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 2009 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
25 */
26
27 /*
28 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
30 */
31
32 #include <sys/param.h>
33 #include <sys/t_lock.h>
34 #include <sys/systm.h>
35 #include <sys/sysmacros.h>
36 #include <sys/user.h>
37 #include <sys/buf.h>
38 #include <sys/stat.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/dirent.h>
42 #include <sys/vnode.h>
43 #include <sys/proc.h>
44 #include <sys/file.h>
45 #include <sys/fcntl.h>
46 #include <sys/uio.h>
47 #include <sys/fs/pc_label.h>
48 #include <sys/fs/pc_fs.h>
49 #include <sys/fs/pc_dir.h>
50 #include <sys/fs/pc_node.h>
51 #include <sys/mman.h>
52 #include <sys/pathname.h>
53 #include <sys/vmsystm.h>
54 #include <sys/cmn_err.h>
55 #include <sys/debug.h>
56 #include <sys/statvfs.h>
57 #include <sys/unistd.h>
58 #include <sys/kmem.h>
59 #include <sys/conf.h>
60 #include <sys/flock.h>
61 #include <sys/policy.h>
62 #include <sys/sdt.h>
63 #include <sys/sunddi.h>
64 #include <sys/types.h>
65 #include <sys/errno.h>
66
67 #include <vm/seg.h>
68 #include <vm/page.h>
69 #include <vm/pvn.h>
70 #include <vm/seg_map.h>
71 #include <vm/seg_vn.h>
72 #include <vm/hat.h>
73 #include <vm/as.h>
74 #include <vm/seg_kmem.h>
75
76 #include <fs/fs_subr.h>
77
78 static int pcfs_open(struct vnode **, int, struct cred *, caller_context_t *ct);
79 static int pcfs_close(struct vnode *, int, int, offset_t, struct cred *,
80 caller_context_t *ct);
81 static int pcfs_read(struct vnode *, struct uio *, int, struct cred *,
82 caller_context_t *);
83 static int pcfs_write(struct vnode *, struct uio *, int, struct cred *,
84 caller_context_t *);
85 static int pcfs_getattr(struct vnode *, struct vattr *, int, struct cred *,
86 caller_context_t *ct);
87 static int pcfs_setattr(struct vnode *, struct vattr *, int, struct cred *,
88 caller_context_t *);
89 static int pcfs_access(struct vnode *, int, int, struct cred *,
90 caller_context_t *ct);
91 static int pcfs_lookup(struct vnode *, char *, struct vnode **,
92 struct pathname *, int, struct vnode *, struct cred *,
93 caller_context_t *, int *, pathname_t *);
94 static int pcfs_create(struct vnode *, char *, struct vattr *,
95 enum vcexcl, int mode, struct vnode **, struct cred *, int,
96 caller_context_t *, vsecattr_t *);
97 static int pcfs_remove(struct vnode *, char *, struct cred *,
98 caller_context_t *, int);
99 static int pcfs_rename(struct vnode *, char *, struct vnode *, char *,
100 struct cred *, caller_context_t *, int);
101 static int pcfs_mkdir(struct vnode *, char *, struct vattr *, struct vnode **,
102 struct cred *, caller_context_t *, int, vsecattr_t *);
103 static int pcfs_rmdir(struct vnode *, char *, struct vnode *, struct cred *,
104 caller_context_t *, int);
105 static int pcfs_readdir(struct vnode *, struct uio *, struct cred *, int *,
106 caller_context_t *, int);
107 static int pcfs_fsync(struct vnode *, int, struct cred *, caller_context_t *);
108 static void pcfs_inactive(struct vnode *, struct cred *, caller_context_t *);
109 static int pcfs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *);
110 static int pcfs_space(struct vnode *, int, struct flock64 *, int,
111 offset_t, cred_t *, caller_context_t *);
112 static int pcfs_getpage(struct vnode *, offset_t, size_t, uint_t *, page_t *[],
113 size_t, struct seg *, caddr_t, enum seg_rw, struct cred *,
114 caller_context_t *);
115 static int pcfs_getapage(struct vnode *, u_offset_t, size_t, uint_t *,
116 page_t *[], size_t, struct seg *, caddr_t, enum seg_rw, struct cred *);
117 static int pcfs_putpage(struct vnode *, offset_t, size_t, int, struct cred *,
118 caller_context_t *);
119 static int pcfs_map(struct vnode *, offset_t, struct as *, caddr_t *, size_t,
120 uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
121 static int pcfs_addmap(struct vnode *, offset_t, struct as *, caddr_t,
122 size_t, uchar_t, uchar_t, uint_t, struct cred *, caller_context_t *);
123 static int pcfs_delmap(struct vnode *, offset_t, struct as *, caddr_t,
124 size_t, uint_t, uint_t, uint_t, struct cred *, caller_context_t *);
125 static int pcfs_seek(struct vnode *, offset_t, offset_t *,
126 caller_context_t *);
127 static int pcfs_pathconf(struct vnode *, int, ulong_t *, struct cred *,
128 caller_context_t *);
129
130 int pcfs_putapage(struct vnode *, page_t *, u_offset_t *, size_t *, int,
131 struct cred *);
132 static int rwpcp(struct pcnode *, struct uio *, enum uio_rw, int);
133 static int get_long_fn_chunk(struct pcdir_lfn *ep, char *buf);
134
135 extern krwlock_t pcnodes_lock;
136
137 #define lround(r) (((r)+sizeof (long long)-1)&(~(sizeof (long long)-1)))
138
139 /*
140 * vnode op vectors for files and directories.
141 */
142 struct vnodeops *pcfs_fvnodeops;
143 struct vnodeops *pcfs_dvnodeops;
144
145 const fs_operation_def_t pcfs_fvnodeops_template[] = {
146 VOPNAME_OPEN, { .vop_open = pcfs_open },
147 VOPNAME_CLOSE, { .vop_close = pcfs_close },
148 VOPNAME_READ, { .vop_read = pcfs_read },
149 VOPNAME_WRITE, { .vop_write = pcfs_write },
150 VOPNAME_GETATTR, { .vop_getattr = pcfs_getattr },
151 VOPNAME_SETATTR, { .vop_setattr = pcfs_setattr },
152 VOPNAME_ACCESS, { .vop_access = pcfs_access },
153 VOPNAME_FSYNC, { .vop_fsync = pcfs_fsync },
154 VOPNAME_INACTIVE, { .vop_inactive = pcfs_inactive },
155 VOPNAME_FID, { .vop_fid = pcfs_fid },
156 VOPNAME_SEEK, { .vop_seek = pcfs_seek },
157 VOPNAME_SPACE, { .vop_space = pcfs_space },
158 VOPNAME_GETPAGE, { .vop_getpage = pcfs_getpage },
159 VOPNAME_PUTPAGE, { .vop_putpage = pcfs_putpage },
160 VOPNAME_MAP, { .vop_map = pcfs_map },
161 VOPNAME_ADDMAP, { .vop_addmap = pcfs_addmap },
162 VOPNAME_DELMAP, { .vop_delmap = pcfs_delmap },
163 VOPNAME_PATHCONF, { .vop_pathconf = pcfs_pathconf },
164 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
165 NULL, NULL
166 };
167
168 const fs_operation_def_t pcfs_dvnodeops_template[] = {
169 VOPNAME_OPEN, { .vop_open = pcfs_open },
170 VOPNAME_CLOSE, { .vop_close = pcfs_close },
171 VOPNAME_GETATTR, { .vop_getattr = pcfs_getattr },
172 VOPNAME_SETATTR, { .vop_setattr = pcfs_setattr },
173 VOPNAME_ACCESS, { .vop_access = pcfs_access },
174 VOPNAME_LOOKUP, { .vop_lookup = pcfs_lookup },
175 VOPNAME_CREATE, { .vop_create = pcfs_create },
176 VOPNAME_REMOVE, { .vop_remove = pcfs_remove },
177 VOPNAME_RENAME, { .vop_rename = pcfs_rename },
178 VOPNAME_MKDIR, { .vop_mkdir = pcfs_mkdir },
179 VOPNAME_RMDIR, { .vop_rmdir = pcfs_rmdir },
180 VOPNAME_READDIR, { .vop_readdir = pcfs_readdir },
181 VOPNAME_FSYNC, { .vop_fsync = pcfs_fsync },
182 VOPNAME_INACTIVE, { .vop_inactive = pcfs_inactive },
183 VOPNAME_FID, { .vop_fid = pcfs_fid },
184 VOPNAME_SEEK, { .vop_seek = pcfs_seek },
185 VOPNAME_PATHCONF, { .vop_pathconf = pcfs_pathconf },
186 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support },
187 NULL, NULL
188 };
189
190
191 /*ARGSUSED*/
192 static int
193 pcfs_open(
194 struct vnode **vpp,
195 int flag,
196 struct cred *cr,
197 caller_context_t *ct)
198 {
199 return (0);
200 }
201
202 /*
203 * files are sync'ed on close to keep floppy up to date
204 */
205
206 /*ARGSUSED*/
207 static int
208 pcfs_close(
209 struct vnode *vp,
210 int flag,
211 int count,
212 offset_t offset,
213 struct cred *cr,
214 caller_context_t *ct)
215 {
216 return (0);
217 }
218
219 /*ARGSUSED*/
220 static int
221 pcfs_read(
222 struct vnode *vp,
223 struct uio *uiop,
224 int ioflag,
225 struct cred *cr,
226 struct caller_context *ct)
227 {
228 struct pcfs *fsp;
229 struct pcnode *pcp;
230 int error;
231
232 fsp = VFSTOPCFS(vp->v_vfsp);
233 if (error = pc_verify(fsp))
234 return (error);
235 error = pc_lockfs(fsp, 0, 0);
236 if (error)
237 return (error);
238 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
239 pc_unlockfs(fsp);
240 return (EIO);
241 }
242 error = rwpcp(pcp, uiop, UIO_READ, ioflag);
243 if ((fsp->pcfs_vfs->vfs_flag & VFS_RDONLY) == 0) {
244 pc_mark_acc(fsp, pcp);
245 }
246 pc_unlockfs(fsp);
247 if (error) {
248 PC_DPRINTF1(1, "pcfs_read: io error = %d\n", error);
249 }
250 return (error);
251 }
252
253 /*ARGSUSED*/
254 static int
255 pcfs_write(
256 struct vnode *vp,
257 struct uio *uiop,
258 int ioflag,
259 struct cred *cr,
260 struct caller_context *ct)
261 {
262 struct pcfs *fsp;
263 struct pcnode *pcp;
264 int error;
265
266 fsp = VFSTOPCFS(vp->v_vfsp);
267 if (error = pc_verify(fsp))
268 return (error);
269 error = pc_lockfs(fsp, 0, 0);
270 if (error)
271 return (error);
272 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
273 pc_unlockfs(fsp);
274 return (EIO);
275 }
276 if (ioflag & FAPPEND) {
277 /*
278 * in append mode start at end of file.
