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 #include "xfs_dahash_test.h"
45 #include "xfs_rtbitmap.h"
46 #include "xfs_exchmaps_item.h"
47 #include "xfs_parent.h"
48 #include "scrub/stats.h"
49 #include "scrub/rcbag_btree.h"
50
51 #include <linux/magic.h>
52 #include <linux/fs_context.h>
53 #include <linux/fs_parser.h>
54
55 static const struct super_operations xfs_super_operations;
56
57 static struct dentry *xfs_debugfs; /* top-level xfs debugfs dir */
58 static struct kset *xfs_kset; /* top-level xfs sysfs dir */
59 #ifdef DEBUG
60 static struct xfs_kobj xfs_dbg_kobj; /* global debug sysfs attrs */
61 #endif
62
63 enum xfs_dax_mode {
64 XFS_DAX_INODE = 0,
65 XFS_DAX_ALWAYS = 1,
66 XFS_DAX_NEVER = 2,
67 };
68
69 static void
70 xfs_mount_set_dax_mode(
71 struct xfs_mount *mp,
72 enum xfs_dax_mode mode)
73 {
74 switch (mode) {
75 case XFS_DAX_INODE:
76 mp->m_features &= ~(XFS_FEAT_DAX_ALWAYS | XFS_FEAT_DAX_NEVER);
77 break;
78 case XFS_DAX_ALWAYS:
79 mp->m_features |= XFS_FEAT_DAX_ALWAYS;
80 mp->m_features &= ~XFS_FEAT_DAX_NEVER;
81 break;
82 case XFS_DAX_NEVER:
83 mp->m_features |= XFS_FEAT_DAX_NEVER;
84 mp->m_features &= ~XFS_FEAT_DAX_ALWAYS;
85 break;
86 }
87 }
88
89 static const struct constant_table dax_param_enums[] = {
90 {"inode", XFS_DAX_INODE },
91 {"always", XFS_DAX_ALWAYS },
92 {"never", XFS_DAX_NEVER },
93 {}
94 };
95
96 /*
97 * Table driven mount option parser.
98 */
99 enum {
100 Opt_logbufs, Opt_logbsize, Opt_logdev, Opt_rtdev,
101 Opt_wsync, Opt_noalign, Opt_swalloc, Opt_sunit, Opt_swidth, Opt_nouuid,
102 Opt_grpid, Opt_nogrpid, Opt_bsdgroups, Opt_sysvgroups,
103 Opt_allocsize, Opt_norecovery, Opt_inode64, Opt_inode32, Opt_ikeep,
104 Opt_noikeep, Opt_largeio, Opt_nolargeio, Opt_attr2, Opt_noattr2,
105 Opt_filestreams, Opt_quota, Opt_noquota, Opt_usrquota, Opt_grpquota,
106 Opt_prjquota, Opt_uquota, Opt_gquota, Opt_pquota,
107 Opt_uqnoenforce, Opt_gqnoenforce, Opt_pqnoenforce, Opt_qnoenforce,
108 Opt_discard, Opt_nodiscard, Opt_dax, Opt_dax_enum,
109 };
110
111 static const struct fs_parameter_spec xfs_fs_parameters[] = {
112 fsparam_u32("logbufs", Opt_logbufs),
113 fsparam_string("logbsize", Opt_logbsize),
114 fsparam_string("logdev", Opt_logdev),
115 fsparam_string("rtdev", Opt_rtdev),
116 fsparam_flag("wsync", Opt_wsync),
117 fsparam_flag("noalign", Opt_noalign),
118 fsparam_flag("swalloc", Opt_swalloc),
119 fsparam_u32("sunit", Opt_sunit),
120 fsparam_u32("swidth", Opt_swidth),
121 fsparam_flag("nouuid", Opt_nouuid),
122 fsparam_flag("grpid", Opt_grpid),
123 fsparam_flag("nogrpid", Opt_nogrpid),
124 fsparam_flag("bsdgroups", Opt_bsdgroups),
125 fsparam_flag("sysvgroups", Opt_sysvgroups),
126 fsparam_string("allocsize", Opt_allocsize),
127 fsparam_flag("norecovery", Opt_norecovery),
128 fsparam_flag("inode64", Opt_inode64),
129 fsparam_flag("inode32", Opt_inode32),
130 fsparam_flag("ikeep", Opt_ikeep),
131 fsparam_flag("noikeep", Opt_noikeep),
132 fsparam_flag("largeio", Opt_largeio),
133 fsparam_flag("nolargeio", Opt_nolargeio),
134 fsparam_flag("attr2", Opt_attr2),
135 fsparam_flag("noattr2", Opt_noattr2),
136 fsparam_flag("filestreams", Opt_filestreams),
137 fsparam_flag("quota", Opt_quota),
138 fsparam_flag("noquota", Opt_noquota),
139 fsparam_flag("usrquota", Opt_usrquota),
140 fsparam_flag("grpquota", Opt_grpquota),
141 fsparam_flag("prjquota", Opt_prjquota),
142 fsparam_flag("uquota", Opt_uquota),
143 fsparam_flag("gquota", Opt_gquota),
144 fsparam_flag("pquota", Opt_pquota),
145 fsparam_flag("uqnoenforce", Opt_uqnoenforce),
146 fsparam_flag("gqnoenforce", Opt_gqnoenforce),
147 fsparam_flag("pqnoenforce", Opt_pqnoenforce),
148 fsparam_flag("qnoenforce", Opt_qnoenforce),
149 fsparam_flag("discard", Opt_discard),
150 fsparam_flag("nodiscard", Opt_nodiscard),
151 fsparam_flag("dax", Opt_dax),
152 fsparam_enum("dax", Opt_dax_enum, dax_param_enums),
153 {}
154 };
155
156 struct proc_xfs_info {
157 uint64_t flag;
158 char *str;
159 };
160
161 static int
162 xfs_fs_show_options(
163 struct seq_file *m,
164 struct dentry *root)
165 {
166 static struct proc_xfs_info xfs_info_set[] = {
167 /* the few simple ones we can get from the mount struct */
168 { XFS_FEAT_IKEEP, ",ikeep" },
169 { XFS_FEAT_WSYNC, ",wsync" },
170 { XFS_FEAT_NOALIGN, ",noalign" },
171 { XFS_FEAT_SWALLOC, ",swalloc" },
172 { XFS_FEAT_NOUUID, ",nouuid" },
173 { XFS_FEAT_NORECOVERY, ",norecovery" },
174 { XFS_FEAT_ATTR2, ",attr2" },
175 { XFS_FEAT_FILESTREAMS, ",filestreams" },
176 { XFS_FEAT_GRPID, ",grpid" },
177 { XFS_FEAT_DISCARD, ",discard" },
178 { XFS_FEAT_LARGE_IOSIZE, ",largeio" },
179 { XFS_FEAT_DAX_ALWAYS, ",dax=always" },
180 { XFS_FEAT_DAX_NEVER, ",dax=never" },
181 { 0, NULL }
182 };
183 struct xfs_mount *mp = XFS_M(root->d_sb);
184 struct proc_xfs_info *xfs_infop;
185
186 for (xfs_infop = xfs_info_set; xfs_infop->flag; xfs_infop++) {
187 if (mp->m_features & xfs_infop->flag)
188 seq_puts(m, xfs_infop->str);
189 }
190
191 seq_printf(m, ",inode%d", xfs_has_small_inums(mp) ? 32 : 64);
192
193 if (xfs_has_allocsize(mp))
194 seq_printf(m, ",allocsize=%dk",
195 (1 << mp->m_allocsize_log) >> 10);
196
197 if (mp->m_logbufs > 0)
198 seq_printf(m, ",logbufs=%d", mp->m_logbufs);
199 if (mp->m_logbsize > 0)
200 seq_printf(m, ",logbsize=%dk", mp->m_logbsize >> 10);
201
202 if (mp->m_logname)
203 seq_show_option(m, "logdev", mp->m_logname);
204 if (mp->m_rtname)
205 seq_show_option(m, "rtdev", mp->m_rtname);
206
207 if (mp->m_dalign > 0)
208 seq_printf(m, ",sunit=%d",
209 (int)XFS_FSB_TO_BB(mp, mp->m_dalign));
210 if (mp->m_swidth > 0)
211 seq_printf(m, ",swidth=%d",
212 (int)XFS_FSB_TO_BB(mp, mp->m_swidth));
213
214 if (mp->m_qflags & XFS_UQUOTA_ENFD)
215 seq_puts(m, ",usrquota");
216 else if (mp->m_qflags & XFS_UQUOTA_ACCT)
217 seq_puts(m, ",uqnoenforce");
218
219 if (mp->m_qflags & XFS_PQUOTA_ENFD)
220 seq_puts(m, ",prjquota");
221 else if (mp->m_qflags & XFS_PQUOTA_ACCT)
222 seq_puts(m, ",pqnoenforce");
223
224 if (mp->m_qflags & XFS_GQUOTA_ENFD)
225 seq_puts(m, ",grpquota");
226 else if (mp->m_qflags & XFS_GQUOTA_ACCT)
227 seq_puts(m, ",gqnoenforce");
228
229 if (!(mp->m_qflags & XFS_ALL_QUOTA_ACCT))
230 seq_puts(m, ",noquota");
231
232 return 0;
233 }
234
235 static bool
236 xfs_set_inode_alloc_perag(
237 struct xfs_perag *pag,
238 xfs_ino_t ino,
239 xfs_agnumber_t max_metadata)
240 {
241 if (!xfs_is_inode32(pag->pag_mount)) {
242 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
243 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
244 return false;
245 }
246
247 if (ino > XFS_MAXINUMBER_32) {
248 clear_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
249 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
250 return false;
251 }
252
253 set_bit(XFS_AGSTATE_ALLOWS_INODES, &pag->pag_opstate);
254 if (pag->pag_agno < max_metadata)
255 set_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
256 else
257 clear_bit(XFS_AGSTATE_PREFERS_METADATA, &pag->pag_opstate);
258 return true;
259 }
260
261 /*
262 * Set parameters for inode allocation heuristics, taking into account
263 * filesystem size and inode32/inode64 mount options; i.e. specifically
264 * whether or not XFS_FEAT_SMALL_INUMS is set.
265 *
266 * Inode allocation patterns are altered only if inode32 is requested
267 * (XFS_FEAT_SMALL_INUMS), and the filesystem is sufficiently large.
268 * If altered, XFS_OPSTATE_INODE32 is set as well.
