xref: /linux/fs/xfs/xfs_fsops.c (revision 765532c8aaac624b5f8687af6d319c6a1138a257)
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_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_inum.h"
23 #include "xfs_log.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_btree.h"
35 #include "xfs_error.h"
36 #include "xfs_alloc.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_rw.h"
43 #include "xfs_filestream.h"
44 #include "xfs_trace.h"
45 
46 /*
47  * File system operations
48  */
49 
50 int
51 xfs_fs_geometry(
52 	xfs_mount_t		*mp,
53 	xfs_fsop_geom_t		*geo,
54 	int			new_version)
55 {
56 	geo->blocksize = mp->m_sb.sb_blocksize;
57 	geo->rtextsize = mp->m_sb.sb_rextsize;
58 	geo->agblocks = mp->m_sb.sb_agblocks;
59 	geo->agcount = mp->m_sb.sb_agcount;
60 	geo->logblocks = mp->m_sb.sb_logblocks;
61 	geo->sectsize = mp->m_sb.sb_sectsize;
62 	geo->inodesize = mp->m_sb.sb_inodesize;
63 	geo->imaxpct = mp->m_sb.sb_imax_pct;
64 	geo->datablocks = mp->m_sb.sb_dblocks;
65 	geo->rtblocks = mp->m_sb.sb_rblocks;
66 	geo->rtextents = mp->m_sb.sb_rextents;
67 	geo->logstart = mp->m_sb.sb_logstart;
68 	ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
69 	memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
70 	if (new_version >= 2) {
71 		geo->sunit = mp->m_sb.sb_unit;
72 		geo->swidth = mp->m_sb.sb_width;
73 	}
74 	if (new_version >= 3) {
75 		geo->version = XFS_FSOP_GEOM_VERSION;
76 		geo->flags =
77 			(xfs_sb_version_hasattr(&mp->m_sb) ?
78 				XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
79 			(xfs_sb_version_hasnlink(&mp->m_sb) ?
80 				XFS_FSOP_GEOM_FLAGS_NLINK : 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_hasshared(&mp->m_sb) ?
88 				XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
89 			(xfs_sb_version_hasextflgbit(&mp->m_sb) ?
90 				XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
91 			(xfs_sb_version_hasdirv2(&mp->m_sb) ?
92 				XFS_FSOP_GEOM_FLAGS_DIRV2 : 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 		geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
102 				mp->m_sb.sb_logsectsize : BBSIZE;
103 		geo->rtsectsize = mp->m_sb.sb_blocksize;
104 		geo->dirblocksize = mp->m_dirblksize;
105 	}
106 	if (new_version >= 4) {
107 		geo->flags |=
108 			(xfs_sb_version_haslogv2(&mp->m_sb) ?
109 				XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
110 		geo->logsunit = mp->m_sb.sb_logsunit;
111 	}
112 	return 0;
113 }
114 
115 static int
116 xfs_growfs_data_private(
117 	xfs_mount_t		*mp,		/* mount point for filesystem */
118 	xfs_growfs_data_t	*in)		/* growfs data input struct */
119 {
120 	xfs_agf_t		*agf;
121 	xfs_agi_t		*agi;
122 	xfs_agnumber_t		agno;
123 	xfs_extlen_t		agsize;
124 	xfs_extlen_t		tmpsize;
125 	xfs_alloc_rec_t		*arec;
126 	struct xfs_btree_block	*block;
127 	xfs_buf_t		*bp;
128 	int			bucket;
129 	int			dpct;
130 	int			error;
131 	xfs_agnumber_t		nagcount;
132 	xfs_agnumber_t		nagimax = 0;
133 	xfs_rfsblock_t		nb, nb_mod;
134 	xfs_rfsblock_t		new;
135 	xfs_rfsblock_t		nfree;
136 	xfs_agnumber_t		oagcount;
137 	int			pct;
138 	xfs_trans_t		*tp;
139 
140 	nb = in->newblocks;
141 	pct = in->imaxpct;
142 	if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
143 		return XFS_ERROR(EINVAL);
144 	if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
145 		return error;
146 	dpct = pct - mp->m_sb.sb_imax_pct;
147 	bp = xfs_buf_read_uncached(mp, mp->m_ddev_targp,
148 				XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
149 				BBTOB(XFS_FSS_TO_BB(mp, 1)), 0);
150 	if (!bp)
151 		return EIO;
152 	xfs_buf_relse(bp);
153 
154 	new = nb;	/* use new as a temporary here */
155 	nb_mod = do_div(new, mp->m_sb.sb_agblocks);
156 	nagcount = new + (nb_mod != 0);
157 	if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
158 		nagcount--;
159 		nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
160 		if (nb < mp->m_sb.sb_dblocks)
161 			return XFS_ERROR(EINVAL);
162 	}
163 	new = nb - mp->m_sb.sb_dblocks;
164 	oagcount = mp->m_sb.sb_agcount;
165 
166 	/* allocate the new per-ag structures */
167 	if (nagcount > oagcount) {
168 		error = xfs_initialize_perag(mp, nagcount, &nagimax);
169 		if (error)
170 			return error;
171 	}
172 
173 	tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
174 	tp->t_flags |= XFS_TRANS_RESERVE;
175 	if ((error = xfs_trans_reserve(tp, XFS_GROWFS_SPACE_RES(mp),
176 			XFS_GROWDATA_LOG_RES(mp), 0, 0, 0))) {
177 		xfs_trans_cancel(tp, 0);
178 		return error;
179 	}
180 
181 	/*
182 	 * Write new AG headers to disk. Non-transactional, but written
183 	 * synchronously so they are completed prior to the growfs transaction
184 	 * being logged.
