1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6
7 #include "xfs.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_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap.h"
17 #include "xfs_alloc.h"
18 #include "xfs_fsops.h"
19 #include "xfs_trans.h"
20 #include "xfs_buf_item.h"
21 #include "xfs_log.h"
22 #include "xfs_log_priv.h"
23 #include "xfs_dir2.h"
24 #include "xfs_extfree_item.h"
25 #include "xfs_mru_cache.h"
26 #include "xfs_inode_item.h"
27 #include "xfs_icache.h"
28 #include "xfs_trace.h"
29 #include "xfs_icreate_item.h"
30 #include "xfs_filestream.h"
31 #include "xfs_quota.h"
32 #include "xfs_sysfs.h"
33 #include "xfs_ondisk.h"
34 #include "xfs_rmap_item.h"
35 #include "xfs_refcount_item.h"
36 #include "xfs_bmap_item.h"
37 #include "xfs_reflink.h"
38 #include "xfs_pwork.h"
39 #include "xfs_ag.h"
40 #include "xfs_defer.h"
41 #include "xfs_attr_item.h"
42 #include "xfs_xattr.h"
43 #include "xfs_iunlink_item.h"
44
45 #include <linux/magic.h>
46 #include <linux/fs_context.h>
47 #include <linux/fs_parser.h>
48
49 static const struct super_operations xfs_super_operations;
50
51 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
52 #ifdef DEBUG
53 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
54 #endif
55
56 #ifdef CONFIG_HOTPLUG_CPU
57 static LIST_HEAD(xfs_mount_list);
58 static DEFINE_SPINLOCK(xfs_mount_list_lock);
59
xfs_mount_list_add(struct xfs_mount * mp)60 static inline void xfs_mount_list_add(struct xfs_mount *mp)
61 {
62 spin_lock(&xfs_mount_list_lock);
63 list_add(&mp->m_mount_list, &xfs_mount_list);
64 spin_unlock(&xfs_mount_list_lock);
65 }
66
xfs_mount_list_del(struct xfs_mount * mp)67 static inline void xfs_mount_list_del(struct xfs_mount *mp)
68 {
69 spin_lock(&xfs_mount_list_lock);
70 list_del(&mp->m_mount_list);
71 spin_unlock(&xfs_mount_list_lock);
72 }
73 #else /* !CONFIG_HOTPLUG_CPU */
xfs_mount_list_add(struct xfs_mount * mp)74 static inline void xfs_mount_list_add(struct xfs_mount *mp) {}
xfs_mount_list_del(struct xfs_mount * mp)75 static inline void xfs_mount_list_del(struct xfs_mount *mp) {}
76 #endif
77
78 enum xfs_dax_mode {
79 XFS_DAX_INODE = 0,
80 XFS_DAX_ALWAYS = 1,
81 XFS_DAX_NEVER = 2,
82 };
83
84 static void
xfs_mount_set_dax_mode(struct xfs_mount * mp,enum xfs_dax_mode mode)85 xfs_mount_set_dax_mode(
86 struct xfs_mount *mp,
87 enum xfs_dax_mode mode)
88 {
89 switch (mode) {
90 case XFS_DAX_INODE:
91 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
92 break;
93 case XFS_DAX_ALWAYS:
94 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
95 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
96 break;
97 case XFS_DAX_NEVER:
98 mp->m_features |= XFS_FEAT_DAX_NEVER;
99 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
100 break;
101 }
102 }
103
104 static const struct constant_table dax_param_enums[] = {
105 {"inode", XFS_DAX_INODE },
106 {"always", XFS_DAX_ALWAYS },
107 {"never", XFS_DAX_NEVER },
108 {}
109 };
110
111 /*
112 * Table driven mount option parser.
113 */
114 enum {
115 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
116 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
117 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
118 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
119 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
120 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
121 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
122 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
123 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
124 };
125
126 static const struct fs_parameter_spec xfs_fs_parameters[] = {
127 fsparam_u32("logbufs", Opt_logbufs),
128 fsparam_string("logbsize", Opt_logbsize),
129 fsparam_string("logdev", Opt_logdev),
130 fsparam_string("rtdev", Opt_rtdev),
131 fsparam_flag("wsync", Opt_wsync),
132 fsparam_flag("noalign", Opt_noalign),
133 fsparam_flag("swalloc", Opt_swalloc),
134 fsparam_u32("sunit", Opt_sunit),
135 fsparam_u32("swidth", Opt_swidth),
136 fsparam_flag("nouuid", Opt_nouuid),
137 fsparam_flag("grpid", Opt_grpid),
138 fsparam_flag("nogrpid", Opt_nogrpid),
139 fsparam_flag("bsdgroups", Opt_bsdgroups),
140 fsparam_flag("sysvgroups", Opt_sysvgroups),
141 fsparam_string("allocsize", Opt_allocsize),
142 fsparam_flag("norecovery", Opt_norecovery),
143 fsparam_flag("inode64", Opt_inode64),
144 fsparam_flag("inode32", Opt_inode32),
145 fsparam_flag("ikeep", Opt_ikeep),
146 fsparam_flag("noikeep", Opt_noikeep),
147 fsparam_flag("largeio", Opt_largeio),
148 fsparam_flag("nolargeio", Opt_nolargeio),
149 fsparam_flag("attr2", Opt_attr2),
150 fsparam_flag("noattr2", Opt_noattr2),
151 fsparam_flag("filestreams", Opt_filestreams),
152 fsparam_flag("quota", Opt_quota),
153 fsparam_flag("noquota", Opt_noquota),
154 fsparam_flag("usrquota", Opt_usrquota),
155 fsparam_flag("grpquota", Opt_grpquota),
156 fsparam_flag("prjquota", Opt_prjquota),
157 fsparam_flag("uquota", Opt_uquota),
158 fsparam_flag("gquota", Opt_gquota),
159 fsparam_flag("pquota", Opt_pquota),
160 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
161 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
162 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
163 fsparam_flag("qnoenforce", Opt_qnoenforce),
164 fsparam_flag("discard", Opt_discard),
165 fsparam_flag("nodiscard", Opt_nodiscard),
166 fsparam_flag("dax", Opt_dax),
167 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
168 {}
169 };
170
171 struct proc_xfs_info {
172 uint64_t flag;
173 char *str;
174 };
175
176 static int
xfs_fs_show_options(struct seq_file * m,struct dentry * root)177 xfs_fs_show_options(
178 struct seq_file *m,
179 struct dentry *root)
180 {
181 static struct proc_xfs_info xfs_info_set[] = {
182 /* the few simple ones we can get from the mount struct */
183 { XFS_FEAT_IKEEP, ",ikeep" },
184 { XFS_FEAT_WSYNC, ",wsync" },
185 { XFS_FEAT_NOALIGN, ",noalign" },
186 { XFS_FEAT_SWALLOC, ",swalloc" },
187 { XFS_FEAT_NOUUID, ",nouuid" },
188 { XFS_FEAT_NORECOVERY, ",norecovery" },
189 { XFS_FEAT_ATTR2, ",attr2" },
190 { XFS_FEAT_FILESTREAMS, ",filestreams" },
191 { XFS_FEAT_GRPID, ",grpid" },
192 { XFS_FEAT_DISCARD, ",discard" },
193 { XFS_FEAT_LARGE_IOSIZE, ",largeio" },
194 { XFS_FEAT_DAX_ALWAYS, ",dax=always" },
195 { XFS_FEAT_DAX_NEVER, ",dax=never" },
196 { 0, NULL }
197 };
198 struct xfs_mount *mp = XFS_M(root->d_sb);
199 struct proc_xfs_info *xfs_infop;
200
201 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
202 if (mp->m_features & xfs_infop->flag)
203 seq_puts(m, xfs_infop->str);
204 }
205
206 seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
207
208 if (xfs_has_allocsize(mp))
209 seq_printf(m, ",allocsize=%dk",
210 (1 << mp->m_allocsize_log) >> 10);
211
212 if (mp->m_logbufs > 0)
213 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
214 if (mp->m_logbsize > 0)
215 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
216
217 if (mp->m_logname)
218 seq_show_option(m, "logdev", mp->m_logname);
219 if (mp->m_rtname)
220 seq_show_option(m, "rtdev", mp->m_rtname);
221
222 if (mp->m_dalign > 0)
223 seq_printf(m, ",sunit=%d",
224 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
225 if (mp->m_swidth > 0)
226 seq_printf(m, ",swidth=%d",
227 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
228
229 if (mp->m_qflags & XFS_UQUOTA_ENFD)
230 seq_puts(m, ",usrquota");
231 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
232 seq_puts(m, ",uqnoenforce");
233
234 if (mp->m_qflags & XFS_PQUOTA_ENFD)
235 seq_puts(m, ",prjquota");
236 else if (mp->m_qflags & XFS_PQUOTA_ACCT)
237 seq_puts(m, ",pqnoenforce");
238
239 if (mp->m_qflags & XFS_GQUOTA_ENFD)
240 seq_puts(m, ",grpquota");
241 else if (mp->m_qflags & XFS_GQUOTA_ACCT)
242 seq_puts(m, ",gqnoenforce");
243
244 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
245 seq_puts(m, ",noquota");
246
247 return 0;
248 }
249
250 /*
251 * Set parameters for inode allocation heuristics, taking into account
252 * filesystem size and inode32/inode64 mount options; i.e. specifically
253 * whether or not XFS_FEAT_SMALL_INUMS is set.
254 *
255 * Inode allocation patterns are altered only if inode32 is requested
256 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
257 * If altered, XFS_OPSTATE_INODE32 is set as well.
258 *
259 * An agcount independent of that in the mount structure is provided
260 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
261 * to the potentially higher ag count.
262 *
263 * Returns the maximum AG index which may contain inodes.
