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
xfs_mount_set_dax_mode(struct xfs_mount * mp,enum xfs_dax_mode mode)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
xfs_fs_show_options(struct seq_file * m,struct dentry * root)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
xfs_set_inode_alloc_perag(struct xfs_perag * pag,xfs_ino_t ino,xfs_agnumber_t max_metadata)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
xfs_set_inode_alloc(struct xfs_mount * mp,xfs_agnumber_t agcount)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
xfs_setup_dax_always(struct xfs_mount * mp)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
xfs_blkdev_get(xfs_mount_t * mp,const char * name,struct block_device ** bdevp)361 xfs_blkdev_get(
362 xfs_mount_t *mp,
363 const char *name,
364 struct block_device **bdevp)
365 {
366 int error = 0;
367
368 *bdevp = blkdev_get_by_path(name, BLK_OPEN_READ | BLK_OPEN_WRITE,
369 mp->m_super, &fs_holder_ops);
370 if (IS_ERR(*bdevp)) {
371 error = PTR_ERR(*bdevp);
372 xfs_warn(mp, "Invalid device [%s], error=%d", name, error);
373 }
374
375 return error;
376 }
377
378 STATIC void
xfs_shutdown_devices(struct xfs_mount * mp)379 xfs_shutdown_devices(
380 struct xfs_mount *mp)
381 {
382 /*
383 * Udev is triggered whenever anyone closes a block device or unmounts
384 * a file systemm on a block device.
385 * The default udev rules invoke blkid to read the fs super and create
386 * symlinks to the bdev under /dev/disk. For this, it uses buffered
387 * reads through the page cache.
388 *
389 * xfs_db also uses buffered reads to examine metadata. There is no
390 * coordination between xfs_db and udev, which means that they can run
391 * concurrently. Note there is no coordination between the kernel and
392 * blkid either.
393 *
394 * On a system with 64k pages, the page cache can cache the superblock
395 * and the root inode (and hence the root directory) with the same 64k
396 * page. If udev spawns blkid after the mkfs and the system is busy
397 * enough that it is still running when xfs_db starts up, they'll both
398 * read from the same page in the pagecache.
399 *
400 * The unmount writes updated inode metadata to disk directly. The XFS
401 * buffer cache does not use the bdev pagecache, so it needs to
402 * invalidate that pagecache on unmount. If the above scenario occurs,
403 * the pagecache no longer reflects what's on disk, xfs_db reads the
404 * stale metadata, and fails to find /a. Most of the time this succeeds
405 * because closing a bdev invalidates the page cache, but when processes
406 * race, everyone loses.
407 */
408 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
409 blkdev_issue_flush(mp->m_logdev_targp->bt_bdev);
410 invalidate_bdev(mp->m_logdev_targp->bt_bdev);
411 }
412 if (mp->m_rtdev_targp) {
413 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
414 invalidate_bdev(mp->m_rtdev_targp->bt_bdev);
415 }
416 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
417 invalidate_bdev(mp->m_ddev_targp->bt_bdev);
418 }
419
420 /*
421 * The file system configurations are:
422 * (1) device (partition) with data and internal log
423 * (2) logical volume with data and log subvolumes.
424 * (3) logical volume with data, log, and realtime subvolumes.
425 *
426 * We only have to handle opening the log and realtime volumes here if
427 * they are present. The data subvolume has already been opened by
428 * get_sb_bdev() and is stored in sb->s_bdev.
429 */
430 STATIC int
xfs_open_devices(struct xfs_mount * mp)431 xfs_open_devices(
432 struct xfs_mount *mp)
433 {
434 struct super_block *sb = mp->m_super;
435 struct block_device *ddev = sb->s_bdev;
436 struct block_device *logdev = NULL, *rtdev = NULL;
437 int error;
438
439 /*
440 * blkdev_put() can't be called under s_umount, see the comment
441 * in get_tree_bdev() for more details
442 */
443 up_write(&sb->s_umount);
444
445 /*
446 * Open real time and log devices - order is important.
