xref: /illumos-gate/usr/src/uts/common/fs/udfs/udf_bmap.c (revision d48be21240dfd051b689384ce2b23479d757f2d8)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2004 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/types.h>
28 #include <sys/t_lock.h>
29 #include <sys/param.h>
30 #include <sys/time.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/resource.h>
34 #include <sys/signal.h>
35 #include <sys/cred.h>
36 #include <sys/user.h>
37 #include <sys/buf.h>
38 #include <sys/vfs.h>
39 #include <sys/stat.h>
40 #include <sys/vnode.h>
41 #include <sys/mode.h>
42 #include <sys/proc.h>
43 #include <sys/disp.h>
44 #include <sys/file.h>
45 #include <sys/fcntl.h>
46 #include <sys/flock.h>
47 #include <sys/kmem.h>
48 #include <sys/uio.h>
49 #include <sys/dnlc.h>
50 #include <sys/conf.h>
51 #include <sys/errno.h>
52 #include <sys/mman.h>
53 #include <sys/fbuf.h>
54 #include <sys/pathname.h>
55 #include <sys/debug.h>
56 #include <sys/vmsystm.h>
57 #include <sys/cmn_err.h>
58 #include <sys/dirent.h>
59 #include <sys/errno.h>
60 #include <sys/modctl.h>
61 #include <sys/statvfs.h>
62 #include <sys/mount.h>
63 #include <sys/sunddi.h>
64 #include <sys/bootconf.h>
65 
66 #include <vm/hat.h>
67 #include <vm/page.h>
68 #include <vm/pvn.h>
69 #include <vm/as.h>
70 #include <vm/seg.h>
71 #include <vm/seg_map.h>
72 #include <vm/seg_kmem.h>
73 #include <vm/seg_vn.h>
74 #include <vm/rm.h>
75 #include <vm/page.h>
76 #include <sys/swap.h>
77 
78 
79 #include <fs/fs_subr.h>
80 
81 
82 #include <sys/fs/udf_volume.h>
83 #include <sys/fs/udf_inode.h>
84 
85 
86 int32_t ud_break_create_new_icb(struct ud_inode *, int32_t, uint32_t);
87 int32_t ud_bump_ext_count(struct ud_inode *, int32_t);
88 void ud_remove_ext_at_index(struct ud_inode *, int32_t);
89 int32_t ud_last_alloc_ext(struct ud_inode *, uint64_t, uint32_t, int32_t);
90 int32_t ud_create_ext(struct ud_inode *, int32_t, uint32_t,
91 	int32_t, uint64_t, uint64_t *);
92 int32_t	ud_zero_it(struct ud_inode *, uint32_t, uint32_t);
93 
94 #define	ALLOC_SPACE	0x01
95 #define	NEW_EXT		0x02
96 
97 #define	MEXT_BITS	30
98 
99 int32_t
100 ud_bmap_has_holes(struct ud_inode *ip)
101 {
102 	int32_t i, error = 0;
103 	struct icb_ext *iext;
104 
105 	ud_printf("ud_bmap_has_holes\n");
106 
107 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
108 
109 	/* ICB_FLAG_ONE_AD is always continuos */
110 	if (ip->i_desc_type != ICB_FLAG_ONE_AD) {
111 		if ((error = ud_read_icb_till_off(ip, ip->i_size)) == 0) {
112 			for (i = 0; i < ip->i_ext_used; i++) {
113 				iext = &ip->i_ext[i];
114 				if (iext->ib_flags == IB_UN_RE_AL) {
115 					error = 1;
116 					break;
117 				}
118 			}
119 		}
120 	}
121 
122 	return (error);
123 }
124 
125 int32_t
126 ud_bmap_read(struct ud_inode *ip, u_offset_t off, daddr_t *bnp, int32_t *lenp)
127 {
128 	struct icb_ext *iext;
129 	daddr_t bno;
130 	int32_t lbmask, i, l2b, l2d, error = 0, count;
131 	uint32_t length, block, dummy;
132 
133 	ud_printf("ud_bmap_read\n");
134 
135 	ASSERT(RW_LOCK_HELD(&ip->i_contents));
136 
137 	lbmask = ip->i_udf->udf_lbmask;
138 	l2b = ip->i_udf->udf_l2b_shift;
139 	l2d = ip->i_udf->udf_l2d_shift;
140 
141 	if ((error = ud_read_icb_till_off(ip, ip->i_size)) == 0) {
142 		for (i = 0; i < ip->i_ext_used; i++) {
143 			iext = &ip->i_ext[i];
144 			if ((iext->ib_offset <= off) &&
145 				(off < (iext->ib_offset + iext->ib_count))) {
146 				length = ((iext->ib_offset +
147 						iext->ib_count - off) +
148 						lbmask) & ~lbmask;
149 				if (iext->ib_flags == IB_UN_RE_AL) {
150 					*bnp = UDF_HOLE;
151 					*lenp = length;
152 					break;
153 				}
154 
155 				block = iext->ib_block +
156 					((off - iext->ib_offset) >> l2b);
157 				count = length >> l2b;
158 
159 				bno = ud_xlate_to_daddr(ip->i_udf,
160 					iext->ib_prn, block, count, &dummy);
161 				ASSERT(dummy != 0);
162 				ASSERT(dummy <= count);
163 				*bnp = bno << l2d;
164 				*lenp = dummy << l2b;
165 
166 				break;
167 			}
168 		}
169 		if (i == ip->i_ext_used) {
170 			error = EINVAL;
171 		}
172 	}
173 
174 	return (error);
