xref: /openbmc/linux/fs/xfs/xfs_fsops.c (revision 5b4cb650)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4  * All Rights Reserved.
5  */
6 #include "xfs.h"
7 #include "xfs_fs.h"
8 #include "xfs_shared.h"
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_sb.h"
13 #include "xfs_mount.h"
14 #include "xfs_defer.h"
15 #include "xfs_trans.h"
16 #include "xfs_error.h"
17 #include "xfs_btree.h"
18 #include "xfs_alloc.h"
19 #include "xfs_fsops.h"
20 #include "xfs_trans_space.h"
21 #include "xfs_rtalloc.h"
22 #include "xfs_trace.h"
23 #include "xfs_log.h"
24 #include "xfs_ag.h"
25 #include "xfs_ag_resv.h"
26 
27 /*
28  * growfs operations
29  */
30 static int
31 xfs_growfs_data_private(
32 	xfs_mount_t		*mp,		/* mount point for filesystem */
33 	xfs_growfs_data_t	*in)		/* growfs data input struct */
34 {
35 	xfs_buf_t		*bp;
36 	int			error;
37 	xfs_agnumber_t		nagcount;
38 	xfs_agnumber_t		nagimax = 0;
39 	xfs_rfsblock_t		nb, nb_mod;
40 	xfs_rfsblock_t		new;
41 	xfs_agnumber_t		oagcount;
42 	xfs_trans_t		*tp;
43 	LIST_HEAD		(buffer_list);
44 	struct aghdr_init_data	id = {};
45 
46 	nb = in->newblocks;
47 	if (nb < mp->m_sb.sb_dblocks)
48 		return -EINVAL;
49 	if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
50 		return error;
51 	error = xfs_buf_read_uncached(mp->m_ddev_targp,
52 				XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
53 				XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
54 	if (error)
55 		return error;
56 	xfs_buf_relse(bp);
57 
58 	new = nb;	/* use new as a temporary here */
59 	nb_mod = do_div(new, mp->m_sb.sb_agblocks);
60 	nagcount = new + (nb_mod != 0);
61 	if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
62 		nagcount--;
63 		nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
64 		if (nb < mp->m_sb.sb_dblocks)
65 			return -EINVAL;
66 	}
67 	new = nb - mp->m_sb.sb_dblocks;
68 	oagcount = mp->m_sb.sb_agcount;
69 
70 	/* allocate the new per-ag structures */
71 	if (nagcount > oagcount) {
72 		error = xfs_initialize_perag(mp, nagcount, &nagimax);
73 		if (error)
74 			return error;
75 	}
76 
77 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
78 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
79 	if (error)
80 		return error;
81 
82 	/*
83 	 * Write new AG headers to disk. Non-transactional, but need to be
84 	 * written and completed prior to the growfs transaction being logged.
85 	 * To do this, we use a delayed write buffer list and wait for
86 	 * submission and IO completion of the list as a whole. This allows the
87 	 * IO subsystem to merge all the AG headers in a single AG into a single
88 	 * IO and hide most of the latency of the IO from us.
89 	 *
90 	 * This also means that if we get an error whilst building the buffer
91 	 * list to write, we can cancel the entire list without having written
92 	 * anything.
93 	 */
94 	INIT_LIST_HEAD(&id.buffer_list);
95 	for (id.agno = nagcount - 1;
96 	     id.agno >= oagcount;
97 	     id.agno--, new -= id.agsize) {
98 
99 		if (id.agno == nagcount - 1)
100 			id.agsize = nb -
101 				(id.agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
102 		else
103 			id.agsize = mp->m_sb.sb_agblocks;
104 
105 		error = xfs_ag_init_headers(mp, &id);
106 		if (error) {
107 			xfs_buf_delwri_cancel(&id.buffer_list);
108 			goto out_trans_cancel;
109 		}
110 	}
111 	error = xfs_buf_delwri_submit(&id.buffer_list);
112 	if (error)
113 		goto out_trans_cancel;
114 
115 	xfs_trans_agblocks_delta(tp, id.nfree);
116 
117 	/* If there are new blocks in the old last AG, extend it. */
118 	if (new) {
119 		error = xfs_ag_extend_space(mp, tp, &id, new);
120 		if (error)
121 			goto out_trans_cancel;
122 	}
123 
124 	/*
125 	 * Update changed superblock fields transactionally. These are not
126 	 * seen by the rest of the world until the transaction commit applies
127 	 * them atomically to the superblock.