447 */
448 if (mp->m_logname) {
449 error = xfs_blkdev_get(mp, mp->m_logname, &logdev);
450 if (error)
451 goto out_relock;
452 }
453
454 if (mp->m_rtname) {
455 error = xfs_blkdev_get(mp, mp->m_rtname, &rtdev);
456 if (error)
457 goto out_close_logdev;
458
459 if (rtdev == ddev || rtdev == logdev) {
460 xfs_warn(mp,
461 "Cannot mount filesystem with identical rtdev and ddev/logdev.");
462 error = -EINVAL;
463 goto out_close_rtdev;
464 }
465 }
466
467 /*
468 * Setup xfs_mount buffer target pointers
469 */
470 error = -ENOMEM;
471 mp->m_ddev_targp = xfs_alloc_buftarg(mp, ddev);
472 if (!mp->m_ddev_targp)
473 goto out_close_rtdev;
474
475 if (rtdev) {
476 mp->m_rtdev_targp = xfs_alloc_buftarg(mp, rtdev);
477 if (!mp->m_rtdev_targp)
478 goto out_free_ddev_targ;
479 }
480
481 if (logdev && logdev != ddev) {
482 mp->m_logdev_targp = xfs_alloc_buftarg(mp, logdev);
483 if (!mp->m_logdev_targp)
484 goto out_free_rtdev_targ;
485 } else {
486 mp->m_logdev_targp = mp->m_ddev_targp;
487 }
488
489 error = 0;
490 out_relock:
491 down_write(&sb->s_umount);
492 return error;
493
494 out_free_rtdev_targ:
495 if (mp->m_rtdev_targp)
496 xfs_free_buftarg(mp->m_rtdev_targp);
497 out_free_ddev_targ:
498 xfs_free_buftarg(mp->m_ddev_targp);
499 out_close_rtdev:
500 if (rtdev)
501 blkdev_put(rtdev, sb);
502 out_close_logdev:
503 if (logdev && logdev != ddev)
504 blkdev_put(logdev, sb);
505 goto out_relock;
506 }
507
508 /*
509 * Setup xfs_mount buffer target pointers based on superblock
510 */
511 STATIC int
xfs_setup_devices(struct xfs_mount * mp)512 xfs_setup_devices(
513 struct xfs_mount *mp)
514 {
515 int error;
516
517 error = xfs_setsize_buftarg(mp->m_ddev_targp, mp->m_sb.sb_sectsize);
518 if (error)
519 return error;
520
521 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp) {
522 unsigned int log_sector_size = BBSIZE;
523
524 if (xfs_has_sector(mp))
525 log_sector_size = mp->m_sb.sb_logsectsize;
526 error = xfs_setsize_buftarg(mp->m_logdev_targp,
527 log_sector_size);
528 if (error)
529 return error;
530 }
531 if (mp->m_rtdev_targp) {
532 error = xfs_setsize_buftarg(mp->m_rtdev_targp,
533 mp->m_sb.sb_sectsize);
534 if (error)
535 return error;
536 }
537
538 return 0;
539 }
540
541 STATIC int
xfs_init_mount_workqueues(struct xfs_mount * mp)542 xfs_init_mount_workqueues(
543 struct xfs_mount *mp)
544 {
545 mp->m_buf_workqueue = alloc_workqueue("xfs-buf/%s",
546 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
547 1, mp->m_super->s_id);
548 if (!mp->m_buf_workqueue)
549 goto out;
550
551 mp->m_unwritten_workqueue = alloc_workqueue("xfs-conv/%s",
552 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
553 0, mp->m_super->s_id);
554 if (!mp->m_unwritten_workqueue)
555 goto out_destroy_buf;
556
557 mp->m_reclaim_workqueue = alloc_workqueue("xfs-reclaim/%s",
558 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
559 0, mp->m_super->s_id);
560 if (!mp->m_reclaim_workqueue)
561 goto out_destroy_unwritten;
562
563 mp->m_blockgc_wq = alloc_workqueue("xfs-blockgc/%s",
564 XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
565 0, mp->m_super->s_id);
566 if (!mp->m_blockgc_wq)
567 goto out_destroy_reclaim;
568
569 mp->m_inodegc_wq = alloc_workqueue("xfs-inodegc/%s",
570 XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
571 1, mp->m_super->s_id);
572 if (!mp->m_inodegc_wq)
573 goto out_destroy_blockgc;
574
575 mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
576 XFS_WQFLAGS(WQ_FREEZABLE), 0, mp->m_super->s_id);
577 if (!mp->m_sync_workqueue)
578 goto out_destroy_inodegc;
579
580 return 0;
581
582 out_destroy_inodegc:
583 destroy_workqueue(mp->m_inodegc_wq);
584 out_destroy_blockgc:
585 destroy_workqueue(mp->m_blockgc_wq);
586 out_destroy_reclaim:
587 destroy_workqueue(mp->m_reclaim_workqueue);
588 out_destroy_unwritten:
589 destroy_workqueue(mp->m_unwritten_workqueue);
590 out_destroy_buf:
591 destroy_workqueue(mp->m_buf_workqueue);
592 out:
593 return -ENOMEM;
594 }
595
596 STATIC void
xfs_destroy_mount_workqueues(struct xfs_mount * mp)597 xfs_destroy_mount_workqueues(
598 struct xfs_mount *mp)
599 {
600 destroy_workqueue(mp->m_sync_workqueue);
601 destroy_workqueue(mp->m_blockgc_wq);
602 destroy_workqueue(mp->m_inodegc_wq);
603 destroy_workqueue(mp->m_reclaim_workqueue);
604 destroy_workqueue(mp->m_unwritten_workqueue);
605 destroy_workqueue(mp->m_buf_workqueue);
606 }
607
608 static void
xfs_flush_inodes_worker(struct work_struct * work)609 xfs_flush_inodes_worker(
610 struct work_struct *work)
611 {
612 struct xfs_mount *mp = container_of(work, struct xfs_mount,
613 m_flush_inodes_work);
614 struct super_block *sb = mp->m_super;
615
616 if (down_read_trylock(&sb->s_umount)) {
617 sync_inodes_sb(sb);
618 up_read(&sb->s_umount);
619 }
620 }
621
622 /*
623 * Flush all dirty data to disk. Must not be called while holding an XFS_ILOCK
624 * or a page lock. We use sync_inodes_sb() here to ensure we block while waiting
625 * for IO to complete so that we effectively throttle multiple callers to the
626 * rate at which IO is completing.
