xref: /openbmc/linux/fs/xfs/libxfs/xfs_inode_fork.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
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_mount.h"
26 #include "xfs_inode.h"
27 #include "xfs_trans.h"
28 #include "xfs_inode_item.h"
29 #include "xfs_btree.h"
30 #include "xfs_bmap_btree.h"
31 #include "xfs_bmap.h"
32 #include "xfs_error.h"
33 #include "xfs_trace.h"
34 #include "xfs_attr_sf.h"
35 #include "xfs_da_format.h"
36 #include "xfs_da_btree.h"
37 #include "xfs_dir2_priv.h"
38 #include "xfs_attr_leaf.h"
39 #include "xfs_shared.h"
40 
41 kmem_zone_t *xfs_ifork_zone;
42 
43 STATIC int xfs_iformat_local(xfs_inode_t *, xfs_dinode_t *, int, int);
44 STATIC int xfs_iformat_extents(xfs_inode_t *, xfs_dinode_t *, int);
45 STATIC int xfs_iformat_btree(xfs_inode_t *, xfs_dinode_t *, int);
46 
47 /*
48  * Copy inode type and data and attr format specific information from the
49  * on-disk inode to the in-core inode and fork structures.  For fifos, devices,
50  * and sockets this means set i_rdev to the proper value.  For files,
51  * directories, and symlinks this means to bring in the in-line data or extent
52  * pointers as well as the attribute fork.  For a fork in B-tree format, only
53  * the root is immediately brought in-core.  The rest will be read in later when
54  * first referenced (see xfs_iread_extents()).
55  */
56 int
57 xfs_iformat_fork(
58 	struct xfs_inode	*ip,
59 	struct xfs_dinode	*dip)
60 {
61 	struct inode		*inode = VFS_I(ip);
62 	struct xfs_attr_shortform *atp;
63 	int			size;
64 	int			error = 0;
65 	xfs_fsize_t             di_size;
66 
67 	switch (inode->i_mode & S_IFMT) {
68 	case S_IFIFO:
69 	case S_IFCHR:
70 	case S_IFBLK:
71 	case S_IFSOCK:
72 		ip->i_d.di_size = 0;
73 		inode->i_rdev = xfs_to_linux_dev_t(xfs_dinode_get_rdev(dip));
74 		break;
75 
76 	case S_IFREG:
77 	case S_IFLNK:
78 	case S_IFDIR:
79 		switch (dip->di_format) {
80 		case XFS_DINODE_FMT_LOCAL:
81 			di_size = be64_to_cpu(dip->di_size);
82 			size = (int)di_size;
83 			error = xfs_iformat_local(ip, dip, XFS_DATA_FORK, size);
84 			break;
85 		case XFS_DINODE_FMT_EXTENTS:
86 			error = xfs_iformat_extents(ip, dip, XFS_DATA_FORK);
87 			break;
88 		case XFS_DINODE_FMT_BTREE:
89 			error = xfs_iformat_btree(ip, dip, XFS_DATA_FORK);
90 			break;
91 		default:
92 			return -EFSCORRUPTED;
93 		}
94 		break;
95 
96 	default:
97 		return -EFSCORRUPTED;
98 	}
99 	if (error)
100 		return error;
101 
102 	if (xfs_is_reflink_inode(ip)) {
103 		ASSERT(ip->i_cowfp == NULL);
104 		xfs_ifork_init_cow(ip);
105 	}
106 
107 	if (!XFS_DFORK_Q(dip))
108 		return 0;
109 
110 	ASSERT(ip->i_afp == NULL);
111 	ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP | KM_NOFS);
112 
113 	switch (dip->di_aformat) {
114 	case XFS_DINODE_FMT_LOCAL:
115 		atp = (xfs_attr_shortform_t *)XFS_DFORK_APTR(dip);
116 		size = be16_to_cpu(atp->hdr.totsize);
117 
118 		error = xfs_iformat_local(ip, dip, XFS_ATTR_FORK, size);
119 		break;
120 	case XFS_DINODE_FMT_EXTENTS:
121 		error = xfs_iformat_extents(ip, dip, XFS_ATTR_FORK);
122 		break;
123 	case XFS_DINODE_FMT_BTREE:
124 		error = xfs_iformat_btree(ip, dip, XFS_ATTR_FORK);
125 		break;
126 	default:
127 		error = -EFSCORRUPTED;
128 		break;
129 	}
130 	if (error) {
131 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
132 		ip->i_afp = NULL;
133 		if (ip->i_cowfp)
134 			kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
135 		ip->i_cowfp = NULL;
136 		xfs_idestroy_fork(ip, XFS_DATA_FORK);
137 	}
138 	return error;
139 }
140 
141 void
142 xfs_init_local_fork(
143 	struct xfs_inode	*ip,
144 	int			whichfork,
145 	const void		*data,
146 	int			size)
147 {
148 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
149 	int			mem_size = size, real_size = 0;
150 	bool			zero_terminate;
151 
152 	/*
153 	 * If we are using the local fork to store a symlink body we need to
154 	 * zero-terminate it so that we can pass it back to the VFS directly.
