• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3  * All Rights Reserved.
4  *
5  * This program is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU General Public License as
7  * published by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it would be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * 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 the Free Software Foundation,
16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
17  */
18 #include <linux/log2.h>
19 
20 #include "xfs.h"
21 #include "xfs_fs.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
25 #include "xfs_inum.h"
26 #include "xfs_sb.h"
27 #include "xfs_ag.h"
28 #include "xfs_mount.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_bmap.h"
34 #include "xfs_error.h"
35 #include "xfs_trace.h"
36 #include "xfs_attr_sf.h"
37 #include "xfs_dinode.h"
38 
39 kmem_zone_t *xfs_ifork_zone;
40 
41 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
42 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
43 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
44 
45 #ifdef DEBUG
46 /*
47  * Make sure that the extents in the given memory buffer
48  * are valid.
49  */
50 void
xfs_validate_extents(xfs_ifork_t * ifp,int nrecs,xfs_exntfmt_t fmt)51 xfs_validate_extents(
52 	xfs_ifork_t		*ifp,
53 	int			nrecs,
54 	xfs_exntfmt_t		fmt)
55 {
56 	xfs_bmbt_irec_t		irec;
57 	xfs_bmbt_rec_host_t	rec;
58 	int			i;
59 
60 	for (i = 0; i < nrecs; i++) {
61 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
62 		rec.l0 = get_unaligned(&ep->l0);
63 		rec.l1 = get_unaligned(&ep->l1);
64 		xfs_bmbt_get_all(&rec, &irec);
65 		if (fmt == XFS_EXTFMT_NOSTATE)
66 			ASSERT(irec.br_state == XFS_EXT_NORM);
67 	}
68 }
69 #else /* DEBUG */
70 #define xfs_validate_extents(ifp, nrecs, fmt)
71 #endif /* DEBUG */
72 
73 
74 /*
75  * Move inode type and inode format specific information from the
76  * on-disk inode to the in-core inode.  For fifos, devs, and sockets
77  * this means set if_rdev to the proper value.  For files, directories,
78  * and symlinks this means to bring in the in-line data or extent
79  * pointers.  For a file in B-tree format, only the root is immediately
80  * brought in-core.  The rest will be in-lined in if_extents when it
81  * is first referenced (see xfs_iread_extents()).
82  */
83 int
xfs_iformat_fork(xfs_inode_t * ip,xfs_dinode_t * dip)84 xfs_iformat_fork(
85 	xfs_inode_t		*ip,
86 	xfs_dinode_t		*dip)
87 {
88 	xfs_attr_shortform_t	*atp;
89 	int			size;
90 	int			error = 0;
91 	xfs_fsize_t             di_size;
92 
93 	if (unlikely(be32_to_cpu(dip->di_nextents) +
94 		     be16_to_cpu(dip->di_anextents) >
95 		     be64_to_cpu(dip->di_nblocks))) {
96 		xfs_warn(ip->i_mount,
97 			"corrupt dinode %Lu, extent total = %d, nblocks = %Lu.",
98 			(unsigned long long)ip->i_ino,
99 			(int)(be32_to_cpu(dip->di_nextents) +
100 			      be16_to_cpu(dip->di_anextents)),
101 			(unsigned long long)
102 				be64_to_cpu(dip->di_nblocks));
103 		XFS_CORRUPTION_ERROR("xfs_iformat(1)", XFS_ERRLEVEL_LOW,
104 				     ip->i_mount, dip);
105 		return -EFSCORRUPTED;
106 	}
107 
108 	if (unlikely(dip->di_forkoff > ip->i_mount->m_sb.sb_inodesize)) {
109 		xfs_warn(ip->i_mount, "corrupt dinode %Lu, forkoff = 0x%x.",
110 			(unsigned long long)ip->i_ino,
111 			dip->di_forkoff);
112 		XFS_CORRUPTION_ERROR("xfs_iformat(2)", XFS_ERRLEVEL_LOW,
113 				     ip->i_mount, dip);
114 		return -EFSCORRUPTED;
115 	}
116 
117 	if (unlikely((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) &&
118 		     !ip->i_mount->m_rtdev_targp)) {
119 		xfs_warn(ip->i_mount,
120 			"corrupt dinode %Lu, has realtime flag set.",
121 			ip->i_ino);
122 		XFS_CORRUPTION_ERROR("xfs_iformat(realtime)",
123 				     XFS_ERRLEVEL_LOW, ip->i_mount, dip);
124 		return -EFSCORRUPTED;
125 	}
126 
127 	switch (ip->i_d.di_mode & S_IFMT) {
128 	case S_IFIFO:
129 	case S_IFCHR:
130 	case S_IFBLK:
131 	case S_IFSOCK:
132 		if (unlikely(dip->di_format != XFS_DINODE_FMT_DEV)) {
133 			XFS_CORRUPTION_ERROR("xfs_iformat(3)", XFS_ERRLEVEL_LOW,
134 					      ip->i_mount, dip);
135 			return -EFSCORRUPTED;
136 		}
137 		ip->i_d.di_size = 0;
138 		ip->i_df.if_u2.if_rdev = xfs_dinode_get_rdev(dip);
139 		break;
140 
141 	case S_IFREG:
142 	case S_IFLNK:
143 	case S_IFDIR:
144 		switch (dip->di_format) {
145 		case XFS_DINODE_FMT_LOCAL:
146 			/*
147 			 * no local regular files yet
148 			 */
149 			if (unlikely(S_ISREG(be16_to_cpu(dip->di_mode)))) {
150 				xfs_warn(ip->i_mount,
151 			"corrupt inode %Lu (local format for regular file).",
152 					(unsigned long long) ip->i_ino);
153 				XFS_CORRUPTION_ERROR("xfs_iformat(4)",
154 						     XFS_ERRLEVEL_LOW,
155 						     ip->i_mount, dip);
156 				return -EFSCORRUPTED;
157 			}
158 
159 			di_size = be64_to_cpu(dip->di_size);
160 			if (unlikely(di_size < 0 ||
161 				     di_size > XFS_DFORK_DSIZE(dip, ip->i_mount))) {
162 				xfs_warn(ip->i_mount,
163 			"corrupt inode %Lu (bad size %Ld for local inode).",
164 					(unsigned long long) ip->i_ino,
165 					(long long) di_size);
166 				XFS_CORRUPTION_ERROR("xfs_iformat(5)",
167 						     XFS_ERRLEVEL_LOW,
168 						     ip->i_mount, dip);
169 				return -EFSCORRUPTED;
170 			}
171 
172 			size = (int)di_size;
173 			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
174 			break;
175 		case XFS_DINODE_FMT_EXTENTS:
176 			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
177 			break;
178 		case XFS_DINODE_FMT_BTREE:
179 			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
180 			break;
181 		default:
182 			XFS_ERROR_REPORT("xfs_iformat(6)", XFS_ERRLEVEL_LOW,
183 					 ip->i_mount);
184 			return -EFSCORRUPTED;
185 		}
186 		break;
187 
188 	default:
189 		XFS_ERROR_REPORT("xfs_iformat(7)", XFS_ERRLEVEL_LOW, ip->i_mount);
190 		return -EFSCORRUPTED;
191 	}
192 	if (error) {
193 		return error;
194 	}
195 	if (!XFS_DFORK_Q(dip))
196 		return 0;
197 
198 	ASSERT(ip->i_afp == NULL);
199 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
200 
201 	switch (dip->di_aformat) {
202 	case XFS_DINODE_FMT_LOCAL:
203 		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
204 		size = be16_to_cpu(atp->hdr.totsize);
205 
206 		if (unlikely(size < sizeof(struct xfs_attr_sf_hdr))) {
207 			xfs_warn(ip->i_mount,
208 				"corrupt inode %Lu (bad attr fork size %Ld).",
209 				(unsigned long long) ip->i_ino,
210 				(long long) size);
211 			XFS_CORRUPTION_ERROR("xfs_iformat(8)",
212 					     XFS_ERRLEVEL_LOW,
213 					     ip->i_mount, dip);
214 			return -EFSCORRUPTED;
215 		}
216 
217 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
218 		break;
219 	case XFS_DINODE_FMT_EXTENTS:
220 		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
221 		break;
222 	case XFS_DINODE_FMT_BTREE:
223 		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
224 		break;
225 	default:
226 		error = -EFSCORRUPTED;
227 		break;
228 	}
229 	if (error) {
230 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
231 		ip->i_afp = NULL;
232 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
233 	}
234 	return error;
235 }
236 
237 /*
238  * The file is in-lined in the on-disk inode.
239  * If it fits into if_inline_data, then copy
240  * it there, otherwise allocate a buffer for it
241  * and copy the data there.  Either way, set
242  * if_data to point at the data.
243  * If we allocate a buffer for the data, make
244  * sure that its size is a multiple of 4 and
245  * record the real size in i_real_bytes.
246  */
247 STATIC int
xfs_iformat_local(xfs_inode_t * ip,xfs_dinode_t * dip,int whichfork,int size)248 xfs_iformat_local(
249 	xfs_inode_t	*ip,
250 	xfs_dinode_t	*dip,
251 	int		whichfork,
252 	int		size)
253 {
254 	xfs_ifork_t	*ifp;
255 	int		real_size;
256 
257 	/*
258 	 * If the size is unreasonable, then something
259 	 * is wrong and we just bail out rather than crash in
260 	 * kmem_alloc() or memcpy() below.
