xref: /openbmc/linux/fs/xfs/xfs_fsops.c (revision c81ceb58e10124b79b9a01c70b9c25b2fe540475)
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 		xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, agno, 0);
356 
357 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
358 		arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
359 		arec->ar_blockcount = cpu_to_be32(
360 			agsize - be32_to_cpu(arec->ar_startblock));
361 
362 		error = xfs_bwrite(bp);
363 		xfs_buf_relse(bp);
364 		if (error)
365 			goto error0;
366 
367 		/*
368 		 * CNT btree root block
369 		 */
370 		bp = xfs_growfs_get_hdr_buf(mp,
371 				XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
372 				BTOBB(mp->m_sb.sb_blocksize), 0,
373 				&xfs_allocbt_buf_ops);
374 		if (!bp) {
375 			error = -ENOMEM;
376 			goto error0;
377 		}
378 
379 		xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, agno, 0);
380 
381 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
382 		arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
383 		arec->ar_blockcount = cpu_to_be32(
384 			agsize - be32_to_cpu(arec->ar_startblock));
385 		nfree += be32_to_cpu(arec->ar_blockcount);
386 
387 		error = xfs_bwrite(bp);
388 		xfs_buf_relse(bp);
389 		if (error)
390 			goto error0;
391 
392 		/* RMAP btree root block */
393 		if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
394 			struct xfs_rmap_rec	*rrec;
395 			struct xfs_btree_block	*block;
396 
397 			bp = xfs_growfs_get_hdr_buf(mp,
398 				XFS_AGB_TO_DADDR(mp, agno, XFS_RMAP_BLOCK(mp)),
399 				BTOBB(mp->m_sb.sb_blocksize), 0,
400 				&xfs_rmapbt_buf_ops);
401 			if (!bp) {
402 				error = -ENOMEM;
403 				goto error0;
404 			}
405 
406 			xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 0,
407 						agno, 0);
408 			block = XFS_BUF_TO_BLOCK(bp);
409 
410 
411 			/*
412 			 * mark the AG header regions as static metadata The BNO
413 			 * btree block is the first block after the headers, so
414 			 * it's location defines the size of region the static
415 			 * metadata consumes.
416 			 *
417 			 * Note: unlike mkfs, we never have to account for log
418 			 * space when growing the data regions
419 			 */
420 			rrec = XFS_RMAP_REC_ADDR(block, 1);
421 			rrec->rm_startblock = 0;
422 			rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
423 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
424 			rrec->rm_offset = 0;
425 			be16_add_cpu(&block->bb_numrecs, 1);
426 
427 			/* account freespace btree root blocks */
428 			rrec = XFS_RMAP_REC_ADDR(block, 2);
429 			rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
430 			rrec->rm_blockcount = cpu_to_be32(2);
431 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
432 			rrec->rm_offset = 0;
433 			be16_add_cpu(&block->bb_numrecs, 1);
434 
435 			/* account inode btree root blocks */
436 			rrec = XFS_RMAP_REC_ADDR(block, 3);
437 			rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
438 			rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
439 							XFS_IBT_BLOCK(mp));
440 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
441 			rrec->rm_offset = 0;
442 			be16_add_cpu(&block->bb_numrecs, 1);
443 
444 			/* account for rmap btree root */
445 			rrec = XFS_RMAP_REC_ADDR(block, 4);
446 			rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
447 			rrec->rm_blockcount = cpu_to_be32(1);
448 			rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
449 			rrec->rm_offset = 0;
450 			be16_add_cpu(&block->bb_numrecs, 1);
451 
452 			/* account for refc btree root */
453 			if (xfs_sb_version_hasreflink(&mp->m_sb)) {
454 				rrec = XFS_RMAP_REC_ADDR(block, 5);
455 				rrec->rm_startblock = cpu_to_be32(
456 						xfs_refc_block(mp));
457 				rrec->rm_blockcount = cpu_to_be32(1);
458 				rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
459 				rrec->rm_offset = 0;
460 				be16_add_cpu(&block->bb_numrecs, 1);
461 			}
462 
463 			error = xfs_bwrite(bp);
464 			xfs_buf_relse(bp);
465 			if (error)
466 				goto error0;
467 		}
468 
469 		/*
470 		 * INO btree root block
471 		 */
472 		bp = xfs_growfs_get_hdr_buf(mp,
473 				XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
474 				BTOBB(mp->m_sb.sb_blocksize), 0,
