xref: /linux/fs/xfs/xfs_fsops.c (revision 80d443e8876602be2c130f79c4de81e12e2a700d)
1 /*
2  * Copyright (c) 2000-2005 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 "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_sb.h"
25 #include "xfs_mount.h"
26 #include "xfs_defer.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
29 #include "xfs_inode.h"
30 #include "xfs_trans.h"
31 #include "xfs_inode_item.h"
32 #include "xfs_error.h"
33 #include "xfs_btree.h"
34 #include "xfs_alloc_btree.h"
35 #include "xfs_alloc.h"
36 #include "xfs_rmap_btree.h"
37 #include "xfs_ialloc.h"
38 #include "xfs_fsops.h"
39 #include "xfs_itable.h"
40 #include "xfs_trans_space.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_trace.h"
43 #include "xfs_log.h"
44 #include "xfs_filestream.h"
45 #include "xfs_rmap.h"
46 #include "xfs_ag_resv.h"
47 
48 /*
49  * File system operations
50  */
51 
52 int
53 xfs_fs_geometry(
54 	xfs_mount_t		*mp,
55 	xfs_fsop_geom_t		*geo,
56 	int			new_version)
57 {
58 
59 	memset(geo, 0, sizeof(*geo));
60 
61 	geo->blocksize = mp->m_sb.sb_blocksize;
62 	geo->rtextsize = mp->m_sb.sb_rextsize;
63 	geo->agblocks = mp->m_sb.sb_agblocks;
64 	geo->agcount = mp->m_sb.sb_agcount;
65 	geo->logblocks = mp->m_sb.sb_logblocks;
66 	geo->sectsize = mp->m_sb.sb_sectsize;
67 	geo->inodesize = mp->m_sb.sb_inodesize;
68 	geo->imaxpct = mp->m_sb.sb_imax_pct;
69 	geo->datablocks = mp->m_sb.sb_dblocks;
70 	geo->rtblocks = mp->m_sb.sb_rblocks;
71 	geo->rtextents = mp->m_sb.sb_rextents;
72 	geo->logstart = mp->m_sb.sb_logstart;
73 	ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
74 	memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
75 	if (new_version >= 2) {
76 		geo->sunit = mp->m_sb.sb_unit;
77 		geo->swidth = mp->m_sb.sb_width;
78 	}
79 	if (new_version >= 3) {
80 		geo->version = XFS_FSOP_GEOM_VERSION;
81 		geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
82 			     XFS_FSOP_GEOM_FLAGS_DIRV2 |
83 			(xfs_sb_version_hasattr(&mp->m_sb) ?
84 				XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
85 			(xfs_sb_version_hasquota(&mp->m_sb) ?
86 				XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
87 			(xfs_sb_version_hasalign(&mp->m_sb) ?
88 				XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
89 			(xfs_sb_version_hasdalign(&mp->m_sb) ?
90 				XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
91 			(xfs_sb_version_hasextflgbit(&mp->m_sb) ?
92 				XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
93 			(xfs_sb_version_hassector(&mp->m_sb) ?
94 				XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
95 			(xfs_sb_version_hasasciici(&mp->m_sb) ?
96 				XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
97 			(xfs_sb_version_haslazysbcount(&mp->m_sb) ?
98 				XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
99 			(xfs_sb_version_hasattr2(&mp->m_sb) ?
100 				XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
101 			(xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
102 				XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
103 			(xfs_sb_version_hascrc(&mp->m_sb) ?
104 				XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
105 			(xfs_sb_version_hasftype(&mp->m_sb) ?
106 				XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
107 			(xfs_sb_version_hasfinobt(&mp->m_sb) ?
108 				XFS_FSOP_GEOM_FLAGS_FINOBT : 0) |
109 			(xfs_sb_version_hassparseinodes(&mp->m_sb) ?
110 				XFS_FSOP_GEOM_FLAGS_SPINODES : 0) |
111 			(xfs_sb_version_hasrmapbt(&mp->m_sb) ?
112 				XFS_FSOP_GEOM_FLAGS_RMAPBT : 0) |
113 			(xfs_sb_version_hasreflink(&mp->m_sb) ?