264 */
265 xfs_agnumber_t
xfs_set_inode_alloc(struct xfs_mount * mp,xfs_agnumber_t agcount)266 xfs_set_inode_alloc(
267 struct xfs_mount *mp,
268 xfs_agnumber_t agcount)
269 {
270 xfs_agnumber_t index;
271 xfs_agnumber_t maxagi = 0;
272 xfs_sb_t *sbp = &mp->m_sb;
273 xfs_agnumber_t max_metadata;
274 xfs_agino_t agino;
275 xfs_ino_t ino;
276
277 /*
278 * Calculate how much should be reserved for inodes to meet
279 * the max inode percentage. Used only for inode32.
280 */
281 if (M_IGEO(mp)->maxicount) {
282 uint64_t icount;
283
284 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
285 do_div(icount, 100);
286 icount += sbp->sb_agblocks - 1;
287 do_div(icount, sbp->sb_agblocks);
288 max_metadata = icount;
289 } else {
290 max_metadata = agcount;
291 }
292
293 /* Get the last possible inode in the filesystem */
294 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
295 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
296
297 /*
298 * If user asked for no more than 32-bit inodes, and the fs is
299 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
300 * the allocator to accommodate the request.
301 */
302 if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
303 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
304 else
305 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
306
307 for (index = 0; index < agcount; index++) {
308 struct xfs_perag *pag;
309
310 ino = XFS_AGINO_TO_INO(mp, index, agino);
311
312 pag = xfs_perag_get(mp, index);
313
314 if (xfs_is_inode32(mp)) {
315 if (ino > XFS_MAXINUMBER_32) {
316 pag->pagi_inodeok = 0;
317 pag->pagf_metadata = 0;
318 } else {
319 pag->pagi_inodeok = 1;
320 maxagi++;
321 if (index < max_metadata)
322 pag->pagf_metadata = 1;
323 else
324 pag->pagf_metadata = 0;
325 }
326 } else {
327 pag->pagi_inodeok = 1;
328 pag->pagf_metadata = 0;
329 }
330
331 xfs_perag_put(pag);
332 }
333
334 return xfs_is_inode32(mp) ? maxagi : agcount;
335 }
336
337 static int
xfs_setup_dax_always(struct xfs_mount * mp)338 xfs_setup_dax_always(
339 struct xfs_mount *mp)
340 {
341 if (!mp->m_ddev_targp->bt_daxdev &&
342 (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
343 xfs_alert(mp,
344 "DAX unsupported by block device. Turning off DAX.");
345 goto disable_dax;
346 }
347
348 if (mp->m_super->s_blocksize != PAGE_SIZE) {
349 xfs_alert(mp,
350 "DAX not supported for blocksize. Turning off DAX.");
351 goto disable_dax;
352 }
353
354 if (xfs_has_reflink(mp) &&
355 bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
356 xfs_alert(mp,
357 "DAX and reflink cannot work with multi-partitions!");
358 return -EINVAL;
359 }
360
361 xfs_warn(mp, "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
362 return 0;
363
364 disable_dax:
365 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
366 return 0;
367 }
368
369 STATIC int
xfs_blkdev_get(xfs_mount_t * mp,const char * name,struct block_device ** bdevp)370 xfs_blkdev_get(
371 xfs_mount_t *mp,
372 const char *name,
373 struct block_device **bdevp)
374 {
375 int error = 0;
376
377 *bdevp = blkdev_get_by_path(name, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
378 mp);
379 if (IS_ERR(*bdevp)) {
380 error = PTR_ERR(*bdevp);
381 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
382 }
383
384 return error;
385 }
386
387 STATIC void
xfs_blkdev_put(struct block_device * bdev)388 xfs_blkdev_put(
389 struct block_device *bdev)
390 {
391 if (bdev)
392 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
393 }
394
395 STATIC void
xfs_close_devices(struct xfs_mount * mp)396 xfs_close_devices(
397 struct xfs_mount *mp)
398 {
399 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
400 struct block_device *logdev = mp->m_logdev_targp->bt_bdev;
401
402 xfs_free_buftarg(mp->m_logdev_targp);
403 xfs_blkdev_put(logdev);
404 }
405 if (mp->m_rtdev_targp) {
406 struct block_device *rtdev = mp->m_rtdev_targp->bt_bdev;
407
408 xfs_free_buftarg(mp->m_rtdev_targp);
409 xfs_blkdev_put(rtdev);
410 }
411 xfs_free_buftarg(mp->m_ddev_targp);
412 }
413
414 /*
415 * The file system configurations are:
416 * (1) device (partition) with data and internal log
417 * (2) logical volume with data and log subvolumes.
418 * (3) logical volume with data, log, and realtime subvolumes.
419 *
420 * We only have to handle opening the log and realtime volumes here if
421 * they are present. The data subvolume has already been opened by
422 * get_sb_bdev() and is stored in sb->s_bdev.
423 */
424 STATIC int
xfs_open_devices(struct xfs_mount * mp)425 xfs_open_devices(
426 struct xfs_mount *mp)
427 {
428 struct block_device *ddev = mp->m_super->s_bdev;
429 struct block_device *logdev = NULL, *rtdev = NULL;
430 int error;
431
432 /*
433 * Open real time and log devices - order is important.
434 */
435 if (mp->m_logname) {
436 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
437 if (error)
438 return error;
439 }
440
441 if (mp->m_rtname) {
442 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
443 if (error)
444 goto out_close_logdev;
445
446 if (rtdev == ddev || rtdev == logdev) {
447 xfs_warn(mp,
448 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
449 error = -EINVAL;
450 goto out_close_rtdev;
451 }
452 }
453
454 /*
455 * Setup xfs_mount buffer target pointers
456 */
457 error = -ENOMEM;
458 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
459 if (!mp->m_ddev_targp)
460 goto out_close_rtdev;
461
462 if (rtdev) {
463 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
464 if (!mp->m_rtdev_targp)
465 goto out_free_ddev_targ;
466 }
467
468 if (logdev && logdev != ddev) {
469 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
470 if (!mp->m_logdev_targp)
471 goto out_free_rtdev_targ;
472 } else {
473 mp->m_logdev_targp = mp->m_ddev_targp;
474 }
475
476 return 0;
477
478 out_free_rtdev_targ:
479 if (mp->m_rtdev_targp)
480 xfs_free_buftarg(mp->m_rtdev_targp);
481 out_free_ddev_targ:
482 xfs_free_buftarg(mp->m_ddev_targp);
483 out_close_rtdev:
484 xfs_blkdev_put(rtdev);
485 out_close_logdev:
486 if (logdev && logdev != ddev)
487 xfs_blkdev_put(logdev);
488 return error;
489 }
490
491 /*
492 * Setup xfs_mount buffer target pointers based on superblock
493 */
494 STATIC int
xfs_setup_devices(struct xfs_mount * mp)495 xfs_setup_devices(
496 struct xfs_mount *mp)
497 {
498 int error;
499
500 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
501 if (error)
502 return error;
503
504 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
505 unsigned int log_sector_size = BBSIZE;
506
507 if (xfs_has_sector(mp))
508 log_sector_size = mp->m_sb.sb_logsectsize;
509 error = xfs_setsize_buftarg(mp->m_logdev_targp,
510 log_sector_size);
511 if (error)
512 return error;
513 }
514 if (mp->m_rtdev_targp) {
515 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
516 mp->m_sb.sb_sectsize);
517 if (error)
518 return error;
519 }
520
521 return 0;
522 }
523
524 STATIC int
xfs_init_mount_workqueues(struct xfs_mount * mp)525 xfs_init_mount_workqueues(
526 struct xfs_mount *mp)
527 {
528 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
529 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
530 1, mp->m_super->s_id);
531 if (!mp->m_buf_workqueue)
532 goto out;
533
534 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
535 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
536 0, mp->m_super->s_id);
537 if (!mp->m_unwritten_workqueue)
538 goto out_destroy_buf;
539
540 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
541 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
542 0, mp->m_super->s_id);
543 if (!mp->m_reclaim_workqueue)
544 goto out_destroy_unwritten;
545
546 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
547 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
548 0, mp->m_super->s_id);
549 if (!mp->m_blockgc_wq)
550 goto out_destroy_reclaim;
551
552 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
553 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
554 1, mp->m_super->s_id);
555 if (!mp->m_inodegc_wq)
556 goto out_destroy_blockgc;
557
558 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
559 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
560 if (!mp->m_sync_workqueue)
561 goto out_destroy_inodegc;
562
563 return 0;
564
565 out_destroy_inodegc:
566 destroy_workqueue(mp->m_inodegc_wq);
567 out_destroy_blockgc:
568 destroy_workqueue(mp->m_blockgc_wq);
569 out_destroy_reclaim:
570 destroy_workqueue(mp->m_reclaim_workqueue);
571 out_destroy_unwritten:
572 destroy_workqueue(mp->m_unwritten_workqueue);
573 out_destroy_buf:
574 destroy_workqueue(mp->m_buf_workqueue);
575 out:
576 return -ENOMEM;
577 }
578
579 STATIC void
xfs_destroy_mount_workqueues(struct xfs_mount * mp)580 xfs_destroy_mount_workqueues(
581 struct xfs_mount *mp)
582 {
583 destroy_workqueue(mp->m_sync_workqueue);
584 destroy_workqueue(mp->m_blockgc_wq);
585 destroy_workqueue(mp->m_inodegc_wq);
586 destroy_workqueue(mp->m_reclaim_workqueue);
587 destroy_workqueue(mp->m_unwritten_workqueue);
588 destroy_workqueue(mp->m_buf_workqueue);
589 }
590
591 static void
xfs_flush_inodes_worker(struct work_struct * work)592 xfs_flush_inodes_worker(
593 struct work_struct *work)
594 {
595 struct xfs_mount *mp = container_of(work, struct xfs_mount,
596 m_flush_inodes_work);
597 struct super_block *sb = mp->m_super;
598
599 if (down_read_trylock(&sb->s_umount)) {
600 sync_inodes_sb(sb);
601 up_read(&sb->s_umount);
602 }
603 }
604
605 /*
606 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
607 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
608 * for IO to complete so that we effectively throttle multiple callers to the
609 * rate at which IO is completing.
