * If not mounted with XATTR support then return EINVAL
2751 */
2752
2753 if (!(ip->i_ufsvfs->vfs_vfs->vfs_flag & VFS_XATTR))
2754 return (EINVAL);
2755 /*
2756 * We don't allow recursive attributes...
2757 * Maybe someday we will.
2758 */
2759 if ((ip->i_cflags & IXATTR)) {
2760 return (EINVAL);
2761 }
2762
2763 if ((vp = dnlc_lookup(dvp, XATTR_DIR_NAME)) == NULL) {
2764 error = ufs_xattr_getattrdir(dvp, &sip, flags, cr);
2765 if (error) {
2766 *vpp = NULL;
2767 goto out;
2768 }
2769
2770 vp = ITOV(sip);
2771 dnlc_update(dvp, XATTR_DIR_NAME, vp);
2772 }
2773
2774 /*
2775 * Check accessibility of directory.
2776 */
2777 if (vp == DNLC_NO_VNODE) {
2778 VN_RELE(vp);
2779 error = ENOENT;
2780 goto out;
2781 }
2782 if ((error = ufs_iaccess(VTOI(vp), IEXEC, cr, 1)) != 0) {
2783 VN_RELE(vp);
2784 goto out;
2785 }
2786
2787 *vpp = vp;
2788 return (0);
2789 }
2790
2791 /*
2792 * Check for a null component, which we should treat as
2793 * looking at dvp from within it's parent, so we don't
2794 * need a call to ufs_iaccess(), as it has already been
2795 * done.
2796 */
2797 if (nm[0] == 0) {
2798 VN_HOLD(dvp);
2799 error = 0;
2800 *vpp = dvp;
2801 goto out;
2802 }
2803
2804 /*
2805 * Check for "." ie itself. this is a quick check and
2806 * avoids adding "." into the dnlc (which have been seen
2807 * to occupy >10% of the cache).
2808 */
2809 if ((nm[0] == '.') && (nm[1] == 0)) {
2810 /*
2811 * Don't return without checking accessibility
2812 * of the directory. We only need the lock if
2813 * we are going to return it.
2814 */
2815 if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) == 0) {
2816 VN_HOLD(dvp);
2817 *vpp = dvp;
2818 }
2819 goto out;
2820 }
2821
2822 /*
2823 * Fast path: Check the directory name lookup cache.
2824 */
2825 if (vp = dnlc_lookup(dvp, nm)) {
2826 /*
2827 * Check accessibility of directory.
2828 */
2829 if ((error = ufs_iaccess(ip, IEXEC, cr, 1)) != 0) {
2830 VN_RELE(vp);
2831 goto out;
2832 }
2833 if (vp == DNLC_NO_VNODE) {
2834 VN_RELE(vp);
2835 error = ENOENT;
2836 goto out;
2837 }
2838 xip = VTOI(vp);
2839 ulp = NULL;
2840 goto fastpath;
2841 }
2842
2843 /*
2844 * Keep the idle queue from getting too long by
2845 * idling two inodes before attempting to allocate another.
2846 * This operation must be performed before entering
2847 * lockfs or a transaction.
2848 */
2849 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2850 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2851 ins.in_lidles.value.ul += ufs_lookup_idle_count;
2852 ufs_idle_some(ufs_lookup_idle_count);
2853 }
2854
2855 retry_lookup:
2856 /*
2857 * Check accessibility of directory.
2858 */
2859 if (error = ufs_diraccess(ip, IEXEC, cr))
2860 goto out;
2861
2862 ufsvfsp = ip->i_ufsvfs;
2863 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK);
2864 if (error)
2865 goto out;
2866
2867 error = ufs_dirlook(ip, nm, &xip, cr, 1, 0);
2868
2869 fastpath:
2870 if (error == 0) {
2871 ip = xip;
2872 *vpp = ITOV(ip);
2873
2874 /*
2875 * If vnode is a device return special vnode instead.
2876 */
2877 if (IS_DEVVP(*vpp)) {
2878 struct vnode *newvp;
2879
2880 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type,
2881 cr);
2882 VN_RELE(*vpp);
2883 if (newvp == NULL)
2884 error = ENOSYS;
2885 else
2886 *vpp = newvp;
2887 } else if (ip->i_cflags & ICOMPRESS) {
2888 struct vnode *newvp;
2889
2890 /*
2891 * Compressed file, substitute dcfs vnode
2892 */
2893 newvp = decompvp(*vpp, cr, ct);
2894 VN_RELE(*vpp);
2895 if (newvp == NULL)
2896 error = ENOSYS;
2897 else
2898 *vpp = newvp;
2899 }
2900 }
2901 if (ulp) {
2902 ufs_lockfs_end(ulp);
2903 }
2904
2905 if (error == EAGAIN)
2906 goto retry_lookup;
2907
2908 out:
2909 return (error);
2910 }
2911
2912 /*ARGSUSED*/
2913 static int
2914 ufs_create(struct vnode *dvp, char *name, struct vattr *vap, enum vcexcl excl,
2915 int mode, struct vnode **vpp, struct cred *cr, int flag,
2916 caller_context_t *ct, vsecattr_t *vsecp)
2917 {
2918 struct inode *ip;
2919 struct inode *xip;
2920 struct inode *dip;
2921 struct vnode *xvp;
2922 struct ufsvfs *ufsvfsp;
2923 struct ulockfs *ulp;
2924 int error;
2925 int issync;
2926 int truncflag;
2927 int trans_size;
2928 int noentry;
2929 int defer_dip_seq_update = 0; /* need to defer update of dip->i_seq */
2930 int retry = 1;
2931 int indeadlock;
2932
2933 again:
2934 ip = VTOI(dvp);
2935 ufsvfsp = ip->i_ufsvfs;
2936 truncflag = 0;
2937
2938 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_CREATE_MASK);
2939 if (error)
2940 goto out;
2941
2942 if (ulp) {
2943 trans_size = (int)TOP_CREATE_SIZE(ip);
2944 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_CREATE, trans_size);
2945 }
2946
2947 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr) != 0)
2948 vap->va_mode &= ~VSVTX;
2949
2950 if (*name == '\0') {
2951 /*
2952 * Null component name refers to the directory itself.
2953 */
2954 VN_HOLD(dvp);
2955 /*
2956 * Even though this is an error case, we need to grab the
2957 * quota lock since the error handling code below is common.
2958 */
2959 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2960 rw_enter(&ip->i_contents, RW_WRITER);
2961 error = EEXIST;
2962 } else {
2963 xip = NULL;
2964 noentry = 0;
2965 /*
2966 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
2967 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
2968 * possible, retries the operation.
2969 */
2970 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_CREATE,
2971 retry_dir);
2972 if (indeadlock)
2973 goto again;
2974
2975 xvp = dnlc_lookup(dvp, name);
2976 if (xvp == DNLC_NO_VNODE) {
2977 noentry = 1;
2978 VN_RELE(xvp);
2979 xvp = NULL;
2980 }
2981 if (xvp) {
2982 rw_exit(&ip->i_rwlock);
2983 if (error = ufs_iaccess(ip, IEXEC, cr, 1)) {
2984 VN_RELE(xvp);
2985 } else {
2986 error = EEXIST;
2987 xip = VTOI(xvp);
2988 }
2989 } else {
2990 /*
2991 * Suppress file system full message if we will retry
2992 */
2993 error = ufs_direnter_cm(ip, name, DE_CREATE,
2994 vap, &xip, cr, (noentry | (retry ? IQUIET : 0)));
2995 if (error == EAGAIN) {
2996 if (ulp) {
2997 TRANS_END_CSYNC(ufsvfsp, error, issync,
2998 TOP_CREATE, trans_size);
2999 ufs_lockfs_end(ulp);
3000 }
3001 goto again;
3002 }
3003 rw_exit(&ip->i_rwlock);
3004 }
3005 ip = xip;
3006 if (ip != NULL) {
3007 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
3008 rw_enter(&ip->i_contents, RW_WRITER);
3009 }
3010 }
3011
3012 /*
3013 * If the file already exists and this is a non-exclusive create,
3014 * check permissions and allow access for non-directories.
3015 * Read-only create of an existing directory is also allowed.
3016 * We fail an exclusive create of anything which already exists.
3017 */
3018 if (error == EEXIST) {
3019 dip = VTOI(dvp);
3020 if (excl == NONEXCL) {
3021 if ((((ip->i_mode & IFMT) == IFDIR) ||
3022 ((ip->i_mode & IFMT) == IFATTRDIR)) &&
3023 (mode & IWRITE))
3024 error = EISDIR;
3025 else if (mode)
3026 error = ufs_iaccess(ip, mode, cr, 0);
3027 else
3028 error = 0;
3029 }
3030 if (error) {
3031 rw_exit(&ip->i_contents);
3032 rw_exit(&ufsvfsp->vfs_dqrwlock);
3033 VN_RELE(ITOV(ip));
3034 goto unlock;
3035 }
3036 /*
3037 * If the error EEXIST was set, then i_seq can not
3038 * have been updated. The sequence number interface
3039 * is defined such that a non-error VOP_CREATE must
3040 * increase the dir va_seq it by at least one. If we
3041 * have cleared the error, increase i_seq. Note that
3042 * we are increasing the dir i_seq and in rare cases
3043 * ip may actually be from the dvp, so we already have
3044 * the locks and it will not be subject to truncation.
3045 * In case we have to update i_seq of the parent
3046 * directory dip, we have to defer it till we have
3047 * released our locks on ip due to lock ordering requirements.
3048 */
3049 if (ip != dip)
3050 defer_dip_seq_update = 1;
3051 else
3052 ip->i_seq++;
3053
3054 if (((ip->i_mode & IFMT) == IFREG) &&
3055 (vap->va_mask & AT_SIZE) && vap->va_size == 0) {
3056 /*
3057 * Truncate regular files, if requested by caller.
3058 * Grab i_rwlock to make sure no one else is
3059 * currently writing to the file (we promised
3060 * bmap we would do this).
3061 * Must get the locks in the correct order.
3062 */
3063 if (ip->i_size == 0) {
3064 ip->i_flag |= ICHG | IUPD;
3065 ip->i_seq++;
3066 TRANS_INODE(ufsvfsp, ip);
3067 } else {
3068 /*
3069 * Large Files: Why this check here?
3070 * Though we do it in vn_create() we really
3071 * want to guarantee that we do not destroy
3072 * Large file data by atomically checking
3073 * the size while holding the contents
3074 * lock.
3075 */
3076 if (flag && !(flag & FOFFMAX) &&
3077 ((ip->i_mode & IFMT) == IFREG) &&
3078 (ip->i_size > (offset_t)MAXOFF32_T)) {
3079 rw_exit(&ip->i_contents);
3080 rw_exit(&ufsvfsp->vfs_dqrwlock);
3081 error = EOVERFLOW;
3082 goto unlock;
3083 }
3084 if (TRANS_ISTRANS(ufsvfsp))
3085 truncflag++;
3086 else {
3087 rw_exit(&ip->i_contents);
3088 rw_exit(&ufsvfsp->vfs_dqrwlock);
3089 ufs_tryirwlock_trans(&ip->i_rwlock,
3090 RW_WRITER, TOP_CREATE,
3091 retry_file);
3092 if (indeadlock) {
3093 VN_RELE(ITOV(ip));
3094 goto again;
3095 }
3096 rw_enter(&ufsvfsp->vfs_dqrwlock,
3097 RW_READER);
3098 rw_enter(&ip->i_contents, RW_WRITER);
3099 (void) ufs_itrunc(ip, (u_offset_t)0, 0,
3100 cr);
3101 rw_exit(&ip->i_rwlock);
3102 }
3103
3104 }
3105 if (error == 0) {
3106 vnevent_create(ITOV(ip), ct);
3107 }
3108 }
3109 }
3110
3111 if (error) {
3112 if (ip != NULL) {
3113 rw_exit(&ufsvfsp->vfs_dqrwlock);
3114 rw_exit(&ip->i_contents);
3115 }
3116 goto unlock;
3117 }
3118
3119 *vpp = ITOV(ip);
3120 ITIMES(ip);
3121 rw_exit(&ip->i_contents);
3122 rw_exit(&ufsvfsp->vfs_dqrwlock);
3123
3124 /*
3125 * If vnode is a device return special vnode instead.
3126 */
3127 if (!error && IS_DEVVP(*vpp)) {
3128 struct vnode *newvp;
3129
3130 newvp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr);
3131 VN_RELE(*vpp);
3132 if (newvp == NULL) {
3133 error = ENOSYS;
3134 goto unlock;
3135 }
3136 truncflag = 0;
3137 *vpp = newvp;
3138 }
3139 unlock:
3140
3141 /*
3142 * Do the deferred update of the parent directory's sequence
3143 * number now.
3144 */
3145 if (defer_dip_seq_update == 1) {
3146 rw_enter(&dip->i_contents, RW_READER);
3147 mutex_enter(&dip->i_tlock);
3148 dip->i_seq++;
3149 mutex_exit(&dip->i_tlock);
3150 rw_exit(&dip->i_contents);
3151 }
3152
3153 if (ulp) {
3154 int terr = 0;
3155
3156 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_CREATE,
3157 trans_size);
3158
3159 /*
3160 * If we haven't had a more interesting failure
3161 * already, then anything that might've happened
3162 * here should be reported.
3163 */
3164 if (error == 0)
3165 error = terr;
3166 }
3167
3168 if (!error && truncflag) {
3169 ufs_tryirwlock(&ip->i_rwlock, RW_WRITER, retry_trunc);
3170 if (indeadlock) {
3171 if (ulp)
3172 ufs_lockfs_end(ulp);
3173 VN_RELE(ITOV(ip));
3174 goto again;
3175 }
3176 (void) TRANS_ITRUNC(ip, (u_offset_t)0, 0, cr);
3177 rw_exit(&ip->i_rwlock);
3178 }
3179
3180 if (ulp)
3181 ufs_lockfs_end(ulp);
3182
3183 /*
3184 * If no inodes available, try to free one up out of the
3185 * pending delete queue.
3186 */
3187 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3188 ufs_delete_drain_wait(ufsvfsp, 1);
3189 retry = 0;
3190 goto again;
3191 }
3192
3193 out:
3194 return (error);
3195 }
3196
3197 extern int ufs_idle_max;
3198 /*ARGSUSED*/
3199 static int
3200 ufs_remove(struct vnode *vp, char *nm, struct cred *cr,
3201 caller_context_t *ct, int flags)
3202 {
3203 struct inode *ip = VTOI(vp);
3204 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3205 struct ulockfs *ulp;
3206 vnode_t *rmvp = NULL; /* Vnode corresponding to name being removed */
3207 int indeadlock;
3208 int error;
3209 int issync;
3210 int trans_size;
3211
3212 /*
3213 * don't let the delete queue get too long
3214 */
3215 if (ufsvfsp == NULL) {
3216 error = EIO;
3217 goto out;
3218 }
3219 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3220 ufs_delete_drain(vp->v_vfsp, 1, 1);
3221
3222 error = ufs_eventlookup(vp, nm, cr, &rmvp);
3223 if (rmvp != NULL) {
3224 /* Only send the event if there were no errors */
3225 if (error == 0)
3226 vnevent_remove(rmvp, vp, nm, ct);
3227 VN_RELE(rmvp);
3228 }
3229
3230 retry_remove:
3231 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_REMOVE_MASK);
3232 if (error)
3233 goto out;
3234
3235 if (ulp)
3236 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_REMOVE,
3237 trans_size = (int)TOP_REMOVE_SIZE(VTOI(vp)));
3238
3239 /*
3240 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3241 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3242 * possible, retries the operation.
3243 */
3244 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_REMOVE, retry);
3245 if (indeadlock)
3246 goto retry_remove;
3247 error = ufs_dirremove(ip, nm, (struct inode *)0, (struct vnode *)0,
3248 DR_REMOVE, cr);
3249 rw_exit(&ip->i_rwlock);
3250
3251 if (ulp) {
3252 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_REMOVE, trans_size);
3253 ufs_lockfs_end(ulp);
3254 }
3255
3256 out:
3257 return (error);
3258 }
3259
3260 /*
3261 * Link a file or a directory. Only privileged processes are allowed to
3262 * make links to directories.
3263 */
3264 /*ARGSUSED*/
3265 static int
3266 ufs_link(struct vnode *tdvp, struct vnode *svp, char *tnm, struct cred *cr,
3267 caller_context_t *ct, int flags)
3268 {
3269 struct inode *sip;
3270 struct inode *tdp = VTOI(tdvp);
3271 struct ufsvfs *ufsvfsp = tdp->i_ufsvfs;
3272 struct ulockfs *ulp;
3273 struct vnode *realvp;
3274 int error;
3275 int issync;
3276 int trans_size;
3277 int isdev;
3278 int indeadlock;
3279
3280 retry_link:
3281 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LINK_MASK);
3282 if (error)
3283 goto out;
3284
3285 if (ulp)
3286 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_LINK,
3287 trans_size = (int)TOP_LINK_SIZE(VTOI(tdvp)));
3288
3289 if (VOP_REALVP(svp, &realvp, ct) == 0)
3290 svp = realvp;
3291
3292 /*
3293 * Make sure link for extended attributes is valid
3294 * We only support hard linking of attr in ATTRDIR to ATTRDIR
3295 *
3296 * Make certain we don't attempt to look at a device node as
3297 * a ufs inode.
