xref: /linux/fs/jfs/jfs_extent.c (revision d6869352cb3c3cf3450637a52349e2e87c1354aa)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2004
4  */
5 
6 #include <linux/fs.h>
7 #include <linux/quotaops.h>
8 #include "jfs_incore.h"
9 #include "jfs_inode.h"
10 #include "jfs_superblock.h"
11 #include "jfs_dmap.h"
12 #include "jfs_extent.h"
13 #include "jfs_debug.h"
14 
15 /*
16  * forward references
17  */
18 static int extBalloc(struct inode *, s64, s64 *, s64 *);
19 #ifdef _NOTYET
20 static int extBrealloc(struct inode *, s64, s64, s64 *, s64 *);
21 #endif
22 static s64 extRoundDown(s64 nb);
23 
24 #define DPD(a)		(printk("(a): %d\n",(a)))
25 #define DPC(a)		(printk("(a): %c\n",(a)))
26 #define DPL1(a)					\
27 {						\
28 	if ((a) >> 32)				\
29 		printk("(a): %x%08x  ",(a));	\
30 	else					\
31 		printk("(a): %x  ",(a) << 32);	\
32 }
33 #define DPL(a)					\
34 {						\
35 	if ((a) >> 32)				\
36 		printk("(a): %x%08x\n",(a));	\
37 	else					\
38 		printk("(a): %x\n",(a) << 32);	\
39 }
40 
41 #define DPD1(a)		(printk("(a): %d  ",(a)))
42 #define DPX(a)		(printk("(a): %08x\n",(a)))
43 #define DPX1(a)		(printk("(a): %08x  ",(a)))
44 #define DPS(a)		(printk("%s\n",(a)))
45 #define DPE(a)		(printk("\nENTERING: %s\n",(a)))
46 #define DPE1(a)		(printk("\nENTERING: %s",(a)))
47 #define DPS1(a)		(printk("  %s  ",(a)))
48 
49 
50 /*
51  * NAME:	extAlloc()
52  *
53  * FUNCTION:	allocate an extent for a specified page range within a
54  *		file.
55  *
56  * PARAMETERS:
57  *	ip	- the inode of the file.
58  *	xlen	- requested extent length.
59  *	pno	- the starting page number with the file.
60  *	xp	- pointer to an xad.  on entry, xad describes an
61  *		  extent that is used as an allocation hint if the
62  *		  xaddr of the xad is non-zero.  on successful exit,
63  *		  the xad describes the newly allocated extent.
64  *	abnr	- bool indicating whether the newly allocated extent
65  *		  should be marked as allocated but not recorded.
66  *
67  * RETURN VALUES:
68  *	0	- success
69  *	-EIO	- i/o error.
70  *	-ENOSPC	- insufficient disk resources.
71  */
72 int
73 extAlloc(struct inode *ip, s64 xlen, s64 pno, xad_t * xp, bool abnr)
74 {
75 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
76 	s64 nxlen, nxaddr, xoff, hint, xaddr = 0;
77 	int rc;
78 	int xflag;
79 
80 	/* This blocks if we are low on resources */
81 	txBeginAnon(ip->i_sb);
82 
83 	/* Avoid race with jfs_commit_inode() */
84 	mutex_lock(&JFS_IP(ip)->commit_mutex);
85 
86 	/* validate extent length */
87 	if (xlen > MAXXLEN)
88 		xlen = MAXXLEN;
89 
90 	/* get the page's starting extent offset */
91 	xoff = pno << sbi->l2nbperpage;
92 
93 	/* check if an allocation hint was provided */
94 	if ((hint = addressXAD(xp))) {
95 		/* get the size of the extent described by the hint */
96 		nxlen = lengthXAD(xp);
97 
98 		/* check if the hint is for the portion of the file
99 		 * immediately previous to the current allocation
100 		 * request and if hint extent has the same abnr
101 		 * value as the current request.  if so, we can
102 		 * extend the hint extent to include the current
103 		 * extent if we can allocate the blocks immediately
104 		 * following the hint extent.
