xref: /openbmc/linux/fs/xfs/xfs_fsops.c (revision 7587eb18)
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_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
29 #include "xfs_trans.h"
30 #include "xfs_inode_item.h"
31 #include "xfs_error.h"
32 #include "xfs_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_alloc.h"
35 #include "xfs_ialloc.h"
36 #include "xfs_fsops.h"
37 #include "xfs_itable.h"
38 #include "xfs_trans_space.h"
39 #include "xfs_rtalloc.h"
40 #include "xfs_trace.h"
41 #include "xfs_log.h"
42 #include "xfs_filestream.h"
43 
44 /*
45  * File system operations
46  */
47 
48 int
49 xfs_fs_geometry(
50 	xfs_mount_t		*mp,
51 	xfs_fsop_geom_t		*geo,
52 	int			new_version)
53 {
54 
55 	memset(geo, 0, sizeof(*geo));
56 
57 	geo->blocksize = mp->m_sb.sb_blocksize;
58 	geo->rtextsize = mp->m_sb.sb_rextsize;
59 	geo->agblocks = mp->m_sb.sb_agblocks;
60 	geo->agcount = mp->m_sb.sb_agcount;
61 	geo->logblocks = mp->m_sb.sb_logblocks;
62 	geo->sectsize = mp->m_sb.sb_sectsize;
63 	geo->inodesize = mp->m_sb.sb_inodesize;
64 	geo->imaxpct = mp->m_sb.sb_imax_pct;
65 	geo->datablocks = mp->m_sb.sb_dblocks;
66 	geo->rtblocks = mp->m_sb.sb_rblocks;
67 	geo->rtextents = mp->m_sb.sb_rextents;
68 	geo->logstart = mp->m_sb.sb_logstart;
69 	ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
70 	memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
71 	if (new_version >= 2) {
72 		geo->sunit = mp->m_sb.sb_unit;
73 		geo->swidth = mp->m_sb.sb_width;
74 	}
75 	if (new_version >= 3) {
76 		geo->version = XFS_FSOP_GEOM_VERSION;
77 		geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
78 			     XFS_FSOP_GEOM_FLAGS_DIRV2 |
79 			(xfs_sb_version_hasattr(&mp->m_sb) ?
80 				XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
81 			(xfs_sb_version_hasquota(&mp->m_sb) ?
82 				XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
83 			(xfs_sb_version_hasalign(&mp->m_sb) ?
84 				XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
85 			(xfs_sb_version_hasdalign(&mp->m_sb) ?
86 				XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
87 			(xfs_sb_version_hasextflgbit(&mp->m_sb) ?
88 				XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
89 			(xfs_sb_version_hassector(&mp->m_sb) ?
90 				XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
91 			(xfs_sb_version_hasasciici(&mp->m_sb) ?
92 				XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
93 			(xfs_sb_version_haslazysbcount(&mp->m_sb) ?
94 				XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
95 			(xfs_sb_version_hasattr2(&mp->m_sb) ?
96 				XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
97 			(xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
98 				XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
99 			(xfs_sb_version_hascrc(&mp->m_sb) ?
100 				XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
101 			(xfs_sb_version_hasftype(&mp->m_sb) ?
102 				XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
103 			(xfs_sb_version_hasfinobt(&mp->m_sb) ?
104 				XFS_FSOP_GEOM_FLAGS_FINOBT : 0) |
105 			(xfs_sb_version_hassparseinodes(&mp->m_sb) ?
106 				XFS_FSOP_GEOM_FLAGS_SPINODES : 0);
107 		geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
108 				mp->m_sb.sb_logsectsize : BBSIZE;
109 		geo->rtsectsize = mp->m_sb.sb_blocksize;
110 		geo->dirblocksize = mp->m_dir_geo->blksize;
111 	}
112 	if (new_version >= 4) {
113 		geo->flags |=
114 			(xfs_sb_version_haslogv2(&mp->m_sb) ?