627 */
628 void
xfs_flush_inodes(struct xfs_mount * mp)629 xfs_flush_inodes(
630 struct xfs_mount *mp)
631 {
632 /*
633 * If flush_work() returns true then that means we waited for a flush
634 * which was already in progress. Don't bother running another scan.
635 */
636 if (flush_work(&mp->m_flush_inodes_work))
637 return;
638
639 queue_work(mp->m_sync_workqueue, &mp->m_flush_inodes_work);
640 flush_work(&mp->m_flush_inodes_work);
641 }
642
643 /* Catch misguided souls that try to use this interface on XFS */
644 STATIC struct inode *
xfs_fs_alloc_inode(struct super_block * sb)645 xfs_fs_alloc_inode(
646 struct super_block *sb)
647 {
648 BUG();
649 return NULL;
650 }
651
652 /*
653 * Now that the generic code is guaranteed not to be accessing
654 * the linux inode, we can inactivate and reclaim the inode.
655 */
656 STATIC void
xfs_fs_destroy_inode(struct inode * inode)657 xfs_fs_destroy_inode(
658 struct inode *inode)
659 {
660 struct xfs_inode *ip = XFS_I(inode);
661
662 trace_xfs_destroy_inode(ip);
663
664 ASSERT(!rwsem_is_locked(&inode->i_rwsem));
665 XFS_STATS_INC(ip->i_mount, vn_rele);
666 XFS_STATS_INC(ip->i_mount, vn_remove);
667 xfs_inode_mark_reclaimable(ip);
668 }
669
670 static void
xfs_fs_dirty_inode(struct inode * inode,int flags)671 xfs_fs_dirty_inode(
672 struct inode *inode,
673 int flags)
674 {
675 struct xfs_inode *ip = XFS_I(inode);
676 struct xfs_mount *mp = ip->i_mount;
677 struct xfs_trans *tp;
678
679 if (!(inode->i_sb->s_flags & SB_LAZYTIME))
680 return;
681
682 /*
683 * Only do the timestamp update if the inode is dirty (I_DIRTY_SYNC)
684 * and has dirty timestamp (I_DIRTY_TIME). I_DIRTY_TIME can be passed
685 * in flags possibly together with I_DIRTY_SYNC.
686 */
687 if ((flags & ~I_DIRTY_TIME) != I_DIRTY_SYNC || !(flags & I_DIRTY_TIME))
688 return;
689
690 if (xfs_trans_alloc(mp, &M_RES(mp)->tr_fsyncts, 0, 0, 0, &tp))
691 return;
692 xfs_ilock(ip, XFS_ILOCK_EXCL);
693 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
694 xfs_trans_log_inode(tp, ip, XFS_ILOG_TIMESTAMP);
695 xfs_trans_commit(tp);
696 }
697
698 /*
699 * Slab object creation initialisation for the XFS inode.
700 * This covers only the idempotent fields in the XFS inode;
701 * all other fields need to be initialised on allocation
702 * from the slab. This avoids the need to repeatedly initialise
703 * fields in the xfs inode that left in the initialise state
704 * when freeing the inode.
705 */
706 STATIC void
xfs_fs_inode_init_once(void * inode)707 xfs_fs_inode_init_once(
708 void *inode)
709 {
710 struct xfs_inode *ip = inode;
711
712 memset(ip, 0, sizeof(struct xfs_inode));
713
714 /* vfs inode */
715 inode_init_once(VFS_I(ip));
716
717 /* xfs inode */
718 atomic_set(&ip->i_pincount, 0);
719 spin_lock_init(&ip->i_flags_lock);
720
721 mrlock_init(&ip->i_lock, MRLOCK_ALLOW_EQUAL_PRI|MRLOCK_BARRIER,
722 "xfsino", ip->i_ino);
723 }
724
725 /*
726 * We do an unlocked check for XFS_IDONTCACHE here because we are already
727 * serialised against cache hits here via the inode->i_lock and igrab() in
728 * xfs_iget_cache_hit(). Hence a lookup that might clear this flag will not be
729 * racing with us, and it avoids needing to grab a spinlock here for every inode
730 * we drop the final reference on.
731 */
732 STATIC int
xfs_fs_drop_inode(struct inode * inode)733 xfs_fs_drop_inode(
734 struct inode *inode)
735 {
736 struct xfs_inode *ip = XFS_I(inode);
737
738 /*
739 * If this unlinked inode is in the middle of recovery, don't
740 * drop the inode just yet; log recovery will take care of
741 * that. See the comment for this inode flag.