269 *
270 * An agcount independent of that in the mount structure is provided
271 * because in the growfs case, mp->m_sb.sb_agcount is not yet updated
272 * to the potentially higher ag count.
273 *
274 * Returns the maximum AG index which may contain inodes.
275 */
276 xfs_agnumber_t
277 xfs_set_inode_alloc(
278 struct xfs_mount *mp,
279 xfs_agnumber_t agcount)
280 {
281 xfs_agnumber_t index;
282 xfs_agnumber_t maxagi = 0;
283 xfs_sb_t *sbp = &mp->m_sb;
284 xfs_agnumber_t max_metadata;
285 xfs_agino_t agino;
286 xfs_ino_t ino;
287
288 /*
289 * Calculate how much should be reserved for inodes to meet
290 * the max inode percentage. Used only for inode32.
291 */
292 if (M_IGEO(mp)->maxicount) {
293 uint64_t icount;
294
295 icount = sbp->sb_dblocks * sbp->sb_imax_pct;
296 do_div(icount, 100);
297 icount += sbp->sb_agblocks - 1;
298 do_div(icount, sbp->sb_agblocks);
299 max_metadata = icount;
300 } else {
301 max_metadata = agcount;
302 }
303
304 /* Get the last possible inode in the filesystem */
305 agino = XFS_AGB_TO_AGINO(mp, sbp->sb_agblocks - 1);
306 ino = XFS_AGINO_TO_INO(mp, agcount - 1, agino);
307
308 /*
309 * If user asked for no more than 32-bit inodes, and the fs is
310 * sufficiently large, set XFS_OPSTATE_INODE32 if we must alter
311 * the allocator to accommodate the request.
312 */
313 if (xfs_has_small_inums(mp) && ino > XFS_MAXINUMBER_32)
314 set_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
315 else
316 clear_bit(XFS_OPSTATE_INODE32, &mp->m_opstate);
317
318 for (index = 0; index < agcount; index++) {
319 struct xfs_perag *pag;
320
321 ino = XFS_AGINO_TO_INO(mp, index, agino);
322
323 pag = xfs_perag_get(mp, index);
324 if (xfs_set_inode_alloc_perag(pag, ino, max_metadata))
325 maxagi++;
326 xfs_perag_put(pag);
327 }
328
329 return xfs_is_inode32(mp) ? maxagi : agcount;
330 }
331
332 static int
333 xfs_setup_dax_always(
334 struct xfs_mount *mp)
335 {
336 if (!mp->m_ddev_targp->bt_daxdev &&
337 (!mp->m_rtdev_targp || !mp->m_rtdev_targp->bt_daxdev)) {
338 xfs_alert(mp,
339 "DAX unsupported by block device. Turning off DAX.");
340 goto disable_dax;
341 }
342
343 if (mp->m_super->s_blocksize != PAGE_SIZE) {
344 xfs_alert(mp,
345 "DAX not supported for blocksize. Turning off DAX.");
346 goto disable_dax;
347 }
348
349 if (xfs_has_reflink(mp) &&
350 bdev_is_partition(mp->m_ddev_targp->bt_bdev)) {
351 xfs_alert(mp,
352 "DAX and reflink cannot work with multi-partitions!");
353 return -EINVAL;
354 }
355
356 return 0;
357
358 disable_dax:
359 xfs_mount_set_dax_mode(mp, XFS_DAX_NEVER);
360 return 0;
361 }
362
363 STATIC int
364 xfs_blkdev_get(
365 xfs_mount_t *mp,
366 const char *name,
367 struct file **bdev_filep)
368 {
369 int error = 0;
370
371 *bdev_filep = bdev_file_open_by_path(name,
372 BLK_OPEN_READ | BLK_OPEN_WRITE | BLK_OPEN_RESTRICT_WRITES,
373 mp->m_super, &fs_holder_ops);
374 if (IS_ERR(*bdev_filep)) {
375 error = PTR_ERR(*bdev_filep);
376 *bdev_filep = NULL;
377 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
378 }
379
380 return error;
381 }
382
383 STATIC void
384 xfs_shutdown_devices(
385 struct xfs_mount *mp)
386 {
387 /*
388 * Udev is triggered whenever anyone closes a block device or unmounts
389 * a file systemm on a block device.
390 * The default udev rules invoke blkid to read the fs super and create
391 * symlinks to the bdev under /dev/disk. For this, it uses buffered
392 * reads through the page cache.
393 *
394 * xfs_db also uses buffered reads to examine metadata. There is no
395 * coordination between xfs_db and udev, which means that they can run
396 * concurrently. Note there is no coordination between the kernel and
397 * blkid either.
398 *
399 * On a system with 64k pages, the page cache can cache the superblock
400 * and the root inode (and hence the root directory) with the same 64k
401 * page. If udev spawns blkid after the mkfs and the system is busy
402 * enough that it is still running when xfs_db starts up, they'll both
403 * read from the same page in the pagecache.
404 *
405 * The unmount writes updated inode metadata to disk directly. The XFS
406 * buffer cache does not use the bdev pagecache, so it needs to
407 * invalidate that pagecache on unmount. If the above scenario occurs,
408 * the pagecache no longer reflects what's on disk, xfs_db reads the
409 * stale metadata, and fails to find /a. Most of the time this succeeds
410 * because closing a bdev invalidates the page cache, but when processes
411 * race, everyone loses.
412 */
413 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
414 blkdev_issue_flush(mp->m_logdev_targp->bt_bdev);
415 invalidate_bdev(mp->m_logdev_targp->bt_bdev);
416 }
417 if (mp->m_rtdev_targp) {
418 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
419 invalidate_bdev(mp->m_rtdev_targp->bt_bdev);
420 }
421 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
422 invalidate_bdev(mp->m_ddev_targp->bt_bdev);
423 }
424
425 /*
426 * The file system configurations are:
427 * (1) device (partition) with data and internal log
428 * (2) logical volume with data and log subvolumes.
429 * (3) logical volume with data, log, and realtime subvolumes.
430 *
431 * We only have to handle opening the log and realtime volumes here if
432 * they are present. The data subvolume has already been opened by
433 * get_sb_bdev() and is stored in sb->s_bdev.
434 */
435 STATIC int
436 xfs_open_devices(
437 struct xfs_mount *mp)
438 {
439 struct super_block *sb = mp->m_super;
440 struct block_device *ddev = sb->s_bdev;
441 struct file *logdev_file = NULL, *rtdev_file = NULL;
442 int error;
443
444 /*
445 * Open real time and log devices - order is important.
446 */
447 if (mp->m_logname) {
448 error = xfs_blkdev_get(mp, mp->m_logname, &logdev_file);
449 if (error)
450 return error;
451 }
452
453 if (mp->m_rtname) {
454 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev_file);
455 if (error)
456 goto out_close_logdev;
457
458 if (file_bdev(rtdev_file) == ddev ||
459 (logdev_file &&
460 file_bdev(rtdev_file) == file_bdev(logdev_file))) {
461 xfs_warn(mp,
462 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
463 error = -EINVAL;
464 goto out_close_rtdev;
465 }
466 }
467
468 /*
469 * Setup xfs_mount buffer target pointers
470 */
471 error = -ENOMEM;
472 mp->m_ddev_targp = xfs_alloc_buftarg(mp, sb->s_bdev_file);
473 if (!mp->m_ddev_targp)
474 goto out_close_rtdev;
475
476 if (rtdev_file) {
477 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev_file);
478 if (!mp->m_rtdev_targp)
479 goto out_free_ddev_targ;
480 }
481
482 if (logdev_file && file_bdev(logdev_file) != ddev) {
483 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev_file);
484 if (!mp->m_logdev_targp)
485 goto out_free_rtdev_targ;
486 } else {
487 mp->m_logdev_targp = mp->m_ddev_targp;
488 /* Handle won't be used, drop it */
489 if (logdev_file)
490 bdev_fput(logdev_file);
491 }
492
493 return 0;
494
495 out_free_rtdev_targ:
496 if (mp->m_rtdev_targp)
497 xfs_free_buftarg(mp->m_rtdev_targp);
498 out_free_ddev_targ:
499 xfs_free_buftarg(mp->m_ddev_targp);
500 out_close_rtdev:
501 if (rtdev_file)
502 bdev_fput(rtdev_file);
503 out_close_logdev:
504 if (logdev_file)
505 bdev_fput(logdev_file);
506 return error;
507 }
508
509 /*
510 * Setup xfs_mount buffer target pointers based on superblock
511 */
512 STATIC int
513 xfs_setup_devices(
514 struct xfs_mount *mp)
515 {
516 int error;
517
518 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
519 if (error)
520 return error;
521
522 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
523 unsigned int log_sector_size = BBSIZE;
524
525 if (xfs_has_sector(mp))
526 log_sector_size = mp->m_sb.sb_logsectsize;
527 error = xfs_setsize_buftarg(mp->m_logdev_targp,
528 log_sector_size);
529 if (error)
530 return error;
531 }
532 if (mp->m_rtdev_targp) {
533 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
534 mp->m_sb.sb_sectsize);
535 if (error)
536 return error;
537 }
538
539 return 0;
540 }
541
542 STATIC int
543 xfs_init_mount_workqueues(
544 struct xfs_mount *mp)
545 {
546 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
547 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
548 1, mp->m_super->s_id);
549 if (!mp->m_buf_workqueue)
550 goto out;
551
552 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
553 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
554 0, mp->m_super->s_id);
555 if (!mp->m_unwritten_workqueue)
556 goto out_destroy_buf;
557
558 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
559 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
560 0, mp->m_super->s_id);
561 if (!mp->m_reclaim_workqueue)
562 goto out_destroy_unwritten;
563
564 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
565 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
566 0, mp->m_super->s_id);
567 if (!mp->m_blockgc_wq)
568 goto out_destroy_reclaim;
569
570 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
571 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
572 1, mp->m_super->s_id);
573 if (!mp->m_inodegc_wq)
574 goto out_destroy_blockgc;
575
576 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
577 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
578 if (!mp->m_sync_workqueue)
579 goto out_destroy_inodegc;
580
581 return 0;
582
583 out_destroy_inodegc:
584 destroy_workqueue(mp->m_inodegc_wq);
585 out_destroy_blockgc:
586 destroy_workqueue(mp->m_blockgc_wq);
587 out_destroy_reclaim:
588 destroy_workqueue(mp->m_reclaim_workqueue);
589 out_destroy_unwritten:
590 destroy_workqueue(mp->m_unwritten_workqueue);
591 out_destroy_buf:
592 destroy_workqueue(mp->m_buf_workqueue);
593 out:
594 return -ENOMEM;
595 }
596
597 STATIC void
598 xfs_destroy_mount_workqueues(
599 struct xfs_mount *mp)
600 {
601 destroy_workqueue(mp->m_sync_workqueue);
602 destroy_workqueue(mp->m_blockgc_wq);
603 destroy_workqueue(mp->m_inodegc_wq);
604 destroy_workqueue(mp->m_reclaim_workqueue);
605 destroy_workqueue(mp->m_unwritten_workqueue);
606 destroy_workqueue(mp->m_buf_workqueue);
607 }
608
609 static void
610 xfs_flush_inodes_worker(
611 struct work_struct *work)
612 {
613 struct xfs_mount *mp = container_of(work, struct xfs_mount,
614 m_flush_inodes_work);
615 struct super_block *sb = mp->m_super;
616
617 if (down_read_trylock(&sb->s_umount)) {
618 sync_inodes_sb(sb);
619 up_read(&sb->s_umount);
620 }
621 }
622
623 /*
624 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
625 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
626 * for IO to complete so that we effectively throttle multiple callers to the
627 * rate at which IO is completing.
