xref: /linux/fs/xfs/xfs_fsops.c (revision d99ff463ecf651437e9e4abe68f331dfb6b5bd9d)
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_trans.h"
15 #include "xfs_error.h"
16 #include "xfs_alloc.h"
17 #include "xfs_fsops.h"
18 #include "xfs_trans_space.h"
19 #include "xfs_log.h"
20 #include "xfs_log_priv.h"
21 #include "xfs_ag.h"
22 #include "xfs_ag_resv.h"
23 #include "xfs_trace.h"
24 
25 /*
26  * Write new AG headers to disk. Non-transactional, but need to be
27  * written and completed prior to the growfs transaction being logged.
28  * To do this, we use a delayed write buffer list and wait for
29  * submission and IO completion of the list as a whole. This allows the
30  * IO subsystem to merge all the AG headers in a single AG into a single
31  * IO and hide most of the latency of the IO from us.
32  *
33  * This also means that if we get an error whilst building the buffer
34  * list to write, we can cancel the entire list without having written
35  * anything.
36  */
37 static int
38 xfs_resizefs_init_new_ags(
39 	struct xfs_trans	*tp,
40 	struct aghdr_init_data	*id,
41 	xfs_agnumber_t		oagcount,
42 	xfs_agnumber_t		nagcount,
43 	xfs_rfsblock_t		delta,
44 	struct xfs_perag	*last_pag,
45 	bool			*lastag_extended)
46 {
47 	struct xfs_mount	*mp = tp->t_mountp;
48 	xfs_rfsblock_t		nb = mp->m_sb.sb_dblocks + delta;
49 	int			error;
50 
51 	*lastag_extended = false;
52 
53 	INIT_LIST_HEAD(&id->buffer_list);
54 	for (id->agno = nagcount - 1;
55 	     id->agno >= oagcount;
56 	     id->agno--, delta -= id->agsize) {
57 
58 		if (id->agno == nagcount - 1)
59 			id->agsize = nb - (id->agno *
60 					(xfs_rfsblock_t)mp->m_sb.sb_agblocks);
61 		else
62 			id->agsize = mp->m_sb.sb_agblocks;
63 
64 		error = xfs_ag_init_headers(mp, id);
65 		if (error) {
66 			xfs_buf_delwri_cancel(&id->buffer_list);
67 			return error;
68 		}
69 	}
70 
71 	error = xfs_buf_delwri_submit(&id->buffer_list);
72 	if (error)
73 		return error;
74 
75 	if (delta) {
76 		*lastag_extended = true;
77 		error = xfs_ag_extend_space(last_pag, tp, delta);
78 	}
79 	return error;
80 }
81 
82 /*
83  * growfs operations
84  */
85 static int
86 xfs_growfs_data_private(
87 	struct xfs_mount	*mp,		/* mount point for filesystem */
88 	struct xfs_growfs_data	*in)		/* growfs data input struct */
89 {
90 	struct xfs_buf		*bp;
91 	int			error;
92 	xfs_agnumber_t		nagcount;
93 	xfs_agnumber_t		nagimax = 0;
94 	xfs_rfsblock_t		nb, nb_div, nb_mod;
95 	int64_t			delta;
96 	bool			lastag_extended = false;
97 	xfs_agnumber_t		oagcount;
98 	struct xfs_trans	*tp;
99 	struct aghdr_init_data	id = {};
100 	struct xfs_perag	*last_pag;
101 
102 	nb = in->newblocks;
103 	error = xfs_sb_validate_fsb_count(&mp->m_sb, nb);
104 	if (error)
105 		return error;
106 
107 	if (nb > mp->m_sb.sb_dblocks) {
108 		error = xfs_buf_read_uncached(mp->m_ddev_targp,
109 				XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
110 				XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
111 		if (error)
112 			return error;
113 		xfs_buf_relse(bp);
114 	}
115 
116 	nb_div = nb;
117 	nb_mod = do_div(nb_div, mp->m_sb.sb_agblocks);
118 	if (nb_mod && nb_mod >= XFS_MIN_AG_BLOCKS)
119 		nb_div++;
120 	else if (nb_mod)
121 		nb = nb_div * mp->m_sb.sb_agblocks;
122 
123 	if (nb_div > XFS_MAX_AGNUMBER + 1) {
124 		nb_div = XFS_MAX_AGNUMBER + 1;
125 		nb = nb_div * mp->m_sb.sb_agblocks;
126 	}
127 	nagcount = nb_div;
128 	delta = nb - mp->m_sb.sb_dblocks;
129 	/*
130 	 * Reject filesystems with a single AG because they are not
131 	 * supported, and reject a shrink operation that would cause a
132 	 * filesystem to become unsupported.