742 */
743 if (ip->i_flags & XFS_IRECOVERY) {
744 ASSERT(xlog_recovery_needed(ip->i_mount->m_log));
745 return 0;
746 }
747
748 return generic_drop_inode(inode);
749 }
750
751 static void
xfs_mount_free(struct xfs_mount * mp)752 xfs_mount_free(
753 struct xfs_mount *mp)
754 {
755 /*
756 * Free the buftargs here because blkdev_put needs to be called outside
757 * of sb->s_umount, which is held around the call to ->put_super.
758 */
759 if (mp->m_logdev_targp && mp->m_logdev_targp != mp->m_ddev_targp)
760 xfs_free_buftarg(mp->m_logdev_targp);
761 if (mp->m_rtdev_targp)
762 xfs_free_buftarg(mp->m_rtdev_targp);
763 if (mp->m_ddev_targp)
764 xfs_free_buftarg(mp->m_ddev_targp);
765
766 debugfs_remove(mp->m_debugfs);
767 kfree(mp->m_rtname);
768 kfree(mp->m_logname);
769 kmem_free(mp);
770 }
771
772 STATIC int
xfs_fs_sync_fs(struct super_block * sb,int wait)773 xfs_fs_sync_fs(
774 struct super_block *sb,
775 int wait)
776 {
777 struct xfs_mount *mp = XFS_M(sb);
778 int error;
779
780 trace_xfs_fs_sync_fs(mp, __return_address);
781
782 /*
783 * Doing anything during the async pass would be counterproductive.
784 */
785 if (!wait)
786 return 0;
787
788 error = xfs_log_force(mp, XFS_LOG_SYNC);
789 if (error)
790 return error;
791
792 if (laptop_mode) {
793 /*
794 * The disk must be active because we're syncing.
795 * We schedule log work now (now that the disk is
796 * active) instead of later (when it might not be).
797 */
798 flush_delayed_work(&mp->m_log->l_work);
799 }
800
801 /*
802 * If we are called with page faults frozen out, it means we are about
803 * to freeze the transaction subsystem. Take the opportunity to shut
804 * down inodegc because once SB_FREEZE_FS is set it's too late to
805 * prevent inactivation races with freeze. The fs doesn't get called
806 * again by the freezing process until after SB_FREEZE_FS has been set,
807 * so it's now or never. Same logic applies to speculative allocation
808 * garbage collection.
809 *
810 * We don't care if this is a normal syncfs call that does this or
811 * freeze that does this - we can run this multiple times without issue
812 * and we won't race with a restart because a restart can only occur
813 * when the state is either SB_FREEZE_FS or SB_FREEZE_COMPLETE.
814 */
815 if (sb->s_writers.frozen == SB_FREEZE_PAGEFAULT) {
816 xfs_inodegc_stop(mp);
817 xfs_blockgc_stop(mp);
818 }
819
820 return 0;
821 }
822
823 STATIC int
xfs_fs_statfs(struct dentry * dentry,struct kstatfs * statp)824 xfs_fs_statfs(
825 struct dentry *dentry,
826 struct kstatfs *statp)
827 {
828 struct xfs_mount *mp = XFS_M(dentry->d_sb);
829 xfs_sb_t *sbp = &mp->m_sb;
830 struct xfs_inode *ip = XFS_I(d_inode(dentry));
831 uint64_t fakeinos, id;
832 uint64_t icount;
833 uint64_t ifree;
834 uint64_t fdblocks;
835 xfs_extlen_t lsize;
836 int64_t ffree;
837
838 /*
839 * Expedite background inodegc but don't wait. We do not want to block
840 * here waiting hours for a billion extent file to be truncated.