628 */
629 void
630 xfs_flush_inodes(
631 struct xfs_mount *mp)
632 {
633 /*
634 * If flush_work() returns true then that means we waited for a flush
635 * which was already in progress. Don't bother running another scan.
636 */
637 if (flush_work(&mp->m_flush_inodes_work))
638 return;
639
640 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
641 flush_work(&mp->m_flush_inodes_work);
642 }
643
644 /* Catch misguided souls that try to use this interface on XFS */
645 STATIC struct inode *
646 xfs_fs_alloc_inode(
647 struct super_block *sb)
648 {
649 BUG();
650 return NULL;
651 }
652
653 /*
654 * Now that the generic code is guaranteed not to be accessing
655 * the linux inode, we can inactivate and reclaim the inode.
656 */
657 STATIC void
658 xfs_fs_destroy_inode(
659 struct inode *inode)
660 {
661 struct xfs_inode *ip = XFS_I(inode);
662
663 trace_xfs_destroy_inode(ip);
664
665 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
666 XFS_STATS_INC(ip->i_mount, vn_rele);
667 XFS_STATS_INC(ip->i_mount, vn_remove);
668 xfs_inode_mark_reclaimable(ip);
669 }
670
671 static void
672 xfs_fs_dirty_inode(
673 struct inode *inode,
674 int flags)
675 {
676 struct xfs_inode *ip = XFS_I(inode);
677 struct xfs_mount *mp = ip->i_mount;
678 struct xfs_trans *tp;
679
680 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
681 return;
682
683 /*
684 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
685 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
686 * in flags possibly together with I_DIRTY_SYNC.
687 */
688 if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
689 return;
690
691 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
692 return;
693 xfs_ilock(ip, XFS_ILOCK_EXCL);
694 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
695 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
696 xfs_trans_commit(tp);
697 }
698
699 /*
700 * Slab object creation initialisation for the XFS inode.
701 * This covers only the idempotent fields in the XFS inode;
702 * all other fields need to be initialised on allocation
703 * from the slab. This avoids the need to repeatedly initialise
704 * fields in the xfs inode that left in the initialise state
705 * when freeing the inode.
706 */
707 STATIC void
708 xfs_fs_inode_init_once(
709 void *inode)
710 {
711 struct xfs_inode *ip = inode;
712
713 memset(ip, 0, sizeof(struct xfs_inode));
714
715 /* vfs inode */
716 inode_init_once(VFS_I(ip));
717
718 /* xfs inode */
719 atomic_set(&ip->i_pincount, 0);
720 spin_lock_init(&ip->i_flags_lock);
721 init_rwsem(&ip->i_lock);
722 }
723
724 /*
725 * We do an unlocked check for XFS_IDONTCACHE here because we are already
726 * serialised against cache hits here via the inode->i_lock and igrab() in
727 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
728 * racing with us, and it avoids needing to grab a spinlock here for every inode
729 * we drop the final reference on.
730 */
731 STATIC int
732 xfs_fs_drop_inode(
733 struct inode *inode)
734 {
735 struct xfs_inode *ip = XFS_I(inode);
736
737 /*
738 * If this unlinked inode is in the middle of recovery, don't
739 * drop the inode just yet; log recovery will take care of
740 * that. See the comment for this inode flag.
741 */
742 if (ip->i_flags & XFS_IRECOVERY) {
743 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
744 return 0;
745 }
746
747 return generic_drop_inode(inode);
748 }
749
750 static void
751 xfs_mount_free(
752 struct xfs_mount *mp)
753 {
754 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
755 xfs_free_buftarg(mp->m_logdev_targp);
756 if (mp->m_rtdev_targp)
757 xfs_free_buftarg(mp->m_rtdev_targp);
758 if (mp->m_ddev_targp)
759 xfs_free_buftarg(mp->m_ddev_targp);
760
761 debugfs_remove(mp->m_debugfs);
762 kfree(mp->m_rtname);
763 kfree(mp->m_logname);
764 kfree(mp);
765 }
766
767 STATIC int
768 xfs_fs_sync_fs(
769 struct super_block *sb,
770 int wait)
771 {
772 struct xfs_mount *mp = XFS_M(sb);
773 int error;
774
775 trace_xfs_fs_sync_fs(mp, __return_address);
776
777 /*
778 * Doing anything during the async pass would be counterproductive.
779 */
780 if (!wait)
781 return 0;
782
783 error = xfs_log_force(mp, XFS_LOG_SYNC);
784 if (error)
785 return error;
786
787 if (laptop_mode) {
788 /*
789 * The disk must be active because we're syncing.
790 * We schedule log work now (now that the disk is
791 * active) instead of later (when it might not be).
792 */
793 flush_delayed_work(&mp->m_log->l_work);
794 }
795
796 /*
797 * If we are called with page faults frozen out, it means we are about
798 * to freeze the transaction subsystem. Take the opportunity to shut
799 * down inodegc because once SB_FREEZE_FS is set it's too late to
800 * prevent inactivation races with freeze. The fs doesn't get called
801 * again by the freezing process until after SB_FREEZE_FS has been set,
802 * so it's now or never. Same logic applies to speculative allocation
803 * garbage collection.
804 *
805 * We don't care if this is a normal syncfs call that does this or
806 * freeze that does this - we can run this multiple times without issue
807 * and we won't race with a restart because a restart can only occur
808 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
809 */
810 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
811 xfs_inodegc_stop(mp);
812 xfs_blockgc_stop(mp);
813 }
814
815 return 0;
816 }
817
818 STATIC int
819 xfs_fs_statfs(
820 struct dentry *dentry,
821 struct kstatfs *statp)
822 {
823 struct xfs_mount *mp = XFS_M(dentry->d_sb);
824 xfs_sb_t *sbp = &mp->m_sb;
825 struct xfs_inode *ip = XFS_I(d_inode(dentry));
826 uint64_t fakeinos, id;
827 uint64_t icount;
828 uint64_t ifree;
829 uint64_t fdblocks;
830 xfs_extlen_t lsize;
831 int64_t ffree;
832
833 /*
834 * Expedite background inodegc but don't wait. We do not want to block
835 * here waiting hours for a billion extent file to be truncated.
836 */
837 xfs_inodegc_push(mp);
838
839 statp->f_type = XFS_SUPER_MAGIC;
840 statp->f_namelen = MAXNAMELEN - 1;
841
842 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
843 statp->f_fsid = u64_to_fsid(id);
844
845 icount = percpu_counter_sum(&mp->m_icount);
846 ifree = percpu_counter_sum(&mp->m_ifree);
847 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
848
849 spin_lock(&mp->m_sb_lock);
850 statp->f_bsize = sbp->sb_blocksize;
851 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
852 statp->f_blocks = sbp->sb_dblocks - lsize;
853 spin_unlock(&mp->m_sb_lock);
854
855 /* make sure statp->f_bfree does not underflow */
856 statp->f_bfree = max_t(int64_t, 0,
857 fdblocks - xfs_fdblocks_unavailable(mp));
858 statp->f_bavail = statp->f_bfree;
859
860 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
861 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
862 if (M_IGEO(mp)->maxicount)
863 statp->f_files = min_t(typeof(statp->f_files),
864 statp->f_files,
865 M_IGEO(mp)->maxicount);
866
867 /* If sb_icount overshot maxicount, report actual allocation */
868 statp->f_files = max_t(typeof(statp->f_files),
869 statp->f_files,
870 sbp->sb_icount);
871
872 /* make sure statp->f_ffree does not underflow */
873 ffree = statp->f_files - (icount - ifree);
874 statp->f_ffree = max_t(int64_t, ffree, 0);
875
876
877 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
878 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
879 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
880 xfs_qm_statvfs(ip, statp);
881
882 if (XFS_IS_REALTIME_MOUNT(mp) &&
883 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
884 s64 freertx;
885
886 statp->f_blocks = sbp->sb_rblocks;
887 freertx = percpu_counter_sum_positive(&mp->m_frextents);
888 statp->f_bavail = statp->f_bfree = xfs_rtx_to_rtb(mp, freertx);
889 }
890
891 return 0;
892 }
893
894 STATIC void
895 xfs_save_resvblks(struct xfs_mount *mp)
896 {
897 mp->m_resblks_save = mp->m_resblks;
898 xfs_reserve_blocks(mp, 0);
899 }
900
901 STATIC void
902 xfs_restore_resvblks(struct xfs_mount *mp)
903 {
904 uint64_t resblks;
905
906 if (mp->m_resblks_save) {
907 resblks = mp->m_resblks_save;
908 mp->m_resblks_save = 0;
909 } else
910 resblks = xfs_default_resblks(mp);
911
912 xfs_reserve_blocks(mp, resblks);
913 }
914
915 /*
916 * Second stage of a freeze. The data is already frozen so we only
917 * need to take care of the metadata. Once that's done sync the superblock
918 * to the log to dirty it in case of a crash while frozen. This ensures that we
919 * will recover the unlinked inode lists on the next mount.
