xref: /netbsd-src/sys/ufs/ext2fs/ext2fs_balloc.c (revision 5aefcfdc06931dd97e76246d2fe0302f7b3fe094)
1 /*	$NetBSD: ext2fs_balloc.c,v 1.8 2000/12/10 06:38:31 chs Exp $	*/
2 
3 /*
4  * Copyright (c) 1997 Manuel Bouyer.
5  * Copyright (c) 1982, 1986, 1989, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgement:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  *	@(#)ffs_balloc.c	8.4 (Berkeley) 9/23/93
37  * Modified for ext2fs by Manuel Bouyer.
38  */
39 
40 #if defined(_KERNEL) && !defined(_LKM)
41 #include "opt_uvmhist.h"
42 #endif
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/buf.h>
47 #include <sys/proc.h>
48 #include <sys/file.h>
49 #include <sys/vnode.h>
50 #include <sys/mount.h>
51 
52 #include <uvm/uvm.h>
53 
54 #include <ufs/ufs/quota.h>
55 #include <ufs/ufs/inode.h>
56 #include <ufs/ufs/ufs_extern.h>
57 
58 #include <ufs/ext2fs/ext2fs.h>
59 #include <ufs/ext2fs/ext2fs_extern.h>
60 
61 /*
62  * Balloc defines the structure of file system storage
63  * by allocating the physical blocks on a device given
64  * the inode and the logical block number in a file.
65  */
66 int
67 ext2fs_balloc(ip, bn, size, cred, bpp, flags)
68 	struct inode *ip;
69 	ufs_daddr_t bn;
70 	int size;
71 	struct ucred *cred;
72 	struct buf **bpp;
73 	int flags;
74 {
75 	struct m_ext2fs *fs;
76 	ufs_daddr_t nb;
77 	struct buf *bp, *nbp;
78 	struct vnode *vp = ITOV(ip);
79 	struct indir indirs[NIADDR + 2];
80 	ufs_daddr_t newb, lbn, *bap, pref;
81 	int num, i, error;
82 	u_int deallocated;
83 	ufs_daddr_t *allocib, *blkp, *allocblk, allociblk[NIADDR + 1];
84 	int unwindidx = -1;
85 	UVMHIST_FUNC("ext2fs_balloc"); UVMHIST_CALLED(ubchist);
86 
87 	UVMHIST_LOG(ubchist, "bn 0x%x", bn,0,0,0);
88 
89 	if (bpp != NULL) {
90 		*bpp = NULL;
91 	}
92 	if (bn < 0)
93 		return (EFBIG);
94 	fs = ip->i_e2fs;
95 	lbn = bn;
96 
97 	/*
98 	 * The first NDADDR blocks are direct blocks
99 	 */
100 	if (bn < NDADDR) {
101 		nb = fs2h32(ip->i_e2fs_blocks[bn]);
102 		if (nb != 0) {
103 
104 			/*
105 			 * the block is already allocated, just read it.
106 			 */
107 
108 			if (bpp != NULL) {
109 				error = bread(vp, bn, fs->e2fs_bsize, NOCRED,
110 					      &bp);
111 				if (error) {
112 					brelse(bp);
113 					return (error);
114 				}
115 				*bpp = bp;
116 			}
117 			return (0);
118 		}
119 
120 		/*
121 		 * allocate a new direct block.
122 		 */
123 
124 		error = ext2fs_alloc(ip, bn,
125 		    ext2fs_blkpref(ip, bn, bn, &ip->i_e2fs_blocks[0]),
126 		    cred, &newb);
127 		if (error)
128 			return (error);
129 		ip->i_e2fs_last_lblk = lbn;
130 		ip->i_e2fs_last_blk = newb;
131 		ip->i_e2fs_blocks[bn] = h2fs32(newb);
132 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
133 		if (bpp != NULL) {
134 			bp = getblk(vp, bn, fs->e2fs_bsize, 0, 0);
135 			bp->b_blkno = fsbtodb(fs, newb);
136 			if (flags & B_CLRBUF)
137 				clrbuf(bp);
138 			*bpp = bp;
139 		}
140 		return (0);
141 	}
142 	/*
143 	 * Determine the number of levels of indirection.
