xref: /linux/fs/xfs/xfs_fsops.c (revision c4101e55974cc7d835fbd2d8e01553a3f61e9e75)
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 	/* No work to do */
138 	if (delta == 0)
139 		return 0;
140 
141 	oagcount = mp->m_sb.sb_agcount;
142 	/* allocate the new per-ag structures */
143 	if (nagcount > oagcount) {
144 		error = xfs_initialize_perag(mp, nagcount, nb, &nagimax);
145 		if (error)
146 			return error;
147 	} else if (nagcount < oagcount) {
148 		/* TODO: shrinking the entire AGs hasn't yet completed */
149 		return -EINVAL;
150 	}
151 
152 	if (delta > 0)
153 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
154 				XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE,
155 				&tp);
156 	else
157 		error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata, -delta, 0,
158 				0, &tp);
159 	if (error)
160 		goto out_free_unused_perag;
161 
162 	last_pag = xfs_perag_get(mp, oagcount - 1);
163 	if (delta > 0) {
164 		error = xfs_resizefs_init_new_ags(tp, &id, oagcount, nagcount,
165 				delta, last_pag, &lastag_extended);
166 	} else {
167 		xfs_warn_mount(mp, XFS_OPSTATE_WARNED_SHRINK,
168 	"EXPERIMENTAL online shrink feature in use. Use at your own risk!");
169 
170 		error = xfs_ag_shrink_space(last_pag, &tp, -delta);
171 	}
172 	xfs_perag_put(last_pag);
173 	if (error)
174 		goto out_trans_cancel;
175 
176 	/*
177 	 * Update changed superblock fields transactionally. These are not
178 	 * seen by the rest of the world until the transaction commit applies
179 	 * them atomically to the superblock.
180 	 */
181 	if (nagcount > oagcount)
182 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
183 	if (delta)
184 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS, delta);
185 	if (id.nfree)
186 		xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, id.nfree);
187 
188 	/*
189 	 * Sync sb counters now to reflect the updated values. This is
190 	 * particularly important for shrink because the write verifier
191 	 * will fail if sb_fdblocks is ever larger than sb_dblocks.
192 	 */
193 	if (xfs_has_lazysbcount(mp))
194 		xfs_log_sb(tp);
195 
196 	xfs_trans_set_sync(tp);
197 	error = xfs_trans_commit(tp);
198 	if (error)
199 		return error;
200 
201 	/* New allocation groups fully initialized, so update mount struct */
202 	if (nagimax)
203 		mp->m_maxagi = nagimax;
204 	xfs_set_low_space_thresholds(mp);
205 	mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
206 
207 	if (delta > 0) {
208 		/*
209 		 * If we expanded the last AG, free the per-AG reservation
210 		 * so we can reinitialize it with the new size.
211 		 */
212 		if (lastag_extended) {
213 			struct xfs_perag	*pag;
214 
215 			pag = xfs_perag_get(mp, id.agno);
216 			error = xfs_ag_resv_free(pag);
217 			xfs_perag_put(pag);
218 			if (error)
219 				return error;
220 		}
221 		/*
222 		 * Reserve AG metadata blocks. ENOSPC here does not mean there
223 		 * was a growfs failure, just that there still isn't space for
224 		 * new user data after the grow has been run.
225 		 */
226 		error = xfs_fs_reserve_ag_blocks(mp);
227 		if (error == -ENOSPC)
228 			error = 0;
229 	}
230 	return error;
231 
232 out_trans_cancel:
233 	xfs_trans_cancel(tp);
234 out_free_unused_perag:
235 	if (nagcount > oagcount)
236 		xfs_free_unused_perag_range(mp, oagcount, nagcount);
237 	return error;
238 }
239 
240 static int
241 xfs_growfs_log_private(
242 	struct xfs_mount	*mp,	/* mount point for filesystem */
243 	struct xfs_growfs_log	*in)	/* growfs log input struct */
244 {
245 	xfs_extlen_t		nb;
246 
247 	nb = in->newblocks;
248 	if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
249 		return -EINVAL;
250 	if (nb == mp->m_sb.sb_logblocks &&
251 	    in->isint == (mp->m_sb.sb_logstart != 0))
252 		return -EINVAL;
253 	/*
254 	 * Moving the log is hard, need new interfaces to sync
255 	 * the log first, hold off all activity while moving it.