3298 */
3299
3300 isdev = IS_DEVVP(svp);
3301 if (((isdev == 0) && ((VTOI(svp)->i_cflags & IXATTR) == 0) &&
3302 ((tdp->i_mode & IFMT) == IFATTRDIR)) ||
3303 ((isdev == 0) && (VTOI(svp)->i_cflags & IXATTR) &&
3304 ((tdp->i_mode & IFMT) == IFDIR))) {
3305 error = EINVAL;
3306 goto unlock;
3307 }
3308
3309 sip = VTOI(svp);
3310 if ((svp->v_type == VDIR &&
3311 secpolicy_fs_linkdir(cr, ufsvfsp->vfs_vfs) != 0) ||
3312 (sip->i_uid != crgetuid(cr) && secpolicy_basic_link(cr) != 0)) {
3313 error = EPERM;
3314 goto unlock;
3315 }
3316
3317 /*
3318 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3319 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3320 * possible, retries the operation.
3321 */
3322 ufs_tryirwlock_trans(&tdp->i_rwlock, RW_WRITER, TOP_LINK, retry);
3323 if (indeadlock)
3324 goto retry_link;
3325 error = ufs_direnter_lr(tdp, tnm, DE_LINK, (struct inode *)0,
3326 sip, cr);
3327 rw_exit(&tdp->i_rwlock);
3328
3329 unlock:
3330 if (ulp) {
3331 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_LINK, trans_size);
3332 ufs_lockfs_end(ulp);
3333 }
3334
3335 if (!error) {
3336 vnevent_link(svp, ct);
3337 }
3338 out:
3339 return (error);
3340 }
3341
3342 uint64_t ufs_rename_retry_cnt;
3343 uint64_t ufs_rename_upgrade_retry_cnt;
3344 uint64_t ufs_rename_dircheck_retry_cnt;
3345 clock_t ufs_rename_backoff_delay = 1;
3346
3347 /*
3348 * Rename a file or directory.
3349 * We are given the vnode and entry string of the source and the
3350 * vnode and entry string of the place we want to move the source
3351 * to (the target). The essential operation is:
3352 * unlink(target);
3353 * link(source, target);
3354 * unlink(source);
3355 * but "atomically". Can't do full commit without saving state in
3356 * the inode on disk, which isn't feasible at this time. Best we
3357 * can do is always guarantee that the TARGET exists.
3358 */
3359
3360 /*ARGSUSED*/
3361 static int
3362 ufs_rename(
3363 struct vnode *sdvp, /* old (source) parent vnode */
3364 char *snm, /* old (source) entry name */
3365 struct vnode *tdvp, /* new (target) parent vnode */
3366 char *tnm, /* new (target) entry name */
3367 struct cred *cr,
3368 caller_context_t *ct,
3369 int flags)
3370 {
3371 struct inode *sip = NULL; /* source inode */
3372 struct inode *ip = NULL; /* check inode */
3373 struct inode *sdp; /* old (source) parent inode */
3374 struct inode *tdp; /* new (target) parent inode */
3375 struct vnode *svp = NULL; /* source vnode */
3376 struct vnode *tvp = NULL; /* target vnode, if it exists */
3377 struct vnode *realvp;
3378 struct ufsvfs *ufsvfsp;
3379 struct ulockfs *ulp = NULL;
3380 struct ufs_slot slot;
3381 timestruc_t now;
3382 int error;
3383 int issync;
3384 int trans_size;
3385 krwlock_t *first_lock;
3386 krwlock_t *second_lock;
3387 krwlock_t *reverse_lock;
3388 int serr, terr;
3389
3390 sdp = VTOI(sdvp);
3391 slot.fbp = NULL;
3392 ufsvfsp = sdp->i_ufsvfs;
3393
3394 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3395 tdvp = realvp;
3396
3397 /* Must do this before taking locks in case of DNLC miss */
3398 terr = ufs_eventlookup(tdvp, tnm, cr, &tvp);
3399 serr = ufs_eventlookup(sdvp, snm, cr, &svp);
3400
3401 if ((serr == 0) && ((terr == 0) || (terr == ENOENT))) {
3402 if (tvp != NULL)
3403 vnevent_pre_rename_dest(tvp, tdvp, tnm, ct);
3404
3405 /*
3406 * Notify the target directory of the rename event
3407 * if source and target directories are not the same.
3408 */
3409 if (sdvp != tdvp)
3410 vnevent_pre_rename_dest_dir(tdvp, svp, tnm, ct);
3411
3412 if (svp != NULL)
3413 vnevent_pre_rename_src(svp, sdvp, snm, ct);
3414 }
3415
3416 if (svp != NULL)
3417 VN_RELE(svp);
3418
3419 retry_rename:
3420 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RENAME_MASK);
3421 if (error)
3422 goto unlock;
3423
3424 if (ulp)
3425 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RENAME,
3426 trans_size = (int)TOP_RENAME_SIZE(sdp));
3427
3428 if (VOP_REALVP(tdvp, &realvp, ct) == 0)
3429 tdvp = realvp;
3430
3431 tdp = VTOI(tdvp);
3432
3433 /*
3434 * We only allow renaming of attributes from ATTRDIR to ATTRDIR.
3435 */
3436 if ((tdp->i_mode & IFMT) != (sdp->i_mode & IFMT)) {
3437 error = EINVAL;
3438 goto unlock;
3439 }
3440
3441 /*
3442 * Check accessibility of directory.
3443 */
3444 if (error = ufs_diraccess(sdp, IEXEC, cr))
3445 goto unlock;
3446
3447 /*
3448 * Look up inode of file we're supposed to rename.
3449 */
3450 gethrestime(&now);
3451 if (error = ufs_dirlook(sdp, snm, &sip, cr, 0, 0)) {
3452 if (error == EAGAIN) {
3453 if (ulp) {
3454 TRANS_END_CSYNC(ufsvfsp, error, issync,
3455 TOP_RENAME, trans_size);
3456 ufs_lockfs_end(ulp);
3457 }
3458 goto retry_rename;
3459 }
3460
3461 goto unlock;
3462 }
3463
3464 /*
3465 * Lock both the source and target directories (they may be
3466 * the same) to provide the atomicity semantics that was
3467 * previously provided by the per file system vfs_rename_lock
3468 *
3469 * with vfs_rename_lock removed to allow simultaneous renames
3470 * within a file system, ufs_dircheckpath can deadlock while
3471 * traversing back to ensure that source is not a parent directory
3472 * of target parent directory. This is because we get into
3473 * ufs_dircheckpath with the sdp and tdp locks held as RW_WRITER.
3474 * If the tdp and sdp of the simultaneous renames happen to be
3475 * in the path of each other, it can lead to a deadlock. This
3476 * can be avoided by getting the locks as RW_READER here and then
3477 * upgrading to RW_WRITER after completing the ufs_dircheckpath.
3478 *
3479 * We hold the target directory's i_rwlock after calling
3480 * ufs_lockfs_begin but in many other operations (like ufs_readdir)
3481 * VOP_RWLOCK is explicitly called by the filesystem independent code
3482 * before calling the file system operation. In these cases the order
3483 * is reversed (i.e i_rwlock is taken first and then ufs_lockfs_begin
3484 * is called). This is fine as long as ufs_lockfs_begin acts as a VOP
3485 * counter but with ufs_quiesce setting the SLOCK bit this becomes a
3486 * synchronizing object which might lead to a deadlock. So we use
3487 * rw_tryenter instead of rw_enter. If we fail to get this lock and
3488 * find that SLOCK bit is set, we call ufs_lockfs_end and restart the
3489 * operation.
3490 */
3491 retry:
3492 first_lock = &tdp->i_rwlock;
3493 second_lock = &sdp->i_rwlock;
3494 retry_firstlock:
3495 if (!rw_tryenter(first_lock, RW_READER)) {
3496 /*
3497 * We didn't get the lock. Check if the SLOCK is set in the
3498 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3499 * and wait for SLOCK to be cleared.
3500 */
3501
3502 if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3503 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3504 trans_size);
3505 ufs_lockfs_end(ulp);
3506 goto retry_rename;
3507
3508 } else {
3509 /*
3510 * SLOCK isn't set so this is a genuine synchronization
3511 * case. Let's try again after giving them a breather.
3512 */
3513 delay(RETRY_LOCK_DELAY);
3514 goto retry_firstlock;
3515 }
3516 }
3517 /*
3518 * Need to check if the tdp and sdp are same !!!
3519 */
3520 if ((tdp != sdp) && (!rw_tryenter(second_lock, RW_READER))) {
3521 /*
3522 * We didn't get the lock. Check if the SLOCK is set in the
3523 * ufsvfs. If yes, we might be in a deadlock. Safer to give up
3524 * and wait for SLOCK to be cleared.
3525 */
3526
3527 rw_exit(first_lock);
3528 if (ulp && ULOCKFS_IS_SLOCK(ulp)) {
3529 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME,
3530 trans_size);
3531 ufs_lockfs_end(ulp);
3532 goto retry_rename;
3533
3534 } else {
3535 /*
3536 * So we couldn't get the second level peer lock *and*
3537 * the SLOCK bit isn't set. Too bad we can be
3538 * contentding with someone wanting these locks otherway
3539 * round. Reverse the locks in case there is a heavy
3540 * contention for the second level lock.
3541 */
3542 reverse_lock = first_lock;
3543 first_lock = second_lock;
3544 second_lock = reverse_lock;
3545 ufs_rename_retry_cnt++;
3546 goto retry_firstlock;
3547 }
3548 }
3549
3550 if (sip == tdp) {
3551 error = EINVAL;
3552 goto errout;
3553 }
3554 /*
3555 * Make sure we can delete the source entry. This requires
3556 * write permission on the containing directory.
3557 * Check for sticky directories.
3558 */
3559 rw_enter(&sdp->i_contents, RW_READER);
3560 rw_enter(&sip->i_contents, RW_READER);
3561 if ((error = ufs_iaccess(sdp, IWRITE, cr, 0)) != 0 ||
3562 (error = ufs_sticky_remove_access(sdp, sip, cr)) != 0) {
3563 rw_exit(&sip->i_contents);
3564 rw_exit(&sdp->i_contents);
3565 goto errout;
3566 }
3567
3568 /*
3569 * If this is a rename of a directory and the parent is
3570 * different (".." must be changed), then the source
3571 * directory must not be in the directory hierarchy
3572 * above the target, as this would orphan everything
3573 * below the source directory. Also the user must have
3574 * write permission in the source so as to be able to
3575 * change "..".
3576 */
3577 if ((((sip->i_mode & IFMT) == IFDIR) ||
3578 ((sip->i_mode & IFMT) == IFATTRDIR)) && sdp != tdp) {
3579 ino_t inum;
3580
3581 if (error = ufs_iaccess(sip, IWRITE, cr, 0)) {
3582 rw_exit(&sip->i_contents);
3583 rw_exit(&sdp->i_contents);
3584 goto errout;
3585 }
3586 inum = sip->i_number;
3587 rw_exit(&sip->i_contents);
3588 rw_exit(&sdp->i_contents);
3589 if ((error = ufs_dircheckpath(inum, tdp, sdp, cr))) {
3590 /*
3591 * If we got EAGAIN ufs_dircheckpath detected a
3592 * potential deadlock and backed out. We need
3593 * to retry the operation since sdp and tdp have
3594 * to be released to avoid the deadlock.
3595 */
3596 if (error == EAGAIN) {
3597 rw_exit(&tdp->i_rwlock);
3598 if (tdp != sdp)
3599 rw_exit(&sdp->i_rwlock);
3600 delay(ufs_rename_backoff_delay);
3601 ufs_rename_dircheck_retry_cnt++;
3602 goto retry;
3603 }
3604 goto errout;
3605 }
3606 } else {
3607 rw_exit(&sip->i_contents);
3608 rw_exit(&sdp->i_contents);
3609 }
3610
3611
3612 /*
3613 * Check for renaming '.' or '..' or alias of '.'
3614 */
3615 if (strcmp(snm, ".") == 0 || strcmp(snm, "..") == 0 || sdp == sip) {
3616 error = EINVAL;
3617 goto errout;
3618 }
3619
3620 /*
3621 * Simultaneous renames can deadlock in ufs_dircheckpath since it
3622 * tries to traverse back the file tree with both tdp and sdp held
3623 * as RW_WRITER. To avoid that we have to hold the tdp and sdp locks
3624 * as RW_READERS till ufs_dircheckpath is done.
3625 * Now that ufs_dircheckpath is done with, we can upgrade the locks
3626 * to RW_WRITER.
3627 */
3628 if (!rw_tryupgrade(&tdp->i_rwlock)) {
3629 /*
3630 * The upgrade failed. We got to give away the lock
3631 * as to avoid deadlocking with someone else who is
3632 * waiting for writer lock. With the lock gone, we
3633 * cannot be sure the checks done above will hold
3634 * good when we eventually get them back as writer.
3635 * So if we can't upgrade we drop the locks and retry
3636 * everything again.
3637 */
3638 rw_exit(&tdp->i_rwlock);
3639 if (tdp != sdp)
3640 rw_exit(&sdp->i_rwlock);
3641 delay(ufs_rename_backoff_delay);
3642 ufs_rename_upgrade_retry_cnt++;
3643 goto retry;
3644 }
3645 if (tdp != sdp) {
3646 if (!rw_tryupgrade(&sdp->i_rwlock)) {
3647 /*
3648 * The upgrade failed. We got to give away the lock
3649 * as to avoid deadlocking with someone else who is
3650 * waiting for writer lock. With the lock gone, we
3651 * cannot be sure the checks done above will hold
3652 * good when we eventually get them back as writer.
3653 * So if we can't upgrade we drop the locks and retry
3654 * everything again.
3655 */
3656 rw_exit(&tdp->i_rwlock);
3657 rw_exit(&sdp->i_rwlock);
3658 delay(ufs_rename_backoff_delay);
3659 ufs_rename_upgrade_retry_cnt++;
3660 goto retry;
3661 }
3662 }
3663
3664 /*
3665 * Now that all the locks are held check to make sure another thread
3666 * didn't slip in and take out the sip.
3667 */
3668 slot.status = NONE;
3669 if ((sip->i_ctime.tv_usec * 1000) > now.tv_nsec ||
3670 sip->i_ctime.tv_sec > now.tv_sec) {
3671 rw_enter(&sdp->i_ufsvfs->vfs_dqrwlock, RW_READER);
3672 rw_enter(&sdp->i_contents, RW_WRITER);
3673 error = ufs_dircheckforname(sdp, snm, strlen(snm), &slot,
3674 &ip, cr, 0);
3675 rw_exit(&sdp->i_contents);
3676 rw_exit(&sdp->i_ufsvfs->vfs_dqrwlock);
3677 if (error) {
3678 goto errout;
3679 }
3680 if (ip == NULL) {
3681 error = ENOENT;
3682 goto errout;
3683 } else {
3684 /*
3685 * If the inode was found need to drop the v_count
3686 * so as not to keep the filesystem from being
3687 * unmounted at a later time.
3688 */
3689 VN_RELE(ITOV(ip));
3690 }
3691
3692 /*
3693 * Release the slot.fbp that has the page mapped and
3694 * locked SE_SHARED, and could be used in in
3695 * ufs_direnter_lr() which needs to get the SE_EXCL lock
3696 * on said page.
3697 */
3698 if (slot.fbp) {
3699 fbrelse(slot.fbp, S_OTHER);
3700 slot.fbp = NULL;
3701 }
3702 }
3703
3704 /*
3705 * Link source to the target.
3706 */
3707 if (error = ufs_direnter_lr(tdp, tnm, DE_RENAME, sdp, sip, cr)) {
3708 /*
3709 * ESAME isn't really an error; it indicates that the
3710 * operation should not be done because the source and target
3711 * are the same file, but that no error should be reported.
3712 */
3713 if (error == ESAME)
3714 error = 0;
3715 goto errout;
3716 }
3717
3718 if (error == 0 && tvp != NULL)
3719 vnevent_rename_dest(tvp, tdvp, tnm, ct);
3720
3721 /*
3722 * Unlink the source.
3723 * Remove the source entry. ufs_dirremove() checks that the entry
3724 * still reflects sip, and returns an error if it doesn't.
3725 * If the entry has changed just forget about it. Release
3726 * the source inode.
3727 */
3728 if ((error = ufs_dirremove(sdp, snm, sip, (struct vnode *)0,
3729 DR_RENAME, cr)) == ENOENT)
3730 error = 0;
3731
3732 if (error == 0) {
3733 vnevent_rename_src(ITOV(sip), sdvp, snm, ct);
3734 vnevent_rename_dest_dir(tdvp, ITOV(sip), tnm, ct);
3735 }
3736
3737 errout:
3738 if (slot.fbp)
3739 fbrelse(slot.fbp, S_OTHER);
3740
3741 rw_exit(&tdp->i_rwlock);
3742 if (sdp != tdp) {
3743 rw_exit(&sdp->i_rwlock);
3744 }
3745
3746 unlock:
3747 if (tvp != NULL)
3748 VN_RELE(tvp);
3749 if (sip != NULL)
3750 VN_RELE(ITOV(sip));
3751
3752 if (ulp) {
3753 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RENAME, trans_size);
3754 ufs_lockfs_end(ulp);
3755 }
3756
3757 return (error);
3758 }
3759
3760 /*ARGSUSED*/
3761 static int
3762 ufs_mkdir(struct vnode *dvp, char *dirname, struct vattr *vap,
3763 struct vnode **vpp, struct cred *cr, caller_context_t *ct, int flags,
3764 vsecattr_t *vsecp)
3765 {
3766 struct inode *ip;
3767 struct inode *xip;
3768 struct ufsvfs *ufsvfsp;
3769 struct ulockfs *ulp;
3770 int error;
3771 int issync;
3772 int trans_size;
3773 int indeadlock;
3774 int retry = 1;
3775
3776 ASSERT((vap->va_mask & (AT_TYPE|AT_MODE)) == (AT_TYPE|AT_MODE));
3777
3778 /*
3779 * Can't make directory in attr hidden dir
3780 */
3781 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
3782 return (EINVAL);
3783
3784 again:
3785 ip = VTOI(dvp);
3786 ufsvfsp = ip->i_ufsvfs;
3787 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_MKDIR_MASK);
3788 if (error)
3789 goto out;
3790 if (ulp)
3791 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_MKDIR,
3792 trans_size = (int)TOP_MKDIR_SIZE(ip));
3793
3794 /*
3795 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3796 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3797 * possible, retries the operation.