920 */
921 STATIC int
922 xfs_fs_freeze(
923 struct super_block *sb)
924 {
925 struct xfs_mount *mp = XFS_M(sb);
926 unsigned int flags;
927 int ret;
928
929 /*
930 * The filesystem is now frozen far enough that memory reclaim
931 * cannot safely operate on the filesystem. Hence we need to
932 * set a GFP_NOFS context here to avoid recursion deadlocks.
933 */
934 flags = memalloc_nofs_save();
935 xfs_save_resvblks(mp);
936 ret = xfs_log_quiesce(mp);
937 memalloc_nofs_restore(flags);
938
939 /*
940 * For read-write filesystems, we need to restart the inodegc on error
941 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
942 * going to be run to restart it now. We are at SB_FREEZE_FS level
943 * here, so we can restart safely without racing with a stop in
944 * xfs_fs_sync_fs().
945 */
946 if (ret && !xfs_is_readonly(mp)) {
947 xfs_blockgc_start(mp);
948 xfs_inodegc_start(mp);
949 }
950
951 return ret;
952 }
953
954 STATIC int
955 xfs_fs_unfreeze(
956 struct super_block *sb)
957 {
958 struct xfs_mount *mp = XFS_M(sb);
959
960 xfs_restore_resvblks(mp);
961 xfs_log_work_queue(mp);
962
963 /*
964 * Don't reactivate the inodegc worker on a readonly filesystem because
965 * inodes are sent directly to reclaim. Don't reactivate the blockgc
966 * worker because there are no speculative preallocations on a readonly
967 * filesystem.
968 */
969 if (!xfs_is_readonly(mp)) {
970 xfs_blockgc_start(mp);
971 xfs_inodegc_start(mp);
972 }
973
974 return 0;
975 }
976
977 /*
978 * This function fills in xfs_mount_t fields based on mount args.
979 * Note: the superblock _has_ now been read in.
980 */
981 STATIC int
982 xfs_finish_flags(
983 struct xfs_mount *mp)
984 {
985 /* Fail a mount where the logbuf is smaller than the log stripe */
986 if (xfs_has_logv2(mp)) {
987 if (mp->m_logbsize <= 0 &&
988 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
989 mp->m_logbsize = mp->m_sb.sb_logsunit;
990 } else if (mp->m_logbsize > 0 &&
991 mp->m_logbsize < mp->m_sb.sb_logsunit) {
992 xfs_warn(mp,
993 "logbuf size must be greater than or equal to log stripe size");
994 return -EINVAL;
995 }
996 } else {
997 /* Fail a mount if the logbuf is larger than 32K */
998 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
999 xfs_warn(mp,
1000 "logbuf size for version 1 logs must be 16K or 32K");
1001 return -EINVAL;
1002 }
1003 }
1004
1005 /*
1006 * V5 filesystems always use attr2 format for attributes.
1007 */
1008 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
1009 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1010 "attr2 is always enabled for V5 filesystems.");
1011 return -EINVAL;
1012 }
1013
1014 /*
1015 * prohibit r/w mounts of read-only filesystems
1016 */
1017 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1018 xfs_warn(mp,
1019 "cannot mount a read-only filesystem as read-write");
1020 return -EROFS;
1021 }
1022
1023 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1024 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1025 !xfs_has_pquotino(mp)) {
1026 xfs_warn(mp,
1027 "Super block does not support project and group quota together");
1028 return -EINVAL;
1029 }
1030
1031 return 0;
1032 }
1033
1034 static int
1035 xfs_init_percpu_counters(
1036 struct xfs_mount *mp)
1037 {
1038 int error;
1039
1040 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1041 if (error)
1042 return -ENOMEM;
1043
1044 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1045 if (error)
1046 goto free_icount;
1047
1048 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1049 if (error)
1050 goto free_ifree;
1051
1052 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1053 if (error)
1054 goto free_fdblocks;
1055
1056 error = percpu_counter_init(&mp->m_delalloc_rtextents, 0, GFP_KERNEL);
1057 if (error)
1058 goto free_delalloc;
1059
1060 error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1061 if (error)
1062 goto free_delalloc_rt;
1063
1064 return 0;
1065
1066 free_delalloc_rt:
1067 percpu_counter_destroy(&mp->m_delalloc_rtextents);
1068 free_delalloc:
1069 percpu_counter_destroy(&mp->m_delalloc_blks);
1070 free_fdblocks:
1071 percpu_counter_destroy(&mp->m_fdblocks);
1072 free_ifree:
1073 percpu_counter_destroy(&mp->m_ifree);
1074 free_icount:
1075 percpu_counter_destroy(&mp->m_icount);
1076 return -ENOMEM;
1077 }
1078
1079 void
1080 xfs_reinit_percpu_counters(
1081 struct xfs_mount *mp)
1082 {
1083 percpu_counter_set(&mp->m_icount, mp->m_sb.sb_icount);
1084 percpu_counter_set(&mp->m_ifree, mp->m_sb.sb_ifree);
1085 percpu_counter_set(&mp->m_fdblocks, mp->m_sb.sb_fdblocks);
1086 percpu_counter_set(&mp->m_frextents, mp->m_sb.sb_frextents);
1087 }
1088
1089 static void
1090 xfs_destroy_percpu_counters(
1091 struct xfs_mount *mp)
1092 {
1093 percpu_counter_destroy(&mp->m_icount);
1094 percpu_counter_destroy(&mp->m_ifree);
1095 percpu_counter_destroy(&mp->m_fdblocks);
1096 ASSERT(xfs_is_shutdown(mp) ||
1097 percpu_counter_sum(&mp->m_delalloc_rtextents) == 0);
1098 percpu_counter_destroy(&mp->m_delalloc_rtextents);
1099 ASSERT(xfs_is_shutdown(mp) ||
1100 percpu_counter_sum(&mp->m_delalloc_blks) == 0);
1101 percpu_counter_destroy(&mp->m_delalloc_blks);
1102 percpu_counter_destroy(&mp->m_frextents);
1103 }
1104
1105 static int
1106 xfs_inodegc_init_percpu(
1107 struct xfs_mount *mp)
1108 {
1109 struct xfs_inodegc *gc;
1110 int cpu;
1111
1112 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1113 if (!mp->m_inodegc)
1114 return -ENOMEM;
1115
1116 for_each_possible_cpu(cpu) {
1117 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1118 gc->cpu = cpu;
1119 gc->mp = mp;
1120 init_llist_head(&gc->list);
1121 gc->items = 0;
1122 gc->error = 0;
1123 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1124 }
1125 return 0;
1126 }
1127
1128 static void
1129 xfs_inodegc_free_percpu(
1130 struct xfs_mount *mp)
1131 {
1132 if (!mp->m_inodegc)
1133 return;
1134 free_percpu(mp->m_inodegc);
1135 }
1136
1137 static void
1138 xfs_fs_put_super(
1139 struct super_block *sb)
1140 {
1141 struct xfs_mount *mp = XFS_M(sb);
1142
1143 xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1144 xfs_filestream_unmount(mp);
1145 xfs_unmountfs(mp);
1146
1147 xfs_freesb(mp);
1148 xchk_mount_stats_free(mp);
1149 free_percpu(mp->m_stats.xs_stats);
1150 xfs_inodegc_free_percpu(mp);
1151 xfs_destroy_percpu_counters(mp);
1152 xfs_destroy_mount_workqueues(mp);
1153 xfs_shutdown_devices(mp);
1154 }
1155
1156 static long
1157 xfs_fs_nr_cached_objects(
1158 struct super_block *sb,
1159 struct shrink_control *sc)
1160 {
1161 /* Paranoia: catch incorrect calls during mount setup or teardown */
1162 if (WARN_ON_ONCE(!sb->s_fs_info))
1163 return 0;
1164 return xfs_reclaim_inodes_count(XFS_M(sb));
1165 }
1166
1167 static long
1168 xfs_fs_free_cached_objects(
1169 struct super_block *sb,
1170 struct shrink_control *sc)
1171 {
1172 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1173 }
1174
1175 static void
1176 xfs_fs_shutdown(
1177 struct super_block *sb)
1178 {
1179 xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED);
1180 }
1181
1182 static const struct super_operations xfs_super_operations = {
1183 .alloc_inode = xfs_fs_alloc_inode,
1184 .destroy_inode = xfs_fs_destroy_inode,
1185 .dirty_inode = xfs_fs_dirty_inode,
1186 .drop_inode = xfs_fs_drop_inode,
1187 .put_super = xfs_fs_put_super,
1188 .sync_fs = xfs_fs_sync_fs,
1189 .freeze_fs = xfs_fs_freeze,
1190 .unfreeze_fs = xfs_fs_unfreeze,
1191 .statfs = xfs_fs_statfs,
1192 .show_options = xfs_fs_show_options,
1193 .nr_cached_objects = xfs_fs_nr_cached_objects,
1194 .free_cached_objects = xfs_fs_free_cached_objects,
1195 .shutdown = xfs_fs_shutdown,
1196 };
1197
1198 static int
1199 suffix_kstrtoint(
1200 const char *s,
1201 unsigned int base,
1202 int *res)
1203 {
1204 int last, shift_left_factor = 0, _res;
1205 char *value;
1206 int ret = 0;
1207
1208 value = kstrdup(s, GFP_KERNEL);
1209 if (!value)
1210 return -ENOMEM;
1211
1212 last = strlen(value) - 1;
1213 if (value[last] == 'K' || value[last] == 'k') {
1214 shift_left_factor = 10;
1215 value[last] = '\0';
1216 }
1217 if (value[last] == 'M' || value[last] == 'm') {
1218 shift_left_factor = 20;
1219 value[last] = '\0';
1220 }
1221 if (value[last] == 'G' || value[last] == 'g') {
1222 shift_left_factor = 30;
1223 value[last] = '\0';
1224 }
1225
1226 if (kstrtoint(value, base, &_res))
1227 ret = -EINVAL;
1228 kfree(value);
1229 *res = _res << shift_left_factor;
1230 return ret;
1231 }
1232
1233 static inline void
1234 xfs_fs_warn_deprecated(
1235 struct fs_context *fc,
1236 struct fs_parameter *param,
1237 uint64_t flag,
1238 bool value)
1239 {
1240 /* Don't print the warning if reconfiguring and current mount point
1241 * already had the flag set
1242 */
1243 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1244 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1245 return;
1246 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1247 }
1248
1249 /*
1250 * Set mount state from a mount option.
1251 *
1252 * NOTE: mp->m_super is NULL here!