114 				XFS_FSOP_GEOM_FLAGS_REFLINK : 0);
115 		geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
116 				mp->m_sb.sb_logsectsize : BBSIZE;
117 		geo->rtsectsize = mp->m_sb.sb_blocksize;
118 		geo->dirblocksize = mp->m_dir_geo->blksize;
119 	}
120 	if (new_version >= 4) {
121 		geo->flags |=
122 			(xfs_sb_version_haslogv2(&mp->m_sb) ?
123 				XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
124 		geo->logsunit = mp->m_sb.sb_logsunit;
125 	}
126 	return 0;
127 }
128 
129 static struct xfs_buf *
130 xfs_growfs_get_hdr_buf(
131 	struct xfs_mount	*mp,
132 	xfs_daddr_t		blkno,
133 	size_t			numblks,
134 	int			flags,
135 	const struct xfs_buf_ops *ops)
136 {
137 	struct xfs_buf		*bp;
138 
139 	bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
140 	if (!bp)
141 		return NULL;
142 
143 	xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
144 	bp->b_bn = blkno;
145 	bp->b_maps[0].bm_bn = blkno;
146 	bp->b_ops = ops;
147 
148 	return bp;
149 }
150 
151 static int
152 xfs_growfs_data_private(
153 	xfs_mount_t		*mp,		/* mount point for filesystem */
154 	xfs_growfs_data_t	*in)		/* growfs data input struct */
155 {
156 	xfs_agf_t		*agf;
157 	struct xfs_agfl		*agfl;
158 	xfs_agi_t		*agi;
159 	xfs_agnumber_t		agno;
160 	xfs_extlen_t		agsize;
161 	xfs_extlen_t		tmpsize;
162 	xfs_alloc_rec_t		*arec;
163 	xfs_buf_t		*bp;
164 	int			bucket;
165 	int			dpct;
166 	int			error, saved_error = 0;
167 	xfs_agnumber_t		nagcount;
168 	xfs_agnumber_t		nagimax = 0;
169 	xfs_rfsblock_t		nb, nb_mod;
170 	xfs_rfsblock_t		new;
171 	xfs_rfsblock_t		nfree;
172 	xfs_agnumber_t		oagcount;
173 	int			pct;
174 	xfs_trans_t		*tp;
175 
176 	nb = in->newblocks;
177 	pct = in->imaxpct;
178 	if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
179 		return -EINVAL;
180 	if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
181 		return error;
182 	dpct = pct - mp->m_sb.sb_imax_pct;
183 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
184 				XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
185 				XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
186 	if (error)
187 		return error;
188 	xfs_buf_relse(bp);
189 
190 	new = nb;	/* use new as a temporary here */
191 	nb_mod = do_div(new, mp->m_sb.sb_agblocks);
192 	nagcount = new + (nb_mod != 0);
193 	if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
194 		nagcount--;
195 		nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
196 		if (nb < mp->m_sb.sb_dblocks)
197 			return -EINVAL;
198 	}
199 	new = nb - mp->m_sb.sb_dblocks;
200 	oagcount = mp->m_sb.sb_agcount;
201 
202 	/* allocate the new per-ag structures */
203 	if (nagcount > oagcount) {
204 		error = xfs_initialize_perag(mp, nagcount, &nagimax);
205 		if (error)
206 			return error;
207 	}
208 
209 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
210 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
211 	if (error)
212 		return error;
213 
214 	/*
215 	 * Write new AG headers to disk. Non-transactional, but written
216 	 * synchronously so they are completed prior to the growfs transaction
217 	 * being logged.