261 	 */
262 	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
263 		xfs_warn(ip->i_mount,
264 	"corrupt inode %Lu (bad size %d for local fork, size = %d).",
265 			(unsigned long long) ip->i_ino, size,
266 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
267 		XFS_CORRUPTION_ERROR("xfs_iformat_local", XFS_ERRLEVEL_LOW,
268 				     ip->i_mount, dip);
269 		return -EFSCORRUPTED;
270 	}
271 	ifp = XFS_IFORK_PTR(ip, whichfork);
272 	real_size = 0;
273 	if (size == 0)
274 		ifp->if_u1.if_data = NULL;
275 	else if (size <= sizeof(ifp->if_u2.if_inline_data))
276 		ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
277 	else {
278 		real_size = roundup(size, 4);
279 		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
280 	}
281 	ifp->if_bytes = size;
282 	ifp->if_real_bytes = real_size;
283 	if (size)
284 		memcpy(ifp->if_u1.if_data, XFS_DFORK_PTR(dip, whichfork), size);
285 	ifp->if_flags &= ~XFS_IFEXTENTS;
286 	ifp->if_flags |= XFS_IFINLINE;
287 	return 0;
288 }
289 
290 /*
291  * The file consists of a set of extents all
292  * of which fit into the on-disk inode.
293  * If there are few enough extents to fit into
294  * the if_inline_ext, then copy them there.
295  * Otherwise allocate a buffer for them and copy
296  * them into it.  Either way, set if_extents
297  * to point at the extents.
298  */
299 STATIC int
xfs_iformat_extents(xfs_inode_t * ip,xfs_dinode_t * dip,int whichfork)300 xfs_iformat_extents(
301 	xfs_inode_t	*ip,
302 	xfs_dinode_t	*dip,
303 	int		whichfork)
304 {
305 	xfs_bmbt_rec_t	*dp;
306 	xfs_ifork_t	*ifp;
307 	int		nex;
308 	int		size;
309 	int		i;
310 
311 	ifp = XFS_IFORK_PTR(ip, whichfork);
312 	nex = XFS_DFORK_NEXTENTS(dip, whichfork);
313 	size = nex * (uint)sizeof(xfs_bmbt_rec_t);
314 
315 	/*
316 	 * If the number of extents is unreasonable, then something
317 	 * is wrong and we just bail out rather than crash in
318 	 * kmem_alloc() or memcpy() below.
319 	 */
320 	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
321 		xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
322 			(unsigned long long) ip->i_ino, nex);
323 		XFS_CORRUPTION_ERROR("xfs_iformat_extents(1)", XFS_ERRLEVEL_LOW,
324 				     ip->i_mount, dip);
325 		return -EFSCORRUPTED;
326 	}
327 
328 	ifp->if_real_bytes = 0;
329 	if (nex == 0)
330 		ifp->if_u1.if_extents = NULL;
331 	else if (nex <= XFS_INLINE_EXTS)
332 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
333 	else
334 		xfs_iext_add(ifp, 0, nex);
335 
336 	ifp->if_bytes = size;
337 	if (size) {
338 		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
339 		xfs_validate_extents(ifp, nex, XFS_EXTFMT_INODE(ip));
340 		for (i = 0; i < nex; i++, dp++) {
341 			xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
342 			ep->l0 = get_unaligned_be64(&dp->l0);
343 			ep->l1 = get_unaligned_be64(&dp->l1);
344 		}
345 		XFS_BMAP_TRACE_EXLIST(ip, nex, whichfork);
346 		if (whichfork != XFS_DATA_FORK ||
347 			XFS_EXTFMT_INODE(ip) == XFS_EXTFMT_NOSTATE)
348 				if (unlikely(xfs_check_nostate_extents(
349 				    ifp, 0, nex))) {
350 					XFS_ERROR_REPORT("xfs_iformat_extents(2)",
351 							 XFS_ERRLEVEL_LOW,
352 							 ip->i_mount);
353 					return -EFSCORRUPTED;
354 				}
355 	}
356 	ifp->if_flags |= XFS_IFEXTENTS;
357 	return 0;
358 }
359 
360 /*
361  * The file has too many extents to fit into
362  * the inode, so they are in B-tree format.
363  * Allocate a buffer for the root of the B-tree
364  * and copy the root into it.  The i_extents
365  * field will remain NULL until all of the
366  * extents are read in (when they are needed).
367  */
368 STATIC int
xfs_iformat_btree(xfs_inode_t * ip,xfs_dinode_t * dip,int whichfork)369 xfs_iformat_btree(
370 	xfs_inode_t		*ip,
371 	xfs_dinode_t		*dip,
372 	int			whichfork)
373 {
374 	struct xfs_mount	*mp = ip->i_mount;
375 	xfs_bmdr_block_t	*dfp;
376 	xfs_ifork_t		*ifp;
377 	/* REFERENCED */
378 	int			nrecs;
379 	int			size;
380 
381 	ifp = XFS_IFORK_PTR(ip, whichfork);
382 	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
383 	size = XFS_BMAP_BROOT_SPACE(mp, dfp);
384 	nrecs = be16_to_cpu(dfp->bb_numrecs);
385 
386 	/*
387 	 * blow out if -- fork has less extents than can fit in
388 	 * fork (fork shouldn't be a btree format), root btree
389 	 * block has more records than can fit into the fork,
390 	 * or the number of extents is greater than the number of
391 	 * blocks.
392 	 */
393 	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
394 					XFS_IFORK_MAXEXT(ip, whichfork) ||
395 		     XFS_BMDR_SPACE_CALC(nrecs) >
396 					XFS_DFORK_SIZE(dip, mp, whichfork) ||
397 		     XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks)) {
398 		xfs_warn(mp, "corrupt inode %Lu (btree).",
399 					(unsigned long long) ip->i_ino);
400 		XFS_CORRUPTION_ERROR("xfs_iformat_btree", XFS_ERRLEVEL_LOW,
401 					 mp, dip);
402 		return -EFSCORRUPTED;
403 	}
404 
405 	ifp->if_broot_bytes = size;
406 	ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
407 	ASSERT(ifp->if_broot != NULL);
408 	/*
409 	 * Copy and convert from the on-disk structure
410 	 * to the in-memory structure.
411 	 */
412 	xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
413 			 ifp->if_broot, size);
414 	ifp->if_flags &= ~XFS_IFEXTENTS;
415 	ifp->if_flags |= XFS_IFBROOT;
416 
417 	return 0;
418 }
419 
420 /*
421  * Read in extents from a btree-format inode.
422  * Allocate and fill in if_extents.  Real work is done in xfs_bmap.c.
423  */
424 int
xfs_iread_extents(xfs_trans_t * tp,xfs_inode_t * ip,int whichfork)425 xfs_iread_extents(
426 	xfs_trans_t	*tp,
427 	xfs_inode_t	*ip,
428 	int		whichfork)
429 {
430 	int		error;
431 	xfs_ifork_t	*ifp;
432 	xfs_extnum_t	nextents;
433 
434 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
435 
436 	if (unlikely(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
437 		XFS_ERROR_REPORT("xfs_iread_extents", XFS_ERRLEVEL_LOW,
438 				 ip->i_mount);
439 		return -EFSCORRUPTED;
440 	}
441 	nextents = XFS_IFORK_NEXTENTS(ip, whichfork);
442 	ifp = XFS_IFORK_PTR(ip, whichfork);
443 
444 	/*
445 	 * We know that the size is valid (it's checked in iformat_btree)
446 	 */
447 	ifp->if_bytes = ifp->if_real_bytes = 0;
448 	ifp->if_flags |= XFS_IFEXTENTS;
449 	xfs_iext_add(ifp, 0, nextents);
450 	error = xfs_bmap_read_extents(tp, ip, whichfork);
451 	if (error) {
452 		xfs_iext_destroy(ifp);
453 		ifp->if_flags &= ~XFS_IFEXTENTS;
454 		return error;
455 	}
456 	xfs_validate_extents(ifp, nextents, XFS_EXTFMT_INODE(ip));
457 	return 0;
458 }
459 /*
460  * Reallocate the space for if_broot based on the number of records
461  * being added or deleted as indicated in rec_diff.  Move the records
462  * and pointers in if_broot to fit the new size.  When shrinking this
463  * will eliminate holes between the records and pointers created by
464  * the caller.  When growing this will create holes to be filled in
465  * by the caller.
466  *
467  * The caller must not request to add more records than would fit in
468  * the on-disk inode root.  If the if_broot is currently NULL, then
469  * if we are adding records, one will be allocated.  The caller must also
470  * not request that the number of records go below zero, although
471  * it can go to zero.
472  *
473  * ip -- the inode whose if_broot area is changing
474  * ext_diff -- the change in the number of records, positive or negative,
475  *	 requested for the if_broot array.
476  */
477 void
xfs_iroot_realloc(xfs_inode_t * ip,int rec_diff,int whichfork)478 xfs_iroot_realloc(
479 	xfs_inode_t		*ip,
480 	int			rec_diff,
481 	int			whichfork)
482 {
483 	struct xfs_mount	*mp = ip->i_mount;
484 	int			cur_max;
485 	xfs_ifork_t		*ifp;
486 	struct xfs_btree_block	*new_broot;
487 	int			new_max;
488 	size_t			new_size;
489 	char			*np;
490 	char			*op;
491 
492 	/*
493 	 * Handle the degenerate case quietly.
494 	 */
495 	if (rec_diff == 0) {
496 		return;
497 	}
498 
499 	ifp = XFS_IFORK_PTR(ip, whichfork);
500 	if (rec_diff > 0) {
501 		/*
502 		 * If there wasn't any memory allocated before, just
503 		 * allocate it now and get out.
504 		 */
505 		if (ifp->if_broot_bytes == 0) {
506 			new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
507 			ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
508 			ifp->if_broot_bytes = (int)new_size;
509 			return;
510 		}
511 
512 		/*
513 		 * If there is already an existing if_broot, then we need
514 		 * to realloc() it and shift the pointers to their new
515 		 * location.  The records don't change location because
516 		 * they are kept butted up against the btree block header.