279 */
280 uiop->uio_loffset = pcp->pc_size;
281 }
282 error = rwpcp(pcp, uiop, UIO_WRITE, ioflag);
283 pcp->pc_flags |= PC_MOD;
284 pc_mark_mod(fsp, pcp);
285 if (ioflag & (FSYNC|FDSYNC))
286 (void) pc_nodeupdate(pcp);
287
288 pc_unlockfs(fsp);
289 if (error) {
290 PC_DPRINTF1(1, "pcfs_write: io error = %d\n", error);
291 }
292 return (error);
293 }
294
295 /*
296 * read or write a vnode
297 */
298 static int
299 rwpcp(
300 struct pcnode *pcp,
301 struct uio *uio,
302 enum uio_rw rw,
303 int ioflag)
304 {
305 struct vnode *vp = PCTOV(pcp);
306 struct pcfs *fsp;
307 daddr_t bn; /* phys block number */
308 int n;
309 offset_t off;
310 caddr_t base;
311 int mapon, pagecreate;
312 int newpage;
313 int error = 0;
314 rlim64_t limit = uio->uio_llimit;
315 int oresid = uio->uio_resid;
316
317 /*
318 * If the filesystem was umounted by force, return immediately.
319 */
320 if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
321 return (EIO);
322
323 PC_DPRINTF4(5, "rwpcp pcp=%p off=%lld resid=%ld size=%u\n", (void *)pcp,
324 uio->uio_loffset, uio->uio_resid, pcp->pc_size);
325
326 ASSERT(rw == UIO_READ || rw == UIO_WRITE);
327 ASSERT(vp->v_type == VREG);
328
329 if (uio->uio_loffset >= UINT32_MAX && rw == UIO_READ) {
330 return (0);
331 }
332
333 if (uio->uio_loffset < 0)
334 return (EINVAL);
335
336 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T)
337 limit = MAXOFFSET_T;
338
339 if (uio->uio_loffset >= limit && rw == UIO_WRITE) {
340 proc_t *p = ttoproc(curthread);
341
342 mutex_enter(&p->p_lock);
343 (void) rctl_action(rctlproc_legacy[RLIMIT_FSIZE], p->p_rctls,
344 p, RCA_UNSAFE_SIGINFO);
345 mutex_exit(&p->p_lock);
346 return (EFBIG);
347 }
348
349 /* the following condition will occur only for write */
350
351 if (uio->uio_loffset >= UINT32_MAX)
352 return (EFBIG);
353
354 if (uio->uio_resid == 0)
355 return (0);
356
357 if (limit > UINT32_MAX)
358 limit = UINT32_MAX;
359
360 fsp = VFSTOPCFS(vp->v_vfsp);
361 if (fsp->pcfs_flags & PCFS_IRRECOV)
362 return (EIO);
363
364 do {
365 /*
366 * Assignments to "n" in this block may appear
367 * to overflow in some cases. However, after careful
368 * analysis it was determined that all assignments to
369 * "n" serve only to make "n" smaller. Since "n"
370 * starts out as no larger than MAXBSIZE, "int" is
371 * safe.
372 */
373 off = uio->uio_loffset & MAXBMASK;
374 mapon = (int)(uio->uio_loffset & MAXBOFFSET);
375 n = MIN(MAXBSIZE - mapon, uio->uio_resid);
376 if (rw == UIO_READ) {
377 offset_t diff;
378
379 diff = pcp->pc_size - uio->uio_loffset;
380 if (diff <= 0)
381 return (0);
382 if (diff < n)
383 n = (int)diff;
384 }
385 /*
386 * Compare limit with the actual offset + n, not the
387 * rounded down offset "off" or we will overflow
388 * the maximum file size after all.
389 */
390 if (rw == UIO_WRITE && uio->uio_loffset + n >= limit) {
391 if (uio->uio_loffset >= limit) {
392 error = EFBIG;
393 break;
394 }
395 n = (int)(limit - uio->uio_loffset);
396 }
397
398 /*
399 * Touch the page and fault it in if it is not in
400 * core before segmap_getmapflt can lock it. This
401 * is to avoid the deadlock if the buffer is mapped
402 * to the same file through mmap which we want to
403 * write to.
404 */
405 uio_prefaultpages((long)n, uio);
406
407 base = segmap_getmap(segkmap, vp, (u_offset_t)off);
408 pagecreate = 0;
409 newpage = 0;
410 if (rw == UIO_WRITE) {
411 /*
412 * If PAGESIZE < MAXBSIZE, perhaps we ought to deal
413 * with one page at a time, instead of one MAXBSIZE
414 * at a time, so we can fully explore pagecreate
415 * optimization??
416 */
417 if (uio->uio_loffset + n > pcp->pc_size) {
418 uint_t ncl, lcn;
419
420 ncl = (uint_t)howmany((offset_t)pcp->pc_size,
421 fsp->pcfs_clsize);
422 if (uio->uio_loffset > pcp->pc_size &&
423 ncl < (uint_t)howmany(uio->uio_loffset,
424 fsp->pcfs_clsize)) {
425 /*
426 * Allocate and zerofill skipped
427 * clusters. This may not be worth the
428 * effort since a small lseek beyond
429 * eof but still within the cluster
430 * will not be zeroed out.
431 */
432 lcn = pc_lblkno(fsp, uio->uio_loffset);
433 error = pc_balloc(pcp, (daddr_t)lcn,
434 1, &bn);
435 ncl = lcn + 1;
436 }
437 if (!error &&
438 ncl < (uint_t)howmany(uio->uio_loffset + n,
439 fsp->pcfs_clsize))
440 /*
441 * allocate clusters w/o zerofill
442 */
443 error = pc_balloc(pcp,
444 (daddr_t)pc_lblkno(fsp,
445 uio->uio_loffset + n - 1),
446 0, &bn);
447
448 pcp->pc_flags |= PC_CHG;
449
450 if (error) {
451 pc_cluster32_t ncl;
452 int nerror;
453
454 /*
455 * figure out new file size from
456 * cluster chain length. If this
457 * is detected to loop, the chain
458 * is corrupted and we'd better
459 * keep our fingers off that file.
460 */
461 nerror = pc_fileclsize(fsp,
462 pcp->pc_scluster, &ncl);
463 if (nerror) {
464 PC_DPRINTF1(2,
465 "cluster chain "
466 "corruption, "
467 "scluster=%d\n",
468 pcp->pc_scluster);
469 pcp->pc_size = 0;
470 pcp->pc_flags |= PC_INVAL;
471 error = nerror;
472 (void) segmap_release(segkmap,
473 base, 0);
474 break;
475 }
476 pcp->pc_size = fsp->pcfs_clsize * ncl;
477
478 if (error == ENOSPC &&
479 (pcp->pc_size - uio->uio_loffset)
480 > 0) {
481 PC_DPRINTF3(2, "rwpcp ENOSPC "
482 "off=%lld n=%d size=%d\n",
483 uio->uio_loffset,
484 n, pcp->pc_size);
485 n = (int)(pcp->pc_size -
486 uio->uio_loffset);
487 } else {
488 PC_DPRINTF1(1,
489 "rwpcp error1=%d\n", error);
490 (void) segmap_release(segkmap,
491 base, 0);
492 break;
493 }
494 } else {
495 pcp->pc_size =
496 (uint_t)(uio->uio_loffset + n);
497 }
498 if (mapon == 0) {
499 newpage = segmap_pagecreate(segkmap,
500 base, (size_t)n, 0);
501 pagecreate = 1;
502 }
503 } else if (n == MAXBSIZE) {
504 newpage = segmap_pagecreate(segkmap, base,
505 (size_t)n, 0);
506 pagecreate = 1;
507 }
508 }
509 error = uiomove(base + mapon, (size_t)n, rw, uio);
510
511 if (pagecreate && uio->uio_loffset <
512 roundup(off + mapon + n, PAGESIZE)) {
513 offset_t nzero, nmoved;
514
515 nmoved = uio->uio_loffset - (off + mapon);
516 nzero = roundup(mapon + n, PAGESIZE) - nmoved;
517 (void) kzero(base + mapon + nmoved, (size_t)nzero);
518 }
519
520 /*
521 * Unlock the pages which have been allocated by
522 * page_create_va() in segmap_pagecreate().
523 */
524 if (newpage) {
525 segmap_pageunlock(segkmap, base, (size_t)n,
526 rw == UIO_WRITE ? S_WRITE : S_READ);
527 }
528
529 if (error) {
530 PC_DPRINTF1(1, "rwpcp error2=%d\n", error);
531 /*
532 * If we failed on a write, we may have already
533 * allocated file blocks as well as pages. It's hard
534 * to undo the block allocation, but we must be sure
535 * to invalidate any pages that may have been
536 * allocated.
537 */
538 if (rw == UIO_WRITE)
539 (void) segmap_release(segkmap, base, SM_INVAL);
540 else
541 (void) segmap_release(segkmap, base, 0);
542 } else {
543 uint_t flags = 0;
544
545 if (rw == UIO_READ) {
546 if (n + mapon == MAXBSIZE ||
547 uio->uio_loffset == pcp->pc_size)
548 flags = SM_DONTNEED;
549 } else if (ioflag & (FSYNC|FDSYNC)) {
550 flags = SM_WRITE;
551 } else if (n + mapon == MAXBSIZE) {
552 flags = SM_WRITE|SM_ASYNC|SM_DONTNEED;
553 }
554 error = segmap_release(segkmap, base, flags);
555 }
556
557 } while (error == 0 && uio->uio_resid > 0 && n != 0);
558
559 if (oresid != uio->uio_resid)
560 error = 0;
561 return (error);
562 }
563
564 /*ARGSUSED*/
565 static int
566 pcfs_getattr(
567 struct vnode *vp,
568 struct vattr *vap,
569 int flags,
570 struct cred *cr,
571 caller_context_t *ct)
572 {
573 struct pcnode *pcp;
574 struct pcfs *fsp;
575 int error;
576 char attr;
577 struct pctime atime;
578 int64_t unixtime;
579
580 PC_DPRINTF1(8, "pcfs_getattr: vp=%p\n", (void *)vp);
581
582 fsp = VFSTOPCFS(vp->v_vfsp);
583 error = pc_lockfs(fsp, 0, 0);
584 if (error)
585 return (error);
586
587 /*
588 * Note that we don't check for "invalid node" (PC_INVAL) here
589 * only in order to make stat() succeed. We allow no I/O on such
590 * a node, but do allow to check for its existence.
591 */
592 if ((pcp = VTOPC(vp)) == NULL) {
593 pc_unlockfs(fsp);
594 return (EIO);
595 }
596 /*
597 * Copy from pcnode.