155 	 * Overallocate the in-memory fork by one for that and add a zero
156 	 * to terminate it below.
157 	 */
158 	zero_terminate = S_ISLNK(VFS_I(ip)->i_mode);
159 	if (zero_terminate)
160 		mem_size++;
161 
162 	if (size) {
163 		real_size = roundup(mem_size, 4);
164 		ifp->if_u1.if_data = kmem_alloc(real_size, KM_SLEEP | KM_NOFS);
165 		memcpy(ifp->if_u1.if_data, data, size);
166 		if (zero_terminate)
167 			ifp->if_u1.if_data[size] = '\0';
168 	} else {
169 		ifp->if_u1.if_data = NULL;
170 	}
171 
172 	ifp->if_bytes = size;
173 	ifp->if_real_bytes = real_size;
174 	ifp->if_flags &= ~(XFS_IFEXTENTS | XFS_IFBROOT);
175 	ifp->if_flags |= XFS_IFINLINE;
176 }
177 
178 /*
179  * The file is in-lined in the on-disk inode.
180  */
181 STATIC int
182 xfs_iformat_local(
183 	xfs_inode_t	*ip,
184 	xfs_dinode_t	*dip,
185 	int		whichfork,
186 	int		size)
187 {
188 	/*
189 	 * If the size is unreasonable, then something
190 	 * is wrong and we just bail out rather than crash in
191 	 * kmem_alloc() or memcpy() below.
192 	 */
193 	if (unlikely(size > XFS_DFORK_SIZE(dip, ip->i_mount, whichfork))) {
194 		xfs_warn(ip->i_mount,
195 	"corrupt inode %Lu (bad size %d for local fork, size = %d).",
196 			(unsigned long long) ip->i_ino, size,
197 			XFS_DFORK_SIZE(dip, ip->i_mount, whichfork));
198 		xfs_inode_verifier_error(ip, -EFSCORRUPTED,
199 				"xfs_iformat_local", dip, sizeof(*dip),
200 				__this_address);
201 		return -EFSCORRUPTED;
202 	}
203 
204 	xfs_init_local_fork(ip, whichfork, XFS_DFORK_PTR(dip, whichfork), size);
205 	return 0;
206 }
207 
208 /*
209  * The file consists of a set of extents all of which fit into the on-disk
210  * inode.
211  */
212 STATIC int
213 xfs_iformat_extents(
214 	struct xfs_inode	*ip,
215 	struct xfs_dinode	*dip,
216 	int			whichfork)
217 {
218 	struct xfs_mount	*mp = ip->i_mount;
219 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
220 	int			state = xfs_bmap_fork_to_state(whichfork);
221 	int			nex = XFS_DFORK_NEXTENTS(dip, whichfork);
222 	int			size = nex * sizeof(xfs_bmbt_rec_t);
223 	struct xfs_iext_cursor	icur;
224 	struct xfs_bmbt_rec	*dp;
225 	struct xfs_bmbt_irec	new;
226 	int			i;
227 
228 	/*
229 	 * If the number of extents is unreasonable, then something is wrong and
230 	 * we just bail out rather than crash in kmem_alloc() or memcpy() below.