517 		 */
518 		cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
519 		new_max = cur_max + rec_diff;
520 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
521 		ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
522 				XFS_BMAP_BROOT_SPACE_CALC(mp, cur_max),
523 				KM_SLEEP | KM_NOFS);
524 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
525 						     ifp->if_broot_bytes);
526 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
527 						     (int)new_size);
528 		ifp->if_broot_bytes = (int)new_size;
529 		ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
530 			XFS_IFORK_SIZE(ip, whichfork));
531 		memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
532 		return;
533 	}
534 
535 	/*
536 	 * rec_diff is less than 0.  In this case, we are shrinking the
537 	 * if_broot buffer.  It must already exist.  If we go to zero
538 	 * records, just get rid of the root and clear the status bit.
539 	 */
540 	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
541 	cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
542 	new_max = cur_max + rec_diff;
543 	ASSERT(new_max >= 0);
544 	if (new_max > 0)
545 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
546 	else
547 		new_size = 0;
548 	if (new_size > 0) {
549 		new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
550 		/*
551 		 * First copy over the btree block header.
552 		 */
553 		memcpy(new_broot, ifp->if_broot,
554 			XFS_BMBT_BLOCK_LEN(ip->i_mount));
555 	} else {
556 		new_broot = NULL;
557 		ifp->if_flags &= ~XFS_IFBROOT;
558 	}
559 
560 	/*
561 	 * Only copy the records and pointers if there are any.
562 	 */
563 	if (new_max > 0) {
564 		/*
565 		 * First copy the records.
566 		 */
567 		op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
568 		np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
569 		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
570 
571 		/*
572 		 * Then copy the pointers.
573 		 */
574 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
575 						     ifp->if_broot_bytes);
576 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
577 						     (int)new_size);
578 		memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
579 	}
580 	kmem_free(ifp->if_broot);
581 	ifp->if_broot = new_broot;
582 	ifp->if_broot_bytes = (int)new_size;
583 	if (ifp->if_broot)
584 		ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
585 			XFS_IFORK_SIZE(ip, whichfork));
586 	return;
587 }
588 
589 
590 /*
591  * This is called when the amount of space needed for if_data
592  * is increased or decreased.  The change in size is indicated by
593  * the number of bytes that need to be added or deleted in the
594  * byte_diff parameter.
595  *
596  * If the amount of space needed has decreased below the size of the
597  * inline buffer, then switch to using the inline buffer.  Otherwise,
598  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
599  * to what is needed.
600  *
601  * ip -- the inode whose if_data area is changing
602  * byte_diff -- the change in the number of bytes, positive or negative,
603  *	 requested for the if_data array.
604  */
605 void
xfs_idata_realloc(xfs_inode_t * ip,int byte_diff,int whichfork)606 xfs_idata_realloc(
607 	xfs_inode_t	*ip,
608 	int		byte_diff,
609 	int		whichfork)
610 {
611 	xfs_ifork_t	*ifp;
612 	int		new_size;
613 	int		real_size;
614 
615 	if (byte_diff == 0) {
616 		return;
617 	}
618 
619 	ifp = XFS_IFORK_PTR(ip, whichfork);
620 	new_size = (int)ifp->if_bytes + byte_diff;
621 	ASSERT(new_size >= 0);
622 
623 	if (new_size == 0) {
624 		if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
625 			kmem_free(ifp->if_u1.if_data);
626 		}
627 		ifp->if_u1.if_data = NULL;
628 		real_size = 0;
629 	} else if (new_size <= sizeof(ifp->if_u2.if_inline_data)) {
630 		/*
631 		 * If the valid extents/data can fit in if_inline_ext/data,
632 		 * copy them from the malloc'd vector and free it.
633 		 */
634 		if (ifp->if_u1.if_data == NULL) {
635 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
636 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
637 			ASSERT(ifp->if_real_bytes != 0);
638 			memcpy(ifp->if_u2.if_inline_data, ifp->if_u1.if_data,
639 			      new_size);
640 			kmem_free(ifp->if_u1.if_data);
641 			ifp->if_u1.if_data = ifp->if_u2.if_inline_data;
642 		}
643 		real_size = 0;
644 	} else {
645 		/*
646 		 * Stuck with malloc/realloc.
647 		 * For inline data, the underlying buffer must be
648 		 * a multiple of 4 bytes in size so that it can be
649 		 * logged and stay on word boundaries.  We enforce
650 		 * that here.
651 		 */
652 		real_size = roundup(new_size, 4);
653 		if (ifp->if_u1.if_data == NULL) {
654 			ASSERT(ifp->if_real_bytes == 0);
655 			ifp->if_u1.if_data = kmem_alloc(real_size,
656 							KM_SLEEP | KM_NOFS);
657 		} else if (ifp->if_u1.if_data != ifp->if_u2.if_inline_data) {
658 			/*
659 			 * Only do the realloc if the underlying size
660 			 * is really changing.
661 			 */
662 			if (ifp->if_real_bytes != real_size) {
663 				ifp->if_u1.if_data =
664 					kmem_realloc(ifp->if_u1.if_data,
665 							real_size,
666 							ifp->if_real_bytes,
667 							KM_SLEEP | KM_NOFS);
668 			}
669 		} else {
670 			ASSERT(ifp->if_real_bytes == 0);
671 			ifp->if_u1.if_data = kmem_alloc(real_size,
672 							KM_SLEEP | KM_NOFS);
673 			memcpy(ifp->if_u1.if_data, ifp->if_u2.if_inline_data,
674 				ifp->if_bytes);
675 		}
676 	}
677 	ifp->if_real_bytes = real_size;
678 	ifp->if_bytes = new_size;
679 	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
680 }
681 
682 void
xfs_idestroy_fork(xfs_inode_t * ip,int whichfork)683 xfs_idestroy_fork(
684 	xfs_inode_t	*ip,
685 	int		whichfork)
686 {
687 	xfs_ifork_t	*ifp;
688 
689 	ifp = XFS_IFORK_PTR(ip, whichfork);
690 	if (ifp->if_broot != NULL) {
691 		kmem_free(ifp->if_broot);
692 		ifp->if_broot = NULL;
693 	}
694 
695 	/*
696 	 * If the format is local, then we can't have an extents
697 	 * array so just look for an inline data array.  If we're
698 	 * not local then we may or may not have an extents list,
699 	 * so check and free it up if we do.
700 	 */
701 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
702 		if ((ifp->if_u1.if_data != ifp->if_u2.if_inline_data) &&
703 		    (ifp->if_u1.if_data != NULL)) {
704 			ASSERT(ifp->if_real_bytes != 0);
705 			kmem_free(ifp->if_u1.if_data);
706 			ifp->if_u1.if_data = NULL;
707 			ifp->if_real_bytes = 0;
708 		}
709 	} else if ((ifp->if_flags & XFS_IFEXTENTS) &&
710 		   ((ifp->if_flags & XFS_IFEXTIREC) ||
711 		    ((ifp->if_u1.if_extents != NULL) &&
712 		     (ifp->if_u1.if_extents != ifp->if_u2.if_inline_ext)))) {
713 		ASSERT(ifp->if_real_bytes != 0);
714 		xfs_iext_destroy(ifp);
715 	}
716 	ASSERT(ifp->if_u1.if_extents == NULL ||
717 	       ifp->if_u1.if_extents == ifp->if_u2.if_inline_ext);
718 	ASSERT(ifp->if_real_bytes == 0);
719 	if (whichfork == XFS_ATTR_FORK) {
720 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
721 		ip->i_afp = NULL;
722 	}
723 }
724 
725 /*
726  * Convert in-core extents to on-disk form
727  *
728  * For either the data or attr fork in extent format, we need to endian convert
729  * the in-core extent as we place them into the on-disk inode.
730  *
731  * In the case of the data fork, the in-core and on-disk fork sizes can be
732  * different due to delayed allocation extents. We only copy on-disk extents
733  * here, so callers must always use the physical fork size to determine the
734  * size of the buffer passed to this routine.  We will return the size actually
735  * used.
736  */
737 int
xfs_iextents_copy(xfs_inode_t * ip,xfs_bmbt_rec_t * dp,int whichfork)738 xfs_iextents_copy(
739 	xfs_inode_t		*ip,
740 	xfs_bmbt_rec_t		*dp,
741 	int			whichfork)
742 {
743 	int			copied;
744 	int			i;
745 	xfs_ifork_t		*ifp;
746 	int			nrecs;
747 	xfs_fsblock_t		start_block;
748 
749 	ifp = XFS_IFORK_PTR(ip, whichfork);
750 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
751 	ASSERT(ifp->if_bytes > 0);
752 
753 	nrecs = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
754 	XFS_BMAP_TRACE_EXLIST(ip, nrecs, whichfork);
755 	ASSERT(nrecs > 0);
756 
757 	/*
758 	 * There are some delayed allocation extents in the
759 	 * inode, so copy the extents one at a time and skip
760 	 * the delayed ones.  There must be at least one
761 	 * non-delayed extent.
762 	 */
763 	copied = 0;
764 	for (i = 0; i < nrecs; i++) {
765 		xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, i);
766 		start_block = xfs_bmbt_get_startblock(ep);
767 		if (isnullstartblock(start_block)) {
768 			/*
769 			 * It's a delayed allocation extent, so skip it.
770 			 */
771 			continue;
772 		}
773 
774 		/* Translate to on disk format */
775 		put_unaligned_be64(ep->l0, &dp->l0);
776 		put_unaligned_be64(ep->l1, &dp->l1);
777 		dp++;
778 		copied++;
779 	}
780 	ASSERT(copied != 0);
781 	xfs_validate_extents(ifp, copied, XFS_EXTFMT_INODE(ip));
782 
783 	return (copied * (uint)sizeof(xfs_bmbt_rec_t));
784 }
785 
786 /*
787  * Each of the following cases stores data into the same region
788  * of the on-disk inode, so only one of them can be valid at
789  * any given time. While it is possible to have conflicting formats
790  * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
791  * in EXTENTS format, this can only happen when the fork has
792  * changed formats after being modified but before being flushed.
