1 /*
2  *   Copyright (C) International Business Machines Corp., 2000-2003
3  *
4  *   This program is free software;  you can redistribute it and/or modify
5  *   it under the terms of the GNU General Public License as published by
6  *   the Free Software Foundation; either version 2 of the License, or
7  *   (at your option) any later version.
8  *
9  *   This program is distributed in the hope that it will be useful,
10  *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
11  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
12  *   the GNU General Public License for more details.
13  *
14  *   You should have received a copy of the GNU General Public License
15  *   along with this program;  if not, write to the Free Software
16  *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17  */
18 
19 #include <linux/fs.h>
20 #include "jfs_incore.h"
21 #include "jfs_superblock.h"
22 #include "jfs_dmap.h"
23 #include "jfs_extent.h"
24 #include "jfs_debug.h"
25 
26 /*
27  * forward references
28  */
29 static int extBalloc(struct inode *, s64, s64 *, s64 *);
30 #ifdef _NOTYET
31 static int extBrealloc(struct inode *, s64, s64, s64 *, s64 *);
32 #endif
33 static s64 extRoundDown(s64 nb);
34 
35 /*
36  * external references
37  */
38 extern int jfs_commit_inode(struct inode *, int);
39 
40 
41 #define DPD(a)          (printk("(a): %d\n",(a)))
42 #define DPC(a)          (printk("(a): %c\n",(a)))
43 #define DPL1(a)					\
44 {						\
45 	if ((a) >> 32)				\
46 		printk("(a): %x%08x  ",(a));	\
47 	else					\
48 		printk("(a): %x  ",(a) << 32);	\
49 }
50 #define DPL(a)					\
51 {						\
52 	if ((a) >> 32)				\
53 		printk("(a): %x%08x\n",(a));	\
54 	else					\
55 		printk("(a): %x\n",(a) << 32);	\
56 }
57 
58 #define DPD1(a)         (printk("(a): %d  ",(a)))
59 #define DPX(a)          (printk("(a): %08x\n",(a)))
60 #define DPX1(a)         (printk("(a): %08x  ",(a)))
61 #define DPS(a)          (printk("%s\n",(a)))
62 #define DPE(a)          (printk("\nENTERING: %s\n",(a)))
63 #define DPE1(a)          (printk("\nENTERING: %s",(a)))
64 #define DPS1(a)         (printk("  %s  ",(a)))
65 
66 
67 /*
68  * NAME:	extAlloc()
69  *
70  * FUNCTION:    allocate an extent for a specified page range within a
71  *		file.
72  *
73  * PARAMETERS:
74  *	ip	- the inode of the file.
75  *	xlen	- requested extent length.
76  *	pno	- the starting page number with the file.
77  *	xp	- pointer to an xad.  on entry, xad describes an
78  *		  extent that is used as an allocation hint if the
79  *		  xaddr of the xad is non-zero.  on successful exit,
80  *		  the xad describes the newly allocated extent.
81  *	abnr	- boolean_t indicating whether the newly allocated extent
82  *		  should be marked as allocated but not recorded.
83  *
84  * RETURN VALUES:
85  *      0       - success
86  *      -EIO	- i/o error.
87  *      -ENOSPC	- insufficient disk resources.
88  */
89 int
extAlloc(struct inode * ip,s64 xlen,s64 pno,xad_t * xp,boolean_t abnr)90 extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, boolean_t abnr)
91 {
92 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
93 	s64 nxlen, nxaddr, xoff, hint, xaddr = 0;
94 	int rc;
95 	int xflag;
96 
97 	/* This blocks if we are low on resources */
98 	txBeginAnon(ip->i_sb);
99 
100 	/* Avoid race with jfs_commit_inode() */
101 	down(&JFS_IP(ip)->commit_sem);
102 
103 	/* validate extent length */
104 	if (xlen > MAXXLEN)
105 		xlen = MAXXLEN;
106 
107 	/* get the page's starting extent offset */
108 	xoff = pno << sbi->l2nbperpage;
109 
110 	/* check if an allocation hint was provided */
111 	if ((hint = addressXAD(xp))) {
112 		/* get the size of the extent described by the hint */
113 		nxlen = lengthXAD(xp);
114 
115 		/* check if the hint is for the portion of the file
116 		 * immediately previous to the current allocation
117 		 * request and if hint extent has the same abnr
118 		 * value as the current request.  if so, we can
119 		 * extend the hint extent to include the current
120 		 * extent if we can allocate the blocks immediately
121 		 * following the hint extent.