105 		 */
106 		if (offsetXAD(xp) + nxlen == xoff &&
107 		    abnr == ((xp->flag & XAD_NOTRECORDED) ? true : false))
108 			xaddr = hint + nxlen;
109 
110 		/* adjust the hint to the last block of the extent */
111 		hint += (nxlen - 1);
112 	}
113 
114 	/* allocate the disk blocks for the extent.  initially, extBalloc()
115 	 * will try to allocate disk blocks for the requested size (xlen).
116 	 * if this fails (xlen contiguous free blocks not available), it'll
117 	 * try to allocate a smaller number of blocks (producing a smaller
118 	 * extent), with this smaller number of blocks consisting of the
119 	 * requested number of blocks rounded down to the next smaller
120 	 * power of 2 number (i.e. 16 -> 8).  it'll continue to round down
121 	 * and retry the allocation until the number of blocks to allocate
122 	 * is smaller than the number of blocks per page.
123 	 */
124 	nxlen = xlen;
125 	if ((rc = extBalloc(ip, hint ? hint : INOHINT(ip), &nxlen, &nxaddr))) {
126 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
127 		return (rc);
128 	}
129 
130 	/* Allocate blocks to quota. */
131 	rc = dquot_alloc_block(ip, nxlen);
132 	if (rc) {
133 		dbFree(ip, nxaddr, (s64) nxlen);
134 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
135 		return rc;
136 	}
137 
138 	/* determine the value of the extent flag */
139 	xflag = abnr ? XAD_NOTRECORDED : 0;
140 
141 	/* if we can extend the hint extent to cover the current request,
142 	 * extend it.  otherwise, insert a new extent to
143 	 * cover the current request.
144 	 */
145 	if (xaddr && xaddr == nxaddr)
146 		rc = xtExtend(0, ip, xoff, (int) nxlen, 0);
147 	else
148 		rc = xtInsert(0, ip, xflag, xoff, (int) nxlen, &nxaddr, 0);
149 
150 	/* if the extend or insert failed,
151 	 * free the newly allocated blocks and return the error.
152 	 */
153 	if (rc) {
154 		dbFree(ip, nxaddr, nxlen);
155 		dquot_free_block(ip, nxlen);
156 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
157 		return (rc);
158 	}
159 
160 	/* set the results of the extent allocation */
161 	XADaddress(xp, nxaddr);
162 	XADlength(xp, nxlen);
163 	XADoffset(xp, xoff);
164 	xp->flag = xflag;
165 
166 	mark_inode_dirty(ip);
167 
168 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
169 	/*
170 	 * COMMIT_SyncList flags an anonymous tlock on page that is on
171 	 * sync list.
172 	 * We need to commit the inode to get the page written disk.
173 	 */
174 	if (test_and_clear_cflag(COMMIT_Synclist,ip))
175 		jfs_commit_inode(ip, 0);
176 
177 	return (0);
178 }
179 
180 
181 #ifdef _NOTYET
182 /*
183  * NAME:	extRealloc()
184  *
185  * FUNCTION:	extend the allocation of a file extent containing a
186  *		partial back last page.
187  *
188  * PARAMETERS:
189  *	ip	- the inode of the file.
190  *	cp	- cbuf for the partial backed last page.
191  *	xlen	- request size of the resulting extent.
192  *	xp	- pointer to an xad. on successful exit, the xad
193  *		  describes the newly allocated extent.
194  *	abnr	- bool indicating whether the newly allocated extent
195  *		  should be marked as allocated but not recorded.
196  *
197  * RETURN VALUES:
198  *	0	- success
199  *	-EIO	- i/o error.
200  *	-ENOSPC	- insufficient disk resources.