185 	 */
186 	nfree = 0;
187 	for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
188 		/*
189 		 * AG freelist header block
190 		 */
191 		bp = xfs_buf_get(mp->m_ddev_targp,
192 				 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
193 				 XFS_FSS_TO_BB(mp, 1), XBF_LOCK | XBF_MAPPED);
194 		agf = XFS_BUF_TO_AGF(bp);
195 		memset(agf, 0, mp->m_sb.sb_sectsize);
196 		agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
197 		agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
198 		agf->agf_seqno = cpu_to_be32(agno);
199 		if (agno == nagcount - 1)
200 			agsize =
201 				nb -
202 				(agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
203 		else
204 			agsize = mp->m_sb.sb_agblocks;
205 		agf->agf_length = cpu_to_be32(agsize);
206 		agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
207 		agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
208 		agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
209 		agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
210 		agf->agf_flfirst = 0;
211 		agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
212 		agf->agf_flcount = 0;
213 		tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
214 		agf->agf_freeblks = cpu_to_be32(tmpsize);
215 		agf->agf_longest = cpu_to_be32(tmpsize);
216 		error = xfs_bwrite(mp, bp);
217 		if (error) {
218 			goto error0;
219 		}
220 		/*
221 		 * AG inode header block
222 		 */
223 		bp = xfs_buf_get(mp->m_ddev_targp,
224 				 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
225 				 XFS_FSS_TO_BB(mp, 1), XBF_LOCK | XBF_MAPPED);
226 		agi = XFS_BUF_TO_AGI(bp);
227 		memset(agi, 0, mp->m_sb.sb_sectsize);
228 		agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
229 		agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
230 		agi->agi_seqno = cpu_to_be32(agno);
231 		agi->agi_length = cpu_to_be32(agsize);
232 		agi->agi_count = 0;
233 		agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
234 		agi->agi_level = cpu_to_be32(1);
235 		agi->agi_freecount = 0;
236 		agi->agi_newino = cpu_to_be32(NULLAGINO);
237 		agi->agi_dirino = cpu_to_be32(NULLAGINO);
238 		for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
239 			agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
240 		error = xfs_bwrite(mp, bp);
241 		if (error) {
242 			goto error0;
243 		}
244 		/*
245 		 * BNO btree root block
246 		 */
247 		bp = xfs_buf_get(mp->m_ddev_targp,
248 				 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
249 				 BTOBB(mp->m_sb.sb_blocksize),
250 				 XBF_LOCK | XBF_MAPPED);
251 		block = XFS_BUF_TO_BLOCK(bp);
252 		memset(block, 0, mp->m_sb.sb_blocksize);
253 		block->bb_magic = cpu_to_be32(XFS_ABTB_MAGIC);
254 		block->bb_level = 0;
255 		block->bb_numrecs = cpu_to_be16(1);
256 		block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
257 		block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
258 		arec = XFS_ALLOC_REC_ADDR(mp, block, 1);
259 		arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
260 		arec->ar_blockcount = cpu_to_be32(
261 			agsize - be32_to_cpu(arec->ar_startblock));
262 		error = xfs_bwrite(mp, bp);
263 		if (error) {
264 			goto error0;
265 		}
266 		/*
267 		 * CNT btree root block
268 		 */
269 		bp = xfs_buf_get(mp->m_ddev_targp,
270 				 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
271 				 BTOBB(mp->m_sb.sb_blocksize),
272 				 XBF_LOCK | XBF_MAPPED);
273 		block = XFS_BUF_TO_BLOCK(bp);
274 		memset(block, 0, mp->m_sb.sb_blocksize);
275 		block->bb_magic = cpu_to_be32(XFS_ABTC_MAGIC);
276 		block->bb_level = 0;
277 		block->bb_numrecs = cpu_to_be16(1);
278 		block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
279 		block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
280 		arec = XFS_ALLOC_REC_ADDR(mp, block, 1);