475 				&xfs_inobt_buf_ops);
476 		if (!bp) {
477 			error = -ENOMEM;
478 			goto error0;
479 		}
480 
481 		xfs_btree_init_block(mp, bp, XFS_BTNUM_INO , 0, 0, agno, 0);
482 
483 		error = xfs_bwrite(bp);
484 		xfs_buf_relse(bp);
485 		if (error)
486 			goto error0;
487 
488 		/*
489 		 * FINO btree root block
490 		 */
491 		if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
492 			bp = xfs_growfs_get_hdr_buf(mp,
493 				XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
494 				BTOBB(mp->m_sb.sb_blocksize), 0,
495 				&xfs_inobt_buf_ops);
496 			if (!bp) {
497 				error = -ENOMEM;
498 				goto error0;
499 			}
500 
501 			xfs_btree_init_block(mp, bp, XFS_BTNUM_FINO,
502 						     0, 0, agno, 0);
503 
504 			error = xfs_bwrite(bp);
505 			xfs_buf_relse(bp);
506 			if (error)
507 				goto error0;
508 		}
509 
510 		/*
511 		 * refcount btree root block
512 		 */
513 		if (xfs_sb_version_hasreflink(&mp->m_sb)) {
514 			bp = xfs_growfs_get_hdr_buf(mp,
515 				XFS_AGB_TO_DADDR(mp, agno, xfs_refc_block(mp)),
516 				BTOBB(mp->m_sb.sb_blocksize), 0,
517 				&xfs_refcountbt_buf_ops);
518 			if (!bp) {
519 				error = -ENOMEM;
520 				goto error0;
521 			}
522 
523 			xfs_btree_init_block(mp, bp, XFS_BTNUM_REFC,
524 					     0, 0, agno, 0);
525 
526 			error = xfs_bwrite(bp);
527 			xfs_buf_relse(bp);
528 			if (error)
529 				goto error0;
530 		}
531 	}
532 	xfs_trans_agblocks_delta(tp, nfree);
533 	/*
534 	 * There are new blocks in the old last a.g.
535 	 */
536 	if (new) {
537 		struct xfs_owner_info	oinfo;
538 
539 		/*
540 		 * Change the agi length.
541 		 */
542 		error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
543 		if (error) {
544 			goto error0;
545 		}
546 		ASSERT(bp);
547 		agi = XFS_BUF_TO_AGI(bp);
548 		be32_add_cpu(&agi->agi_length, new);
549 		ASSERT(nagcount == oagcount ||
550 		       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
551 		xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
552 		/*
553 		 * Change agf length.
554 		 */
555 		error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
556 		if (error) {
557 			goto error0;
558 		}
559 		ASSERT(bp);
560 		agf = XFS_BUF_TO_AGF(bp);
561 		be32_add_cpu(&agf->agf_length, new);
562 		ASSERT(be32_to_cpu(agf->agf_length) ==
563 		       be32_to_cpu(agi->agi_length));
564 
565 		xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
566 
567 		/*
568 		 * Free the new space.
569 		 *
570 		 * XFS_RMAP_OWN_NULL is used here to tell the rmap btree that
571 		 * this doesn't actually exist in the rmap btree.
572 		 */
573 		xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_NULL);
574 		error = xfs_rmap_free(tp, bp, agno,
575 				be32_to_cpu(agf->agf_length) - new,
576 				new, &oinfo);
577 		if (error)
578 			goto error0;
579 		error = xfs_free_extent(tp,
580 				XFS_AGB_TO_FSB(mp, agno,
581 					be32_to_cpu(agf->agf_length) - new),
582 				new, &oinfo, XFS_AG_RESV_NONE);
583 		if (error)
584 			goto error0;
585 	}
586 
587 	/*
588 	 * Update changed superblock fields transactionally. These are not
589 	 * seen by the rest of the world until the transaction commit applies
590 	 * them atomically to the superblock.
591 	 */
592 	if (nagcount > oagcount)
593 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
594 	if (nb > mp->m_sb.sb_dblocks)
595 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
596 				 nb - mp->m_sb.sb_dblocks);
597 	if (nfree)
598 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
599 	if (dpct)
600 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
601 	xfs_trans_set_sync(tp);
602 	error = xfs_trans_commit(tp);
603 	if (error)
604 		return error;
605 
606 	/* New allocation groups fully initialized, so update mount struct */
607 	if (nagimax)
608 		mp->m_maxagi = nagimax;
609 	if (mp->m_sb.sb_imax_pct) {
610 		uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
611 		do_div(icount, 100);
612 		mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
613 	} else
614 		mp->m_maxicount = 0;
615 	xfs_set_low_space_thresholds(mp);
616 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
617 
618 	/*
619 	 * If we expanded the last AG, free the per-AG reservation
620 	 * so we can reinitialize it with the new size.