218 	 */
219 	nfree = 0;
220 	for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
221 		__be32	*agfl_bno;
222 
223 		/*
224 		 * AG freespace header block
225 		 */
226 		bp = xfs_growfs_get_hdr_buf(mp,
227 				XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
228 				XFS_FSS_TO_BB(mp, 1), 0,
229 				&xfs_agf_buf_ops);
230 		if (!bp) {
231 			error = -ENOMEM;
232 			goto error0;
233 		}
234 
235 		agf = XFS_BUF_TO_AGF(bp);
236 		agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
237 		agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
238 		agf->agf_seqno = cpu_to_be32(agno);
239 		if (agno == nagcount - 1)
240 			agsize =
241 				nb -
242 				(agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
243 		else
244 			agsize = mp->m_sb.sb_agblocks;
245 		agf->agf_length = cpu_to_be32(agsize);
246 		agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
247 		agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
248 		agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
249 		agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
250 		if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
251 			agf->agf_roots[XFS_BTNUM_RMAPi] =
252 						cpu_to_be32(XFS_RMAP_BLOCK(mp));
253 			agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
254 			agf->agf_rmap_blocks = cpu_to_be32(1);
255 		}
256 
257 		agf->agf_flfirst = cpu_to_be32(1);
258 		agf->agf_fllast = 0;
259 		agf->agf_flcount = 0;
260 		tmpsize = agsize - mp->m_ag_prealloc_blocks;
261 		agf->agf_freeblks = cpu_to_be32(tmpsize);
262 		agf->agf_longest = cpu_to_be32(tmpsize);
263 		if (xfs_sb_version_hascrc(&mp->m_sb))
264 			uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
265 		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
266 			agf->agf_refcount_root = cpu_to_be32(
267 					xfs_refc_block(mp));
268 			agf->agf_refcount_level = cpu_to_be32(1);
269 			agf->agf_refcount_blocks = cpu_to_be32(1);
270 		}
271 
272 		error = xfs_bwrite(bp);
273 		xfs_buf_relse(bp);
274 		if (error)
275 			goto error0;
276 
277 		/*
278 		 * AG freelist header block
279 		 */
280 		bp = xfs_growfs_get_hdr_buf(mp,
281 				XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
282 				XFS_FSS_TO_BB(mp, 1), 0,
283 				&xfs_agfl_buf_ops);
284 		if (!bp) {
285 			error = -ENOMEM;
286 			goto error0;
287 		}
288 
289 		agfl = XFS_BUF_TO_AGFL(bp);
290 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
291 			agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
292 			agfl->agfl_seqno = cpu_to_be32(agno);
293 			uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
294 		}
295 
296 		agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
297 		for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
298 			agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
299 
300 		error = xfs_bwrite(bp);
301 		xfs_buf_relse(bp);
302 		if (error)
303 			goto error0;
304 
305 		/*
306 		 * AG inode header block
307 		 */
308 		bp = xfs_growfs_get_hdr_buf(mp,
309 				XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
310 				XFS_FSS_TO_BB(mp, 1), 0,
311 				&xfs_agi_buf_ops);
312 		if (!bp) {
313 			error = -ENOMEM;
314 			goto error0;
315 		}
316 
317 		agi = XFS_BUF_TO_AGI(bp);
318 		agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
319 		agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
320 		agi->agi_seqno = cpu_to_be32(agno);
321 		agi->agi_length = cpu_to_be32(agsize);
322 		agi->agi_count = 0;
323 		agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
324 		agi->agi_level = cpu_to_be32(1);
325 		agi->agi_freecount = 0;
326 		agi->agi_newino = cpu_to_be32(NULLAGINO);
327 		agi->agi_dirino = cpu_to_be32(NULLAGINO);
328 		if (xfs_sb_version_hascrc(&mp->m_sb))
329 			uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
330 		if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
331 			agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
332 			agi->agi_free_level = cpu_to_be32(1);
333 		}
334 		for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
335 			agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
336 
337 		error = xfs_bwrite(bp);
338 		xfs_buf_relse(bp);
339 		if (error)
340 			goto error0;
341 
342 		/*
343 		 * BNO btree root block
344 		 */
345 		bp = xfs_growfs_get_hdr_buf(mp,
346 				XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
347 				BTOBB(mp->m_sb.sb_blocksize), 0,
348 				&xfs_allocbt_buf_ops);
349 
350 		if (!bp) {
351 			error = -ENOMEM;
352 			goto error0;
353 		}
354 
355 		if (xfs_sb_version_hascrc(&mp->m_sb))