598 */
599 vap->va_type = vp->v_type;
600 attr = pcp->pc_entry.pcd_attr;
601 if (PCA_IS_HIDDEN(fsp, attr))
602 vap->va_mode = 0;
603 else if (attr & PCA_LABEL)
604 vap->va_mode = 0444;
605 else if (attr & PCA_RDONLY)
606 vap->va_mode = 0555;
607 else if (fsp->pcfs_flags & PCFS_BOOTPART) {
608 vap->va_mode = 0755;
609 } else {
610 vap->va_mode = 0777;
611 }
612
613 if (attr & PCA_DIR)
614 vap->va_mode |= S_IFDIR;
615 else
616 vap->va_mode |= S_IFREG;
617 if (fsp->pcfs_flags & PCFS_BOOTPART) {
618 vap->va_uid = 0;
619 vap->va_gid = 0;
620 } else {
621 vap->va_uid = crgetuid(cr);
622 vap->va_gid = crgetgid(cr);
623 }
624 vap->va_fsid = vp->v_vfsp->vfs_dev;
625 vap->va_nodeid = (ino64_t)pc_makenodeid(pcp->pc_eblkno,
626 pcp->pc_eoffset, pcp->pc_entry.pcd_attr,
627 pc_getstartcluster(fsp, &pcp->pc_entry), pc_direntpersec(fsp));
628 vap->va_nlink = 1;
629 vap->va_size = (u_offset_t)pcp->pc_size;
630 vap->va_rdev = 0;
631 vap->va_nblocks =
632 (fsblkcnt64_t)howmany((offset_t)pcp->pc_size, DEV_BSIZE);
633 vap->va_blksize = fsp->pcfs_clsize;
634
635 /*
636 * FAT root directories have no timestamps. In order not to return
637 * "time zero" (1/1/1970), we record the time of the mount and give
638 * that. This breaks less expectations.
639 */
640 if (vp->v_flag & VROOT) {
641 vap->va_mtime = fsp->pcfs_mounttime;
642 vap->va_atime = fsp->pcfs_mounttime;
643 vap->va_ctime = fsp->pcfs_mounttime;
644 pc_unlockfs(fsp);
645 return (0);
646 }
647
648 pc_pcttotv(&pcp->pc_entry.pcd_mtime, &unixtime);
649 if ((fsp->pcfs_flags & PCFS_NOCLAMPTIME) == 0) {
650 if (unixtime > INT32_MAX)
651 DTRACE_PROBE1(pcfs__mtimeclamped, int64_t, unixtime);
652 unixtime = MIN(unixtime, INT32_MAX);
653 } else if (unixtime > INT32_MAX &&
654 get_udatamodel() == DATAMODEL_ILP32) {
655 pc_unlockfs(fsp);
656 DTRACE_PROBE1(pcfs__mtimeoverflowed, int64_t, unixtime);
657 return (EOVERFLOW);
658 }
659
660 vap->va_mtime.tv_sec = (time_t)unixtime;
661 vap->va_mtime.tv_nsec = 0;
662
663 /*
664 * FAT doesn't know about POSIX ctime.
665 * Best approximation is to always set it to mtime.
666 */
667 vap->va_ctime = vap->va_mtime;
668
669 /*
670 * FAT only stores "last access date". If that's the
671 * same as the date of last modification then the time
672 * of last access is known. Otherwise, use midnight.
673 */
674 atime.pct_date = pcp->pc_entry.pcd_ladate;
675 if (atime.pct_date == pcp->pc_entry.pcd_mtime.pct_date)
676 atime.pct_time = pcp->pc_entry.pcd_mtime.pct_time;
677 else
678 atime.pct_time = 0;
679 pc_pcttotv(&atime, &unixtime);
680 if ((fsp->pcfs_flags & PCFS_NOCLAMPTIME) == 0) {
681 if (unixtime > INT32_MAX)
682 DTRACE_PROBE1(pcfs__atimeclamped, int64_t, unixtime);
683 unixtime = MIN(unixtime, INT32_MAX);
684 } else if (unixtime > INT32_MAX &&
685 get_udatamodel() == DATAMODEL_ILP32) {
686 pc_unlockfs(fsp);
687 DTRACE_PROBE1(pcfs__atimeoverflowed, int64_t, unixtime);
688 return (EOVERFLOW);
689 }
690
691 vap->va_atime.tv_sec = (time_t)unixtime;
692 vap->va_atime.tv_nsec = 0;
693
694 pc_unlockfs(fsp);
695 return (0);
696 }
697
698
699 /*ARGSUSED*/
700 static int
701 pcfs_setattr(
702 struct vnode *vp,
703 struct vattr *vap,
704 int flags,
705 struct cred *cr,
706 caller_context_t *ct)
707 {
708 struct pcnode *pcp;
709 mode_t mask = vap->va_mask;
710 int error;
711 struct pcfs *fsp;
712 timestruc_t now, *timep;
713
714 PC_DPRINTF2(6, "pcfs_setattr: vp=%p mask=%x\n", (void *)vp, (int)mask);
715 /*
716 * cannot set these attributes
717 */
718 if (mask & (AT_NOSET | AT_UID | AT_GID)) {
719 return (EINVAL);
720 }
721 /*
722 * pcfs_setattr is now allowed on directories to avoid silly warnings
723 * from 'tar' when it tries to set times on a directory, and console
724 * printf's on the NFS server when it gets EINVAL back on such a
725 * request. One possible problem with that since a directory entry
726 * identifies a file, '.' and all the '..' entries in subdirectories
727 * may get out of sync when the directory is updated since they're
728 * treated like separate files. We could fix that by looking for
729 * '.' and giving it the same attributes, and then looking for
730 * all the subdirectories and updating '..', but that's pretty
731 * expensive for something that doesn't seem likely to matter.
732 */
733 /* can't do some ops on directories anyway */
734 if ((vp->v_type == VDIR) &&
735 (mask & AT_SIZE)) {
736 return (EINVAL);
737 }
738
739 fsp = VFSTOPCFS(vp->v_vfsp);
740 error = pc_lockfs(fsp, 0, 0);
741 if (error)
742 return (error);
743 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
744 pc_unlockfs(fsp);
745 return (EIO);
746 }
747
748 if (fsp->pcfs_flags & PCFS_BOOTPART) {
749 if (secpolicy_pcfs_modify_bootpartition(cr) != 0) {
750 pc_unlockfs(fsp);
751 return (EACCES);
752 }
753 }
754
755 /*
756 * Change file access modes.
757 * If nobody has write permission, file is marked readonly.
758 * Otherwise file is writable by anyone.
759 */
760 if ((mask & AT_MODE) && (vap->va_mode != (mode_t)-1)) {
761 if ((vap->va_mode & 0222) == 0)
762 pcp->pc_entry.pcd_attr |= PCA_RDONLY;
763 else
764 pcp->pc_entry.pcd_attr &= ~PCA_RDONLY;
765 pcp->pc_flags |= PC_CHG;
766 }
767 /*
768 * Truncate file. Must have write permission.
769 */
770 if ((mask & AT_SIZE) && (vap->va_size != (u_offset_t)-1)) {
771 if (pcp->pc_entry.pcd_attr & PCA_RDONLY) {
772 error = EACCES;
773 goto out;
774 }
775 if (vap->va_size > UINT32_MAX) {
776 error = EFBIG;
777 goto out;
778 }
779 error = pc_truncate(pcp, (uint_t)vap->va_size);
780
781 if (error)
782 goto out;
783
784 if (vap->va_size == 0) {
785 vnevent_truncate(vp, ct);
786 } else {
787 vnevent_resize(vp, ct);
788 }
789 }
790 /*
791 * Change file modified times.
792 */
793 if (mask & (AT_MTIME | AT_CTIME)) {
794 /*
795 * If SysV-compatible option to set access and
796 * modified times if privileged, owner, or write access,
797 * use current time rather than va_mtime.
798 *
799 * XXX - va_mtime.tv_sec == -1 flags this.
800 */
801 timep = &vap->va_mtime;
802 if (vap->va_mtime.tv_sec == -1) {
803 gethrestime(&now);
804 timep = &now;
805 }
806 if ((fsp->pcfs_flags & PCFS_NOCLAMPTIME) == 0 &&
807 timep->tv_sec > INT32_MAX) {
808 error = EOVERFLOW;
809 goto out;
810 }
811 error = pc_tvtopct(timep, &pcp->pc_entry.pcd_mtime);
812 if (error)
813 goto out;
814 pcp->pc_flags |= PC_CHG;
815 }
816 /*
817 * Change file access times.
818 */
819 if (mask & AT_ATIME) {
820 /*
821 * If SysV-compatible option to set access and
822 * modified times if privileged, owner, or write access,
823 * use current time rather than va_mtime.
824 *
825 * XXX - va_atime.tv_sec == -1 flags this.
826 */
827 struct pctime atime;
828
829 timep = &vap->va_atime;
830 if (vap->va_atime.tv_sec == -1) {
831 gethrestime(&now);
832 timep = &now;
833 }
834 if ((fsp->pcfs_flags & PCFS_NOCLAMPTIME) == 0 &&
835 timep->tv_sec > INT32_MAX) {
836 error = EOVERFLOW;
837 goto out;
838 }
839 error = pc_tvtopct(timep, &atime);
840 if (error)
841 goto out;
842 pcp->pc_entry.pcd_ladate = atime.pct_date;
843 pcp->pc_flags |= PC_CHG;
844 }
845 out:
846 pc_unlockfs(fsp);
847 return (error);
848 }
849
850
851 /*ARGSUSED*/
852 static int
853 pcfs_access(
854 struct vnode *vp,
855 int mode,
856 int flags,
857 struct cred *cr,
858 caller_context_t *ct)
859 {
860 struct pcnode *pcp;
861 struct pcfs *fsp;
862
863
864 fsp = VFSTOPCFS(vp->v_vfsp);
865
866 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL)
867 return (EIO);
868 if ((mode & VWRITE) && (pcp->pc_entry.pcd_attr & PCA_RDONLY))
869 return (EACCES);
870
871 /*
872 * If this is a boot partition, privileged users have full access while
873 * others have read-only access.
874 */
875 if (fsp->pcfs_flags & PCFS_BOOTPART) {
876 if ((mode & VWRITE) &&
877 secpolicy_pcfs_modify_bootpartition(cr) != 0)
878 return (EACCES);
879 }
880 return (0);
881 }
882
883
884 /*ARGSUSED*/
885 static int
886 pcfs_fsync(
887 struct vnode *vp,
888 int syncflag,
889 struct cred *cr,
890 caller_context_t *ct)
891 {
892 struct pcfs *fsp;
893 struct pcnode *pcp;
894 int error;
895
896 fsp = VFSTOPCFS(vp->v_vfsp);
897 if (error = pc_verify(fsp))
898 return (error);
899 error = pc_lockfs(fsp, 0, 0);
900 if (error)
901 return (error);
902 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
903 pc_unlockfs(fsp);
904 return (EIO);
905 }
906 rw_enter(&pcnodes_lock, RW_WRITER);
907 error = pc_nodesync(pcp);
908 rw_exit(&pcnodes_lock);
909 pc_unlockfs(fsp);
910 return (error);
911 }
912
913
914 /*ARGSUSED*/
915 static void
916 pcfs_inactive(
917 struct vnode *vp,
918 struct cred *cr,
919 caller_context_t *ct)
920 {
921 struct pcnode *pcp;
922 struct pcfs *fsp;
923 int error;
924
925 fsp = VFSTOPCFS(vp->v_vfsp);
926 error = pc_lockfs(fsp, 0, 1);
927
928 /*
929 * If the filesystem was umounted by force, all dirty
930 * pages associated with this vnode are invalidated
931 * and then the vnode will be freed.