201  */
202 int extRealloc(struct inode *ip, s64 nxlen, xad_t * xp, bool abnr)
203 {
204 	struct super_block *sb = ip->i_sb;
205 	s64 xaddr, xlen, nxaddr, delta, xoff;
206 	s64 ntail, nextend, ninsert;
207 	int rc, nbperpage = JFS_SBI(sb)->nbperpage;
208 	int xflag;
209 
210 	/* This blocks if we are low on resources */
211 	txBeginAnon(ip->i_sb);
212 
213 	mutex_lock(&JFS_IP(ip)->commit_mutex);
214 	/* validate extent length */
215 	if (nxlen > MAXXLEN)
216 		nxlen = MAXXLEN;
217 
218 	/* get the extend (partial) page's disk block address and
219 	 * number of blocks.
220 	 */
221 	xaddr = addressXAD(xp);
222 	xlen = lengthXAD(xp);
223 	xoff = offsetXAD(xp);
224 
225 	/* if the extend page is abnr and if the request is for
226 	 * the extent to be allocated and recorded,
227 	 * make the page allocated and recorded.
228 	 */
229 	if ((xp->flag & XAD_NOTRECORDED) && !abnr) {
230 		xp->flag = 0;
231 		if ((rc = xtUpdate(0, ip, xp)))
232 			goto exit;
233 	}
234 
235 	/* try to allocated the request number of blocks for the
236 	 * extent.  dbRealloc() first tries to satisfy the request
237 	 * by extending the allocation in place. otherwise, it will
238 	 * try to allocate a new set of blocks large enough for the
239 	 * request.  in satisfying a request, dbReAlloc() may allocate
240 	 * less than what was request but will always allocate enough
241 	 * space as to satisfy the extend page.
242 	 */
243 	if ((rc = extBrealloc(ip, xaddr, xlen, &nxlen, &nxaddr)))
244 		goto exit;
245 
246 	/* Allocat blocks to quota. */
247 	rc = dquot_alloc_block(ip, nxlen);
248 	if (rc) {
249 		dbFree(ip, nxaddr, (s64) nxlen);
250 		mutex_unlock(&JFS_IP(ip)->commit_mutex);
251 		return rc;
252 	}
253 
254 	delta = nxlen - xlen;
255 
256 	/* check if the extend page is not abnr but the request is abnr
257 	 * and the allocated disk space is for more than one page.  if this
258 	 * is the case, there is a miss match of abnr between the extend page
259 	 * and the one or more pages following the extend page.  as a result,
260 	 * two extents will have to be manipulated. the first will be that
261 	 * of the extent of the extend page and will be manipulated thru
262 	 * an xtExtend() or an xtTailgate(), depending upon whether the
263 	 * disk allocation occurred as an inplace extension.  the second
264 	 * extent will be manipulated (created) through an xtInsert() and
265 	 * will be for the pages following the extend page.
266 	 */
267 	if (abnr && (!(xp->flag & XAD_NOTRECORDED)) && (nxlen > nbperpage)) {
268 		ntail = nbperpage;
269 		nextend = ntail - xlen;
270 		ninsert = nxlen - nbperpage;
271 
272 		xflag = XAD_NOTRECORDED;
273 	} else {
274 		ntail = nxlen;
275 		nextend = delta;
276 		ninsert = 0;
277 
278 		xflag = xp->flag;
279 	}
280 
281 	/* if we were able to extend the disk allocation in place,
282 	 * extend the extent.  otherwise, move the extent to a
283 	 * new disk location.
284 	 */
285 	if (xaddr == nxaddr) {
286 		/* extend the extent */
287 		if ((rc = xtExtend(0, ip, xoff + xlen, (int) nextend, 0))) {
288 			dbFree(ip, xaddr + xlen, delta);
289 			dquot_free_block(ip, nxlen);
290 			goto exit;
291 		}
292 	} else {
293 		/*
294 		 * move the extent to a new location:
295 		 *
296 		 * xtTailgate() accounts for relocated tail extent;
297 		 */
298 		if ((rc = xtTailgate(0, ip, xoff, (int) ntail, nxaddr, 0))) {
299 			dbFree(ip, nxaddr, nxlen);
300 			dquot_free_block(ip, nxlen);
301 			goto exit;
302 		}
303 	}
304 
305 
306 	/* check if we need to also insert a new extent */
307 	if (ninsert) {
308 		/* perform the insert.  if it fails, free the blocks
309 		 * to be inserted and make it appear that we only did
310 		 * the xtExtend() or xtTailgate() above.