610 */
611 void
xfs_flush_inodes(struct xfs_mount * mp)612 xfs_flush_inodes(
613 struct xfs_mount *mp)
614 {
615 /*
616 * If flush_work() returns true then that means we waited for a flush
617 * which was already in progress. Don't bother running another scan.
618 */
619 if (flush_work(&mp->m_flush_inodes_work))
620 return;
621
622 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
623 flush_work(&mp->m_flush_inodes_work);
624 }
625
626 /* Catch misguided souls that try to use this interface on XFS */
627 STATIC struct inode *
xfs_fs_alloc_inode(struct super_block * sb)628 xfs_fs_alloc_inode(
629 struct super_block *sb)
630 {
631 BUG();
632 return NULL;
633 }
634
635 /*
636 * Now that the generic code is guaranteed not to be accessing
637 * the linux inode, we can inactivate and reclaim the inode.
638 */
639 STATIC void
xfs_fs_destroy_inode(struct inode * inode)640 xfs_fs_destroy_inode(
641 struct inode *inode)
642 {
643 struct xfs_inode *ip = XFS_I(inode);
644
645 trace_xfs_destroy_inode(ip);
646
647 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
648 XFS_STATS_INC(ip->i_mount, vn_rele);
649 XFS_STATS_INC(ip->i_mount, vn_remove);
650 xfs_inode_mark_reclaimable(ip);
651 }
652
653 static void
xfs_fs_dirty_inode(struct inode * inode,int flags)654 xfs_fs_dirty_inode(
655 struct inode *inode,
656 int flags)
657 {
658 struct xfs_inode *ip = XFS_I(inode);
659 struct xfs_mount *mp = ip->i_mount;
660 struct xfs_trans *tp;
661
662 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
663 return;
664
665 /*
666 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
667 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
668 * in flags possibly together with I_DIRTY_SYNC.
669 */
670 if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
671 return;
672
673 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
674 return;
675 xfs_ilock(ip, XFS_ILOCK_EXCL);
676 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
677 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
678 xfs_trans_commit(tp);
679 }
680
681 /*
682 * Slab object creation initialisation for the XFS inode.
683 * This covers only the idempotent fields in the XFS inode;
684 * all other fields need to be initialised on allocation
685 * from the slab. This avoids the need to repeatedly initialise
686 * fields in the xfs inode that left in the initialise state
687 * when freeing the inode.
688 */
689 STATIC void
xfs_fs_inode_init_once(void * inode)690 xfs_fs_inode_init_once(
691 void *inode)
692 {
693 struct xfs_inode *ip = inode;
694
695 memset(ip, 0, sizeof(struct xfs_inode));
696
697 /* vfs inode */
698 inode_init_once(VFS_I(ip));
699
700 /* xfs inode */
701 atomic_set(&ip->i_pincount, 0);
702 spin_lock_init(&ip->i_flags_lock);
703
704 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
705 "xfsino", ip->i_ino);
706 }
707
708 /*
709 * We do an unlocked check for XFS_IDONTCACHE here because we are already
710 * serialised against cache hits here via the inode->i_lock and igrab() in
711 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
712 * racing with us, and it avoids needing to grab a spinlock here for every inode
713 * we drop the final reference on.
714 */
715 STATIC int
xfs_fs_drop_inode(struct inode * inode)716 xfs_fs_drop_inode(
717 struct inode *inode)
718 {
719 struct xfs_inode *ip = XFS_I(inode);
720
721 /*
722 * If this unlinked inode is in the middle of recovery, don't
723 * drop the inode just yet; log recovery will take care of
724 * that. See the comment for this inode flag.
725 */
726 if (ip->i_flags & XFS_IRECOVERY) {
727 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
728 return 0;
729 }
730
731 return generic_drop_inode(inode);
732 }
733
734 static void
xfs_mount_free(struct xfs_mount * mp)735 xfs_mount_free(
736 struct xfs_mount *mp)
737 {
738 kfree(mp->m_rtname);
739 kfree(mp->m_logname);
740 kmem_free(mp);
741 }
742
743 STATIC int
xfs_fs_sync_fs(struct super_block * sb,int wait)744 xfs_fs_sync_fs(
745 struct super_block *sb,
746 int wait)
747 {
748 struct xfs_mount *mp = XFS_M(sb);
749 int error;
750
751 trace_xfs_fs_sync_fs(mp, __return_address);
752
753 /*
754 * Doing anything during the async pass would be counterproductive.
755 */
756 if (!wait)
757 return 0;
758
759 error = xfs_log_force(mp, XFS_LOG_SYNC);
760 if (error)
761 return error;
762
763 if (laptop_mode) {
764 /*
765 * The disk must be active because we're syncing.
766 * We schedule log work now (now that the disk is
767 * active) instead of later (when it might not be).
768 */
769 flush_delayed_work(&mp->m_log->l_work);
770 }
771
772 /*
773 * If we are called with page faults frozen out, it means we are about
774 * to freeze the transaction subsystem. Take the opportunity to shut
775 * down inodegc because once SB_FREEZE_FS is set it's too late to
776 * prevent inactivation races with freeze. The fs doesn't get called
777 * again by the freezing process until after SB_FREEZE_FS has been set,
778 * so it's now or never. Same logic applies to speculative allocation
779 * garbage collection.
780 *
781 * We don't care if this is a normal syncfs call that does this or
782 * freeze that does this - we can run this multiple times without issue
783 * and we won't race with a restart because a restart can only occur
784 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
785 */
786 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
787 xfs_inodegc_stop(mp);
788 xfs_blockgc_stop(mp);
789 }
790
791 return 0;
792 }
793
794 STATIC int
xfs_fs_statfs(struct dentry * dentry,struct kstatfs * statp)795 xfs_fs_statfs(
796 struct dentry *dentry,
797 struct kstatfs *statp)
798 {
799 struct xfs_mount *mp = XFS_M(dentry->d_sb);
800 xfs_sb_t *sbp = &mp->m_sb;
801 struct xfs_inode *ip = XFS_I(d_inode(dentry));
802 uint64_t fakeinos, id;
803 uint64_t icount;
804 uint64_t ifree;
805 uint64_t fdblocks;
806 xfs_extlen_t lsize;
807 int64_t ffree;
808
809 /*
810 * Expedite background inodegc but don't wait. We do not want to block
811 * here waiting hours for a billion extent file to be truncated.
812 */
813 xfs_inodegc_push(mp);
814
815 statp->f_type = XFS_SUPER_MAGIC;
816 statp->f_namelen = MAXNAMELEN - 1;
817
818 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
819 statp->f_fsid = u64_to_fsid(id);
820
821 icount = percpu_counter_sum(&mp->m_icount);
822 ifree = percpu_counter_sum(&mp->m_ifree);
823 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
824
825 spin_lock(&mp->m_sb_lock);
826 statp->f_bsize = sbp->sb_blocksize;
827 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
828 statp->f_blocks = sbp->sb_dblocks - lsize;
829 spin_unlock(&mp->m_sb_lock);
830
831 /* make sure statp->f_bfree does not underflow */
832 statp->f_bfree = max_t(int64_t, 0,
833 fdblocks - xfs_fdblocks_unavailable(mp));
834 statp->f_bavail = statp->f_bfree;
835
836 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
837 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
838 if (M_IGEO(mp)->maxicount)
839 statp->f_files = min_t(typeof(statp->f_files),
840 statp->f_files,
841 M_IGEO(mp)->maxicount);
842
843 /* If sb_icount overshot maxicount, report actual allocation */
844 statp->f_files = max_t(typeof(statp->f_files),
845 statp->f_files,
846 sbp->sb_icount);
847
848 /* make sure statp->f_ffree does not underflow */
849 ffree = statp->f_files - (icount - ifree);
850 statp->f_ffree = max_t(int64_t, ffree, 0);
851
852
853 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
854 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
855 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
856 xfs_qm_statvfs(ip, statp);
857
858 if (XFS_IS_REALTIME_MOUNT(mp) &&
859 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
860 s64 freertx;
861
862 statp->f_blocks = sbp->sb_rblocks;
863 freertx = percpu_counter_sum_positive(&mp->m_frextents);
864 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
865 }
866
867 return 0;
868 }
869
870 STATIC void
xfs_save_resvblks(struct xfs_mount * mp)871 xfs_save_resvblks(struct xfs_mount *mp)
872 {
873 uint64_t resblks = 0;
874
875 mp->m_resblks_save = mp->m_resblks;
876 xfs_reserve_blocks(mp, &resblks, NULL);
877 }
878
879 STATIC void
xfs_restore_resvblks(struct xfs_mount * mp)880 xfs_restore_resvblks(struct xfs_mount *mp)
881 {
882 uint64_t resblks;
883
884 if (mp->m_resblks_save) {
885 resblks = mp->m_resblks_save;
886 mp->m_resblks_save = 0;
887 } else
888 resblks = xfs_default_resblks(mp);
889
890 xfs_reserve_blocks(mp, &resblks, NULL);
891 }
892
893 /*
894 * Second stage of a freeze. The data is already frozen so we only
895 * need to take care of the metadata. Once that's done sync the superblock
896 * to the log to dirty it in case of a crash while frozen. This ensures that we
897 * will recover the unlinked inode lists on the next mount.