3798 */
3799 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_MKDIR, retry);
3800 if (indeadlock)
3801 goto again;
3802
3803 error = ufs_direnter_cm(ip, dirname, DE_MKDIR, vap, &xip, cr,
3804 (retry ? IQUIET : 0));
3805 if (error == EAGAIN) {
3806 if (ulp) {
3807 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_MKDIR,
3808 trans_size);
3809 ufs_lockfs_end(ulp);
3810 }
3811 goto again;
3812 }
3813
3814 rw_exit(&ip->i_rwlock);
3815 if (error == 0) {
3816 ip = xip;
3817 *vpp = ITOV(ip);
3818 } else if (error == EEXIST)
3819 VN_RELE(ITOV(xip));
3820
3821 if (ulp) {
3822 int terr = 0;
3823 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_MKDIR, trans_size);
3824 ufs_lockfs_end(ulp);
3825 if (error == 0)
3826 error = terr;
3827 }
3828 out:
3829 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
3830 ufs_delete_drain_wait(ufsvfsp, 1);
3831 retry = 0;
3832 goto again;
3833 }
3834
3835 return (error);
3836 }
3837
3838 /*ARGSUSED*/
3839 static int
3840 ufs_rmdir(struct vnode *vp, char *nm, struct vnode *cdir, struct cred *cr,
3841 caller_context_t *ct, int flags)
3842 {
3843 struct inode *ip = VTOI(vp);
3844 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
3845 struct ulockfs *ulp;
3846 vnode_t *rmvp = NULL; /* Vnode of removed directory */
3847 int error;
3848 int issync;
3849 int trans_size;
3850 int indeadlock;
3851
3852 /*
3853 * don't let the delete queue get too long
3854 */
3855 if (ufsvfsp == NULL) {
3856 error = EIO;
3857 goto out;
3858 }
3859 if (ufsvfsp->vfs_delete.uq_ne > ufs_idle_max)
3860 ufs_delete_drain(vp->v_vfsp, 1, 1);
3861
3862 error = ufs_eventlookup(vp, nm, cr, &rmvp);
3863 if (rmvp != NULL) {
3864 /* Only send the event if there were no errors */
3865 if (error == 0)
3866 vnevent_rmdir(rmvp, vp, nm, ct);
3867 VN_RELE(rmvp);
3868 }
3869
3870 retry_rmdir:
3871 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_RMDIR_MASK);
3872 if (error)
3873 goto out;
3874
3875 if (ulp)
3876 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_RMDIR,
3877 trans_size = TOP_RMDIR_SIZE);
3878
3879 /*
3880 * ufs_tryirwlock_trans uses rw_tryenter and checks for SLOCK
3881 * to avoid i_rwlock, ufs_lockfs_begin deadlock. If deadlock
3882 * possible, retries the operation.
3883 */
3884 ufs_tryirwlock_trans(&ip->i_rwlock, RW_WRITER, TOP_RMDIR, retry);
3885 if (indeadlock)
3886 goto retry_rmdir;
3887 error = ufs_dirremove(ip, nm, (struct inode *)0, cdir, DR_RMDIR, cr);
3888
3889 rw_exit(&ip->i_rwlock);
3890
3891 if (ulp) {
3892 TRANS_END_CSYNC(ufsvfsp, error, issync, TOP_RMDIR,
3893 trans_size);
3894 ufs_lockfs_end(ulp);
3895 }
3896
3897 out:
3898 return (error);
3899 }
3900
3901 /* ARGSUSED */
3902 static int
3903 ufs_readdir(
3904 struct vnode *vp,
3905 struct uio *uiop,
3906 struct cred *cr,
3907 int *eofp,
3908 caller_context_t *ct,
3909 int flags)
3910 {
3911 struct iovec *iovp;
3912 struct inode *ip;
3913 struct direct *idp;
3914 struct dirent64 *odp;
3915 struct fbuf *fbp;
3916 struct ufsvfs *ufsvfsp;
3917 struct ulockfs *ulp;
3918 caddr_t outbuf;
3919 size_t bufsize;
3920 uint_t offset;
3921 uint_t bytes_wanted, total_bytes_wanted;
3922 int incount = 0;
3923 int outcount = 0;
3924 int error;
3925
3926 ip = VTOI(vp);
3927 ASSERT(RW_READ_HELD(&ip->i_rwlock));
3928
3929 if (uiop->uio_loffset >= MAXOFF32_T) {
3930 if (eofp)
3931 *eofp = 1;
3932 return (0);
3933 }
3934
3935 /*
3936 * Check if we have been called with a valid iov_len
3937 * and bail out if not, otherwise we may potentially loop
3938 * forever further down.
3939 */
3940 if (uiop->uio_iov->iov_len <= 0) {
3941 error = EINVAL;
3942 goto out;
3943 }
3944
3945 /*
3946 * Large Files: When we come here we are guaranteed that
3947 * uio_offset can be used safely. The high word is zero.
3948 */
3949
3950 ufsvfsp = ip->i_ufsvfs;
3951 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_READDIR_MASK);
3952 if (error)
3953 goto out;
3954
3955 iovp = uiop->uio_iov;
3956 total_bytes_wanted = iovp->iov_len;
3957
3958 /* Large Files: directory files should not be "large" */
3959
3960 ASSERT(ip->i_size <= MAXOFF32_T);
3961
3962 /* Force offset to be valid (to guard against bogus lseek() values) */
3963 offset = (uint_t)uiop->uio_offset & ~(DIRBLKSIZ - 1);
3964
3965 /* Quit if at end of file or link count of zero (posix) */
3966 if (offset >= (uint_t)ip->i_size || ip->i_nlink <= 0) {
3967 if (eofp)
3968 *eofp = 1;
3969 error = 0;
3970 goto unlock;
3971 }
3972
3973 /*
3974 * Get space to change directory entries into fs independent format.
3975 * Do fast alloc for the most commonly used-request size (filesystem
3976 * block size).
3977 */
3978 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1) {
3979 bufsize = total_bytes_wanted;
3980 outbuf = kmem_alloc(bufsize, KM_SLEEP);
3981 odp = (struct dirent64 *)outbuf;
3982 } else {
3983 bufsize = total_bytes_wanted;
3984 odp = (struct dirent64 *)iovp->iov_base;
3985 }
3986
3987 nextblk:
3988 bytes_wanted = total_bytes_wanted;
3989
3990 /* Truncate request to file size */
3991 if (offset + bytes_wanted > (int)ip->i_size)
3992 bytes_wanted = (int)(ip->i_size - offset);
3993
3994 /* Comply with MAXBSIZE boundary restrictions of fbread() */
3995 if ((offset & MAXBOFFSET) + bytes_wanted > MAXBSIZE)
3996 bytes_wanted = MAXBSIZE - (offset & MAXBOFFSET);
3997
3998 /*
3999 * Read in the next chunk.
4000 * We are still holding the i_rwlock.
4001 */
4002 error = fbread(vp, (offset_t)offset, bytes_wanted, S_OTHER, &fbp);
4003
4004 if (error)
4005 goto update_inode;
4006 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) && (ip->i_fs->fs_ronly == 0) &&
4007 (!ufsvfsp->vfs_noatime)) {
4008 ip->i_flag |= IACC;
4009 }
4010 incount = 0;
4011 idp = (struct direct *)fbp->fb_addr;
4012 if (idp->d_ino == 0 && idp->d_reclen == 0 && idp->d_namlen == 0) {
4013 cmn_err(CE_WARN, "ufs_readdir: bad dir, inumber = %llu, "
4014 "fs = %s\n",
4015 (u_longlong_t)ip->i_number, ufsvfsp->vfs_fs->fs_fsmnt);
4016 fbrelse(fbp, S_OTHER);
4017 error = ENXIO;
4018 goto update_inode;
4019 }
4020 /* Transform to file-system independent format */
4021 while (incount < bytes_wanted) {
4022 /*
4023 * If the current directory entry is mangled, then skip
4024 * to the next block. It would be nice to set the FSBAD
4025 * flag in the super-block so that a fsck is forced on
4026 * next reboot, but locking is a problem.
4027 */
4028 if (idp->d_reclen & 0x3) {
4029 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4030 break;
4031 }
4032
4033 /* Skip to requested offset and skip empty entries */
4034 if (idp->d_ino != 0 && offset >= (uint_t)uiop->uio_offset) {
4035 ushort_t this_reclen =
4036 DIRENT64_RECLEN(idp->d_namlen);
4037 /* Buffer too small for any entries */
4038 if (!outcount && this_reclen > bufsize) {
4039 fbrelse(fbp, S_OTHER);
4040 error = EINVAL;
4041 goto update_inode;
4042 }
4043 /* If would overrun the buffer, quit */
4044 if (outcount + this_reclen > bufsize) {
4045 break;
4046 }
4047 /* Take this entry */
4048 odp->d_ino = (ino64_t)idp->d_ino;
4049 odp->d_reclen = (ushort_t)this_reclen;
4050 odp->d_off = (offset_t)(offset + idp->d_reclen);
4051
4052 /* use strncpy(9f) to zero out uninitialized bytes */
4053
4054 ASSERT(strlen(idp->d_name) + 1 <=
4055 DIRENT64_NAMELEN(this_reclen));
4056 (void) strncpy(odp->d_name, idp->d_name,
4057 DIRENT64_NAMELEN(this_reclen));
4058 outcount += odp->d_reclen;
4059 odp = (struct dirent64 *)
4060 ((intptr_t)odp + odp->d_reclen);
4061 ASSERT(outcount <= bufsize);
4062 }
4063 if (idp->d_reclen) {
4064 incount += idp->d_reclen;
4065 offset += idp->d_reclen;
4066 idp = (struct direct *)((intptr_t)idp + idp->d_reclen);
4067 } else {
4068 offset = (offset + DIRBLKSIZ) & ~(DIRBLKSIZ-1);
4069 break;
4070 }
4071 }
4072 /* Release the chunk */
4073 fbrelse(fbp, S_OTHER);
4074
4075 /* Read whole block, but got no entries, read another if not eof */
4076
4077 /*
4078 * Large Files: casting i_size to int here is not a problem
4079 * because directory sizes are always less than MAXOFF32_T.
4080 * See assertion above.
4081 */
4082
4083 if (offset < (int)ip->i_size && !outcount)
4084 goto nextblk;
4085
4086 /* Copy out the entry data */
4087 if (uiop->uio_segflg == UIO_SYSSPACE && uiop->uio_iovcnt == 1) {
4088 iovp->iov_base += outcount;
4089 iovp->iov_len -= outcount;
4090 uiop->uio_resid -= outcount;
4091 uiop->uio_offset = offset;
4092 } else if ((error = uiomove(outbuf, (long)outcount, UIO_READ,
4093 uiop)) == 0)
4094 uiop->uio_offset = offset;
4095 update_inode:
4096 ITIMES(ip);
4097 if (uiop->uio_segflg != UIO_SYSSPACE || uiop->uio_iovcnt != 1)
4098 kmem_free(outbuf, bufsize);
4099
4100 if (eofp && error == 0)
4101 *eofp = (uiop->uio_offset >= (int)ip->i_size);
4102 unlock:
4103 if (ulp) {
4104 ufs_lockfs_end(ulp);
4105 }
4106 out:
4107 return (error);
4108 }
4109
4110 /*ARGSUSED*/
4111 static int
4112 ufs_symlink(
4113 struct vnode *dvp, /* ptr to parent dir vnode */
4114 char *linkname, /* name of symbolic link */
4115 struct vattr *vap, /* attributes */
4116 char *target, /* target path */
4117 struct cred *cr, /* user credentials */
4118 caller_context_t *ct,
4119 int flags)
4120 {
4121 struct inode *ip, *dip = VTOI(dvp);
4122 struct ufsvfs *ufsvfsp = dip->i_ufsvfs;
4123 struct ulockfs *ulp;
4124 int error;
4125 int issync;
4126 int trans_size;
4127 int residual;
4128 int ioflag;
4129 int retry = 1;
4130
4131 /*
4132 * No symlinks in attrdirs at this time
4133 */
4134 if ((VTOI(dvp)->i_mode & IFMT) == IFATTRDIR)
4135 return (EINVAL);
4136
4137 again:
4138 ip = (struct inode *)NULL;
4139 vap->va_type = VLNK;
4140 vap->va_rdev = 0;
4141
4142 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SYMLINK_MASK);
4143 if (error)
4144 goto out;
4145
4146 if (ulp)
4147 TRANS_BEGIN_CSYNC(ufsvfsp, issync, TOP_SYMLINK,
4148 trans_size = (int)TOP_SYMLINK_SIZE(dip));
4149
4150 /*
4151 * We must create the inode before the directory entry, to avoid
4152 * racing with readlink(). ufs_dirmakeinode requires that we
4153 * hold the quota lock as reader, and directory locks as writer.
4154 */
4155
4156 rw_enter(&dip->i_rwlock, RW_WRITER);
4157 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4158 rw_enter(&dip->i_contents, RW_WRITER);
4159
4160 /*
4161 * Suppress any out of inodes messages if we will retry on
4162 * ENOSP
4163 */
4164 if (retry)
4165 dip->i_flag |= IQUIET;
4166
4167 error = ufs_dirmakeinode(dip, &ip, vap, DE_SYMLINK, cr);
4168
4169 dip->i_flag &= ~IQUIET;
4170
4171 rw_exit(&dip->i_contents);
4172 rw_exit(&ufsvfsp->vfs_dqrwlock);
4173 rw_exit(&dip->i_rwlock);
4174
4175 if (error)
4176 goto unlock;
4177
4178 /*
4179 * OK. The inode has been created. Write out the data of the
4180 * symbolic link. Since symbolic links are metadata, and should
4181 * remain consistent across a system crash, we need to force the
4182 * data out synchronously.
4183 *
4184 * (This is a change from the semantics in earlier releases, which
4185 * only created symbolic links synchronously if the semi-documented
4186 * 'syncdir' option was set, or if we were being invoked by the NFS
4187 * server, which requires symbolic links to be created synchronously.)
4188 *
4189 * We need to pass in a pointer for the residual length; otherwise
4190 * ufs_rdwri() will always return EIO if it can't write the data,
4191 * even if the error was really ENOSPC or EDQUOT.
4192 */
4193
4194 ioflag = FWRITE | FDSYNC;
4195 residual = 0;
4196
4197 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4198 rw_enter(&ip->i_contents, RW_WRITER);
4199
4200 /*
4201 * Suppress file system full messages if we will retry
4202 */
4203 if (retry)
4204 ip->i_flag |= IQUIET;
4205
4206 error = ufs_rdwri(UIO_WRITE, ioflag, ip, target, strlen(target),
4207 (offset_t)0, UIO_SYSSPACE, &residual, cr);
4208
4209 ip->i_flag &= ~IQUIET;
4210
4211 if (error) {
4212 rw_exit(&ip->i_contents);
4213 rw_exit(&ufsvfsp->vfs_dqrwlock);
4214 goto remove;
4215 }
4216
4217 /*
4218 * If the link's data is small enough, we can cache it in the inode.
4219 * This is a "fast symbolic link". We don't use the first direct
4220 * block because that's actually used to point at the symbolic link's
4221 * contents on disk; but we know that none of the other direct or
4222 * indirect blocks can be used because symbolic links are restricted
4223 * to be smaller than a file system block.
4224 */
4225
4226 ASSERT(MAXPATHLEN <= VBSIZE(ITOV(ip)));
4227
4228 if (ip->i_size > 0 && ip->i_size <= FSL_SIZE) {
4229 if (kcopy(target, &ip->i_db[1], ip->i_size) == 0) {
4230 ip->i_flag |= IFASTSYMLNK;
4231 } else {
4232 int i;
4233 /* error, clear garbage left behind */
4234 for (i = 1; i < NDADDR; i++)
4235 ip->i_db[i] = 0;
4236 for (i = 0; i < NIADDR; i++)
4237 ip->i_ib[i] = 0;
4238 }
4239 }
4240
4241 rw_exit(&ip->i_contents);
4242 rw_exit(&ufsvfsp->vfs_dqrwlock);
4243
4244 /*
4245 * OK. We've successfully created the symbolic link. All that
4246 * remains is to insert it into the appropriate directory.