621 	 */
622 	if (new) {
623 		struct xfs_perag	*pag;
624 
625 		pag = xfs_perag_get(mp, agno);
626 		error = xfs_ag_resv_free(pag);
627 		xfs_perag_put(pag);
628 		if (error)
629 			goto out;
630 	}
631 
632 	/* Reserve AG metadata blocks. */
633 	error = xfs_fs_reserve_ag_blocks(mp);
634 	if (error && error != -ENOSPC)
635 		goto out;
636 
637 	/* update secondary superblocks. */
638 	for (agno = 1; agno < nagcount; agno++) {
639 		error = 0;
640 		/*
641 		 * new secondary superblocks need to be zeroed, not read from
642 		 * disk as the contents of the new area we are growing into is
643 		 * completely unknown.
644 		 */
645 		if (agno < oagcount) {
646 			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
647 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
648 				  XFS_FSS_TO_BB(mp, 1), 0, &bp,
649 				  &xfs_sb_buf_ops);
650 		} else {
651 			bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
652 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
653 				  XFS_FSS_TO_BB(mp, 1), 0);
654 			if (bp) {
655 				bp->b_ops = &xfs_sb_buf_ops;
656 				xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
657 			} else
658 				error = -ENOMEM;
659 		}
660 
661 		/*
662 		 * If we get an error reading or writing alternate superblocks,
663 		 * continue.  xfs_repair chooses the "best" superblock based
664 		 * on most matches; if we break early, we'll leave more
665 		 * superblocks un-updated than updated, and xfs_repair may
666 		 * pick them over the properly-updated primary.
667 		 */
668 		if (error) {
669 			xfs_warn(mp,
670 		"error %d reading secondary superblock for ag %d",
671 				error, agno);
672 			saved_error = error;
673 			continue;
674 		}
675 		xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
676 
677 		error = xfs_bwrite(bp);
678 		xfs_buf_relse(bp);
679 		if (error) {
680 			xfs_warn(mp,
681 		"write error %d updating secondary superblock for ag %d",
682 				error, agno);
683 			saved_error = error;
684 			continue;
685 		}
686 	}
687 
688  out:
689 	return saved_error ? saved_error : error;
690 
691  error0:
692 	xfs_trans_cancel(tp);
693 	return error;
694 }
695 
696 static int
697 xfs_growfs_log_private(
698 	xfs_mount_t		*mp,	/* mount point for filesystem */
699 	xfs_growfs_log_t	*in)	/* growfs log input struct */
700 {
701 	xfs_extlen_t		nb;
702 
703 	nb = in->newblocks;
704 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
705 		return -EINVAL;
706 	if (nb == mp->m_sb.sb_logblocks &&
707 	    in->isint == (mp->m_sb.sb_logstart != 0))
708 		return -EINVAL;
709 	/*
710 	 * Moving the log is hard, need new interfaces to sync
711 	 * the log first, hold off all activity while moving it.
712 	 * Can have shorter or longer log in the same space,
713 	 * or transform internal to external log or vice versa.
714 	 */
715 	return -ENOSYS;
716 }
717 
718 /*
719  * protected versions of growfs function acquire and release locks on the mount
720  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
721  * XFS_IOC_FSGROWFSRT
722  */
723 
724 
725 int
726 xfs_growfs_data(
727 	xfs_mount_t		*mp,
728 	xfs_growfs_data_t	*in)
729 {
730 	int error;
731 
732 	if (!capable(CAP_SYS_ADMIN))
733 		return -EPERM;
734 	if (!mutex_trylock(&mp->m_growlock))
735 		return -EWOULDBLOCK;
736 	error = xfs_growfs_data_private(mp, in);
737 	/*
738 	 * Increment the generation unconditionally, the error could be from
739 	 * updating the secondary superblocks, in which case the new size
740 	 * is live already.