144 	 */
145 	pref = 0;
146 	if ((error = ufs_getlbns(vp, bn, indirs, &num)) != 0)
147 		return(error);
148 #ifdef DIAGNOSTIC
149 	if (num < 1)
150 		panic ("ext2fs_balloc: ufs_getlbns returned indirect block\n");
151 #endif
152 	/*
153 	 * Fetch the first indirect block allocating if necessary.
154 	 */
155 	--num;
156 	nb = fs2h32(ip->i_e2fs_blocks[NDADDR + indirs[0].in_off]);
157 	allocib = NULL;
158 	allocblk = allociblk;
159 	if (nb == 0) {
160 		pref = ext2fs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
161 		error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
162 		if (error)
163 			return (error);
164 		nb = newb;
165 		*allocblk++ = nb;
166 		ip->i_e2fs_last_blk = newb;
167 		bp = getblk(vp, indirs[1].in_lbn, fs->e2fs_bsize, 0, 0);
168 		bp->b_blkno = fsbtodb(fs, newb);
169 		clrbuf(bp);
170 		/*
171 		 * Write synchronously so that indirect blocks
172 		 * never point at garbage.
173 		 */
174 		if ((error = bwrite(bp)) != 0)
175 			goto fail;
176 		unwindidx = 0;
177 		allocib = &ip->i_e2fs_blocks[NDADDR + indirs[0].in_off];
178 		*allocib = h2fs32(newb);
179 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
180 	}
181 	/*
182 	 * Fetch through the indirect blocks, allocating as necessary.
183 	 */
184 	for (i = 1;;) {
185 		error = bread(vp,
186 		    indirs[i].in_lbn, (int)fs->e2fs_bsize, NOCRED, &bp);
187 		if (error) {
188 			brelse(bp);
189 			goto fail;
190 		}
191 		bap = (ufs_daddr_t *)bp->b_data;
192 		nb = fs2h32(bap[indirs[i].in_off]);
193 		if (i == num)
194 			break;
195 		i++;
196 		if (nb != 0) {
197 			brelse(bp);
198 			continue;
199 		}
200 		pref = ext2fs_blkpref(ip, lbn, 0, (ufs_daddr_t *)0);
201 		error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
202 		if (error) {
203 			brelse(bp);
204 			goto fail;
205 		}
206 		nb = newb;
207 		*allocblk++ = nb;
208 		ip->i_e2fs_last_blk = newb;
209 		nbp = getblk(vp, indirs[i].in_lbn, fs->e2fs_bsize, 0, 0);
210 		nbp->b_blkno = fsbtodb(fs, nb);
211 		clrbuf(nbp);
212 		/*
213 		 * Write synchronously so that indirect blocks
214 		 * never point at garbage.
215 		 */
216 		if ((error = bwrite(nbp)) != 0) {
217 			brelse(bp);
218 			goto fail;
219 		}
220 		if (unwindidx < 0)
221 			unwindidx = i - 1;
222 		bap[indirs[i - 1].in_off] = h2fs32(nb);
223 		/*
224 		 * If required, write synchronously, otherwise use
225 		 * delayed write.
226 		 */
227 		if (flags & B_SYNC) {
228 			bwrite(bp);
229 		} else {
230 			bdwrite(bp);
231 		}
232 	}
233 	/*
234 	 * Get the data block, allocating if necessary.
235 	 */
236 	if (nb == 0) {
237 		pref = ext2fs_blkpref(ip, lbn, indirs[num].in_off, &bap[0]);
238 		error = ext2fs_alloc(ip, lbn, pref, cred, &newb);
239 		if (error) {
240 			brelse(bp);
241 			goto fail;
242 		}
243 		nb = newb;
244 		*allocblk++ = nb;
245 		ip->i_e2fs_last_lblk = lbn;
246 		ip->i_e2fs_last_blk = newb;
247 		bap[indirs[num].in_off] = h2fs32(nb);
248 		/*
249 		 * If required, write synchronously, otherwise use
250 		 * delayed write.