4247 */
4248
4249 rw_enter(&dip->i_rwlock, RW_WRITER);
4250 error = ufs_direnter_lr(dip, linkname, DE_SYMLINK, NULL, ip, cr);
4251 rw_exit(&dip->i_rwlock);
4252
4253 /*
4254 * Fall through into remove-on-error code. We're either done, or we
4255 * need to remove the inode (if we couldn't insert it).
4256 */
4257
4258 remove:
4259 if (error && (ip != NULL)) {
4260 rw_enter(&ip->i_contents, RW_WRITER);
4261 ip->i_nlink--;
4262 ip->i_flag |= ICHG;
4263 ip->i_seq++;
4264 ufs_setreclaim(ip);
4265 rw_exit(&ip->i_contents);
4266 }
4267
4268 unlock:
4269 if (ip != NULL)
4270 VN_RELE(ITOV(ip));
4271
4272 if (ulp) {
4273 int terr = 0;
4274
4275 TRANS_END_CSYNC(ufsvfsp, terr, issync, TOP_SYMLINK,
4276 trans_size);
4277 ufs_lockfs_end(ulp);
4278 if (error == 0)
4279 error = terr;
4280 }
4281
4282 /*
4283 * We may have failed due to lack of an inode or of a block to
4284 * store the target in. Try flushing the delete queue to free
4285 * logically-available things up and try again.
4286 */
4287 if ((error == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
4288 ufs_delete_drain_wait(ufsvfsp, 1);
4289 retry = 0;
4290 goto again;
4291 }
4292
4293 out:
4294 return (error);
4295 }
4296
4297 /*
4298 * Ufs specific routine used to do ufs io.
4299 */
4300 int
4301 ufs_rdwri(enum uio_rw rw, int ioflag, struct inode *ip, caddr_t base,
4302 ssize_t len, offset_t offset, enum uio_seg seg, int *aresid,
4303 struct cred *cr)
4304 {
4305 struct uio auio;
4306 struct iovec aiov;
4307 int error;
4308
4309 ASSERT(RW_LOCK_HELD(&ip->i_contents));
4310
4311 bzero((caddr_t)&auio, sizeof (uio_t));
4312 bzero((caddr_t)&aiov, sizeof (iovec_t));
4313
4314 aiov.iov_base = base;
4315 aiov.iov_len = len;
4316 auio.uio_iov = &aiov;
4317 auio.uio_iovcnt = 1;
4318 auio.uio_loffset = offset;
4319 auio.uio_segflg = (short)seg;
4320 auio.uio_resid = len;
4321
4322 if (rw == UIO_WRITE) {
4323 auio.uio_fmode = FWRITE;
4324 auio.uio_extflg = UIO_COPY_DEFAULT;
4325 auio.uio_llimit = curproc->p_fsz_ctl;
4326 error = wrip(ip, &auio, ioflag, cr);
4327 } else {
4328 auio.uio_fmode = FREAD;
4329 auio.uio_extflg = UIO_COPY_CACHED;
4330 auio.uio_llimit = MAXOFFSET_T;
4331 error = rdip(ip, &auio, ioflag, cr);
4332 }
4333
4334 if (aresid) {
4335 *aresid = auio.uio_resid;
4336 } else if (auio.uio_resid) {
4337 error = EIO;
4338 }
4339 return (error);
4340 }
4341
4342 /*ARGSUSED*/
4343 static int
4344 ufs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
4345 {
4346 struct ufid *ufid;
4347 struct inode *ip = VTOI(vp);
4348
4349 if (ip->i_ufsvfs == NULL)
4350 return (EIO);
4351
4352 if (fidp->fid_len < (sizeof (struct ufid) - sizeof (ushort_t))) {
4353 fidp->fid_len = sizeof (struct ufid) - sizeof (ushort_t);
4354 return (ENOSPC);
4355 }
4356
4357 ufid = (struct ufid *)fidp;
4358 bzero((char *)ufid, sizeof (struct ufid));
4359 ufid->ufid_len = sizeof (struct ufid) - sizeof (ushort_t);
4360 ufid->ufid_ino = ip->i_number;
4361 ufid->ufid_gen = ip->i_gen;
4362
4363 return (0);
4364 }
4365
4366 /* ARGSUSED2 */
4367 static int
4368 ufs_rwlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4369 {
4370 struct inode *ip = VTOI(vp);
4371 struct ufsvfs *ufsvfsp;
4372 int forcedirectio;
4373
4374 /*
4375 * Read case is easy.
4376 */
4377 if (!write_lock) {
4378 rw_enter(&ip->i_rwlock, RW_READER);
4379 return (V_WRITELOCK_FALSE);
4380 }
4381
4382 /*
4383 * Caller has requested a writer lock, but that inhibits any
4384 * concurrency in the VOPs that follow. Acquire the lock shared
4385 * and defer exclusive access until it is known to be needed in
4386 * other VOP handlers. Some cases can be determined here.
4387 */
4388
4389 /*
4390 * If directio is not set, there is no chance of concurrency,
4391 * so just acquire the lock exclusive. Beware of a forced
4392 * unmount before looking at the mount option.
4393 */
4394 ufsvfsp = ip->i_ufsvfs;
4395 forcedirectio = ufsvfsp ? ufsvfsp->vfs_forcedirectio : 0;
4396 if (!(ip->i_flag & IDIRECTIO || forcedirectio) ||
4397 !ufs_allow_shared_writes) {
4398 rw_enter(&ip->i_rwlock, RW_WRITER);
4399 return (V_WRITELOCK_TRUE);
4400 }
4401
4402 /*
4403 * Mandatory locking forces acquiring i_rwlock exclusive.
4404 */
4405 if (MANDLOCK(vp, ip->i_mode)) {
4406 rw_enter(&ip->i_rwlock, RW_WRITER);
4407 return (V_WRITELOCK_TRUE);
4408 }
4409
4410 /*
4411 * Acquire the lock shared in case a concurrent write follows.
4412 * Mandatory locking could have become enabled before the lock
4413 * was acquired. Re-check and upgrade if needed.
4414 */
4415 rw_enter(&ip->i_rwlock, RW_READER);
4416 if (MANDLOCK(vp, ip->i_mode)) {
4417 rw_exit(&ip->i_rwlock);
4418 rw_enter(&ip->i_rwlock, RW_WRITER);
4419 return (V_WRITELOCK_TRUE);
4420 }
4421 return (V_WRITELOCK_FALSE);
4422 }
4423
4424 /*ARGSUSED*/
4425 static void
4426 ufs_rwunlock(struct vnode *vp, int write_lock, caller_context_t *ctp)
4427 {
4428 struct inode *ip = VTOI(vp);
4429
4430 rw_exit(&ip->i_rwlock);
4431 }
4432
4433 /* ARGSUSED */
4434 static int
4435 ufs_seek(struct vnode *vp, offset_t ooff, offset_t *noffp,
4436 caller_context_t *ct)
4437 {
4438 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0);
4439 }
4440
4441 /* ARGSUSED */
4442 static int
4443 ufs_frlock(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4444 offset_t offset, struct flk_callback *flk_cbp, struct cred *cr,
4445 caller_context_t *ct)
4446 {
4447 struct inode *ip = VTOI(vp);
4448
4449 if (ip->i_ufsvfs == NULL)
4450 return (EIO);
4451
4452 /*
4453 * If file is being mapped, disallow frlock.
4454 * XXX I am not holding tlock while checking i_mapcnt because the
4455 * current locking strategy drops all locks before calling fs_frlock.
4456 * So, mapcnt could change before we enter fs_frlock making is
4457 * meaningless to have held tlock in the first place.
4458 */
4459 if (ip->i_mapcnt > 0 && MANDLOCK(vp, ip->i_mode))
4460 return (EAGAIN);
4461 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct));
4462 }
4463
4464 /* ARGSUSED */
4465 static int
4466 ufs_space(struct vnode *vp, int cmd, struct flock64 *bfp, int flag,
4467 offset_t offset, cred_t *cr, caller_context_t *ct)
4468 {
4469 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
4470 struct ulockfs *ulp;
4471 int error;
4472
4473 if ((error = convoff(vp, bfp, 0, offset)) == 0) {
4474 if (cmd == F_FREESP) {
4475 error = ufs_lockfs_begin(ufsvfsp, &ulp,
4476 ULOCKFS_SPACE_MASK);
4477 if (error)
4478 return (error);
4479 error = ufs_freesp(vp, bfp, flag, cr);
4480
4481 if (error == 0) {
4482 if (bfp->l_start == 0) {
4483 vnevent_truncate(vp, ct);
4484 } else {
4485 vnevent_resize(vp, ct);
4486 }
4487 }
4488 } else if (cmd == F_ALLOCSP) {
4489 error = ufs_lockfs_begin(ufsvfsp, &ulp,
4490 ULOCKFS_FALLOCATE_MASK);
4491 if (error)
4492 return (error);
4493 error = ufs_allocsp(vp, bfp, cr);
4494 } else
4495 return (EINVAL); /* Command not handled here */
4496
4497 if (ulp)
4498 ufs_lockfs_end(ulp);
4499
4500 }
4501 return (error);
4502 }
4503
4504 /*
4505 * Used to determine if read ahead should be done. Also used to
4506 * to determine when write back occurs.
4507 */
4508 #define CLUSTSZ(ip) ((ip)->i_ufsvfs->vfs_ioclustsz)
4509
4510 /*
4511 * A faster version of ufs_getpage.
4512 *
4513 * We optimize by inlining the pvn_getpages iterator, eliminating
4514 * calls to bmap_read if file doesn't have UFS holes, and avoiding
4515 * the overhead of page_exists().
4516 *
4517 * When files has UFS_HOLES and ufs_getpage is called with S_READ,
4518 * we set *protp to PROT_READ to avoid calling bmap_read. This approach
4519 * victimizes performance when a file with UFS holes is faulted
4520 * first in the S_READ mode, and then in the S_WRITE mode. We will get
4521 * two MMU faults in this case.
4522 *
4523 * XXX - the inode fields which control the sequential mode are not
4524 * protected by any mutex. The read ahead will act wild if
4525 * multiple processes will access the file concurrently and
4526 * some of them in sequential mode. One particulary bad case
4527 * is if another thread will change the value of i_nextrio between
4528 * the time this thread tests the i_nextrio value and then reads it
4529 * again to use it as the offset for the read ahead.
4530 */
4531 /*ARGSUSED*/
4532 static int
4533 ufs_getpage(struct vnode *vp, offset_t off, size_t len, uint_t *protp,
4534 page_t *plarr[], size_t plsz, struct seg *seg, caddr_t addr,
4535 enum seg_rw rw, struct cred *cr, caller_context_t *ct)
4536 {
4537 u_offset_t uoff = (u_offset_t)off; /* type conversion */
4538 u_offset_t pgoff;
4539 u_offset_t eoff;
4540 struct inode *ip = VTOI(vp);
4541 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
4542 struct fs *fs;
4543 struct ulockfs *ulp;
4544 page_t **pl;
4545 caddr_t pgaddr;
4546 krw_t rwtype;
4547 int err;
4548 int has_holes;
4549 int beyond_eof;
4550 int seqmode;
4551 int pgsize = PAGESIZE;
4552 int dolock;
4553 int do_qlock;
4554 int trans_size;
4555
4556 ASSERT((uoff & PAGEOFFSET) == 0);
4557
4558 if (protp)
4559 *protp = PROT_ALL;
4560
4561 /*
4562 * Obey the lockfs protocol
4563 */
4564 err = ufs_lockfs_begin_getpage(ufsvfsp, &ulp, seg,
4565 rw == S_READ || rw == S_EXEC, protp);
4566 if (err)
4567 goto out;
4568
4569 fs = ufsvfsp->vfs_fs;
4570
4571 if (ulp && (rw == S_CREATE || rw == S_WRITE) &&
4572 !(vp->v_flag & VISSWAP)) {
4573 /*
4574 * Try to start a transaction, will return if blocking is
4575 * expected to occur and the address space is not the
4576 * kernel address space.
4577 */
4578 trans_size = TOP_GETPAGE_SIZE(ip);
4579 if (seg->s_as != &kas) {
4580 TRANS_TRY_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE,
4581 trans_size, err)
4582 if (err == EWOULDBLOCK) {
4583 /*
4584 * Use EDEADLK here because the VM code
4585 * can normally never see this error.
4586 */
4587 err = EDEADLK;
4588 ufs_lockfs_end(ulp);
4589 goto out;
4590 }
4591 } else {
4592 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4593 }
4594 }
4595
4596 if (vp->v_flag & VNOMAP) {
4597 err = ENOSYS;
4598 goto unlock;
4599 }
4600
4601 seqmode = ip->i_nextr == uoff && rw != S_CREATE;
4602
4603 rwtype = RW_READER; /* start as a reader */
4604 dolock = (rw_owner(&ip->i_contents) != curthread);
4605 /*
4606 * If this thread owns the lock, i.e., this thread grabbed it
4607 * as writer somewhere above, then we don't need to grab the
4608 * lock as reader in this routine.
4609 */
4610 do_qlock = (rw_owner(&ufsvfsp->vfs_dqrwlock) != curthread);
4611
4612 retrylock:
4613 if (dolock) {
4614 /*
4615 * Grab the quota lock if we need to call
4616 * bmap_write() below (with i_contents as writer).
4617 */
4618 if (do_qlock && rwtype == RW_WRITER)
4619 rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
4620 rw_enter(&ip->i_contents, rwtype);
4621 }
4622
4623 /*
4624 * We may be getting called as a side effect of a bmap using
4625 * fbread() when the blocks might be being allocated and the
4626 * size has not yet been up'ed. In this case we want to be
4627 * able to return zero pages if we get back UFS_HOLE from
4628 * calling bmap for a non write case here. We also might have
4629 * to read some frags from the disk into a page if we are
4630 * extending the number of frags for a given lbn in bmap().
4631 * Large Files: The read of i_size here is atomic because
4632 * i_contents is held here. If dolock is zero, the lock
4633 * is held in bmap routines.
4634 */
4635 beyond_eof = uoff + len >
4636 P2ROUNDUP_TYPED(ip->i_size, PAGESIZE, u_offset_t);
4637 if (beyond_eof && seg != segkmap) {
4638 if (dolock) {
4639 rw_exit(&ip->i_contents);
4640 if (do_qlock && rwtype == RW_WRITER)
4641 rw_exit(&ufsvfsp->vfs_dqrwlock);
4642 }
4643 err = EFAULT;
4644 goto unlock;
4645 }
4646
4647 /*
4648 * Must hold i_contents lock throughout the call to pvn_getpages
4649 * since locked pages are returned from each call to ufs_getapage.
4650 * Must *not* return locked pages and then try for contents lock
4651 * due to lock ordering requirements (inode > page)
4652 */
4653
4654 has_holes = bmap_has_holes(ip);
4655
4656 if ((rw == S_WRITE || rw == S_CREATE) && has_holes && !beyond_eof) {
4657 int blk_size;
4658 u_offset_t offset;
4659
4660 /*
4661 * We must acquire the RW_WRITER lock in order to
4662 * call bmap_write().
4663 */
4664 if (dolock && rwtype == RW_READER) {
4665 rwtype = RW_WRITER;
4666
4667 /*
4668 * Grab the quota lock before
4669 * upgrading i_contents, but if we can't grab it
4670 * don't wait here due to lock order:
4671 * vfs_dqrwlock > i_contents.
4672 */
4673 if (do_qlock &&
4674 rw_tryenter(&ufsvfsp->vfs_dqrwlock, RW_READER)
4675 == 0) {
4676 rw_exit(&ip->i_contents);
4677 goto retrylock;
4678 }
4679 if (!rw_tryupgrade(&ip->i_contents)) {
4680 rw_exit(&ip->i_contents);
4681 if (do_qlock)
4682 rw_exit(&ufsvfsp->vfs_dqrwlock);
4683 goto retrylock;
4684 }
4685 }
4686
4687 /*
4688 * May be allocating disk blocks for holes here as
4689 * a result of mmap faults. write(2) does the bmap_write
4690 * in rdip/wrip, not here. We are not dealing with frags
4691 * in this case.
4692 */
4693 /*
4694 * Large Files: We cast fs_bmask field to offset_t
4695 * just as we do for MAXBMASK because uoff is a 64-bit
4696 * data type. fs_bmask will still be a 32-bit type
4697 * as we cannot change any ondisk data structures.
4698 */
4699
4700 offset = uoff & (offset_t)fs->fs_bmask;
4701 while (offset < uoff + len) {
4702 blk_size = (int)blksize(fs, ip, lblkno(fs, offset));
4703 err = bmap_write(ip, offset, blk_size,
4704 BI_NORMAL, NULL, cr);
4705 if (ip->i_flag & (ICHG|IUPD))
4706 ip->i_seq++;
4707 if (err)
4708 goto update_inode;
4709 offset += blk_size; /* XXX - make this contig */
4710 }
4711 }
4712
4713 /*
4714 * Can be a reader from now on.
4715 */
4716 if (dolock && rwtype == RW_WRITER) {
4717 rw_downgrade(&ip->i_contents);
4718 /*
4719 * We can release vfs_dqrwlock early so do it, but make
4720 * sure we don't try to release it again at the bottom.
4721 */
4722 if (do_qlock) {
4723 rw_exit(&ufsvfsp->vfs_dqrwlock);
4724 do_qlock = 0;
4725 }
4726 }
4727
4728 /*
4729 * We remove PROT_WRITE in cases when the file has UFS holes
4730 * because we don't want to call bmap_read() to check each
4731 * page if it is backed with a disk block.