793  * In these cases, the format always takes precedence, because the
794  * format indicates the current state of the fork.
795  */
796 void
xfs_iflush_fork(xfs_inode_t * ip,xfs_dinode_t * dip,xfs_inode_log_item_t * iip,int whichfork)797 xfs_iflush_fork(
798 	xfs_inode_t		*ip,
799 	xfs_dinode_t		*dip,
800 	xfs_inode_log_item_t	*iip,
801 	int			whichfork)
802 {
803 	char			*cp;
804 	xfs_ifork_t		*ifp;
805 	xfs_mount_t		*mp;
806 	static const short	brootflag[2] =
807 		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
808 	static const short	dataflag[2] =
809 		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
810 	static const short	extflag[2] =
811 		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
812 
813 	if (!iip)
814 		return;
815 	ifp = XFS_IFORK_PTR(ip, whichfork);
816 	/*
817 	 * This can happen if we gave up in iformat in an error path,
818 	 * for the attribute fork.
819 	 */
820 	if (!ifp) {
821 		ASSERT(whichfork == XFS_ATTR_FORK);
822 		return;
823 	}
824 	cp = XFS_DFORK_PTR(dip, whichfork);
825 	mp = ip->i_mount;
826 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
827 	case XFS_DINODE_FMT_LOCAL:
828 		if ((iip->ili_fields & dataflag[whichfork]) &&
829 		    (ifp->if_bytes > 0)) {
830 			ASSERT(ifp->if_u1.if_data != NULL);
831 			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
832 			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
833 		}
834 		break;
835 
836 	case XFS_DINODE_FMT_EXTENTS:
837 		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
838 		       !(iip->ili_fields & extflag[whichfork]));
839 		if ((iip->ili_fields & extflag[whichfork]) &&
840 		    (ifp->if_bytes > 0)) {
841 			ASSERT(xfs_iext_get_ext(ifp, 0));
842 			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
843 			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
844 				whichfork);
845 		}
846 		break;
847 
848 	case XFS_DINODE_FMT_BTREE:
849 		if ((iip->ili_fields & brootflag[whichfork]) &&
850 		    (ifp->if_broot_bytes > 0)) {
851 			ASSERT(ifp->if_broot != NULL);
852 			ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
853 			        XFS_IFORK_SIZE(ip, whichfork));
854 			xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
855 				(xfs_bmdr_block_t *)cp,
856 				XFS_DFORK_SIZE(dip, mp, whichfork));
857 		}
858 		break;
859 
860 	case XFS_DINODE_FMT_DEV:
861 		if (iip->ili_fields & XFS_ILOG_DEV) {
862 			ASSERT(whichfork == XFS_DATA_FORK);
863 			xfs_dinode_put_rdev(dip, ip->i_df.if_u2.if_rdev);
864 		}
865 		break;
866 
867 	case XFS_DINODE_FMT_UUID:
868 		if (iip->ili_fields & XFS_ILOG_UUID) {
869 			ASSERT(whichfork == XFS_DATA_FORK);
870 			memcpy(XFS_DFORK_DPTR(dip),
871 			       &ip->i_df.if_u2.if_uuid,
872 			       sizeof(uuid_t));
873 		}
874 		break;
875 
876 	default:
877 		ASSERT(0);
878 		break;
879 	}
880 }
881 
882 /*
883  * Return a pointer to the extent record at file index idx.
884  */
885 xfs_bmbt_rec_host_t *
xfs_iext_get_ext(xfs_ifork_t * ifp,xfs_extnum_t idx)886 xfs_iext_get_ext(
887 	xfs_ifork_t	*ifp,		/* inode fork pointer */
888 	xfs_extnum_t	idx)		/* index of target extent */
889 {
890 	ASSERT(idx >= 0);
891 	ASSERT(idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
892 
893 	if ((ifp->if_flags & XFS_IFEXTIREC) && (idx == 0)) {
894 		return ifp->if_u1.if_ext_irec->er_extbuf;
895 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
896 		xfs_ext_irec_t	*erp;		/* irec pointer */
897 		int		erp_idx = 0;	/* irec index */
898 		xfs_extnum_t	page_idx = idx;	/* ext index in target list */
899 
900 		erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 0);
901 		return &erp->er_extbuf[page_idx];
902 	} else if (ifp->if_bytes) {
903 		return &ifp->if_u1.if_extents[idx];
904 	} else {
905 		return NULL;
906 	}
907 }
908 
909 /*
910  * Insert new item(s) into the extent records for incore inode
911  * fork 'ifp'.  'count' new items are inserted at index 'idx'.
912  */
913 void
xfs_iext_insert(xfs_inode_t * ip,xfs_extnum_t idx,xfs_extnum_t count,xfs_bmbt_irec_t * new,int state)914 xfs_iext_insert(
915 	xfs_inode_t	*ip,		/* incore inode pointer */
916 	xfs_extnum_t	idx,		/* starting index of new items */
917 	xfs_extnum_t	count,		/* number of inserted items */
918 	xfs_bmbt_irec_t	*new,		/* items to insert */
919 	int		state)		/* type of extent conversion */
920 {
921 	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
922 	xfs_extnum_t	i;		/* extent record index */
923 
924 	trace_xfs_iext_insert(ip, idx, new, state, _RET_IP_);
925 
926 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
927 	xfs_iext_add(ifp, idx, count);
928 	for (i = idx; i < idx + count; i++, new++)
929 		xfs_bmbt_set_all(xfs_iext_get_ext(ifp, i), new);
930 }
931 
932 /*
933  * This is called when the amount of space required for incore file
934  * extents needs to be increased. The ext_diff parameter stores the
935  * number of new extents being added and the idx parameter contains
936  * the extent index where the new extents will be added. If the new
937  * extents are being appended, then we just need to (re)allocate and
938  * initialize the space. Otherwise, if the new extents are being
939  * inserted into the middle of the existing entries, a bit more work
940  * is required to make room for the new extents to be inserted. The
941  * caller is responsible for filling in the new extent entries upon
942  * return.
943  */
944 void
xfs_iext_add(xfs_ifork_t * ifp,xfs_extnum_t idx,int ext_diff)945 xfs_iext_add(
946 	xfs_ifork_t	*ifp,		/* inode fork pointer */
947 	xfs_extnum_t	idx,		/* index to begin adding exts */
948 	int		ext_diff)	/* number of extents to add */
949 {
950 	int		byte_diff;	/* new bytes being added */
951 	int		new_size;	/* size of extents after adding */
952 	xfs_extnum_t	nextents;	/* number of extents in file */
953 
954 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
955 	ASSERT((idx >= 0) && (idx <= nextents));
956 	byte_diff = ext_diff * sizeof(xfs_bmbt_rec_t);
957 	new_size = ifp->if_bytes + byte_diff;
958 	/*
959 	 * If the new number of extents (nextents + ext_diff)
960 	 * fits inside the inode, then continue to use the inline
961 	 * extent buffer.
962 	 */
963 	if (nextents + ext_diff <= XFS_INLINE_EXTS) {
964 		if (idx < nextents) {
965 			memmove(&ifp->if_u2.if_inline_ext[idx + ext_diff],
966 				&ifp->if_u2.if_inline_ext[idx],
967 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
968 			memset(&ifp->if_u2.if_inline_ext[idx], 0, byte_diff);
969 		}
970 		ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
971 		ifp->if_real_bytes = 0;
972 	}
973 	/*
974 	 * Otherwise use a linear (direct) extent list.
975 	 * If the extents are currently inside the inode,
976 	 * xfs_iext_realloc_direct will switch us from
977 	 * inline to direct extent allocation mode.
978 	 */
979 	else if (nextents + ext_diff <= XFS_LINEAR_EXTS) {
980 		xfs_iext_realloc_direct(ifp, new_size);
981 		if (idx < nextents) {
982 			memmove(&ifp->if_u1.if_extents[idx + ext_diff],
983 				&ifp->if_u1.if_extents[idx],
984 				(nextents - idx) * sizeof(xfs_bmbt_rec_t));
985 			memset(&ifp->if_u1.if_extents[idx], 0, byte_diff);
986 		}
987 	}
988 	/* Indirection array */
989 	else {
990 		xfs_ext_irec_t	*erp;
991 		int		erp_idx = 0;
992 		int		page_idx = idx;
993 
994 		ASSERT(nextents + ext_diff > XFS_LINEAR_EXTS);
995 		if (ifp->if_flags & XFS_IFEXTIREC) {
996 			erp = xfs_iext_idx_to_irec(ifp, &page_idx, &erp_idx, 1);
997 		} else {
998 			xfs_iext_irec_init(ifp);
999 			ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1000 			erp = ifp->if_u1.if_ext_irec;
1001 		}
1002 		/* Extents fit in target extent page */
1003 		if (erp && erp->er_extcount + ext_diff <= XFS_LINEAR_EXTS) {
1004 			if (page_idx < erp->er_extcount) {
1005 				memmove(&erp->er_extbuf[page_idx + ext_diff],
1006 					&erp->er_extbuf[page_idx],
1007 					(erp->er_extcount - page_idx) *
1008 					sizeof(xfs_bmbt_rec_t));
1009 				memset(&erp->er_extbuf[page_idx], 0, byte_diff);
1010 			}
1011 			erp->er_extcount += ext_diff;
1012 			xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1013 		}
1014 		/* Insert a new extent page */
1015 		else if (erp) {
1016 			xfs_iext_add_indirect_multi(ifp,
1017 				erp_idx, page_idx, ext_diff);
1018 		}
1019 		/*
1020 		 * If extent(s) are being appended to the last page in
1021 		 * the indirection array and the new extent(s) don't fit
1022 		 * in the page, then erp is NULL and erp_idx is set to
1023 		 * the next index needed in the indirection array.