133 	 */
134 	if (delta < 0 && nagcount < 2)
135 		return -EINVAL;
136 
137 	oagcount = mp->m_sb.sb_agcount;
138 	/* allocate the new per-ag structures */
139 	if (nagcount > oagcount) {
140 		error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
141 		if (error)
142 			return error;
143 	} else if (nagcount < oagcount) {
144 		/* TODO: shrinking the entire AGs hasn't yet completed */
145 		return -EINVAL;
146 	}
147 
148 	if (delta > 0)
149 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
150 				XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
151 				&tp);
152 	else
153 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0,
154 				0, &tp);
155 	if (error)
156 		return error;
157 
158 	last_pag = xfs_perag_get(mp, oagcount - 1);
159 	if (delta > 0) {
160 		error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
161 				delta, last_pag, &lastag_extended);
162 	} else {
163 		xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
164 	"EXPERIMENTAL online shrink feature in use. Use at your own risk!");
165 
166 		error = xfs_ag_shrink_space(last_pag, &tp, -delta);
167 	}
168 	xfs_perag_put(last_pag);
169 	if (error)
170 		goto out_trans_cancel;
171 
172 	/*
173 	 * Update changed superblock fields transactionally. These are not
174 	 * seen by the rest of the world until the transaction commit applies
175 	 * them atomically to the superblock.
176 	 */
177 	if (nagcount > oagcount)
178 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
179 	if (delta)
180 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
181 	if (id.nfree)
182 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
183 
184 	/*
185 	 * Sync sb counters now to reflect the updated values. This is
186 	 * particularly important for shrink because the write verifier
187 	 * will fail if sb_fdblocks is ever larger than sb_dblocks.
188 	 */
189 	if (xfs_has_lazysbcount(mp))
190 		xfs_log_sb(tp);
191 
192 	xfs_trans_set_sync(tp);
193 	error = xfs_trans_commit(tp);
194 	if (error)
195 		return error;
196 
197 	/* New allocation groups fully initialized, so update mount struct */
198 	if (nagimax)
199 		mp->m_maxagi = nagimax;
200 	xfs_set_low_space_thresholds(mp);
201 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
202 
203 	if (delta > 0) {
204 		/*
205 		 * If we expanded the last AG, free the per-AG reservation
206 		 * so we can reinitialize it with the new size.
207 		 */
208 		if (lastag_extended) {
209 			struct xfs_perag	*pag;
210 
211 			pag = xfs_perag_get(mp, id.agno);
212 			error = xfs_ag_resv_free(pag);
213 			xfs_perag_put(pag);
214 			if (error)
215 				return error;
216 		}
217 		/*
218 		 * Reserve AG metadata blocks. ENOSPC here does not mean there
219 		 * was a growfs failure, just that there still isn't space for
220 		 * new user data after the grow has been run.
221 		 */
222 		error = xfs_fs_reserve_ag_blocks(mp);
223 		if (error == -ENOSPC)
224 			error = 0;
225 	}
226 	return error;
227 
228 out_trans_cancel:
229 	xfs_trans_cancel(tp);
230 	return error;
231 }
232 
233 static int
234 xfs_growfs_log_private(
235 	struct xfs_mount	*mp,	/* mount point for filesystem */
236 	struct xfs_growfs_log	*in)	/* growfs log input struct */
237 {
238 	xfs_extlen_t		nb;
239 
240 	nb = in->newblocks;
241 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
242 		return -EINVAL;
243 	if (nb == mp->m_sb.sb_logblocks &&
244 	    in->isint == (mp->m_sb.sb_logstart != 0))
245 		return -EINVAL;
246 	/*
247 	 * Moving the log is hard, need new interfaces to sync
248 	 * the log first, hold off all activity while moving it.