356 			xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
357 						agno, XFS_BTREE_CRC_BLOCKS);
358 		else
359 			xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
360 						agno, 0);
361 
362 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
363 		arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
364 		arec->ar_blockcount = cpu_to_be32(
365 			agsize - be32_to_cpu(arec->ar_startblock));
366 
367 		error = xfs_bwrite(bp);
368 		xfs_buf_relse(bp);
369 		if (error)
370 			goto error0;
371 
372 		/*
373 		 * CNT btree root block
374 		 */
375 		bp = xfs_growfs_get_hdr_buf(mp,
376 				XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
377 				BTOBB(mp->m_sb.sb_blocksize), 0,
378 				&xfs_allocbt_buf_ops);
379 		if (!bp) {
380 			error = -ENOMEM;
381 			goto error0;
382 		}
383 
384 		if (xfs_sb_version_hascrc(&mp->m_sb))
385 			xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
386 						agno, XFS_BTREE_CRC_BLOCKS);
387 		else
388 			xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
389 						agno, 0);
390 
391 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
392 		arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
393 		arec->ar_blockcount = cpu_to_be32(
394 			agsize - be32_to_cpu(arec->ar_startblock));
395 		nfree += be32_to_cpu(arec->ar_blockcount);
396 
397 		error = xfs_bwrite(bp);
398 		xfs_buf_relse(bp);
399 		if (error)
400 			goto error0;
401 
402 		/* RMAP btree root block */
403 		if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
404 			struct xfs_rmap_rec	*rrec;
405 			struct xfs_btree_block	*block;
406 
407 			bp = xfs_growfs_get_hdr_buf(mp,
408 				XFS_AGB_TO_DADDR(mp, agno, XFS_RMAP_BLOCK(mp)),
409 				BTOBB(mp->m_sb.sb_blocksize), 0,
410 				&xfs_rmapbt_buf_ops);
411 			if (!bp) {
412 				error = -ENOMEM;
413 				goto error0;
414 			}
415 
416 			xfs_btree_init_block(mp, bp, XFS_RMAP_CRC_MAGIC, 0, 0,
417 						agno, XFS_BTREE_CRC_BLOCKS);
418 			block = XFS_BUF_TO_BLOCK(bp);
419 
420 
421 			/*
422 			 * mark the AG header regions as static metadata The BNO
423 			 * btree block is the first block after the headers, so
424 			 * it's location defines the size of region the static
425 			 * metadata consumes.
426 			 *
427 			 * Note: unlike mkfs, we never have to account for log
428 			 * space when growing the data regions
429 			 */
430 			rrec = XFS_RMAP_REC_ADDR(block, 1);
431 			rrec->rm_startblock = 0;
432 			rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
433 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
434 			rrec->rm_offset = 0;
435 			be16_add_cpu(&block->bb_numrecs, 1);
436 
437 			/* account freespace btree root blocks */
438 			rrec = XFS_RMAP_REC_ADDR(block, 2);
439 			rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
440 			rrec->rm_blockcount = cpu_to_be32(2);
441 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
442 			rrec->rm_offset = 0;
443 			be16_add_cpu(&block->bb_numrecs, 1);
444 
445 			/* account inode btree root blocks */
446 			rrec = XFS_RMAP_REC_ADDR(block, 3);
447 			rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
448 			rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
449 							XFS_IBT_BLOCK(mp));
450 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
451 			rrec->rm_offset = 0;
452 			be16_add_cpu(&block->bb_numrecs, 1);
453 
454 			/* account for rmap btree root */
455 			rrec = XFS_RMAP_REC_ADDR(block, 4);
456 			rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
457 			rrec->rm_blockcount = cpu_to_be32(1);
458 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
459 			rrec->rm_offset = 0;
460 			be16_add_cpu(&block->bb_numrecs, 1);
461 
462 			/* account for refc btree root */
463 			if (xfs_sb_version_hasreflink(&mp->m_sb)) {
464 				rrec = XFS_RMAP_REC_ADDR(block, 5);
465 				rrec->rm_startblock = cpu_to_be32(
466 						xfs_refc_block(mp));
467 				rrec->rm_blockcount = cpu_to_be32(1);
468 				rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
469 				rrec->rm_offset = 0;
470 				be16_add_cpu(&block->bb_numrecs, 1);
471 			}
472 
473 			error = xfs_bwrite(bp);
474 			xfs_buf_relse(bp);
475 			if (error)
476 				goto error0;
477 		}
478 
479 		/*
480 		 * INO btree root block
481 		 */
482 		bp = xfs_growfs_get_hdr_buf(mp,
483 				XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
484 				BTOBB(mp->m_sb.sb_blocksize), 0,
485 				&xfs_inobt_buf_ops);
486 		if (!bp) {
487 			error = -ENOMEM;
488 			goto error0;
489 		}
490 
491 		if (xfs_sb_version_hascrc(&mp->m_sb))
492 			xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
493 						agno, XFS_BTREE_CRC_BLOCKS);