175 }
176 
177 
178 /*
179  * Extent allocation in the inode
180  * Initially when the inode is allocated we
181  * will allocate EXT_PER_MALLOC extents and once these
182  * are used we allocate another 10 and copy
183  * the old extents and start using the others
184  */
185 #define	BASE(count)	((count) & ~lbmask)
186 #define	CEIL(count)	(((count) + lbmask) & ~lbmask)
187 
188 #define	PBASE(count)	((count) & PAGEMASK)
189 #define	PCEIL(count)	(((count) + PAGEOFFSET) & PAGEMASK)
190 
191 
192 /* ARGSUSED3 */
193 int32_t
194 ud_bmap_write(struct ud_inode *ip,
195 	u_offset_t off, int32_t size, int32_t alloc_only, struct cred *cr)
196 {
197 	int32_t error = 0, i, isdir, issync;
198 	struct udf_vfs *udf_vfsp;
199 	struct icb_ext *iext, *pext;
200 	uint32_t blkno, sz;
201 	u_offset_t isize;
202 	uint32_t acount, prox;
203 	int32_t blkcount, next;
204 	int32_t lbmask, l2b;
205 	uint64_t end_req, end_ext, mext_sz, icb_offset, count;
206 	int32_t dtype_changed = 0, memory_allocated = 0;
207 	struct	fbuf *fbp = NULL;
208 
209 
210 	ud_printf("ud_bmap_write\n");
211 
212 	ASSERT(RW_WRITE_HELD(&ip->i_contents));
213 
214 	udf_vfsp = ip->i_udf;
215 	lbmask = udf_vfsp->udf_lbmask;
216 	l2b = udf_vfsp->udf_l2b_shift;
217 	mext_sz = (1 << MEXT_BITS) - PAGESIZE;
218 
219 	if (lblkno(udf_vfsp, off) < 0) {
220 		return (EFBIG);
221 	}
222 
223 	issync = ((ip->i_flag & ISYNC) != 0);
224 
225 	isdir = (ip->i_type == VDIR);
226 	if (isdir || issync) {
227 		alloc_only = 0;		/* make sure */
228 	}
229 
230 	end_req = BASE(off) + size;
231 	if (ip->i_desc_type == ICB_FLAG_ONE_AD) {
232 		if (end_req < ip->i_max_emb) {
233 			goto out;
234 		}
235 
236 		if (ip->i_size != 0) {
237 			error = fbread(ITOV(ip), 0, ip->i_size, S_OTHER, &fbp);
238 			if (error != 0) {
239 				goto out;
240 			}
241 		} else {
242 			fbp = NULL;
243 		}
244 		/*
245 		 * Change the desc_type
246 		 */
247 		ip->i_desc_type = ICB_FLAG_SHORT_AD;
248 		dtype_changed = 1;
249 
250 one_ad_no_i_ext:
251 		ASSERT(ip->i_ext == NULL);
252 		ASSERT(ip->i_astrat == STRAT_TYPE4);
253 
254 		ip->i_ext_used = 0;
255 		ip->i_cur_max_ext = ip->i_max_emb / sizeof (struct short_ad);
256 		ip->i_cur_max_ext --;
257 		if (end_req > mext_sz) {
258 			next = end_req / mext_sz;
259 		} else {
260 			next = 1;
261 		}
262 		ip->i_ext_count =
263 			((next / EXT_PER_MALLOC) + 1) * EXT_PER_MALLOC;
264 		iext = ip->i_ext = (struct icb_ext  *)kmem_zalloc(
265 			ip->i_ext_count * sizeof (struct icb_ext), KM_SLEEP);
266 		memory_allocated = 1;
267 
268 		/* There will be atleast EXT_PER_MALLOC icb_ext's allocated */
269 
270 one_ad_i_ext:
271 		icb_offset = 0;
272 		count = end_req;
273 
274 		/* Can we create a HOLE */
275 
276 		if ((PCEIL(ip->i_size) < PBASE(off)) &&
277 			((PBASE(off) - PCEIL(ip->i_size)) >= PAGESIZE)) {
278 
279 			if (ip->i_size != 0) {
280 
281 				/*
282 				 * Allocate one block for
283 				 * old data.(cannot be more than one page)
284 				 */
285 
286 				count = PAGESIZE;
287 				if (error = ud_create_ext(ip, ip->i_ext_used,
288 					ALLOC_SPACE | NEW_EXT, alloc_only,
289 					icb_offset, &count)) {
290 					goto embedded_error;
291 				}
292 				icb_offset = PAGESIZE;
293 			}
294 
295 			/*
296 			 * Allocate a hole from PCEIL(ip->i_size) to PBASE(off)
297 			 */
298 
299 			count = PBASE(off) - PCEIL(ip->i_size);
300 			(void) ud_create_ext(ip, ip->i_ext_used, NEW_EXT,
301 					alloc_only, icb_offset, &count);
302 			icb_offset = PBASE(off);
303 
304 			/*
305 			 * Allocate the rest of the space PBASE(off) to end_req
306 			 */
307 			count = end_req - PBASE(off);
308 		} else {
309 			/*
310 			 * If no hole can be created then allocate
311 			 * space till the end of the request
312 			 */
313 			count = end_req;
314 		}
315 
316 
317 
318 		if (error = ud_create_ext(ip, ip->i_ext_used,
319 				ALLOC_SPACE | NEW_EXT,
320 				alloc_only, icb_offset, &count)) {
321 embedded_error:
322 			/*
323 			 * Something error
324 			 * most probable file system is full
325 			 * we know that the file came in as a embedded file.