281 		arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
282 		arec->ar_blockcount = cpu_to_be32(
283 			agsize - be32_to_cpu(arec->ar_startblock));
284 		nfree += be32_to_cpu(arec->ar_blockcount);
285 		error = xfs_bwrite(mp, bp);
286 		if (error) {
287 			goto error0;
288 		}
289 		/*
290 		 * INO btree root block
291 		 */
292 		bp = xfs_buf_get(mp->m_ddev_targp,
293 				 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
294 				 BTOBB(mp->m_sb.sb_blocksize),
295 				 XBF_LOCK | XBF_MAPPED);
296 		block = XFS_BUF_TO_BLOCK(bp);
297 		memset(block, 0, mp->m_sb.sb_blocksize);
298 		block->bb_magic = cpu_to_be32(XFS_IBT_MAGIC);
299 		block->bb_level = 0;
300 		block->bb_numrecs = 0;
301 		block->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
302 		block->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
303 		error = xfs_bwrite(mp, bp);
304 		if (error) {
305 			goto error0;
306 		}
307 	}
308 	xfs_trans_agblocks_delta(tp, nfree);
309 	/*
310 	 * There are new blocks in the old last a.g.
311 	 */
312 	if (new) {
313 		/*
314 		 * Change the agi length.
315 		 */
316 		error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
317 		if (error) {
318 			goto error0;
319 		}
320 		ASSERT(bp);
321 		agi = XFS_BUF_TO_AGI(bp);
322 		be32_add_cpu(&agi->agi_length, new);
323 		ASSERT(nagcount == oagcount ||
324 		       be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
325 		xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
326 		/*
327 		 * Change agf length.
328 		 */
329 		error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
330 		if (error) {
331 			goto error0;
332 		}
333 		ASSERT(bp);
334 		agf = XFS_BUF_TO_AGF(bp);
335 		be32_add_cpu(&agf->agf_length, new);
336 		ASSERT(be32_to_cpu(agf->agf_length) ==
337 		       be32_to_cpu(agi->agi_length));
338 
339 		xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
340 		/*
341 		 * Free the new space.
342 		 */
343 		error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
344 			be32_to_cpu(agf->agf_length) - new), new);
345 		if (error) {
346 			goto error0;
347 		}
348 	}
349 
350 	/*
351 	 * Update changed superblock fields transactionally. These are not
352 	 * seen by the rest of the world until the transaction commit applies
353 	 * them atomically to the superblock.
354 	 */
355 	if (nagcount > oagcount)
356 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
357 	if (nb > mp->m_sb.sb_dblocks)
358 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
359 				 nb - mp->m_sb.sb_dblocks);
360 	if (nfree)
361 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
362 	if (dpct)
363 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
364 	error = xfs_trans_commit(tp, 0);
365 	if (error)
366 		return error;
367 
368 	/* New allocation groups fully initialized, so update mount struct */
369 	if (nagimax)
370 		mp->m_maxagi = nagimax;
371 	if (mp->m_sb.sb_imax_pct) {
372 		__uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
373 		do_div(icount, 100);
374 		mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
375 	} else
376 		mp->m_maxicount = 0;
377 
378 	/* update secondary superblocks. */
379 	for (agno = 1; agno < nagcount; agno++) {
380 		error = xfs_read_buf(mp, mp->m_ddev_targp,
381 				  XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
382 				  XFS_FSS_TO_BB(mp, 1), 0, &bp);
383 		if (error) {
384 			xfs_fs_cmn_err(CE_WARN, mp,
385 			"error %d reading secondary superblock for ag %d",
386 				error, agno);
387 			break;
388 		}
389 		xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
390 		/*
391 		 * If we get an error writing out the alternate superblocks,
392 		 * just issue a warning and continue.  The real work is
393 		 * already done and committed.