494 		else
495 			xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
496 						agno, 0);
497 
498 		error = xfs_bwrite(bp);
499 		xfs_buf_relse(bp);
500 		if (error)
501 			goto error0;
502 
503 		/*
504 		 * FINO btree root block
505 		 */
506 		if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
507 			bp = xfs_growfs_get_hdr_buf(mp,
508 				XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
509 				BTOBB(mp->m_sb.sb_blocksize), 0,
510 				&xfs_inobt_buf_ops);
511 			if (!bp) {
512 				error = -ENOMEM;
513 				goto error0;
514 			}
515 
516 			if (xfs_sb_version_hascrc(&mp->m_sb))
517 				xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
518 						     0, 0, agno,
519 						     XFS_BTREE_CRC_BLOCKS);
520 			else
521 				xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
522 						     0, agno, 0);
523 
524 			error = xfs_bwrite(bp);
525 			xfs_buf_relse(bp);
526 			if (error)
527 				goto error0;
528 		}
529 
530 		/*
531 		 * refcount btree root block
532 		 */
533 		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
534 			bp = xfs_growfs_get_hdr_buf(mp,
535 				XFS_AGB_TO_DADDR(mp, agno, xfs_refc_block(mp)),
536 				BTOBB(mp->m_sb.sb_blocksize), 0,
537 				&xfs_refcountbt_buf_ops);
538 			if (!bp) {
539 				error = -ENOMEM;
540 				goto error0;
541 			}
542 
543 			xfs_btree_init_block(mp, bp, XFS_REFC_CRC_MAGIC,
544 					     0, 0, agno,
545 					     XFS_BTREE_CRC_BLOCKS);
546 
547 			error = xfs_bwrite(bp);
548 			xfs_buf_relse(bp);
549 			if (error)
550 				goto error0;
551 		}
552 	}
553 	xfs_trans_agblocks_delta(tp, nfree);
554 	/*
555 	 * There are new blocks in the old last a.g.
556 	 */
557 	if (new) {
558 		struct xfs_owner_info	oinfo;
559 
560 		/*
561 		 * Change the agi length.
562 		 */
563 		error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
564 		if (error) {
565 			goto error0;
566 		}
567 		ASSERT(bp);
568 		agi = XFS_BUF_TO_AGI(bp);
569 		be32_add_cpu(&agi->agi_length, new);
570 		ASSERT(nagcount == oagcount ||
571 		       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
572 		xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
573 		/*
574 		 * Change agf length.
575 		 */
576 		error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
577 		if (error) {
578 			goto error0;
579 		}
580 		ASSERT(bp);
581 		agf = XFS_BUF_TO_AGF(bp);
582 		be32_add_cpu(&agf->agf_length, new);
583 		ASSERT(be32_to_cpu(agf->agf_length) ==
584 		       be32_to_cpu(agi->agi_length));
585 
586 		xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
587 
588 		/*
589 		 * Free the new space.
590 		 *
591 		 * XFS_RMAP_OWN_NULL is used here to tell the rmap btree that
592 		 * this doesn't actually exist in the rmap btree.
593 		 */
594 		xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_NULL);
595 		error = xfs_free_extent(tp,
596 				XFS_AGB_TO_FSB(mp, agno,
597 					be32_to_cpu(agf->agf_length) - new),
598 				new, &oinfo, XFS_AG_RESV_NONE);
599 		if (error)
600 			goto error0;
601 	}
602 
603 	/*
604 	 * Update changed superblock fields transactionally. These are not
605 	 * seen by the rest of the world until the transaction commit applies
606 	 * them atomically to the superblock.
607 	 */
608 	if (nagcount > oagcount)
609 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
610 	if (nb > mp->m_sb.sb_dblocks)
611 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
612 				 nb - mp->m_sb.sb_dblocks);
613 	if (nfree)
614 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
615 	if (dpct)
616 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
617 	xfs_trans_set_sync(tp);
618 	error = xfs_trans_commit(tp);
619 	if (error)
620 		return error;
621 
622 	/* New allocation groups fully initialized, so update mount struct */
623 	if (nagimax)
624 		mp->m_maxagi = nagimax;
625 	if (mp->m_sb.sb_imax_pct) {
626 		__uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
627 		do_div(icount, 100);
628 		mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
629 	} else
630 		mp->m_maxicount = 0;
631 	xfs_set_low_space_thresholds(mp);
632 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
633 
634 	/* Reserve AG metadata blocks. */
635 	error = xfs_fs_reserve_ag_blocks(mp);
636 	if (error && error != -ENOSPC)
637 		goto out;
638 
639 	/* update secondary superblocks. */
640 	for (agno = 1; agno < nagcount; agno++) {
641 		error = 0;
642 		/*
643 		 * new secondary superblocks need to be zeroed, not read from
644 		 * disk as the contents of the new area we are growing into is
645 		 * completely unknown.