231 	 */
232 	if (unlikely(size < 0 || size > XFS_DFORK_SIZE(dip, mp, whichfork))) {
233 		xfs_warn(ip->i_mount, "corrupt inode %Lu ((a)extents = %d).",
234 			(unsigned long long) ip->i_ino, nex);
235 		xfs_inode_verifier_error(ip, -EFSCORRUPTED,
236 				"xfs_iformat_extents(1)", dip, sizeof(*dip),
237 				__this_address);
238 		return -EFSCORRUPTED;
239 	}
240 
241 	ifp->if_real_bytes = 0;
242 	ifp->if_bytes = 0;
243 	ifp->if_u1.if_root = NULL;
244 	ifp->if_height = 0;
245 	if (size) {
246 		dp = (xfs_bmbt_rec_t *) XFS_DFORK_PTR(dip, whichfork);
247 
248 		xfs_iext_first(ifp, &icur);
249 		for (i = 0; i < nex; i++, dp++) {
250 			xfs_failaddr_t	fa;
251 
252 			xfs_bmbt_disk_get_all(dp, &new);
253 			fa = xfs_bmap_validate_extent(ip, whichfork, &new);
254 			if (fa) {
255 				xfs_inode_verifier_error(ip, -EFSCORRUPTED,
256 						"xfs_iformat_extents(2)",
257 						dp, sizeof(*dp), fa);
258 				return -EFSCORRUPTED;
259 			}
260 
261 			xfs_iext_insert(ip, &icur, &new, state);
262 			trace_xfs_read_extent(ip, &icur, state, _THIS_IP_);
263 			xfs_iext_next(ifp, &icur);
264 		}
265 	}
266 	ifp->if_flags |= XFS_IFEXTENTS;
267 	return 0;
268 }
269 
270 /*
271  * The file has too many extents to fit into
272  * the inode, so they are in B-tree format.
273  * Allocate a buffer for the root of the B-tree
274  * and copy the root into it.  The i_extents
275  * field will remain NULL until all of the
276  * extents are read in (when they are needed).
277  */
278 STATIC int
279 xfs_iformat_btree(
280 	xfs_inode_t		*ip,
281 	xfs_dinode_t		*dip,
282 	int			whichfork)
283 {
284 	struct xfs_mount	*mp = ip->i_mount;
285 	xfs_bmdr_block_t	*dfp;
286 	xfs_ifork_t		*ifp;
287 	/* REFERENCED */
288 	int			nrecs;
289 	int			size;
290 	int			level;
291 
292 	ifp = XFS_IFORK_PTR(ip, whichfork);
293 	dfp = (xfs_bmdr_block_t *)XFS_DFORK_PTR(dip, whichfork);
294 	size = XFS_BMAP_BROOT_SPACE(mp, dfp);
295 	nrecs = be16_to_cpu(dfp->bb_numrecs);
296 	level = be16_to_cpu(dfp->bb_level);
297 
298 	/*
299 	 * blow out if -- fork has less extents than can fit in
300 	 * fork (fork shouldn't be a btree format), root btree
301 	 * block has more records than can fit into the fork,
302 	 * or the number of extents is greater than the number of
303 	 * blocks.
304 	 */
305 	if (unlikely(XFS_IFORK_NEXTENTS(ip, whichfork) <=
306 					XFS_IFORK_MAXEXT(ip, whichfork) ||
307 		     nrecs == 0 ||
308 		     XFS_BMDR_SPACE_CALC(nrecs) >
309 					XFS_DFORK_SIZE(dip, mp, whichfork) ||
310 		     XFS_IFORK_NEXTENTS(ip, whichfork) > ip->i_d.di_nblocks) ||
311 		     level == 0 || level > XFS_BTREE_MAXLEVELS) {
312 		xfs_warn(mp, "corrupt inode %Lu (btree).",
313 					(unsigned long long) ip->i_ino);
314 		xfs_inode_verifier_error(ip, -EFSCORRUPTED,
315 				"xfs_iformat_btree", dfp, size,
316 				__this_address);
317 		return -EFSCORRUPTED;
318 	}
319 
320 	ifp->if_broot_bytes = size;
321 	ifp->if_broot = kmem_alloc(size, KM_SLEEP | KM_NOFS);
322 	ASSERT(ifp->if_broot != NULL);
323 	/*
324 	 * Copy and convert from the on-disk structure
325 	 * to the in-memory structure.