1024 		 */
1025 		else {
1026 			uint	count = ext_diff;
1027 
1028 			while (count) {
1029 				erp = xfs_iext_irec_new(ifp, erp_idx);
1030 				erp->er_extcount = min(count, XFS_LINEAR_EXTS);
1031 				count -= erp->er_extcount;
1032 				if (count)
1033 					erp_idx++;
1034 			}
1035 		}
1036 	}
1037 	ifp->if_bytes = new_size;
1038 }
1039 
1040 /*
1041  * This is called when incore extents are being added to the indirection
1042  * array and the new extents do not fit in the target extent list. The
1043  * erp_idx parameter contains the irec index for the target extent list
1044  * in the indirection array, and the idx parameter contains the extent
1045  * index within the list. The number of extents being added is stored
1046  * in the count parameter.
1047  *
1048  *    |-------|   |-------|
1049  *    |       |   |       |    idx - number of extents before idx
1050  *    |  idx  |   | count |
1051  *    |       |   |       |    count - number of extents being inserted at idx
1052  *    |-------|   |-------|
1053  *    | count |   | nex2  |    nex2 - number of extents after idx + count
1054  *    |-------|   |-------|
1055  */
1056 void
xfs_iext_add_indirect_multi(xfs_ifork_t * ifp,int erp_idx,xfs_extnum_t idx,int count)1057 xfs_iext_add_indirect_multi(
1058 	xfs_ifork_t	*ifp,			/* inode fork pointer */
1059 	int		erp_idx,		/* target extent irec index */
1060 	xfs_extnum_t	idx,			/* index within target list */
1061 	int		count)			/* new extents being added */
1062 {
1063 	int		byte_diff;		/* new bytes being added */
1064 	xfs_ext_irec_t	*erp;			/* pointer to irec entry */
1065 	xfs_extnum_t	ext_diff;		/* number of extents to add */
1066 	xfs_extnum_t	ext_cnt;		/* new extents still needed */
1067 	xfs_extnum_t	nex2;			/* extents after idx + count */
1068 	xfs_bmbt_rec_t	*nex2_ep = NULL;	/* temp list for nex2 extents */
1069 	int		nlists;			/* number of irec's (lists) */
1070 
1071 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1072 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
1073 	nex2 = erp->er_extcount - idx;
1074 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1075 
1076 	/*
1077 	 * Save second part of target extent list
1078 	 * (all extents past */
1079 	if (nex2) {
1080 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
1081 		nex2_ep = (xfs_bmbt_rec_t *) kmem_alloc(byte_diff, KM_NOFS);
1082 		memmove(nex2_ep, &erp->er_extbuf[idx], byte_diff);
1083 		erp->er_extcount -= nex2;
1084 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -nex2);
1085 		memset(&erp->er_extbuf[idx], 0, byte_diff);
1086 	}
1087 
1088 	/*
1089 	 * Add the new extents to the end of the target
1090 	 * list, then allocate new irec record(s) and
1091 	 * extent buffer(s) as needed to store the rest
1092 	 * of the new extents.
1093 	 */
1094 	ext_cnt = count;
1095 	ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS - erp->er_extcount);
1096 	if (ext_diff) {
1097 		erp->er_extcount += ext_diff;
1098 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1099 		ext_cnt -= ext_diff;
1100 	}
1101 	while (ext_cnt) {
1102 		erp_idx++;
1103 		erp = xfs_iext_irec_new(ifp, erp_idx);
1104 		ext_diff = MIN(ext_cnt, (int)XFS_LINEAR_EXTS);
1105 		erp->er_extcount = ext_diff;
1106 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, ext_diff);
1107 		ext_cnt -= ext_diff;
1108 	}
1109 
1110 	/* Add nex2 extents back to indirection array */
1111 	if (nex2) {
1112 		xfs_extnum_t	ext_avail;
1113 		int		i;
1114 
1115 		byte_diff = nex2 * sizeof(xfs_bmbt_rec_t);
1116 		ext_avail = XFS_LINEAR_EXTS - erp->er_extcount;
1117 		i = 0;
1118 		/*
1119 		 * If nex2 extents fit in the current page, append
1120 		 * nex2_ep after the new extents.
1121 		 */
1122 		if (nex2 <= ext_avail) {
1123 			i = erp->er_extcount;
1124 		}
1125 		/*
1126 		 * Otherwise, check if space is available in the
1127 		 * next page.
1128 		 */
1129 		else if ((erp_idx < nlists - 1) &&
1130 			 (nex2 <= (ext_avail = XFS_LINEAR_EXTS -
1131 			  ifp->if_u1.if_ext_irec[erp_idx+1].er_extcount))) {
1132 			erp_idx++;
1133 			erp++;
1134 			/* Create a hole for nex2 extents */
1135 			memmove(&erp->er_extbuf[nex2], erp->er_extbuf,
1136 				erp->er_extcount * sizeof(xfs_bmbt_rec_t));
1137 		}
1138 		/*
1139 		 * Final choice, create a new extent page for
1140 		 * nex2 extents.
1141 		 */
1142 		else {
1143 			erp_idx++;
1144 			erp = xfs_iext_irec_new(ifp, erp_idx);
1145 		}
1146 		memmove(&erp->er_extbuf[i], nex2_ep, byte_diff);
1147 		kmem_free(nex2_ep);
1148 		erp->er_extcount += nex2;
1149 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, nex2);
1150 	}
1151 }
1152 
1153 /*
1154  * This is called when the amount of space required for incore file
1155  * extents needs to be decreased. The ext_diff parameter stores the
1156  * number of extents to be removed and the idx parameter contains
1157  * the extent index where the extents will be removed from.
1158  *
1159  * If the amount of space needed has decreased below the linear
1160  * limit, XFS_IEXT_BUFSZ, then switch to using the contiguous
1161  * extent array.  Otherwise, use kmem_realloc() to adjust the
1162  * size to what is needed.
1163  */
1164 void
xfs_iext_remove(xfs_inode_t * ip,xfs_extnum_t idx,int ext_diff,int state)1165 xfs_iext_remove(
1166 	xfs_inode_t	*ip,		/* incore inode pointer */
1167 	xfs_extnum_t	idx,		/* index to begin removing exts */
1168 	int		ext_diff,	/* number of extents to remove */
1169 	int		state)		/* type of extent conversion */
1170 {
1171 	xfs_ifork_t	*ifp = (state & BMAP_ATTRFORK) ? ip->i_afp : &ip->i_df;
1172 	xfs_extnum_t	nextents;	/* number of extents in file */
1173 	int		new_size;	/* size of extents after removal */
1174 
1175 	trace_xfs_iext_remove(ip, idx, state, _RET_IP_);
1176 
1177 	ASSERT(ext_diff > 0);
1178 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1179 	new_size = (nextents - ext_diff) * sizeof(xfs_bmbt_rec_t);
1180 
1181 	if (new_size == 0) {
1182 		xfs_iext_destroy(ifp);
1183 	} else if (ifp->if_flags & XFS_IFEXTIREC) {
1184 		xfs_iext_remove_indirect(ifp, idx, ext_diff);
1185 	} else if (ifp->if_real_bytes) {
1186 		xfs_iext_remove_direct(ifp, idx, ext_diff);
1187 	} else {
1188 		xfs_iext_remove_inline(ifp, idx, ext_diff);
1189 	}
1190 	ifp->if_bytes = new_size;
1191 }
1192 
1193 /*
1194  * This removes ext_diff extents from the inline buffer, beginning
1195  * at extent index idx.
1196  */
1197 void
xfs_iext_remove_inline(xfs_ifork_t * ifp,xfs_extnum_t idx,int ext_diff)1198 xfs_iext_remove_inline(
1199 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1200 	xfs_extnum_t	idx,		/* index to begin removing exts */
1201 	int		ext_diff)	/* number of extents to remove */
1202 {
1203 	int		nextents;	/* number of extents in file */
1204 
1205 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1206 	ASSERT(idx < XFS_INLINE_EXTS);
1207 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1208 	ASSERT(((nextents - ext_diff) > 0) &&
1209 		(nextents - ext_diff) < XFS_INLINE_EXTS);
1210 
1211 	if (idx + ext_diff < nextents) {
1212 		memmove(&ifp->if_u2.if_inline_ext[idx],
1213 			&ifp->if_u2.if_inline_ext[idx + ext_diff],
1214 			(nextents - (idx + ext_diff)) *
1215 			 sizeof(xfs_bmbt_rec_t));
1216 		memset(&ifp->if_u2.if_inline_ext[nextents - ext_diff],
1217 			0, ext_diff * sizeof(xfs_bmbt_rec_t));
1218 	} else {
1219 		memset(&ifp->if_u2.if_inline_ext[idx], 0,
1220 			ext_diff * sizeof(xfs_bmbt_rec_t));
1221 	}
1222 }
1223 
1224 /*
1225  * This removes ext_diff extents from a linear (direct) extent list,
1226  * beginning at extent index idx. If the extents are being removed
1227  * from the end of the list (ie. truncate) then we just need to re-
1228  * allocate the list to remove the extra space. Otherwise, if the
1229  * extents are being removed from the middle of the existing extent
1230  * entries, then we first need to move the extent records beginning
1231  * at idx + ext_diff up in the list to overwrite the records being
1232  * removed, then remove the extra space via kmem_realloc.