646 		 */
647 		if (agno < oagcount) {
648 			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
649 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
650 				  XFS_FSS_TO_BB(mp, 1), 0, &bp,
651 				  &xfs_sb_buf_ops);
652 		} else {
653 			bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
654 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
655 				  XFS_FSS_TO_BB(mp, 1), 0);
656 			if (bp) {
657 				bp->b_ops = &xfs_sb_buf_ops;
658 				xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
659 			} else
660 				error = -ENOMEM;
661 		}
662 
663 		/*
664 		 * If we get an error reading or writing alternate superblocks,
665 		 * continue.  xfs_repair chooses the "best" superblock based
666 		 * on most matches; if we break early, we'll leave more
667 		 * superblocks un-updated than updated, and xfs_repair may
668 		 * pick them over the properly-updated primary.
669 		 */
670 		if (error) {
671 			xfs_warn(mp,
672 		"error %d reading secondary superblock for ag %d",
673 				error, agno);
674 			saved_error = error;
675 			continue;
676 		}
677 		xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
678 
679 		error = xfs_bwrite(bp);
680 		xfs_buf_relse(bp);
681 		if (error) {
682 			xfs_warn(mp,
683 		"write error %d updating secondary superblock for ag %d",
684 				error, agno);
685 			saved_error = error;
686 			continue;
687 		}
688 	}
689 
690  out:
691 	return saved_error ? saved_error : error;
692 
693  error0:
694 	xfs_trans_cancel(tp);
695 	return error;
696 }
697 
698 static int
699 xfs_growfs_log_private(
700 	xfs_mount_t		*mp,	/* mount point for filesystem */
701 	xfs_growfs_log_t	*in)	/* growfs log input struct */
702 {
703 	xfs_extlen_t		nb;
704 
705 	nb = in->newblocks;
706 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
707 		return -EINVAL;
708 	if (nb == mp->m_sb.sb_logblocks &&
709 	    in->isint == (mp->m_sb.sb_logstart != 0))
710 		return -EINVAL;
711 	/*
712 	 * Moving the log is hard, need new interfaces to sync
713 	 * the log first, hold off all activity while moving it.
714 	 * Can have shorter or longer log in the same space,
715 	 * or transform internal to external log or vice versa.
716 	 */
717 	return -ENOSYS;
718 }
719 
720 /*
721  * protected versions of growfs function acquire and release locks on the mount
722  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
723  * XFS_IOC_FSGROWFSRT
724  */
725 
726 
727 int
728 xfs_growfs_data(
729 	xfs_mount_t		*mp,
730 	xfs_growfs_data_t	*in)
731 {
732 	int error;
733 
734 	if (!capable(CAP_SYS_ADMIN))
735 		return -EPERM;
736 	if (!mutex_trylock(&mp->m_growlock))
737 		return -EWOULDBLOCK;
738 	error = xfs_growfs_data_private(mp, in);
739 	/*
740 	 * Increment the generation unconditionally, the error could be from
741 	 * updating the secondary superblocks, in which case the new size
742 	 * is live already.
743 	 */
744 	mp->m_generation++;
745 	mutex_unlock(&mp->m_growlock);
746 	return error;
747 }
748 
749 int
750 xfs_growfs_log(
751 	xfs_mount_t		*mp,
752 	xfs_growfs_log_t	*in)
753 {
754 	int error;
755 
756 	if (!capable(CAP_SYS_ADMIN))
757 		return -EPERM;
758 	if (!mutex_trylock(&mp->m_growlock))
759 		return -EWOULDBLOCK;
760 	error = xfs_growfs_log_private(mp, in);
761 	mutex_unlock(&mp->m_growlock);
762 	return error;
763 }
764 
765 /*
766  * exported through ioctl XFS_IOC_FSCOUNTS
767  */
768 
769 int
770 xfs_fs_counts(
771 	xfs_mount_t		*mp,
772 	xfs_fsop_counts_t	*cnt)
773 {
774 	cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
775 	cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
776 	cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
777 						mp->m_alloc_set_aside;
778 
779 	spin_lock(&mp->m_sb_lock);
780 	cnt->freertx = mp->m_sb.sb_frextents;
781 	spin_unlock(&mp->m_sb_lock);
782 	return 0;
783 }
784 
785 /*
786  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
787  *
788  * xfs_reserve_blocks is called to set m_resblks
789  * in the in-core mount table. The number of unused reserved blocks
790  * is kept in m_resblks_avail.