115 				XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
116 		geo->logsunit = mp->m_sb.sb_logsunit;
117 	}
118 	return 0;
119 }
120 
121 static struct xfs_buf *
122 xfs_growfs_get_hdr_buf(
123 	struct xfs_mount	*mp,
124 	xfs_daddr_t		blkno,
125 	size_t			numblks,
126 	int			flags,
127 	const struct xfs_buf_ops *ops)
128 {
129 	struct xfs_buf		*bp;
130 
131 	bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
132 	if (!bp)
133 		return NULL;
134 
135 	xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
136 	bp->b_bn = blkno;
137 	bp->b_maps[0].bm_bn = blkno;
138 	bp->b_ops = ops;
139 
140 	return bp;
141 }
142 
143 static int
144 xfs_growfs_data_private(
145 	xfs_mount_t		*mp,		/* mount point for filesystem */
146 	xfs_growfs_data_t	*in)		/* growfs data input struct */
147 {
148 	xfs_agf_t		*agf;
149 	struct xfs_agfl		*agfl;
150 	xfs_agi_t		*agi;
151 	xfs_agnumber_t		agno;
152 	xfs_extlen_t		agsize;
153 	xfs_extlen_t		tmpsize;
154 	xfs_alloc_rec_t		*arec;
155 	xfs_buf_t		*bp;
156 	int			bucket;
157 	int			dpct;
158 	int			error, saved_error = 0;
159 	xfs_agnumber_t		nagcount;
160 	xfs_agnumber_t		nagimax = 0;
161 	xfs_rfsblock_t		nb, nb_mod;
162 	xfs_rfsblock_t		new;
163 	xfs_rfsblock_t		nfree;
164 	xfs_agnumber_t		oagcount;
165 	int			pct;
166 	xfs_trans_t		*tp;
167 
168 	nb = in->newblocks;
169 	pct = in->imaxpct;
170 	if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
171 		return -EINVAL;
172 	if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
173 		return error;
174 	dpct = pct - mp->m_sb.sb_imax_pct;
175 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
176 				XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
177 				XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
178 	if (error)
179 		return error;
180 	xfs_buf_relse(bp);
181 
182 	new = nb;	/* use new as a temporary here */
183 	nb_mod = do_div(new, mp->m_sb.sb_agblocks);
184 	nagcount = new + (nb_mod != 0);
185 	if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
186 		nagcount--;
187 		nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
188 		if (nb < mp->m_sb.sb_dblocks)
189 			return -EINVAL;
190 	}
191 	new = nb - mp->m_sb.sb_dblocks;
192 	oagcount = mp->m_sb.sb_agcount;
193 
194 	/* allocate the new per-ag structures */
195 	if (nagcount > oagcount) {
196 		error = xfs_initialize_perag(mp, nagcount, &nagimax);
197 		if (error)
198 			return error;
199 	}
200 
201 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
202 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
203 	if (error)
204 		return error;
205 
206 	/*
207 	 * Write new AG headers to disk. Non-transactional, but written
208 	 * synchronously so they are completed prior to the growfs transaction
209 	 * being logged.
210 	 */
211 	nfree = 0;
212 	for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
213 		__be32	*agfl_bno;
214 
215 		/*
216 		 * AG freespace header block
217 		 */
218 		bp = xfs_growfs_get_hdr_buf(mp,
219 				XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
220 				XFS_FSS_TO_BB(mp, 1), 0,
221 				&xfs_agf_buf_ops);
222 		if (!bp) {
223 			error = -ENOMEM;
224 			goto error0;
225 		}
226 
227 		agf = XFS_BUF_TO_AGF(bp);
228 		agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
229 		agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
230 		agf->agf_seqno = cpu_to_be32(agno);
231 		if (agno == nagcount - 1)
232 			agsize =
233 				nb -
234 				(agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
235 		else
236 			agsize = mp->m_sb.sb_agblocks;
237 		agf->agf_length = cpu_to_be32(agsize);
238 		agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
239 		agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
240 		agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
241 		agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
242 		agf->agf_flfirst = cpu_to_be32(1);
243 		agf->agf_fllast = 0;
244 		agf->agf_flcount = 0;
245 		tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
246 		agf->agf_freeblks = cpu_to_be32(tmpsize);
247 		agf->agf_longest = cpu_to_be32(tmpsize);
248 		if (xfs_sb_version_hascrc(&mp->m_sb))
249 			uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
250 
251 		error = xfs_bwrite(bp);
252 		xfs_buf_relse(bp);
253 		if (error)
254 			goto error0;
255 
256 		/*
257 		 * AG freelist header block
258 		 */
259 		bp = xfs_growfs_get_hdr_buf(mp,
260 				XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
261 				XFS_FSS_TO_BB(mp, 1), 0,
262 				&xfs_agfl_buf_ops);
263 		if (!bp) {
264 			error = -ENOMEM;
265 			goto error0;
266 		}
267 
268 		agfl = XFS_BUF_TO_AGFL(bp);
269 		if (xfs_sb_version_hascrc(&mp->m_sb)) {
270 			agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
271 			agfl->agfl_seqno = cpu_to_be32(agno);
272 			uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
273 		}
274 