898 */
899 STATIC int
xfs_fs_freeze(struct super_block * sb)900 xfs_fs_freeze(
901 struct super_block *sb)
902 {
903 struct xfs_mount *mp = XFS_M(sb);
904 unsigned int flags;
905 int ret;
906
907 /*
908 * The filesystem is now frozen far enough that memory reclaim
909 * cannot safely operate on the filesystem. Hence we need to
910 * set a GFP_NOFS context here to avoid recursion deadlocks.
911 */
912 flags = memalloc_nofs_save();
913 xfs_save_resvblks(mp);
914 ret = xfs_log_quiesce(mp);
915 memalloc_nofs_restore(flags);
916
917 /*
918 * For read-write filesystems, we need to restart the inodegc on error
919 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
920 * going to be run to restart it now. We are at SB_FREEZE_FS level
921 * here, so we can restart safely without racing with a stop in
922 * xfs_fs_sync_fs().
923 */
924 if (ret && !xfs_is_readonly(mp)) {
925 xfs_blockgc_start(mp);
926 xfs_inodegc_start(mp);
927 }
928
929 return ret;
930 }
931
932 STATIC int
xfs_fs_unfreeze(struct super_block * sb)933 xfs_fs_unfreeze(
934 struct super_block *sb)
935 {
936 struct xfs_mount *mp = XFS_M(sb);
937
938 xfs_restore_resvblks(mp);
939 xfs_log_work_queue(mp);
940
941 /*
942 * Don't reactivate the inodegc worker on a readonly filesystem because
943 * inodes are sent directly to reclaim. Don't reactivate the blockgc
944 * worker because there are no speculative preallocations on a readonly
945 * filesystem.
946 */
947 if (!xfs_is_readonly(mp)) {
948 xfs_blockgc_start(mp);
949 xfs_inodegc_start(mp);
950 }
951
952 return 0;
953 }
954
955 /*
956 * This function fills in xfs_mount_t fields based on mount args.
957 * Note: the superblock _has_ now been read in.
958 */
959 STATIC int
xfs_finish_flags(struct xfs_mount * mp)960 xfs_finish_flags(
961 struct xfs_mount *mp)
962 {
963 /* Fail a mount where the logbuf is smaller than the log stripe */
964 if (xfs_has_logv2(mp)) {
965 if (mp->m_logbsize <= 0 &&
966 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
967 mp->m_logbsize = mp->m_sb.sb_logsunit;
968 } else if (mp->m_logbsize > 0 &&
969 mp->m_logbsize < mp->m_sb.sb_logsunit) {
970 xfs_warn(mp,
971 "logbuf size must be greater than or equal to log stripe size");
972 return -EINVAL;
973 }
974 } else {
975 /* Fail a mount if the logbuf is larger than 32K */
976 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
977 xfs_warn(mp,
978 "logbuf size for version 1 logs must be 16K or 32K");
979 return -EINVAL;
980 }
981 }
982
983 /*
984 * V5 filesystems always use attr2 format for attributes.
985 */
986 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
987 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
988 "attr2 is always enabled for V5 filesystems.");
989 return -EINVAL;
990 }
991
992 /*
993 * prohibit r/w mounts of read-only filesystems
994 */
995 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
996 xfs_warn(mp,
997 "cannot mount a read-only filesystem as read-write");
998 return -EROFS;
999 }
1000
1001 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1002 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1003 !xfs_has_pquotino(mp)) {
1004 xfs_warn(mp,
1005 "Super block does not support project and group quota together");
1006 return -EINVAL;
1007 }
1008
1009 return 0;
1010 }
1011
1012 static int
xfs_init_percpu_counters(struct xfs_mount * mp)1013 xfs_init_percpu_counters(
1014 struct xfs_mount *mp)
1015 {
1016 int error;
1017
1018 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1019 if (error)
1020 return -ENOMEM;
1021
1022 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1023 if (error)
1024 goto free_icount;
1025
1026 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1027 if (error)
1028 goto free_ifree;
1029
1030 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1031 if (error)
1032 goto free_fdblocks;
1033
1034 error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1035 if (error)
1036 goto free_delalloc;
1037
1038 return 0;
1039
1040 free_delalloc:
1041 percpu_counter_destroy(&mp->m_delalloc_blks);
1042 free_fdblocks:
1043 percpu_counter_destroy(&mp->m_fdblocks);
1044 free_ifree:
1045 percpu_counter_destroy(&mp->m_ifree);
1046 free_icount:
1047 percpu_counter_destroy(&mp->m_icount);
1048 return -ENOMEM;
1049 }
1050
1051 void
xfs_reinit_percpu_counters(struct xfs_mount * mp)1052 xfs_reinit_percpu_counters(
1053 struct xfs_mount *mp)
1054 {
1055 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1056 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1057 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1058 percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1059 }
1060
1061 static void
xfs_destroy_percpu_counters(struct xfs_mount * mp)1062 xfs_destroy_percpu_counters(
1063 struct xfs_mount *mp)
1064 {
1065 percpu_counter_destroy(&mp->m_icount);
1066 percpu_counter_destroy(&mp->m_ifree);
1067 percpu_counter_destroy(&mp->m_fdblocks);
1068 ASSERT(xfs_is_shutdown(mp) ||
1069 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1070 percpu_counter_destroy(&mp->m_delalloc_blks);
1071 percpu_counter_destroy(&mp->m_frextents);
1072 }
1073
1074 static int
xfs_inodegc_init_percpu(struct xfs_mount * mp)1075 xfs_inodegc_init_percpu(
1076 struct xfs_mount *mp)
1077 {
1078 struct xfs_inodegc *gc;
1079 int cpu;
1080
1081 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1082 if (!mp->m_inodegc)
1083 return -ENOMEM;
1084
1085 for_each_possible_cpu(cpu) {
1086 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1087 #if defined(DEBUG) || defined(XFS_WARN)
1088 gc->cpu = cpu;
1089 #endif
1090 init_llist_head(&gc->list);
1091 gc->items = 0;
1092 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1093 }
1094 return 0;
1095 }
1096
1097 static void
xfs_inodegc_free_percpu(struct xfs_mount * mp)1098 xfs_inodegc_free_percpu(
1099 struct xfs_mount *mp)
1100 {
1101 if (!mp->m_inodegc)
1102 return;
1103 free_percpu(mp->m_inodegc);
1104 }
1105
1106 static void
xfs_fs_put_super(struct super_block * sb)1107 xfs_fs_put_super(
1108 struct super_block *sb)
1109 {
1110 struct xfs_mount *mp = XFS_M(sb);
1111
1112 /* if ->fill_super failed, we have no mount to tear down */
1113 if (!sb->s_fs_info)
1114 return;
1115
1116 xfs_notice(mp, "Unmounting Filesystem");
1117 xfs_filestream_unmount(mp);
1118 xfs_unmountfs(mp);
1119
1120 xfs_freesb(mp);
1121 free_percpu(mp->m_stats.xs_stats);
1122 xfs_mount_list_del(mp);
1123 xfs_inodegc_free_percpu(mp);
1124 xfs_destroy_percpu_counters(mp);
1125 xfs_destroy_mount_workqueues(mp);
1126 xfs_close_devices(mp);
1127
1128 sb->s_fs_info = NULL;
1129 xfs_mount_free(mp);
1130 }
1131
1132 static long
xfs_fs_nr_cached_objects(struct super_block * sb,struct shrink_control * sc)1133 xfs_fs_nr_cached_objects(
1134 struct super_block *sb,
1135 struct shrink_control *sc)
1136 {
1137 /* Paranoia: catch incorrect calls during mount setup or teardown */
1138 if (WARN_ON_ONCE(!sb->s_fs_info))
1139 return 0;
1140 return xfs_reclaim_inodes_count(XFS_M(sb));
1141 }
1142
1143 static long
xfs_fs_free_cached_objects(struct super_block * sb,struct shrink_control * sc)1144 xfs_fs_free_cached_objects(
1145 struct super_block *sb,
1146 struct shrink_control *sc)
1147 {
1148 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1149 }
1150
1151 static const struct super_operations xfs_super_operations = {
1152 .alloc_inode = xfs_fs_alloc_inode,
1153 .destroy_inode = xfs_fs_destroy_inode,
1154 .dirty_inode = xfs_fs_dirty_inode,
1155 .drop_inode = xfs_fs_drop_inode,
1156 .put_super = xfs_fs_put_super,
1157 .sync_fs = xfs_fs_sync_fs,
1158 .freeze_fs = xfs_fs_freeze,
1159 .unfreeze_fs = xfs_fs_unfreeze,
1160 .statfs = xfs_fs_statfs,
1161 .show_options = xfs_fs_show_options,
1162 .nr_cached_objects = xfs_fs_nr_cached_objects,
1163 .free_cached_objects = xfs_fs_free_cached_objects,
1164 };
1165
1166 static int
suffix_kstrtoint(const char * s,unsigned int base,int * res)1167 suffix_kstrtoint(
1168 const char *s,
1169 unsigned int base,
1170 int *res)
1171 {
1172 int last, shift_left_factor = 0, _res;
1173 char *value;
1174 int ret = 0;
1175
1176 value = kstrdup(s, GFP_KERNEL);
1177 if (!value)
1178 return -ENOMEM;
1179
1180 last = strlen(value) - 1;
1181 if (value[last] == 'K' || value[last] == 'k') {
1182 shift_left_factor = 10;
1183 value[last] = '\0';
1184 }
1185 if (value[last] == 'M' || value[last] == 'm') {
1186 shift_left_factor = 20;
1187 value[last] = '\0';
1188 }
1189 if (value[last] == 'G' || value[last] == 'g') {
1190 shift_left_factor = 30;
1191 value[last] = '\0';
1192 }
1193
1194 if (kstrtoint(value, base, &_res))
1195 ret = -EINVAL;
1196 kfree(value);
1197 *res = _res << shift_left_factor;
1198 return ret;
1199 }
1200
1201 static inline void
xfs_fs_warn_deprecated(struct fs_context * fc,struct fs_parameter * param,uint64_t flag,bool value)1202 xfs_fs_warn_deprecated(
1203 struct fs_context *fc,
1204 struct fs_parameter *param,
1205 uint64_t flag,
1206 bool value)
1207 {
1208 /* Don't print the warning if reconfiguring and current mount point
1209 * already had the flag set
1210 */
1211 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1212 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1213 return;
1214 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1215 }
1216
1217 /*
1218 * Set mount state from a mount option.