791  *
792  * Reserve the requested number of blocks if available. Otherwise return
793  * as many as possible to satisfy the request. The actual number
794  * reserved are returned in outval
795  *
796  * A null inval pointer indicates that only the current reserved blocks
797  * available  should  be returned no settings are changed.
798  */
799 
800 int
801 xfs_reserve_blocks(
802 	xfs_mount_t             *mp,
803 	__uint64_t              *inval,
804 	xfs_fsop_resblks_t      *outval)
805 {
806 	__int64_t		lcounter, delta;
807 	__int64_t		fdblks_delta = 0;
808 	__uint64_t		request;
809 	__int64_t		free;
810 	int			error = 0;
811 
812 	/* If inval is null, report current values and return */
813 	if (inval == (__uint64_t *)NULL) {
814 		if (!outval)
815 			return -EINVAL;
816 		outval->resblks = mp->m_resblks;
817 		outval->resblks_avail = mp->m_resblks_avail;
818 		return 0;
819 	}
820 
821 	request = *inval;
822 
823 	/*
824 	 * With per-cpu counters, this becomes an interesting problem. we need
825 	 * to work out if we are freeing or allocation blocks first, then we can
826 	 * do the modification as necessary.
827 	 *
828 	 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
829 	 * hold out any changes while we work out what to do. This means that
830 	 * the amount of free space can change while we do this, so we need to
831 	 * retry if we end up trying to reserve more space than is available.
832 	 */
833 	spin_lock(&mp->m_sb_lock);
834 
835 	/*
836 	 * If our previous reservation was larger than the current value,
837 	 * then move any unused blocks back to the free pool. Modify the resblks
838 	 * counters directly since we shouldn't have any problems unreserving
839 	 * space.
840 	 */
841 	if (mp->m_resblks > request) {
842 		lcounter = mp->m_resblks_avail - request;
843 		if (lcounter  > 0) {		/* release unused blocks */
844 			fdblks_delta = lcounter;
845 			mp->m_resblks_avail -= lcounter;
846 		}
847 		mp->m_resblks = request;
848 		if (fdblks_delta) {
849 			spin_unlock(&mp->m_sb_lock);
850 			error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
851 			spin_lock(&mp->m_sb_lock);
852 		}
853 
854 		goto out;
855 	}
856 
857 	/*
858 	 * If the request is larger than the current reservation, reserve the
859 	 * blocks before we update the reserve counters. Sample m_fdblocks and
860 	 * perform a partial reservation if the request exceeds free space.
861 	 */
862 	error = -ENOSPC;
863 	do {
864 		free = percpu_counter_sum(&mp->m_fdblocks) -
865 						mp->m_alloc_set_aside;
866 		if (!free)
867 			break;
868 
869 		delta = request - mp->m_resblks;
870 		lcounter = free - delta;
871 		if (lcounter < 0)
872 			/* We can't satisfy the request, just get what we can */
873 			fdblks_delta = free;
874 		else
875 			fdblks_delta = delta;
876 
877 		/*
878 		 * We'll either succeed in getting space from the free block
879 		 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
880 		 * things changed while we were calculating fdblks_delta and so
881 		 * we should try again to see if there is anything left to
882 		 * reserve.
883 		 *
884 		 * Don't set the reserved flag here - we don't want to reserve
885 		 * the extra reserve blocks from the reserve.....
886 		 */
887 		spin_unlock(&mp->m_sb_lock);
888 		error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
889 		spin_lock(&mp->m_sb_lock);
890 	} while (error == -ENOSPC);
891 
892 	/*
893 	 * Update the reserve counters if blocks have been successfully
894 	 * allocated.