841 */
842 xfs_inodegc_push(mp);
843
844 statp->f_type = XFS_SUPER_MAGIC;
845 statp->f_namelen = MAXNAMELEN - 1;
846
847 id = huge_encode_dev(mp->m_ddev_targp->bt_dev);
848 statp->f_fsid = u64_to_fsid(id);
849
850 icount = percpu_counter_sum(&mp->m_icount);
851 ifree = percpu_counter_sum(&mp->m_ifree);
852 fdblocks = percpu_counter_sum(&mp->m_fdblocks);
853
854 spin_lock(&mp->m_sb_lock);
855 statp->f_bsize = sbp->sb_blocksize;
856 lsize = sbp->sb_logstart ? sbp->sb_logblocks : 0;
857 statp->f_blocks = sbp->sb_dblocks - lsize;
858 spin_unlock(&mp->m_sb_lock);
859
860 /* make sure statp->f_bfree does not underflow */
861 statp->f_bfree = max_t(int64_t, 0,
862 fdblocks - xfs_fdblocks_unavailable(mp));
863 statp->f_bavail = statp->f_bfree;
864
865 fakeinos = XFS_FSB_TO_INO(mp, statp->f_bfree);
866 statp->f_files = min(icount + fakeinos, (uint64_t)XFS_MAXINUMBER);
867 if (M_IGEO(mp)->maxicount)
868 statp->f_files = min_t(typeof(statp->f_files),
869 statp->f_files,
870 M_IGEO(mp)->maxicount);
871
872 /* If sb_icount overshot maxicount, report actual allocation */
873 statp->f_files = max_t(typeof(statp->f_files),
874 statp->f_files,
875 sbp->sb_icount);
876
877 /* make sure statp->f_ffree does not underflow */
878 ffree = statp->f_files - (icount - ifree);
879 statp->f_ffree = max_t(int64_t, ffree, 0);
880
881 if (XFS_IS_REALTIME_MOUNT(mp) &&
882 (ip->i_diflags & (XFS_DIFLAG_RTINHERIT | XFS_DIFLAG_REALTIME))) {
883 s64 freertx;
884
885 statp->f_blocks = sbp->sb_rblocks;
886 freertx = percpu_counter_sum_positive(&mp->m_frextents);
887 statp->f_bavail = statp->f_bfree = freertx * sbp->sb_rextsize;
888 }
889
890 if ((ip->i_diflags & XFS_DIFLAG_PROJINHERIT) &&
891 ((mp->m_qflags & (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))) ==
892 (XFS_PQUOTA_ACCT|XFS_PQUOTA_ENFD))
893 xfs_qm_statvfs(ip, statp);
894
895 return 0;
896 }
897
898 STATIC void
xfs_save_resvblks(struct xfs_mount * mp)899 xfs_save_resvblks(struct xfs_mount *mp)
900 {
901 uint64_t resblks = 0;
902
903 mp->m_resblks_save = mp->m_resblks;
904 xfs_reserve_blocks(mp, &resblks, NULL);
905 }
906
907 STATIC void
xfs_restore_resvblks(struct xfs_mount * mp)908 xfs_restore_resvblks(struct xfs_mount *mp)
909 {
910 uint64_t resblks;
911
912 if (mp->m_resblks_save) {
913 resblks = mp->m_resblks_save;
914 mp->m_resblks_save = 0;
915 } else
916 resblks = xfs_default_resblks(mp);
917
918 xfs_reserve_blocks(mp, &resblks, NULL);
919 }
920
921 /*
922 * Second stage of a freeze. The data is already frozen so we only
923 * need to take care of the metadata. Once that's done sync the superblock
924 * to the log to dirty it in case of a crash while frozen. This ensures that we
925 * will recover the unlinked inode lists on the next mount.
926 */
927 STATIC int
xfs_fs_freeze(struct super_block * sb)928 xfs_fs_freeze(
929 struct super_block *sb)
930 {
931 struct xfs_mount *mp = XFS_M(sb);
932 unsigned int flags;
933 int ret;
934
935 /*
936 * The filesystem is now frozen far enough that memory reclaim
937 * cannot safely operate on the filesystem. Hence we need to
938 * set a GFP_NOFS context here to avoid recursion deadlocks.
939 */
940 flags = memalloc_nofs_save();
941 xfs_save_resvblks(mp);
942 ret = xfs_log_quiesce(mp);
943 memalloc_nofs_restore(flags);
944
945 /*
946 * For read-write filesystems, we need to restart the inodegc on error
947 * because we stopped it at SB_FREEZE_PAGEFAULT level and a thaw is not
948 * going to be run to restart it now. We are at SB_FREEZE_FS level
949 * here, so we can restart safely without racing with a stop in
950 * xfs_fs_sync_fs().
951 */
952 if (ret && !xfs_is_readonly(mp)) {
953 xfs_blockgc_start(mp);
954 xfs_inodegc_start(mp);
955 }
956
957 return ret;
958 }
959
960 STATIC int
xfs_fs_unfreeze(struct super_block * sb)961 xfs_fs_unfreeze(
962 struct super_block *sb)
963 {
964 struct xfs_mount *mp = XFS_M(sb);
965
966 xfs_restore_resvblks(mp);
967 xfs_log_work_queue(mp);
968
969 /*
970 * Don't reactivate the inodegc worker on a readonly filesystem because
971 * inodes are sent directly to reclaim. Don't reactivate the blockgc
972 * worker because there are no speculative preallocations on a readonly
973 * filesystem.
974 */
975 if (!xfs_is_readonly(mp)) {
976 xfs_blockgc_start(mp);
977 xfs_inodegc_start(mp);
978 }
979
980 return 0;
981 }
982
983 /*
984 * This function fills in xfs_mount_t fields based on mount args.
985 * Note: the superblock _has_ now been read in.
986 */
987 STATIC int
xfs_finish_flags(struct xfs_mount * mp)988 xfs_finish_flags(
989 struct xfs_mount *mp)
990 {
991 /* Fail a mount where the logbuf is smaller than the log stripe */
992 if (xfs_has_logv2(mp)) {
993 if (mp->m_logbsize <= 0 &&
994 mp->m_sb.sb_logsunit > XLOG_BIG_RECORD_BSIZE) {
995 mp->m_logbsize = mp->m_sb.sb_logsunit;
996 } else if (mp->m_logbsize > 0 &&
997 mp->m_logbsize < mp->m_sb.sb_logsunit) {
998 xfs_warn(mp,
999 "logbuf size must be greater than or equal to log stripe size");
1000 return -EINVAL;
1001 }
1002 } else {
1003 /* Fail a mount if the logbuf is larger than 32K */
1004 if (mp->m_logbsize > XLOG_BIG_RECORD_BSIZE) {
1005 xfs_warn(mp,
1006 "logbuf size for version 1 logs must be 16K or 32K");
1007 return -EINVAL;
1008 }
1009 }
1010
1011 /*
1012 * V5 filesystems always use attr2 format for attributes.