122 		 */
123 		if (offsetXAD(xp) + nxlen == xoff &&
124 		    abnr == ((xp->flag & XAD_NOTRECORDED) ? TRUE : FALSE))
125 			xaddr = hint + nxlen;
126 
127 		/* adjust the hint to the last block of the extent */
128 		hint += (nxlen - 1);
129 	}
130 
131 	/* allocate the disk blocks for the extent.  initially, extBalloc()
132 	 * will try to allocate disk blocks for the requested size (xlen).
133 	 * if this fails (xlen contigious free blocks not avaliable), it'll
134 	 * try to allocate a smaller number of blocks (producing a smaller
135 	 * extent), with this smaller number of blocks consisting of the
136 	 * requested number of blocks rounded down to the next smaller
137 	 * power of 2 number (i.e. 16 -> 8).  it'll continue to round down
138 	 * and retry the allocation until the number of blocks to allocate
139 	 * is smaller than the number of blocks per page.
140 	 */
141 	nxlen = xlen;
142 	if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
143 		up(&JFS_IP(ip)->commit_sem);
144 		return (rc);
145 	}
146 
147 	/* determine the value of the extent flag */
148 	xflag = (abnr == TRUE) ? XAD_NOTRECORDED : 0;
149 
150 	/* if we can extend the hint extent to cover the current request,
151 	 * extend it.  otherwise, insert a new extent to
152 	 * cover the current request.
153 	 */
154 	if (xaddr && xaddr == nxaddr)
155 		rc = xtExtend(0, ip, xoff, (int) nxlen, 0);
156 	else
157 		rc = xtInsert(0, ip, xflag, xoff, (int) nxlen, &nxaddr, 0);
158 
159 	/* if the extend or insert failed,
160 	 * free the newly allocated blocks and return the error.
161 	 */
162 	if (rc) {
163 		dbFree(ip, nxaddr, nxlen);
164 		up(&JFS_IP(ip)->commit_sem);
165 		return (rc);
166 	}
167 
168 	/* update the number of blocks allocated to the file */
169 	ip->i_blocks += LBLK2PBLK(ip->i_sb, nxlen);
170 
171 	/* set the results of the extent allocation */
172 	XADaddress(xp, nxaddr);
173 	XADlength(xp, nxlen);
174 	XADoffset(xp, xoff);
175 	xp->flag = xflag;
176 
177 	mark_inode_dirty(ip);
178 	set_cflag(COMMIT_Syncdata, ip);
179 
180 	up(&JFS_IP(ip)->commit_sem);
181 	/*
182 	 * COMMIT_SyncList flags an anonymous tlock on page that is on
183 	 * sync list.
184 	 * We need to commit the inode to get the page written disk.
185 	 */
186 	if (test_and_clear_cflag(COMMIT_Synclist,ip))
187 		jfs_commit_inode(ip, 0);
188 
189 	return (0);
190 }
191 
192 
193 #ifdef _NOTYET
194 /*
195  * NAME:        extRealloc()
196  *
197  * FUNCTION:    extend the allocation of a file extent containing a
198  *		partial back last page.
199  *
200  * PARAMETERS:
201  *	ip	- the inode of the file.
202  *	cp	- cbuf for the partial backed last page.
203  *	xlen	- request size of the resulting extent.
204  *	xp	- pointer to an xad. on successful exit, the xad
205  *		  describes the newly allocated extent.
206  *	abnr	- boolean_t indicating whether the newly allocated extent
207  *		  should be marked as allocated but not recorded.