1233  */
1234 void
xfs_iext_remove_direct(xfs_ifork_t * ifp,xfs_extnum_t idx,int ext_diff)1235 xfs_iext_remove_direct(
1236 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1237 	xfs_extnum_t	idx,		/* index to begin removing exts */
1238 	int		ext_diff)	/* number of extents to remove */
1239 {
1240 	xfs_extnum_t	nextents;	/* number of extents in file */
1241 	int		new_size;	/* size of extents after removal */
1242 
1243 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1244 	new_size = ifp->if_bytes -
1245 		(ext_diff * sizeof(xfs_bmbt_rec_t));
1246 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1247 
1248 	if (new_size == 0) {
1249 		xfs_iext_destroy(ifp);
1250 		return;
1251 	}
1252 	/* Move extents up in the list (if needed) */
1253 	if (idx + ext_diff < nextents) {
1254 		memmove(&ifp->if_u1.if_extents[idx],
1255 			&ifp->if_u1.if_extents[idx + ext_diff],
1256 			(nextents - (idx + ext_diff)) *
1257 			 sizeof(xfs_bmbt_rec_t));
1258 	}
1259 	memset(&ifp->if_u1.if_extents[nextents - ext_diff],
1260 		0, ext_diff * sizeof(xfs_bmbt_rec_t));
1261 	/*
1262 	 * Reallocate the direct extent list. If the extents
1263 	 * will fit inside the inode then xfs_iext_realloc_direct
1264 	 * will switch from direct to inline extent allocation
1265 	 * mode for us.
1266 	 */
1267 	xfs_iext_realloc_direct(ifp, new_size);
1268 	ifp->if_bytes = new_size;
1269 }
1270 
1271 /*
1272  * This is called when incore extents are being removed from the
1273  * indirection array and the extents being removed span multiple extent
1274  * buffers. The idx parameter contains the file extent index where we
1275  * want to begin removing extents, and the count parameter contains
1276  * how many extents need to be removed.
1277  *
1278  *    |-------|   |-------|
1279  *    | nex1  |   |       |    nex1 - number of extents before idx
1280  *    |-------|   | count |
1281  *    |       |   |       |    count - number of extents being removed at idx
1282  *    | count |   |-------|
1283  *    |       |   | nex2  |    nex2 - number of extents after idx + count
1284  *    |-------|   |-------|
1285  */
1286 void
xfs_iext_remove_indirect(xfs_ifork_t * ifp,xfs_extnum_t idx,int count)1287 xfs_iext_remove_indirect(
1288 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1289 	xfs_extnum_t	idx,		/* index to begin removing extents */
1290 	int		count)		/* number of extents to remove */
1291 {
1292 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
1293 	int		erp_idx = 0;	/* indirection array index */
1294 	xfs_extnum_t	ext_cnt;	/* extents left to remove */
1295 	xfs_extnum_t	ext_diff;	/* extents to remove in current list */
1296 	xfs_extnum_t	nex1;		/* number of extents before idx */
1297 	xfs_extnum_t	nex2;		/* extents after idx + count */
1298 	int		page_idx = idx;	/* index in target extent list */
1299 
1300 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1301 	erp = xfs_iext_idx_to_irec(ifp,  &page_idx, &erp_idx, 0);
1302 	ASSERT(erp != NULL);
1303 	nex1 = page_idx;
1304 	ext_cnt = count;
1305 	while (ext_cnt) {
1306 		nex2 = MAX((erp->er_extcount - (nex1 + ext_cnt)), 0);
1307 		ext_diff = MIN(ext_cnt, (erp->er_extcount - nex1));
1308 		/*
1309 		 * Check for deletion of entire list;
1310 		 * xfs_iext_irec_remove() updates extent offsets.
1311 		 */
1312 		if (ext_diff == erp->er_extcount) {
1313 			xfs_iext_irec_remove(ifp, erp_idx);
1314 			ext_cnt -= ext_diff;
1315 			nex1 = 0;
1316 			if (ext_cnt) {
1317 				ASSERT(erp_idx < ifp->if_real_bytes /
1318 					XFS_IEXT_BUFSZ);
1319 				erp = &ifp->if_u1.if_ext_irec[erp_idx];
1320 				nex1 = 0;
1321 				continue;
1322 			} else {
1323 				break;
1324 			}
1325 		}
1326 		/* Move extents up (if needed) */
1327 		if (nex2) {
1328 			memmove(&erp->er_extbuf[nex1],
1329 				&erp->er_extbuf[nex1 + ext_diff],
1330 				nex2 * sizeof(xfs_bmbt_rec_t));
1331 		}
1332 		/* Zero out rest of page */
1333 		memset(&erp->er_extbuf[nex1 + nex2], 0, (XFS_IEXT_BUFSZ -
1334 			((nex1 + nex2) * sizeof(xfs_bmbt_rec_t))));
1335 		/* Update remaining counters */
1336 		erp->er_extcount -= ext_diff;
1337 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1, -ext_diff);
1338 		ext_cnt -= ext_diff;
1339 		nex1 = 0;
1340 		erp_idx++;
1341 		erp++;
1342 	}
1343 	ifp->if_bytes -= count * sizeof(xfs_bmbt_rec_t);
1344 	xfs_iext_irec_compact(ifp);
1345 }
1346 
1347 /*
1348  * Create, destroy, or resize a linear (direct) block of extents.
1349  */
1350 void
xfs_iext_realloc_direct(xfs_ifork_t * ifp,int new_size)1351 xfs_iext_realloc_direct(
1352 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1353 	int		new_size)	/* new size of extents after adding */
1354 {
1355 	int		rnew_size;	/* real new size of extents */
1356 
1357 	rnew_size = new_size;
1358 
1359 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC) ||
1360 		((new_size >= 0) && (new_size <= XFS_IEXT_BUFSZ) &&
1361 		 (new_size != ifp->if_real_bytes)));
1362 
1363 	/* Free extent records */
1364 	if (new_size == 0) {
1365 		xfs_iext_destroy(ifp);
1366 	}
1367 	/* Resize direct extent list and zero any new bytes */
1368 	else if (ifp->if_real_bytes) {
1369 		/* Check if extents will fit inside the inode */
1370 		if (new_size <= XFS_INLINE_EXTS * sizeof(xfs_bmbt_rec_t)) {
1371 			xfs_iext_direct_to_inline(ifp, new_size /
1372 				(uint)sizeof(xfs_bmbt_rec_t));
1373 			ifp->if_bytes = new_size;
1374 			return;
1375 		}
1376 		if (!is_power_of_2(new_size)){
1377 			rnew_size = roundup_pow_of_two(new_size);
1378 		}
1379 		if (rnew_size != ifp->if_real_bytes) {
1380 			ifp->if_u1.if_extents =
1381 				kmem_realloc(ifp->if_u1.if_extents,
1382 						rnew_size,
1383 						ifp->if_real_bytes, KM_NOFS);
1384 		}
1385 		if (rnew_size > ifp->if_real_bytes) {
1386 			memset(&ifp->if_u1.if_extents[ifp->if_bytes /
1387 				(uint)sizeof(xfs_bmbt_rec_t)], 0,
1388 				rnew_size - ifp->if_real_bytes);
1389 		}
1390 	}
1391 	/* Switch from the inline extent buffer to a direct extent list */
1392 	else {
1393 		if (!is_power_of_2(new_size)) {
1394 			rnew_size = roundup_pow_of_two(new_size);
1395 		}
1396 		xfs_iext_inline_to_direct(ifp, rnew_size);
1397 	}
1398 	ifp->if_real_bytes = rnew_size;
1399 	ifp->if_bytes = new_size;
1400 }
1401 
1402 /*
1403  * Switch from linear (direct) extent records to inline buffer.
1404  */
1405 void
xfs_iext_direct_to_inline(xfs_ifork_t * ifp,xfs_extnum_t nextents)1406 xfs_iext_direct_to_inline(
1407 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1408 	xfs_extnum_t	nextents)	/* number of extents in file */
1409 {
1410 	ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1411 	ASSERT(nextents <= XFS_INLINE_EXTS);
1412 	/*
1413 	 * The inline buffer was zeroed when we switched
1414 	 * from inline to direct extent allocation mode,
1415 	 * so we don't need to clear it here.
1416 	 */
1417 	memcpy(ifp->if_u2.if_inline_ext, ifp->if_u1.if_extents,
1418 		nextents * sizeof(xfs_bmbt_rec_t));
1419 	kmem_free(ifp->if_u1.if_extents);
1420 	ifp->if_u1.if_extents = ifp->if_u2.if_inline_ext;
1421 	ifp->if_real_bytes = 0;
1422 }
1423 
1424 /*
1425  * Switch from inline buffer to linear (direct) extent records.
1426  * new_size should already be rounded up to the next power of 2
1427  * by the caller (when appropriate), so use new_size as it is.
1428  * However, since new_size may be rounded up, we can't update
1429  * if_bytes here. It is the caller's responsibility to update
1430  * if_bytes upon return.
1431  */
1432 void
xfs_iext_inline_to_direct(xfs_ifork_t * ifp,int new_size)1433 xfs_iext_inline_to_direct(
1434 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1435 	int		new_size)	/* number of extents in file */
1436 {
1437 	ifp->if_u1.if_extents = kmem_alloc(new_size, KM_NOFS);
1438 	memset(ifp->if_u1.if_extents, 0, new_size);
1439 	if (ifp->if_bytes) {
1440 		memcpy(ifp->if_u1.if_extents, ifp->if_u2.if_inline_ext,
1441 			ifp->if_bytes);
1442 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
1443 			sizeof(xfs_bmbt_rec_t));
1444 	}
1445 	ifp->if_real_bytes = new_size;
1446 }
1447 
1448 /*
1449  * Resize an extent indirection array to new_size bytes.
1450  */
1451 STATIC void
xfs_iext_realloc_indirect(xfs_ifork_t * ifp,int new_size)1452 xfs_iext_realloc_indirect(
1453 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1454 	int		new_size)	/* new indirection array size */
1455 {
1456 	int		nlists;		/* number of irec's (ex lists) */
1457 	int		size;		/* current indirection array size */
1458 
1459 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1460 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1461 	size = nlists * sizeof(xfs_ext_irec_t);
1462 	ASSERT(ifp->if_real_bytes);
1463 	ASSERT((new_size >= 0) && (new_size != size));
1464 	if (new_size == 0) {
1465 		xfs_iext_destroy(ifp);
1466 	} else {
1467 		ifp->if_u1.if_ext_irec = (xfs_ext_irec_t *)
1468 			kmem_realloc(ifp->if_u1.if_ext_irec,
1469 				new_size, size, KM_NOFS);
1470 	}
1471 }
1472 
1473 /*
1474  * Switch from indirection array to linear (direct) extent allocations.