4732 */
4733 if (protp && has_holes && rw != S_WRITE && rw != S_CREATE)
4734 *protp &= ~PROT_WRITE;
4735
4736 err = 0;
4737
4738 /*
4739 * The loop looks up pages in the range [off, off + len).
4740 * For each page, we first check if we should initiate an asynchronous
4741 * read ahead before we call page_lookup (we may sleep in page_lookup
4742 * for a previously initiated disk read).
4743 */
4744 eoff = (uoff + len);
4745 for (pgoff = uoff, pgaddr = addr, pl = plarr;
4746 pgoff < eoff; /* empty */) {
4747 page_t *pp;
4748 u_offset_t nextrio;
4749 se_t se;
4750 int retval;
4751
4752 se = ((rw == S_CREATE || rw == S_OTHER) ? SE_EXCL : SE_SHARED);
4753
4754 /* Handle async getpage (faultahead) */
4755 if (plarr == NULL) {
4756 ip->i_nextrio = pgoff;
4757 (void) ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4758 pgoff += pgsize;
4759 pgaddr += pgsize;
4760 continue;
4761 }
4762 /*
4763 * Check if we should initiate read ahead of next cluster.
4764 * We call page_exists only when we need to confirm that
4765 * we have the current page before we initiate the read ahead.
4766 */
4767 nextrio = ip->i_nextrio;
4768 if (seqmode &&
4769 pgoff + CLUSTSZ(ip) >= nextrio && pgoff <= nextrio &&
4770 nextrio < ip->i_size && page_exists(vp, pgoff)) {
4771 retval = ufs_getpage_ra(vp, pgoff, seg, pgaddr);
4772 /*
4773 * We always read ahead the next cluster of data
4774 * starting from i_nextrio. If the page (vp,nextrio)
4775 * is actually in core at this point, the routine
4776 * ufs_getpage_ra() will stop pre-fetching data
4777 * until we read that page in a synchronized manner
4778 * through ufs_getpage_miss(). So, we should increase
4779 * i_nextrio if the page (vp, nextrio) exists.
4780 */
4781 if ((retval == 0) && page_exists(vp, nextrio)) {
4782 ip->i_nextrio = nextrio + pgsize;
4783 }
4784 }
4785
4786 if ((pp = page_lookup(vp, pgoff, se)) != NULL) {
4787 /*
4788 * We found the page in the page cache.
4789 */
4790 *pl++ = pp;
4791 pgoff += pgsize;
4792 pgaddr += pgsize;
4793 len -= pgsize;
4794 plsz -= pgsize;
4795 } else {
4796 /*
4797 * We have to create the page, or read it from disk.
4798 */
4799 if (err = ufs_getpage_miss(vp, pgoff, len, seg, pgaddr,
4800 pl, plsz, rw, seqmode))
4801 goto error;
4802
4803 while (*pl != NULL) {
4804 pl++;
4805 pgoff += pgsize;
4806 pgaddr += pgsize;
4807 len -= pgsize;
4808 plsz -= pgsize;
4809 }
4810 }
4811 }
4812
4813 /*
4814 * Return pages up to plsz if they are in the page cache.
4815 * We cannot return pages if there is a chance that they are
4816 * backed with a UFS hole and rw is S_WRITE or S_CREATE.
4817 */
4818 if (plarr && !(has_holes && (rw == S_WRITE || rw == S_CREATE))) {
4819
4820 ASSERT((protp == NULL) ||
4821 !(has_holes && (*protp & PROT_WRITE)));
4822
4823 eoff = pgoff + plsz;
4824 while (pgoff < eoff) {
4825 page_t *pp;
4826
4827 if ((pp = page_lookup_nowait(vp, pgoff,
4828 SE_SHARED)) == NULL)
4829 break;
4830
4831 *pl++ = pp;
4832 pgoff += pgsize;
4833 plsz -= pgsize;
4834 }
4835 }
4836
4837 if (plarr)
4838 *pl = NULL; /* Terminate page list */
4839 ip->i_nextr = pgoff;
4840
4841 error:
4842 if (err && plarr) {
4843 /*
4844 * Release any pages we have locked.
4845 */
4846 while (pl > &plarr[0])
4847 page_unlock(*--pl);
4848
4849 plarr[0] = NULL;
4850 }
4851
4852 update_inode:
4853 /*
4854 * If the inode is not already marked for IACC (in rdip() for read)
4855 * and the inode is not marked for no access time update (in wrip()
4856 * for write) then update the inode access time and mod time now.
4857 */
4858 if ((ip->i_flag & (IACC | INOACC)) == 0) {
4859 if ((rw != S_OTHER) && (ip->i_mode & IFMT) != IFDIR) {
4860 if (!ULOCKFS_IS_NOIACC(ITOUL(ip)) &&
4861 (fs->fs_ronly == 0) &&
4862 (!ufsvfsp->vfs_noatime)) {
4863 mutex_enter(&ip->i_tlock);
4864 ip->i_flag |= IACC;
4865 ITIMES_NOLOCK(ip);
4866 mutex_exit(&ip->i_tlock);
4867 }
4868 }
4869 }
4870
4871 if (dolock) {
4872 rw_exit(&ip->i_contents);
4873 if (do_qlock && rwtype == RW_WRITER)
4874 rw_exit(&ufsvfsp->vfs_dqrwlock);
4875 }
4876
4877 unlock:
4878 if (ulp) {
4879 if ((rw == S_CREATE || rw == S_WRITE) &&
4880 !(vp->v_flag & VISSWAP)) {
4881 TRANS_END_ASYNC(ufsvfsp, TOP_GETPAGE, trans_size);
4882 }
4883 ufs_lockfs_end(ulp);
4884 }
4885 out:
4886 return (err);
4887 }
4888
4889 /*
4890 * ufs_getpage_miss is called when ufs_getpage missed the page in the page
4891 * cache. The page is either read from the disk, or it's created.
4892 * A page is created (without disk read) if rw == S_CREATE, or if
4893 * the page is not backed with a real disk block (UFS hole).
4894 */
4895 /* ARGSUSED */
4896 static int
4897 ufs_getpage_miss(struct vnode *vp, u_offset_t off, size_t len, struct seg *seg,
4898 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw, int seq)
4899 {
4900 struct inode *ip = VTOI(vp);
4901 page_t *pp;
4902 daddr_t bn;
4903 size_t io_len;
4904 int crpage = 0;
4905 int err;
4906 int contig;
4907 int bsize = ip->i_fs->fs_bsize;
4908
4909 /*
4910 * Figure out whether the page can be created, or must be
4911 * must be read from the disk.
4912 */
4913 if (rw == S_CREATE)
4914 crpage = 1;
4915 else {
4916 contig = 0;
4917 if (err = bmap_read(ip, off, &bn, &contig))
4918 return (err);
4919
4920 crpage = (bn == UFS_HOLE);
4921
4922 /*
4923 * If its also a fallocated block that hasn't been written to
4924 * yet, we will treat it just like a UFS_HOLE and create
4925 * a zero page for it
4926 */
4927 if (ISFALLOCBLK(ip, bn))
4928 crpage = 1;
4929 }
4930
4931 if (crpage) {
4932 if ((pp = page_create_va(vp, off, PAGESIZE, PG_WAIT, seg,
4933 addr)) == NULL) {
4934 return (ufs_fault(vp,
4935 "ufs_getpage_miss: page_create == NULL"));
4936 }
4937
4938 if (rw != S_CREATE)
4939 pagezero(pp, 0, PAGESIZE);
4940
4941 io_len = PAGESIZE;
4942 } else {
4943 u_offset_t io_off;
4944 uint_t xlen;
4945 struct buf *bp;
4946 ufsvfs_t *ufsvfsp = ip->i_ufsvfs;
4947
4948 /*
4949 * If access is not in sequential order, we read from disk
4950 * in bsize units.
4951 *
4952 * We limit the size of the transfer to bsize if we are reading
4953 * from the beginning of the file. Note in this situation we
4954 * will hedge our bets and initiate an async read ahead of
4955 * the second block.
4956 */
4957 if (!seq || off == 0)
4958 contig = MIN(contig, bsize);
4959
4960 pp = pvn_read_kluster(vp, off, seg, addr, &io_off,
4961 &io_len, off, contig, 0);
4962
4963 /*
4964 * Some other thread has entered the page.
4965 * ufs_getpage will retry page_lookup.
4966 */
4967 if (pp == NULL) {
4968 pl[0] = NULL;
4969 return (0);
4970 }
4971
4972 /*
4973 * Zero part of the page which we are not
4974 * going to read from the disk.
4975 */
4976 xlen = io_len & PAGEOFFSET;
4977 if (xlen != 0)
4978 pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
4979
4980 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ);
4981 bp->b_edev = ip->i_dev;
4982 bp->b_dev = cmpdev(ip->i_dev);
4983 bp->b_blkno = bn;
4984 bp->b_un.b_addr = (caddr_t)0;
4985 bp->b_file = ip->i_vnode;
4986 bp->b_offset = off;
4987
4988 if (ufsvfsp->vfs_log) {
4989 lufs_read_strategy(ufsvfsp->vfs_log, bp);
4990 } else if (ufsvfsp->vfs_snapshot) {
4991 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
4992 } else {
4993 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
4994 ub.ub_getpages.value.ul++;
4995 (void) bdev_strategy(bp);
4996 lwp_stat_update(LWP_STAT_INBLK, 1);
4997 }
4998
4999 ip->i_nextrio = off + ((io_len + PAGESIZE - 1) & PAGEMASK);
5000
5001 /*
5002 * If the file access is sequential, initiate read ahead
5003 * of the next cluster.
5004 */
5005 if (seq && ip->i_nextrio < ip->i_size)
5006 (void) ufs_getpage_ra(vp, off, seg, addr);
5007 err = biowait(bp);
5008 pageio_done(bp);
5009
5010 if (err) {
5011 pvn_read_done(pp, B_ERROR);
5012 return (err);
5013 }
5014 }
5015
5016 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
5017 return (0);
5018 }
5019
5020 /*
5021 * Read ahead a cluster from the disk. Returns the length in bytes.
5022 */
5023 static int
5024 ufs_getpage_ra(struct vnode *vp, u_offset_t off, struct seg *seg, caddr_t addr)
5025 {
5026 struct inode *ip = VTOI(vp);
5027 page_t *pp;
5028 u_offset_t io_off = ip->i_nextrio;
5029 ufsvfs_t *ufsvfsp;
5030 caddr_t addr2 = addr + (io_off - off);
5031 struct buf *bp;
5032 daddr_t bn;
5033 size_t io_len;
5034 int err;
5035 int contig;
5036 int xlen;
5037 int bsize = ip->i_fs->fs_bsize;
5038
5039 /*
5040 * If the directio advisory is in effect on this file,
5041 * then do not do buffered read ahead. Read ahead makes
5042 * it more difficult on threads using directio as they
5043 * will be forced to flush the pages from this vnode.
5044 */
5045 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5046 return (0);
5047 if (ip->i_flag & IDIRECTIO || ufsvfsp->vfs_forcedirectio)
5048 return (0);
5049
5050 /*
5051 * Is this test needed?
5052 */
5053 if (addr2 >= seg->s_base + seg->s_size)
5054 return (0);
5055
5056 contig = 0;
5057 err = bmap_read(ip, io_off, &bn, &contig);
5058 /*
5059 * If its a UFS_HOLE or a fallocated block, do not perform
5060 * any read ahead's since there probably is nothing to read ahead
5061 */
5062 if (err || bn == UFS_HOLE || ISFALLOCBLK(ip, bn))
5063 return (0);
5064
5065 /*
5066 * Limit the transfer size to bsize if this is the 2nd block.
5067 */
5068 if (io_off == (u_offset_t)bsize)
5069 contig = MIN(contig, bsize);
5070
5071 if ((pp = pvn_read_kluster(vp, io_off, seg, addr2, &io_off,
5072 &io_len, io_off, contig, 1)) == NULL)
5073 return (0);
5074
5075 /*
5076 * Zero part of page which we are not going to read from disk
5077 */
5078 if ((xlen = (io_len & PAGEOFFSET)) > 0)
5079 pagezero(pp->p_prev, xlen, PAGESIZE - xlen);
5080
5081 ip->i_nextrio = (io_off + io_len + PAGESIZE - 1) & PAGEMASK;
5082
5083 bp = pageio_setup(pp, io_len, ip->i_devvp, B_READ | B_ASYNC);
5084 bp->b_edev = ip->i_dev;
5085 bp->b_dev = cmpdev(ip->i_dev);
5086 bp->b_blkno = bn;
5087 bp->b_un.b_addr = (caddr_t)0;
5088 bp->b_file = ip->i_vnode;
5089 bp->b_offset = off;
5090
5091 if (ufsvfsp->vfs_log) {
5092 lufs_read_strategy(ufsvfsp->vfs_log, bp);
5093 } else if (ufsvfsp->vfs_snapshot) {
5094 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5095 } else {
5096 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5097 ub.ub_getras.value.ul++;
5098 (void) bdev_strategy(bp);
5099 lwp_stat_update(LWP_STAT_INBLK, 1);
5100 }
5101
5102 return (io_len);
5103 }
5104
5105 int ufs_delay = 1;
5106 /*
5107 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE, B_ASYNC}
5108 *
5109 * LMXXX - the inode really ought to contain a pointer to one of these
5110 * async args. Stuff gunk in there and just hand the whole mess off.
5111 * This would replace i_delaylen, i_delayoff.
5112 */
5113 /*ARGSUSED*/
5114 static int
5115 ufs_putpage(struct vnode *vp, offset_t off, size_t len, int flags,
5116 struct cred *cr, caller_context_t *ct)
5117 {
5118 struct inode *ip = VTOI(vp);
5119 int err = 0;
5120
5121 if (vp->v_count == 0) {
5122 return (ufs_fault(vp, "ufs_putpage: bad v_count == 0"));
5123 }
5124
5125 /*
5126 * XXX - Why should this check be made here?
5127 */
5128 if (vp->v_flag & VNOMAP) {
5129 err = ENOSYS;
5130 goto errout;
5131 }
5132
5133 if (ip->i_ufsvfs == NULL) {
5134 err = EIO;
5135 goto errout;
5136 }
5137
5138 if (flags & B_ASYNC) {
5139 if (ufs_delay && len &&
5140 (flags & ~(B_ASYNC|B_DONTNEED|B_FREE)) == 0) {
5141 mutex_enter(&ip->i_tlock);
5142 /*
5143 * If nobody stalled, start a new cluster.
5144 */
5145 if (ip->i_delaylen == 0) {
5146 ip->i_delayoff = off;
5147 ip->i_delaylen = len;
5148 mutex_exit(&ip->i_tlock);
5149 goto errout;
5150 }
5151 /*
5152 * If we have a full cluster or they are not contig,
5153 * then push last cluster and start over.
5154 */
5155 if (ip->i_delaylen >= CLUSTSZ(ip) ||
5156 ip->i_delayoff + ip->i_delaylen != off) {
5157 u_offset_t doff;
5158 size_t dlen;
5159
5160 doff = ip->i_delayoff;
5161 dlen = ip->i_delaylen;
5162 ip->i_delayoff = off;
5163 ip->i_delaylen = len;
5164 mutex_exit(&ip->i_tlock);
5165 err = ufs_putpages(vp, doff, dlen,
5166 flags, cr);
5167 /* LMXXX - flags are new val, not old */
5168 goto errout;
5169 }
5170 /*
5171 * There is something there, it's not full, and
5172 * it is contig.
5173 */
5174 ip->i_delaylen += len;
5175 mutex_exit(&ip->i_tlock);
5176 goto errout;
5177 }
5178 /*
5179 * Must have weird flags or we are not clustering.
5180 */
5181 }
5182
5183 err = ufs_putpages(vp, off, len, flags, cr);
5184
5185 errout:
5186 return (err);
5187 }
5188
5189 /*
5190 * If len == 0, do from off to EOF.
5191 *
5192 * The normal cases should be len == 0 & off == 0 (entire vp list),
5193 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
5194 * (from pageout).
5195 */
5196 /*ARGSUSED*/
5197 static int
5198 ufs_putpages(
5199 struct vnode *vp,
5200 offset_t off,
5201 size_t len,
5202 int flags,
5203 struct cred *cr)
5204 {
5205 u_offset_t io_off;
5206 u_offset_t eoff;
5207 struct inode *ip = VTOI(vp);
5208 page_t *pp;
5209 size_t io_len;
5210 int err = 0;
5211 int dolock;
5212
5213 if (vp->v_count == 0)
5214 return (ufs_fault(vp, "ufs_putpages: v_count == 0"));
5215 /*
5216 * Acquire the readers/write inode lock before locking
5217 * any pages in this inode.
5218 * The inode lock is held during i/o.
5219 */
5220 if (len == 0) {
5221 mutex_enter(&ip->i_tlock);
5222 ip->i_delayoff = ip->i_delaylen = 0;
5223 mutex_exit(&ip->i_tlock);
5224 }
5225 dolock = (rw_owner(&ip->i_contents) != curthread);
5226 if (dolock) {
5227 /*
5228 * Must synchronize this thread and any possible thread
5229 * operating in the window of vulnerability in wrip().