249 	 * Can have shorter or longer log in the same space,
250 	 * or transform internal to external log or vice versa.
251 	 */
252 	return -ENOSYS;
253 }
254 
255 static int
256 xfs_growfs_imaxpct(
257 	struct xfs_mount	*mp,
258 	__u32			imaxpct)
259 {
260 	struct xfs_trans	*tp;
261 	int			dpct;
262 	int			error;
263 
264 	if (imaxpct > 100)
265 		return -EINVAL;
266 
267 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
268 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
269 	if (error)
270 		return error;
271 
272 	dpct = imaxpct - mp->m_sb.sb_imax_pct;
273 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
274 	xfs_trans_set_sync(tp);
275 	return xfs_trans_commit(tp);
276 }
277 
278 /*
279  * protected versions of growfs function acquire and release locks on the mount
280  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
281  * XFS_IOC_FSGROWFSRT
282  */
283 int
284 xfs_growfs_data(
285 	struct xfs_mount	*mp,
286 	struct xfs_growfs_data	*in)
287 {
288 	int			error = 0;
289 
290 	if (!capable(CAP_SYS_ADMIN))
291 		return -EPERM;
292 	if (!mutex_trylock(&mp->m_growlock))
293 		return -EWOULDBLOCK;
294 
295 	/* update imaxpct separately to the physical grow of the filesystem */
296 	if (in->imaxpct != mp->m_sb.sb_imax_pct) {
297 		error = xfs_growfs_imaxpct(mp, in->imaxpct);
298 		if (error)
299 			goto out_error;
300 	}
301 
302 	if (in->newblocks != mp->m_sb.sb_dblocks) {
303 		error = xfs_growfs_data_private(mp, in);
304 		if (error)
305 			goto out_error;
306 	}
307 
308 	/* Post growfs calculations needed to reflect new state in operations */
309 	if (mp->m_sb.sb_imax_pct) {
310 		uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
311 		do_div(icount, 100);
312 		M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
313 	} else
314 		M_IGEO(mp)->maxicount = 0;
315 
316 	/* Update secondary superblocks now the physical grow has completed */
317 	error = xfs_update_secondary_sbs(mp);
318 
319 out_error:
320 	/*
321 	 * Increment the generation unconditionally, the error could be from
322 	 * updating the secondary superblocks, in which case the new size
323 	 * is live already.
324 	 */
325 	mp->m_generation++;
326 	mutex_unlock(&mp->m_growlock);
327 	return error;
328 }
329 
330 int
331 xfs_growfs_log(
332 	xfs_mount_t		*mp,
333 	struct xfs_growfs_log	*in)
334 {
335 	int error;
336 
337 	if (!capable(CAP_SYS_ADMIN))
338 		return -EPERM;
339 	if (!mutex_trylock(&mp->m_growlock))
340 		return -EWOULDBLOCK;
341 	error = xfs_growfs_log_private(mp, in);
342 	mutex_unlock(&mp->m_growlock);
343 	return error;
344 }
345 
346 /*
347  * exported through ioctl XFS_IOC_FSCOUNTS
348  */
349 
350 void
351 xfs_fs_counts(
352 	xfs_mount_t		*mp,
353 	xfs_fsop_counts_t	*cnt)
354 {
355 	cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
356 	cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
357 	cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
358 						xfs_fdblocks_unavailable(mp);
359 	cnt->freertx = percpu_counter_read_positive(&mp->m_frextents);
360 }
361 
362 /*
363  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
364  *
365  * xfs_reserve_blocks is called to set m_resblks
366  * in the in-core mount table. The number of unused reserved blocks
367  * is kept in m_resblks_avail.
368  *
369  * Reserve the requested number of blocks if available. Otherwise return
370  * as many as possible to satisfy the request. The actual number
371  * reserved are returned in outval
372  *
373  * A null inval pointer indicates that only the current reserved blocks
374  * available  should  be returned no settings are changed.