1013 */
1014 if (xfs_has_crc(mp) && xfs_has_noattr2(mp)) {
1015 xfs_warn(mp, "Cannot mount a V5 filesystem as noattr2. "
1016 "attr2 is always enabled for V5 filesystems.");
1017 return -EINVAL;
1018 }
1019
1020 /*
1021 * prohibit r/w mounts of read-only filesystems
1022 */
1023 if ((mp->m_sb.sb_flags & XFS_SBF_READONLY) && !xfs_is_readonly(mp)) {
1024 xfs_warn(mp,
1025 "cannot mount a read-only filesystem as read-write");
1026 return -EROFS;
1027 }
1028
1029 if ((mp->m_qflags & XFS_GQUOTA_ACCT) &&
1030 (mp->m_qflags & XFS_PQUOTA_ACCT) &&
1031 !xfs_has_pquotino(mp)) {
1032 xfs_warn(mp,
1033 "Super block does not support project and group quota together");
1034 return -EINVAL;
1035 }
1036
1037 return 0;
1038 }
1039
1040 static int
xfs_init_percpu_counters(struct xfs_mount * mp)1041 xfs_init_percpu_counters(
1042 struct xfs_mount *mp)
1043 {
1044 int error;
1045
1046 error = percpu_counter_init(&mp->m_icount, 0, GFP_KERNEL);
1047 if (error)
1048 return -ENOMEM;
1049
1050 error = percpu_counter_init(&mp->m_ifree, 0, GFP_KERNEL);
1051 if (error)
1052 goto free_icount;
1053
1054 error = percpu_counter_init(&mp->m_fdblocks, 0, GFP_KERNEL);
1055 if (error)
1056 goto free_ifree;
1057
1058 error = percpu_counter_init(&mp->m_delalloc_blks, 0, GFP_KERNEL);
1059 if (error)
1060 goto free_fdblocks;
1061
1062 error = percpu_counter_init(&mp->m_frextents, 0, GFP_KERNEL);
1063 if (error)
1064 goto free_delalloc;
1065
1066 return 0;
1067
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
xfs_reinit_percpu_counters(struct xfs_mount * mp)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
xfs_destroy_percpu_counters(struct xfs_mount * mp)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_blks) == 0);
1098 percpu_counter_destroy(&mp->m_delalloc_blks);
1099 percpu_counter_destroy(&mp->m_frextents);
1100 }
1101
1102 static int
xfs_inodegc_init_percpu(struct xfs_mount * mp)1103 xfs_inodegc_init_percpu(
1104 struct xfs_mount *mp)
1105 {
1106 struct xfs_inodegc *gc;
1107 int cpu;
1108
1109 mp->m_inodegc = alloc_percpu(struct xfs_inodegc);
1110 if (!mp->m_inodegc)
1111 return -ENOMEM;
1112
1113 for_each_possible_cpu(cpu) {
1114 gc = per_cpu_ptr(mp->m_inodegc, cpu);
1115 gc->cpu = cpu;
1116 gc->mp = mp;
1117 init_llist_head(&gc->list);
1118 gc->items = 0;
1119 gc->error = 0;
1120 INIT_DELAYED_WORK(&gc->work, xfs_inodegc_worker);
1121 }
1122 return 0;
1123 }
1124
1125 static void
xfs_inodegc_free_percpu(struct xfs_mount * mp)1126 xfs_inodegc_free_percpu(
1127 struct xfs_mount *mp)
1128 {
1129 if (!mp->m_inodegc)
1130 return;
1131 free_percpu(mp->m_inodegc);
1132 }
1133
1134 static void
xfs_fs_put_super(struct super_block * sb)1135 xfs_fs_put_super(
1136 struct super_block *sb)
1137 {
1138 struct xfs_mount *mp = XFS_M(sb);
1139
1140 xfs_notice(mp, "Unmounting Filesystem %pU", &mp->m_sb.sb_uuid);
1141 xfs_filestream_unmount(mp);
1142 xfs_unmountfs(mp);
1143
1144 xfs_freesb(mp);
1145 xchk_mount_stats_free(mp);
1146 free_percpu(mp->m_stats.xs_stats);
1147 xfs_inodegc_free_percpu(mp);
1148 xfs_destroy_percpu_counters(mp);