326 	 */
327 	xfs_bmdr_to_bmbt(ip, dfp, XFS_DFORK_SIZE(dip, ip->i_mount, whichfork),
328 			 ifp->if_broot, size);
329 	ifp->if_flags &= ~XFS_IFEXTENTS;
330 	ifp->if_flags |= XFS_IFBROOT;
331 
332 	ifp->if_real_bytes = 0;
333 	ifp->if_bytes = 0;
334 	ifp->if_u1.if_root = NULL;
335 	ifp->if_height = 0;
336 	return 0;
337 }
338 
339 /*
340  * Reallocate the space for if_broot based on the number of records
341  * being added or deleted as indicated in rec_diff.  Move the records
342  * and pointers in if_broot to fit the new size.  When shrinking this
343  * will eliminate holes between the records and pointers created by
344  * the caller.  When growing this will create holes to be filled in
345  * by the caller.
346  *
347  * The caller must not request to add more records than would fit in
348  * the on-disk inode root.  If the if_broot is currently NULL, then
349  * if we are adding records, one will be allocated.  The caller must also
350  * not request that the number of records go below zero, although
351  * it can go to zero.
352  *
353  * ip -- the inode whose if_broot area is changing
354  * ext_diff -- the change in the number of records, positive or negative,
355  *	 requested for the if_broot array.
356  */
357 void
358 xfs_iroot_realloc(
359 	xfs_inode_t		*ip,
360 	int			rec_diff,
361 	int			whichfork)
362 {
363 	struct xfs_mount	*mp = ip->i_mount;
364 	int			cur_max;
365 	xfs_ifork_t		*ifp;
366 	struct xfs_btree_block	*new_broot;
367 	int			new_max;
368 	size_t			new_size;
369 	char			*np;
370 	char			*op;
371 
372 	/*
373 	 * Handle the degenerate case quietly.
374 	 */
375 	if (rec_diff == 0) {
376 		return;
377 	}
378 
379 	ifp = XFS_IFORK_PTR(ip, whichfork);
380 	if (rec_diff > 0) {
381 		/*
382 		 * If there wasn't any memory allocated before, just
383 		 * allocate it now and get out.
384 		 */
385 		if (ifp->if_broot_bytes == 0) {
386 			new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, rec_diff);
387 			ifp->if_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
388 			ifp->if_broot_bytes = (int)new_size;
389 			return;
390 		}
391 
392 		/*
393 		 * If there is already an existing if_broot, then we need
394 		 * to realloc() it and shift the pointers to their new
395 		 * location.  The records don't change location because
396 		 * they are kept butted up against the btree block header.
397 		 */
398 		cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
399 		new_max = cur_max + rec_diff;
400 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
401 		ifp->if_broot = kmem_realloc(ifp->if_broot, new_size,
402 				KM_SLEEP | KM_NOFS);
403 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
404 						     ifp->if_broot_bytes);
405 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
406 						     (int)new_size);
407 		ifp->if_broot_bytes = (int)new_size;
408 		ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
409 			XFS_IFORK_SIZE(ip, whichfork));
410 		memmove(np, op, cur_max * (uint)sizeof(xfs_fsblock_t));
411 		return;
412 	}
413 
414 	/*
415 	 * rec_diff is less than 0.  In this case, we are shrinking the
416 	 * if_broot buffer.  It must already exist.  If we go to zero
417 	 * records, just get rid of the root and clear the status bit.
418 	 */
419 	ASSERT((ifp->if_broot != NULL) && (ifp->if_broot_bytes > 0));
420 	cur_max = xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0);
421 	new_max = cur_max + rec_diff;
422 	ASSERT(new_max >= 0);
423 	if (new_max > 0)
424 		new_size = XFS_BMAP_BROOT_SPACE_CALC(mp, new_max);
425 	else
426 		new_size = 0;
427 	if (new_size > 0) {
428 		new_broot = kmem_alloc(new_size, KM_SLEEP | KM_NOFS);
429 		/*
430 		 * First copy over the btree block header.
431 		 */
432 		memcpy(new_broot, ifp->if_broot,
433 			XFS_BMBT_BLOCK_LEN(ip->i_mount));
434 	} else {
435 		new_broot = NULL;
436 		ifp->if_flags &= ~XFS_IFBROOT;
437 	}
438 
439 	/*
440 	 * Only copy the records and pointers if there are any.
441 	 */
442 	if (new_max > 0) {
443 		/*
444 		 * First copy the records.