311 		 */
312 		xaddr = nxaddr + ntail;
313 		if (xtInsert (0, ip, xflag, xoff + ntail, (int) ninsert,
314 			      &xaddr, 0)) {
315 			dbFree(ip, xaddr, (s64) ninsert);
316 			delta = nextend;
317 			nxlen = ntail;
318 			xflag = 0;
319 		}
320 	}
321 
322 	/* set the return results */
323 	XADaddress(xp, nxaddr);
324 	XADlength(xp, nxlen);
325 	XADoffset(xp, xoff);
326 	xp->flag = xflag;
327 
328 	mark_inode_dirty(ip);
329 exit:
330 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
331 	return (rc);
332 }
333 #endif			/* _NOTYET */
334 
335 
336 /*
337  * NAME:	extHint()
338  *
339  * FUNCTION:	produce an extent allocation hint for a file offset.
340  *
341  * PARAMETERS:
342  *	ip	- the inode of the file.
343  *	offset  - file offset for which the hint is needed.
344  *	xp	- pointer to the xad that is to be filled in with
345  *		  the hint.
346  *
347  * RETURN VALUES:
348  *	0	- success
349  *	-EIO	- i/o error.
350  */
351 int extHint(struct inode *ip, s64 offset, xad_t * xp)
352 {
353 	struct super_block *sb = ip->i_sb;
354 	int nbperpage = JFS_SBI(sb)->nbperpage;
355 	s64 prev;
356 	int rc = 0;
357 	s64 xaddr;
358 	int xlen;
359 	int xflag;
360 
361 	/* init the hint as "no hint provided" */
362 	XADaddress(xp, 0);
363 
364 	/* determine the starting extent offset of the page previous
365 	 * to the page containing the offset.
366 	 */
367 	prev = ((offset & ~POFFSET) >> JFS_SBI(sb)->l2bsize) - nbperpage;
368 
369 	/* if the offset is in the first page of the file, no hint provided.
370 	 */
371 	if (prev < 0)
372 		goto out;
373 
374 	rc = xtLookup(ip, prev, nbperpage, &xflag, &xaddr, &xlen, 0);
375 
376 	if ((rc == 0) && xlen) {
377 		if (xlen != nbperpage) {
378 			jfs_error(ip->i_sb, "corrupt xtree\n");
379 			rc = -EIO;
380 		}
381 		XADaddress(xp, xaddr);
382 		XADlength(xp, xlen);
383 		XADoffset(xp, prev);
384 		/*
385 		 * only preserve the abnr flag within the xad flags
386 		 * of the returned hint.
387 		 */
388 		xp->flag  = xflag & XAD_NOTRECORDED;
389 	} else
390 		rc = 0;
391 
392 out:
393 	return (rc);
394 }
395 
396 
397 /*
398  * NAME:	extRecord()
399  *
400  * FUNCTION:	change a page with a file from not recorded to recorded.
401  *
402  * PARAMETERS:
403  *	ip	- inode of the file.
404  *	cp	- cbuf of the file page.
405  *
406  * RETURN VALUES:
407  *	0	- success
408  *	-EIO	- i/o error.
409  *	-ENOSPC	- insufficient disk resources.
410  */
411 int extRecord(struct inode *ip, xad_t * xp)
412 {
413 	int rc;
414 
415 	txBeginAnon(ip->i_sb);
416 
417 	mutex_lock(&JFS_IP(ip)->commit_mutex);
418 
419 	/* update the extent */
420 	rc = xtUpdate(0, ip, xp);
421 
422 	mutex_unlock(&JFS_IP(ip)->commit_mutex);
423 	return rc;
424 }
425 
426 
427 #ifdef _NOTYET
428 /*
429  * NAME:	extFill()
430  *
431  * FUNCTION:	allocate disk space for a file page that represents
432  *		a file hole.