256 	 * Can have shorter or longer log in the same space,
257 	 * or transform internal to external log or vice versa.
258 	 */
259 	return -ENOSYS;
260 }
261 
262 static int
263 xfs_growfs_imaxpct(
264 	struct xfs_mount	*mp,
265 	__u32			imaxpct)
266 {
267 	struct xfs_trans	*tp;
268 	int			dpct;
269 	int			error;
270 
271 	if (imaxpct > 100)
272 		return -EINVAL;
273 
274 	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
275 			XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
276 	if (error)
277 		return error;
278 
279 	dpct = imaxpct - mp->m_sb.sb_imax_pct;
280 	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
281 	xfs_trans_set_sync(tp);
282 	return xfs_trans_commit(tp);
283 }
284 
285 /*
286  * protected versions of growfs function acquire and release locks on the mount
287  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
288  * XFS_IOC_FSGROWFSRT
289  */
290 int
291 xfs_growfs_data(
292 	struct xfs_mount	*mp,
293 	struct xfs_growfs_data	*in)
294 {
295 	int			error = 0;
296 
297 	if (!capable(CAP_SYS_ADMIN))
298 		return -EPERM;
299 	if (!mutex_trylock(&mp->m_growlock))
300 		return -EWOULDBLOCK;
301 
302 	/* update imaxpct separately to the physical grow of the filesystem */
303 	if (in->imaxpct != mp->m_sb.sb_imax_pct) {
304 		error = xfs_growfs_imaxpct(mp, in->imaxpct);
305 		if (error)
306 			goto out_error;
307 	}
308 
309 	if (in->newblocks != mp->m_sb.sb_dblocks) {
310 		error = xfs_growfs_data_private(mp, in);
311 		if (error)
312 			goto out_error;
313 	}
314 
315 	/* Post growfs calculations needed to reflect new state in operations */
316 	if (mp->m_sb.sb_imax_pct) {
317 		uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
318 		do_div(icount, 100);
319 		M_IGEO(mp)->maxicount = XFS_FSB_TO_INO(mp, icount);
320 	} else
321 		M_IGEO(mp)->maxicount = 0;
322 
323 	/* Update secondary superblocks now the physical grow has completed */
324 	error = xfs_update_secondary_sbs(mp);
325 
326 out_error:
327 	/*
328 	 * Increment the generation unconditionally, the error could be from
329 	 * updating the secondary superblocks, in which case the new size
330 	 * is live already.
331 	 */
332 	mp->m_generation++;
333 	mutex_unlock(&mp->m_growlock);
334 	return error;
335 }
336 
337 int
338 xfs_growfs_log(
339 	xfs_mount_t		*mp,
340 	struct xfs_growfs_log	*in)
341 {
342 	int error;
343 
344 	if (!capable(CAP_SYS_ADMIN))
345 		return -EPERM;
346 	if (!mutex_trylock(&mp->m_growlock))
347 		return -EWOULDBLOCK;
348 	error = xfs_growfs_log_private(mp, in);
349 	mutex_unlock(&mp->m_growlock);
350 	return error;
351 }
352 
353 /*
354  * Reserve the requested number of blocks if available. Otherwise return
355  * as many as possible to satisfy the request. The actual number
356  * reserved are returned in outval.