5230 * It is dangerous to allow both a thread doing a putpage
5231 * and a thread writing, so serialize them. The exception
5232 * is when the thread in wrip() does something which causes
5233 * a putpage operation. Then, the thread must be allowed
5234 * to continue. It may encounter a bmap_read problem in
5235 * ufs_putapage, but that is handled in ufs_putapage.
5236 * Allow async writers to proceed, we don't want to block
5237 * the pageout daemon.
5238 */
5239 if (ip->i_writer == curthread)
5240 rw_enter(&ip->i_contents, RW_READER);
5241 else {
5242 for (;;) {
5243 rw_enter(&ip->i_contents, RW_READER);
5244 mutex_enter(&ip->i_tlock);
5245 /*
5246 * If there is no thread in the critical
5247 * section of wrip(), then proceed.
5248 * Otherwise, wait until there isn't one.
5249 */
5250 if (ip->i_writer == NULL) {
5251 mutex_exit(&ip->i_tlock);
5252 break;
5253 }
5254 rw_exit(&ip->i_contents);
5255 /*
5256 * Bounce async writers when we have a writer
5257 * working on this file so we don't deadlock
5258 * the pageout daemon.
5259 */
5260 if (flags & B_ASYNC) {
5261 mutex_exit(&ip->i_tlock);
5262 return (0);
5263 }
5264 cv_wait(&ip->i_wrcv, &ip->i_tlock);
5265 mutex_exit(&ip->i_tlock);
5266 }
5267 }
5268 }
5269
5270 if (!vn_has_cached_data(vp)) {
5271 if (dolock)
5272 rw_exit(&ip->i_contents);
5273 return (0);
5274 }
5275
5276 if (len == 0) {
5277 /*
5278 * Search the entire vp list for pages >= off.
5279 */
5280 err = pvn_vplist_dirty(vp, (u_offset_t)off, ufs_putapage,
5281 flags, cr);
5282 } else {
5283 /*
5284 * Loop over all offsets in the range looking for
5285 * pages to deal with.
5286 */
5287 if ((eoff = blkroundup(ip->i_fs, ip->i_size)) != 0)
5288 eoff = MIN(off + len, eoff);
5289 else
5290 eoff = off + len;
5291
5292 for (io_off = off; io_off < eoff; io_off += io_len) {
5293 /*
5294 * If we are not invalidating, synchronously
5295 * freeing or writing pages, use the routine
5296 * page_lookup_nowait() to prevent reclaiming
5297 * them from the free list.
5298 */
5299 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
5300 pp = page_lookup(vp, io_off,
5301 (flags & (B_INVAL | B_FREE)) ?
5302 SE_EXCL : SE_SHARED);
5303 } else {
5304 pp = page_lookup_nowait(vp, io_off,
5305 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
5306 }
5307
5308 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
5309 io_len = PAGESIZE;
5310 else {
5311 u_offset_t *io_offp = &io_off;
5312
5313 err = ufs_putapage(vp, pp, io_offp, &io_len,
5314 flags, cr);
5315 if (err != 0)
5316 break;
5317 /*
5318 * "io_off" and "io_len" are returned as
5319 * the range of pages we actually wrote.
5320 * This allows us to skip ahead more quickly
5321 * since several pages may've been dealt
5322 * with by this iteration of the loop.
5323 */
5324 }
5325 }
5326 }
5327 if (err == 0 && off == 0 && (len == 0 || len >= ip->i_size)) {
5328 /*
5329 * We have just sync'ed back all the pages on
5330 * the inode, turn off the IMODTIME flag.
5331 */
5332 mutex_enter(&ip->i_tlock);
5333 ip->i_flag &= ~IMODTIME;
5334 mutex_exit(&ip->i_tlock);
5335 }
5336 if (dolock)
5337 rw_exit(&ip->i_contents);
5338 return (err);
5339 }
5340
5341 static void
5342 ufs_iodone(buf_t *bp)
5343 {
5344 struct inode *ip;
5345
5346 ASSERT((bp->b_pages->p_vnode != NULL) && !(bp->b_flags & B_READ));
5347
5348 bp->b_iodone = NULL;
5349
5350 ip = VTOI(bp->b_pages->p_vnode);
5351
5352 mutex_enter(&ip->i_tlock);
5353 if (ip->i_writes >= ufs_LW) {
5354 if ((ip->i_writes -= bp->b_bcount) <= ufs_LW)
5355 if (ufs_WRITES)
5356 cv_broadcast(&ip->i_wrcv); /* wake all up */
5357 } else {
5358 ip->i_writes -= bp->b_bcount;
5359 }
5360
5361 mutex_exit(&ip->i_tlock);
5362 iodone(bp);
5363 }
5364
5365 /*
5366 * Write out a single page, possibly klustering adjacent
5367 * dirty pages. The inode lock must be held.
5368 *
5369 * LMXXX - bsize < pagesize not done.
5370 */
5371 /*ARGSUSED*/
5372 int
5373 ufs_putapage(
5374 struct vnode *vp,
5375 page_t *pp,
5376 u_offset_t *offp,
5377 size_t *lenp, /* return values */
5378 int flags,
5379 struct cred *cr)
5380 {
5381 u_offset_t io_off;
5382 u_offset_t off;
5383 struct inode *ip = VTOI(vp);
5384 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
5385 struct fs *fs;
5386 struct buf *bp;
5387 size_t io_len;
5388 daddr_t bn;
5389 int err;
5390 int contig;
5391 int dotrans;
5392
5393 ASSERT(RW_LOCK_HELD(&ip->i_contents));
5394
5395 if (ufsvfsp == NULL) {
5396 err = EIO;
5397 goto out_trace;
5398 }
5399
5400 fs = ip->i_fs;
5401 ASSERT(fs->fs_ronly == 0);
5402
5403 /*
5404 * If the modified time on the inode has not already been
5405 * set elsewhere (e.g. for write/setattr) we set the time now.
5406 * This gives us approximate modified times for mmap'ed files
5407 * which are modified via stores in the user address space.
5408 */
5409 if ((ip->i_flag & IMODTIME) == 0) {
5410 mutex_enter(&ip->i_tlock);
5411 ip->i_flag |= IUPD;
5412 ip->i_seq++;
5413 ITIMES_NOLOCK(ip);
5414 mutex_exit(&ip->i_tlock);
5415 }
5416
5417 /*
5418 * Align the request to a block boundry (for old file systems),
5419 * and go ask bmap() how contiguous things are for this file.
5420 */
5421 off = pp->p_offset & (offset_t)fs->fs_bmask; /* block align it */
5422 contig = 0;
5423 err = bmap_read(ip, off, &bn, &contig);
5424 if (err)
5425 goto out;
5426 if (bn == UFS_HOLE) { /* putpage never allocates */
5427 /*
5428 * logging device is in error mode; simply return EIO
5429 */
5430 if (TRANS_ISERROR(ufsvfsp)) {
5431 err = EIO;
5432 goto out;
5433 }
5434 /*
5435 * Oops, the thread in the window in wrip() did some
5436 * sort of operation which caused a putpage in the bad
5437 * range. In this case, just return an error which will
5438 * cause the software modified bit on the page to set
5439 * and the page will get written out again later.
5440 */
5441 if (ip->i_writer == curthread) {
5442 err = EIO;
5443 goto out;
5444 }
5445 /*
5446 * If the pager is trying to push a page in the bad range
5447 * just tell him to try again later when things are better.
5448 */
5449 if (flags & B_ASYNC) {
5450 err = EAGAIN;
5451 goto out;
5452 }
5453 err = ufs_fault(ITOV(ip), "ufs_putapage: bn == UFS_HOLE");
5454 goto out;
5455 }
5456
5457 /*
5458 * If it is an fallocate'd block, reverse the negativity since
5459 * we are now writing to it
5460 */
5461 if (ISFALLOCBLK(ip, bn)) {
5462 err = bmap_set_bn(vp, off, dbtofsb(fs, -bn));
5463 if (err)
5464 goto out;
5465
5466 bn = -bn;
5467 }
5468
5469 /*
5470 * Take the length (of contiguous bytes) passed back from bmap()
5471 * and _try_ and get a set of pages covering that extent.
5472 */
5473 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, off, contig, flags);
5474
5475 /*
5476 * May have run out of memory and not clustered backwards.
5477 * off p_offset
5478 * [ pp - 1 ][ pp ]
5479 * [ block ]
5480 * We told bmap off, so we have to adjust the bn accordingly.
5481 */
5482 if (io_off > off) {
5483 bn += btod(io_off - off);
5484 contig -= (io_off - off);
5485 }
5486
5487 /*
5488 * bmap was carefull to tell us the right size so use that.
5489 * There might be unallocated frags at the end.
5490 * LMXXX - bzero the end of the page? We must be writing after EOF.
5491 */
5492 if (io_len > contig) {
5493 ASSERT(io_len - contig < fs->fs_bsize);
5494 io_len -= (io_len - contig);
5495 }
5496
5497 /*
5498 * Handle the case where we are writing the last page after EOF.
5499 *
5500 * XXX - just a patch for i-mt3.
5501 */
5502 if (io_len == 0) {
5503 ASSERT(pp->p_offset >=
5504 (u_offset_t)(roundup(ip->i_size, PAGESIZE)));
5505 io_len = PAGESIZE;
5506 }
5507
5508 bp = pageio_setup(pp, io_len, ip->i_devvp, B_WRITE | flags);
5509
5510 ULOCKFS_SET_MOD(ITOUL(ip));
5511
5512 bp->b_edev = ip->i_dev;
5513 bp->b_dev = cmpdev(ip->i_dev);
5514 bp->b_blkno = bn;
5515 bp->b_un.b_addr = (caddr_t)0;
5516 bp->b_file = ip->i_vnode;
5517
5518 /*
5519 * File contents of shadow or quota inodes are metadata, and updates
5520 * to these need to be put into a logging transaction. All direct
5521 * callers in UFS do that, but fsflush can come here _before_ the
5522 * normal codepath. An example would be updating ACL information, for
5523 * which the normal codepath would be:
5524 * ufs_si_store()
5525 * ufs_rdwri()
5526 * wrip()
5527 * segmap_release()
5528 * VOP_PUTPAGE()
5529 * Here, fsflush can pick up the dirty page before segmap_release()
5530 * forces it out. If that happens, there's no transaction.
5531 * We therefore need to test whether a transaction exists, and if not
5532 * create one - for fsflush.
5533 */
5534 dotrans =
5535 (((ip->i_mode & IFMT) == IFSHAD || ufsvfsp->vfs_qinod == ip) &&
5536 ((curthread->t_flag & T_DONTBLOCK) == 0) &&
5537 (TRANS_ISTRANS(ufsvfsp)));
5538
5539 if (dotrans) {
5540 curthread->t_flag |= T_DONTBLOCK;
5541 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5542 }
5543 if (TRANS_ISTRANS(ufsvfsp)) {
5544 if ((ip->i_mode & IFMT) == IFSHAD) {
5545 TRANS_BUF(ufsvfsp, 0, io_len, bp, DT_SHAD);
5546 } else if (ufsvfsp->vfs_qinod == ip) {
5547 TRANS_DELTA(ufsvfsp, ldbtob(bn), bp->b_bcount, DT_QR,
5548 0, 0);
5549 }
5550 }
5551 if (dotrans) {
5552 TRANS_END_ASYNC(ufsvfsp, TOP_PUTPAGE, TOP_PUTPAGE_SIZE(ip));
5553 curthread->t_flag &= ~T_DONTBLOCK;
5554 }
5555
5556 /* write throttle */
5557
5558 ASSERT(bp->b_iodone == NULL);
5559 bp->b_iodone = (int (*)())ufs_iodone;
5560 mutex_enter(&ip->i_tlock);
5561 ip->i_writes += bp->b_bcount;
5562 mutex_exit(&ip->i_tlock);
5563
5564 if (bp->b_flags & B_ASYNC) {
5565 if (ufsvfsp->vfs_log) {
5566 lufs_write_strategy(ufsvfsp->vfs_log, bp);
5567 } else if (ufsvfsp->vfs_snapshot) {
5568 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5569 } else {
5570 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5571 ub.ub_putasyncs.value.ul++;
5572 (void) bdev_strategy(bp);
5573 lwp_stat_update(LWP_STAT_OUBLK, 1);
5574 }
5575 } else {
5576 if (ufsvfsp->vfs_log) {
5577 lufs_write_strategy(ufsvfsp->vfs_log, bp);
5578 } else if (ufsvfsp->vfs_snapshot) {
5579 fssnap_strategy(&ufsvfsp->vfs_snapshot, bp);
5580 } else {
5581 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
5582 ub.ub_putsyncs.value.ul++;
5583 (void) bdev_strategy(bp);
5584 lwp_stat_update(LWP_STAT_OUBLK, 1);
5585 }
5586 err = biowait(bp);
5587 pageio_done(bp);
5588 pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
5589 }
5590
5591 pp = NULL;
5592
5593 out:
5594 if (err != 0 && pp != NULL)
5595 pvn_write_done(pp, B_ERROR | B_WRITE | flags);
5596
5597 if (offp)
5598 *offp = io_off;
5599 if (lenp)
5600 *lenp = io_len;
5601 out_trace:
5602 return (err);
5603 }
5604
5605 uint64_t ufs_map_alock_retry_cnt;
5606 uint64_t ufs_map_lockfs_retry_cnt;
5607
5608 /* ARGSUSED */
5609 static int
5610 ufs_map(struct vnode *vp,
5611 offset_t off,
5612 struct as *as,
5613 caddr_t *addrp,
5614 size_t len,
5615 uchar_t prot,
5616 uchar_t maxprot,
5617 uint_t flags,
5618 struct cred *cr,
5619 caller_context_t *ct)
5620 {
5621 struct segvn_crargs vn_a;
5622 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5623 struct ulockfs *ulp;
5624 int error, sig;
5625 k_sigset_t smask;
5626 caddr_t hint = *addrp;
5627
5628 if (vp->v_flag & VNOMAP) {
5629 error = ENOSYS;
5630 goto out;
5631 }
5632
5633 if (off < (offset_t)0 || (offset_t)(off + len) < (offset_t)0) {
5634 error = ENXIO;
5635 goto out;
5636 }
5637
5638 if (vp->v_type != VREG) {
5639 error = ENODEV;
5640 goto out;
5641 }
5642
5643 retry_map:
5644 *addrp = hint;
5645 /*
5646 * If file is being locked, disallow mapping.
5647 */
5648 if (vn_has_mandatory_locks(vp, VTOI(vp)->i_mode)) {
5649 error = EAGAIN;
5650 goto out;
5651 }
5652
5653 as_rangelock(as);
5654 /*
5655 * Note that if we are retrying (because ufs_lockfs_trybegin failed in
5656 * the previous attempt), some other thread could have grabbed
5657 * the same VA range if MAP_FIXED is set. In that case, choose_addr
5658 * would unmap the valid VA range, that is ok.
5659 */
5660 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
5661 if (error != 0) {
5662 as_rangeunlock(as);
5663 goto out;
5664 }
5665
5666 /*
5667 * a_lock has to be acquired before entering the lockfs protocol
5668 * because that is the order in which pagefault works. Also we cannot
5669 * block on a_lock here because this waiting writer will prevent
5670 * further readers like ufs_read from progressing and could cause
5671 * deadlock between ufs_read/ufs_map/pagefault when a quiesce is
5672 * pending.
5673 */
5674 while (!AS_LOCK_TRYENTER(as, RW_WRITER)) {
5675 ufs_map_alock_retry_cnt++;
5676 delay(RETRY_LOCK_DELAY);
5677 }
5678
5679 /*
5680 * We can't hold as->a_lock and wait for lockfs to succeed because
5681 * the proc tools might hang on a_lock, so call ufs_lockfs_trybegin()
5682 * instead.
5683 */
5684 if (error = ufs_lockfs_trybegin(ufsvfsp, &ulp, ULOCKFS_MAP_MASK)) {
5685 /*
5686 * ufs_lockfs_trybegin() did not succeed. It is safer to give up
5687 * as->a_lock and wait for ulp->ul_fs_lock status to change.