251 		 */
252 		if (flags & B_SYNC) {
253 			bwrite(bp);
254 		} else {
255 			bdwrite(bp);
256 		}
257 		if (bpp != NULL) {
258 			nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
259 			nbp->b_blkno = fsbtodb(fs, nb);
260 			if (flags & B_CLRBUF)
261 				clrbuf(nbp);
262 			*bpp = nbp;
263 		}
264 		return (0);
265 	}
266 	brelse(bp);
267 	if (bpp != NULL) {
268 		if (flags & B_CLRBUF) {
269 			error = bread(vp, lbn, (int)fs->e2fs_bsize, NOCRED,
270 				      &nbp);
271 			if (error) {
272 				brelse(nbp);
273 				goto fail;
274 			}
275 		} else {
276 			nbp = getblk(vp, lbn, fs->e2fs_bsize, 0, 0);
277 			nbp->b_blkno = fsbtodb(fs, nb);
278 		}
279 		*bpp = nbp;
280 	}
281 	return (0);
282 fail:
283 	/*
284 	 * If we have failed part way through block allocation, we
285 	 * have to deallocate any indirect blocks that we have allocated.
286 	 */
287 	for (deallocated = 0, blkp = allociblk; blkp < allocblk; blkp++) {
288 		ext2fs_blkfree(ip, *blkp);
289 		deallocated += fs->e2fs_bsize;
290 	}
291 	if (unwindidx >= 0) {
292 		if (unwindidx == 0) {
293 			*allocib = 0;
294 		} else {
295 			int r;
296 
297 			r = bread(vp, indirs[unwindidx].in_lbn,
298 			    (int)fs->e2fs_bsize, NOCRED, &bp);
299 			if (r) {
300 				panic("Could not unwind indirect block, error %d", r);
301 				brelse(bp);
302 			} else {
303 				bap = (ufs_daddr_t *)bp->b_data;
304 				bap[indirs[unwindidx].in_off] = 0;
305 				if (flags & B_SYNC)
306 					bwrite(bp);
307 				else
308 					bdwrite(bp);
309 			}
310 		}
311 		for (i = unwindidx + 1; i <= num; i++) {
312 			bp = getblk(vp, indirs[i].in_lbn, (int)fs->e2fs_bsize,
313 			    0, 0);
314 			bp->b_flags |= B_INVAL;
315 			brelse(bp);
316 		}
317 	}
318 	if (deallocated) {
319 		ip->i_e2fs_nblock -= btodb(deallocated);
320 		ip->i_e2fs_flags |= IN_CHANGE | IN_UPDATE;
321 	}
322 	return error;
323 }
324 
325 int
326 ext2fs_ballocn(v)
327 	void *v;
328 {
329 	struct vop_ballocn_args /* {
330 		struct vnode *a_vp;
331 		off_t a_offset;
332 		off_t a_length;
333 		struct ucred *a_cred;
334 		int a_flags;
335 	} */ *ap = v;
336 	off_t off, len;
337 	struct vnode *vp = ap->a_vp;
338 	struct inode *ip = VTOI(vp);
339 	struct m_ext2fs *fs = ip->i_e2fs;
340 	int error, delta, bshift, bsize;
341 	UVMHIST_FUNC("ext2fs_ballocn"); UVMHIST_CALLED(ubchist);
342 
343 	bshift = fs->e2fs_bshift;
344 	bsize = 1 << bshift;
345 
346 	off = ap->a_offset;
347 	len = ap->a_length;
348 
349 	delta = off & (bsize - 1);
350 	off -= delta;
351 	len += delta;
352 
353 	while (len > 0) {
354 		bsize = min(bsize, len);
355 		UVMHIST_LOG(ubchist, "off 0x%x len 0x%x bsize 0x%x",
356 			    off, len, bsize, 0);
357 
358 		error = ext2fs_balloc(ip, lblkno(fs, off), bsize, ap->a_cred,
359 		    NULL, ap->a_flags);
360 		if (error) {
361 			UVMHIST_LOG(ubchist, "error %d", error, 0,0,0);
362 			return error;
363 		}
364 
365 		/*
366 		 * increase file size now, VOP_BALLOC() requires that
367 		 * EOF be up-to-date before each call.