1253 */
1254 static int
1255 xfs_fs_parse_param(
1256 struct fs_context *fc,
1257 struct fs_parameter *param)
1258 {
1259 struct xfs_mount *parsing_mp = fc->s_fs_info;
1260 struct fs_parse_result result;
1261 int size = 0;
1262 int opt;
1263
1264 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1265 if (opt < 0)
1266 return opt;
1267
1268 switch (opt) {
1269 case Opt_logbufs:
1270 parsing_mp->m_logbufs = result.uint_32;
1271 return 0;
1272 case Opt_logbsize:
1273 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1274 return -EINVAL;
1275 return 0;
1276 case Opt_logdev:
1277 kfree(parsing_mp->m_logname);
1278 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1279 if (!parsing_mp->m_logname)
1280 return -ENOMEM;
1281 return 0;
1282 case Opt_rtdev:
1283 kfree(parsing_mp->m_rtname);
1284 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1285 if (!parsing_mp->m_rtname)
1286 return -ENOMEM;
1287 return 0;
1288 case Opt_allocsize:
1289 if (suffix_kstrtoint(param->string, 10, &size))
1290 return -EINVAL;
1291 parsing_mp->m_allocsize_log = ffs(size) - 1;
1292 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1293 return 0;
1294 case Opt_grpid:
1295 case Opt_bsdgroups:
1296 parsing_mp->m_features |= XFS_FEAT_GRPID;
1297 return 0;
1298 case Opt_nogrpid:
1299 case Opt_sysvgroups:
1300 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1301 return 0;
1302 case Opt_wsync:
1303 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1304 return 0;
1305 case Opt_norecovery:
1306 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1307 return 0;
1308 case Opt_noalign:
1309 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1310 return 0;
1311 case Opt_swalloc:
1312 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1313 return 0;
1314 case Opt_sunit:
1315 parsing_mp->m_dalign = result.uint_32;
1316 return 0;
1317 case Opt_swidth:
1318 parsing_mp->m_swidth = result.uint_32;
1319 return 0;
1320 case Opt_inode32:
1321 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1322 return 0;
1323 case Opt_inode64:
1324 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1325 return 0;
1326 case Opt_nouuid:
1327 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1328 return 0;
1329 case Opt_largeio:
1330 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1331 return 0;
1332 case Opt_nolargeio:
1333 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1334 return 0;
1335 case Opt_filestreams:
1336 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1337 return 0;
1338 case Opt_noquota:
1339 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1340 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1341 return 0;
1342 case Opt_quota:
1343 case Opt_uquota:
1344 case Opt_usrquota:
1345 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1346 return 0;
1347 case Opt_qnoenforce:
1348 case Opt_uqnoenforce:
1349 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1350 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1351 return 0;
1352 case Opt_pquota:
1353 case Opt_prjquota:
1354 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1355 return 0;
1356 case Opt_pqnoenforce:
1357 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1358 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1359 return 0;
1360 case Opt_gquota:
1361 case Opt_grpquota:
1362 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1363 return 0;
1364 case Opt_gqnoenforce:
1365 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1366 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1367 return 0;
1368 case Opt_discard:
1369 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1370 return 0;
1371 case Opt_nodiscard:
1372 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1373 return 0;
1374 #ifdef CONFIG_FS_DAX
1375 case Opt_dax:
1376 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1377 return 0;
1378 case Opt_dax_enum:
1379 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1380 return 0;
1381 #endif
1382 /* Following mount options will be removed in September 2025 */
1383 case Opt_ikeep:
1384 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1385 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1386 return 0;
1387 case Opt_noikeep:
1388 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1389 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1390 return 0;
1391 case Opt_attr2:
1392 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1393 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1394 return 0;
1395 case Opt_noattr2:
1396 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1397 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1398 return 0;
1399 default:
1400 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1401 return -EINVAL;
1402 }
1403
1404 return 0;
1405 }
1406
1407 static int
1408 xfs_fs_validate_params(
1409 struct xfs_mount *mp)
1410 {
1411 /* No recovery flag requires a read-only mount */
1412 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1413 xfs_warn(mp, "no-recovery mounts must be read-only.");
1414 return -EINVAL;
1415 }
1416
1417 /*
1418 * We have not read the superblock at this point, so only the attr2
1419 * mount option can set the attr2 feature by this stage.
1420 */
1421 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1422 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1423 return -EINVAL;
1424 }
1425
1426
1427 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1428 xfs_warn(mp,
1429 "sunit and swidth options incompatible with the noalign option");
1430 return -EINVAL;
1431 }
1432
1433 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1434 xfs_warn(mp, "quota support not available in this kernel.");
1435 return -EINVAL;
1436 }
1437
1438 if ((mp->m_dalign && !mp->m_swidth) ||
1439 (!mp->m_dalign && mp->m_swidth)) {
1440 xfs_warn(mp, "sunit and swidth must be specified together");
1441 return -EINVAL;
1442 }
1443
1444 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1445 xfs_warn(mp,
1446 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1447 mp->m_swidth, mp->m_dalign);
1448 return -EINVAL;
1449 }
1450
1451 if (mp->m_logbufs != -1 &&
1452 mp->m_logbufs != 0 &&
1453 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1454 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1455 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1456 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1457 return -EINVAL;
1458 }
1459
1460 if (mp->m_logbsize != -1 &&
1461 mp->m_logbsize != 0 &&
1462 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1463 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1464 !is_power_of_2(mp->m_logbsize))) {
1465 xfs_warn(mp,
1466 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1467 mp->m_logbsize);
1468 return -EINVAL;
1469 }
1470
1471 if (xfs_has_allocsize(mp) &&
1472 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1473 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1474 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1475 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1476 return -EINVAL;
1477 }
1478
1479 return 0;
1480 }
1481
1482 struct dentry *
1483 xfs_debugfs_mkdir(
1484 const char *name,
1485 struct dentry *parent)
1486 {
1487 struct dentry *child;
1488
1489 /* Apparently we're expected to ignore error returns?? */
1490 child = debugfs_create_dir(name, parent);
1491 if (IS_ERR(child))
1492 return NULL;
1493
1494 return child;
1495 }
1496
1497 static int
1498 xfs_fs_fill_super(
1499 struct super_block *sb,
1500 struct fs_context *fc)
1501 {
1502 struct xfs_mount *mp = sb->s_fs_info;
1503 struct inode *root;
1504 int flags = 0, error;
1505
1506 mp->m_super = sb;
1507
1508 /*
1509 * Copy VFS mount flags from the context now that all parameter parsing
1510 * is guaranteed to have been completed by either the old mount API or
1511 * the newer fsopen/fsconfig API.
1512 */
1513 if (fc->sb_flags & SB_RDONLY)
1514 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1515 if (fc->sb_flags & SB_DIRSYNC)
1516 mp->m_features |= XFS_FEAT_DIRSYNC;
1517 if (fc->sb_flags & SB_SYNCHRONOUS)
1518 mp->m_features |= XFS_FEAT_WSYNC;
1519
1520 error = xfs_fs_validate_params(mp);
1521 if (error)
1522 return error;
1523
1524 sb_min_blocksize(sb, BBSIZE);
1525 sb->s_xattr = xfs_xattr_handlers;
1526 sb->s_export_op = &xfs_export_operations;
1527 #ifdef CONFIG_XFS_QUOTA
1528 sb->s_qcop = &xfs_quotactl_operations;
1529 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1530 #endif
1531 sb->s_op = &xfs_super_operations;
1532
1533 /*
1534 * Delay mount work if the debug hook is set. This is debug
1535 * instrumention to coordinate simulation of xfs mount failures with
1536 * VFS superblock operations
1537 */
1538 if (xfs_globals.mount_delay) {
1539 xfs_notice(mp, "Delaying mount for %d seconds.",
1540 xfs_globals.mount_delay);
1541 msleep(xfs_globals.mount_delay * 1000);
1542 }
1543
1544 if (fc->sb_flags & SB_SILENT)
1545 flags |= XFS_MFSI_QUIET;
1546
1547 error = xfs_open_devices(mp);
1548 if (error)
1549 return error;
1550
1551 if (xfs_debugfs) {
1552 mp->m_debugfs = xfs_debugfs_mkdir(mp->m_super->s_id,
1553 xfs_debugfs);
1554 } else {
1555 mp->m_debugfs = NULL;
1556 }
1557
1558 error = xfs_init_mount_workqueues(mp);
1559 if (error)
1560 goto out_shutdown_devices;
1561
1562 error = xfs_init_percpu_counters(mp);
1563 if (error)
1564 goto out_destroy_workqueues;
1565
1566 error = xfs_inodegc_init_percpu(mp);
1567 if (error)
1568 goto out_destroy_counters;
1569
1570 /* Allocate stats memory before we do operations that might use it */
1571 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1572 if (!mp->m_stats.xs_stats) {
1573 error = -ENOMEM;
1574 goto out_destroy_inodegc;
1575 }
1576
1577 error = xchk_mount_stats_alloc(mp);
1578 if (error)
1579 goto out_free_stats;
1580
1581 error = xfs_readsb(mp, flags);
1582 if (error)
1583 goto out_free_scrub_stats;
1584
1585 error = xfs_finish_flags(mp);
1586 if (error)
1587 goto out_free_sb;
1588
1589 error = xfs_setup_devices(mp);
1590 if (error)
1591 goto out_free_sb;
1592
1593 /*
1594 * V4 support is undergoing deprecation.
1595 *
1596 * Note: this has to use an open coded m_features check as xfs_has_crc
1597 * always returns false for !CONFIG_XFS_SUPPORT_V4.
1598 */
1599 if (!(mp->m_features & XFS_FEAT_CRC)) {
1600 if (!IS_ENABLED(CONFIG_XFS_SUPPORT_V4)) {
1601 xfs_warn(mp,
1602 "Deprecated V4 format (crc=0) not supported by kernel.");
1603 error = -EINVAL;
1604 goto out_free_sb;
1605 }
1606 xfs_warn_once(mp,
1607 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1608 }
1609
1610 /* ASCII case insensitivity is undergoing deprecation. */
1611 if (xfs_has_asciici(mp)) {
1612 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1613 xfs_warn_once(mp,
1614 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1615 #else
1616 xfs_warn(mp,
1617 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1618 error = -EINVAL;
1619 goto out_free_sb;
1620 #endif
1621 }
1622
1623 /* Filesystem claims it needs repair, so refuse the mount. */
1624 if (xfs_has_needsrepair(mp)) {
1625 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1626 error = -EFSCORRUPTED;
1627 goto out_free_sb;
1628 }
1629
1630 /*
1631 * Don't touch the filesystem if a user tool thinks it owns the primary
1632 * superblock. mkfs doesn't clear the flag from secondary supers, so
1633 * we don't check them at all.