445 		 */
446 		op = (char *)XFS_BMBT_REC_ADDR(mp, ifp->if_broot, 1);
447 		np = (char *)XFS_BMBT_REC_ADDR(mp, new_broot, 1);
448 		memcpy(np, op, new_max * (uint)sizeof(xfs_bmbt_rec_t));
449 
450 		/*
451 		 * Then copy the pointers.
452 		 */
453 		op = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, ifp->if_broot, 1,
454 						     ifp->if_broot_bytes);
455 		np = (char *)XFS_BMAP_BROOT_PTR_ADDR(mp, new_broot, 1,
456 						     (int)new_size);
457 		memcpy(np, op, new_max * (uint)sizeof(xfs_fsblock_t));
458 	}
459 	kmem_free(ifp->if_broot);
460 	ifp->if_broot = new_broot;
461 	ifp->if_broot_bytes = (int)new_size;
462 	if (ifp->if_broot)
463 		ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
464 			XFS_IFORK_SIZE(ip, whichfork));
465 	return;
466 }
467 
468 
469 /*
470  * This is called when the amount of space needed for if_data
471  * is increased or decreased.  The change in size is indicated by
472  * the number of bytes that need to be added or deleted in the
473  * byte_diff parameter.
474  *
475  * If the amount of space needed has decreased below the size of the
476  * inline buffer, then switch to using the inline buffer.  Otherwise,
477  * use kmem_realloc() or kmem_alloc() to adjust the size of the buffer
478  * to what is needed.
479  *
480  * ip -- the inode whose if_data area is changing
481  * byte_diff -- the change in the number of bytes, positive or negative,
482  *	 requested for the if_data array.
483  */
484 void
485 xfs_idata_realloc(
486 	xfs_inode_t	*ip,
487 	int		byte_diff,
488 	int		whichfork)
489 {
490 	xfs_ifork_t	*ifp;
491 	int		new_size;
492 	int		real_size;
493 
494 	if (byte_diff == 0) {
495 		return;
496 	}
497 
498 	ifp = XFS_IFORK_PTR(ip, whichfork);
499 	new_size = (int)ifp->if_bytes + byte_diff;
500 	ASSERT(new_size >= 0);
501 
502 	if (new_size == 0) {
503 		kmem_free(ifp->if_u1.if_data);
504 		ifp->if_u1.if_data = NULL;
505 		real_size = 0;
506 	} else {
507 		/*
508 		 * Stuck with malloc/realloc.
509 		 * For inline data, the underlying buffer must be
510 		 * a multiple of 4 bytes in size so that it can be
511 		 * logged and stay on word boundaries.  We enforce
512 		 * that here.
513 		 */
514 		real_size = roundup(new_size, 4);
515 		if (ifp->if_u1.if_data == NULL) {
516 			ASSERT(ifp->if_real_bytes == 0);
517 			ifp->if_u1.if_data = kmem_alloc(real_size,
518 							KM_SLEEP | KM_NOFS);
519 		} else {
520 			/*
521 			 * Only do the realloc if the underlying size
522 			 * is really changing.
523 			 */
524 			if (ifp->if_real_bytes != real_size) {
525 				ifp->if_u1.if_data =
526 					kmem_realloc(ifp->if_u1.if_data,
527 							real_size,
528 							KM_SLEEP | KM_NOFS);
529 			}
530 		}
531 	}
532 	ifp->if_real_bytes = real_size;
533 	ifp->if_bytes = new_size;
534 	ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
535 }
536 
537 void
538 xfs_idestroy_fork(
539 	xfs_inode_t	*ip,
540 	int		whichfork)
541 {
542 	xfs_ifork_t	*ifp;
543 
544 	ifp = XFS_IFORK_PTR(ip, whichfork);
545 	if (ifp->if_broot != NULL) {
546 		kmem_free(ifp->if_broot);
547 		ifp->if_broot = NULL;
548 	}
549 
550 	/*
551 	 * If the format is local, then we can't have an extents
552 	 * array so just look for an inline data array.  If we're
553 	 * not local then we may or may not have an extents list,
554 	 * so check and free it up if we do.