208  *
209  * RETURN VALUES:
210  *      0       - success
211  *      -EIO	- i/o error.
212  *      -ENOSPC	- insufficient disk resources.
213  */
extRealloc(struct inode * ip,s64 nxlen,xad_t * xp,boolean_t abnr)214 int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, boolean_t abnr)
215 {
216 	struct super_block *sb = ip->i_sb;
217 	s64 xaddr, xlen, nxaddr, delta, xoff;
218 	s64 ntail, nextend, ninsert;
219 	int rc, nbperpage = JFS_SBI(sb)->nbperpage;
220 	int xflag;
221 
222 	/* This blocks if we are low on resources */
223 	txBeginAnon(ip->i_sb);
224 
225 	down(&JFS_IP(ip)->commit_sem);
226 	/* validate extent length */
227 	if (nxlen > MAXXLEN)
228 		nxlen = MAXXLEN;
229 
230 	/* get the extend (partial) page's disk block address and
231 	 * number of blocks.
232 	 */
233 	xaddr = addressXAD(xp);
234 	xlen = lengthXAD(xp);
235 	xoff = offsetXAD(xp);
236 
237 	/* if the extend page is abnr and if the request is for
238 	 * the extent to be allocated and recorded,
239 	 * make the page allocated and recorded.
240 	 */
241 	if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
242 		xp->flag = 0;
243 		if ((rc = xtUpdate(0, ip, xp)))
244 			goto exit;
245 	}
246 
247 	/* try to allocated the request number of blocks for the
248 	 * extent.  dbRealloc() first tries to satisfy the request
249 	 * by extending the allocation in place. otherwise, it will
250 	 * try to allocate a new set of blocks large enough for the
251 	 * request.  in satisfying a request, dbReAlloc() may allocate
252 	 * less than what was request but will always allocate enough
253 	 * space as to satisfy the extend page.
254 	 */
255 	if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
256 		goto exit;
257 
258 	delta = nxlen - xlen;
259 
260 	/* check if the extend page is not abnr but the request is abnr
261 	 * and the allocated disk space is for more than one page.  if this
262 	 * is the case, there is a miss match of abnr between the extend page
263 	 * and the one or more pages following the extend page.  as a result,
264 	 * two extents will have to be manipulated. the first will be that
265 	 * of the extent of the extend page and will be manipulated thru
266 	 * an xtExtend() or an xtTailgate(), depending upon whether the
267 	 * disk allocation occurred as an inplace extension.  the second
268 	 * extent will be manipulated (created) through an xtInsert() and
269 	 * will be for the pages following the extend page.
270 	 */
271 	if (abnr && (!(xp->flag & XAD_NOTRECORDED)) && (nxlen > nbperpage)) {
272 		ntail = nbperpage;
273 		nextend = ntail - xlen;
274 		ninsert = nxlen - nbperpage;
275 
276 		xflag = XAD_NOTRECORDED;
277 	} else {
278 		ntail = nxlen;
279 		nextend = delta;
280 		ninsert = 0;
281 
282 		xflag = xp->flag;
283 	}
284 
285 	/* if we were able to extend the disk allocation in place,
286 	 * extend the extent.  otherwise, move the extent to a
287 	 * new disk location.
288 	 */
289 	if (xaddr == nxaddr) {
290 		/* extend the extent */
291 		if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
292 			dbFree(ip, xaddr + xlen, delta);
293 			goto exit;
294 		}
295 	} else {
296 		/*
297 		 * move the extent to a new location:
298 		 *
299 		 * xtTailgate() accounts for relocated tail extent;
300 		 */
301 		if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
302 			dbFree(ip, nxaddr, nxlen);
303 			goto exit;
304 		}
305 	}
306 
307 
308 	/* check if we need to also insert a new extent */
309 	if (ninsert) {
310 		/* perform the insert.  if it fails, free the blocks
311 		 * to be inserted and make it appear that we only did
312 		 * the xtExtend() or xtTailgate() above.