326 			 * undo what ever we did in this block of code
327 			 */
328 			if (dtype_changed) {
329 				ip->i_desc_type = ICB_FLAG_ONE_AD;
330 			}
331 			for (i = 0; i < ip->i_ext_used; i++) {
332 				iext = &ip->i_ext[i];
333 				if (iext->ib_flags != IB_UN_RE_AL) {
334 					ud_free_space(ip->i_udf->udf_vfs,
335 						iext->ib_prn, iext->ib_block,
336 						(iext->ib_count + lbmask) >>
337 							l2b);
338 				}
339 			}
340 			if (memory_allocated) {
341 				kmem_free(ip->i_ext,
342 					ip->i_ext_count *
343 					sizeof (struct icb_ext));
344 				ip->i_ext = NULL;
345 				ip->i_ext_count = ip->i_ext_used = 0;
346 			}
347 		}
348 
349 		if (fbp != NULL) {
350 			fbrelse(fbp, S_WRITE);
351 		}
352 
353 		return (error);
354 	} else {
355 
356 		/*
357 		 * Type 4 directories being created
358 		 */
359 		if (ip->i_ext == NULL) {
360 			goto one_ad_no_i_ext;
361 		}
362 
363 		/*
364 		 * Read the entire icb's to memory
365 		 */
366 		if (ud_read_icb_till_off(ip, ip->i_size) != 0) {
367 			error = EINVAL;
368 			goto out;
369 		}
370 
371 		isize = CEIL(ip->i_size);
372 
373 		if (end_req > isize) {
374 
375 			/*
376 			 * The new file size is greater
377 			 * than the old size
378 			 */
379 
380 			if (ip->i_ext == NULL) {
381 				goto one_ad_no_i_ext;
382 			} else if (ip->i_ext_used == 0) {
383 				goto one_ad_i_ext;
384 			}
385 
386 			error = ud_last_alloc_ext(ip, off, size, alloc_only);
387 
388 			return (error);
389 		} else {
390 
391 			/*
392 			 * File growing the new size will be less than
393 			 * iext->ib_offset + CEIL(iext->ib_count)
394 			 */
395 
396 			iext = &ip->i_ext[ip->i_ext_used - 1];
397 
398 			if (end_req > (iext->ib_offset + iext->ib_count)) {
399 
400 				iext->ib_count = end_req - iext->ib_offset;
401 
402 				if (iext->ib_flags != IB_UN_RE_AL) {
403 					error = 0;
404 					goto out;
405 				}
406 			}
407 		}
408 	}
409 
410 	/* By this point the end of last extent is >= BASE(off) + size */
411 
412 	ASSERT(ip->i_ext);
413 
414 	/*
415 	 * Figure out the icb_ext that has offset "off"
416 	 */
417 	for (i = 0; i < ip->i_ext_used; i++) {
418 		iext = &ip->i_ext[i];
419 		if ((iext->ib_offset <= off) &&
420 			((iext->ib_offset + iext->ib_count) > off)) {
421 			break;
422 		}
423 	}
424 
425 	/*
426 	 * iext will have offset "off"
427 	 */
428 
429 
430 	do {
431 		iext = &ip->i_ext[i];
432 
433 		if ((iext->ib_flags & IB_UN_RE_AL) == 0) {
434 
435 			/*
436 			 * Already allocated do nothing
437 			 */
438 
439 			i++;
440 		} else {
441 
442 			/*
443 			 * We are in a hole.
444 			 * allocate the required space
445 			 * while trying to create smaller holes
446 			 */
447 
448 			if ((PBASE(off) > PBASE(iext->ib_offset)) &&
449 				((PBASE(off) - PBASE(iext->ib_offset)) >=
450 						PAGESIZE)) {
451 
452 				/*
453 				 * Allocate space from begining of
454 				 * old hole to the begining of new hole
455 				 * We want all holes created by us
456 				 * to be MMUPAGE Aligned
457 				 */
458 
459 				if (PBASE(iext->ib_offset) !=
460 						BASE(iext->ib_offset)) {
461 					if ((error = ud_break_create_new_icb(
462 						ip, i, BASE(iext->ib_offset) -
463 						PBASE(iext->ib_offset))) != 0) {
464 						return (error);
465 					}
466 					goto alloc_cur_ext;
467 				}
468 
469 				/*
470 				 * Create the new hole
471 				 */
472 
473 				if ((error = ud_break_create_new_icb(ip, i,
474 					PBASE(off) - iext->ib_offset)) != 0) {
475 					return (error);
476 				}
477 				iext = &ip->i_ext[i];
478 				i++;
479 				continue;
480 			}
481 
482 			end_ext = iext->ib_offset + iext->ib_count;
483 
484 			if ((PBASE(end_ext) > PCEIL(end_req)) &&
485 				((PBASE(end_ext) - PCEIL(end_req)) >=
486 							PAGESIZE)) {
487 				/*
488 				 * We can create a hole
489 				 * from PCEIL(end_req) - BASE(end_ext)
490 				 */
491 				if ((error = ud_break_create_new_icb(ip, i,
492 				PCEIL(end_req) - iext->ib_offset)) != 0) {
493 					return (error);
494 				}
495 			}
496 
497 
498 alloc_cur_ext:
499 			/*
500 			 * Allocate the current extent
501 			 */
502 
503 
504 			/*
505 			 * If the previous extent
506 			 * is allocated then try to allocate
507 			 * adjascent to the previous extent
508 			 */
509 			prox = 0;
510 			if (i != 0) {
511 				pext = &ip->i_ext[i - 1];
512 				if (pext->ib_flags != IB_UN_RE_AL) {
513 					prox = pext->ib_block +
514 						(CEIL(pext->ib_count) >> l2b);
515 				}
516 			}
517 
518 			iext = &ip->i_ext[i];
519 			blkcount = CEIL(iext->ib_count) >> l2b;