932 */
933 if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED) {
934 pcp = VTOPC(vp);
935 if (vn_has_cached_data(vp)) {
936 (void) pvn_vplist_dirty(vp, (u_offset_t)0,
937 pcfs_putapage, B_INVAL, (struct cred *)NULL);
938 }
939 remque(pcp);
940 if (error == 0)
941 pc_unlockfs(fsp);
942 vn_free(vp);
943 kmem_free(pcp, sizeof (struct pcnode));
944 VFS_RELE(PCFSTOVFS(fsp));
945 return;
946 }
947
948 mutex_enter(&vp->v_lock);
949 ASSERT(vp->v_count >= 1);
950 if (vp->v_count > 1) {
951 vp->v_count--; /* release our hold from vn_rele */
952 mutex_exit(&vp->v_lock);
953 pc_unlockfs(fsp);
954 return;
955 }
956 mutex_exit(&vp->v_lock);
957
958 /*
959 * Check again to confirm that no intervening I/O error
960 * with a subsequent pc_diskchanged() call has released
961 * the pcnode. If it has then release the vnode as above.
962 */
963 pcp = VTOPC(vp);
964 if (pcp == NULL || pcp->pc_flags & PC_INVAL) {
965 if (vn_has_cached_data(vp))
966 (void) pvn_vplist_dirty(vp, (u_offset_t)0,
967 pcfs_putapage, B_INVAL | B_TRUNC,
968 (struct cred *)NULL);
969 }
970
971 if (pcp == NULL) {
972 vn_free(vp);
973 } else {
974 pc_rele(pcp);
975 }
976
977 if (!error)
978 pc_unlockfs(fsp);
979 }
980
981 /*ARGSUSED*/
982 static int
983 pcfs_lookup(
984 struct vnode *dvp,
985 char *nm,
986 struct vnode **vpp,
987 struct pathname *pnp,
988 int flags,
989 struct vnode *rdir,
990 struct cred *cr,
991 caller_context_t *ct,
992 int *direntflags,
993 pathname_t *realpnp)
994 {
995 struct pcfs *fsp;
996 struct pcnode *pcp;
997 int error;
998
999 /*
1000 * If the filesystem was umounted by force, return immediately.
1001 */
1002 if (dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1003 return (EIO);
1004
1005 /*
1006 * verify that the dvp is still valid on the disk
1007 */
1008 fsp = VFSTOPCFS(dvp->v_vfsp);
1009 if (error = pc_verify(fsp))
1010 return (error);
1011 error = pc_lockfs(fsp, 0, 0);
1012 if (error)
1013 return (error);
1014 if (VTOPC(dvp) == NULL || VTOPC(dvp)->pc_flags & PC_INVAL) {
1015 pc_unlockfs(fsp);
1016 return (EIO);
1017 }
1018 /*
1019 * Null component name is a synonym for directory being searched.
1020 */
1021 if (*nm == '\0') {
1022 VN_HOLD(dvp);
1023 *vpp = dvp;
1024 pc_unlockfs(fsp);
1025 return (0);
1026 }
1027
1028 error = pc_dirlook(VTOPC(dvp), nm, &pcp);
1029 if (!error) {
1030 *vpp = PCTOV(pcp);
1031 pcp->pc_flags |= PC_EXTERNAL;
1032 }
1033 pc_unlockfs(fsp);
1034 return (error);
1035 }
1036
1037
1038 /*ARGSUSED*/
1039 static int
1040 pcfs_create(
1041 struct vnode *dvp,
1042 char *nm,
1043 struct vattr *vap,
1044 enum vcexcl exclusive,
1045 int mode,
1046 struct vnode **vpp,
1047 struct cred *cr,
1048 int flag,
1049 caller_context_t *ct,
1050 vsecattr_t *vsecp)
1051 {
1052 int error;
1053 struct pcnode *pcp;
1054 struct vnode *vp;
1055 struct pcfs *fsp;
1056
1057 /*
1058 * can't create directories. use pcfs_mkdir.
1059 * can't create anything other than files.
1060 */
1061 if (vap->va_type == VDIR)
1062 return (EISDIR);
1063 else if (vap->va_type != VREG)
1064 return (EINVAL);
1065
1066 pcp = NULL;
1067 fsp = VFSTOPCFS(dvp->v_vfsp);
1068 error = pc_lockfs(fsp, 0, 0);
1069 if (error)
1070 return (error);
1071 if (VTOPC(dvp) == NULL || VTOPC(dvp)->pc_flags & PC_INVAL) {
1072 pc_unlockfs(fsp);
1073 return (EIO);
1074 }
1075
1076 if (fsp->pcfs_flags & PCFS_BOOTPART) {
1077 if (secpolicy_pcfs_modify_bootpartition(cr) != 0) {
1078 pc_unlockfs(fsp);
1079 return (EACCES);
1080 }
1081 }
1082
1083 if (*nm == '\0') {
1084 /*
1085 * Null component name refers to the directory itself.
1086 */
1087 VN_HOLD(dvp);
1088 pcp = VTOPC(dvp);
1089 error = EEXIST;
1090 } else {
1091 error = pc_direnter(VTOPC(dvp), nm, vap, &pcp);
1092 }
1093 /*
1094 * if file exists and this is a nonexclusive create,
1095 * check for access permissions
1096 */
1097 if (error == EEXIST) {
1098 vp = PCTOV(pcp);
1099 if (exclusive == NONEXCL) {
1100 if (vp->v_type == VDIR) {
1101 error = EISDIR;
1102 } else if (mode) {
1103 error = pcfs_access(PCTOV(pcp), mode, 0,
1104 cr, ct);
1105 } else {
1106 error = 0;
1107 }
1108 }
1109 if (error) {
1110 VN_RELE(PCTOV(pcp));
1111 } else if ((vp->v_type == VREG) && (vap->va_mask & AT_SIZE) &&
1112 (vap->va_size == 0)) {
1113 error = pc_truncate(pcp, 0L);
1114 if (error) {
1115 VN_RELE(PCTOV(pcp));
1116 } else {
1117 vnevent_create(PCTOV(pcp), ct);
1118 }
1119 }
1120 }
1121 if (error) {
1122 pc_unlockfs(fsp);
1123 return (error);
1124 }
1125 *vpp = PCTOV(pcp);
1126 pcp->pc_flags |= PC_EXTERNAL;
1127 pc_unlockfs(fsp);
1128 return (error);
1129 }
1130
1131 /*ARGSUSED*/
1132 static int
1133 pcfs_remove(
1134 struct vnode *vp,
1135 char *nm,
1136 struct cred *cr,
1137 caller_context_t *ct,
1138 int flags)
1139 {
1140 struct pcfs *fsp;
1141 struct pcnode *pcp;
1142 int error;
1143
1144 fsp = VFSTOPCFS(vp->v_vfsp);
1145 if (error = pc_verify(fsp))
1146 return (error);
1147 error = pc_lockfs(fsp, 0, 0);
1148 if (error)
1149 return (error);
1150 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
1151 pc_unlockfs(fsp);
1152 return (EIO);
1153 }
1154 if (fsp->pcfs_flags & PCFS_BOOTPART) {
1155 if (secpolicy_pcfs_modify_bootpartition(cr) != 0) {
1156 pc_unlockfs(fsp);
1157 return (EACCES);
1158 }
1159 }
1160 error = pc_dirremove(pcp, nm, (struct vnode *)0, VREG, ct);
1161 pc_unlockfs(fsp);
1162 return (error);
1163 }
1164
1165 /*
1166 * Rename a file or directory
1167 * This rename is restricted to only rename files within a directory.
1168 * XX should make rename more general
1169 */
1170 /*ARGSUSED*/
1171 static int
1172 pcfs_rename(
1173 struct vnode *sdvp, /* old (source) parent vnode */
1174 char *snm, /* old (source) entry name */
1175 struct vnode *tdvp, /* new (target) parent vnode */
1176 char *tnm, /* new (target) entry name */
1177 struct cred *cr,
1178 caller_context_t *ct,
1179 int flags)
1180 {
1181 struct pcfs *fsp;
1182 struct pcnode *dp; /* parent pcnode */
1183 struct pcnode *tdp;
1184 int error;
1185
1186 fsp = VFSTOPCFS(sdvp->v_vfsp);
1187 if (error = pc_verify(fsp))
1188 return (error);
1189
1190 /*
1191 * make sure we can muck with this directory.
1192 */
1193 error = pcfs_access(sdvp, VWRITE, 0, cr, ct);
1194 if (error) {
1195 return (error);
1196 }
1197 error = pc_lockfs(fsp, 0, 0);
1198 if (error)
1199 return (error);
1200 if (((dp = VTOPC(sdvp)) == NULL) || ((tdp = VTOPC(tdvp)) == NULL) ||
1201 (dp->pc_flags & PC_INVAL) || (tdp->pc_flags & PC_INVAL)) {
1202 pc_unlockfs(fsp);
1203 return (EIO);
1204 }
1205 error = pc_rename(dp, tdp, snm, tnm, ct);
1206 pc_unlockfs(fsp);
1207 return (error);
1208 }
1209
1210 /*ARGSUSED*/
1211 static int
1212 pcfs_mkdir(
1213 struct vnode *dvp,
1214 char *nm,
1215 struct vattr *vap,
1216 struct vnode **vpp,
1217 struct cred *cr,
1218 caller_context_t *ct,
1219 int flags,
1220 vsecattr_t *vsecp)
1221 {
1222 struct pcfs *fsp;
1223 struct pcnode *pcp;
1224 int error;
1225
1226 fsp = VFSTOPCFS(dvp->v_vfsp);
1227 if (error = pc_verify(fsp))
1228 return (error);
1229 error = pc_lockfs(fsp, 0, 0);
1230 if (error)
1231 return (error);
1232 if (VTOPC(dvp) == NULL || VTOPC(dvp)->pc_flags & PC_INVAL) {
1233 pc_unlockfs(fsp);
1234 return (EIO);
1235 }
1236
1237 if (fsp->pcfs_flags & PCFS_BOOTPART) {
1238 if (secpolicy_pcfs_modify_bootpartition(cr) != 0) {
1239 pc_unlockfs(fsp);
1240 return (EACCES);
1241 }
1242 }
1243
1244 error = pc_direnter(VTOPC(dvp), nm, vap, &pcp);
1245
1246 if (!error) {
1247 pcp -> pc_flags |= PC_EXTERNAL;
1248 *vpp = PCTOV(pcp);
1249 } else if (error == EEXIST) {
1250 VN_RELE(PCTOV(pcp));
1251 }
1252 pc_unlockfs(fsp);
1253 return (error);
1254 }
1255
1256 /*ARGSUSED*/
1257 static int
1258 pcfs_rmdir(
1259 struct vnode *dvp,
1260 char *nm,
1261 struct vnode *cdir,
1262 struct cred *cr,
1263 caller_context_t *ct,
1264 int flags)
1265 {
1266 struct pcfs *fsp;
1267 struct pcnode *pcp;
1268 int error;
1269
1270 fsp = VFSTOPCFS(dvp -> v_vfsp);
1271 if (error = pc_verify(fsp))
1272 return (error);
1273 if (error = pc_lockfs(fsp, 0, 0))
1274 return (error);
1275
1276 if ((pcp = VTOPC(dvp)) == NULL || pcp->pc_flags & PC_INVAL) {
1277 pc_unlockfs(fsp);
1278 return (EIO);
1279 }
1280
1281 if (fsp->pcfs_flags & PCFS_BOOTPART) {
1282 if (secpolicy_pcfs_modify_bootpartition(cr) != 0) {
1283 pc_unlockfs(fsp);
1284 return (EACCES);
1285 }
1286 }
1287
1288 error = pc_dirremove(pcp, nm, cdir, VDIR, ct);
1289 pc_unlockfs(fsp);
1290 return (error);
1291 }
1292
1293 /*
1294 * read entries in a directory.