1149 xfs_destroy_mount_workqueues(mp);
1150 xfs_shutdown_devices(mp);
1151 }
1152
1153 static long
xfs_fs_nr_cached_objects(struct super_block * sb,struct shrink_control * sc)1154 xfs_fs_nr_cached_objects(
1155 struct super_block *sb,
1156 struct shrink_control *sc)
1157 {
1158 /* Paranoia: catch incorrect calls during mount setup or teardown */
1159 if (WARN_ON_ONCE(!sb->s_fs_info))
1160 return 0;
1161 return xfs_reclaim_inodes_count(XFS_M(sb));
1162 }
1163
1164 static long
xfs_fs_free_cached_objects(struct super_block * sb,struct shrink_control * sc)1165 xfs_fs_free_cached_objects(
1166 struct super_block *sb,
1167 struct shrink_control *sc)
1168 {
1169 return xfs_reclaim_inodes_nr(XFS_M(sb), sc->nr_to_scan);
1170 }
1171
1172 static void
xfs_fs_shutdown(struct super_block * sb)1173 xfs_fs_shutdown(
1174 struct super_block *sb)
1175 {
1176 xfs_force_shutdown(XFS_M(sb), SHUTDOWN_DEVICE_REMOVED);
1177 }
1178
1179 static const struct super_operations xfs_super_operations = {
1180 .alloc_inode = xfs_fs_alloc_inode,
1181 .destroy_inode = xfs_fs_destroy_inode,
1182 .dirty_inode = xfs_fs_dirty_inode,
1183 .drop_inode = xfs_fs_drop_inode,
1184 .put_super = xfs_fs_put_super,
1185 .sync_fs = xfs_fs_sync_fs,
1186 .freeze_fs = xfs_fs_freeze,
1187 .unfreeze_fs = xfs_fs_unfreeze,
1188 .statfs = xfs_fs_statfs,
1189 .show_options = xfs_fs_show_options,
1190 .nr_cached_objects = xfs_fs_nr_cached_objects,
1191 .free_cached_objects = xfs_fs_free_cached_objects,
1192 .shutdown = xfs_fs_shutdown,
1193 };
1194
1195 static int
suffix_kstrtoint(const char * s,unsigned int base,int * res)1196 suffix_kstrtoint(
1197 const char *s,
1198 unsigned int base,
1199 int *res)
1200 {
1201 int last, shift_left_factor = 0, _res;
1202 char *value;
1203 int ret = 0;
1204
1205 value = kstrdup(s, GFP_KERNEL);
1206 if (!value)
1207 return -ENOMEM;
1208
1209 last = strlen(value) - 1;
1210 if (value[last] == 'K' || value[last] == 'k') {
1211 shift_left_factor = 10;
1212 value[last] = '\0';
1213 }
1214 if (value[last] == 'M' || value[last] == 'm') {
1215 shift_left_factor = 20;
1216 value[last] = '\0';
1217 }
1218 if (value[last] == 'G' || value[last] == 'g') {
1219 shift_left_factor = 30;
1220 value[last] = '\0';
1221 }
1222
1223 if (kstrtoint(value, base, &_res))
1224 ret = -EINVAL;
1225 kfree(value);
1226 *res = _res << shift_left_factor;
1227 return ret;
1228 }
1229
1230 static inline void
xfs_fs_warn_deprecated(struct fs_context * fc,struct fs_parameter * param,uint64_t flag,bool value)1231 xfs_fs_warn_deprecated(
1232 struct fs_context *fc,
1233 struct fs_parameter *param,
1234 uint64_t flag,
1235 bool value)
1236 {
1237 /* Don't print the warning if reconfiguring and current mount point
1238 * already had the flag set
1239 */
1240 if ((fc->purpose & FS_CONTEXT_FOR_RECONFIGURE) &&
1241 !!(XFS_M(fc->root->d_sb)->m_features & flag) == value)
1242 return;
1243 xfs_warn(fc->s_fs_info, "%s mount option is deprecated.", param->key);
1244 }
1245
1246 /*
1247 * Set mount state from a mount option.
1248 *
1249 * NOTE: mp->m_super is NULL here!