394 		 */
395 		if (!(error = xfs_bwrite(mp, bp))) {
396 			continue;
397 		} else {
398 			xfs_fs_cmn_err(CE_WARN, mp,
399 		"write error %d updating secondary superblock for ag %d",
400 				error, agno);
401 			break; /* no point in continuing */
402 		}
403 	}
404 	return 0;
405 
406  error0:
407 	xfs_trans_cancel(tp, XFS_TRANS_ABORT);
408 	return error;
409 }
410 
411 static int
412 xfs_growfs_log_private(
413 	xfs_mount_t		*mp,	/* mount point for filesystem */
414 	xfs_growfs_log_t	*in)	/* growfs log input struct */
415 {
416 	xfs_extlen_t		nb;
417 
418 	nb = in->newblocks;
419 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
420 		return XFS_ERROR(EINVAL);
421 	if (nb == mp->m_sb.sb_logblocks &&
422 	    in->isint == (mp->m_sb.sb_logstart != 0))
423 		return XFS_ERROR(EINVAL);
424 	/*
425 	 * Moving the log is hard, need new interfaces to sync
426 	 * the log first, hold off all activity while moving it.
427 	 * Can have shorter or longer log in the same space,
428 	 * or transform internal to external log or vice versa.
429 	 */
430 	return XFS_ERROR(ENOSYS);
431 }
432 
433 /*
434  * protected versions of growfs function acquire and release locks on the mount
435  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
436  * XFS_IOC_FSGROWFSRT
437  */
438 
439 
440 int
441 xfs_growfs_data(
442 	xfs_mount_t		*mp,
443 	xfs_growfs_data_t	*in)
444 {
445 	int error;
446 
447 	if (!capable(CAP_SYS_ADMIN))
448 		return XFS_ERROR(EPERM);
449 	if (!mutex_trylock(&mp->m_growlock))
450 		return XFS_ERROR(EWOULDBLOCK);
451 	error = xfs_growfs_data_private(mp, in);
452 	mutex_unlock(&mp->m_growlock);
453 	return error;
454 }
455 
456 int
457 xfs_growfs_log(
458 	xfs_mount_t		*mp,
459 	xfs_growfs_log_t	*in)
460 {
461 	int error;
462 
463 	if (!capable(CAP_SYS_ADMIN))
464 		return XFS_ERROR(EPERM);
465 	if (!mutex_trylock(&mp->m_growlock))
466 		return XFS_ERROR(EWOULDBLOCK);
467 	error = xfs_growfs_log_private(mp, in);
468 	mutex_unlock(&mp->m_growlock);
469 	return error;
470 }
471 
472 /*
473  * exported through ioctl XFS_IOC_FSCOUNTS
474  */
475 
476 int
477 xfs_fs_counts(
478 	xfs_mount_t		*mp,
479 	xfs_fsop_counts_t	*cnt)
480 {
481 	xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
482 	spin_lock(&mp->m_sb_lock);
483 	cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
484 	cnt->freertx = mp->m_sb.sb_frextents;
485 	cnt->freeino = mp->m_sb.sb_ifree;
486 	cnt->allocino = mp->m_sb.sb_icount;
487 	spin_unlock(&mp->m_sb_lock);
488 	return 0;
489 }
490 
491 /*
492  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
493  *
494  * xfs_reserve_blocks is called to set m_resblks
495  * in the in-core mount table. The number of unused reserved blocks
496  * is kept in m_resblks_avail.
497  *
498  * Reserve the requested number of blocks if available. Otherwise return
499  * as many as possible to satisfy the request. The actual number
500  * reserved are returned in outval
501  *
502  * A null inval pointer indicates that only the current reserved blocks
503  * available  should  be returned no settings are changed.
504  */
505 
506 int
507 xfs_reserve_blocks(
508 	xfs_mount_t             *mp,
509 	__uint64_t              *inval,
510 	xfs_fsop_resblks_t      *outval)
511 {
512 	__int64_t		lcounter, delta, fdblks_delta;
513 	__uint64_t		request;
514 
515 	/* If inval is null, report current values and return */
516 	if (inval == (__uint64_t *)NULL) {
517 		if (!outval)
518 			return EINVAL;
519 		outval->resblks = mp->m_resblks;
520 		outval->resblks_avail = mp->m_resblks_avail;
521 		return 0;
522 	}
523 
524 	request = *inval;
525 
526 	/*
527 	 * With per-cpu counters, this becomes an interesting
528 	 * problem. we needto work out if we are freeing or allocation
529 	 * blocks first, then we can do the modification as necessary.