1295 * we must convert pc format to unix format
1296 */
1297
1298 /*ARGSUSED*/
1299 static int
1300 pcfs_readdir(
1301 struct vnode *dvp,
1302 struct uio *uiop,
1303 struct cred *cr,
1304 int *eofp,
1305 caller_context_t *ct,
1306 int flags)
1307 {
1308 struct pcnode *pcp;
1309 struct pcfs *fsp;
1310 struct pcdir *ep;
1311 struct buf *bp = NULL;
1312 offset_t offset;
1313 int boff;
1314 struct pc_dirent lbp;
1315 struct pc_dirent *ld = &lbp;
1316 int error;
1317
1318 /*
1319 * If the filesystem was umounted by force, return immediately.
1320 */
1321 if (dvp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1322 return (EIO);
1323
1324 if ((uiop->uio_iovcnt != 1) ||
1325 (uiop->uio_loffset % sizeof (struct pcdir)) != 0) {
1326 return (EINVAL);
1327 }
1328 fsp = VFSTOPCFS(dvp->v_vfsp);
1329 /*
1330 * verify that the dp is still valid on the disk
1331 */
1332 if (error = pc_verify(fsp)) {
1333 return (error);
1334 }
1335 error = pc_lockfs(fsp, 0, 0);
1336 if (error)
1337 return (error);
1338 if ((pcp = VTOPC(dvp)) == NULL || pcp->pc_flags & PC_INVAL) {
1339 pc_unlockfs(fsp);
1340 return (EIO);
1341 }
1342
1343 bzero(ld, sizeof (*ld));
1344
1345 if (eofp != NULL)
1346 *eofp = 0;
1347 offset = uiop->uio_loffset;
1348
1349 if (dvp->v_flag & VROOT) {
1350 /*
1351 * kludge up entries for "." and ".." in the root.
1352 */
1353 if (offset == 0) {
1354 (void) strcpy(ld->d_name, ".");
1355 ld->d_reclen = DIRENT64_RECLEN(1);
1356 ld->d_off = (off64_t)sizeof (struct pcdir);
1357 ld->d_ino = (ino64_t)UINT_MAX;
1358 if (ld->d_reclen > uiop->uio_resid) {
1359 pc_unlockfs(fsp);
1360 return (ENOSPC);
1361 }
1362 (void) uiomove(ld, ld->d_reclen, UIO_READ, uiop);
1363 uiop->uio_loffset = ld->d_off;
1364 offset = uiop->uio_loffset;
1365 }
1366 if (offset == sizeof (struct pcdir)) {
1367 (void) strcpy(ld->d_name, "..");
1368 ld->d_reclen = DIRENT64_RECLEN(2);
1369 if (ld->d_reclen > uiop->uio_resid) {
1370 pc_unlockfs(fsp);
1371 return (ENOSPC);
1372 }
1373 ld->d_off = (off64_t)(uiop->uio_loffset +
1374 sizeof (struct pcdir));
1375 ld->d_ino = (ino64_t)UINT_MAX;
1376 (void) uiomove(ld, ld->d_reclen, UIO_READ, uiop);
1377 uiop->uio_loffset = ld->d_off;
1378 offset = uiop->uio_loffset;
1379 }
1380 offset -= 2 * sizeof (struct pcdir);
1381 /* offset now has the real offset value into directory file */
1382 }
1383
1384 for (;;) {
1385 boff = pc_blkoff(fsp, offset);
1386 if (boff == 0 || bp == NULL || boff >= bp->b_bcount) {
1387 if (bp != NULL) {
1388 brelse(bp);
1389 bp = NULL;
1390 }
1391 error = pc_blkatoff(pcp, offset, &bp, &ep);
1392 if (error) {
1393 if (error == ENOENT) {
1394 error = 0;
1395 if (eofp)
1396 *eofp = 1;
1397 }
1398 break;
1399 }
1400 }
1401 if (ep->pcd_filename[0] == PCD_UNUSED) {
1402 if (eofp)
1403 *eofp = 1;
1404 break;
1405 }
1406 /*
1407 * Don't display label because it may contain funny characters.
1408 */
1409 if (ep->pcd_filename[0] == PCD_ERASED) {
1410 uiop->uio_loffset += sizeof (struct pcdir);
1411 offset += sizeof (struct pcdir);
1412 ep++;
1413 continue;
1414 }
1415 if (PCDL_IS_LFN(ep)) {
1416 if (pc_read_long_fn(dvp, uiop, ld, &ep, &offset, &bp) !=
1417 0)
1418 break;
1419 continue;
1420 }
1421
1422 if (pc_read_short_fn(dvp, uiop, ld, &ep, &offset, &bp) != 0)
1423 break;
1424 }
1425 if (bp)
1426 brelse(bp);
1427 pc_unlockfs(fsp);
1428 return (error);
1429 }
1430
1431
1432 /*
1433 * Called from pvn_getpages to get a particular page. When we are called
1434 * the pcfs is already locked.
1435 */
1436 /*ARGSUSED*/
1437 static int
1438 pcfs_getapage(
1439 struct vnode *vp,
1440 u_offset_t off,
1441 size_t len,
1442 uint_t *protp,
1443 page_t *pl[], /* NULL if async IO is requested */
1444 size_t plsz,
1445 struct seg *seg,
1446 caddr_t addr,
1447 enum seg_rw rw,
1448 struct cred *cr)
1449 {
1450 struct pcnode *pcp;
1451 struct pcfs *fsp = VFSTOPCFS(vp->v_vfsp);
1452 struct vnode *devvp;
1453 page_t *pp;
1454 page_t *pagefound;
1455 int err;
1456
1457 /*
1458 * If the filesystem was umounted by force, return immediately.
1459 */
1460 if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1461 return (EIO);
1462
1463 PC_DPRINTF3(5, "pcfs_getapage: vp=%p off=%lld len=%lu\n",
1464 (void *)vp, off, len);
1465
1466 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL)
1467 return (EIO);
1468 devvp = fsp->pcfs_devvp;
1469
1470 /* pcfs doesn't do readaheads */
1471 if (pl == NULL)
1472 return (0);
1473
1474 pl[0] = NULL;
1475 err = 0;
1476 /*
1477 * If the accessed time on the pcnode has not already been
1478 * set elsewhere (e.g. for read/setattr) we set the time now.
1479 * This gives us approximate modified times for mmap'ed files
1480 * which are accessed via loads in the user address space.
1481 */
1482 if ((pcp->pc_flags & PC_ACC) == 0 &&
1483 ((fsp->pcfs_vfs->vfs_flag & VFS_RDONLY) == 0)) {
1484 pc_mark_acc(fsp, pcp);
1485 }
1486 reread:
1487 if ((pagefound = page_exists(vp, off)) == NULL) {
1488 /*
1489 * Need to really do disk IO to get the page(s).
1490 */
1491 struct buf *bp;
1492 daddr_t lbn, bn;
1493 u_offset_t io_off;
1494 size_t io_len;
1495 u_offset_t lbnoff, xferoffset;
1496 u_offset_t pgoff;
1497 uint_t xfersize;
1498 int err1;
1499
1500 lbn = pc_lblkno(fsp, off);
1501 lbnoff = off & ~(fsp->pcfs_clsize - 1);
1502 xferoffset = off & ~(fsp->pcfs_secsize - 1);
1503
1504 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, &io_len,
1505 off, (size_t)MIN(pc_blksize(fsp, pcp, off), PAGESIZE), 0);
1506 if (pp == NULL)
1507 /*
1508 * XXX - If pcfs is made MT-hot, this should go
1509 * back to reread.
1510 */
1511 panic("pcfs_getapage pvn_read_kluster");
1512
1513 for (pgoff = 0; pgoff < PAGESIZE && xferoffset < pcp->pc_size;
1514 pgoff += xfersize,
1515 lbn += howmany(xfersize, fsp->pcfs_clsize),
1516 lbnoff += xfersize, xferoffset += xfersize) {
1517 /*
1518 * read as many contiguous blocks as possible to
1519 * fill this page
1520 */
1521 xfersize = PAGESIZE - pgoff;
1522 err1 = pc_bmap(pcp, lbn, &bn, &xfersize);
1523 if (err1) {
1524 PC_DPRINTF1(1, "pc_getapage err=%d", err1);
1525 err = err1;
1526 goto out;
1527 }
1528 bp = pageio_setup(pp, xfersize, devvp, B_READ);
1529 bp->b_edev = devvp->v_rdev;
1530 bp->b_dev = cmpdev(devvp->v_rdev);
1531 bp->b_blkno = bn + btodt(xferoffset - lbnoff);
1532 bp->b_un.b_addr = (caddr_t)(uintptr_t)pgoff;
1533 bp->b_file = vp;
1534 bp->b_offset = (offset_t)(off + pgoff);
1535
1536 (void) bdev_strategy(bp);
1537
1538 lwp_stat_update(LWP_STAT_INBLK, 1);
1539
1540 if (err == 0)
1541 err = biowait(bp);
1542 else
1543 (void) biowait(bp);
1544 pageio_done(bp);
1545 if (err)
1546 goto out;
1547 }
1548 if (pgoff < PAGESIZE) {
1549 pagezero(pp->p_prev, pgoff, PAGESIZE - pgoff);
1550 }
1551 pvn_plist_init(pp, pl, plsz, off, io_len, rw);
1552 }
1553 out:
1554 if (err) {
1555 if (pp != NULL)
1556 pvn_read_done(pp, B_ERROR);
1557 return (err);
1558 }
1559
1560 if (pagefound) {
1561 /*
1562 * Page exists in the cache, acquire the "shared"
1563 * lock. If this fails, go back to reread.