520 
521 			if ((error = ud_alloc_space(ip->i_vfs,
522 					ip->i_icb_prn, prox, blkcount,
523 					&blkno, &sz, 1, 0)) != 0) {
524 				return (error);
525 			}
526 			ip->i_lbr += sz;
527 			if (sz == 0) {
528 				return (ENOSPC);
529 			}
530 
531 			if (alloc_only == 0) {
532 				error = ud_zero_it(ip, blkno, sz);
533 			}
534 
535 			acount = sz << l2b;
536 			if ((prox == blkno) &&
537 				((pext->ib_count + acount) < mext_sz)) {
538 
539 				/*
540 				 * We are able to allocate adjascent to
541 				 * the previous extent. Increment the
542 				 * previous extent count if the size
543 				 * of the extent is not greater than
544 				 * max extent size
545 				 */
546 
547 				pext = &ip->i_ext[i - 1];
548 				pext->ib_count += acount;
549 
550 				if (sz == blkcount) {
551 					/*
552 					 * and get rid of the current
553 					 * extent since we have
554 					 * allocated all of its size
555 					 * and incremented the
556 					 * previous extents count
557 					 */
558 					ud_remove_ext_at_index(ip, i);
559 				} else {
560 					/*
561 					 * reduce the count of the
562 					 * current extent by the amount
563 					 * allocated in the last extent
564 					 */
565 					ASSERT(acount < iext->ib_count);
566 					iext->ib_count -= acount;
567 					iext->ib_offset += acount;
568 				}
569 			} else {
570 				if (sz < blkcount) {
571 					if ((error = ud_break_create_new_icb(
572 						ip, i, sz << l2b)) != 0) {
573 						return (error);
574 					}
575 				}
576 				iext = &ip->i_ext[i];
577 				count -= CEIL(iext->ib_count);
578 				iext->ib_prn = ip->i_icb_prn;
579 				iext->ib_block = blkno;
580 				iext->ib_flags &= ~IB_UN_RE_AL;
581 /*
582  *				iext->ib_flags |= IB_UN_REC;
583  */
584 				i++;
585 				continue;
586 			}
587 		}
588 	} while ((iext->ib_offset + iext->ib_count) < end_req);
589 
590 out:
591 	return (error);
592 }
593 
594 
595 /*
596  * increase i_con/i_ext arrays and set new elements
597  * using long or short allocation descriptors
598  */
599 static void
600 ud_common_ad(struct ud_inode *ip, struct buf *bp)
601 {
602 	int32_t ndesc, count, lbmask;
603 	uint32_t length;
604 	struct alloc_ext_desc *aed;
605 	struct icb_ext *iext, *con;
606 	u_offset_t offset;
607 	long_ad_t *lad;
608 	short_ad_t *sad;
609 	int islong;
610 	void *addr;
611 
612 	addr = bp->b_un.b_addr + sizeof (struct alloc_ext_desc);
613 	aed = (struct alloc_ext_desc *)bp->b_un.b_addr;
614 	length = SWAP_32(aed->aed_len_aed);
615 	if (ip->i_desc_type == ICB_FLAG_LONG_AD) {
616 		islong = 1;
617 		lad = addr;
618 		ndesc = length / sizeof (*lad);
619 	} else if (ip->i_desc_type == ICB_FLAG_SHORT_AD) {
620 		islong = 0;
621 		sad = addr;
622 		ndesc = length / sizeof (*sad);
623 	} else
624 		return;
625 
626 	/*
627 	 * realloc i_ext array
628 	 */
629 	count = (((ip->i_ext_used + ndesc) / EXT_PER_MALLOC) + 1) *
630 	    EXT_PER_MALLOC;
631 	addr = kmem_zalloc(count * sizeof (struct icb_ext), KM_SLEEP);
632 	bcopy(ip->i_ext, addr, ip->i_ext_used * sizeof (struct icb_ext));
633 	kmem_free(ip->i_ext, ip->i_ext_count * sizeof (struct icb_ext));
634 	ip->i_ext = addr;
635 	ip->i_ext_count = count;
636 
637 	/*
638 	 * scan descriptors
639 	 */
640 	lbmask = ip->i_udf->udf_lbmask;
641 	iext = &ip->i_ext[ip->i_ext_used - 1];
642 	offset = iext->ib_offset + iext->ib_count;
643 	iext++;
644 	while (ndesc--) {
645 		if (islong)
646 			length = SWAP_32(lad->lad_ext_len);
647 		else
648 			length = SWAP_32(sad->sad_ext_len);
649 
650 		if ((length & 0x3FFFFFFF) == 0)
651 			break;
652 		else if (((length >> 30) & IB_MASK) == IB_CON) {
653 			if (ip->i_con_used == ip->i_con_count) {
654 				struct icb_ext *old;
655 				int32_t old_count;
656 
657 				old = ip->i_con;
658 				old_count = ip->i_con_count *
659 				    sizeof (struct icb_ext);
660 				ip->i_con_count += EXT_PER_MALLOC;
661 				ip->i_con = kmem_zalloc(ip->i_con_count *
662 				    sizeof (struct icb_ext), KM_SLEEP);
663 
664 				if (old) {
665 					bcopy(old, ip->i_con, old_count);
666 					kmem_free(old, old_count);
667 				}
668 			}
669 			con = &ip->i_con[ip->i_con_used];
670 			if (islong) {
671 				con->ib_prn = SWAP_16(lad->lad_ext_prn);
672 				con->ib_block = SWAP_32(lad->lad_ext_loc);
673 			} else {
674 				con->ib_prn = ip->i_icb_prn;
675 				con->ib_block = SWAP_32(sad->sad_ext_loc);
676 			}
677 			con->ib_count = length & 0x3FFFFFFF;
678 			con->ib_flags = (length >> 30) & IB_MASK;
679 			ip->i_con_used++;
680 			break;
681 		}
682 
683 		if (islong) {