375  */
376 
377 int
378 xfs_reserve_blocks(
379 	xfs_mount_t             *mp,
380 	uint64_t              *inval,
381 	xfs_fsop_resblks_t      *outval)
382 {
383 	int64_t			lcounter, delta;
384 	int64_t			fdblks_delta = 0;
385 	uint64_t		request;
386 	int64_t			free;
387 	int			error = 0;
388 
389 	/* If inval is null, report current values and return */
390 	if (inval == (uint64_t *)NULL) {
391 		if (!outval)
392 			return -EINVAL;
393 		outval->resblks = mp->m_resblks;
394 		outval->resblks_avail = mp->m_resblks_avail;
395 		return 0;
396 	}
397 
398 	request = *inval;
399 
400 	/*
401 	 * With per-cpu counters, this becomes an interesting problem. we need
402 	 * to work out if we are freeing or allocation blocks first, then we can
403 	 * do the modification as necessary.
404 	 *
405 	 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
406 	 * hold out any changes while we work out what to do. This means that
407 	 * the amount of free space can change while we do this, so we need to
408 	 * retry if we end up trying to reserve more space than is available.
409 	 */
410 	spin_lock(&mp->m_sb_lock);
411 
412 	/*
413 	 * If our previous reservation was larger than the current value,
414 	 * then move any unused blocks back to the free pool. Modify the resblks
415 	 * counters directly since we shouldn't have any problems unreserving
416 	 * space.
417 	 */
418 	if (mp->m_resblks > request) {
419 		lcounter = mp->m_resblks_avail - request;
420 		if (lcounter  > 0) {		/* release unused blocks */
421 			fdblks_delta = lcounter;
422 			mp->m_resblks_avail -= lcounter;
423 		}
424 		mp->m_resblks = request;
425 		if (fdblks_delta) {
426 			spin_unlock(&mp->m_sb_lock);
427 			error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
428 			spin_lock(&mp->m_sb_lock);
429 		}
430 
431 		goto out;
432 	}
433 
434 	/*
435 	 * If the request is larger than the current reservation, reserve the
436 	 * blocks before we update the reserve counters. Sample m_fdblocks and
437 	 * perform a partial reservation if the request exceeds free space.
438 	 *
439 	 * The code below estimates how many blocks it can request from
440 	 * fdblocks to stash in the reserve pool.  This is a classic TOCTOU
441 	 * race since fdblocks updates are not always coordinated via
442 	 * m_sb_lock.  Set the reserve size even if there's not enough free
443 	 * space to fill it because mod_fdblocks will refill an undersized
444 	 * reserve when it can.
445 	 */
446 	free = percpu_counter_sum(&mp->m_fdblocks) -
447 						xfs_fdblocks_unavailable(mp);
448 	delta = request - mp->m_resblks;
449 	mp->m_resblks = request;
450 	if (delta > 0 && free > 0) {
451 		/*
452 		 * We'll either succeed in getting space from the free block
453 		 * count or we'll get an ENOSPC.  Don't set the reserved flag
454 		 * here - we don't want to reserve the extra reserve blocks
455 		 * from the reserve.
456 		 *
457 		 * The desired reserve size can change after we drop the lock.
458 		 * Use mod_fdblocks to put the space into the reserve or into
459 		 * fdblocks as appropriate.
460 		 */
461 		fdblks_delta = min(free, delta);
462 		spin_unlock(&mp->m_sb_lock);
463 		error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
464 		if (!error)
465 			xfs_mod_fdblocks(mp, fdblks_delta, 0);
466 		spin_lock(&mp->m_sb_lock);
467 	}
468 out:
469 	if (outval) {
470 		outval->resblks = mp->m_resblks;
471 		outval->resblks_avail = mp->m_resblks_avail;
472 	}
473 
474 	spin_unlock(&mp->m_sb_lock);
475 	return error;
476 }
477 
478 int
479 xfs_fs_goingdown(
480 	xfs_mount_t	*mp,
481 	uint32_t	inflags)
482 {
483 	switch (inflags) {
484 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
485 		if (!freeze_bdev(mp->m_super->s_bdev)) {
486 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
487 			thaw_bdev(mp->m_super->s_bdev);
488 		}
489 		break;
490 	}
491 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
492 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
493 		break;
494 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
495 		xfs_force_shutdown(mp,
496 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
497 		break;
498 	default:
499 		return -EINVAL;
500 	}
501 
502 	return 0;
503 }
504 
505 /*
506  * Force a shutdown of the filesystem instantly while keeping the filesystem
507  * consistent. We don't do an unmount here; just shutdown the shop, make sure
508  * that absolutely nothing persistent happens to this filesystem after this
509  * point.