1219 *
1220 * NOTE: mp->m_super is NULL here!
1221 */
1222 static int
xfs_fs_parse_param(struct fs_context * fc,struct fs_parameter * param)1223 xfs_fs_parse_param(
1224 struct fs_context *fc,
1225 struct fs_parameter *param)
1226 {
1227 struct xfs_mount *parsing_mp = fc->s_fs_info;
1228 struct fs_parse_result result;
1229 int size = 0;
1230 int opt;
1231
1232 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1233 if (opt < 0)
1234 return opt;
1235
1236 switch (opt) {
1237 case Opt_logbufs:
1238 parsing_mp->m_logbufs = result.uint_32;
1239 return 0;
1240 case Opt_logbsize:
1241 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1242 return -EINVAL;
1243 return 0;
1244 case Opt_logdev:
1245 kfree(parsing_mp->m_logname);
1246 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1247 if (!parsing_mp->m_logname)
1248 return -ENOMEM;
1249 return 0;
1250 case Opt_rtdev:
1251 kfree(parsing_mp->m_rtname);
1252 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1253 if (!parsing_mp->m_rtname)
1254 return -ENOMEM;
1255 return 0;
1256 case Opt_allocsize:
1257 if (suffix_kstrtoint(param->string, 10, &size))
1258 return -EINVAL;
1259 parsing_mp->m_allocsize_log = ffs(size) - 1;
1260 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1261 return 0;
1262 case Opt_grpid:
1263 case Opt_bsdgroups:
1264 parsing_mp->m_features |= XFS_FEAT_GRPID;
1265 return 0;
1266 case Opt_nogrpid:
1267 case Opt_sysvgroups:
1268 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1269 return 0;
1270 case Opt_wsync:
1271 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1272 return 0;
1273 case Opt_norecovery:
1274 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1275 return 0;
1276 case Opt_noalign:
1277 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1278 return 0;
1279 case Opt_swalloc:
1280 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1281 return 0;
1282 case Opt_sunit:
1283 parsing_mp->m_dalign = result.uint_32;
1284 return 0;
1285 case Opt_swidth:
1286 parsing_mp->m_swidth = result.uint_32;
1287 return 0;
1288 case Opt_inode32:
1289 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1290 return 0;
1291 case Opt_inode64:
1292 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1293 return 0;
1294 case Opt_nouuid:
1295 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1296 return 0;
1297 case Opt_largeio:
1298 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1299 return 0;
1300 case Opt_nolargeio:
1301 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1302 return 0;
1303 case Opt_filestreams:
1304 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1305 return 0;
1306 case Opt_noquota:
1307 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1308 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1309 return 0;
1310 case Opt_quota:
1311 case Opt_uquota:
1312 case Opt_usrquota:
1313 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1314 return 0;
1315 case Opt_qnoenforce:
1316 case Opt_uqnoenforce:
1317 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1318 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1319 return 0;
1320 case Opt_pquota:
1321 case Opt_prjquota:
1322 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1323 return 0;
1324 case Opt_pqnoenforce:
1325 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1326 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1327 return 0;
1328 case Opt_gquota:
1329 case Opt_grpquota:
1330 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1331 return 0;
1332 case Opt_gqnoenforce:
1333 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1334 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1335 return 0;
1336 case Opt_discard:
1337 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1338 return 0;
1339 case Opt_nodiscard:
1340 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1341 return 0;
1342 #ifdef CONFIG_FS_DAX
1343 case Opt_dax:
1344 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1345 return 0;
1346 case Opt_dax_enum:
1347 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1348 return 0;
1349 #endif
1350 /* Following mount options will be removed in September 2025 */
1351 case Opt_ikeep:
1352 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1353 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1354 return 0;
1355 case Opt_noikeep:
1356 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1357 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1358 return 0;
1359 case Opt_attr2:
1360 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1361 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1362 return 0;
1363 case Opt_noattr2:
1364 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1365 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1366 return 0;
1367 default:
1368 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1369 return -EINVAL;
1370 }
1371
1372 return 0;
1373 }
1374
1375 static int
xfs_fs_validate_params(struct xfs_mount * mp)1376 xfs_fs_validate_params(
1377 struct xfs_mount *mp)
1378 {
1379 /* No recovery flag requires a read-only mount */
1380 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1381 xfs_warn(mp, "no-recovery mounts must be read-only.");
1382 return -EINVAL;
1383 }
1384
1385 /*
1386 * We have not read the superblock at this point, so only the attr2
1387 * mount option can set the attr2 feature by this stage.
1388 */
1389 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1390 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1391 return -EINVAL;
1392 }
1393
1394
1395 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1396 xfs_warn(mp,
1397 "sunit and swidth options incompatible with the noalign option");
1398 return -EINVAL;
1399 }
1400
1401 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1402 xfs_warn(mp, "quota support not available in this kernel.");
1403 return -EINVAL;
1404 }
1405
1406 if ((mp->m_dalign && !mp->m_swidth) ||
1407 (!mp->m_dalign && mp->m_swidth)) {
1408 xfs_warn(mp, "sunit and swidth must be specified together");
1409 return -EINVAL;
1410 }
1411
1412 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1413 xfs_warn(mp,
1414 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1415 mp->m_swidth, mp->m_dalign);
1416 return -EINVAL;
1417 }
1418
1419 if (mp->m_logbufs != -1 &&
1420 mp->m_logbufs != 0 &&
1421 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1422 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1423 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1424 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1425 return -EINVAL;
1426 }
1427
1428 if (mp->m_logbsize != -1 &&
1429 mp->m_logbsize != 0 &&
1430 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1431 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1432 !is_power_of_2(mp->m_logbsize))) {
1433 xfs_warn(mp,
1434 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1435 mp->m_logbsize);
1436 return -EINVAL;
1437 }
1438
1439 if (xfs_has_allocsize(mp) &&
1440 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1441 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1442 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1443 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1444 return -EINVAL;
1445 }
1446
1447 return 0;
1448 }
1449
1450 static int
xfs_fs_fill_super(struct super_block * sb,struct fs_context * fc)1451 xfs_fs_fill_super(
1452 struct super_block *sb,
1453 struct fs_context *fc)
1454 {
1455 struct xfs_mount *mp = sb->s_fs_info;
1456 struct inode *root;
1457 int flags = 0, error;
1458
1459 mp->m_super = sb;
1460
1461 /*
1462 * Copy VFS mount flags from the context now that all parameter parsing
1463 * is guaranteed to have been completed by either the old mount API or
1464 * the newer fsopen/fsconfig API.
1465 */
1466 if (fc->sb_flags & SB_RDONLY)
1467 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1468 if (fc->sb_flags & SB_DIRSYNC)
1469 mp->m_features |= XFS_FEAT_DIRSYNC;
1470 if (fc->sb_flags & SB_SYNCHRONOUS)
1471 mp->m_features |= XFS_FEAT_WSYNC;
1472
1473 error = xfs_fs_validate_params(mp);
1474 if (error)
1475 goto out_free_names;
1476
1477 sb_min_blocksize(sb, BBSIZE);
1478 sb->s_xattr = xfs_xattr_handlers;
1479 sb->s_export_op = &xfs_export_operations;
1480 #ifdef CONFIG_XFS_QUOTA
1481 sb->s_qcop = &xfs_quotactl_operations;
1482 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1483 #endif
1484 sb->s_op = &xfs_super_operations;
1485
1486 /*
1487 * Delay mount work if the debug hook is set. This is debug
1488 * instrumention to coordinate simulation of xfs mount failures with
1489 * VFS superblock operations
1490 */
1491 if (xfs_globals.mount_delay) {
1492 xfs_notice(mp, "Delaying mount for %d seconds.",
1493 xfs_globals.mount_delay);
1494 msleep(xfs_globals.mount_delay * 1000);
1495 }
1496
1497 if (fc->sb_flags & SB_SILENT)
1498 flags |= XFS_MFSI_QUIET;
1499
1500 error = xfs_open_devices(mp);
1501 if (error)
1502 goto out_free_names;
1503
1504 error = xfs_init_mount_workqueues(mp);
1505 if (error)
1506 goto out_close_devices;
1507
1508 error = xfs_init_percpu_counters(mp);
1509 if (error)
1510 goto out_destroy_workqueues;
1511
1512 error = xfs_inodegc_init_percpu(mp);
1513 if (error)
1514 goto out_destroy_counters;
1515
1516 /*
1517 * All percpu data structures requiring cleanup when a cpu goes offline
1518 * must be allocated before adding this @mp to the cpu-dead handler's
1519 * mount list.
1520 */
1521 xfs_mount_list_add(mp);
1522
1523 /* Allocate stats memory before we do operations that might use it */
1524 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1525 if (!mp->m_stats.xs_stats) {
1526 error = -ENOMEM;
1527 goto out_destroy_inodegc;
1528 }
1529
1530 error = xfs_readsb(mp, flags);
1531 if (error)
1532 goto out_free_stats;
1533
1534 error = xfs_finish_flags(mp);
1535 if (error)
1536 goto out_free_sb;
1537
1538 error = xfs_setup_devices(mp);
1539 if (error)
1540 goto out_free_sb;
1541
1542 /* V4 support is undergoing deprecation. */
1543 if (!xfs_has_crc(mp)) {
1544 #ifdef CONFIG_XFS_SUPPORT_V4
1545 xfs_warn_once(mp,
1546 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1547 #else
1548 xfs_warn(mp,
1549 "Deprecated V4 format (crc=0) not supported by kernel.");
1550 error = -EINVAL;
1551 goto out_free_sb;
1552 #endif
1553 }
1554
1555 /* Filesystem claims it needs repair, so refuse the mount. */
1556 if (xfs_has_needsrepair(mp)) {
1557 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1558 error = -EFSCORRUPTED;
1559 goto out_free_sb;
1560 }
1561
1562 /*
1563 * Don't touch the filesystem if a user tool thinks it owns the primary
1564 * superblock. mkfs doesn't clear the flag from secondary supers, so
1565 * we don't check them at all.