275 		agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
276 		for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
277 			agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
278 
279 		error = xfs_bwrite(bp);
280 		xfs_buf_relse(bp);
281 		if (error)
282 			goto error0;
283 
284 		/*
285 		 * AG inode header block
286 		 */
287 		bp = xfs_growfs_get_hdr_buf(mp,
288 				XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
289 				XFS_FSS_TO_BB(mp, 1), 0,
290 				&xfs_agi_buf_ops);
291 		if (!bp) {
292 			error = -ENOMEM;
293 			goto error0;
294 		}
295 
296 		agi = XFS_BUF_TO_AGI(bp);
297 		agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
298 		agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
299 		agi->agi_seqno = cpu_to_be32(agno);
300 		agi->agi_length = cpu_to_be32(agsize);
301 		agi->agi_count = 0;
302 		agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
303 		agi->agi_level = cpu_to_be32(1);
304 		agi->agi_freecount = 0;
305 		agi->agi_newino = cpu_to_be32(NULLAGINO);
306 		agi->agi_dirino = cpu_to_be32(NULLAGINO);
307 		if (xfs_sb_version_hascrc(&mp->m_sb))
308 			uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
309 		if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
310 			agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
311 			agi->agi_free_level = cpu_to_be32(1);
312 		}
313 		for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
314 			agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
315 
316 		error = xfs_bwrite(bp);
317 		xfs_buf_relse(bp);
318 		if (error)
319 			goto error0;
320 
321 		/*
322 		 * BNO btree root block
323 		 */
324 		bp = xfs_growfs_get_hdr_buf(mp,
325 				XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
326 				BTOBB(mp->m_sb.sb_blocksize), 0,
327 				&xfs_allocbt_buf_ops);
328 
329 		if (!bp) {
330 			error = -ENOMEM;
331 			goto error0;
332 		}
333 
334 		if (xfs_sb_version_hascrc(&mp->m_sb))
335 			xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
336 						agno, XFS_BTREE_CRC_BLOCKS);
337 		else
338 			xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
339 						agno, 0);
340 
341 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
342 		arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
343 		arec->ar_blockcount = cpu_to_be32(
344 			agsize - be32_to_cpu(arec->ar_startblock));
345 
346 		error = xfs_bwrite(bp);
347 		xfs_buf_relse(bp);
348 		if (error)
349 			goto error0;
350 
351 		/*
352 		 * CNT btree root block
353 		 */
354 		bp = xfs_growfs_get_hdr_buf(mp,
355 				XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
356 				BTOBB(mp->m_sb.sb_blocksize), 0,
357 				&xfs_allocbt_buf_ops);
358 		if (!bp) {
359 			error = -ENOMEM;
360 			goto error0;
361 		}
362 
363 		if (xfs_sb_version_hascrc(&mp->m_sb))
364 			xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
365 						agno, XFS_BTREE_CRC_BLOCKS);
366 		else
367 			xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
368 						agno, 0);
369 
370 		arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
371 		arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
372 		arec->ar_blockcount = cpu_to_be32(
373 			agsize - be32_to_cpu(arec->ar_startblock));
374 		nfree += be32_to_cpu(arec->ar_blockcount);
375 
376 		error = xfs_bwrite(bp);
377 		xfs_buf_relse(bp);
378 		if (error)
379 			goto error0;
380 
381 		/*
382 		 * INO btree root block
383 		 */
384 		bp = xfs_growfs_get_hdr_buf(mp,
385 				XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
386 				BTOBB(mp->m_sb.sb_blocksize), 0,
387 				&xfs_inobt_buf_ops);
388 		if (!bp) {
389 			error = -ENOMEM;
390 			goto error0;
391 		}
392 
393 		if (xfs_sb_version_hascrc(&mp->m_sb))
394 			xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
395 						agno, XFS_BTREE_CRC_BLOCKS);
396 		else
397 			xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
398 						agno, 0);
399 
400 		error = xfs_bwrite(bp);
401 		xfs_buf_relse(bp);
402 		if (error)
403 			goto error0;
404 
405 		/*
406 		 * FINO btree root block
407 		 */
408 		if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
409 			bp = xfs_growfs_get_hdr_buf(mp,
410 				XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
411 				BTOBB(mp->m_sb.sb_blocksize), 0,
412 				&xfs_inobt_buf_ops);
413 			if (!bp) {
414 				error = -ENOMEM;
415 				goto error0;
416 			}
417 
418 			if (xfs_sb_version_hascrc(&mp->m_sb))
419 				xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
420 						     0, 0, agno,
421 						     XFS_BTREE_CRC_BLOCKS);
422 			else
423 				xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
424 						     0, agno, 0);
425 
426 			error = xfs_bwrite(bp);
427 			xfs_buf_relse(bp);
428 			if (error)
429 				goto error0;
430 		}
431 
432 	}
433 	xfs_trans_agblocks_delta(tp, nfree);
434 	/*
435 	 * There are new blocks in the old last a.g.