1564 */
1565 if ((pp = page_lookup(vp, off, SE_SHARED)) == NULL) {
1566 goto reread;
1567 }
1568 pl[0] = pp;
1569 pl[1] = NULL;
1570 }
1571 return (err);
1572 }
1573
1574 /*
1575 * Return all the pages from [off..off+len] in given file
1576 */
1577 /* ARGSUSED */
1578 static int
1579 pcfs_getpage(
1580 struct vnode *vp,
1581 offset_t off,
1582 size_t len,
1583 uint_t *protp,
1584 page_t *pl[],
1585 size_t plsz,
1586 struct seg *seg,
1587 caddr_t addr,
1588 enum seg_rw rw,
1589 struct cred *cr,
1590 caller_context_t *ct)
1591 {
1592 struct pcfs *fsp = VFSTOPCFS(vp->v_vfsp);
1593 int err;
1594
1595 PC_DPRINTF0(6, "pcfs_getpage\n");
1596 if (err = pc_verify(fsp))
1597 return (err);
1598 if (vp->v_flag & VNOMAP)
1599 return (ENOSYS);
1600 ASSERT(off <= UINT32_MAX);
1601 err = pc_lockfs(fsp, 0, 0);
1602 if (err)
1603 return (err);
1604 if (protp != NULL)
1605 *protp = PROT_ALL;
1606
1607 ASSERT((off & PAGEOFFSET) == 0);
1608 err = pvn_getpages(pcfs_getapage, vp, off, len, protp, pl, plsz,
1609 seg, addr, rw, cr);
1610
1611 pc_unlockfs(fsp);
1612 return (err);
1613 }
1614
1615
1616 /*
1617 * Flags are composed of {B_INVAL, B_FREE, B_DONTNEED, B_FORCE}
1618 * If len == 0, do from off to EOF.
1619 *
1620 * The normal cases should be len == 0 & off == 0 (entire vp list),
1621 * len == MAXBSIZE (from segmap_release actions), and len == PAGESIZE
1622 * (from pageout).
1623 *
1624 */
1625 /*ARGSUSED*/
1626 static int
1627 pcfs_putpage(
1628 struct vnode *vp,
1629 offset_t off,
1630 size_t len,
1631 int flags,
1632 struct cred *cr,
1633 caller_context_t *ct)
1634 {
1635 struct pcnode *pcp;
1636 page_t *pp;
1637 struct pcfs *fsp;
1638 u_offset_t io_off;
1639 size_t io_len;
1640 offset_t eoff;
1641 int err;
1642
1643 /*
1644 * If the filesystem was umounted by force, return immediately.
1645 */
1646 if (vp->v_vfsp->vfs_flag & VFS_UNMOUNTED)
1647 return (EIO);
1648
1649 PC_DPRINTF1(6, "pcfs_putpage vp=0x%p\n", (void *)vp);
1650 if (vp->v_flag & VNOMAP)
1651 return (ENOSYS);
1652
1653 fsp = VFSTOPCFS(vp->v_vfsp);
1654
1655 if (err = pc_verify(fsp))
1656 return (err);
1657 if ((pcp = VTOPC(vp)) == NULL) {
1658 PC_DPRINTF1(3, "pcfs_putpage NULL vp=0x%p\n", (void *)vp);
1659 return (EIO);
1660 }
1661 if (pcp->pc_flags & PC_INVAL)
1662 return (EIO);
1663
1664 if (curproc == proc_pageout) {
1665 /*
1666 * XXX - This is a quick hack to avoid blocking
1667 * pageout. Also to avoid pcfs_getapage deadlocking
1668 * with putpage when memory is running out,
1669 * since we only have one global lock and we don't
1670 * support async putpage.
1671 * It should be fixed someday.
1672 *
1673 * Interestingly, this used to be a test of NOMEMWAIT().
1674 * We only ever got here once pcfs started supporting
1675 * NFS sharing, and then only because the NFS server
1676 * threads seem to do writes in sched's process context.
1677 * Since everyone else seems to just care about pageout,
1678 * the test was changed to look for pageout directly.
1679 */
1680 return (ENOMEM);
1681 }
1682
1683 ASSERT(off <= UINT32_MAX);
1684
1685 flags &= ~B_ASYNC; /* XXX should fix this later */
1686
1687 err = pc_lockfs(fsp, 0, 0);
1688 if (err)
1689 return (err);
1690 if (!vn_has_cached_data(vp) || off >= pcp->pc_size) {
1691 pc_unlockfs(fsp);
1692 return (0);
1693 }
1694
1695 if (len == 0) {
1696 /*
1697 * Search the entire vp list for pages >= off
1698 */
1699 err = pvn_vplist_dirty(vp, off,
1700 pcfs_putapage, flags, cr);
1701 } else {
1702 eoff = off + len;
1703
1704 for (io_off = off; io_off < eoff &&
1705 io_off < pcp->pc_size; io_off += io_len) {
1706 /*
1707 * If we are not invalidating, synchronously
1708 * freeing or writing pages use the routine
1709 * page_lookup_nowait() to prevent reclaiming
1710 * them from the free list.
1711 */
1712 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) {
1713 pp = page_lookup(vp, io_off,
1714 (flags & (B_INVAL | B_FREE)) ?
1715 SE_EXCL : SE_SHARED);
1716 } else {
1717 pp = page_lookup_nowait(vp, io_off,
1718 (flags & B_FREE) ? SE_EXCL : SE_SHARED);
1719 }
1720
1721 if (pp == NULL || pvn_getdirty(pp, flags) == 0)
1722 io_len = PAGESIZE;
1723 else {
1724 err = pcfs_putapage(vp, pp, &io_off, &io_len,
1725 flags, cr);
1726 if (err != 0)
1727 break;
1728 /*
1729 * "io_off" and "io_len" are returned as
1730 * the range of pages we actually wrote.
1731 * This allows us to skip ahead more quickly
1732 * since several pages may've been dealt
1733 * with by this iteration of the loop.
1734 */
1735 }
1736 }
1737 }
1738 if (err == 0 && (flags & B_INVAL) &&
1739 off == 0 && len == 0 && vn_has_cached_data(vp)) {
1740 /*
1741 * If doing "invalidation", make sure that
1742 * all pages on the vnode list are actually
1743 * gone.
1744 */
1745 cmn_err(CE_PANIC,
1746 "pcfs_putpage: B_INVAL, pages not gone");
1747 } else if (err) {
1748 PC_DPRINTF1(1, "pcfs_putpage err=%d\n", err);
1749 }
1750 pc_unlockfs(fsp);
1751 return (err);
1752 }
1753
1754 /*
1755 * Write out a single page, possibly klustering adjacent dirty pages.
1756 */
1757 /*ARGSUSED*/
1758 int
1759 pcfs_putapage(
1760 struct vnode *vp,
1761 page_t *pp,
1762 u_offset_t *offp,
1763 size_t *lenp,
1764 int flags,
1765 struct cred *cr)
1766 {
1767 struct pcnode *pcp;
1768 struct pcfs *fsp;
1769 struct vnode *devvp;
1770 size_t io_len;
1771 daddr_t bn;
1772 u_offset_t lbn, lbnoff, xferoffset;
1773 uint_t pgoff, xfersize;
1774 int err = 0;
1775 u_offset_t io_off;
1776
1777 pcp = VTOPC(vp);
1778 fsp = VFSTOPCFS(vp->v_vfsp);
1779 devvp = fsp->pcfs_devvp;
1780
1781 /*
1782 * If the modified time on the inode has not already been
1783 * set elsewhere (e.g. for write/setattr) and this is not
1784 * a call from msync (B_FORCE) we set the time now.
1785 * This gives us approximate modified times for mmap'ed files
1786 * which are modified via stores in the user address space.
1787 */
1788 if ((pcp->pc_flags & PC_MOD) == 0 || (flags & B_FORCE)) {
1789 pcp->pc_flags |= PC_MOD;
1790 pc_mark_mod(fsp, pcp);
1791 }
1792 pp = pvn_write_kluster(vp, pp, &io_off, &io_len, pp->p_offset,
1793 PAGESIZE, flags);
1794
1795 if (fsp->pcfs_flags & PCFS_IRRECOV) {
1796 goto out;
1797 }
1798
1799 PC_DPRINTF1(7, "pc_putpage writing dirty page off=%llu\n", io_off);
1800
1801 lbn = pc_lblkno(fsp, io_off);
1802 lbnoff = io_off & ~(fsp->pcfs_clsize - 1);
1803 xferoffset = io_off & ~(fsp->pcfs_secsize - 1);
1804
1805 for (pgoff = 0; pgoff < io_len && xferoffset < pcp->pc_size;
1806 pgoff += xfersize,
1807 lbn += howmany(xfersize, fsp->pcfs_clsize),
1808 lbnoff += xfersize, xferoffset += xfersize) {
1809
1810 struct buf *bp;
1811 int err1;
1812
1813 /*
1814 * write as many contiguous blocks as possible from this page
1815 */
1816 xfersize = io_len - pgoff;
1817 err1 = pc_bmap(pcp, (daddr_t)lbn, &bn, &xfersize);
1818 if (err1) {
1819 err = err1;
1820 goto out;
1821 }
1822 bp = pageio_setup(pp, xfersize, devvp, B_WRITE | flags);
1823 bp->b_edev = devvp->v_rdev;
1824 bp->b_dev = cmpdev(devvp->v_rdev);
1825 bp->b_blkno = bn + btodt(xferoffset - lbnoff);
1826 bp->b_un.b_addr = (caddr_t)(uintptr_t)pgoff;
1827 bp->b_file = vp;
1828 bp->b_offset = (offset_t)(io_off + pgoff);
1829
1830 (void) bdev_strategy(bp);
1831
1832 lwp_stat_update(LWP_STAT_OUBLK, 1);
1833
1834 if (err == 0)
1835 err = biowait(bp);
1836 else
1837 (void) biowait(bp);
1838 pageio_done(bp);
1839 }
1840 pvn_write_done(pp, ((err) ? B_ERROR : 0) | B_WRITE | flags);
1841 pp = NULL;
1842
1843 out:
1844 if ((fsp->pcfs_flags & PCFS_IRRECOV) && pp != NULL) {
1845 pvn_write_done(pp, B_WRITE | flags);
1846 } else if (err != 0 && pp != NULL) {
1847 pvn_write_done(pp, B_ERROR | B_WRITE | flags);
1848 }
1849
1850 if (offp)
1851 *offp = io_off;
1852 if (lenp)
1853 *lenp = io_len;
1854 PC_DPRINTF4(4, "pcfs_putapage: vp=%p pp=%p off=%lld len=%lu\n",
1855 (void *)vp, (void *)pp, io_off, io_len);
1856 if (err) {
1857 PC_DPRINTF1(1, "pcfs_putapage err=%d", err);
1858 }
1859 return (err);
1860 }
1861
1862 /*ARGSUSED*/
1863 static int
1864 pcfs_map(
1865 struct vnode *vp,
1866 offset_t off,
1867 struct as *as,
1868 caddr_t *addrp,
1869 size_t len,
1870 uchar_t prot,
1871 uchar_t maxprot,
1872 uint_t flags,
1873 struct cred *cr,
1874 caller_context_t *ct)
1875 {
1876 struct segvn_crargs vn_a;
1877 int error;
1878
1879 PC_DPRINTF0(6, "pcfs_map\n");
1880 if (vp->v_flag & VNOMAP)
1881 return (ENOSYS);
1882
1883 if (off > UINT32_MAX || off + len > UINT32_MAX)
1884 return (ENXIO);
1885
1886 as_rangelock(as);
1887 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags);
1888 if (error != 0) {
1889 as_rangeunlock(as);
1890 return (error);
1891 }
1892
1893 vn_a.vp = vp;
1894 vn_a.offset = off;
1895 vn_a.type = flags & MAP_TYPE;
1896 vn_a.prot = prot;
1897 vn_a.maxprot = maxprot;
1898 vn_a.flags = flags & ~MAP_TYPE;
1899 vn_a.cred = cr;
1900 vn_a.amp = NULL;
1901 vn_a.szc = 0;
1902 vn_a.lgrp_mem_policy_flags = 0;
1903
1904 error = as_map(as, *addrp, len, segvn_create, &vn_a);
1905 as_rangeunlock(as);
1906 return (error);
1907 }
1908
1909 /* ARGSUSED */
1910 static int
1911 pcfs_seek(
1912 struct vnode *vp,
1913 offset_t ooff,
1914 offset_t *noffp,
1915 caller_context_t *ct)
1916 {
1917 if (*noffp < 0)
1918 return (EINVAL);
1919 else if (*noffp > MAXOFFSET_T)
1920 return (EINVAL);
1921 else
1922 return (0);
1923 }
1924
1925 /* ARGSUSED */
1926 static int
1927 pcfs_addmap(
1928 struct vnode *vp,
1929 offset_t off,
1930 struct as *as,
1931 caddr_t addr,
1932 size_t len,
1933 uchar_t prot,
1934 uchar_t maxprot,
1935 uint_t flags,
1936 struct cred *cr,
1937 caller_context_t *ct)
1938 {
1939 if (vp->v_flag & VNOMAP)
1940 return (ENOSYS);
1941 return (0);
1942 }
1943
1944 /*ARGSUSED*/
1945 static int
1946 pcfs_delmap(
1947 struct vnode *vp,
1948 offset_t off,
1949 struct as *as,
1950 caddr_t addr,
1951 size_t len,
1952 uint_t prot,
1953 uint_t maxprot,
1954 uint_t flags,
1955 struct cred *cr,
1956 caller_context_t *ct)
1957 {
1958 if (vp->v_flag & VNOMAP)
1959 return (ENOSYS);
1960 return (0);
1961 }
1962
1963 /*
1964 * POSIX pathconf() support.