357  */
358 int
359 xfs_reserve_blocks(
360 	struct xfs_mount	*mp,
361 	uint64_t		request)
362 {
363 	int64_t			lcounter, delta;
364 	int64_t			fdblks_delta = 0;
365 	int64_t			free;
366 	int			error = 0;
367 
368 	/*
369 	 * With per-cpu counters, this becomes an interesting problem. we need
370 	 * to work out if we are freeing or allocation blocks first, then we can
371 	 * do the modification as necessary.
372 	 *
373 	 * We do this under the m_sb_lock so that if we are near ENOSPC, we will
374 	 * hold out any changes while we work out what to do. This means that
375 	 * the amount of free space can change while we do this, so we need to
376 	 * retry if we end up trying to reserve more space than is available.
377 	 */
378 	spin_lock(&mp->m_sb_lock);
379 
380 	/*
381 	 * If our previous reservation was larger than the current value,
382 	 * then move any unused blocks back to the free pool. Modify the resblks
383 	 * counters directly since we shouldn't have any problems unreserving
384 	 * space.
385 	 */
386 	if (mp->m_resblks > request) {
387 		lcounter = mp->m_resblks_avail - request;
388 		if (lcounter  > 0) {		/* release unused blocks */
389 			fdblks_delta = lcounter;
390 			mp->m_resblks_avail -= lcounter;
391 		}
392 		mp->m_resblks = request;
393 		if (fdblks_delta) {
394 			spin_unlock(&mp->m_sb_lock);
395 			error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
396 			spin_lock(&mp->m_sb_lock);
397 		}
398 
399 		goto out;
400 	}
401 
402 	/*
403 	 * If the request is larger than the current reservation, reserve the
404 	 * blocks before we update the reserve counters. Sample m_fdblocks and
405 	 * perform a partial reservation if the request exceeds free space.
406 	 *
407 	 * The code below estimates how many blocks it can request from
408 	 * fdblocks to stash in the reserve pool.  This is a classic TOCTOU
409 	 * race since fdblocks updates are not always coordinated via
410 	 * m_sb_lock.  Set the reserve size even if there's not enough free
411 	 * space to fill it because mod_fdblocks will refill an undersized
412 	 * reserve when it can.
413 	 */
414 	free = percpu_counter_sum(&mp->m_fdblocks) -
415 						xfs_fdblocks_unavailable(mp);
416 	delta = request - mp->m_resblks;
417 	mp->m_resblks = request;
418 	if (delta > 0 && free > 0) {
419 		/*
420 		 * We'll either succeed in getting space from the free block
421 		 * count or we'll get an ENOSPC.  Don't set the reserved flag
422 		 * here - we don't want to reserve the extra reserve blocks
423 		 * from the reserve.
424 		 *
425 		 * The desired reserve size can change after we drop the lock.
426 		 * Use mod_fdblocks to put the space into the reserve or into
427 		 * fdblocks as appropriate.
428 		 */
429 		fdblks_delta = min(free, delta);
430 		spin_unlock(&mp->m_sb_lock);
431 		error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
432 		if (!error)
433 			xfs_mod_fdblocks(mp, fdblks_delta, 0);
434 		spin_lock(&mp->m_sb_lock);
435 	}
436 out:
437 	spin_unlock(&mp->m_sb_lock);
438 	return error;
439 }
440 
441 int
442 xfs_fs_goingdown(
443 	xfs_mount_t	*mp,
444 	uint32_t	inflags)
445 {
446 	switch (inflags) {
447 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
448 		if (!bdev_freeze(mp->m_super->s_bdev)) {
449 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
450 			bdev_thaw(mp->m_super->s_bdev);
451 		}
452 		break;
453 	}
454 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
455 		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
456 		break;
457 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
458 		xfs_force_shutdown(mp,
459 				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
460 		break;
461 	default:
462 		return -EINVAL;
463 	}
464 
465 	return 0;
466 }
467 
468 /*
469  * Force a shutdown of the filesystem instantly while keeping the filesystem
470  * consistent. We don't do an unmount here; just shutdown the shop, make sure
471  * that absolutely nothing persistent happens to this filesystem after this
472  * point.