1634 */
1635 if (mp->m_sb.sb_inprogress) {
1636 xfs_warn(mp, "Offline file system operation in progress!");
1637 error = -EFSCORRUPTED;
1638 goto out_free_sb;
1639 }
1640
1641 /*
1642 * Until this is fixed only page-sized or smaller data blocks work.
1643 */
1644 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1645 xfs_warn(mp,
1646 "File system with blocksize %d bytes. "
1647 "Only pagesize (%ld) or less will currently work.",
1648 mp->m_sb.sb_blocksize, PAGE_SIZE);
1649 error = -ENOSYS;
1650 goto out_free_sb;
1651 }
1652
1653 /* Ensure this filesystem fits in the page cache limits */
1654 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1655 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1656 xfs_warn(mp,
1657 "file system too large to be mounted on this system.");
1658 error = -EFBIG;
1659 goto out_free_sb;
1660 }
1661
1662 /*
1663 * XFS block mappings use 54 bits to store the logical block offset.
1664 * This should suffice to handle the maximum file size that the VFS
1665 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1666 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1667 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1668 * to check this assertion.
1669 *
1670 * Avoid integer overflow by comparing the maximum bmbt offset to the
1671 * maximum pagecache offset in units of fs blocks.
1672 */
1673 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1674 xfs_warn(mp,
1675 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1676 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1677 XFS_MAX_FILEOFF);
1678 error = -EINVAL;
1679 goto out_free_sb;
1680 }
1681
1682 error = xfs_filestream_mount(mp);
1683 if (error)
1684 goto out_free_sb;
1685
1686 /*
1687 * we must configure the block size in the superblock before we run the
1688 * full mount process as the mount process can lookup and cache inodes.
1689 */
1690 sb->s_magic = XFS_SUPER_MAGIC;
1691 sb->s_blocksize = mp->m_sb.sb_blocksize;
1692 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1693 sb->s_maxbytes = MAX_LFS_FILESIZE;
1694 sb->s_max_links = XFS_MAXLINK;
1695 sb->s_time_gran = 1;
1696 if (xfs_has_bigtime(mp)) {
1697 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1698 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1699 } else {
1700 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1701 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1702 }
1703 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1704 sb->s_iflags |= SB_I_CGROUPWB;
1705
1706 set_posix_acl_flag(sb);
1707
1708 /* version 5 superblocks support inode version counters. */
1709 if (xfs_has_crc(mp))
1710 sb->s_flags |= SB_I_VERSION;
1711
1712 if (xfs_has_dax_always(mp)) {
1713 error = xfs_setup_dax_always(mp);
1714 if (error)
1715 goto out_filestream_unmount;
1716 }
1717
1718 if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1719 xfs_warn(mp,
1720 "mounting with \"discard\" option, but the device does not support discard");
1721 mp->m_features &= ~XFS_FEAT_DISCARD;
1722 }
1723
1724 if (xfs_has_reflink(mp)) {
1725 if (mp->m_sb.sb_rblocks) {
1726 xfs_alert(mp,
1727 "reflink not compatible with realtime device!");
1728 error = -EINVAL;
1729 goto out_filestream_unmount;
1730 }
1731
1732 if (xfs_globals.always_cow) {
1733 xfs_info(mp, "using DEBUG-only always_cow mode.");
1734 mp->m_always_cow = true;
1735 }
1736 }
1737
1738 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1739 xfs_alert(mp,
1740 "reverse mapping btree not compatible with realtime device!");
1741 error = -EINVAL;
1742 goto out_filestream_unmount;
1743 }
1744
1745 if (xfs_has_exchange_range(mp))
1746 xfs_warn(mp,
1747 "EXPERIMENTAL exchange-range feature enabled. Use at your own risk!");
1748
1749 if (xfs_has_parent(mp))
1750 xfs_warn(mp,
1751 "EXPERIMENTAL parent pointer feature enabled. Use at your own risk!");
1752
1753 error = xfs_mountfs(mp);
1754 if (error)
1755 goto out_filestream_unmount;
1756
1757 root = igrab(VFS_I(mp->m_rootip));
1758 if (!root) {
1759 error = -ENOENT;
1760 goto out_unmount;
1761 }
1762 sb->s_root = d_make_root(root);
1763 if (!sb->s_root) {
1764 error = -ENOMEM;
1765 goto out_unmount;
1766 }
1767
1768 return 0;
1769
1770 out_filestream_unmount:
1771 xfs_filestream_unmount(mp);
1772 out_free_sb:
1773 xfs_freesb(mp);
1774 out_free_scrub_stats:
1775 xchk_mount_stats_free(mp);
1776 out_free_stats:
1777 free_percpu(mp->m_stats.xs_stats);
1778 out_destroy_inodegc:
1779 xfs_inodegc_free_percpu(mp);
1780 out_destroy_counters:
1781 xfs_destroy_percpu_counters(mp);
1782 out_destroy_workqueues:
1783 xfs_destroy_mount_workqueues(mp);
1784 out_shutdown_devices:
1785 xfs_shutdown_devices(mp);
1786 return error;
1787
1788 out_unmount:
1789 xfs_filestream_unmount(mp);
1790 xfs_unmountfs(mp);
1791 goto out_free_sb;
1792 }
1793
1794 static int
1795 xfs_fs_get_tree(
1796 struct fs_context *fc)
1797 {
1798 return get_tree_bdev(fc, xfs_fs_fill_super);
1799 }
1800
1801 static int
1802 xfs_remount_rw(
1803 struct xfs_mount *mp)
1804 {
1805 struct xfs_sb *sbp = &mp->m_sb;
1806 int error;
1807
1808 if (xfs_has_norecovery(mp)) {
1809 xfs_warn(mp,
1810 "ro->rw transition prohibited on norecovery mount");
1811 return -EINVAL;
1812 }
1813
1814 if (xfs_sb_is_v5(sbp) &&
1815 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1816 xfs_warn(mp,
1817 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1818 (sbp->sb_features_ro_compat &
1819 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1820 return -EINVAL;
1821 }
1822
1823 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1824
1825 /*
1826 * If this is the first remount to writeable state we might have some
1827 * superblock changes to update.
1828 */
1829 if (mp->m_update_sb) {
1830 error = xfs_sync_sb(mp, false);
1831 if (error) {
1832 xfs_warn(mp, "failed to write sb changes");
1833 return error;
1834 }
1835 mp->m_update_sb = false;
1836 }
1837
1838 /*
1839 * Fill out the reserve pool if it is empty. Use the stashed value if
1840 * it is non-zero, otherwise go with the default.
1841 */
1842 xfs_restore_resvblks(mp);
1843 xfs_log_work_queue(mp);
1844 xfs_blockgc_start(mp);
1845
1846 /* Create the per-AG metadata reservation pool .*/
1847 error = xfs_fs_reserve_ag_blocks(mp);
1848 if (error && error != -ENOSPC)
1849 return error;
1850
1851 /* Re-enable the background inode inactivation worker. */
1852 xfs_inodegc_start(mp);
1853
1854 return 0;
1855 }
1856
1857 static int
1858 xfs_remount_ro(
1859 struct xfs_mount *mp)
1860 {
1861 struct xfs_icwalk icw = {
1862 .icw_flags = XFS_ICWALK_FLAG_SYNC,
1863 };
1864 int error;
1865
1866 /* Flush all the dirty data to disk. */
1867 error = sync_filesystem(mp->m_super);
1868 if (error)
1869 return error;
1870
1871 /*
1872 * Cancel background eofb scanning so it cannot race with the final
1873 * log force+buftarg wait and deadlock the remount.
1874 */
1875 xfs_blockgc_stop(mp);
1876
1877 /*
1878 * Clear out all remaining COW staging extents and speculative post-EOF
1879 * preallocations so that we don't leave inodes requiring inactivation
1880 * cleanups during reclaim on a read-only mount. We must process every
1881 * cached inode, so this requires a synchronous cache scan.
1882 */
1883 error = xfs_blockgc_free_space(mp, &icw);
1884 if (error) {
1885 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1886 return error;
1887 }
1888
1889 /*
1890 * Stop the inodegc background worker. xfs_fs_reconfigure already
1891 * flushed all pending inodegc work when it sync'd the filesystem.
1892 * The VFS holds s_umount, so we know that inodes cannot enter
1893 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1894 * we send inodes straight to reclaim, so no inodes will be queued.
1895 */
1896 xfs_inodegc_stop(mp);
1897
1898 /* Free the per-AG metadata reservation pool. */
1899 xfs_fs_unreserve_ag_blocks(mp);
1900
1901 /*
1902 * Before we sync the metadata, we need to free up the reserve block
1903 * pool so that the used block count in the superblock on disk is
1904 * correct at the end of the remount. Stash the current* reserve pool
1905 * size so that if we get remounted rw, we can return it to the same
1906 * size.
1907 */
1908 xfs_save_resvblks(mp);
1909
1910 xfs_log_clean(mp);
1911 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1912
1913 return 0;
1914 }
1915
1916 /*
1917 * Logically we would return an error here to prevent users from believing
1918 * they might have changed mount options using remount which can't be changed.
1919 *
1920 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1921 * arguments in some cases so we can't blindly reject options, but have to
1922 * check for each specified option if it actually differs from the currently
1923 * set option and only reject it if that's the case.
1924 *
1925 * Until that is implemented we return success for every remount request, and
1926 * silently ignore all options that we can't actually change.
1927 */
1928 static int
1929 xfs_fs_reconfigure(
1930 struct fs_context *fc)
1931 {
1932 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1933 struct xfs_mount *new_mp = fc->s_fs_info;
1934 int flags = fc->sb_flags;
1935 int error;
1936
1937 /* version 5 superblocks always support version counters. */
1938 if (xfs_has_crc(mp))
1939 fc->sb_flags |= SB_I_VERSION;
1940
1941 error = xfs_fs_validate_params(new_mp);
1942 if (error)
1943 return error;
1944
1945 /* inode32 -> inode64 */
1946 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1947 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1948 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1949 }
1950
1951 /* inode64 -> inode32 */
1952 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1953 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1954 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1955 }
1956
1957 /* ro -> rw */
1958 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1959 error = xfs_remount_rw(mp);
1960 if (error)
1961 return error;
1962 }
1963
1964 /* rw -> ro */
1965 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1966 error = xfs_remount_ro(mp);
1967 if (error)
1968 return error;
1969 }
1970
1971 return 0;
1972 }
1973
1974 static void
1975 xfs_fs_free(
1976 struct fs_context *fc)
1977 {
1978 struct xfs_mount *mp = fc->s_fs_info;
1979
1980 /*
1981 * mp is stored in the fs_context when it is initialized.