555 	 */
556 	if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
557 		if (ifp->if_u1.if_data != NULL) {
558 			ASSERT(ifp->if_real_bytes != 0);
559 			kmem_free(ifp->if_u1.if_data);
560 			ifp->if_u1.if_data = NULL;
561 			ifp->if_real_bytes = 0;
562 		}
563 	} else if ((ifp->if_flags & XFS_IFEXTENTS) && ifp->if_height) {
564 		xfs_iext_destroy(ifp);
565 	}
566 
567 	ASSERT(ifp->if_real_bytes == 0);
568 
569 	if (whichfork == XFS_ATTR_FORK) {
570 		kmem_zone_free(xfs_ifork_zone, ip->i_afp);
571 		ip->i_afp = NULL;
572 	} else if (whichfork == XFS_COW_FORK) {
573 		kmem_zone_free(xfs_ifork_zone, ip->i_cowfp);
574 		ip->i_cowfp = NULL;
575 	}
576 }
577 
578 /*
579  * Convert in-core extents to on-disk form
580  *
581  * In the case of the data fork, the in-core and on-disk fork sizes can be
582  * different due to delayed allocation extents. We only copy on-disk extents
583  * here, so callers must always use the physical fork size to determine the
584  * size of the buffer passed to this routine.  We will return the size actually
585  * used.
586  */
587 int
588 xfs_iextents_copy(
589 	struct xfs_inode	*ip,
590 	struct xfs_bmbt_rec	*dp,
591 	int			whichfork)
592 {
593 	int			state = xfs_bmap_fork_to_state(whichfork);
594 	struct xfs_ifork	*ifp = XFS_IFORK_PTR(ip, whichfork);
595 	struct xfs_iext_cursor	icur;
596 	struct xfs_bmbt_irec	rec;
597 	int			copied = 0;
598 
599 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL | XFS_ILOCK_SHARED));
600 	ASSERT(ifp->if_bytes > 0);
601 
602 	for_each_xfs_iext(ifp, &icur, &rec) {
603 		if (isnullstartblock(rec.br_startblock))
604 			continue;
605 		ASSERT(xfs_bmap_validate_extent(ip, whichfork, &rec) == NULL);
606 		xfs_bmbt_disk_set_all(dp, &rec);
607 		trace_xfs_write_extent(ip, &icur, state, _RET_IP_);
608 		copied += sizeof(struct xfs_bmbt_rec);
609 		dp++;
610 	}
611 
612 	ASSERT(copied > 0);
613 	ASSERT(copied <= ifp->if_bytes);
614 	return copied;
615 }
616 
617 /*
618  * Each of the following cases stores data into the same region
619  * of the on-disk inode, so only one of them can be valid at
620  * any given time. While it is possible to have conflicting formats
621  * and log flags, e.g. having XFS_ILOG_?DATA set when the fork is
622  * in EXTENTS format, this can only happen when the fork has
623  * changed formats after being modified but before being flushed.
624  * In these cases, the format always takes precedence, because the
625  * format indicates the current state of the fork.
626  */
627 void
628 xfs_iflush_fork(
629 	xfs_inode_t		*ip,
630 	xfs_dinode_t		*dip,
631 	xfs_inode_log_item_t	*iip,
632 	int			whichfork)
633 {
634 	char			*cp;
635 	xfs_ifork_t		*ifp;
636 	xfs_mount_t		*mp;
637 	static const short	brootflag[2] =
638 		{ XFS_ILOG_DBROOT, XFS_ILOG_ABROOT };
639 	static const short	dataflag[2] =
640 		{ XFS_ILOG_DDATA, XFS_ILOG_ADATA };
641 	static const short	extflag[2] =
642 		{ XFS_ILOG_DEXT, XFS_ILOG_AEXT };
643 
644 	if (!iip)
645 		return;
646 	ifp = XFS_IFORK_PTR(ip, whichfork);
647 	/*
648 	 * This can happen if we gave up in iformat in an error path,
649 	 * for the attribute fork.