741 	 */
742 	mp->m_generation++;
743 	mutex_unlock(&mp->m_growlock);
744 	return error;
745 }
746 
747 int
748 xfs_growfs_log(
749 	xfs_mount_t		*mp,
750 	xfs_growfs_log_t	*in)
751 {
752 	int error;
753 
754 	if (!capable(CAP_SYS_ADMIN))
755 		return -EPERM;
756 	if (!mutex_trylock(&mp->m_growlock))
757 		return -EWOULDBLOCK;
758 	error = xfs_growfs_log_private(mp, in);
759 	mutex_unlock(&mp->m_growlock);
760 	return error;
761 }
762 
763 /*
764  * exported through ioctl XFS_IOC_FSCOUNTS
765  */
766 
767 int
768 xfs_fs_counts(
769 	xfs_mount_t		*mp,
770 	xfs_fsop_counts_t	*cnt)
771 {
772 	cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
773 	cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
774 	cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
775 						mp->m_alloc_set_aside;
776 
777 	spin_lock(&mp->m_sb_lock);
778 	cnt->freertx = mp->m_sb.sb_frextents;
779 	spin_unlock(&mp->m_sb_lock);
780 	return 0;
781 }
782 
783 /*
784  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
785  *
786  * xfs_reserve_blocks is called to set m_resblks
787  * in the in-core mount table. The number of unused reserved blocks
788  * is kept in m_resblks_avail.
789  *
790  * Reserve the requested number of blocks if available. Otherwise return
791  * as many as possible to satisfy the request. The actual number
792  * reserved are returned in outval
793  *
794  * A null inval pointer indicates that only the current reserved blocks
795  * available  should  be returned no settings are changed.
796  */
797 
798 int
799 xfs_reserve_blocks(
800 	xfs_mount_t             *mp,
801 	uint64_t              *inval,
802 	xfs_fsop_resblks_t      *outval)
803 {
804 	int64_t			lcounter, delta;
805 	int64_t			fdblks_delta = 0;
806 	uint64_t		request;
807 	int64_t			free;
808 	int			error = 0;
809 
810 	/* If inval is null, report current values and return */
811 	if (inval == (uint64_t *)NULL) {
812 		if (!outval)
813 			return -EINVAL;
814 		outval->resblks = mp->m_resblks;
815 		outval->resblks_avail = mp->m_resblks_avail;
816 		return 0;
817 	}
818 
819 	request = *inval;
820 
821 	/*
822 	 * With per-cpu counters, this becomes an interesting problem. we need
823 	 * to work out if we are freeing or allocation blocks first, then we can
824 	 * do the modification as necessary.
825 	 *
826 	 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
827 	 * hold out any changes while we work out what to do. This means that
828 	 * the amount of free space can change while we do this, so we need to
829 	 * retry if we end up trying to reserve more space than is available.
830 	 */
831 	spin_lock(&mp->m_sb_lock);
832 
833 	/*
834 	 * If our previous reservation was larger than the current value,
835 	 * then move any unused blocks back to the free pool. Modify the resblks
836 	 * counters directly since we shouldn't have any problems unreserving
837 	 * space.
838 	 */
839 	if (mp->m_resblks > request) {
840 		lcounter = mp->m_resblks_avail - request;
841 		if (lcounter  > 0) {		/* release unused blocks */
842 			fdblks_delta = lcounter;
843 			mp->m_resblks_avail -= lcounter;
844 		}
845 		mp->m_resblks = request;
846 		if (fdblks_delta) {
847 			spin_unlock(&mp->m_sb_lock);
848 			error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
849 			spin_lock(&mp->m_sb_lock);
850 		}
851 
852 		goto out;
853 	}
854 
855 	/*
856 	 * If the request is larger than the current reservation, reserve the
857 	 * blocks before we update the reserve counters. Sample m_fdblocks and
858 	 * perform a partial reservation if the request exceeds free space.
859 	 */
860 	error = -ENOSPC;
861 	do {
862 		free = percpu_counter_sum(&mp->m_fdblocks) -
863 						mp->m_alloc_set_aside;
864 		if (!free)
865 			break;
866 
867 		delta = request - mp->m_resblks;
868 		lcounter = free - delta;
869 		if (lcounter < 0)
870 			/* We can't satisfy the request, just get what we can */
871 			fdblks_delta = free;
872 		else
873 			fdblks_delta = delta;
874 
875 		/*
876 		 * We'll either succeed in getting space from the free block
877 		 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
878 		 * things changed while we were calculating fdblks_delta and so
879 		 * we should try again to see if there is anything left to
880 		 * reserve.
881 		 *
882 		 * Don't set the reserved flag here - we don't want to reserve
883 		 * the extra reserve blocks from the reserve.....
884 		 */
885 		spin_unlock(&mp->m_sb_lock);
886 		error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
887 		spin_lock(&mp->m_sb_lock);
888 	} while (error == -ENOSPC);
889 
890 	/*
891 	 * Update the reserve counters if blocks have been successfully
892 	 * allocated.