1475  */
1476 STATIC void
xfs_iext_indirect_to_direct(xfs_ifork_t * ifp)1477 xfs_iext_indirect_to_direct(
1478 	 xfs_ifork_t	*ifp)		/* inode fork pointer */
1479 {
1480 	xfs_bmbt_rec_host_t *ep;	/* extent record pointer */
1481 	xfs_extnum_t	nextents;	/* number of extents in file */
1482 	int		size;		/* size of file extents */
1483 
1484 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1485 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1486 	ASSERT(nextents <= XFS_LINEAR_EXTS);
1487 	size = nextents * sizeof(xfs_bmbt_rec_t);
1488 
1489 	xfs_iext_irec_compact_pages(ifp);
1490 	ASSERT(ifp->if_real_bytes == XFS_IEXT_BUFSZ);
1491 
1492 	ep = ifp->if_u1.if_ext_irec->er_extbuf;
1493 	kmem_free(ifp->if_u1.if_ext_irec);
1494 	ifp->if_flags &= ~XFS_IFEXTIREC;
1495 	ifp->if_u1.if_extents = ep;
1496 	ifp->if_bytes = size;
1497 	if (nextents < XFS_LINEAR_EXTS) {
1498 		xfs_iext_realloc_direct(ifp, size);
1499 	}
1500 }
1501 
1502 /*
1503  * Free incore file extents.
1504  */
1505 void
xfs_iext_destroy(xfs_ifork_t * ifp)1506 xfs_iext_destroy(
1507 	xfs_ifork_t	*ifp)		/* inode fork pointer */
1508 {
1509 	if (ifp->if_flags & XFS_IFEXTIREC) {
1510 		int	erp_idx;
1511 		int	nlists;
1512 
1513 		nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1514 		for (erp_idx = nlists - 1; erp_idx >= 0 ; erp_idx--) {
1515 			xfs_iext_irec_remove(ifp, erp_idx);
1516 		}
1517 		ifp->if_flags &= ~XFS_IFEXTIREC;
1518 	} else if (ifp->if_real_bytes) {
1519 		kmem_free(ifp->if_u1.if_extents);
1520 	} else if (ifp->if_bytes) {
1521 		memset(ifp->if_u2.if_inline_ext, 0, XFS_INLINE_EXTS *
1522 			sizeof(xfs_bmbt_rec_t));
1523 	}
1524 	ifp->if_u1.if_extents = NULL;
1525 	ifp->if_real_bytes = 0;
1526 	ifp->if_bytes = 0;
1527 }
1528 
1529 /*
1530  * Return a pointer to the extent record for file system block bno.
1531  */
1532 xfs_bmbt_rec_host_t *			/* pointer to found extent record */
xfs_iext_bno_to_ext(xfs_ifork_t * ifp,xfs_fileoff_t bno,xfs_extnum_t * idxp)1533 xfs_iext_bno_to_ext(
1534 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1535 	xfs_fileoff_t	bno,		/* block number to search for */
1536 	xfs_extnum_t	*idxp)		/* index of target extent */
1537 {
1538 	xfs_bmbt_rec_host_t *base;	/* pointer to first extent */
1539 	xfs_filblks_t	blockcount = 0;	/* number of blocks in extent */
1540 	xfs_bmbt_rec_host_t *ep = NULL;	/* pointer to target extent */
1541 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
1542 	int		high;		/* upper boundary in search */
1543 	xfs_extnum_t	idx = 0;	/* index of target extent */
1544 	int		low;		/* lower boundary in search */
1545 	xfs_extnum_t	nextents;	/* number of file extents */
1546 	xfs_fileoff_t	startoff = 0;	/* start offset of extent */
1547 
1548 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1549 	if (nextents == 0) {
1550 		*idxp = 0;
1551 		return NULL;
1552 	}
1553 	low = 0;
1554 	if (ifp->if_flags & XFS_IFEXTIREC) {
1555 		/* Find target extent list */
1556 		int	erp_idx = 0;
1557 		erp = xfs_iext_bno_to_irec(ifp, bno, &erp_idx);
1558 		base = erp->er_extbuf;
1559 		high = erp->er_extcount - 1;
1560 	} else {
1561 		base = ifp->if_u1.if_extents;
1562 		high = nextents - 1;
1563 	}
1564 	/* Binary search extent records */
1565 	while (low <= high) {
1566 		idx = (low + high) >> 1;
1567 		ep = base + idx;
1568 		startoff = xfs_bmbt_get_startoff(ep);
1569 		blockcount = xfs_bmbt_get_blockcount(ep);
1570 		if (bno < startoff) {
1571 			high = idx - 1;
1572 		} else if (bno >= startoff + blockcount) {
1573 			low = idx + 1;
1574 		} else {
1575 			/* Convert back to file-based extent index */
1576 			if (ifp->if_flags & XFS_IFEXTIREC) {
1577 				idx += erp->er_extoff;
1578 			}
1579 			*idxp = idx;
1580 			return ep;
1581 		}
1582 	}
1583 	/* Convert back to file-based extent index */
1584 	if (ifp->if_flags & XFS_IFEXTIREC) {
1585 		idx += erp->er_extoff;
1586 	}
1587 	if (bno >= startoff + blockcount) {
1588 		if (++idx == nextents) {
1589 			ep = NULL;
1590 		} else {
1591 			ep = xfs_iext_get_ext(ifp, idx);
1592 		}
1593 	}
1594 	*idxp = idx;
1595 	return ep;
1596 }
1597 
1598 /*
1599  * Return a pointer to the indirection array entry containing the
1600  * extent record for filesystem block bno. Store the index of the
1601  * target irec in *erp_idxp.
1602  */
1603 xfs_ext_irec_t *			/* pointer to found extent record */
xfs_iext_bno_to_irec(xfs_ifork_t * ifp,xfs_fileoff_t bno,int * erp_idxp)1604 xfs_iext_bno_to_irec(
1605 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1606 	xfs_fileoff_t	bno,		/* block number to search for */
1607 	int		*erp_idxp)	/* irec index of target ext list */
1608 {
1609 	xfs_ext_irec_t	*erp = NULL;	/* indirection array pointer */
1610 	xfs_ext_irec_t	*erp_next;	/* next indirection array entry */
1611 	int		erp_idx;	/* indirection array index */
1612 	int		nlists;		/* number of extent irec's (lists) */
1613 	int		high;		/* binary search upper limit */
1614 	int		low;		/* binary search lower limit */
1615 
1616 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1617 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1618 	erp_idx = 0;
1619 	low = 0;
1620 	high = nlists - 1;
1621 	while (low <= high) {
1622 		erp_idx = (low + high) >> 1;
1623 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
1624 		erp_next = erp_idx < nlists - 1 ? erp + 1 : NULL;
1625 		if (bno < xfs_bmbt_get_startoff(erp->er_extbuf)) {
1626 			high = erp_idx - 1;
1627 		} else if (erp_next && bno >=
1628 			   xfs_bmbt_get_startoff(erp_next->er_extbuf)) {
1629 			low = erp_idx + 1;
1630 		} else {
1631 			break;
1632 		}
1633 	}
1634 	*erp_idxp = erp_idx;
1635 	return erp;
1636 }
1637 
1638 /*
1639  * Return a pointer to the indirection array entry containing the
1640  * extent record at file extent index *idxp. Store the index of the
1641  * target irec in *erp_idxp and store the page index of the target
1642  * extent record in *idxp.
1643  */
1644 xfs_ext_irec_t *
xfs_iext_idx_to_irec(xfs_ifork_t * ifp,xfs_extnum_t * idxp,int * erp_idxp,int realloc)1645 xfs_iext_idx_to_irec(
1646 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1647 	xfs_extnum_t	*idxp,		/* extent index (file -> page) */
1648 	int		*erp_idxp,	/* pointer to target irec */
1649 	int		realloc)	/* new bytes were just added */
1650 {
1651 	xfs_ext_irec_t	*prev;		/* pointer to previous irec */
1652 	xfs_ext_irec_t	*erp = NULL;	/* pointer to current irec */
1653 	int		erp_idx;	/* indirection array index */
1654 	int		nlists;		/* number of irec's (ex lists) */
1655 	int		high;		/* binary search upper limit */
1656 	int		low;		/* binary search lower limit */
1657 	xfs_extnum_t	page_idx = *idxp; /* extent index in target list */
1658 
1659 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1660 	ASSERT(page_idx >= 0);
1661 	ASSERT(page_idx <= ifp->if_bytes / sizeof(xfs_bmbt_rec_t));
1662 	ASSERT(page_idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t) || realloc);
1663 
1664 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1665 	erp_idx = 0;
1666 	low = 0;
1667 	high = nlists - 1;
1668 
1669 	/* Binary search extent irec's */
1670 	while (low <= high) {
1671 		erp_idx = (low + high) >> 1;
1672 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
1673 		prev = erp_idx > 0 ? erp - 1 : NULL;
1674 		if (page_idx < erp->er_extoff || (page_idx == erp->er_extoff &&
1675 		     realloc && prev && prev->er_extcount < XFS_LINEAR_EXTS)) {
1676 			high = erp_idx - 1;
1677 		} else if (page_idx > erp->er_extoff + erp->er_extcount ||
1678 			   (page_idx == erp->er_extoff + erp->er_extcount &&
1679 			    !realloc)) {
1680 			low = erp_idx + 1;
1681 		} else if (page_idx == erp->er_extoff + erp->er_extcount &&
1682 			   erp->er_extcount == XFS_LINEAR_EXTS) {
1683 			ASSERT(realloc);
1684 			page_idx = 0;
1685 			erp_idx++;
1686 			erp = erp_idx < nlists ? erp + 1 : NULL;
1687 			break;
1688 		} else {
1689 			page_idx -= erp->er_extoff;
1690 			break;
1691 		}
1692 	}
1693 	*idxp = page_idx;
1694 	*erp_idxp = erp_idx;
1695 	return erp;
1696 }
1697 
1698 /*
1699  * Allocate and initialize an indirection array once the space needed
1700  * for incore extents increases above XFS_IEXT_BUFSZ.