1965 */
1966 /* ARGSUSED */
1967 static int
1968 pcfs_pathconf(
1969 struct vnode *vp,
1970 int cmd,
1971 ulong_t *valp,
1972 struct cred *cr,
1973 caller_context_t *ct)
1974 {
1975 struct pcfs *fsp = VFSTOPCFS(vp->v_vfsp);
1976
1977 switch (cmd) {
1978 case _PC_LINK_MAX:
1979 *valp = 1;
1980 return (0);
1981
1982 case _PC_CASE_BEHAVIOR:
1983 return (EINVAL);
1984
1985 case _PC_FILESIZEBITS:
1986 /*
1987 * Both FAT16 and FAT32 support 4GB - 1 byte for file size.
1988 * FAT12 can only go up to the maximum filesystem capacity
1989 * which is ~509MB.
1990 */
1991 *valp = IS_FAT12(fsp) ? 30 : 33;
1992 return (0);
1993
1994 case _PC_TIMESTAMP_RESOLUTION:
1995 /*
1996 * PCFS keeps track of modification times, it its own
1997 * internal format, to a resolution of 2 seconds.
1998 * Since 2000 million is representable in an int32_t
1999 * without overflow (or becoming negative), we allow
2000 * this value to be returned.
2001 */
2002 *valp = 2000000000L;
2003 return (0);
2004
2005 default:
2006 return (fs_pathconf(vp, cmd, valp, cr, ct));
2007 }
2008
2009 }
2010
2011 /* ARGSUSED */
2012 static int
2013 pcfs_space(
2014 struct vnode *vp,
2015 int cmd,
2016 struct flock64 *bfp,
2017 int flag,
2018 offset_t offset,
2019 cred_t *cr,
2020 caller_context_t *ct)
2021 {
2022 struct vattr vattr;
2023 int error;
2024
2025 if (cmd != F_FREESP)
2026 return (EINVAL);
2027
2028 if ((error = convoff(vp, bfp, 0, offset)) == 0) {
2029 if ((bfp->l_start > UINT32_MAX) || (bfp->l_len > UINT32_MAX))
2030 return (EFBIG);
2031 /*
2032 * we only support the special case of l_len == 0,
2033 * meaning free to end of file at this moment.
2034 */
2035 if (bfp->l_len != 0)
2036 return (EINVAL);
2037 vattr.va_mask = AT_SIZE;
2038 vattr.va_size = bfp->l_start;
2039 error = VOP_SETATTR(vp, (vattr_t *)&vattr, 0, cr, ct);
2040 }
2041 return (error);
2042 }
2043
2044 /*
2045 * Break up 'len' chars from 'buf' into a long file name chunk.
2046 * Pad with '0xff' to make Norton Disk Doctor and Microsoft ScanDisk happy.
2047 */
2048 void
2049 set_long_fn_chunk(struct pcdir_lfn *ep, char *buf, int len)
2050 {
2051 int i;
2052
2053 ASSERT(buf != NULL);
2054
2055 for (i = 0; i < PCLF_FIRSTNAMESIZE; i += 2) {
2056 if (len > 0) {
2057 ep->pcdl_firstfilename[i] = *buf++;
2058 ep->pcdl_firstfilename[i + 1] = *buf++;
2059 len -= 2;
2060 } else {
2061 ep->pcdl_firstfilename[i] = (uchar_t)0xff;
2062 ep->pcdl_firstfilename[i + 1] = (uchar_t)0xff;
2063 }
2064 }
2065
2066 for (i = 0; i < PCLF_SECONDNAMESIZE; i += 2) {
2067 if (len > 0) {
2068 ep->pcdl_secondfilename[i] = *buf++;
2069 ep->pcdl_secondfilename[i + 1] = *buf++;
2070 len -= 2;
2071 } else {
2072 ep->pcdl_secondfilename[i] = (uchar_t)0xff;
2073 ep->pcdl_secondfilename[i + 1] = (uchar_t)0xff;
2074 }
2075 }
2076 for (i = 0; i < PCLF_THIRDNAMESIZE; i += 2) {
2077 if (len > 0) {
2078 ep->pcdl_thirdfilename[i] = *buf++;
2079 ep->pcdl_thirdfilename[i + 1] = *buf++;
2080 len -= 2;
2081 } else {
2082 ep->pcdl_thirdfilename[i] = (uchar_t)0xff;
2083 ep->pcdl_thirdfilename[i + 1] = (uchar_t)0xff;
2084 }
2085 }
2086 }
2087
2088 /*
2089 * Extract the characters from the long filename chunk into 'buf'.
2090 * Return the number of characters extracted.
2091 */
2092 static int
2093 get_long_fn_chunk(struct pcdir_lfn *ep, char *buf)
2094 {
2095 char *tmp = buf;
2096 int i;
2097
2098 /* Copy all the names, no filtering now */
2099
2100 for (i = 0; i < PCLF_FIRSTNAMESIZE; i += 2, tmp += 2) {
2101 *tmp = ep->pcdl_firstfilename[i];
2102 *(tmp + 1) = ep->pcdl_firstfilename[i + 1];
2103
2104 if ((*tmp == '\0') && (*(tmp+1) == '\0'))
2105 return (tmp - buf);
2106 }
2107 for (i = 0; i < PCLF_SECONDNAMESIZE; i += 2, tmp += 2) {
2108 *tmp = ep->pcdl_secondfilename[i];
2109 *(tmp + 1) = ep->pcdl_secondfilename[i + 1];
2110
2111 if ((*tmp == '\0') && (*(tmp+1) == '\0'))
2112 return (tmp - buf);
2113 }
2114 for (i = 0; i < PCLF_THIRDNAMESIZE; i += 2, tmp += 2) {
2115 *tmp = ep->pcdl_thirdfilename[i];
2116 *(tmp + 1) = ep->pcdl_thirdfilename[i + 1];
2117
2118 if ((*tmp == '\0') && (*(tmp+1) == '\0'))
2119 return (tmp - buf);
2120 }
2121 return (tmp - buf);
2122 }
2123
2124
2125 /*
2126 * Checksum the passed in short filename.
2127 * This is used to validate each component of the long name to make
2128 * sure the long name is valid (it hasn't been "detached" from the
2129 * short filename). This algorithm was found in FreeBSD.
2130 * (sys/fs/msdosfs/msdosfs_conv.c:winChksum(), Wolfgang Solfrank)
2131 */
2132
2133 uchar_t
2134 pc_checksum_long_fn(char *name, char *ext)
2135 {
2136 uchar_t c;
2137 char b[11];
2138
2139 bcopy(name, b, 8);
2140 bcopy(ext, b+8, 3);
2141
2142 c = b[0];
2143 c = ((c << 7) | (c >> 1)) + b[1];
2144 c = ((c << 7) | (c >> 1)) + b[2];
2145 c = ((c << 7) | (c >> 1)) + b[3];
2146 c = ((c << 7) | (c >> 1)) + b[4];
2147 c = ((c << 7) | (c >> 1)) + b[5];
2148 c = ((c << 7) | (c >> 1)) + b[6];
2149 c = ((c << 7) | (c >> 1)) + b[7];
2150 c = ((c << 7) | (c >> 1)) + b[8];
2151 c = ((c << 7) | (c >> 1)) + b[9];
2152 c = ((c << 7) | (c >> 1)) + b[10];
2153
2154 return (c);
2155 }
2156
2157 /*
2158 * Read a chunk of long filename entries into 'namep'.
2159 * Return with offset pointing to short entry (on success), or next
2160 * entry to read (if this wasn't a valid lfn really).
2161 * Uses the passed-in buffer if it can, otherwise kmem_allocs() room for
2162 * a long filename.
2163 *
2164 * Can also be called with a NULL namep, in which case it just returns
2165 * whether this was really a valid long filename and consumes it
2166 * (used by pc_dirempty()).