1982 * mp is transferred to the superblock on a successful mount,
1983 * but if an error occurs before the transfer we have to free
1984 * it here.
1985 */
1986 if (mp)
1987 xfs_mount_free(mp);
1988 }
1989
1990 static const struct fs_context_operations xfs_context_ops = {
1991 .parse_param = xfs_fs_parse_param,
1992 .get_tree = xfs_fs_get_tree,
1993 .reconfigure = xfs_fs_reconfigure,
1994 .free = xfs_fs_free,
1995 };
1996
1997 /*
1998 * WARNING: do not initialise any parameters in this function that depend on
1999 * mount option parsing having already been performed as this can be called from
2000 * fsopen() before any parameters have been set.
2001 */
2002 static int xfs_init_fs_context(
2003 struct fs_context *fc)
2004 {
2005 struct xfs_mount *mp;
2006
2007 mp = kzalloc(sizeof(struct xfs_mount), GFP_KERNEL | __GFP_NOFAIL);
2008 if (!mp)
2009 return -ENOMEM;
2010
2011 spin_lock_init(&mp->m_sb_lock);
2012 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
2013 spin_lock_init(&mp->m_perag_lock);
2014 mutex_init(&mp->m_growlock);
2015 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
2016 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
2017 mp->m_kobj.kobject.kset = xfs_kset;
2018 /*
2019 * We don't create the finobt per-ag space reservation until after log
2020 * recovery, so we must set this to true so that an ifree transaction
2021 * started during log recovery will not depend on space reservations
2022 * for finobt expansion.
2023 */
2024 mp->m_finobt_nores = true;
2025
2026 /*
2027 * These can be overridden by the mount option parsing.
2028 */
2029 mp->m_logbufs = -1;
2030 mp->m_logbsize = -1;
2031 mp->m_allocsize_log = 16; /* 64k */
2032
2033 xfs_hooks_init(&mp->m_dir_update_hooks);
2034
2035 fc->s_fs_info = mp;
2036 fc->ops = &xfs_context_ops;
2037
2038 return 0;
2039 }
2040
2041 static void
2042 xfs_kill_sb(
2043 struct super_block *sb)
2044 {
2045 kill_block_super(sb);
2046 xfs_mount_free(XFS_M(sb));
2047 }
2048
2049 static struct file_system_type xfs_fs_type = {
2050 .owner = THIS_MODULE,
2051 .name = "xfs",
2052 .init_fs_context = xfs_init_fs_context,
2053 .parameters = xfs_fs_parameters,
2054 .kill_sb = xfs_kill_sb,
2055 .fs_flags = FS_REQUIRES_DEV | FS_ALLOW_IDMAP,
2056 };
2057 MODULE_ALIAS_FS("xfs");
2058
2059 STATIC int __init
2060 xfs_init_caches(void)
2061 {
2062 int error;
2063
2064 xfs_buf_cache = kmem_cache_create("xfs_buf", sizeof(struct xfs_buf), 0,
2065 SLAB_HWCACHE_ALIGN |
2066 SLAB_RECLAIM_ACCOUNT,
2067 NULL);
2068 if (!xfs_buf_cache)
2069 goto out;
2070
2071 xfs_log_ticket_cache = kmem_cache_create("xfs_log_ticket",
2072 sizeof(struct xlog_ticket),
2073 0, 0, NULL);
2074 if (!xfs_log_ticket_cache)
2075 goto out_destroy_buf_cache;
2076
2077 error = xfs_btree_init_cur_caches();
2078 if (error)
2079 goto out_destroy_log_ticket_cache;
2080
2081 error = rcbagbt_init_cur_cache();
2082 if (error)
2083 goto out_destroy_btree_cur_cache;
2084
2085 error = xfs_defer_init_item_caches();
2086 if (error)
2087 goto out_destroy_rcbagbt_cur_cache;
2088
2089 xfs_da_state_cache = kmem_cache_create("xfs_da_state",
2090 sizeof(struct xfs_da_state),
2091 0, 0, NULL);
2092 if (!xfs_da_state_cache)
2093 goto out_destroy_defer_item_cache;
2094
2095 xfs_ifork_cache = kmem_cache_create("xfs_ifork",
2096 sizeof(struct xfs_ifork),
2097 0, 0, NULL);
2098 if (!xfs_ifork_cache)
2099 goto out_destroy_da_state_cache;
2100
2101 xfs_trans_cache = kmem_cache_create("xfs_trans",
2102 sizeof(struct xfs_trans),
2103 0, 0, NULL);
2104 if (!xfs_trans_cache)
2105 goto out_destroy_ifork_cache;
2106
2107
2108 /*
2109 * The size of the cache-allocated buf log item is the maximum
2110 * size possible under XFS. This wastes a little bit of memory,
2111 * but it is much faster.
2112 */
2113 xfs_buf_item_cache = kmem_cache_create("xfs_buf_item",
2114 sizeof(struct xfs_buf_log_item),
2115 0, 0, NULL);
2116 if (!xfs_buf_item_cache)
2117 goto out_destroy_trans_cache;
2118
2119 xfs_efd_cache = kmem_cache_create("xfs_efd_item",
2120 xfs_efd_log_item_sizeof(XFS_EFD_MAX_FAST_EXTENTS),
2121 0, 0, NULL);
2122 if (!xfs_efd_cache)
2123 goto out_destroy_buf_item_cache;
2124
2125 xfs_efi_cache = kmem_cache_create("xfs_efi_item",
2126 xfs_efi_log_item_sizeof(XFS_EFI_MAX_FAST_EXTENTS),
2127 0, 0, NULL);
2128 if (!xfs_efi_cache)
2129 goto out_destroy_efd_cache;
2130
2131 xfs_inode_cache = kmem_cache_create("xfs_inode",
2132 sizeof(struct xfs_inode), 0,
2133 (SLAB_HWCACHE_ALIGN |
2134 SLAB_RECLAIM_ACCOUNT |
2135 SLAB_ACCOUNT),
2136 xfs_fs_inode_init_once);
2137 if (!xfs_inode_cache)
2138 goto out_destroy_efi_cache;
2139
2140 xfs_ili_cache = kmem_cache_create("xfs_ili",
2141 sizeof(struct xfs_inode_log_item), 0,
2142 SLAB_RECLAIM_ACCOUNT,
2143 NULL);
2144 if (!xfs_ili_cache)
2145 goto out_destroy_inode_cache;
2146
2147 xfs_icreate_cache = kmem_cache_create("xfs_icr",
2148 sizeof(struct xfs_icreate_item),
2149 0, 0, NULL);
2150 if (!xfs_icreate_cache)
2151 goto out_destroy_ili_cache;
2152
2153 xfs_rud_cache = kmem_cache_create("xfs_rud_item",
2154 sizeof(struct xfs_rud_log_item),
2155 0, 0, NULL);
2156 if (!xfs_rud_cache)
2157 goto out_destroy_icreate_cache;
2158
2159 xfs_rui_cache = kmem_cache_create("xfs_rui_item",
2160 xfs_rui_log_item_sizeof(XFS_RUI_MAX_FAST_EXTENTS),
2161 0, 0, NULL);
2162 if (!xfs_rui_cache)
2163 goto out_destroy_rud_cache;
2164
2165 xfs_cud_cache = kmem_cache_create("xfs_cud_item",
2166 sizeof(struct xfs_cud_log_item),
2167 0, 0, NULL);
2168 if (!xfs_cud_cache)
2169 goto out_destroy_rui_cache;
2170
2171 xfs_cui_cache = kmem_cache_create("xfs_cui_item",
2172 xfs_cui_log_item_sizeof(XFS_CUI_MAX_FAST_EXTENTS),
2173 0, 0, NULL);
2174 if (!xfs_cui_cache)
2175 goto out_destroy_cud_cache;
2176
2177 xfs_bud_cache = kmem_cache_create("xfs_bud_item",
2178 sizeof(struct xfs_bud_log_item),
2179 0, 0, NULL);
2180 if (!xfs_bud_cache)
2181 goto out_destroy_cui_cache;
2182
2183 xfs_bui_cache = kmem_cache_create("xfs_bui_item",
2184 xfs_bui_log_item_sizeof(XFS_BUI_MAX_FAST_EXTENTS),
2185 0, 0, NULL);
2186 if (!xfs_bui_cache)
2187 goto out_destroy_bud_cache;
2188
2189 xfs_attrd_cache = kmem_cache_create("xfs_attrd_item",
2190 sizeof(struct xfs_attrd_log_item),
2191 0, 0, NULL);
2192 if (!xfs_attrd_cache)
2193 goto out_destroy_bui_cache;
2194
2195 xfs_attri_cache = kmem_cache_create("xfs_attri_item",
2196 sizeof(struct xfs_attri_log_item),
2197 0, 0, NULL);
2198 if (!xfs_attri_cache)
2199 goto out_destroy_attrd_cache;
2200
2201 xfs_iunlink_cache = kmem_cache_create("xfs_iul_item",
2202 sizeof(struct xfs_iunlink_item),
2203 0, 0, NULL);
2204 if (!xfs_iunlink_cache)
2205 goto out_destroy_attri_cache;
2206
2207 xfs_xmd_cache = kmem_cache_create("xfs_xmd_item",
2208 sizeof(struct xfs_xmd_log_item),
2209 0, 0, NULL);
2210 if (!xfs_xmd_cache)
2211 goto out_destroy_iul_cache;
2212
2213 xfs_xmi_cache = kmem_cache_create("xfs_xmi_item",
2214 sizeof(struct xfs_xmi_log_item),
2215 0, 0, NULL);
2216 if (!xfs_xmi_cache)
2217 goto out_destroy_xmd_cache;
2218
2219 xfs_parent_args_cache = kmem_cache_create("xfs_parent_args",
2220 sizeof(struct xfs_parent_args),
2221 0, 0, NULL);
2222 if (!xfs_parent_args_cache)
2223 goto out_destroy_xmi_cache;
2224
2225 return 0;
2226
2227 out_destroy_xmi_cache:
2228 kmem_cache_destroy(xfs_xmi_cache);
2229 out_destroy_xmd_cache:
2230 kmem_cache_destroy(xfs_xmd_cache);
2231 out_destroy_iul_cache:
2232 kmem_cache_destroy(xfs_iunlink_cache);
2233 out_destroy_attri_cache:
2234 kmem_cache_destroy(xfs_attri_cache);
2235 out_destroy_attrd_cache:
2236 kmem_cache_destroy(xfs_attrd_cache);
2237 out_destroy_bui_cache:
2238 kmem_cache_destroy(xfs_bui_cache);
2239 out_destroy_bud_cache:
2240 kmem_cache_destroy(xfs_bud_cache);
2241 out_destroy_cui_cache:
2242 kmem_cache_destroy(xfs_cui_cache);
2243 out_destroy_cud_cache:
2244 kmem_cache_destroy(xfs_cud_cache);
2245 out_destroy_rui_cache:
2246 kmem_cache_destroy(xfs_rui_cache);
2247 out_destroy_rud_cache:
2248 kmem_cache_destroy(xfs_rud_cache);
2249 out_destroy_icreate_cache:
2250 kmem_cache_destroy(xfs_icreate_cache);
2251 out_destroy_ili_cache:
2252 kmem_cache_destroy(xfs_ili_cache);
2253 out_destroy_inode_cache:
2254 kmem_cache_destroy(xfs_inode_cache);
2255 out_destroy_efi_cache:
2256 kmem_cache_destroy(xfs_efi_cache);
2257 out_destroy_efd_cache:
2258 kmem_cache_destroy(xfs_efd_cache);
2259 out_destroy_buf_item_cache:
2260 kmem_cache_destroy(xfs_buf_item_cache);
2261 out_destroy_trans_cache:
2262 kmem_cache_destroy(xfs_trans_cache);
2263 out_destroy_ifork_cache:
2264 kmem_cache_destroy(xfs_ifork_cache);
2265 out_destroy_da_state_cache:
2266 kmem_cache_destroy(xfs_da_state_cache);
2267 out_destroy_defer_item_cache:
2268 xfs_defer_destroy_item_caches();
2269 out_destroy_rcbagbt_cur_cache:
2270 rcbagbt_destroy_cur_cache();
2271 out_destroy_btree_cur_cache:
2272 xfs_btree_destroy_cur_caches();
2273 out_destroy_log_ticket_cache:
2274 kmem_cache_destroy(xfs_log_ticket_cache);
2275 out_destroy_buf_cache:
2276 kmem_cache_destroy(xfs_buf_cache);
2277 out:
2278 return -ENOMEM;
2279 }
2280
2281 STATIC void
2282 xfs_destroy_caches(void)
2283 {
2284 /*
2285 * Make sure all delayed rcu free are flushed before we
2286 * destroy caches.