313 		 */
314 		xaddr = nxaddr + ntail;
315 		if (xtInsert (0, ip, xflag, xoff + ntail, (int) ninsert,
316 			      &xaddr, 0)) {
317 			dbFree(ip, xaddr, (s64) ninsert);
318 			delta = nextend;
319 			nxlen = ntail;
320 			xflag = 0;
321 		}
322 	}
323 
324 	/* update the inode with the number of blocks allocated */
325 	ip->i_blocks += LBLK2PBLK(sb, delta);
326 
327 	/* set the return results */
328 	XADaddress(xp, nxaddr);
329 	XADlength(xp, nxlen);
330 	XADoffset(xp, xoff);
331 	xp->flag = xflag;
332 
333 	mark_inode_dirty(ip);
334 exit:
335 	up(&JFS_IP(ip)->commit_sem);
336 	return (rc);
337 }
338 #endif			/* _NOTYET */
339 
340 
341 /*
342  * NAME:        extHint()
343  *
344  * FUNCTION:    produce an extent allocation hint for a file offset.
345  *
346  * PARAMETERS:
347  *	ip	- the inode of the file.
348  *	offset  - file offset for which the hint is needed.
349  *	xp	- pointer to the xad that is to be filled in with
350  *		  the hint.
351  *
352  * RETURN VALUES:
353  *      0       - success
354  *      -EIO	- i/o error.
355  */
extHint(struct inode * ip,s64 offset,xad_t * xp)356 int extHint(struct inode *ip, s64 offset, xad_t * xp)
357 {
358 	struct super_block *sb = ip->i_sb;
359 	struct xadlist xadl;
360 	struct lxdlist lxdl;
361 	lxd_t lxd;
362 	s64 prev;
363 	int rc, nbperpage = JFS_SBI(sb)->nbperpage;
364 
365 	/* init the hint as "no hint provided" */
366 	XADaddress(xp, 0);
367 
368 	/* determine the starting extent offset of the page previous
369 	 * to the page containing the offset.
370 	 */
371 	prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;
372 
373 	/* if the offsets in the first page of the file,
374 	 * no hint provided.
375 	 */
376 	if (prev < 0)
377 		return (0);
378 
379 	/* prepare to lookup the previous page's extent info */
380 	lxdl.maxnlxd = 1;
381 	lxdl.nlxd = 1;
382 	lxdl.lxd = &lxd;
383 	LXDoffset(&lxd, prev)
384 	    LXDlength(&lxd, nbperpage);
385 
386 	xadl.maxnxad = 1;
387 	xadl.nxad = 0;
388 	xadl.xad = xp;
389 
390 	/* perform the lookup */
391 	if ((rc = xtLookupList(ip, &lxdl, &xadl, 0)))
392 		return (rc);
393 
394 	/* check if not extent exists for the previous page.
395 	 * this is possible for sparse files.
396 	 */
397 	if (xadl.nxad == 0) {
398 //              assert(ISSPARSE(ip));
399 		return (0);
400 	}
401 
402 	/* only preserve the abnr flag within the xad flags
403 	 * of the returned hint.
404 	 */
405 	xp->flag &= XAD_NOTRECORDED;
406 
407         if(xadl.nxad != 1 || lengthXAD(xp) != nbperpage) {
408 		jfs_error(ip->i_sb, "extHint: corrupt xtree");
409 		return -EIO;
410         }
411 
412 	return (0);
413 }
414 
415 
416 /*
417  * NAME:        extRecord()
418  *
419  * FUNCTION:    change a page with a file from not recorded to recorded.
420  *
421  * PARAMETERS:
422  *	ip	- inode of the file.
423  *	cp	- cbuf of the file page.
424  *
425  * RETURN VALUES:
426  *      0       - success
427  *      -EIO	- i/o error.
428  *      -ENOSPC	- insufficient disk resources.