433  *
434  * PARAMETERS:
435  *	ip	- the inode of the file.
436  *	cp	- cbuf of the file page represent the hole.
437  *
438  * RETURN VALUES:
439  *	0	- success
440  *	-EIO	- i/o error.
441  *	-ENOSPC	- insufficient disk resources.
442  */
443 int extFill(struct inode *ip, xad_t * xp)
444 {
445 	int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
446 	s64 blkno = offsetXAD(xp) >> ip->i_blkbits;
447 
448 //	assert(ISSPARSE(ip));
449 
450 	/* initialize the extent allocation hint */
451 	XADaddress(xp, 0);
452 
453 	/* allocate an extent to fill the hole */
454 	if ((rc = extAlloc(ip, nbperpage, blkno, xp, false)))
455 		return (rc);
456 
457 	assert(lengthPXD(xp) == nbperpage);
458 
459 	return (0);
460 }
461 #endif			/* _NOTYET */
462 
463 
464 /*
465  * NAME:	extBalloc()
466  *
467  * FUNCTION:	allocate disk blocks to form an extent.
468  *
469  *		initially, we will try to allocate disk blocks for the
470  *		requested size (nblocks).  if this fails (nblocks
471  *		contiguous free blocks not available), we'll try to allocate
472  *		a smaller number of blocks (producing a smaller extent), with
473  *		this smaller number of blocks consisting of the requested
474  *		number of blocks rounded down to the next smaller power of 2
475  *		number (i.e. 16 -> 8).  we'll continue to round down and
476  *		retry the allocation until the number of blocks to allocate
477  *		is smaller than the number of blocks per page.
478  *
479  * PARAMETERS:
480  *	ip	 - the inode of the file.
481  *	hint	 - disk block number to be used as an allocation hint.
482  *	*nblocks - pointer to an s64 value.  on entry, this value specifies
483  *		   the desired number of block to be allocated. on successful
484  *		   exit, this value is set to the number of blocks actually
485  *		   allocated.
486  *	blkno	 - pointer to a block address that is filled in on successful
487  *		   return with the starting block number of the newly
488  *		   allocated block range.
489  *
490  * RETURN VALUES:
491  *	0	- success
492  *	-EIO	- i/o error.
493  *	-ENOSPC	- insufficient disk resources.
494  */
495 static int
496 extBalloc(struct inode *ip, s64 hint, s64 * nblocks, s64 * blkno)
497 {
498 	struct jfs_inode_info *ji = JFS_IP(ip);
499 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
500 	s64 nb, nblks, daddr, max;
501 	int rc, nbperpage = sbi->nbperpage;
502 	struct bmap *bmp = sbi->bmap;
503 	int ag;
504 
505 	/* get the number of blocks to initially attempt to allocate.
506 	 * we'll first try the number of blocks requested unless this
507 	 * number is greater than the maximum number of contiguous free
508 	 * blocks in the map. in that case, we'll start off with the
509 	 * maximum free.
510 	 */
511 	max = (s64) 1 << bmp->db_maxfreebud;
512 	if (*nblocks >= max && *nblocks > nbperpage)
513 		nb = nblks = (max > nbperpage) ? max : nbperpage;
514 	else
515 		nb = nblks = *nblocks;
516 
517 	/* try to allocate blocks */
518 	while ((rc = dbAlloc(ip, hint, nb, &daddr)) != 0) {
519 		/* if something other than an out of space error,
520 		 * stop and return this error.