5688 */
5689 ufs_map_lockfs_retry_cnt++;
5690 AS_LOCK_EXIT(as);
5691 as_rangeunlock(as);
5692 if (error == EIO)
5693 goto out;
5694
5695 mutex_enter(&ulp->ul_lock);
5696 while (ulp->ul_fs_lock & ULOCKFS_MAP_MASK) {
5697 if (ULOCKFS_IS_SLOCK(ulp) || ufsvfsp->vfs_nointr) {
5698 cv_wait(&ulp->ul_cv, &ulp->ul_lock);
5699 } else {
5700 sigintr(&smask, 1);
5701 sig = cv_wait_sig(&ulp->ul_cv, &ulp->ul_lock);
5702 sigunintr(&smask);
5703 if (((ulp->ul_fs_lock & ULOCKFS_MAP_MASK) &&
5704 !sig) || ufsvfsp->vfs_dontblock) {
5705 mutex_exit(&ulp->ul_lock);
5706 return (EINTR);
5707 }
5708 }
5709 }
5710 mutex_exit(&ulp->ul_lock);
5711 goto retry_map;
5712 }
5713
5714 vn_a.vp = vp;
5715 vn_a.offset = (u_offset_t)off;
5716 vn_a.type = flags & MAP_TYPE;
5717 vn_a.prot = prot;
5718 vn_a.maxprot = maxprot;
5719 vn_a.cred = cr;
5720 vn_a.amp = NULL;
5721 vn_a.flags = flags & ~MAP_TYPE;
5722 vn_a.szc = 0;
5723 vn_a.lgrp_mem_policy_flags = 0;
5724
5725 error = as_map_locked(as, *addrp, len, segvn_create, &vn_a);
5726 if (ulp)
5727 ufs_lockfs_end(ulp);
5728 as_rangeunlock(as);
5729 out:
5730 return (error);
5731 }
5732
5733 /* ARGSUSED */
5734 static int
5735 ufs_addmap(struct vnode *vp,
5736 offset_t off,
5737 struct as *as,
5738 caddr_t addr,
5739 size_t len,
5740 uchar_t prot,
5741 uchar_t maxprot,
5742 uint_t flags,
5743 struct cred *cr,
5744 caller_context_t *ct)
5745 {
5746 struct inode *ip = VTOI(vp);
5747
5748 if (vp->v_flag & VNOMAP) {
5749 return (ENOSYS);
5750 }
5751
5752 mutex_enter(&ip->i_tlock);
5753 ip->i_mapcnt += btopr(len);
5754 mutex_exit(&ip->i_tlock);
5755 return (0);
5756 }
5757
5758 /*ARGSUSED*/
5759 static int
5760 ufs_delmap(struct vnode *vp, offset_t off, struct as *as, caddr_t addr,
5761 size_t len, uint_t prot, uint_t maxprot, uint_t flags,
5762 struct cred *cr, caller_context_t *ct)
5763 {
5764 struct inode *ip = VTOI(vp);
5765
5766 if (vp->v_flag & VNOMAP) {
5767 return (ENOSYS);
5768 }
5769
5770 mutex_enter(&ip->i_tlock);
5771 ip->i_mapcnt -= btopr(len); /* Count released mappings */
5772 ASSERT(ip->i_mapcnt >= 0);
5773 mutex_exit(&ip->i_tlock);
5774 return (0);
5775 }
5776 /*
5777 * Return the answer requested to poll() for non-device files
5778 */
5779 struct pollhead ufs_pollhd;
5780
5781 /* ARGSUSED */
5782 int
5783 ufs_poll(vnode_t *vp, short ev, int any, short *revp, struct pollhead **phpp,
5784 caller_context_t *ct)
5785 {
5786 struct ufsvfs *ufsvfsp;
5787
5788 *revp = 0;
5789 ufsvfsp = VTOI(vp)->i_ufsvfs;
5790
5791 if (!ufsvfsp) {
5792 *revp = POLLHUP;
5793 goto out;
5794 }
5795
5796 if (ULOCKFS_IS_HLOCK(&ufsvfsp->vfs_ulockfs) ||
5797 ULOCKFS_IS_ELOCK(&ufsvfsp->vfs_ulockfs)) {
5798 *revp |= POLLERR;
5799
5800 } else {
5801 if ((ev & POLLOUT) && !ufsvfsp->vfs_fs->fs_ronly &&
5802 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5803 *revp |= POLLOUT;
5804
5805 if ((ev & POLLWRBAND) && !ufsvfsp->vfs_fs->fs_ronly &&
5806 !ULOCKFS_IS_WLOCK(&ufsvfsp->vfs_ulockfs))
5807 *revp |= POLLWRBAND;
5808
5809 if (ev & POLLIN)
5810 *revp |= POLLIN;
5811
5812 if (ev & POLLRDNORM)
5813 *revp |= POLLRDNORM;
5814
5815 if (ev & POLLRDBAND)
5816 *revp |= POLLRDBAND;
5817 }
5818
5819 if ((ev & POLLPRI) && (*revp & (POLLERR|POLLHUP)))
5820 *revp |= POLLPRI;
5821 out:
5822 *phpp = !any && !*revp ? &ufs_pollhd : (struct pollhead *)NULL;
5823
5824 return (0);
5825 }
5826
5827 /* ARGSUSED */
5828 static int
5829 ufs_l_pathconf(struct vnode *vp, int cmd, ulong_t *valp, struct cred *cr,
5830 caller_context_t *ct)
5831 {
5832 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
5833 struct ulockfs *ulp = NULL;
5834 struct inode *sip = NULL;
5835 int error;
5836 struct inode *ip = VTOI(vp);
5837 int issync;
5838
5839 error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_PATHCONF_MASK);
5840 if (error)
5841 return (error);
5842
5843 switch (cmd) {
5844 /*
5845 * Have to handle _PC_NAME_MAX here, because the normal way
5846 * [fs_pathconf() -> VOP_STATVFS() -> ufs_statvfs()]
5847 * results in a lock ordering reversal between
5848 * ufs_lockfs_{begin,end}() and
5849 * ufs_thread_{suspend,continue}().
5850 *
5851 * Keep in sync with ufs_statvfs().
5852 */
5853 case _PC_NAME_MAX:
5854 *valp = MAXNAMLEN;
5855 break;
5856
5857 case _PC_FILESIZEBITS:
5858 if (ufsvfsp->vfs_lfflags & UFS_LARGEFILES)
5859 *valp = UFS_FILESIZE_BITS;
5860 else
5861 *valp = 32;
5862 break;
5863
5864 case _PC_XATTR_EXISTS:
5865 if (vp->v_vfsp->vfs_flag & VFS_XATTR) {
5866
5867 error =
5868 ufs_xattr_getattrdir(vp, &sip, LOOKUP_XATTR, cr);
5869 if (error == 0 && sip != NULL) {
5870 /* Start transaction */
5871 if (ulp) {
5872 TRANS_BEGIN_CSYNC(ufsvfsp, issync,
5873 TOP_RMDIR, TOP_RMDIR_SIZE);
5874 }
5875 /*
5876 * Is directory empty
5877 */
5878 rw_enter(&sip->i_rwlock, RW_WRITER);
5879 rw_enter(&sip->i_contents, RW_WRITER);
5880 if (ufs_xattrdirempty(sip,
5881 sip->i_number, CRED())) {
5882 rw_enter(&ip->i_contents, RW_WRITER);
5883 ufs_unhook_shadow(ip, sip);
5884 rw_exit(&ip->i_contents);
5885
5886 *valp = 0;
5887
5888 } else
5889 *valp = 1;
5890 rw_exit(&sip->i_contents);
5891 rw_exit(&sip->i_rwlock);
5892 if (ulp) {
5893 TRANS_END_CSYNC(ufsvfsp, error, issync,
5894 TOP_RMDIR, TOP_RMDIR_SIZE);
5895 }
5896 VN_RELE(ITOV(sip));
5897 } else if (error == ENOENT) {
5898 *valp = 0;
5899 error = 0;
5900 }
5901 } else {
5902 error = fs_pathconf(vp, cmd, valp, cr, ct);
5903 }
5904 break;
5905
5906 case _PC_ACL_ENABLED:
5907 *valp = _ACL_ACLENT_ENABLED;
5908 break;
5909
5910 case _PC_MIN_HOLE_SIZE:
5911 *valp = (ulong_t)ip->i_fs->fs_bsize;
5912 break;
5913
5914 case _PC_SATTR_ENABLED:
5915 case _PC_SATTR_EXISTS:
5916 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) &&
5917 (vp->v_type == VREG || vp->v_type == VDIR);
5918 break;
5919
5920 case _PC_TIMESTAMP_RESOLUTION:
5921 /*
5922 * UFS keeps only microsecond timestamp resolution.
5923 * This is historical and will probably never change.
5924 */
5925 *valp = 1000L;
5926 break;
5927
5928 default:
5929 error = fs_pathconf(vp, cmd, valp, cr, ct);
5930 break;
5931 }
5932
5933 if (ulp != NULL) {
5934 ufs_lockfs_end(ulp);
5935 }
5936 return (error);
5937 }
5938
5939 int ufs_pageio_writes, ufs_pageio_reads;
5940
5941 /*ARGSUSED*/
5942 static int
5943 ufs_pageio(struct vnode *vp, page_t *pp, u_offset_t io_off, size_t io_len,
5944 int flags, struct cred *cr, caller_context_t *ct)
5945 {
5946 struct inode *ip = VTOI(vp);
5947 struct ufsvfs *ufsvfsp;
5948 page_t *npp = NULL, *opp = NULL, *cpp = pp;
5949 struct buf *bp;
5950 daddr_t bn;
5951 size_t done_len = 0, cur_len = 0;
5952 int err = 0;
5953 int contig = 0;
5954 int dolock;
5955 int vmpss = 0;
5956 struct ulockfs *ulp;
5957
5958 if ((flags & B_READ) && pp != NULL && pp->p_vnode == vp &&
5959 vp->v_mpssdata != NULL) {
5960 vmpss = 1;
5961 }
5962
5963 dolock = (rw_owner(&ip->i_contents) != curthread);
5964 /*
5965 * We need a better check. Ideally, we would use another
5966 * vnodeops so that hlocked and forcibly unmounted file
5967 * systems would return EIO where appropriate and w/o the
5968 * need for these checks.
5969 */
5970 if ((ufsvfsp = ip->i_ufsvfs) == NULL)
5971 return (EIO);
5972
5973 /*
5974 * For vmpss (pp can be NULL) case respect the quiesce protocol.
5975 * ul_lock must be taken before locking pages so we can't use it here
5976 * if pp is non NULL because segvn already locked pages
5977 * SE_EXCL. Instead we rely on the fact that a forced umount or
5978 * applying a filesystem lock via ufs_fiolfs() will block in the
5979 * implicit call to ufs_flush() until we unlock the pages after the
5980 * return to segvn. Other ufs_quiesce() callers keep ufs_quiesce_pend
5981 * above 0 until they are done. We have to be careful not to increment
5982 * ul_vnops_cnt here after forceful unmount hlocks the file system.
5983 *
5984 * If pp is NULL use ul_lock to make sure we don't increment
5985 * ul_vnops_cnt after forceful unmount hlocks the file system.
5986 */
5987 if (vmpss || pp == NULL) {
5988 ulp = &ufsvfsp->vfs_ulockfs;
5989 if (pp == NULL)
5990 mutex_enter(&ulp->ul_lock);
5991 if (ulp->ul_fs_lock & ULOCKFS_GETREAD_MASK) {
5992 if (pp == NULL) {
5993 mutex_exit(&ulp->ul_lock);
5994 }
5995 return (vmpss ? EIO : EINVAL);
5996 }
5997 atomic_inc_ulong(&ulp->ul_vnops_cnt);
5998 if (pp == NULL)
5999 mutex_exit(&ulp->ul_lock);
6000 if (ufs_quiesce_pend) {
6001 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6002 cv_broadcast(&ulp->ul_cv);
6003 return (vmpss ? EIO : EINVAL);
6004 }
6005 }
6006
6007 if (dolock) {
6008 /*
6009 * segvn may call VOP_PAGEIO() instead of VOP_GETPAGE() to
6010 * handle a fault against a segment that maps vnode pages with
6011 * large mappings. Segvn creates pages and holds them locked
6012 * SE_EXCL during VOP_PAGEIO() call. In this case we have to
6013 * use rw_tryenter() to avoid a potential deadlock since in
6014 * lock order i_contents needs to be taken first.
6015 * Segvn will retry via VOP_GETPAGE() if VOP_PAGEIO() fails.
6016 */
6017 if (!vmpss) {
6018 rw_enter(&ip->i_contents, RW_READER);
6019 } else if (!rw_tryenter(&ip->i_contents, RW_READER)) {
6020 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6021 cv_broadcast(&ulp->ul_cv);
6022 return (EDEADLK);
6023 }
6024 }
6025
6026 /*
6027 * Return an error to segvn because the pagefault request is beyond
6028 * PAGESIZE rounded EOF.
6029 */
6030 if (vmpss && btopr(io_off + io_len) > btopr(ip->i_size)) {
6031 if (dolock)
6032 rw_exit(&ip->i_contents);
6033 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6034 cv_broadcast(&ulp->ul_cv);
6035 return (EFAULT);
6036 }
6037
6038 if (pp == NULL) {
6039 if (bmap_has_holes(ip)) {
6040 err = ENOSYS;
6041 } else {
6042 err = EINVAL;
6043 }
6044 if (dolock)
6045 rw_exit(&ip->i_contents);
6046 if (!atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6047 cv_broadcast(&ulp->ul_cv);
6048 return (err);
6049 }
6050
6051 /*
6052 * Break the io request into chunks, one for each contiguous
6053 * stretch of disk blocks in the target file.
6054 */
6055 while (done_len < io_len) {
6056 ASSERT(cpp);
6057 contig = 0;
6058 if (err = bmap_read(ip, (u_offset_t)(io_off + done_len),
6059 &bn, &contig))
6060 break;
6061
6062 if (bn == UFS_HOLE) { /* No holey swapfiles */
6063 if (vmpss) {
6064 err = EFAULT;
6065 break;
6066 }
6067 err = ufs_fault(ITOV(ip), "ufs_pageio: bn == UFS_HOLE");
6068 break;
6069 }
6070
6071 cur_len = MIN(io_len - done_len, contig);
6072 /*
6073 * Zero out a page beyond EOF, when the last block of
6074 * a file is a UFS fragment so that ufs_pageio() can be used
6075 * instead of ufs_getpage() to handle faults against
6076 * segvn segments that use large pages.
6077 */
6078 page_list_break(&cpp, &npp, btopr(cur_len));
6079 if ((flags & B_READ) && (cur_len & PAGEOFFSET)) {
6080 size_t xlen = cur_len & PAGEOFFSET;
6081 pagezero(cpp->p_prev, xlen, PAGESIZE - xlen);
6082 }
6083
6084 bp = pageio_setup(cpp, cur_len, ip->i_devvp, flags);
6085 ASSERT(bp != NULL);
6086
6087 bp->b_edev = ip->i_dev;
6088 bp->b_dev = cmpdev(ip->i_dev);
6089 bp->b_blkno = bn;
6090 bp->b_un.b_addr = (caddr_t)0;
6091 bp->b_file = ip->i_vnode;
6092
6093 ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
6094 ub.ub_pageios.value.ul++;
6095 if (ufsvfsp->vfs_snapshot)
6096 fssnap_strategy(&(ufsvfsp->vfs_snapshot), bp);
6097 else
6098 (void) bdev_strategy(bp);
6099
6100 if (flags & B_READ)
6101 ufs_pageio_reads++;
6102 else
6103 ufs_pageio_writes++;
6104 if (flags & B_READ)
6105 lwp_stat_update(LWP_STAT_INBLK, 1);
6106 else
6107 lwp_stat_update(LWP_STAT_OUBLK, 1);
6108 /*
6109 * If the request is not B_ASYNC, wait for i/o to complete
6110 * and re-assemble the page list to return to the caller.
6111 * If it is B_ASYNC we leave the page list in pieces and
6112 * cleanup() will dispose of them.
6113 */
6114 if ((flags & B_ASYNC) == 0) {
6115 err = biowait(bp);
6116 pageio_done(bp);
6117 if (err)
6118 break;
6119 page_list_concat(&opp, &cpp);
6120 }
6121 cpp = npp;
6122 npp = NULL;
6123 if (flags & B_READ)
6124 cur_len = P2ROUNDUP_TYPED(cur_len, PAGESIZE, size_t);
6125 done_len += cur_len;
6126 }
6127 ASSERT(err || (cpp == NULL && npp == NULL && done_len == io_len));
6128 if (err) {
6129 if (flags & B_ASYNC) {
6130 /* Cleanup unprocessed parts of list */
6131 page_list_concat(&cpp, &npp);
6132 if (flags & B_READ)
6133 pvn_read_done(cpp, B_ERROR);
6134 else
6135 pvn_write_done(cpp, B_ERROR);
6136 } else {
6137 /* Re-assemble list and let caller clean up */
6138 page_list_concat(&opp, &cpp);
6139 page_list_concat(&opp, &npp);
6140 }
6141 }
6142
6143 if (vmpss && !(ip->i_flag & IACC) && !ULOCKFS_IS_NOIACC(ulp) &&
6144 ufsvfsp->vfs_fs->fs_ronly == 0 && !ufsvfsp->vfs_noatime) {
6145 mutex_enter(&ip->i_tlock);
6146 ip->i_flag |= IACC;
6147 ITIMES_NOLOCK(ip);
6148 mutex_exit(&ip->i_tlock);
6149 }
6150
6151 if (dolock)
6152 rw_exit(&ip->i_contents);
6153 if (vmpss && !atomic_dec_ulong_nv(&ulp->ul_vnops_cnt))
6154 cv_broadcast(&ulp->ul_cv);
6155 return (err);
6156 }
6157
6158 /*
6159 * Called when the kernel is in a frozen state to dump data
6160 * directly to the device. It uses a private dump data structure,
6161 * set up by dump_ctl, to locate the correct disk block to which to dump.
6162 */
6163 /*ARGSUSED*/
6164 static int
6165 ufs_dump(vnode_t *vp, caddr_t addr, offset_t ldbn, offset_t dblks,
6166 caller_context_t *ct)
6167 {
6168 u_offset_t file_size;
6169 struct inode *ip = VTOI(vp);
6170 struct fs *fs = ip->i_fs;
6171 daddr_t dbn, lfsbn;
6172 int disk_blks = fs->fs_bsize >> DEV_BSHIFT;
6173 int error = 0;
6174 int ndbs, nfsbs;
6175
6176 /*
6177 * forced unmount case
6178 */
6179 if (ip->i_ufsvfs == NULL)
6180 return (EIO);
6181 /*
6182 * Validate the inode that it has not been modified since
6183 * the dump structure is allocated.