429  */
extRecord(struct inode * ip,xad_t * xp)430 int extRecord(struct inode *ip, xad_t * xp)
431 {
432 	int rc;
433 
434 	txBeginAnon(ip->i_sb);
435 
436 	down(&JFS_IP(ip)->commit_sem);
437 
438 	/* update the extent */
439 	rc = xtUpdate(0, ip, xp);
440 
441 	up(&JFS_IP(ip)->commit_sem);
442 	return rc;
443 }
444 
445 
446 #ifdef _NOTYET
447 /*
448  * NAME:        extFill()
449  *
450  * FUNCTION:    allocate disk space for a file page that represents
451  *		a file hole.
452  *
453  * PARAMETERS:
454  *	ip	- the inode of the file.
455  *	cp	- cbuf of the file page represent the hole.
456  *
457  * RETURN VALUES:
458  *      0       - success
459  *      -EIO	- i/o error.
460  *      -ENOSPC	- insufficient disk resources.
461  */
extFill(struct inode * ip,xad_t * xp)462 int extFill(struct inode *ip, xad_t * xp)
463 {
464 	int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
465 	s64 blkno = offsetXAD(xp) >> ip->i_blksize;
466 
467 //      assert(ISSPARSE(ip));
468 
469 	/* initialize the extent allocation hint */
470 	XADaddress(xp, 0);
471 
472 	/* allocate an extent to fill the hole */
473 	if ((rc = extAlloc(ip, nbperpage, blkno, xp, FALSE)))
474 		return (rc);
475 
476 	assert(lengthPXD(xp) == nbperpage);
477 
478 	return (0);
479 }
480 #endif			/* _NOTYET */
481 
482 
483 /*
484  * NAME:	extBalloc()
485  *
486  * FUNCTION:    allocate disk blocks to form an extent.
487  *
488  *		initially, we will try to allocate disk blocks for the
489  *		requested size (nblocks).  if this fails (nblocks
490  *		contigious free blocks not avaliable), we'll try to allocate
491  *		a smaller number of blocks (producing a smaller extent), with
492  *		this smaller number of blocks consisting of the requested
493  *		number of blocks rounded down to the next smaller power of 2
494  *		number (i.e. 16 -> 8).  we'll continue to round down and
495  *		retry the allocation until the number of blocks to allocate
496  *		is smaller than the number of blocks per page.
497  *
498  * PARAMETERS:
499  *	ip	 - the inode of the file.
500  *	hint	 - disk block number to be used as an allocation hint.
501  *	*nblocks - pointer to an s64 value.  on entry, this value specifies
502  *		   the desired number of block to be allocated. on successful
503  *		   exit, this value is set to the number of blocks actually
504  *		   allocated.
505  *	blkno	 - pointer to a block address that is filled in on successful
506  *		   return with the starting block number of the newly
507  *		   allocated block range.
508  *
509  * RETURN VALUES:
510  *      0       - success
511  *      -EIO	- i/o error.
512  *      -ENOSPC	- insufficient disk resources.
513  */
514 static int
extBalloc(struct inode * ip,s64 hint,s64 * nblocks,s64 * blkno)515 extBalloc(struct inode *ip, s64 hint, s64 * nblocks, s64 * blkno)
516 {
517 	struct jfs_inode_info *ji = JFS_IP(ip);
518 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
519 	s64 nb, nblks, daddr, max;
520 	int rc, nbperpage = sbi->nbperpage;
521 	struct bmap *bmp = sbi->bmap;
522 	int ag;
523 
524 	/* get the number of blocks to initially attempt to allocate.
525 	 * we'll first try the number of blocks requested unless this
526 	 * number is greater than the maximum number of contigious free
527 	 * blocks in the map. in that case, we'll start off with the
528 	 * maximum free.
529 	 */
530 	max = (s64) 1 << bmp->db_maxfreebud;
531 	if (*nblocks >= max && *nblocks > nbperpage)
532 		nb = nblks = (max > nbperpage) ? max : nbperpage;
533 	else
534 		nb = nblks = *nblocks;
535 
536 	/* try to allocate blocks */
537 	while ((rc = dbAlloc(ip, hint, nb, &daddr))) {
538 		/* if something other than an out of space error,
539 		 * stop and return this error.