2287 */
2288 rcu_barrier();
2289 kmem_cache_destroy(xfs_parent_args_cache);
2290 kmem_cache_destroy(xfs_xmd_cache);
2291 kmem_cache_destroy(xfs_xmi_cache);
2292 kmem_cache_destroy(xfs_iunlink_cache);
2293 kmem_cache_destroy(xfs_attri_cache);
2294 kmem_cache_destroy(xfs_attrd_cache);
2295 kmem_cache_destroy(xfs_bui_cache);
2296 kmem_cache_destroy(xfs_bud_cache);
2297 kmem_cache_destroy(xfs_cui_cache);
2298 kmem_cache_destroy(xfs_cud_cache);
2299 kmem_cache_destroy(xfs_rui_cache);
2300 kmem_cache_destroy(xfs_rud_cache);
2301 kmem_cache_destroy(xfs_icreate_cache);
2302 kmem_cache_destroy(xfs_ili_cache);
2303 kmem_cache_destroy(xfs_inode_cache);
2304 kmem_cache_destroy(xfs_efi_cache);
2305 kmem_cache_destroy(xfs_efd_cache);
2306 kmem_cache_destroy(xfs_buf_item_cache);
2307 kmem_cache_destroy(xfs_trans_cache);
2308 kmem_cache_destroy(xfs_ifork_cache);
2309 kmem_cache_destroy(xfs_da_state_cache);
2310 xfs_defer_destroy_item_caches();
2311 rcbagbt_destroy_cur_cache();
2312 xfs_btree_destroy_cur_caches();
2313 kmem_cache_destroy(xfs_log_ticket_cache);
2314 kmem_cache_destroy(xfs_buf_cache);
2315 }
2316
2317 STATIC int __init
2318 xfs_init_workqueues(void)
2319 {
2320 /*
2321 * The allocation workqueue can be used in memory reclaim situations
2322 * (writepage path), and parallelism is only limited by the number of
2323 * AGs in all the filesystems mounted. Hence use the default large
2324 * max_active value for this workqueue.
2325 */
2326 xfs_alloc_wq = alloc_workqueue("xfsalloc",
2327 XFS_WQFLAGS(WQ_MEM_RECLAIM | WQ_FREEZABLE), 0);
2328 if (!xfs_alloc_wq)
2329 return -ENOMEM;
2330
2331 xfs_discard_wq = alloc_workqueue("xfsdiscard", XFS_WQFLAGS(WQ_UNBOUND),
2332 0);
2333 if (!xfs_discard_wq)
2334 goto out_free_alloc_wq;
2335
2336 return 0;
2337 out_free_alloc_wq:
2338 destroy_workqueue(xfs_alloc_wq);
2339 return -ENOMEM;
2340 }
2341
2342 STATIC void
2343 xfs_destroy_workqueues(void)
2344 {
2345 destroy_workqueue(xfs_discard_wq);
2346 destroy_workqueue(xfs_alloc_wq);
2347 }
2348
2349 STATIC int __init
2350 init_xfs_fs(void)
2351 {
2352 int error;
2353
2354 xfs_check_ondisk_structs();
2355
2356 error = xfs_dahash_test();
2357 if (error)
2358 return error;
2359
2360 printk(KERN_INFO XFS_VERSION_STRING " with "
2361 XFS_BUILD_OPTIONS " enabled\n");
2362
2363 xfs_dir_startup();
2364
2365 error = xfs_init_caches();
2366 if (error)
2367 goto out;
2368
2369 error = xfs_init_workqueues();
2370 if (error)
2371 goto out_destroy_caches;
2372
2373 error = xfs_mru_cache_init();
2374 if (error)
2375 goto out_destroy_wq;
2376
2377 error = xfs_init_procfs();
2378 if (error)
2379 goto out_mru_cache_uninit;
2380
2381 error = xfs_sysctl_register();
2382 if (error)
2383 goto out_cleanup_procfs;
2384
2385 xfs_debugfs = xfs_debugfs_mkdir("xfs", NULL);
2386
2387 xfs_kset = kset_create_and_add("xfs", NULL, fs_kobj);
2388 if (!xfs_kset) {
2389 error = -ENOMEM;
2390 goto out_debugfs_unregister;
2391 }
2392
2393 xfsstats.xs_kobj.kobject.kset = xfs_kset;
2394
2395 xfsstats.xs_stats = alloc_percpu(struct xfsstats);
2396 if (!xfsstats.xs_stats) {
2397 error = -ENOMEM;
2398 goto out_kset_unregister;
2399 }
2400
2401 error = xfs_sysfs_init(&xfsstats.xs_kobj, &xfs_stats_ktype, NULL,
2402 "stats");
2403 if (error)
2404 goto out_free_stats;
2405
2406 error = xchk_global_stats_setup(xfs_debugfs);
2407 if (error)
2408 goto out_remove_stats_kobj;
2409
2410 #ifdef DEBUG
2411 xfs_dbg_kobj.kobject.kset = xfs_kset;
2412 error = xfs_sysfs_init(&xfs_dbg_kobj, &xfs_dbg_ktype, NULL, "debug");
2413 if (error)
2414 goto out_remove_scrub_stats;
2415 #endif
2416
2417 error = xfs_qm_init();
2418 if (error)
2419 goto out_remove_dbg_kobj;
2420
2421 error = register_filesystem(&xfs_fs_type);
2422 if (error)
2423 goto out_qm_exit;
2424 return 0;
2425
2426 out_qm_exit:
2427 xfs_qm_exit();
2428 out_remove_dbg_kobj:
2429 #ifdef DEBUG
2430 xfs_sysfs_del(&xfs_dbg_kobj);
2431 out_remove_scrub_stats:
2432 #endif
2433 xchk_global_stats_teardown();
2434 out_remove_stats_kobj:
2435 xfs_sysfs_del(&xfsstats.xs_kobj);
2436 out_free_stats:
2437 free_percpu(xfsstats.xs_stats);
2438 out_kset_unregister:
2439 kset_unregister(xfs_kset);
2440 out_debugfs_unregister:
2441 debugfs_remove(xfs_debugfs);
2442 xfs_sysctl_unregister();
2443 out_cleanup_procfs:
2444 xfs_cleanup_procfs();
2445 out_mru_cache_uninit:
2446 xfs_mru_cache_uninit();
2447 out_destroy_wq:
2448 xfs_destroy_workqueues();
2449 out_destroy_caches:
2450 xfs_destroy_caches();
2451 out:
2452 return error;
2453 }
2454
2455 STATIC void __exit
2456 exit_xfs_fs(void)
2457 {
2458 xfs_qm_exit();
2459 unregister_filesystem(&xfs_fs_type);
2460 #ifdef DEBUG
2461 xfs_sysfs_del(&xfs_dbg_kobj);
2462 #endif
2463 xchk_global_stats_teardown();
2464 xfs_sysfs_del(&xfsstats.xs_kobj);
2465 free_percpu(xfsstats.xs_stats);
2466 kset_unregister(xfs_kset);
2467 debugfs_remove(xfs_debugfs);
2468 xfs_sysctl_unregister();
2469 xfs_cleanup_procfs();
2470 xfs_mru_cache_uninit();
2471 xfs_destroy_workqueues();
2472 xfs_destroy_caches();
2473 xfs_uuid_table_free();
2474 }
2475
2476 module_init(init_xfs_fs);
2477 module_exit(exit_xfs_fs);
2478
2479 MODULE_AUTHOR("Silicon Graphics, Inc.");
2480 MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled");
2481 MODULE_LICENSE("GPL");
2482
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.