684 			iext->ib_prn = SWAP_16(lad->lad_ext_prn);
685 			iext->ib_block = SWAP_32(lad->lad_ext_loc);
686 			lad++;
687 		} else {
688 			iext->ib_prn = 0;
689 			iext->ib_block = SWAP_32(sad->sad_ext_loc);
690 			sad++;
691 		}
692 		iext->ib_count = length & 0x3FFFFFFF;
693 		iext->ib_offset = offset;
694 		iext->ib_marker1 = (uint32_t)0xAAAAAAAA;
695 		iext->ib_marker2 = (uint32_t)0xBBBBBBBB;
696 		offset += (iext->ib_count + lbmask) & (~lbmask);
697 		iext->ib_flags = (length >> 30) & IB_MASK;
698 		ip->i_ext_used++;
699 		iext++;
700 	}
701 }
702 
703 
704 static int32_t
705 ud_read_next_cont(struct ud_inode *ip)
706 {
707 	uint32_t dummy, error = 0;
708 	struct alloc_ext_desc *aed;
709 	struct icb_ext *cont;
710 	struct buf *bp;
711 	daddr_t bno;
712 
713 	cont = &ip->i_con[ip->i_con_read];
714 	ASSERT(cont->ib_count > 0);
715 
716 	bno = ud_xlate_to_daddr(ip->i_udf, cont->ib_prn, cont->ib_block,
717 	    1, &dummy);
718 	bp = ud_bread(ip->i_dev, bno << ip->i_udf->udf_l2d_shift,
719 	    cont->ib_count);
720 	if (bp->b_flags & B_ERROR)
721 		error = bp->b_error;
722 	else {
723 		aed = (struct alloc_ext_desc *)bp->b_un.b_addr;
724 		if (ud_verify_tag_and_desc(&aed->aed_tag, UD_ALLOC_EXT_DESC,
725 		    cont->ib_block, 1, cont->ib_count))
726 			error = EINVAL;
727 	}
728 
729 	if (error == 0)
730 		ud_common_ad(ip, bp);
731 
732 	brelse(bp);
733 	return (error);
734 }
735 
736 
737 int32_t
738 ud_read_icb_till_off(struct ud_inode *ip, u_offset_t offset)
739 {
740 	int32_t error = 0;
741 	struct icb_ext *iext;
742 
743 	ud_printf("ud_read_icb_till_off\n");
744 
745 	if (ip->i_desc_type == ICB_FLAG_ONE_AD)
746 		return (0);
747 	else if ((ip->i_astrat != STRAT_TYPE4) &&
748 	    (ip->i_astrat != STRAT_TYPE4096))
749 		return (EINVAL);
750 	else if (ip->i_ext_used == 0)
751 		return ((ip->i_size == 0) ? 0 : EINVAL);
752 
753 	/*
754 	 * supported allocation strategies are
755 	 * STRAT_TYPE4 and STRAT_TYPE4096
756 	 */
757 
758 	mutex_enter(&ip->i_con_lock);
759 	iext = &ip->i_ext[ip->i_ext_used - 1];
760 	while ((iext->ib_offset + iext->ib_count) < offset) {
761 		if (ip->i_con_used == ip->i_con_read) {
762 			error = EINVAL;
763 			break;
764 		}
765 		if (error = ud_read_next_cont(ip))
766 			break;
767 		ip->i_con_read++;
768 		iext = &ip->i_ext[ip->i_ext_used - 1];
769 	}
770 	mutex_exit(&ip->i_con_lock);
771 
772 	return (error);
773 }
774 
775 
776 /*
777  * Assumption is the off is beyond ip->i_size
778  * And we will have atleast one ext used
779  */
780 int32_t
781 ud_last_alloc_ext(struct ud_inode *ip, uint64_t off,
782 		uint32_t size, int32_t alloc_only)
783 {
784 	struct icb_ext *iext;
785 	struct udf_vfs *udf_vfsp;
786 	int32_t lbsize, lbmask;
787 	uint64_t end_req, end_count, icb_offset;
788 	uint64_t count;
789 	int32_t error = 0;
790 
791 
792 	udf_vfsp = ip->i_udf;
793 	lbsize = udf_vfsp->udf_lbsize;
794 	lbmask = udf_vfsp->udf_lbmask;
795 
796 	end_req = BASE(off) + size;
797 
798 
799 	/*
800 	 * If we are here it means the file
801 	 * is growing beyond the end of the
802 	 * current block. So round up the
803 	 * last extent
804 	 */
805 
806 	iext = &ip->i_ext[ip->i_ext_used - 1];
807 	iext->ib_count = CEIL(iext->ib_count);
808 
809 	/*
810 	 * Figure out if we can create
811 	 * a hole here
812 	 */
813 
814 
815 	end_count = iext->ib_offset + iext->ib_count;
816 
817 	if ((PCEIL(end_count) < PBASE(off)) &&
818 		((PBASE(off) - PCEIL(end_count)) >= PAGESIZE)) {
819 
820 		count = PCEIL(end_count) - CEIL(end_count);
821 		if (count >= lbsize) {
822 
823 			/*
824 			 * There is space between the begining
825 			 * of the hole to be created and
826 			 * end of the last offset
827 			 * Allocate blocks for it
828 			 */
829 
830 			iext = &ip->i_ext[ip->i_ext_used - 1];
831 			icb_offset = iext->ib_offset + CEIL(iext->ib_count);
832 
833 			if (iext->ib_flags == IB_UN_RE_AL) {
834 
835 				/*
836 				 * Previous extent is a unallocated
837 				 * extent. Create a new allocated
838 				 * extent
839 				 */
840 
841 				error = ud_create_ext(ip, ip->i_ext_used,
842 					ALLOC_SPACE | NEW_EXT,
843 					alloc_only, icb_offset, &count);
844 
845 			} else {
846 
847 				/*
848 				 * Last extent is allocated
849 				 * try to allocate adjascent to the
850 				 * last extent
851 				 */
852 
853 				error = ud_create_ext(ip, ip->i_ext_used - 1,
854 						ALLOC_SPACE, alloc_only,
855 						icb_offset, &count);
856 			}
857 
858 			if (error != 0) {
859 				return (error);
860 			}
861 		}