436 	 */
437 	if (new) {
438 		/*
439 		 * Change the agi length.
440 		 */
441 		error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
442 		if (error) {
443 			goto error0;
444 		}
445 		ASSERT(bp);
446 		agi = XFS_BUF_TO_AGI(bp);
447 		be32_add_cpu(&agi->agi_length, new);
448 		ASSERT(nagcount == oagcount ||
449 		       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
450 		xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
451 		/*
452 		 * Change agf length.
453 		 */
454 		error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
455 		if (error) {
456 			goto error0;
457 		}
458 		ASSERT(bp);
459 		agf = XFS_BUF_TO_AGF(bp);
460 		be32_add_cpu(&agf->agf_length, new);
461 		ASSERT(be32_to_cpu(agf->agf_length) ==
462 		       be32_to_cpu(agi->agi_length));
463 
464 		xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
465 		/*
466 		 * Free the new space.
467 		 */
468 		error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
469 			be32_to_cpu(agf->agf_length) - new), new);
470 		if (error) {
471 			goto error0;
472 		}
473 	}
474 
475 	/*
476 	 * Update changed superblock fields transactionally. These are not
477 	 * seen by the rest of the world until the transaction commit applies
478 	 * them atomically to the superblock.
479 	 */
480 	if (nagcount > oagcount)
481 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
482 	if (nb > mp->m_sb.sb_dblocks)
483 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
484 				 nb - mp->m_sb.sb_dblocks);
485 	if (nfree)
486 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
487 	if (dpct)
488 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
489 	xfs_trans_set_sync(tp);
490 	error = xfs_trans_commit(tp);
491 	if (error)
492 		return error;
493 
494 	/* New allocation groups fully initialized, so update mount struct */
495 	if (nagimax)
496 		mp->m_maxagi = nagimax;
497 	if (mp->m_sb.sb_imax_pct) {
498 		__uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
499 		do_div(icount, 100);
500 		mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
501 	} else
502 		mp->m_maxicount = 0;
503 	xfs_set_low_space_thresholds(mp);
504 
505 	/* update secondary superblocks. */
506 	for (agno = 1; agno < nagcount; agno++) {
507 		error = 0;
508 		/*
509 		 * new secondary superblocks need to be zeroed, not read from
510 		 * disk as the contents of the new area we are growing into is
511 		 * completely unknown.
512 		 */
513 		if (agno < oagcount) {
514 			error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
515 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
516 				  XFS_FSS_TO_BB(mp, 1), 0, &bp,
517 				  &xfs_sb_buf_ops);
518 		} else {
519 			bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
520 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
521 				  XFS_FSS_TO_BB(mp, 1), 0);
522 			if (bp) {
523 				bp->b_ops = &xfs_sb_buf_ops;
524 				xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
525 			} else
526 				error = -ENOMEM;
527 		}
528 
529 		/*
530 		 * If we get an error reading or writing alternate superblocks,
531 		 * continue.  xfs_repair chooses the "best" superblock based
532 		 * on most matches; if we break early, we'll leave more
533 		 * superblocks un-updated than updated, and xfs_repair may
534 		 * pick them over the properly-updated primary.
535 		 */
536 		if (error) {
537 			xfs_warn(mp,
538 		"error %d reading secondary superblock for ag %d",
539 				error, agno);
540 			saved_error = error;
541 			continue;
542 		}
543 		xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
544 
545 		error = xfs_bwrite(bp);
546 		xfs_buf_relse(bp);
547 		if (error) {
548 			xfs_warn(mp,
549 		"write error %d updating secondary superblock for ag %d",
550 				error, agno);
551 			saved_error = error;
552 			continue;
553 		}
554 	}
555 	return saved_error ? saved_error : error;
556 
557  error0:
558 	xfs_trans_cancel(tp);
559 	return error;
560 }
561 
562 static int
563 xfs_growfs_log_private(
564 	xfs_mount_t		*mp,	/* mount point for filesystem */
565 	xfs_growfs_log_t	*in)	/* growfs log input struct */
566 {
567 	xfs_extlen_t		nb;
568 
569 	nb = in->newblocks;
570 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
571 		return -EINVAL;
572 	if (nb == mp->m_sb.sb_logblocks &&
573 	    in->isint == (mp->m_sb.sb_logstart != 0))
574 		return -EINVAL;
575 	/*
576 	 * Moving the log is hard, need new interfaces to sync
577 	 * the log first, hold off all activity while moving it.