521 		 */
522 		if (rc != -ENOSPC)
523 			return (rc);
524 
525 		/* decrease the allocation request size */
526 		nb = min(nblks, extRoundDown(nb));
527 
528 		/* give up if we cannot cover a page */
529 		if (nb < nbperpage)
530 			return (rc);
531 	}
532 
533 	*nblocks = nb;
534 	*blkno = daddr;
535 
536 	if (S_ISREG(ip->i_mode) && (ji->fileset == FILESYSTEM_I)) {
537 		ag = BLKTOAG(daddr, sbi);
538 		spin_lock_irq(&ji->ag_lock);
539 		if (ji->active_ag == -1) {
540 			atomic_inc(&bmp->db_active[ag]);
541 			ji->active_ag = ag;
542 		} else if (ji->active_ag != ag) {
543 			atomic_dec(&bmp->db_active[ji->active_ag]);
544 			atomic_inc(&bmp->db_active[ag]);
545 			ji->active_ag = ag;
546 		}
547 		spin_unlock_irq(&ji->ag_lock);
548 	}
549 
550 	return (0);
551 }
552 
553 
554 #ifdef _NOTYET
555 /*
556  * NAME:	extBrealloc()
557  *
558  * FUNCTION:	attempt to extend an extent's allocation.
559  *
560  *		Initially, we will try to extend the extent's allocation
561  *		in place.  If this fails, we'll try to move the extent
562  *		to a new set of blocks.  If moving the extent, we initially
563  *		will try to allocate disk blocks for the requested size
564  *		(newnblks).  if this fails (new contiguous free blocks not
565  *		available), we'll try to allocate a smaller number of
566  *		blocks (producing a smaller extent), with this smaller
567  *		number of blocks consisting of the requested number of
568  *		blocks rounded down to the next smaller power of 2
569  *		number (i.e. 16 -> 8).  We'll continue to round down and
570  *		retry the allocation until the number of blocks to allocate
571  *		is smaller than the number of blocks per page.
572  *
573  * PARAMETERS:
574  *	ip	 - the inode of the file.
575  *	blkno	 - starting block number of the extents current allocation.
576  *	nblks	 - number of blocks within the extents current allocation.
577  *	newnblks - pointer to a s64 value.  on entry, this value is the
578  *		   the new desired extent size (number of blocks).  on
579  *		   successful exit, this value is set to the extent's actual
580  *		   new size (new number of blocks).
581  *	newblkno - the starting block number of the extents new allocation.
582  *
583  * RETURN VALUES:
584  *	0	- success
585  *	-EIO	- i/o error.
586  *	-ENOSPC	- insufficient disk resources.
587  */
588 static int
589 extBrealloc(struct inode *ip,
590 	    s64 blkno, s64 nblks, s64 * newnblks, s64 * newblkno)
591 {
592 	int rc;
593 
594 	/* try to extend in place */
595 	if ((rc = dbExtend(ip, blkno, nblks, *newnblks - nblks)) == 0) {
596 		*newblkno = blkno;
597 		return (0);
598 	} else {
599 		if (rc != -ENOSPC)
600 			return (rc);
601 	}
602 
603 	/* in place extension not possible.
604 	 * try to move the extent to a new set of blocks.
605 	 */
606 	return (extBalloc(ip, blkno, newnblks, newblkno));
607 }
608 #endif			/* _NOTYET */
609 
610 
611 /*
612  * NAME:	extRoundDown()
613  *
614  * FUNCTION:	round down a specified number of blocks to the next
615  *		smallest power of 2 number.
616  *
617  * PARAMETERS:
618  *	nb	- the inode of the file.
619  *
620  * RETURN VALUES:
621  *	next smallest power of 2 number.
622  */
623 static s64 extRoundDown(s64 nb)
624 {
625 	int i;
626 	u64 m, k;
627 
628 	for (i = 0, m = (u64) 1 << 63; i < 64; i++, m >>= 1) {
629 		if (m & nb)
630 			break;
631 	}
632 
633 	i = 63 - i;
634 	k = (u64) 1 << i;
635 	k = ((k - 1) & nb) ? k : k >> 1;
636 
637 	return (k);
638 }
639