650 	 */
651 	if (!ifp) {
652 		ASSERT(whichfork == XFS_ATTR_FORK);
653 		return;
654 	}
655 	cp = XFS_DFORK_PTR(dip, whichfork);
656 	mp = ip->i_mount;
657 	switch (XFS_IFORK_FORMAT(ip, whichfork)) {
658 	case XFS_DINODE_FMT_LOCAL:
659 		if ((iip->ili_fields & dataflag[whichfork]) &&
660 		    (ifp->if_bytes > 0)) {
661 			ASSERT(ifp->if_u1.if_data != NULL);
662 			ASSERT(ifp->if_bytes <= XFS_IFORK_SIZE(ip, whichfork));
663 			memcpy(cp, ifp->if_u1.if_data, ifp->if_bytes);
664 		}
665 		break;
666 
667 	case XFS_DINODE_FMT_EXTENTS:
668 		ASSERT((ifp->if_flags & XFS_IFEXTENTS) ||
669 		       !(iip->ili_fields & extflag[whichfork]));
670 		if ((iip->ili_fields & extflag[whichfork]) &&
671 		    (ifp->if_bytes > 0)) {
672 			ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) > 0);
673 			(void)xfs_iextents_copy(ip, (xfs_bmbt_rec_t *)cp,
674 				whichfork);
675 		}
676 		break;
677 
678 	case XFS_DINODE_FMT_BTREE:
679 		if ((iip->ili_fields & brootflag[whichfork]) &&
680 		    (ifp->if_broot_bytes > 0)) {
681 			ASSERT(ifp->if_broot != NULL);
682 			ASSERT(XFS_BMAP_BMDR_SPACE(ifp->if_broot) <=
683 			        XFS_IFORK_SIZE(ip, whichfork));
684 			xfs_bmbt_to_bmdr(mp, ifp->if_broot, ifp->if_broot_bytes,
685 				(xfs_bmdr_block_t *)cp,
686 				XFS_DFORK_SIZE(dip, mp, whichfork));
687 		}
688 		break;
689 
690 	case XFS_DINODE_FMT_DEV:
691 		if (iip->ili_fields & XFS_ILOG_DEV) {
692 			ASSERT(whichfork == XFS_DATA_FORK);
693 			xfs_dinode_put_rdev(dip,
694 					linux_to_xfs_dev_t(VFS_I(ip)->i_rdev));
695 		}
696 		break;
697 
698 	default:
699 		ASSERT(0);
700 		break;
701 	}
702 }
703 
704 /* Convert bmap state flags to an inode fork. */
705 struct xfs_ifork *
706 xfs_iext_state_to_fork(
707 	struct xfs_inode	*ip,
708 	int			state)
709 {
710 	if (state & BMAP_COWFORK)
711 		return ip->i_cowfp;
712 	else if (state & BMAP_ATTRFORK)
713 		return ip->i_afp;
714 	return &ip->i_df;
715 }
716 
717 /*
718  * Initialize an inode's copy-on-write fork.
719  */
720 void
721 xfs_ifork_init_cow(
722 	struct xfs_inode	*ip)
723 {
724 	if (ip->i_cowfp)
725 		return;
726 
727 	ip->i_cowfp = kmem_zone_zalloc(xfs_ifork_zone,
728 				       KM_SLEEP | KM_NOFS);
729 	ip->i_cowfp->if_flags = XFS_IFEXTENTS;
730 	ip->i_cformat = XFS_DINODE_FMT_EXTENTS;
731 	ip->i_cnextents = 0;
732 }
733 
734 /* Default fork content verifiers. */
735 struct xfs_ifork_ops xfs_default_ifork_ops = {
736 	.verify_attr	= xfs_attr_shortform_verify,
737 	.verify_dir	= xfs_dir2_sf_verify,
738 	.verify_symlink	= xfs_symlink_shortform_verify,
739 };
740 
741 /* Verify the inline contents of the data fork of an inode. */
742 xfs_failaddr_t
743 xfs_ifork_verify_data(
744 	struct xfs_inode	*ip,
745 	struct xfs_ifork_ops	*ops)
746 {
747 	/* Non-local data fork, we're done. */
748 	if (ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
749 		return NULL;
750 
751 	/* Check the inline data fork if there is one. */
752 	switch (VFS_I(ip)->i_mode & S_IFMT) {
753 	case S_IFDIR:
754 		return ops->verify_dir(ip);
755 	case S_IFLNK:
756 		return ops->verify_symlink(ip);
757 	default:
758 		return NULL;
759 	}
760 }
761 
762 /* Verify the inline contents of the attr fork of an inode. */
763 xfs_failaddr_t
764 xfs_ifork_verify_attr(
765 	struct xfs_inode	*ip,
766 	struct xfs_ifork_ops	*ops)
767 {
768 	/* There has to be an attr fork allocated if aformat is local. */
769 	if (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
770 		return NULL;
771 	if (!XFS_IFORK_PTR(ip, XFS_ATTR_FORK))
772 		return __this_address;
773 	return ops->verify_attr(ip);
774 }
775