578 	 * Can have shorter or longer log in the same space,
579 	 * or transform internal to external log or vice versa.
580 	 */
581 	return -ENOSYS;
582 }
583 
584 /*
585  * protected versions of growfs function acquire and release locks on the mount
586  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
587  * XFS_IOC_FSGROWFSRT
588  */
589 
590 
591 int
592 xfs_growfs_data(
593 	xfs_mount_t		*mp,
594 	xfs_growfs_data_t	*in)
595 {
596 	int error;
597 
598 	if (!capable(CAP_SYS_ADMIN))
599 		return -EPERM;
600 	if (!mutex_trylock(&mp->m_growlock))
601 		return -EWOULDBLOCK;
602 	error = xfs_growfs_data_private(mp, in);
603 	/*
604 	 * Increment the generation unconditionally, the error could be from
605 	 * updating the secondary superblocks, in which case the new size
606 	 * is live already.
607 	 */
608 	mp->m_generation++;
609 	mutex_unlock(&mp->m_growlock);
610 	return error;
611 }
612 
613 int
614 xfs_growfs_log(
615 	xfs_mount_t		*mp,
616 	xfs_growfs_log_t	*in)
617 {
618 	int error;
619 
620 	if (!capable(CAP_SYS_ADMIN))
621 		return -EPERM;
622 	if (!mutex_trylock(&mp->m_growlock))
623 		return -EWOULDBLOCK;
624 	error = xfs_growfs_log_private(mp, in);
625 	mutex_unlock(&mp->m_growlock);
626 	return error;
627 }
628 
629 /*
630  * exported through ioctl XFS_IOC_FSCOUNTS
631  */
632 
633 int
634 xfs_fs_counts(
635 	xfs_mount_t		*mp,
636 	xfs_fsop_counts_t	*cnt)
637 {
638 	cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
639 	cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
640 	cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
641 							XFS_ALLOC_SET_ASIDE(mp);
642 
643 	spin_lock(&mp->m_sb_lock);
644 	cnt->freertx = mp->m_sb.sb_frextents;
645 	spin_unlock(&mp->m_sb_lock);
646 	return 0;
647 }
648 
649 /*
650  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
651  *
652  * xfs_reserve_blocks is called to set m_resblks
653  * in the in-core mount table. The number of unused reserved blocks
654  * is kept in m_resblks_avail.
655  *
656  * Reserve the requested number of blocks if available. Otherwise return
657  * as many as possible to satisfy the request. The actual number
658  * reserved are returned in outval
659  *
660  * A null inval pointer indicates that only the current reserved blocks
661  * available  should  be returned no settings are changed.
662  */
663 
664 int
665 xfs_reserve_blocks(
666 	xfs_mount_t             *mp,
667 	__uint64_t              *inval,
668 	xfs_fsop_resblks_t      *outval)
669 {
670 	__int64_t		lcounter, delta, fdblks_delta;
671 	__uint64_t		request;
672 
673 	/* If inval is null, report current values and return */
674 	if (inval == (__uint64_t *)NULL) {
675 		if (!outval)
676 			return -EINVAL;
677 		outval->resblks = mp->m_resblks;
678 		outval->resblks_avail = mp->m_resblks_avail;
679 		return 0;
680 	}
681 
682 	request = *inval;
683 
684 	/*
685 	 * With per-cpu counters, this becomes an interesting
686 	 * problem. we needto work out if we are freeing or allocation
687 	 * blocks first, then we can do the modification as necessary.
688 	 *
689 	 * We do this under the m_sb_lock so that if we are near
690 	 * ENOSPC, we will hold out any changes while we work out
691 	 * what to do. This means that the amount of free space can
692 	 * change while we do this, so we need to retry if we end up
693 	 * trying to reserve more space than is available.
694 	 */
695 retry:
696 	spin_lock(&mp->m_sb_lock);
697 
698 	/*
699 	 * If our previous reservation was larger than the current value,
700 	 * then move any unused blocks back to the free pool.