128 	 */
129 	if (nagcount > oagcount)
130 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
131 	if (nb > mp->m_sb.sb_dblocks)
132 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
133 				 nb - mp->m_sb.sb_dblocks);
134 	if (id.nfree)
135 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
136 	xfs_trans_set_sync(tp);
137 	error = xfs_trans_commit(tp);
138 	if (error)
139 		return error;
140 
141 	/* New allocation groups fully initialized, so update mount struct */
142 	if (nagimax)
143 		mp->m_maxagi = nagimax;
144 	xfs_set_low_space_thresholds(mp);
145 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
146 
147 	/*
148 	 * If we expanded the last AG, free the per-AG reservation
149 	 * so we can reinitialize it with the new size.
150 	 */
151 	if (new) {
152 		struct xfs_perag	*pag;
153 
154 		pag = xfs_perag_get(mp, id.agno);
155 		error = xfs_ag_resv_free(pag);
156 		xfs_perag_put(pag);
157 		if (error)
158 			return error;
159 	}
160 
161 	/*
162 	 * Reserve AG metadata blocks. ENOSPC here does not mean there was a
163 	 * growfs failure, just that there still isn't space for new user data
164 	 * after the grow has been run.
165 	 */
166 	error = xfs_fs_reserve_ag_blocks(mp);
167 	if (error == -ENOSPC)
168 		error = 0;
169 	return error;
170 
171 out_trans_cancel:
172 	xfs_trans_cancel(tp);
173 	return error;
174 }
175 
176 static int
177 xfs_growfs_log_private(
178 	xfs_mount_t		*mp,	/* mount point for filesystem */
179 	xfs_growfs_log_t	*in)	/* growfs log input struct */
180 {
181 	xfs_extlen_t		nb;
182 
183 	nb = in->newblocks;
184 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
185 		return -EINVAL;
186 	if (nb == mp->m_sb.sb_logblocks &&
187 	    in->isint == (mp->m_sb.sb_logstart != 0))
188 		return -EINVAL;
189 	/*
190 	 * Moving the log is hard, need new interfaces to sync
191 	 * the log first, hold off all activity while moving it.
192 	 * Can have shorter or longer log in the same space,
193 	 * or transform internal to external log or vice versa.
194 	 */
195 	return -ENOSYS;
196 }
197 
198 static int
199 xfs_growfs_imaxpct(
200 	struct xfs_mount	*mp,
201 	__u32			imaxpct)
202 {
203 	struct xfs_trans	*tp;
204 	int			dpct;
205 	int			error;
206 
207 	if (imaxpct > 100)
208 		return -EINVAL;
209 
210 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
211 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
212 	if (error)
213 		return error;
214 
215 	dpct = imaxpct - mp->m_sb.sb_imax_pct;
216 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
217 	xfs_trans_set_sync(tp);
218 	return xfs_trans_commit(tp);
219 }
220 
221 /*
222  * protected versions of growfs function acquire and release locks on the mount
223  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
224  * XFS_IOC_FSGROWFSRT
225  */
226 int
227 xfs_growfs_data(
228 	struct xfs_mount	*mp,
229 	struct xfs_growfs_data	*in)
230 {
231 	int			error = 0;
232 
233 	if (!capable(CAP_SYS_ADMIN))
234 		return -EPERM;
235 	if (!mutex_trylock(&mp->m_growlock))
236 		return -EWOULDBLOCK;
237 
238 	/* update imaxpct separately to the physical grow of the filesystem */
239 	if (in->imaxpct != mp->m_sb.sb_imax_pct) {
240 		error = xfs_growfs_imaxpct(mp, in->imaxpct);
241 		if (error)
242 			goto out_error;
243 	}
244 
245 	if (in->newblocks != mp->m_sb.sb_dblocks) {
246 		error = xfs_growfs_data_private(mp, in);
247 		if (error)
248 			goto out_error;
249 	}
250 
251 	/* Post growfs calculations needed to reflect new state in operations */
252 	if (mp->m_sb.sb_imax_pct) {
253 		uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
254 		do_div(icount, 100);
255 		mp->m_maxicount = XFS_FSB_TO_INO(mp, icount);
256 	} else
257 		mp->m_maxicount = 0;
258 
259 	/* Update secondary superblocks now the physical grow has completed */
260 	error = xfs_update_secondary_sbs(mp);
261 
262 out_error:
263 	/*
264 	 * Increment the generation unconditionally, the error could be from
265 	 * updating the secondary superblocks, in which case the new size
266 	 * is live already.