1701  */
1702 void
xfs_iext_irec_init(xfs_ifork_t * ifp)1703 xfs_iext_irec_init(
1704 	xfs_ifork_t	*ifp)		/* inode fork pointer */
1705 {
1706 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
1707 	xfs_extnum_t	nextents;	/* number of extents in file */
1708 
1709 	ASSERT(!(ifp->if_flags & XFS_IFEXTIREC));
1710 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1711 	ASSERT(nextents <= XFS_LINEAR_EXTS);
1712 
1713 	erp = kmem_alloc(sizeof(xfs_ext_irec_t), KM_NOFS);
1714 
1715 	if (nextents == 0) {
1716 		ifp->if_u1.if_extents = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
1717 	} else if (!ifp->if_real_bytes) {
1718 		xfs_iext_inline_to_direct(ifp, XFS_IEXT_BUFSZ);
1719 	} else if (ifp->if_real_bytes < XFS_IEXT_BUFSZ) {
1720 		xfs_iext_realloc_direct(ifp, XFS_IEXT_BUFSZ);
1721 	}
1722 	erp->er_extbuf = ifp->if_u1.if_extents;
1723 	erp->er_extcount = nextents;
1724 	erp->er_extoff = 0;
1725 
1726 	ifp->if_flags |= XFS_IFEXTIREC;
1727 	ifp->if_real_bytes = XFS_IEXT_BUFSZ;
1728 	ifp->if_bytes = nextents * sizeof(xfs_bmbt_rec_t);
1729 	ifp->if_u1.if_ext_irec = erp;
1730 
1731 	return;
1732 }
1733 
1734 /*
1735  * Allocate and initialize a new entry in the indirection array.
1736  */
1737 xfs_ext_irec_t *
xfs_iext_irec_new(xfs_ifork_t * ifp,int erp_idx)1738 xfs_iext_irec_new(
1739 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1740 	int		erp_idx)	/* index for new irec */
1741 {
1742 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
1743 	int		i;		/* loop counter */
1744 	int		nlists;		/* number of irec's (ex lists) */
1745 
1746 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1747 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1748 
1749 	/* Resize indirection array */
1750 	xfs_iext_realloc_indirect(ifp, ++nlists *
1751 				  sizeof(xfs_ext_irec_t));
1752 	/*
1753 	 * Move records down in the array so the
1754 	 * new page can use erp_idx.
1755 	 */
1756 	erp = ifp->if_u1.if_ext_irec;
1757 	for (i = nlists - 1; i > erp_idx; i--) {
1758 		memmove(&erp[i], &erp[i-1], sizeof(xfs_ext_irec_t));
1759 	}
1760 	ASSERT(i == erp_idx);
1761 
1762 	/* Initialize new extent record */
1763 	erp = ifp->if_u1.if_ext_irec;
1764 	erp[erp_idx].er_extbuf = kmem_alloc(XFS_IEXT_BUFSZ, KM_NOFS);
1765 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
1766 	memset(erp[erp_idx].er_extbuf, 0, XFS_IEXT_BUFSZ);
1767 	erp[erp_idx].er_extcount = 0;
1768 	erp[erp_idx].er_extoff = erp_idx > 0 ?
1769 		erp[erp_idx-1].er_extoff + erp[erp_idx-1].er_extcount : 0;
1770 	return (&erp[erp_idx]);
1771 }
1772 
1773 /*
1774  * Remove a record from the indirection array.
1775  */
1776 void
xfs_iext_irec_remove(xfs_ifork_t * ifp,int erp_idx)1777 xfs_iext_irec_remove(
1778 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1779 	int		erp_idx)	/* irec index to remove */
1780 {
1781 	xfs_ext_irec_t	*erp;		/* indirection array pointer */
1782 	int		i;		/* loop counter */
1783 	int		nlists;		/* number of irec's (ex lists) */
1784 
1785 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1786 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1787 	erp = &ifp->if_u1.if_ext_irec[erp_idx];
1788 	if (erp->er_extbuf) {
1789 		xfs_iext_irec_update_extoffs(ifp, erp_idx + 1,
1790 			-erp->er_extcount);
1791 		kmem_free(erp->er_extbuf);
1792 	}
1793 	/* Compact extent records */
1794 	erp = ifp->if_u1.if_ext_irec;
1795 	for (i = erp_idx; i < nlists - 1; i++) {
1796 		memmove(&erp[i], &erp[i+1], sizeof(xfs_ext_irec_t));
1797 	}
1798 	/*
1799 	 * Manually free the last extent record from the indirection
1800 	 * array.  A call to xfs_iext_realloc_indirect() with a size
1801 	 * of zero would result in a call to xfs_iext_destroy() which
1802 	 * would in turn call this function again, creating a nasty
1803 	 * infinite loop.
1804 	 */
1805 	if (--nlists) {
1806 		xfs_iext_realloc_indirect(ifp,
1807 			nlists * sizeof(xfs_ext_irec_t));
1808 	} else {
1809 		kmem_free(ifp->if_u1.if_ext_irec);
1810 	}
1811 	ifp->if_real_bytes = nlists * XFS_IEXT_BUFSZ;
1812 }
1813 
1814 /*
1815  * This is called to clean up large amounts of unused memory allocated
1816  * by the indirection array.  Before compacting anything though, verify
1817  * that the indirection array is still needed and switch back to the
1818  * linear extent list (or even the inline buffer) if possible.  The
1819  * compaction policy is as follows:
1820  *
1821  *    Full Compaction: Extents fit into a single page (or inline buffer)
1822  * Partial Compaction: Extents occupy less than 50% of allocated space
1823  *      No Compaction: Extents occupy at least 50% of allocated space
1824  */
1825 void
xfs_iext_irec_compact(xfs_ifork_t * ifp)1826 xfs_iext_irec_compact(
1827 	xfs_ifork_t	*ifp)		/* inode fork pointer */
1828 {
1829 	xfs_extnum_t	nextents;	/* number of extents in file */
1830 	int		nlists;		/* number of irec's (ex lists) */
1831 
1832 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1833 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1834 	nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1835 
1836 	if (nextents == 0) {
1837 		xfs_iext_destroy(ifp);
1838 	} else if (nextents <= XFS_INLINE_EXTS) {
1839 		xfs_iext_indirect_to_direct(ifp);
1840 		xfs_iext_direct_to_inline(ifp, nextents);
1841 	} else if (nextents <= XFS_LINEAR_EXTS) {
1842 		xfs_iext_indirect_to_direct(ifp);
1843 	} else if (nextents < (nlists * XFS_LINEAR_EXTS) >> 1) {
1844 		xfs_iext_irec_compact_pages(ifp);
1845 	}
1846 }
1847 
1848 /*
1849  * Combine extents from neighboring extent pages.
1850  */
1851 void
xfs_iext_irec_compact_pages(xfs_ifork_t * ifp)1852 xfs_iext_irec_compact_pages(
1853 	xfs_ifork_t	*ifp)		/* inode fork pointer */
1854 {
1855 	xfs_ext_irec_t	*erp, *erp_next;/* pointers to irec entries */
1856 	int		erp_idx = 0;	/* indirection array index */
1857 	int		nlists;		/* number of irec's (ex lists) */
1858 
1859 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1860 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1861 	while (erp_idx < nlists - 1) {
1862 		erp = &ifp->if_u1.if_ext_irec[erp_idx];
1863 		erp_next = erp + 1;
1864 		if (erp_next->er_extcount <=
1865 		    (XFS_LINEAR_EXTS - erp->er_extcount)) {
1866 			memcpy(&erp->er_extbuf[erp->er_extcount],
1867 				erp_next->er_extbuf, erp_next->er_extcount *
1868 				sizeof(xfs_bmbt_rec_t));
1869 			erp->er_extcount += erp_next->er_extcount;
1870 			/*
1871 			 * Free page before removing extent record
1872 			 * so er_extoffs don't get modified in
1873 			 * xfs_iext_irec_remove.
1874 			 */
1875 			kmem_free(erp_next->er_extbuf);
1876 			erp_next->er_extbuf = NULL;
1877 			xfs_iext_irec_remove(ifp, erp_idx + 1);
1878 			nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1879 		} else {
1880 			erp_idx++;
1881 		}
1882 	}
1883 }
1884 
1885 /*
1886  * This is called to update the er_extoff field in the indirection
1887  * array when extents have been added or removed from one of the
1888  * extent lists. erp_idx contains the irec index to begin updating
1889  * at and ext_diff contains the number of extents that were added
1890  * or removed.
1891  */
1892 void
xfs_iext_irec_update_extoffs(xfs_ifork_t * ifp,int erp_idx,int ext_diff)1893 xfs_iext_irec_update_extoffs(
1894 	xfs_ifork_t	*ifp,		/* inode fork pointer */
1895 	int		erp_idx,	/* irec index to update */
1896 	int		ext_diff)	/* number of new extents */
1897 {
1898 	int		i;		/* loop counter */
1899 	int		nlists;		/* number of irec's (ex lists */
1900 
1901 	ASSERT(ifp->if_flags & XFS_IFEXTIREC);
1902 	nlists = ifp->if_real_bytes / XFS_IEXT_BUFSZ;
1903 	for (i = erp_idx; i < nlists; i++) {
1904 		ifp->if_u1.if_ext_irec[i].er_extoff += ext_diff;
1905 	}
1906 }
1907