530 	 *
531 	 * We do this under the m_sb_lock so that if we are near
532 	 * ENOSPC, we will hold out any changes while we work out
533 	 * what to do. This means that the amount of free space can
534 	 * change while we do this, so we need to retry if we end up
535 	 * trying to reserve more space than is available.
536 	 *
537 	 * We also use the xfs_mod_incore_sb() interface so that we
538 	 * don't have to care about whether per cpu counter are
539 	 * enabled, disabled or even compiled in....
540 	 */
541 retry:
542 	spin_lock(&mp->m_sb_lock);
543 	xfs_icsb_sync_counters_locked(mp, 0);
544 
545 	/*
546 	 * If our previous reservation was larger than the current value,
547 	 * then move any unused blocks back to the free pool.
548 	 */
549 	fdblks_delta = 0;
550 	if (mp->m_resblks > request) {
551 		lcounter = mp->m_resblks_avail - request;
552 		if (lcounter  > 0) {		/* release unused blocks */
553 			fdblks_delta = lcounter;
554 			mp->m_resblks_avail -= lcounter;
555 		}
556 		mp->m_resblks = request;
557 	} else {
558 		__int64_t	free;
559 
560 		free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
561 		if (!free)
562 			goto out; /* ENOSPC and fdblks_delta = 0 */
563 
564 		delta = request - mp->m_resblks;
565 		lcounter = free - delta;
566 		if (lcounter < 0) {
567 			/* We can't satisfy the request, just get what we can */
568 			mp->m_resblks += free;
569 			mp->m_resblks_avail += free;
570 			fdblks_delta = -free;
571 		} else {
572 			fdblks_delta = -delta;
573 			mp->m_resblks = request;
574 			mp->m_resblks_avail += delta;
575 		}
576 	}
577 out:
578 	if (outval) {
579 		outval->resblks = mp->m_resblks;
580 		outval->resblks_avail = mp->m_resblks_avail;
581 	}
582 	spin_unlock(&mp->m_sb_lock);
583 
584 	if (fdblks_delta) {
585 		/*
586 		 * If we are putting blocks back here, m_resblks_avail is
587 		 * already at its max so this will put it in the free pool.
588 		 *
589 		 * If we need space, we'll either succeed in getting it
590 		 * from the free block count or we'll get an enospc. If
591 		 * we get a ENOSPC, it means things changed while we were
592 		 * calculating fdblks_delta and so we should try again to
593 		 * see if there is anything left to reserve.
594 		 *
595 		 * Don't set the reserved flag here - we don't want to reserve
596 		 * the extra reserve blocks from the reserve.....
597 		 */
598 		int error;
599 		error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
600 						 fdblks_delta, 0);
601 		if (error == ENOSPC)
602 			goto retry;
603 	}
604 	return 0;
605 }
606 
607 /*
608  * Dump a transaction into the log that contains no real change. This is needed
609  * to be able to make the log dirty or stamp the current tail LSN into the log
610  * during the covering operation.
611  *
612  * We cannot use an inode here for this - that will push dirty state back up
613  * into the VFS and then periodic inode flushing will prevent log covering from
614  * making progress. Hence we log a field in the superblock instead.
615  */
616 int
617 xfs_fs_log_dummy(
618 	xfs_mount_t	*mp,
619 	int		flags)
620 {
621 	xfs_trans_t	*tp;
622 	int		error;
623 
624 	tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
625 	error = xfs_trans_reserve(tp, 0, mp->m_sb.sb_sectsize + 128, 0, 0,
626 					XFS_DEFAULT_LOG_COUNT);
627 	if (error) {
628 		xfs_trans_cancel(tp, 0);
629 		return error;
630 	}
631 
632 	/* log the UUID because it is an unchanging field */
633 	xfs_mod_sb(tp, XFS_SB_UUID);
634 	if (flags & SYNC_WAIT)
635 		xfs_trans_set_sync(tp);
636 	return xfs_trans_commit(tp, 0);
637 }
638 
639 int
640 xfs_fs_goingdown(
641 	xfs_mount_t	*mp,
642 	__uint32_t	inflags)
643 {
644 	switch (inflags) {
645 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
646 		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
647 
648 		if (sb && !IS_ERR(sb)) {
649 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
650 			thaw_bdev(sb->s_bdev, sb);
651 		}
652 
653 		break;
654 	}
655 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
656 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
657 		break;
658 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
659 		xfs_force_shutdown(mp,
660 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
661 		break;
662 	default:
663 		return XFS_ERROR(EINVAL);
664 	}
665 
666 	return 0;
667 }
668