267 	 */
268 	mp->m_generation++;
269 	mutex_unlock(&mp->m_growlock);
270 	return error;
271 }
272 
273 int
274 xfs_growfs_log(
275 	xfs_mount_t		*mp,
276 	xfs_growfs_log_t	*in)
277 {
278 	int error;
279 
280 	if (!capable(CAP_SYS_ADMIN))
281 		return -EPERM;
282 	if (!mutex_trylock(&mp->m_growlock))
283 		return -EWOULDBLOCK;
284 	error = xfs_growfs_log_private(mp, in);
285 	mutex_unlock(&mp->m_growlock);
286 	return error;
287 }
288 
289 /*
290  * exported through ioctl XFS_IOC_FSCOUNTS
291  */
292 
293 int
294 xfs_fs_counts(
295 	xfs_mount_t		*mp,
296 	xfs_fsop_counts_t	*cnt)
297 {
298 	cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
299 	cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
300 	cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
301 						mp->m_alloc_set_aside;
302 
303 	spin_lock(&mp->m_sb_lock);
304 	cnt->freertx = mp->m_sb.sb_frextents;
305 	spin_unlock(&mp->m_sb_lock);
306 	return 0;
307 }
308 
309 /*
310  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
311  *
312  * xfs_reserve_blocks is called to set m_resblks
313  * in the in-core mount table. The number of unused reserved blocks
314  * is kept in m_resblks_avail.
315  *
316  * Reserve the requested number of blocks if available. Otherwise return
317  * as many as possible to satisfy the request. The actual number
318  * reserved are returned in outval
319  *
320  * A null inval pointer indicates that only the current reserved blocks
321  * available  should  be returned no settings are changed.
322  */
323 
324 int
325 xfs_reserve_blocks(
326 	xfs_mount_t             *mp,
327 	uint64_t              *inval,
328 	xfs_fsop_resblks_t      *outval)
329 {
330 	int64_t			lcounter, delta;
331 	int64_t			fdblks_delta = 0;
332 	uint64_t		request;
333 	int64_t			free;
334 	int			error = 0;
335 
336 	/* If inval is null, report current values and return */
337 	if (inval == (uint64_t *)NULL) {
338 		if (!outval)
339 			return -EINVAL;
340 		outval->resblks = mp->m_resblks;
341 		outval->resblks_avail = mp->m_resblks_avail;
342 		return 0;
343 	}
344 
345 	request = *inval;
346 
347 	/*
348 	 * With per-cpu counters, this becomes an interesting problem. we need
349 	 * to work out if we are freeing or allocation blocks first, then we can
350 	 * do the modification as necessary.
351 	 *
352 	 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
353 	 * hold out any changes while we work out what to do. This means that
354 	 * the amount of free space can change while we do this, so we need to
355 	 * retry if we end up trying to reserve more space than is available.
356 	 */
357 	spin_lock(&mp->m_sb_lock);
358 
359 	/*
360 	 * If our previous reservation was larger than the current value,
361 	 * then move any unused blocks back to the free pool. Modify the resblks
362 	 * counters directly since we shouldn't have any problems unreserving
363 	 * space.
364 	 */
365 	if (mp->m_resblks > request) {
366 		lcounter = mp->m_resblks_avail - request;
367 		if (lcounter  > 0) {		/* release unused blocks */
368 			fdblks_delta = lcounter;
369 			mp->m_resblks_avail -= lcounter;
370 		}
371 		mp->m_resblks = request;
372 		if (fdblks_delta) {
373 			spin_unlock(&mp->m_sb_lock);
374 			error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
375 			spin_lock(&mp->m_sb_lock);
376 		}
377 
378 		goto out;
379 	}
380 
381 	/*
382 	 * If the request is larger than the current reservation, reserve the
383 	 * blocks before we update the reserve counters. Sample m_fdblocks and
384 	 * perform a partial reservation if the request exceeds free space.
385 	 */
386 	error = -ENOSPC;
387 	do {
388 		free = percpu_counter_sum(&mp->m_fdblocks) -
389 						mp->m_alloc_set_aside;
390 		if (free <= 0)
391 			break;
392 
393 		delta = request - mp->m_resblks;
394 		lcounter = free - delta;
395 		if (lcounter < 0)
396 			/* We can't satisfy the request, just get what we can */
397 			fdblks_delta = free;
398 		else
399 			fdblks_delta = delta;
400 
401 		/*
402 		 * We'll either succeed in getting space from the free block
403 		 * count or we'll get an ENOSPC. If we get a ENOSPC, it means
404 		 * things changed while we were calculating fdblks_delta and so
405 		 * we should try again to see if there is anything left to
406 		 * reserve.
407 		 *
408 		 * Don't set the reserved flag here - we don't want to reserve
409 		 * the extra reserve blocks from the reserve.....
410 		 */
411 		spin_unlock(&mp->m_sb_lock);
412 		error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
413 		spin_lock(&mp->m_sb_lock);
414 	} while (error == -ENOSPC);
415 
416 	/*
417 	 * Update the reserve counters if blocks have been successfully
418 	 * allocated.