862 
863 		iext = &ip->i_ext[ip->i_ext_used - 1];
864 		end_count = iext->ib_offset + iext->ib_count;
865 		count = PBASE(off) - PCEIL(end_count);
866 		icb_offset = PCEIL(end_count);
867 
868 		if (iext->ib_flags == IB_UN_RE_AL) {
869 
870 			/*
871 			 * The last extent is unallocated
872 			 * Just bump the extent count
873 			 */
874 			(void) ud_create_ext(ip, ip->i_ext_used - 1,
875 					0, alloc_only, icb_offset, &count);
876 		} else {
877 
878 			/*
879 			 * Last extent is allocated
880 			 * round up the size of the extent to
881 			 * lbsize and allocate a new unallocated extent
882 			 */
883 			iext->ib_count = CEIL(iext->ib_count);
884 			(void) ud_create_ext(ip, ip->i_ext_used,
885 				NEW_EXT, alloc_only, icb_offset, &count);
886 		}
887 
888 		icb_offset = PBASE(off);
889 	} else {
890 
891 		/*
892 		 * We cannot create any hole inbetween
893 		 * the last extent and the off so
894 		 * round up the count in the last extent
895 		 */
896 
897 		iext = &ip->i_ext[ip->i_ext_used - 1];
898 		iext->ib_count = CEIL(iext->ib_count);
899 
900 	}
901 
902 
903 	iext = &ip->i_ext[ip->i_ext_used - 1];
904 	count = end_req - (iext->ib_offset + iext->ib_count);
905 	icb_offset = iext->ib_offset + CEIL(iext->ib_count);
906 
907 	if (iext->ib_flags == IB_UN_RE_AL) {
908 
909 		/*
910 		 * Last extent was a unallocated extent
911 		 * create a new extent
912 		 */
913 
914 		error = ud_create_ext(ip, ip->i_ext_used,
915 			ALLOC_SPACE | NEW_EXT, alloc_only, icb_offset, &count);
916 	} else {
917 
918 		/*
919 		 * Last extent was an allocated extent
920 		 * try to allocate adjascent to the old blocks
921 		 */
922 
923 		error = ud_create_ext(ip, ip->i_ext_used - 1,
924 			ALLOC_SPACE, alloc_only, icb_offset, &count);
925 	}
926 
927 	return (error);
928 }
929 
930 /*
931  * Break up the icb_ext at index
932  * into two icb_ext,
933  * one at index ib_count "count" and
934  * the other at index+1 with ib_count = old_ib_count - count
935  */
936 int32_t
937 ud_break_create_new_icb(struct ud_inode *ip,
938 	int32_t index, uint32_t count)
939 {
940 	int32_t i, error;
941 	struct icb_ext *iext, *next;
942 
943 
944 	ud_printf("ud_break_create_new_icb\n");
945 	iext = &ip->i_ext[index];
946 
947 	ASSERT(count < iext->ib_count);
948 
949 	if ((error = ud_bump_ext_count(ip, KM_SLEEP)) != 0) {
950 		return (error);
951 	}
952 
953 	for (i = ip->i_ext_used; i > index; i--) {
954 		ip->i_ext[i] = ip->i_ext[i - 1];
955 	}
956 
957 	next = &ip->i_ext[index + 1];
958 	iext = &ip->i_ext[index];
959 
960 	iext->ib_count = count;
961 	next->ib_count -= count;
962 	next->ib_offset = iext->ib_offset + iext->ib_count;
963 	if (iext->ib_flags != IB_UN_RE_AL) {
964 		next->ib_block = iext->ib_block +
965 			iext->ib_count >> ip->i_udf->udf_l2b_shift;
966 	}
967 	ip->i_ext_used++;
968 	return (0);
969 }
970 
971 void
972 ud_remove_ext_at_index(struct ud_inode *ip, int32_t index)
973 {
974 	int32_t i;
975 
976 	ASSERT(index <= ip->i_ext_used);
977 
978 	for (i = index; i < ip->i_ext_used; i++) {
979 		if ((i + 1) < ip->i_ext_count) {
980 			ip->i_ext[i] = ip->i_ext[i + 1];
981 		} else {
982 			bzero(&ip->i_ext[i], sizeof (struct icb_ext));
983 		}
984 	}
985 	ip->i_ext_used --;
986 }
987 
988 int32_t
989 ud_bump_ext_count(struct ud_inode *ip, int32_t sleep_flag)
990 {
991 	int32_t error = 0;
992 	struct icb_ext *iext;
993 	uint32_t old_count, elen;
994 
995 	ASSERT(ip);
996 	ASSERT(sleep_flag == KM_SLEEP);
997 
998 	ud_printf("ud_bump_ext_count\n");
999 
1000 	if (ip->i_ext_used >= ip->i_ext_count) {
1001 
1002 		old_count = sizeof (struct icb_ext) * ip->i_ext_count;
1003 		ip->i_ext_count += EXT_PER_MALLOC;
1004 		iext = kmem_zalloc(sizeof (struct icb_ext) *
1005 				ip->i_ext_count, sleep_flag);
1006 		bcopy(ip->i_ext, iext, old_count);
1007 		kmem_free(ip->i_ext, old_count);
1008 		ip->i_ext = iext;
1009 	}
1010 
1011 	if (ip->i_ext_used >= ip->i_cur_max_ext) {
1012 		int32_t prox;
1013 		struct icb_ext *icon;
1014 		uint32_t blkno, sz;
1015 		int32_t lbmask, l2b;
1016 
1017 		lbmask = ip->i_udf->udf_lbmask;
1018 		l2b = ip->i_udf->udf_l2b_shift;
1019 
1020 		if ((error = ud_read_icb_till_off(ip, ip->i_size)) != 0) {
1021 			return (error);
1022 		}
1023 
1024 		/*
1025 		 * If there are any old cont extents
1026 		 * allocate the new one ajscant to the old one
1027 		 */
1028 		if (ip->i_con_used != 0) {
1029 			icon = &ip->i_con[ip->i_con_used - 1];
1030 			prox = icon->ib_block + (CEIL(icon->ib_count) >> l2b);