701 	 */
702 	fdblks_delta = 0;
703 	if (mp->m_resblks > request) {
704 		lcounter = mp->m_resblks_avail - request;
705 		if (lcounter  > 0) {		/* release unused blocks */
706 			fdblks_delta = lcounter;
707 			mp->m_resblks_avail -= lcounter;
708 		}
709 		mp->m_resblks = request;
710 	} else {
711 		__int64_t	free;
712 
713 		free = percpu_counter_sum(&mp->m_fdblocks) -
714 							XFS_ALLOC_SET_ASIDE(mp);
715 		if (!free)
716 			goto out; /* ENOSPC and fdblks_delta = 0 */
717 
718 		delta = request - mp->m_resblks;
719 		lcounter = free - delta;
720 		if (lcounter < 0) {
721 			/* We can't satisfy the request, just get what we can */
722 			mp->m_resblks += free;
723 			mp->m_resblks_avail += free;
724 			fdblks_delta = -free;
725 		} else {
726 			fdblks_delta = -delta;
727 			mp->m_resblks = request;
728 			mp->m_resblks_avail += delta;
729 		}
730 	}
731 out:
732 	if (outval) {
733 		outval->resblks = mp->m_resblks;
734 		outval->resblks_avail = mp->m_resblks_avail;
735 	}
736 	spin_unlock(&mp->m_sb_lock);
737 
738 	if (fdblks_delta) {
739 		/*
740 		 * If we are putting blocks back here, m_resblks_avail is
741 		 * already at its max so this will put it in the free pool.
742 		 *
743 		 * If we need space, we'll either succeed in getting it
744 		 * from the free block count or we'll get an enospc. If
745 		 * we get a ENOSPC, it means things changed while we were
746 		 * calculating fdblks_delta and so we should try again to
747 		 * see if there is anything left to reserve.
748 		 *
749 		 * Don't set the reserved flag here - we don't want to reserve
750 		 * the extra reserve blocks from the reserve.....
751 		 */
752 		int error;
753 		error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
754 		if (error == -ENOSPC)
755 			goto retry;
756 	}
757 	return 0;
758 }
759 
760 int
761 xfs_fs_goingdown(
762 	xfs_mount_t	*mp,
763 	__uint32_t	inflags)
764 {
765 	switch (inflags) {
766 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
767 		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
768 
769 		if (sb && !IS_ERR(sb)) {
770 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
771 			thaw_bdev(sb->s_bdev, sb);
772 		}
773 
774 		break;
775 	}
776 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
777 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
778 		break;
779 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
780 		xfs_force_shutdown(mp,
781 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
782 		break;
783 	default:
784 		return -EINVAL;
785 	}
786 
787 	return 0;
788 }
789 
790 /*
791  * Force a shutdown of the filesystem instantly while keeping the filesystem
792  * consistent. We don't do an unmount here; just shutdown the shop, make sure
793  * that absolutely nothing persistent happens to this filesystem after this
794  * point.
795  */
796 void
797 xfs_do_force_shutdown(
798 	xfs_mount_t	*mp,
799 	int		flags,
800 	char		*fname,
801 	int		lnnum)
802 {
803 	int		logerror;
804 
805 	logerror = flags & SHUTDOWN_LOG_IO_ERROR;
806 
807 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
808 		xfs_notice(mp,
809 	"%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
810 			__func__, flags, lnnum, fname, __return_address);
811 	}
812 	/*
813 	 * No need to duplicate efforts.
814 	 */
815 	if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
816 		return;
817 
818 	/*
819 	 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
820 	 * queue up anybody new on the log reservations, and wakes up
821 	 * everybody who's sleeping on log reservations to tell them
822 	 * the bad news.
823 	 */
824 	if (xfs_log_force_umount(mp, logerror))
825 		return;
826 
827 	if (flags & SHUTDOWN_CORRUPT_INCORE) {
828 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
829     "Corruption of in-memory data detected.  Shutting down filesystem");
830 		if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
831 			xfs_stack_trace();
832 	} else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
833 		if (logerror) {
834 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
835 		"Log I/O Error Detected.  Shutting down filesystem");
836 		} else if (flags & SHUTDOWN_DEVICE_REQ) {
837 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
838 		"All device paths lost.  Shutting down filesystem");
839 		} else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
840 			xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
841 		"I/O Error Detected. Shutting down filesystem");
842 		}
843 	}
844 	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
845 		xfs_alert(mp,
846 	"Please umount the filesystem and rectify the problem(s)");
847 	}
848 }
849