1566 */
1567 if (mp->m_sb.sb_inprogress) {
1568 xfs_warn(mp, "Offline file system operation in progress!");
1569 error = -EFSCORRUPTED;
1570 goto out_free_sb;
1571 }
1572
1573 /*
1574 * Until this is fixed only page-sized or smaller data blocks work.
1575 */
1576 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1577 xfs_warn(mp,
1578 "File system with blocksize %d bytes. "
1579 "Only pagesize (%ld) or less will currently work.",
1580 mp->m_sb.sb_blocksize, PAGE_SIZE);
1581 error = -ENOSYS;
1582 goto out_free_sb;
1583 }
1584
1585 /* Ensure this filesystem fits in the page cache limits */
1586 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1587 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1588 xfs_warn(mp,
1589 "file system too large to be mounted on this system.");
1590 error = -EFBIG;
1591 goto out_free_sb;
1592 }
1593
1594 /*
1595 * XFS block mappings use 54 bits to store the logical block offset.
1596 * This should suffice to handle the maximum file size that the VFS
1597 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1598 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1599 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1600 * to check this assertion.
1601 *
1602 * Avoid integer overflow by comparing the maximum bmbt offset to the
1603 * maximum pagecache offset in units of fs blocks.
1604 */
1605 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1606 xfs_warn(mp,
1607 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1608 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1609 XFS_MAX_FILEOFF);
1610 error = -EINVAL;
1611 goto out_free_sb;
1612 }
1613
1614 error = xfs_filestream_mount(mp);
1615 if (error)
1616 goto out_free_sb;
1617
1618 /*
1619 * we must configure the block size in the superblock before we run the
1620 * full mount process as the mount process can lookup and cache inodes.
1621 */
1622 sb->s_magic = XFS_SUPER_MAGIC;
1623 sb->s_blocksize = mp->m_sb.sb_blocksize;
1624 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1625 sb->s_maxbytes = MAX_LFS_FILESIZE;
1626 sb->s_max_links = XFS_MAXLINK;
1627 sb->s_time_gran = 1;
1628 if (xfs_has_bigtime(mp)) {
1629 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1630 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1631 } else {
1632 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1633 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1634 }
1635 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1636 sb->s_iflags |= SB_I_CGROUPWB;
1637
1638 set_posix_acl_flag(sb);
1639
1640 /* version 5 superblocks support inode version counters. */
1641 if (xfs_has_crc(mp))
1642 sb->s_flags |= SB_I_VERSION;
1643
1644 if (xfs_has_dax_always(mp)) {
1645 error = xfs_setup_dax_always(mp);
1646 if (error)
1647 goto out_filestream_unmount;
1648 }
1649
1650 if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1651 xfs_warn(mp,
1652 "mounting with \"discard\" option, but the device does not support discard");
1653 mp->m_features &= ~XFS_FEAT_DISCARD;
1654 }
1655
1656 if (xfs_has_reflink(mp)) {
1657 if (mp->m_sb.sb_rblocks) {
1658 xfs_alert(mp,
1659 "reflink not compatible with realtime device!");
1660 error = -EINVAL;
1661 goto out_filestream_unmount;
1662 }
1663
1664 if (xfs_globals.always_cow) {
1665 xfs_info(mp, "using DEBUG-only always_cow mode.");
1666 mp->m_always_cow = true;
1667 }
1668 }
1669
1670 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1671 xfs_alert(mp,
1672 "reverse mapping btree not compatible with realtime device!");
1673 error = -EINVAL;
1674 goto out_filestream_unmount;
1675 }
1676
1677 if (xfs_has_large_extent_counts(mp))
1678 xfs_warn(mp,
1679 "EXPERIMENTAL Large extent counts feature in use. Use at your own risk!");
1680
1681 error = xfs_mountfs(mp);
1682 if (error)
1683 goto out_filestream_unmount;
1684
1685 root = igrab(VFS_I(mp->m_rootip));
1686 if (!root) {
1687 error = -ENOENT;
1688 goto out_unmount;
1689 }
1690 sb->s_root = d_make_root(root);
1691 if (!sb->s_root) {
1692 error = -ENOMEM;
1693 goto out_unmount;
1694 }
1695
1696 return 0;
1697
1698 out_filestream_unmount:
1699 xfs_filestream_unmount(mp);
1700 out_free_sb:
1701 xfs_freesb(mp);
1702 out_free_stats:
1703 free_percpu(mp->m_stats.xs_stats);
1704 out_destroy_inodegc:
1705 xfs_mount_list_del(mp);
1706 xfs_inodegc_free_percpu(mp);
1707 out_destroy_counters:
1708 xfs_destroy_percpu_counters(mp);
1709 out_destroy_workqueues:
1710 xfs_destroy_mount_workqueues(mp);
1711 out_close_devices:
1712 xfs_close_devices(mp);
1713 out_free_names:
1714 sb->s_fs_info = NULL;
1715 xfs_mount_free(mp);
1716 return error;
1717
1718 out_unmount:
1719 xfs_filestream_unmount(mp);
1720 xfs_unmountfs(mp);
1721 goto out_free_sb;
1722 }
1723
1724 static int
xfs_fs_get_tree(struct fs_context * fc)1725 xfs_fs_get_tree(
1726 struct fs_context *fc)
1727 {
1728 return get_tree_bdev(fc, xfs_fs_fill_super);
1729 }
1730
1731 static int
xfs_remount_rw(struct xfs_mount * mp)1732 xfs_remount_rw(
1733 struct xfs_mount *mp)
1734 {
1735 struct xfs_sb *sbp = &mp->m_sb;
1736 int error;
1737
1738 if (xfs_has_norecovery(mp)) {
1739 xfs_warn(mp,
1740 "ro->rw transition prohibited on norecovery mount");
1741 return -EINVAL;
1742 }
1743
1744 if (xfs_sb_is_v5(sbp) &&
1745 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1746 xfs_warn(mp,
1747 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1748 (sbp->sb_features_ro_compat &
1749 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1750 return -EINVAL;
1751 }
1752
1753 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1754
1755 /*
1756 * If this is the first remount to writeable state we might have some
1757 * superblock changes to update.
1758 */
1759 if (mp->m_update_sb) {
1760 error = xfs_sync_sb(mp, false);
1761 if (error) {
1762 xfs_warn(mp, "failed to write sb changes");
1763 return error;
1764 }
1765 mp->m_update_sb = false;
1766 }
1767
1768 /*
1769 * Fill out the reserve pool if it is empty. Use the stashed value if
1770 * it is non-zero, otherwise go with the default.
1771 */
1772 xfs_restore_resvblks(mp);
1773 xfs_log_work_queue(mp);
1774 xfs_blockgc_start(mp);
1775
1776 /* Create the per-AG metadata reservation pool .*/
1777 error = xfs_fs_reserve_ag_blocks(mp);
1778 if (error && error != -ENOSPC)
1779 return error;
1780
1781 /* Re-enable the background inode inactivation worker. */
1782 xfs_inodegc_start(mp);
1783
1784 return 0;
1785 }
1786
1787 static int
xfs_remount_ro(struct xfs_mount * mp)1788 xfs_remount_ro(
1789 struct xfs_mount *mp)
1790 {
1791 struct xfs_icwalk icw = {
1792 .icw_flags = XFS_ICWALK_FLAG_SYNC,
1793 };
1794 int error;
1795
1796 /* Flush all the dirty data to disk. */
1797 error = sync_filesystem(mp->m_super);
1798 if (error)
1799 return error;
1800
1801 /*
1802 * Cancel background eofb scanning so it cannot race with the final
1803 * log force+buftarg wait and deadlock the remount.
1804 */
1805 xfs_blockgc_stop(mp);
1806
1807 /*
1808 * Clear out all remaining COW staging extents and speculative post-EOF
1809 * preallocations so that we don't leave inodes requiring inactivation
1810 * cleanups during reclaim on a read-only mount. We must process every
1811 * cached inode, so this requires a synchronous cache scan.
1812 */
1813 error = xfs_blockgc_free_space(mp, &icw);
1814 if (error) {
1815 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1816 return error;
1817 }
1818
1819 /*
1820 * Stop the inodegc background worker. xfs_fs_reconfigure already
1821 * flushed all pending inodegc work when it sync'd the filesystem.
1822 * The VFS holds s_umount, so we know that inodes cannot enter
1823 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1824 * we send inodes straight to reclaim, so no inodes will be queued.
1825 */
1826 xfs_inodegc_stop(mp);
1827
1828 /* Free the per-AG metadata reservation pool. */
1829 error = xfs_fs_unreserve_ag_blocks(mp);
1830 if (error) {
1831 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1832 return error;
1833 }
1834
1835 /*
1836 * Before we sync the metadata, we need to free up the reserve block
1837 * pool so that the used block count in the superblock on disk is
1838 * correct at the end of the remount. Stash the current* reserve pool
1839 * size so that if we get remounted rw, we can return it to the same
1840 * size.