368 		 */
369 
370 		if (ip->i_e2fs_size < off + bsize) {
371 			UVMHIST_LOG(ubchist, "old 0x%x new 0x%x",
372 				    ip->i_e2fs_size, off + bsize,0,0);
373 			ip->i_e2fs_size = off + bsize;
374 			if (vp->v_uvm.u_size < ip->i_e2fs_size) {
375 				uvm_vnp_setsize(vp, ip->i_e2fs_size);
376 			}
377 		}
378 
379 		off += bsize;
380 		len -= bsize;
381 	}
382 	return 0;
383 }
384 
385 /*
386  * allocate a range of blocks in a file.
387  * after this function returns, any page entirely contained within the range
388  * will map to invalid data and thus must be overwritten before it is made
389  * accessible to others.
390  */
391 
392 int
393 ext2fs_balloc_range(vp, off, len, cred, flags)
394 	struct vnode *vp;
395 	off_t off, len;
396 	struct ucred *cred;
397 	int flags;
398 {
399 	off_t oldeof, eof, pagestart;
400 	struct uvm_object *uobj;
401 	int i, delta, error, npages;
402 	int bshift = vp->v_mount->mnt_fs_bshift;
403 	int bsize = 1 << bshift;
404 	int ppb = max(bsize >> PAGE_SHIFT, 1);
405 	struct vm_page *pgs[ppb];
406 	UVMHIST_FUNC("ext2fs_balloc_range"); UVMHIST_CALLED(ubchist);
407 	UVMHIST_LOG(ubchist, "vp %p off 0x%x len 0x%x u_size 0x%x",
408 		    vp, off, len, vp->v_uvm.u_size);
409 
410 	error = 0;
411 	uobj = &vp->v_uvm.u_obj;
412 	oldeof = vp->v_uvm.u_size;
413 	eof = max(oldeof, off + len);
414 	UVMHIST_LOG(ubchist, "new eof 0x%x", eof,0,0,0);
415 	pgs[0] = NULL;
416 
417 	/*
418 	 * cache the new range of the file.  this will create zeroed pages
419 	 * where the new block will be and keep them locked until the
420 	 * new block is allocated, so there will be no window where
421 	 * the old contents of the new block is visible to racing threads.
422 	 */
423 
424 	pagestart = trunc_page(off) & ~(bsize - 1);
425 	npages = min(ppb, (round_page(eof) - pagestart) >> PAGE_SHIFT);
426 	memset(pgs, 0, npages);
427 	simple_lock(&uobj->vmobjlock);
428 	error = VOP_GETPAGES(vp, pagestart, pgs, &npages, 0,
429 	    VM_PROT_READ, 0, PGO_SYNCIO | PGO_PASTEOF);
430 	if (error) {
431 		UVMHIST_LOG(ubchist, "getpages %d", error,0,0,0);
432 		goto errout;
433 	}
434 	for (i = 0; i < npages; i++) {
435 		UVMHIST_LOG(ubchist, "got pgs[%d] %p", i, pgs[i],0,0);
436 		KASSERT((pgs[i]->flags & PG_RELEASED) == 0);
437 		pgs[i]->flags &= ~PG_CLEAN;
438 		uvm_pageactivate(pgs[i]);
439 	}
440 
441 	/*
442 	 * adjust off to be block-aligned.
443 	 */
444 
445 	delta = off & (bsize - 1);
446 	off -= delta;
447 	len += delta;
448 
449 	/*
450 	 * now allocate the range.
451 	 */
452 
453 	lockmgr(&vp->v_glock, LK_EXCLUSIVE, NULL);
454 	error = VOP_BALLOCN(vp, off, len, cred, flags);
455 	UVMHIST_LOG(ubchist, "ballocn %d", error,0,0,0);
456 	lockmgr(&vp->v_glock, LK_RELEASE, NULL);
457 
458 	/*
459 	 * unbusy any pages we are holding.
460 	 */
461 
462 errout:
463 	simple_lock(&uobj->vmobjlock);
464 	if (error) {
465 		(void) (uobj->pgops->pgo_flush)(uobj, oldeof, pagestart + ppb,
466 		    PGO_FREE);
467 	}
468 	if (pgs[0] != NULL) {
469 		uvm_page_unbusy(pgs, npages);
470 	}
471 	simple_unlock(&uobj->vmobjlock);
472 	return (error);
473 }
474