510  *
511  * The shutdown state change is atomic, resulting in the first and only the
512  * first shutdown call processing the shutdown. This means we only shutdown the
513  * log once as it requires, and we don't spam the logs when multiple concurrent
514  * shutdowns race to set the shutdown flags.
515  */
516 void
517 xfs_do_force_shutdown(
518 	struct xfs_mount *mp,
519 	uint32_t	flags,
520 	char		*fname,
521 	int		lnnum)
522 {
523 	int		tag;
524 	const char	*why;
525 
526 
527 	if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
528 		xlog_shutdown_wait(mp->m_log);
529 		return;
530 	}
531 	if (mp->m_sb_bp)
532 		mp->m_sb_bp->b_flags |= XBF_DONE;
533 
534 	if (flags & SHUTDOWN_FORCE_UMOUNT)
535 		xfs_alert(mp, "User initiated shutdown received.");
536 
537 	if (xlog_force_shutdown(mp->m_log, flags)) {
538 		tag = XFS_PTAG_SHUTDOWN_LOGERROR;
539 		why = "Log I/O Error";
540 	} else if (flags & SHUTDOWN_CORRUPT_INCORE) {
541 		tag = XFS_PTAG_SHUTDOWN_CORRUPT;
542 		why = "Corruption of in-memory data";
543 	} else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
544 		tag = XFS_PTAG_SHUTDOWN_CORRUPT;
545 		why = "Corruption of on-disk metadata";
546 	} else if (flags & SHUTDOWN_DEVICE_REMOVED) {
547 		tag = XFS_PTAG_SHUTDOWN_IOERROR;
548 		why = "Block device removal";
549 	} else {
550 		tag = XFS_PTAG_SHUTDOWN_IOERROR;
551 		why = "Metadata I/O Error";
552 	}
553 
554 	trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
555 
556 	xfs_alert_tag(mp, tag,
557 "%s (0x%x) detected at %pS (%s:%d).  Shutting down filesystem.",
558 			why, flags, __return_address, fname, lnnum);
559 	xfs_alert(mp,
560 		"Please unmount the filesystem and rectify the problem(s)");
561 	if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
562 		xfs_stack_trace();
563 }
564 
565 /*
566  * Reserve free space for per-AG metadata.
567  */
568 int
569 xfs_fs_reserve_ag_blocks(
570 	struct xfs_mount	*mp)
571 {
572 	xfs_agnumber_t		agno;
573 	struct xfs_perag	*pag;
574 	int			error = 0;
575 	int			err2;
576 
577 	mp->m_finobt_nores = false;
578 	for_each_perag(mp, agno, pag) {
579 		err2 = xfs_ag_resv_init(pag, NULL);
580 		if (err2 && !error)
581 			error = err2;
582 	}
583 
584 	if (error && error != -ENOSPC) {
585 		xfs_warn(mp,
586 	"Error %d reserving per-AG metadata reserve pool.", error);
587 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
588 	}
589 
590 	return error;
591 }
592 
593 /*
594  * Free space reserved for per-AG metadata.
595  */
596 int
597 xfs_fs_unreserve_ag_blocks(
598 	struct xfs_mount	*mp)
599 {
600 	xfs_agnumber_t		agno;
601 	struct xfs_perag	*pag;
602 	int			error = 0;
603 	int			err2;
604 
605 	for_each_perag(mp, agno, pag) {
606 		err2 = xfs_ag_resv_free(pag);
607 		if (err2 && !error)
608 			error = err2;
609 	}
610 
611 	if (error)
612 		xfs_warn(mp,
613 	"Error %d freeing per-AG metadata reserve pool.", error);
614 
615 	return error;
616 }
617