6184 */
6185 mutex_enter(&ip->i_tlock);
6186 if ((dump_info == NULL) ||
6187 (dump_info->ip != ip) ||
6188 (dump_info->time.tv_sec != ip->i_mtime.tv_sec) ||
6189 (dump_info->time.tv_usec != ip->i_mtime.tv_usec)) {
6190 mutex_exit(&ip->i_tlock);
6191 return (-1);
6192 }
6193 mutex_exit(&ip->i_tlock);
6194
6195 /*
6196 * See that the file has room for this write
6197 */
6198 UFS_GET_ISIZE(&file_size, ip);
6199
6200 if (ldbtob(ldbn + dblks) > file_size)
6201 return (ENOSPC);
6202
6203 /*
6204 * Find the physical disk block numbers from the dump
6205 * private data structure directly and write out the data
6206 * in contiguous block lumps
6207 */
6208 while (dblks > 0 && !error) {
6209 lfsbn = (daddr_t)lblkno(fs, ldbtob(ldbn));
6210 dbn = fsbtodb(fs, dump_info->dblk[lfsbn]) + ldbn % disk_blks;
6211 nfsbs = 1;
6212 ndbs = disk_blks - ldbn % disk_blks;
6213 while (ndbs < dblks && fsbtodb(fs, dump_info->dblk[lfsbn +
6214 nfsbs]) == dbn + ndbs) {
6215 nfsbs++;
6216 ndbs += disk_blks;
6217 }
6218 if (ndbs > dblks)
6219 ndbs = dblks;
6220 error = bdev_dump(ip->i_dev, addr, dbn, ndbs);
6221 addr += ldbtob((offset_t)ndbs);
6222 dblks -= ndbs;
6223 ldbn += ndbs;
6224 }
6225 return (error);
6226
6227 }
6228
6229 /*
6230 * Prepare the file system before and after the dump operation.
6231 *
6232 * action = DUMP_ALLOC:
6233 * Preparation before dump, allocate dump private data structure
6234 * to hold all the direct and indirect block info for dump.
6235 *
6236 * action = DUMP_FREE:
6237 * Clean up after dump, deallocate the dump private data structure.
6238 *
6239 * action = DUMP_SCAN:
6240 * Scan dump_info for *blkp DEV_BSIZE blocks of contig fs space;
6241 * if found, the starting file-relative DEV_BSIZE lbn is written
6242 * to *bklp; that lbn is intended for use with VOP_DUMP()
6243 */
6244 /*ARGSUSED*/
6245 static int
6246 ufs_dumpctl(vnode_t *vp, int action, offset_t *blkp, caller_context_t *ct)
6247 {
6248 struct inode *ip = VTOI(vp);
6249 ufsvfs_t *ufsvfsp = ip->i_ufsvfs;
6250 struct fs *fs;
6251 daddr32_t *dblk, *storeblk;
6252 daddr32_t *nextblk, *endblk;
6253 struct buf *bp;
6254 int i, entry, entries;
6255 int n, ncontig;
6256
6257 /*
6258 * check for forced unmount
6259 */
6260 if (ufsvfsp == NULL)
6261 return (EIO);
6262
6263 if (action == DUMP_ALLOC) {
6264 /*
6265 * alloc and record dump_info
6266 */
6267 if (dump_info != NULL)
6268 return (EINVAL);
6269
6270 ASSERT(vp->v_type == VREG);
6271 fs = ufsvfsp->vfs_fs;
6272
6273 rw_enter(&ip->i_contents, RW_READER);
6274
6275 if (bmap_has_holes(ip)) {
6276 rw_exit(&ip->i_contents);
6277 return (EFAULT);
6278 }
6279
6280 /*
6281 * calculate and allocate space needed according to i_size
6282 */
6283 entries = (int)lblkno(fs, blkroundup(fs, ip->i_size));
6284 dump_info = kmem_alloc(sizeof (struct dump) +
6285 (entries - 1) * sizeof (daddr32_t), KM_NOSLEEP);
6286 if (dump_info == NULL) {
6287 rw_exit(&ip->i_contents);
6288 return (ENOMEM);
6289 }
6290
6291 /* Start saving the info */
6292 dump_info->fsbs = entries;
6293 dump_info->ip = ip;
6294 storeblk = &dump_info->dblk[0];
6295
6296 /* Direct Blocks */
6297 for (entry = 0; entry < NDADDR && entry < entries; entry++)
6298 *storeblk++ = ip->i_db[entry];
6299
6300 /* Indirect Blocks */
6301 for (i = 0; i < NIADDR; i++) {
6302 int error = 0;
6303
6304 bp = UFS_BREAD(ufsvfsp,
6305 ip->i_dev, fsbtodb(fs, ip->i_ib[i]), fs->fs_bsize);
6306 if (bp->b_flags & B_ERROR)
6307 error = EIO;
6308 else {
6309 dblk = bp->b_un.b_daddr;
6310 if ((storeblk = save_dblks(ip, ufsvfsp,
6311 storeblk, dblk, i, entries)) == NULL)
6312 error = EIO;
6313 }
6314
6315 brelse(bp);
6316
6317 if (error != 0) {
6318 kmem_free(dump_info, sizeof (struct dump) +
6319 (entries - 1) * sizeof (daddr32_t));
6320 rw_exit(&ip->i_contents);
6321 dump_info = NULL;
6322 return (error);
6323 }
6324 }
6325 /* and time stamp the information */
6326 mutex_enter(&ip->i_tlock);
6327 dump_info->time = ip->i_mtime;
6328 mutex_exit(&ip->i_tlock);
6329
6330 rw_exit(&ip->i_contents);
6331 } else if (action == DUMP_FREE) {
6332 /*
6333 * free dump_info
6334 */
6335 if (dump_info == NULL)
6336 return (EINVAL);
6337 entries = dump_info->fsbs - 1;
6338 kmem_free(dump_info, sizeof (struct dump) +
6339 entries * sizeof (daddr32_t));
6340 dump_info = NULL;
6341 } else if (action == DUMP_SCAN) {
6342 /*
6343 * scan dump_info
6344 */
6345 if (dump_info == NULL)
6346 return (EINVAL);
6347
6348 dblk = dump_info->dblk;
6349 nextblk = dblk + 1;
6350 endblk = dblk + dump_info->fsbs - 1;
6351 fs = ufsvfsp->vfs_fs;
6352 ncontig = *blkp >> (fs->fs_bshift - DEV_BSHIFT);
6353
6354 /*
6355 * scan dblk[] entries; contig fs space is found when:
6356 * ((current blkno + frags per block) == next blkno)
6357 */
6358 n = 0;
6359 while (n < ncontig && dblk < endblk) {
6360 if ((*dblk + fs->fs_frag) == *nextblk)
6361 n++;
6362 else
6363 n = 0;
6364 dblk++;
6365 nextblk++;
6366 }
6367
6368 /*
6369 * index is where size bytes of contig space begins;
6370 * conversion from index to the file's DEV_BSIZE lbn
6371 * is equivalent to: (index * fs_bsize) / DEV_BSIZE
6372 */
6373 if (n == ncontig) {
6374 i = (dblk - dump_info->dblk) - ncontig;
6375 *blkp = i << (fs->fs_bshift - DEV_BSHIFT);
6376 } else
6377 return (EFAULT);
6378 }
6379 return (0);
6380 }
6381
6382 /*
6383 * Recursive helper function for ufs_dumpctl(). It follows the indirect file
6384 * system blocks until it reaches the the disk block addresses, which are
6385 * then stored into the given buffer, storeblk.
6386 */
6387 static daddr32_t *
6388 save_dblks(struct inode *ip, struct ufsvfs *ufsvfsp, daddr32_t *storeblk,
6389 daddr32_t *dblk, int level, int entries)
6390 {
6391 struct fs *fs = ufsvfsp->vfs_fs;
6392 struct buf *bp;
6393 int i;
6394
6395 if (level == 0) {
6396 for (i = 0; i < NINDIR(fs); i++) {
6397 if (storeblk - dump_info->dblk >= entries)
6398 break;
6399 *storeblk++ = dblk[i];
6400 }
6401 return (storeblk);
6402 }
6403 for (i = 0; i < NINDIR(fs); i++) {
6404 if (storeblk - dump_info->dblk >= entries)
6405 break;
6406 bp = UFS_BREAD(ufsvfsp,
6407 ip->i_dev, fsbtodb(fs, dblk[i]), fs->fs_bsize);
6408 if (bp->b_flags & B_ERROR) {
6409 brelse(bp);
6410 return (NULL);
6411 }
6412 storeblk = save_dblks(ip, ufsvfsp, storeblk, bp->b_un.b_daddr,
6413 level - 1, entries);
6414 brelse(bp);
6415
6416 if (storeblk == NULL)
6417 return (NULL);
6418 }
6419 return (storeblk);
6420 }
6421
6422 /* ARGSUSED */
6423 static int
6424 ufs_getsecattr(struct vnode *vp, vsecattr_t *vsap, int flag,
6425 struct cred *cr, caller_context_t *ct)
6426 {
6427 struct inode *ip = VTOI(vp);
6428 struct ulockfs *ulp;
6429 struct ufsvfs *ufsvfsp = ip->i_ufsvfs;
6430 ulong_t vsa_mask = vsap->vsa_mask;
6431 int err = EINVAL;
6432
6433 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6434
6435 /*
6436 * Only grab locks if needed - they're not needed to check vsa_mask
6437 * or if the mask contains no acl flags.
6438 */
6439 if (vsa_mask != 0) {
6440 if (err = ufs_lockfs_begin(ufsvfsp, &ulp,
6441 ULOCKFS_GETATTR_MASK))
6442 return (err);
6443
6444 rw_enter(&ip->i_contents, RW_READER);
6445 err = ufs_acl_get(ip, vsap, flag, cr);
6446 rw_exit(&ip->i_contents);
6447
6448 if (ulp)
6449 ufs_lockfs_end(ulp);
6450 }
6451 return (err);
6452 }
6453
6454 /* ARGSUSED */
6455 static int
6456 ufs_setsecattr(struct vnode *vp, vsecattr_t *vsap, int flag, struct cred *cr,
6457 caller_context_t *ct)
6458 {
6459 struct inode *ip = VTOI(vp);
6460 struct ulockfs *ulp = NULL;
6461 struct ufsvfs *ufsvfsp = VTOI(vp)->i_ufsvfs;
6462 ulong_t vsa_mask = vsap->vsa_mask;
6463 int err;
6464 int haverwlock = 1;
6465 int trans_size;
6466 int donetrans = 0;
6467 int retry = 1;
6468
6469 ASSERT(RW_LOCK_HELD(&ip->i_rwlock));
6470
6471 /* Abort now if the request is either empty or invalid. */
6472 vsa_mask &= (VSA_ACL | VSA_ACLCNT | VSA_DFACL | VSA_DFACLCNT);
6473 if ((vsa_mask == 0) ||
6474 ((vsap->vsa_aclentp == NULL) &&
6475 (vsap->vsa_dfaclentp == NULL))) {
6476 err = EINVAL;
6477 goto out;
6478 }
6479
6480 /*
6481 * Following convention, if this is a directory then we acquire the
6482 * inode's i_rwlock after starting a UFS logging transaction;
6483 * otherwise, we acquire it beforehand. Since we were called (and
6484 * must therefore return) with the lock held, we will have to drop it,
6485 * and later reacquire it, if operating on a directory.
6486 */
6487 if (vp->v_type == VDIR) {
6488 rw_exit(&ip->i_rwlock);
6489 haverwlock = 0;
6490 } else {
6491 /* Upgrade the lock if required. */
6492 if (!rw_write_held(&ip->i_rwlock)) {
6493 rw_exit(&ip->i_rwlock);
6494 rw_enter(&ip->i_rwlock, RW_WRITER);
6495 }
6496 }
6497
6498 again:
6499 ASSERT(!(vp->v_type == VDIR && haverwlock));
6500 if (err = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_SETATTR_MASK)) {
6501 ulp = NULL;
6502 retry = 0;
6503 goto out;
6504 }
6505
6506 /*
6507 * Check that the file system supports this operation. Note that
6508 * ufs_lockfs_begin() will have checked that the file system had
6509 * not been forcibly unmounted.
6510 */
6511 if (ufsvfsp->vfs_fs->fs_ronly) {
6512 err = EROFS;
6513 goto out;
6514 }
6515 if (ufsvfsp->vfs_nosetsec) {
6516 err = ENOSYS;
6517 goto out;
6518 }
6519
6520 if (ulp) {
6521 TRANS_BEGIN_ASYNC(ufsvfsp, TOP_SETSECATTR,
6522 trans_size = TOP_SETSECATTR_SIZE(VTOI(vp)));
6523 donetrans = 1;
6524 }
6525
6526 if (vp->v_type == VDIR) {
6527 rw_enter(&ip->i_rwlock, RW_WRITER);
6528 haverwlock = 1;
6529 }
6530
6531 ASSERT(haverwlock);
6532
6533 /* Do the actual work. */
6534 rw_enter(&ip->i_contents, RW_WRITER);
6535 /*
6536 * Suppress out of inodes messages if we will retry.
6537 */
6538 if (retry)
6539 ip->i_flag |= IQUIET;
6540 err = ufs_acl_set(ip, vsap, flag, cr);
6541 ip->i_flag &= ~IQUIET;
6542 rw_exit(&ip->i_contents);
6543
6544 out:
6545 if (ulp) {
6546 if (donetrans) {
6547 /*
6548 * top_end_async() can eventually call
6549 * top_end_sync(), which can block. We must
6550 * therefore observe the lock-ordering protocol
6551 * here as well.
6552 */
6553 if (vp->v_type == VDIR) {
6554 rw_exit(&ip->i_rwlock);
6555 haverwlock = 0;
6556 }
6557 TRANS_END_ASYNC(ufsvfsp, TOP_SETSECATTR, trans_size);
6558 }
6559 ufs_lockfs_end(ulp);
6560 }
6561 /*
6562 * If no inodes available, try scaring a logically-
6563 * free one out of the delete queue to someplace
6564 * that we can find it.
6565 */
6566 if ((err == ENOSPC) && retry && TRANS_ISTRANS(ufsvfsp)) {
6567 ufs_delete_drain_wait(ufsvfsp, 1);
6568 retry = 0;
6569 if (vp->v_type == VDIR && haverwlock) {
6570 rw_exit(&ip->i_rwlock);
6571 haverwlock = 0;
6572 }
6573 goto again;
6574 }
6575 /*
6576 * If we need to reacquire the lock then it is safe to do so
6577 * as a reader. This is because ufs_rwunlock(), which will be
6578 * called by our caller after we return, does not differentiate
6579 * between shared and exclusive locks.
6580 */
6581 if (!haverwlock) {
6582 ASSERT(vp->v_type == VDIR);
6583 rw_enter(&ip->i_rwlock, RW_READER);
6584 }
6585
6586 return (err);
6587 }
6588
6589 /*
6590 * Locate the vnode to be used for an event notification. As this will
6591 * be called prior to the name space change perform basic verification
6592 * that the change will be allowed.
6593 */
6594
6595 static int
6596 ufs_eventlookup(struct vnode *dvp, char *nm, struct cred *cr,
6597 struct vnode **vpp)
6598 {
6599 int namlen;
6600 int error;
6601 struct vnode *vp;
6602 struct inode *ip;
6603 struct inode *xip;
6604 struct ufsvfs *ufsvfsp;
6605 struct ulockfs *ulp;
6606
6607 ip = VTOI(dvp);
6608 *vpp = NULL;
6609
6610 if ((namlen = strlen(nm)) == 0)
6611 return (EINVAL);
6612
6613 if (nm[0] == '.') {
6614 if (namlen == 1)
6615 return (EINVAL);
6616 else if ((namlen == 2) && nm[1] == '.') {
6617 return (EEXIST);
6618 }
6619 }
6620
6621 /*
6622 * Check accessibility and write access of parent directory as we
6623 * only want to post the event if we're able to make a change.
6624 */
6625 if (error = ufs_diraccess(ip, IEXEC|IWRITE, cr))
6626 return (error);
6627
6628 if (vp = dnlc_lookup(dvp, nm)) {
6629 if (vp == DNLC_NO_VNODE) {
6630 VN_RELE(vp);
6631 return (ENOENT);
6632 }
6633
6634 *vpp = vp;
6635 return (0);
6636 }
6637
6638 /*
6639 * Keep the idle queue from getting too long by idling two
6640 * inodes before attempting to allocate another.
6641 * This operation must be performed before entering lockfs
6642 * or a transaction.
6643 */
6644 if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
6645 if ((curthread->t_flag & T_DONTBLOCK) == 0) {
6646 ins.in_lidles.value.ul += ufs_lookup_idle_count;
6647 ufs_idle_some(ufs_lookup_idle_count);
6648 }
6649
6650 ufsvfsp = ip->i_ufsvfs;
6651
6652 retry_lookup:
6653 if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_LOOKUP_MASK))
6654 return (error);
6655
6656 if ((error = ufs_dirlook(ip, nm, &xip, cr, 1, 1)) == 0) {
6657 vp = ITOV(xip);
6658 *vpp = vp;
6659 }
6660
6661 if (ulp) {
6662 ufs_lockfs_end(ulp);
6663 }
6664
6665 if (error == EAGAIN)
6666 goto retry_lookup;
6667
6668 return (error);
6669 }