2167 */
2168 int
2169 pc_extract_long_fn(struct pcnode *pcp, char *namep,
2170 struct pcdir **epp, offset_t *offset, struct buf **bp)
2171 {
2172 struct pcdir *ep = *epp;
2173 struct pcdir_lfn *lep = (struct pcdir_lfn *)ep;
2174 struct vnode *dvp = PCTOV(pcp);
2175 struct pcfs *fsp = VFSTOPCFS(dvp->v_vfsp);
2176 char *lfn;
2177 char *lfn_base;
2178 int boff;
2179 int i, cs;
2180 char *buf;
2181 uchar_t cksum;
2182 int detached = 0;
2183 int error = 0;
2184 int foldcase;
2185 int count = 0;
2186 size_t u16l = 0, u8l = 0;
2187 char *outbuf;
2188 size_t ret, inlen, outlen;
2189
2190 foldcase = (fsp->pcfs_flags & PCFS_FOLDCASE);
2191 lfn_base = kmem_alloc(PCMAXNAM_UTF16, KM_SLEEP);
2192 lfn = lfn_base + PCMAXNAM_UTF16 - sizeof (uint16_t);
2193 *lfn = '\0';
2194 *(lfn + 1) = '\0';
2195 cksum = lep->pcdl_checksum;
2196
2197 buf = kmem_alloc(PCMAXNAM_UTF16, KM_SLEEP);
2198 for (i = (lep->pcdl_ordinal & ~0xc0); i > 0; i--) {
2199 /* read next block if necessary */
2200 boff = pc_blkoff(fsp, *offset);
2201 if (boff == 0 || *bp == NULL || boff >= (*bp)->b_bcount) {
2202 if (*bp != NULL) {
2203 brelse(*bp);
2204 *bp = NULL;
2205 }
2206 error = pc_blkatoff(pcp, *offset, bp, &ep);
2207 if (error) {
2208 kmem_free(lfn_base, PCMAXNAM_UTF16);
2209 kmem_free(buf, PCMAXNAM_UTF16);
2210 return (error);
2211 }
2212 lep = (struct pcdir_lfn *)ep;
2213 }
2214 /* can this happen? Bad fs? */
2215 if (!PCDL_IS_LFN((struct pcdir *)lep)) {
2216 detached = 1;
2217 break;
2218 }
2219 if (cksum != lep->pcdl_checksum)
2220 detached = 1;
2221 /* process current entry */
2222 cs = get_long_fn_chunk(lep, buf);
2223 count += cs;
2224 for (; cs > 0; cs--) {
2225 /* see if we underflow */
2226 if (lfn >= lfn_base)
2227 *--lfn = buf[cs - 1];
2228 else
2229 detached = 1;
2230 }
2231 lep++;
2232 *offset += sizeof (struct pcdir);
2233 }
2234 kmem_free(buf, PCMAXNAM_UTF16);
2235 /* read next block if necessary */
2236 boff = pc_blkoff(fsp, *offset);
2237 ep = (struct pcdir *)lep;
2238 if (boff == 0 || *bp == NULL || boff >= (*bp)->b_bcount) {
2239 if (*bp != NULL) {
2240 brelse(*bp);
2241 *bp = NULL;
2242 }
2243 error = pc_blkatoff(pcp, *offset, bp, &ep);
2244 if (error) {
2245 kmem_free(lfn_base, PCMAXNAM_UTF16);
2246 return (error);
2247 }
2248 }
2249 /* should be on the short one */
2250 if (PCDL_IS_LFN(ep) || ((ep->pcd_filename[0] == PCD_UNUSED) ||
2251 (ep->pcd_filename[0] == PCD_ERASED))) {
2252 detached = 1;
2253 }
2254 if (detached ||
2255 (cksum != pc_checksum_long_fn(ep->pcd_filename, ep->pcd_ext)) ||
2256 !pc_valid_long_fn(lfn, 0)) {
2257 /*
2258 * process current entry again. This may end up another lfn
2259 * or a short name.
2260 */
2261 *epp = ep;
2262 kmem_free(lfn_base, PCMAXNAM_UTF16);
2263 return (EINVAL);
2264 }
2265 if (PCA_IS_HIDDEN(fsp, ep->pcd_attr)) {
2266 /*
2267 * Don't display label because it may contain
2268 * funny characters.
2269 */
2270 *offset += sizeof (struct pcdir);
2271 ep++;
2272 *epp = ep;
2273 kmem_free(lfn_base, PCMAXNAM_UTF16);
2274 return (EINVAL);
2275 }
2276 if (namep) {
2277 u16l = count / 2;
2278 u8l = PCMAXNAMLEN;
2279 error = uconv_u16tou8((const uint16_t *)lfn, &u16l,
2280 (uchar_t *)namep, &u8l, UCONV_IN_LITTLE_ENDIAN);
2281 /*
2282 * uconv_u16tou8() will catch conversion errors including
2283 * the case where there is not enough room to write the
2284 * converted result and the u8l will never go over the given
2285 * PCMAXNAMLEN.
2286 */
2287 if (error != 0) {
2288 kmem_free(lfn_base, PCMAXNAM_UTF16);
2289 return (EINVAL);
2290 }
2291 namep[u8l] = '\0';
2292 if (foldcase) {
2293 inlen = strlen(namep);
2294 outlen = PCMAXNAMLEN;
2295 outbuf = kmem_alloc(PCMAXNAMLEN + 1, KM_SLEEP);
2296 ret = u8_textprep_str(namep, &inlen, outbuf,
2297 &outlen, U8_TEXTPREP_TOLOWER, U8_UNICODE_LATEST,
2298 &error);
2299 if (ret == -1) {
2300 kmem_free(outbuf, PCMAXNAMLEN + 1);
2301 kmem_free(lfn_base, PCMAXNAM_UTF16);
2302 return (EINVAL);
2303 }
2304 outbuf[PCMAXNAMLEN - outlen] = '\0';
2305 (void) strncpy(namep, outbuf, PCMAXNAMLEN + 1);
2306 kmem_free(outbuf, PCMAXNAMLEN + 1);
2307 }
2308 }
2309 kmem_free(lfn_base, PCMAXNAM_UTF16);
2310 *epp = ep;
2311 return (0);
2312 }
2313 /*
2314 * Read a long filename into the pc_dirent structure and copy it out.
2315 */
2316 int
2317 pc_read_long_fn(struct vnode *dvp, struct uio *uiop, struct pc_dirent *ld,
2318 struct pcdir **epp, offset_t *offset, struct buf **bp)
2319 {
2320 struct pcdir *ep;
2321 struct pcnode *pcp = VTOPC(dvp);
2322 struct pcfs *fsp = VFSTOPCFS(dvp->v_vfsp);
2323 offset_t uiooffset = uiop->uio_loffset;
2324 int error = 0;
2325 offset_t oldoffset;
2326
2327 oldoffset = *offset;
2328 error = pc_extract_long_fn(pcp, ld->d_name, epp, offset, bp);
2329 if (error) {
2330 if (error == EINVAL) {
2331 uiop->uio_loffset += *offset - oldoffset;
2332 return (0);
2333 } else
2334 return (error);
2335 }
2336
2337 ep = *epp;
2338 uiop->uio_loffset += *offset - oldoffset;
2339 ld->d_reclen = DIRENT64_RECLEN(strlen(ld->d_name));
2340 if (ld->d_reclen > uiop->uio_resid) {
2341 uiop->uio_loffset = uiooffset;
2342 return (ENOSPC);
2343 }
2344 ld->d_off = uiop->uio_loffset + sizeof (struct pcdir);
2345 ld->d_ino = pc_makenodeid(pc_daddrdb(fsp, (*bp)->b_blkno),
2346 pc_blkoff(fsp, *offset), ep->pcd_attr,
2347 pc_getstartcluster(fsp, ep), pc_direntpersec(fsp));
2348 (void) uiomove((caddr_t)ld, ld->d_reclen, UIO_READ, uiop);
2349 uiop->uio_loffset = ld->d_off;
2350 *offset += sizeof (struct pcdir);
2351 ep++;
2352 *epp = ep;
2353 return (0);
2354 }
2355
2356 /*
2357 * Read a short filename into the pc_dirent structure and copy it out.
2358 */
2359 int
2360 pc_read_short_fn(struct vnode *dvp, struct uio *uiop, struct pc_dirent *ld,
2361 struct pcdir **epp, offset_t *offset, struct buf **bp)
2362 {
2363 struct pcfs *fsp = VFSTOPCFS(dvp->v_vfsp);
2364 int boff = pc_blkoff(fsp, *offset);
2365 struct pcdir *ep = *epp;
2366 offset_t oldoffset = uiop->uio_loffset;
2367 int error;
2368 int foldcase;
2369
2370 if (PCA_IS_HIDDEN(fsp, ep->pcd_attr)) {
2371 uiop->uio_loffset += sizeof (struct pcdir);
2372 *offset += sizeof (struct pcdir);
2373 ep++;
2374 *epp = ep;
2375 return (0);
2376 }
2377 ld->d_ino = (ino64_t)pc_makenodeid(pc_daddrdb(fsp, (*bp)->b_blkno),
2378 boff, ep->pcd_attr, pc_getstartcluster(fsp, ep),
2379 pc_direntpersec(fsp));
2380 foldcase = (fsp->pcfs_flags & PCFS_FOLDCASE);
2381 error = pc_fname_ext_to_name(&ld->d_name[0], &ep->pcd_filename[0],
2382 &ep->pcd_ext[0], foldcase);
2383 if (error == 0) {
2384 ld->d_reclen = DIRENT64_RECLEN(strlen(ld->d_name));
2385 if (ld->d_reclen > uiop->uio_resid) {
2386 uiop->uio_loffset = oldoffset;
2387 return (ENOSPC);
2388 }
2389 ld->d_off = (off64_t)(uiop->uio_loffset +
2390 sizeof (struct pcdir));
2391 (void) uiomove((caddr_t)ld,
2392 ld->d_reclen, UIO_READ, uiop);
2393 uiop->uio_loffset = ld->d_off;
2394 } else {
2395 uiop->uio_loffset += sizeof (struct pcdir);
2396 }
2397 *offset += sizeof (struct pcdir);
2398 ep++;
2399 *epp = ep;
2400 return (0);
2401 }
2402
2403 /* ARGSUSED */
2404 static int
2405 pcfs_fid(struct vnode *vp, struct fid *fidp, caller_context_t *ct)
2406 {
2407 struct pc_fid *pcfid;
2408 struct pcnode *pcp;
2409 struct pcfs *fsp;
2410 int error;
2411
2412 fsp = VFSTOPCFS(vp->v_vfsp);
2413 if (fsp == NULL)
2414 return (EIO);
2415 error = pc_lockfs(fsp, 0, 0);
2416 if (error)
2417 return (error);
2418 if ((pcp = VTOPC(vp)) == NULL || pcp->pc_flags & PC_INVAL) {
2419 pc_unlockfs(fsp);
2420 return (EIO);
2421 }
2422 if (fidp->fid_len < (sizeof (struct pc_fid) - sizeof (ushort_t))) {
2423 fidp->fid_len = sizeof (struct pc_fid) - sizeof (ushort_t);
2424 pc_unlockfs(fsp);
2425 return (ENOSPC);
2426 }
2427
2428 pcfid = (struct pc_fid *)fidp;
2429 bzero(pcfid, sizeof (struct pc_fid));
2430 pcfid->pcfid_len = sizeof (struct pc_fid) - sizeof (ushort_t);
2431 if (vp->v_flag & VROOT) {
2432 pcfid->pcfid_block = 0;
2433 pcfid->pcfid_offset = 0;
2434 pcfid->pcfid_ctime = 0;
2435 } else {
2436 pcfid->pcfid_block = pcp->pc_eblkno;
2437 pcfid->pcfid_offset = pcp->pc_eoffset;
2438 pcfid->pcfid_ctime = pcp->pc_entry.pcd_crtime.pct_time;
2439 }
2440 pc_unlockfs(fsp);
2441 return (0);
2442 }