1841 */
1842 xfs_save_resvblks(mp);
1843
1844 xfs_log_clean(mp);
1845 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1846
1847 return 0;
1848 }
1849
1850 /*
1851 * Logically we would return an error here to prevent users from believing
1852 * they might have changed mount options using remount which can't be changed.
1853 *
1854 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1855 * arguments in some cases so we can't blindly reject options, but have to
1856 * check for each specified option if it actually differs from the currently
1857 * set option and only reject it if that's the case.
1858 *
1859 * Until that is implemented we return success for every remount request, and
1860 * silently ignore all options that we can't actually change.
1861 */
1862 static int
xfs_fs_reconfigure(struct fs_context * fc)1863 xfs_fs_reconfigure(
1864 struct fs_context *fc)
1865 {
1866 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1867 struct xfs_mount *new_mp = fc->s_fs_info;
1868 int flags = fc->sb_flags;
1869 int error;
1870
1871 /* version 5 superblocks always support version counters. */
1872 if (xfs_has_crc(mp))
1873 fc->sb_flags |= SB_I_VERSION;
1874
1875 error = xfs_fs_validate_params(new_mp);
1876 if (error)
1877 return error;
1878
1879 /* inode32 -> inode64 */
1880 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1881 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1882 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1883 }
1884
1885 /* inode64 -> inode32 */
1886 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1887 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1888 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1889 }
1890
1891 /* ro -> rw */
1892 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1893 error = xfs_remount_rw(mp);
1894 if (error)
1895 return error;
1896 }
1897
1898 /* rw -> ro */
1899 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1900 error = xfs_remount_ro(mp);
1901 if (error)
1902 return error;
1903 }
1904
1905 return 0;
1906 }
1907
xfs_fs_free(struct fs_context * fc)1908 static void xfs_fs_free(
1909 struct fs_context *fc)
1910 {
1911 struct xfs_mount *mp = fc->s_fs_info;
1912
1913 /*
1914 * mp is stored in the fs_context when it is initialized.
1915 * mp is transferred to the superblock on a successful mount,
1916 * but if an error occurs before the transfer we have to free
1917 * it here.
1918 */
1919 if (mp)
1920 xfs_mount_free(mp);
1921 }
1922
1923 static const struct fs_context_operations xfs_context_ops = {
1924 .parse_param = xfs_fs_parse_param,
1925 .get_tree = xfs_fs_get_tree,
1926 .reconfigure = xfs_fs_reconfigure,
1927 .free = xfs_fs_free,
1928 };
1929
1930 /*
1931 * WARNING: do not initialise any parameters in this function that depend on
1932 * mount option parsing having already been performed as this can be called from
1933 * fsopen() before any parameters have been set.
1934 */
xfs_init_fs_context(struct fs_context * fc)1935 static int xfs_init_fs_context(
1936 struct fs_context *fc)
1937 {
1938 struct xfs_mount *mp;
1939
1940 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1941 if (!mp)
1942 return -ENOMEM;
1943
1944 spin_lock_init(&mp->m_sb_lock);
1945 spin_lock_init(&mp->m_agirotor_lock);
1946 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
1947 spin_lock_init(&mp->m_perag_lock);
1948 mutex_init(&mp->m_growlock);
1949 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
1950 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
1951 mp->m_kobj.kobject.kset = xfs_kset;
1952 /*
1953 * We don't create the finobt per-ag space reservation until after log
1954 * recovery, so we must set this to true so that an ifree transaction
1955 * started during log recovery will not depend on space reservations
1956 * for finobt expansion.
1957 */
1958 mp->m_finobt_nores = true;
1959
1960 /*
1961 * These can be overridden by the mount option parsing.
1962 */
1963 mp->m_logbufs = -1;
1964 mp->m_logbsize = -1;
1965 mp->m_allocsize_log = 16; /* 64k */
1966
1967 fc->s_fs_info = mp;
1968 fc->ops = &xfs_context_ops;
1969
1970 return 0;
1971 }
1972
1973 static struct file_system_type xfs_fs_type = {
1974 .owner = THIS_MODULE,
1975 .name = "xfs",
1976 .init_fs_context = xfs_init_fs_context,
1977 .parameters = xfs_fs_parameters,
1978 .kill_sb = kill_block_super,
1979 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
1980 };
1981 MODULE_ALIAS_FS("xfs");
1982
1983 STATIC int __init
xfs_init_caches(void)1984 xfs_init_caches(void)
1985 {
1986 int error;
1987
1988 xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
1989 SLAB_HWCACHE_ALIGN |
1990 SLAB_RECLAIM_ACCOUNT |
1991 SLAB_MEM_SPREAD,
1992 NULL);
1993 if (!xfs_buf_cache)
1994 goto out;
1995
1996 xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
1997 sizeof(struct xlog_ticket),
1998 0, 0, NULL);
1999 if (!xfs_log_ticket_cache)
2000 goto out_destroy_buf_cache;
2001
2002 error = xfs_btree_init_cur_caches();
2003 if (error)
2004 goto out_destroy_log_ticket_cache;
2005
2006 error = xfs_defer_init_item_caches();
2007 if (error)
2008 goto out_destroy_btree_cur_cache;
2009
2010 xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2011 sizeof(struct xfs_da_state),
2012 0, 0, NULL);
2013 if (!xfs_da_state_cache)
2014 goto out_destroy_defer_item_cache;
2015
2016 xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2017 sizeof(struct xfs_ifork),
2018 0, 0, NULL);
2019 if (!xfs_ifork_cache)
2020 goto out_destroy_da_state_cache;
2021
2022 xfs_trans_cache = kmem_cache_create("xfs_trans",
2023 sizeof(struct xfs_trans),
2024 0, 0, NULL);
2025 if (!xfs_trans_cache)
2026 goto out_destroy_ifork_cache;
2027
2028
2029 /*
2030 * The size of the cache-allocated buf log item is the maximum
2031 * size possible under XFS. This wastes a little bit of memory,
2032 * but it is much faster.
2033 */
2034 xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2035 sizeof(struct xfs_buf_log_item),
2036 0, 0, NULL);
2037 if (!xfs_buf_item_cache)
2038 goto out_destroy_trans_cache;
2039
2040 xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2041 xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2042 0, 0, NULL);
2043 if (!xfs_efd_cache)
2044 goto out_destroy_buf_item_cache;
2045
2046 xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2047 xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2048 0, 0, NULL);
2049 if (!xfs_efi_cache)
2050 goto out_destroy_efd_cache;
2051
2052 xfs_inode_cache = kmem_cache_create("xfs_inode",
2053 sizeof(struct xfs_inode), 0,
2054 (SLAB_HWCACHE_ALIGN |
2055 SLAB_RECLAIM_ACCOUNT |
2056 SLAB_MEM_SPREAD | SLAB_ACCOUNT),
2057 xfs_fs_inode_init_once);
2058 if (!xfs_inode_cache)
2059 goto out_destroy_efi_cache;
2060
2061 xfs_ili_cache = kmem_cache_create("xfs_ili",
2062 sizeof(struct xfs_inode_log_item), 0,
2063 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
2064 NULL);
2065 if (!xfs_ili_cache)
2066 goto out_destroy_inode_cache;
2067
2068 xfs_icreate_cache = kmem_cache_create("xfs_icr",
2069 sizeof(struct xfs_icreate_item),
2070 0, 0, NULL);
2071 if (!xfs_icreate_cache)
2072 goto out_destroy_ili_cache;
2073
2074 xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2075 sizeof(struct xfs_rud_log_item),
2076 0, 0, NULL);
2077 if (!xfs_rud_cache)
2078 goto out_destroy_icreate_cache;
2079
2080 xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2081 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2082 0, 0, NULL);
2083 if (!xfs_rui_cache)
2084 goto out_destroy_rud_cache;
2085
2086 xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2087 sizeof(struct xfs_cud_log_item),
2088 0, 0, NULL);
2089 if (!xfs_cud_cache)
2090 goto out_destroy_rui_cache;
2091
2092 xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2093 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2094 0, 0, NULL);
2095 if (!xfs_cui_cache)
2096 goto out_destroy_cud_cache;
2097
2098 xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2099 sizeof(struct xfs_bud_log_item),
2100 0, 0, NULL);
2101 if (!xfs_bud_cache)
2102 goto out_destroy_cui_cache;
2103
2104 xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2105 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2106 0, 0, NULL);
2107 if (!xfs_bui_cache)
2108 goto out_destroy_bud_cache;
2109
2110 xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2111 sizeof(struct xfs_attrd_log_item),
2112 0, 0, NULL);
2113 if (!xfs_attrd_cache)
2114 goto out_destroy_bui_cache;
2115
2116 xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2117 sizeof(struct xfs_attri_log_item),
2118 0, 0, NULL);
2119 if (!xfs_attri_cache)
2120 goto out_destroy_attrd_cache;
2121
2122 xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2123 sizeof(struct xfs_iunlink_item),
2124 0, 0, NULL);
2125 if (!xfs_iunlink_cache)
2126 goto out_destroy_attri_cache;
2127
2128 return 0;
2129
2130 out_destroy_attri_cache:
2131 kmem_cache_destroy(xfs_attri_cache);
2132 out_destroy_attrd_cache:
2133 kmem_cache_destroy(xfs_attrd_cache);
2134 out_destroy_bui_cache:
2135 kmem_cache_destroy(xfs_bui_cache);
2136 out_destroy_bud_cache:
2137 kmem_cache_destroy(xfs_bud_cache);
2138 out_destroy_cui_cache:
2139 kmem_cache_destroy(xfs_cui_cache);
2140 out_destroy_cud_cache:
2141 kmem_cache_destroy(xfs_cud_cache);
2142 out_destroy_rui_cache:
2143 kmem_cache_destroy(xfs_rui_cache);
2144 out_destroy_rud_cache:
2145 kmem_cache_destroy(xfs_rud_cache);
2146 out_destroy_icreate_cache:
2147 kmem_cache_destroy(xfs_icreate_cache);
2148 out_destroy_ili_cache:
2149 kmem_cache_destroy(xfs_ili_cache);
2150 out_destroy_inode_cache:
2151 kmem_cache_destroy(xfs_inode_cache);
2152 out_destroy_efi_cache:
2153 kmem_cache_destroy(xfs_efi_cache);
2154 out_destroy_efd_cache:
2155 kmem_cache_destroy(xfs_efd_cache);
2156 out_destroy_buf_item_cache:
2157 kmem_cache_destroy(xfs_buf_item_cache);
2158 out_destroy_trans_cache:
2159 kmem_cache_destroy(xfs_trans_cache);
2160 out_destroy_ifork_cache:
2161 kmem_cache_destroy(xfs_ifork_cache);
2162 out_destroy_da_state_cache:
2163 kmem_cache_destroy(xfs_da_state_cache);
2164 out_destroy_defer_item_cache:
2165 xfs_defer_destroy_item_caches();
2166 out_destroy_btree_cur_cache:
2167 xfs_btree_destroy_cur_caches();
2168 out_destroy_log_ticket_cache:
2169 kmem_cache_destroy(xfs_log_ticket_cache);
2170 out_destroy_buf_cache:
2171 kmem_cache_destroy(xfs_buf_cache);
2172 out:
2173 return -ENOMEM;
2174 }
2175
2176 STATIC void
xfs_destroy_caches(void)2177 xfs_destroy_caches(void)
2178 {
2179 /*
2180 * Make sure all delayed rcu free are flushed before we
2181 * destroy caches.