1031 		} else {
1032 			prox = 0;
1033 		}
1034 
1035 		/*
1036 		 * Allocate space
1037 		 */
1038 		if ((error = ud_alloc_space(ip->i_vfs, ip->i_icb_prn,
1039 				prox, 1, &blkno, &sz, 0, 0)) != 0) {
1040 			return (error);
1041 		}
1042 		if (sz == 0) {
1043 			return (ENOSPC);
1044 		}
1045 
1046 		sz <<= l2b;
1047 
1048 		if (ip->i_con_used == ip->i_con_count) {
1049 			struct icb_ext *old;
1050 			int32_t old_count;
1051 
1052 			old = ip->i_con;
1053 			old_count = ip->i_con_count *
1054 				sizeof (struct icb_ext);
1055 			ip->i_con_count += EXT_PER_MALLOC;
1056 			ip->i_con = kmem_zalloc(ip->i_con_count *
1057 				sizeof (struct icb_ext), KM_SLEEP);
1058 			if (old != 0) {
1059 				bcopy(old, ip->i_con, old_count);
1060 				kmem_free(old, old_count);
1061 			}
1062 		}
1063 		icon = &ip->i_con[ip->i_con_used++];
1064 		icon->ib_flags = IB_CON;
1065 		icon->ib_prn = ip->i_icb_prn;
1066 		icon->ib_block = blkno;
1067 		icon->ib_count = sz;
1068 		icon->ib_offset = 0;
1069 		icon->ib_marker1 = (uint32_t)0xAAAAAAAA;
1070 		icon->ib_marker2 = (uint32_t)0xBBBBBBBB;
1071 
1072 		/*
1073 		 * Bump the i_cur_max_ext according to
1074 		 * the space allocated
1075 		 */
1076 		if (ip->i_desc_type == ICB_FLAG_SHORT_AD) {
1077 			elen = sizeof (struct short_ad);
1078 		} else if (ip->i_desc_type == ICB_FLAG_LONG_AD) {
1079 			elen = sizeof (struct long_ad);
1080 		} else {
1081 			return (ENOSPC);
1082 		}
1083 		sz = sz - (sizeof (struct alloc_ext_desc) + elen);
1084 		ip->i_cur_max_ext += sz / elen;
1085 	}
1086 	return (error);
1087 }
1088 
1089 int32_t
1090 ud_create_ext(struct ud_inode *ip, int32_t index, uint32_t flags,
1091 	int32_t alloc_only, uint64_t offset, uint64_t *count)
1092 {
1093 	struct icb_ext *iext, *pext;
1094 	struct udf_vfs *udf_vfsp;
1095 	int32_t error = 0, blkcount, acount;
1096 	uint32_t blkno, sz, prox, mext_sz;
1097 	int32_t lbmask, l2b;
1098 
1099 	if (*count == 0) {
1100 		return (0);
1101 	}
1102 
1103 begin:
1104 	udf_vfsp = ip->i_udf;
1105 	lbmask = udf_vfsp->udf_lbmask;
1106 	l2b = udf_vfsp->udf_l2b_shift;
1107 	mext_sz = (1 << MEXT_BITS) - PAGESIZE;
1108 
1109 	if ((error = ud_bump_ext_count(ip, KM_SLEEP)) != 0) {
1110 		return (error);
1111 	}
1112 
1113 	iext = &ip->i_ext[index];
1114 	if (flags & ALLOC_SPACE) {
1115 		if ((flags & NEW_EXT) ||
1116 			(ip->i_ext_count == 0)) {
1117 
1118 			iext->ib_flags = 0;
1119 			iext->ib_prn = ip->i_icb_prn;
1120 			if (*count > mext_sz) {
1121 				blkcount = mext_sz >> l2b;
1122 			} else {
1123 				blkcount = CEIL(*count) >> l2b;
1124 			}
1125 			if ((error = ud_alloc_space(ip->i_vfs,
1126 					ip->i_icb_prn, 0, blkcount,
1127 					&blkno, &sz, 1, 0)) != 0) {
1128 				return (error);
1129 			}
1130 			if (sz == 0) {
1131 				return (ENOSPC);
1132 			}
1133 			ip->i_lbr += sz;
1134 			iext->ib_block = blkno;
1135 			acount = sz << l2b;
1136 			if ((sz << l2b) > *count) {
1137 				iext->ib_count = *count;
1138 				*count = 0;
1139 			} else {
1140 				iext->ib_count = sz << l2b;
1141 				*count -= iext->ib_count;
1142 			}
1143 			iext->ib_offset = offset;
1144 			if (ip->i_ext_used <= index)
1145 				ip->i_ext_used ++;
1146 		} else {
1147 			if ((iext->ib_count + *count) > mext_sz) {
1148 				blkcount = (mext_sz - iext->ib_count) >> l2b;
1149 			} else {
1150 				blkcount = CEIL(*count) >> l2b;
1151 			}
1152 			if (blkcount == 0) {
1153 				flags |= NEW_EXT;
1154 				index++;
1155 				goto begin;
1156 			}
1157 			prox = iext->ib_block + (CEIL(iext->ib_count) >> l2b);
1158 			if ((error = ud_alloc_space(ip->i_vfs,
1159 					ip->i_icb_prn, prox, blkcount,
1160 					&blkno, &sz, 1, 0)) != 0) {
1161 				return (error);
1162 			}
1163 			if (sz == 0) {
1164 				return (ENOSPC);
1165 			}
1166 			acount = sz << l2b;
1167 			if (acount > *count) {
1168 				acount = *count;
1169 				*count = 0;
1170 			} else {
1171 				*count -= acount;
1172 			}
1173 			ip->i_lbr += sz;
1174 			if (prox == blkno) {
1175 				iext->ib_count += acount;
1176 			} else {
1177 				if ((error = ud_bump_ext_count(ip, KM_SLEEP))
1178 						!= 0) {
1179 					return (error);
1180 				}
1181 				pext = &ip->i_ext[index];
1182 				iext = &ip->i_ext[index + 1];
1183 				iext->ib_flags = 0;
1184 				iext->ib_prn = ip->i_icb_prn;
1185 				iext->ib_block = blkno;
1186 				iext->ib_offset =
1187 					pext->ib_offset + pext->ib_count;
1188 				iext->ib_count = acount;
1189 				/*
1190 				 * Increment the index, since we have used
1191 				 * the extent at [index+1] above.