419 	 */
420 	if (!error && fdblks_delta) {
421 		mp->m_resblks += fdblks_delta;
422 		mp->m_resblks_avail += fdblks_delta;
423 	}
424 
425 out:
426 	if (outval) {
427 		outval->resblks = mp->m_resblks;
428 		outval->resblks_avail = mp->m_resblks_avail;
429 	}
430 
431 	spin_unlock(&mp->m_sb_lock);
432 	return error;
433 }
434 
435 int
436 xfs_fs_goingdown(
437 	xfs_mount_t	*mp,
438 	uint32_t	inflags)
439 {
440 	switch (inflags) {
441 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
442 		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
443 
444 		if (sb && !IS_ERR(sb)) {
445 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
446 			thaw_bdev(sb->s_bdev, sb);
447 		}
448 
449 		break;
450 	}
451 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
452 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
453 		break;
454 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
455 		xfs_force_shutdown(mp,
456 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
457 		break;
458 	default:
459 		return -EINVAL;
460 	}
461 
462 	return 0;
463 }
464 
465 /*
466  * Force a shutdown of the filesystem instantly while keeping the filesystem
467  * consistent. We don't do an unmount here; just shutdown the shop, make sure
468  * that absolutely nothing persistent happens to this filesystem after this
469  * point.
470  */
471 void
472 xfs_do_force_shutdown(
473 	struct xfs_mount *mp,
474 	int		flags,
475 	char		*fname,
476 	int		lnnum)
477 {
478 	bool		logerror = flags & SHUTDOWN_LOG_IO_ERROR;
479 
480 	/*
481 	 * No need to duplicate efforts.
482 	 */
483 	if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
484 		return;
485 
486 	/*
487 	 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
488 	 * queue up anybody new on the log reservations, and wakes up
489 	 * everybody who's sleeping on log reservations to tell them
490 	 * the bad news.
491 	 */
492 	if (xfs_log_force_umount(mp, logerror))
493 		return;
494 
495 	if (flags & SHUTDOWN_FORCE_UMOUNT) {
496 		xfs_alert(mp,
497 "User initiated shutdown received. Shutting down filesystem");
498 		return;
499 	}
500 
501 	xfs_notice(mp,
502 "%s(0x%x) called from line %d of file %s. Return address = "PTR_FMT,
503 		__func__, flags, lnnum, fname, __return_address);
504 
505 	if (flags & SHUTDOWN_CORRUPT_INCORE) {
506 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
507 "Corruption of in-memory data detected.  Shutting down filesystem");
508 		if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
509 			xfs_stack_trace();
510 	} else if (logerror) {
511 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
512 			"Log I/O Error Detected. Shutting down filesystem");
513 	} else if (flags & SHUTDOWN_DEVICE_REQ) {
514 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
515 			"All device paths lost. Shutting down filesystem");
516 	} else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
517 		xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
518 			"I/O Error Detected. Shutting down filesystem");
519 	}
520 
521 	xfs_alert(mp,
522 		"Please unmount the filesystem and rectify the problem(s)");
523 }
524 
525 /*
526  * Reserve free space for per-AG metadata.
527  */
528 int
529 xfs_fs_reserve_ag_blocks(
530 	struct xfs_mount	*mp)
531 {
532 	xfs_agnumber_t		agno;
533 	struct xfs_perag	*pag;
534 	int			error = 0;
535 	int			err2;
536 
537 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
538 		pag = xfs_perag_get(mp, agno);
539 		err2 = xfs_ag_resv_init(pag, NULL);
540 		xfs_perag_put(pag);
541 		if (err2 && !error)
542 			error = err2;
543 	}
544 
545 	if (error && error != -ENOSPC) {
546 		xfs_warn(mp,
547 	"Error %d reserving per-AG metadata reserve pool.", error);
548 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
549 	}
550 
551 	return error;
552 }
553 
554 /*
555  * Free space reserved for per-AG metadata.
556  */
557 int
558 xfs_fs_unreserve_ag_blocks(
559 	struct xfs_mount	*mp)
560 {
561 	xfs_agnumber_t		agno;
562 	struct xfs_perag	*pag;
563 	int			error = 0;
564 	int			err2;
565 
566 	for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
567 		pag = xfs_perag_get(mp, agno);
568 		err2 = xfs_ag_resv_free(pag);
569 		xfs_perag_put(pag);
570 		if (err2 && !error)
571 			error = err2;
572 	}
573 
574 	if (error)
575 		xfs_warn(mp,
576 	"Error %d freeing per-AG metadata reserve pool.", error);
577 
578 	return error;
579 }
580