895 	 */
896 	if (!error && fdblks_delta) {
897 		mp->m_resblks += fdblks_delta;
898 		mp->m_resblks_avail += fdblks_delta;
899 	}
900 
901 out:
902 	if (outval) {
903 		outval->resblks = mp->m_resblks;
904 		outval->resblks_avail = mp->m_resblks_avail;
905 	}
906 
907 	spin_unlock(&mp->m_sb_lock);
908 	return error;
909 }
910 
911 int
912 xfs_fs_goingdown(
913 	xfs_mount_t	*mp,
914 	__uint32_t	inflags)
915 {
916 	switch (inflags) {
917 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
918 		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
919 
920 		if (sb && !IS_ERR(sb)) {
921 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
922 			thaw_bdev(sb->s_bdev, sb);
923 		}
924 
925 		break;
926 	}
927 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
928 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
929 		break;
930 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
931 		xfs_force_shutdown(mp,
932 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
933 		break;
934 	default:
935 		return -EINVAL;
936 	}
937 
938 	return 0;
939 }
940 
941 /*
942  * Force a shutdown of the filesystem instantly while keeping the filesystem
943  * consistent. We don't do an unmount here; just shutdown the shop, make sure
944  * that absolutely nothing persistent happens to this filesystem after this
945  * point.
946  */
947 void
948 xfs_do_force_shutdown(
949 	xfs_mount_t	*mp,
950 	int		flags,
951 	char		*fname,
952 	int		lnnum)
953 {
954 	int		logerror;
955 
956 	logerror = flags & SHUTDOWN_LOG_IO_ERROR;
957 
958 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
959 		xfs_notice(mp,
960 	"%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
961 			__func__, flags, lnnum, fname, __return_address);
962 	}
963 	/*
964 	 * No need to duplicate efforts.
965 	 */
966 	if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
967 		return;
968 
969 	/*
970 	 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
971 	 * queue up anybody new on the log reservations, and wakes up
972 	 * everybody who's sleeping on log reservations to tell them
973 	 * the bad news.
974 	 */
975 	if (xfs_log_force_umount(mp, logerror))
976 		return;
977 
978 	if (flags & SHUTDOWN_CORRUPT_INCORE) {
979 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
980     "Corruption of in-memory data detected.  Shutting down filesystem");
981 		if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
982 			xfs_stack_trace();
983 	} else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
984 		if (logerror) {
985 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
986 		"Log I/O Error Detected.  Shutting down filesystem");
987 		} else if (flags & SHUTDOWN_DEVICE_REQ) {
988 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
989 		"All device paths lost.  Shutting down filesystem");
990 		} else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
991 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
992 		"I/O Error Detected. Shutting down filesystem");
993 		}
994 	}
995 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
996 		xfs_alert(mp,
997 	"Please umount the filesystem and rectify the problem(s)");
998 	}
999 }
1000 
1001 /*
1002  * Reserve free space for per-AG metadata.
1003  */
1004 int
1005 xfs_fs_reserve_ag_blocks(
1006 	struct xfs_mount	*mp)
1007 {
1008 	xfs_agnumber_t		agno;
1009 	struct xfs_perag	*pag;
1010 	int			error = 0;
1011 	int			err2;
1012 
1013 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1014 		pag = xfs_perag_get(mp, agno);
1015 		err2 = xfs_ag_resv_init(pag);
1016 		xfs_perag_put(pag);
1017 		if (err2 && !error)
1018 			error = err2;
1019 	}
1020 
1021 	if (error && error != -ENOSPC) {
1022 		xfs_warn(mp,
1023 	"Error %d reserving per-AG metadata reserve pool.", error);
1024 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1025 	}
1026 
1027 	return error;
1028 }
1029 
1030 /*
1031  * Free space reserved for per-AG metadata.
1032  */
1033 int
1034 xfs_fs_unreserve_ag_blocks(
1035 	struct xfs_mount	*mp)
1036 {
1037 	xfs_agnumber_t		agno;
1038 	struct xfs_perag	*pag;
1039 	int			error = 0;
1040 	int			err2;
1041 
1042 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1043 		pag = xfs_perag_get(mp, agno);
1044 		err2 = xfs_ag_resv_free(pag);
1045 		xfs_perag_put(pag);
1046 		if (err2 && !error)
1047 			error = err2;
1048 	}
1049 
1050 	if (error)
1051 		xfs_warn(mp,
1052 	"Error %d freeing per-AG metadata reserve pool.", error);
1053 
1054 	return error;
1055 }
1056