473  *
474  * The shutdown state change is atomic, resulting in the first and only the
475  * first shutdown call processing the shutdown. This means we only shutdown the
476  * log once as it requires, and we don't spam the logs when multiple concurrent
477  * shutdowns race to set the shutdown flags.
478  */
479 void
480 xfs_do_force_shutdown(
481 	struct xfs_mount *mp,
482 	uint32_t	flags,
483 	char		*fname,
484 	int		lnnum)
485 {
486 	int		tag;
487 	const char	*why;
488 
489 
490 	if (test_and_set_bit(XFS_OPSTATE_SHUTDOWN, &mp->m_opstate)) {
491 		xlog_shutdown_wait(mp->m_log);
492 		return;
493 	}
494 	if (mp->m_sb_bp)
495 		mp->m_sb_bp->b_flags |= XBF_DONE;
496 
497 	if (flags & SHUTDOWN_FORCE_UMOUNT)
498 		xfs_alert(mp, "User initiated shutdown received.");
499 
500 	if (xlog_force_shutdown(mp->m_log, flags)) {
501 		tag = XFS_PTAG_SHUTDOWN_LOGERROR;
502 		why = "Log I/O Error";
503 	} else if (flags & SHUTDOWN_CORRUPT_INCORE) {
504 		tag = XFS_PTAG_SHUTDOWN_CORRUPT;
505 		why = "Corruption of in-memory data";
506 	} else if (flags & SHUTDOWN_CORRUPT_ONDISK) {
507 		tag = XFS_PTAG_SHUTDOWN_CORRUPT;
508 		why = "Corruption of on-disk metadata";
509 	} else if (flags & SHUTDOWN_DEVICE_REMOVED) {
510 		tag = XFS_PTAG_SHUTDOWN_IOERROR;
511 		why = "Block device removal";
512 	} else {
513 		tag = XFS_PTAG_SHUTDOWN_IOERROR;
514 		why = "Metadata I/O Error";
515 	}
516 
517 	trace_xfs_force_shutdown(mp, tag, flags, fname, lnnum);
518 
519 	xfs_alert_tag(mp, tag,
520 "%s (0x%x) detected at %pS (%s:%d).  Shutting down filesystem.",
521 			why, flags, __return_address, fname, lnnum);
522 	xfs_alert(mp,
523 		"Please unmount the filesystem and rectify the problem(s)");
524 	if (xfs_error_level >= XFS_ERRLEVEL_HIGH)
525 		xfs_stack_trace();
526 }
527 
528 /*
529  * Reserve free space for per-AG metadata.
530  */
531 int
532 xfs_fs_reserve_ag_blocks(
533 	struct xfs_mount	*mp)
534 {
535 	xfs_agnumber_t		agno;
536 	struct xfs_perag	*pag;
537 	int			error = 0;
538 	int			err2;
539 
540 	mp->m_finobt_nores = false;
541 	for_each_perag(mp, agno, pag) {
542 		err2 = xfs_ag_resv_init(pag, NULL);
543 		if (err2 && !error)
544 			error = err2;
545 	}
546 
547 	if (error && error != -ENOSPC) {
548 		xfs_warn(mp,
549 	"Error %d reserving per-AG metadata reserve pool.", error);
550 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
551 	}
552 
553 	return error;
554 }
555 
556 /*
557  * Free space reserved for per-AG metadata.
558  */
559 int
560 xfs_fs_unreserve_ag_blocks(
561 	struct xfs_mount	*mp)
562 {
563 	xfs_agnumber_t		agno;
564 	struct xfs_perag	*pag;
565 	int			error = 0;
566 	int			err2;
567 
568 	for_each_perag(mp, agno, pag) {
569 		err2 = xfs_ag_resv_free(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