893 	 */
894 	if (!error && fdblks_delta) {
895 		mp->m_resblks += fdblks_delta;
896 		mp->m_resblks_avail += fdblks_delta;
897 	}
898 
899 out:
900 	if (outval) {
901 		outval->resblks = mp->m_resblks;
902 		outval->resblks_avail = mp->m_resblks_avail;
903 	}
904 
905 	spin_unlock(&mp->m_sb_lock);
906 	return error;
907 }
908 
909 int
910 xfs_fs_goingdown(
911 	xfs_mount_t	*mp,
912 	uint32_t	inflags)
913 {
914 	switch (inflags) {
915 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
916 		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
917 
918 		if (sb && !IS_ERR(sb)) {
919 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
920 			thaw_bdev(sb->s_bdev, sb);
921 		}
922 
923 		break;
924 	}
925 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
926 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
927 		break;
928 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
929 		xfs_force_shutdown(mp,
930 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
931 		break;
932 	default:
933 		return -EINVAL;
934 	}
935 
936 	return 0;
937 }
938 
939 /*
940  * Force a shutdown of the filesystem instantly while keeping the filesystem
941  * consistent. We don't do an unmount here; just shutdown the shop, make sure
942  * that absolutely nothing persistent happens to this filesystem after this
943  * point.
944  */
945 void
946 xfs_do_force_shutdown(
947 	xfs_mount_t	*mp,
948 	int		flags,
949 	char		*fname,
950 	int		lnnum)
951 {
952 	int		logerror;
953 
954 	logerror = flags & SHUTDOWN_LOG_IO_ERROR;
955 
956 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
957 		xfs_notice(mp,
958 	"%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
959 			__func__, flags, lnnum, fname, __return_address);
960 	}
961 	/*
962 	 * No need to duplicate efforts.
963 	 */
964 	if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
965 		return;
966 
967 	/*
968 	 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
969 	 * queue up anybody new on the log reservations, and wakes up
970 	 * everybody who's sleeping on log reservations to tell them
971 	 * the bad news.
972 	 */
973 	if (xfs_log_force_umount(mp, logerror))
974 		return;
975 
976 	if (flags & SHUTDOWN_CORRUPT_INCORE) {
977 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
978     "Corruption of in-memory data detected.  Shutting down filesystem");
979 		if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
980 			xfs_stack_trace();
981 	} else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
982 		if (logerror) {
983 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
984 		"Log I/O Error Detected.  Shutting down filesystem");
985 		} else if (flags & SHUTDOWN_DEVICE_REQ) {
986 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
987 		"All device paths lost.  Shutting down filesystem");
988 		} else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
989 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
990 		"I/O Error Detected. Shutting down filesystem");
991 		}
992 	}
993 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
994 		xfs_alert(mp,
995 	"Please umount the filesystem and rectify the problem(s)");
996 	}
997 }
998 
999 /*
1000  * Reserve free space for per-AG metadata.
1001  */
1002 int
1003 xfs_fs_reserve_ag_blocks(
1004 	struct xfs_mount	*mp)
1005 {
1006 	xfs_agnumber_t		agno;
1007 	struct xfs_perag	*pag;
1008 	int			error = 0;
1009 	int			err2;
1010 
1011 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1012 		pag = xfs_perag_get(mp, agno);
1013 		err2 = xfs_ag_resv_init(pag);
1014 		xfs_perag_put(pag);
1015 		if (err2 && !error)
1016 			error = err2;
1017 	}
1018 
1019 	if (error && error != -ENOSPC) {
1020 		xfs_warn(mp,
1021 	"Error %d reserving per-AG metadata reserve pool.", error);
1022 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1023 	}
1024 
1025 	return error;
1026 }
1027 
1028 /*
1029  * Free space reserved for per-AG metadata.
1030  */
1031 int
1032 xfs_fs_unreserve_ag_blocks(
1033 	struct xfs_mount	*mp)
1034 {
1035 	xfs_agnumber_t		agno;
1036 	struct xfs_perag	*pag;
1037 	int			error = 0;
1038 	int			err2;
1039 
1040 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
1041 		pag = xfs_perag_get(mp, agno);
1042 		err2 = xfs_ag_resv_free(pag);
1043 		xfs_perag_put(pag);
1044 		if (err2 && !error)
1045 			error = err2;
1046 	}
1047 
1048 	if (error)
1049 		xfs_warn(mp,
1050 	"Error %d freeing per-AG metadata reserve pool.", error);
1051 
1052 	return error;
1053 }
1054