540 		 */
541 		if (rc != -ENOSPC)
542 			return (rc);
543 
544 		/* decrease the allocation request size */
545 		nb = min(nblks, extRoundDown(nb));
546 
547 		/* give up if we cannot cover a page */
548 		if (nb < nbperpage)
549 			return (rc);
550 	}
551 
552 	*nblocks = nb;
553 	*blkno = daddr;
554 
555 	if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
556 		ag = BLKTOAG(daddr, sbi);
557 		spin_lock_irq(&ji->ag_lock);
558 		if (ji->active_ag == -1) {
559 			atomic_inc(&bmp->db_active[ag]);
560 			ji->active_ag = ag;
561 		} else if (ji->active_ag != ag) {
562 			atomic_dec(&bmp->db_active[ji->active_ag]);
563 			atomic_inc(&bmp->db_active[ag]);
564 			ji->active_ag = ag;
565 		}
566 		spin_unlock_irq(&ji->ag_lock);
567 	}
568 
569 	return (0);
570 }
571 
572 
573 #ifdef _NOTYET
574 /*
575  * NAME:	extBrealloc()
576  *
577  * FUNCTION:    attempt to extend an extent's allocation.
578  *
579  *		initially, we will try to extend the extent's allocation
580  *		in place.  if this fails, we'll try to move the extent
581  *		to a new set of blocks. if moving the extent, we initially
582  *		will try to allocate disk blocks for the requested size
583  *		(nnew).  if this fails 	(nnew contigious free blocks not
584  *		avaliable), we'll try  to allocate a smaller number of
585  *		blocks (producing a smaller extent), with this smaller
586  *		number of blocks consisting of the requested number of
587  *		blocks rounded down to the next smaller power of 2
588  *		number (i.e. 16 -> 8).  we'll continue to round down and
589  *		retry the allocation until the number of blocks to allocate
590  *		is smaller than the number of blocks per page.
591  *
592  * PARAMETERS:
593  *	ip	 - the inode of the file.
594  *	blkno    - starting block number of the extents current allocation.
595  *	nblks    - number of blocks within the extents current allocation.
596  *	newnblks - pointer to a s64 value.  on entry, this value is the
597  *		   the new desired extent size (number of blocks).  on
598  *		   successful exit, this value is set to the extent's actual
599  *		   new size (new number of blocks).
600  *	newblkno - the starting block number of the extents new allocation.
601  *
602  * RETURN VALUES:
603  *      0       - success
604  *      -EIO	- i/o error.
605  *      -ENOSPC	- insufficient disk resources.
606  */
607 static int
extBrealloc(struct inode * ip,s64 blkno,s64 nblks,s64 * newnblks,s64 * newblkno)608 extBrealloc(struct inode *ip,
609 	    s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
610 {
611 	int rc;
612 
613 	/* try to extend in place */
614 	if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
615 		*newblkno = blkno;
616 		return (0);
617 	} else {
618 		if (rc != -ENOSPC)
619 			return (rc);
620 	}
621 
622 	/* in place extension not possible.
623 	 * try to move the extent to a new set of blocks.
624 	 */
625 	return (extBalloc(ip, blkno, newnblks, newblkno));
626 }
627 #endif			/* _NOTYET */
628 
629 
630 /*
631  * NAME:        extRoundDown()
632  *
633  * FUNCTION:    round down a specified number of blocks to the next
634  *		smallest power of 2 number.
635  *
636  * PARAMETERS:
637  *	nb	- the inode of the file.
638  *
639  * RETURN VALUES:
640  *      next smallest power of 2 number.
641  */
extRoundDown(s64 nb)642 static s64 extRoundDown(s64 nb)
643 {
644 	int i;
645 	u64 m, k;
646 
647 	for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
648 		if (m & nb)
649 			break;
650 	}
651 
652 	i = 63 - i;
653 	k = (u64) 1 << i;
654 	k = ((k - 1) & nb) ? k : k >> 1;
655 
656 	return (k);
657 }
658