2182 */
2183 rcu_barrier();
2184 kmem_cache_destroy(xfs_iunlink_cache);
2185 kmem_cache_destroy(xfs_attri_cache);
2186 kmem_cache_destroy(xfs_attrd_cache);
2187 kmem_cache_destroy(xfs_bui_cache);
2188 kmem_cache_destroy(xfs_bud_cache);
2189 kmem_cache_destroy(xfs_cui_cache);
2190 kmem_cache_destroy(xfs_cud_cache);
2191 kmem_cache_destroy(xfs_rui_cache);
2192 kmem_cache_destroy(xfs_rud_cache);
2193 kmem_cache_destroy(xfs_icreate_cache);
2194 kmem_cache_destroy(xfs_ili_cache);
2195 kmem_cache_destroy(xfs_inode_cache);
2196 kmem_cache_destroy(xfs_efi_cache);
2197 kmem_cache_destroy(xfs_efd_cache);
2198 kmem_cache_destroy(xfs_buf_item_cache);
2199 kmem_cache_destroy(xfs_trans_cache);
2200 kmem_cache_destroy(xfs_ifork_cache);
2201 kmem_cache_destroy(xfs_da_state_cache);
2202 xfs_defer_destroy_item_caches();
2203 xfs_btree_destroy_cur_caches();
2204 kmem_cache_destroy(xfs_log_ticket_cache);
2205 kmem_cache_destroy(xfs_buf_cache);
2206 }
2207
2208 STATIC int __init
xfs_init_workqueues(void)2209 xfs_init_workqueues(void)
2210 {
2211 /*
2212 * The allocation workqueue can be used in memory reclaim situations
2213 * (writepage path), and parallelism is only limited by the number of
2214 * AGs in all the filesystems mounted. Hence use the default large
2215 * max_active value for this workqueue.
2216 */
2217 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2218 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2219 if (!xfs_alloc_wq)
2220 return -ENOMEM;
2221
2222 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2223 0);
2224 if (!xfs_discard_wq)
2225 goto out_free_alloc_wq;
2226
2227 return 0;
2228 out_free_alloc_wq:
2229 destroy_workqueue(xfs_alloc_wq);
2230 return -ENOMEM;
2231 }
2232
2233 STATIC void
xfs_destroy_workqueues(void)2234 xfs_destroy_workqueues(void)
2235 {
2236 destroy_workqueue(xfs_discard_wq);
2237 destroy_workqueue(xfs_alloc_wq);
2238 }
2239
2240 #ifdef CONFIG_HOTPLUG_CPU
2241 static int
xfs_cpu_dead(unsigned int cpu)2242 xfs_cpu_dead(
2243 unsigned int cpu)
2244 {
2245 struct xfs_mount *mp, *n;
2246
2247 spin_lock(&xfs_mount_list_lock);
2248 list_for_each_entry_safe(mp, n, &xfs_mount_list, m_mount_list) {
2249 spin_unlock(&xfs_mount_list_lock);
2250 xfs_inodegc_cpu_dead(mp, cpu);
2251 xlog_cil_pcp_dead(mp->m_log, cpu);
2252 spin_lock(&xfs_mount_list_lock);
2253 }
2254 spin_unlock(&xfs_mount_list_lock);
2255 return 0;
2256 }
2257
2258 static int __init
xfs_cpu_hotplug_init(void)2259 xfs_cpu_hotplug_init(void)
2260 {
2261 int error;
2262
2263 error = cpuhp_setup_state_nocalls(CPUHP_XFS_DEAD, "xfs:dead", NULL,
2264 xfs_cpu_dead);
2265 if (error < 0)
2266 xfs_alert(NULL,
2267 "Failed to initialise CPU hotplug, error %d. XFS is non-functional.",
2268 error);
2269 return error;
2270 }
2271
2272 static void
xfs_cpu_hotplug_destroy(void)2273 xfs_cpu_hotplug_destroy(void)
2274 {
2275 cpuhp_remove_state_nocalls(CPUHP_XFS_DEAD);
2276 }
2277
2278 #else /* !CONFIG_HOTPLUG_CPU */
xfs_cpu_hotplug_init(void)2279 static inline int xfs_cpu_hotplug_init(void) { return 0; }
xfs_cpu_hotplug_destroy(void)2280 static inline void xfs_cpu_hotplug_destroy(void) {}
2281 #endif
2282
2283 STATIC int __init
init_xfs_fs(void)2284 init_xfs_fs(void)
2285 {
2286 int error;
2287
2288 xfs_check_ondisk_structs();
2289
2290 printk(KERN_INFO XFS_VERSION_STRING " with "
2291 XFS_BUILD_OPTIONS " enabled\n");
2292
2293 xfs_dir_startup();
2294
2295 error = xfs_cpu_hotplug_init();
2296 if (error)
2297 goto out;
2298
2299 error = xfs_init_caches();
2300 if (error)
2301 goto out_destroy_hp;
2302
2303 error = xfs_init_workqueues();
2304 if (error)
2305 goto out_destroy_caches;
2306
2307 error = xfs_mru_cache_init();
2308 if (error)
2309 goto out_destroy_wq;
2310
2311 error = xfs_init_procfs();
2312 if (error)
2313 goto out_mru_cache_uninit;
2314
2315 error = xfs_sysctl_register();
2316 if (error)
2317 goto out_cleanup_procfs;
2318
2319 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2320 if (!xfs_kset) {
2321 error = -ENOMEM;
2322 goto out_sysctl_unregister;
2323 }
2324
2325 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2326
2327 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2328 if (!xfsstats.xs_stats) {
2329 error = -ENOMEM;
2330 goto out_kset_unregister;
2331 }
2332
2333 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2334 "stats");
2335 if (error)
2336 goto out_free_stats;
2337
2338 #ifdef DEBUG
2339 xfs_dbg_kobj.kobject.kset = xfs_kset;
2340 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2341 if (error)
2342 goto out_remove_stats_kobj;
2343 #endif
2344
2345 error = xfs_qm_init();
2346 if (error)
2347 goto out_remove_dbg_kobj;
2348
2349 error = register_filesystem(&xfs_fs_type);
2350 if (error)
2351 goto out_qm_exit;
2352 return 0;
2353
2354 out_qm_exit:
2355 xfs_qm_exit();
2356 out_remove_dbg_kobj:
2357 #ifdef DEBUG
2358 xfs_sysfs_del(&xfs_dbg_kobj);
2359 out_remove_stats_kobj:
2360 #endif
2361 xfs_sysfs_del(&xfsstats.xs_kobj);
2362 out_free_stats:
2363 free_percpu(xfsstats.xs_stats);
2364 out_kset_unregister:
2365 kset_unregister(xfs_kset);
2366 out_sysctl_unregister:
2367 xfs_sysctl_unregister();
2368 out_cleanup_procfs:
2369 xfs_cleanup_procfs();
2370 out_mru_cache_uninit:
2371 xfs_mru_cache_uninit();
2372 out_destroy_wq:
2373 xfs_destroy_workqueues();
2374 out_destroy_caches:
2375 xfs_destroy_caches();
2376 out_destroy_hp:
2377 xfs_cpu_hotplug_destroy();
2378 out:
2379 return error;
2380 }
2381
2382 STATIC void __exit
exit_xfs_fs(void)2383 exit_xfs_fs(void)
2384 {
2385 xfs_qm_exit();
2386 unregister_filesystem(&xfs_fs_type);
2387 #ifdef DEBUG
2388 xfs_sysfs_del(&xfs_dbg_kobj);
2389 #endif
2390 xfs_sysfs_del(&xfsstats.xs_kobj);
2391 free_percpu(xfsstats.xs_stats);
2392 kset_unregister(xfs_kset);
2393 xfs_sysctl_unregister();
2394 xfs_cleanup_procfs();
2395 xfs_mru_cache_uninit();
2396 xfs_destroy_workqueues();
2397 xfs_destroy_caches();
2398 xfs_uuid_table_free();
2399 xfs_cpu_hotplug_destroy();
2400 }
2401
2402 module_init(init_xfs_fs);
2403 module_exit(exit_xfs_fs);
2404
2405 MODULE_AUTHOR("Silicon Graphics, Inc.");
2406 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2407 MODULE_LICENSE("GPL");
2408