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