1192 				 */
1193 				index++;
1194 				if (ip->i_ext_used <= index)
1195 					ip->i_ext_used ++;
1196 			}
1197 		}
1198 		if (alloc_only == 0) {
1199 			error = ud_zero_it(ip, blkno, sz);
1200 		}
1201 		if (*count) {
1202 			offset = iext->ib_offset + CEIL(iext->ib_count);
1203 			flags |= NEW_EXT;
1204 			index++;
1205 			goto begin;
1206 		}
1207 	} else {
1208 		if (flags & NEW_EXT) {
1209 			iext->ib_flags = IB_UN_RE_AL;
1210 			iext->ib_prn = 0;
1211 			iext->ib_block = 0;
1212 			if (*count > mext_sz) {
1213 				iext->ib_count = mext_sz;
1214 				*count -= iext->ib_count;
1215 			} else {
1216 				iext->ib_count = *count;
1217 				*count = 0;
1218 			}
1219 			iext->ib_offset = offset;
1220 			if (ip->i_ext_used <= index)
1221 				ip->i_ext_used ++;
1222 		} else {
1223 			ASSERT(iext->ib_flags == IB_UN_RE_AL);
1224 			if ((iext->ib_count + *count) > mext_sz) {
1225 				acount = mext_sz - iext->ib_count;
1226 				iext->ib_count += acount;
1227 				*count -= acount;
1228 			} else {
1229 				iext->ib_count += *count;
1230 				*count = 0;
1231 			}
1232 		}
1233 		if (*count != 0) {
1234 			offset = iext->ib_offset + CEIL(iext->ib_count);
1235 			flags |= NEW_EXT;
1236 			index++;
1237 			goto begin;
1238 		}
1239 	}
1240 	iext->ib_marker1 = (uint32_t)0xAAAAAAAA;
1241 	iext->ib_marker2 = (uint32_t)0xBBBBBBBB;
1242 	return (error);
1243 }
1244 
1245 #undef	CEIL
1246 #undef	BASE
1247 
1248 int32_t
1249 ud_zero_it(struct ud_inode *ip, uint32_t start_block, uint32_t block_count)
1250 {
1251 	struct udf_vfs *udf_vfsp;
1252 	uint32_t bno, dummy;
1253 	int32_t error;
1254 	struct buf *bp;
1255 
1256 	/*
1257 	 * Donot use bio routines
1258 	 * since the buffer can sit
1259 	 * long enough in cache for the space
1260 	 * to be allocated/freed and
1261 	 * then allocated
1262 	 */
1263 	udf_vfsp = ip->i_udf;
1264 	bno = ud_xlate_to_daddr(udf_vfsp,
1265 		ip->i_icb_prn, start_block, block_count, &dummy);
1266 
1267 	dummy = block_count << udf_vfsp->udf_l2b_shift;
1268 	bp = (struct buf *)kmem_zalloc(biosize(), KM_SLEEP);
1269 	sema_init(&bp->b_sem, 0, NULL, SEMA_DEFAULT, NULL);
1270 	sema_init(&bp->b_io, 0, NULL, SEMA_DEFAULT, NULL);
1271 
1272 	bp->b_flags = B_WRITE | B_BUSY;
1273 	bp->b_edev = ip->i_dev;
1274 	bp->b_dev = cmpdev(ip->i_dev);
1275 	bp->b_blkno = bno << udf_vfsp->udf_l2d_shift;
1276 	bp->b_bcount = dummy;
1277 	bp->b_un.b_addr = kmem_zalloc(bp->b_bcount, KM_SLEEP);
1278 	bp->b_file = ip->i_vnode;
1279 	bp->b_offset = -1;
1280 
1281 	(void) bdev_strategy(bp);
1282 	if (error = biowait(bp)) {
1283 		cmn_err(CE_WARN, "error in write\n");
1284 	}
1285 
1286 	kmem_free(bp->b_un.b_addr, dummy);
1287 	sema_destroy(&bp->b_io);
1288 	sema_destroy(&bp->b_sem);
1289 	kmem_free((caddr_t)bp, biosize());
1290 
1291 	return (error);
1292 }
1293