1250 */
1251 static int
xfs_fs_parse_param(struct fs_context * fc,struct fs_parameter * param)1252 xfs_fs_parse_param(
1253 struct fs_context *fc,
1254 struct fs_parameter *param)
1255 {
1256 struct xfs_mount *parsing_mp = fc->s_fs_info;
1257 struct fs_parse_result result;
1258 int size = 0;
1259 int opt;
1260
1261 opt = fs_parse(fc, xfs_fs_parameters, param, &result);
1262 if (opt < 0)
1263 return opt;
1264
1265 switch (opt) {
1266 case Opt_logbufs:
1267 parsing_mp->m_logbufs = result.uint_32;
1268 return 0;
1269 case Opt_logbsize:
1270 if (suffix_kstrtoint(param->string, 10, &parsing_mp->m_logbsize))
1271 return -EINVAL;
1272 return 0;
1273 case Opt_logdev:
1274 kfree(parsing_mp->m_logname);
1275 parsing_mp->m_logname = kstrdup(param->string, GFP_KERNEL);
1276 if (!parsing_mp->m_logname)
1277 return -ENOMEM;
1278 return 0;
1279 case Opt_rtdev:
1280 kfree(parsing_mp->m_rtname);
1281 parsing_mp->m_rtname = kstrdup(param->string, GFP_KERNEL);
1282 if (!parsing_mp->m_rtname)
1283 return -ENOMEM;
1284 return 0;
1285 case Opt_allocsize:
1286 if (suffix_kstrtoint(param->string, 10, &size))
1287 return -EINVAL;
1288 parsing_mp->m_allocsize_log = ffs(size) - 1;
1289 parsing_mp->m_features |= XFS_FEAT_ALLOCSIZE;
1290 return 0;
1291 case Opt_grpid:
1292 case Opt_bsdgroups:
1293 parsing_mp->m_features |= XFS_FEAT_GRPID;
1294 return 0;
1295 case Opt_nogrpid:
1296 case Opt_sysvgroups:
1297 parsing_mp->m_features &= ~XFS_FEAT_GRPID;
1298 return 0;
1299 case Opt_wsync:
1300 parsing_mp->m_features |= XFS_FEAT_WSYNC;
1301 return 0;
1302 case Opt_norecovery:
1303 parsing_mp->m_features |= XFS_FEAT_NORECOVERY;
1304 return 0;
1305 case Opt_noalign:
1306 parsing_mp->m_features |= XFS_FEAT_NOALIGN;
1307 return 0;
1308 case Opt_swalloc:
1309 parsing_mp->m_features |= XFS_FEAT_SWALLOC;
1310 return 0;
1311 case Opt_sunit:
1312 parsing_mp->m_dalign = result.uint_32;
1313 return 0;
1314 case Opt_swidth:
1315 parsing_mp->m_swidth = result.uint_32;
1316 return 0;
1317 case Opt_inode32:
1318 parsing_mp->m_features |= XFS_FEAT_SMALL_INUMS;
1319 return 0;
1320 case Opt_inode64:
1321 parsing_mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1322 return 0;
1323 case Opt_nouuid:
1324 parsing_mp->m_features |= XFS_FEAT_NOUUID;
1325 return 0;
1326 case Opt_largeio:
1327 parsing_mp->m_features |= XFS_FEAT_LARGE_IOSIZE;
1328 return 0;
1329 case Opt_nolargeio:
1330 parsing_mp->m_features &= ~XFS_FEAT_LARGE_IOSIZE;
1331 return 0;
1332 case Opt_filestreams:
1333 parsing_mp->m_features |= XFS_FEAT_FILESTREAMS;
1334 return 0;
1335 case Opt_noquota:
1336 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ACCT;
1337 parsing_mp->m_qflags &= ~XFS_ALL_QUOTA_ENFD;
1338 return 0;
1339 case Opt_quota:
1340 case Opt_uquota:
1341 case Opt_usrquota:
1342 parsing_mp->m_qflags |= (XFS_UQUOTA_ACCT | XFS_UQUOTA_ENFD);
1343 return 0;
1344 case Opt_qnoenforce:
1345 case Opt_uqnoenforce:
1346 parsing_mp->m_qflags |= XFS_UQUOTA_ACCT;
1347 parsing_mp->m_qflags &= ~XFS_UQUOTA_ENFD;
1348 return 0;
1349 case Opt_pquota:
1350 case Opt_prjquota:
1351 parsing_mp->m_qflags |= (XFS_PQUOTA_ACCT | XFS_PQUOTA_ENFD);
1352 return 0;
1353 case Opt_pqnoenforce:
1354 parsing_mp->m_qflags |= XFS_PQUOTA_ACCT;
1355 parsing_mp->m_qflags &= ~XFS_PQUOTA_ENFD;
1356 return 0;
1357 case Opt_gquota:
1358 case Opt_grpquota:
1359 parsing_mp->m_qflags |= (XFS_GQUOTA_ACCT | XFS_GQUOTA_ENFD);
1360 return 0;
1361 case Opt_gqnoenforce:
1362 parsing_mp->m_qflags |= XFS_GQUOTA_ACCT;
1363 parsing_mp->m_qflags &= ~XFS_GQUOTA_ENFD;
1364 return 0;
1365 case Opt_discard:
1366 parsing_mp->m_features |= XFS_FEAT_DISCARD;
1367 return 0;
1368 case Opt_nodiscard:
1369 parsing_mp->m_features &= ~XFS_FEAT_DISCARD;
1370 return 0;
1371 #ifdef CONFIG_FS_DAX
1372 case Opt_dax:
1373 xfs_mount_set_dax_mode(parsing_mp, XFS_DAX_ALWAYS);
1374 return 0;
1375 case Opt_dax_enum:
1376 xfs_mount_set_dax_mode(parsing_mp, result.uint_32);
1377 return 0;
1378 #endif
1379 /* Following mount options will be removed in September 2025 */
1380 case Opt_ikeep:
1381 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, true);
1382 parsing_mp->m_features |= XFS_FEAT_IKEEP;
1383 return 0;
1384 case Opt_noikeep:
1385 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_IKEEP, false);
1386 parsing_mp->m_features &= ~XFS_FEAT_IKEEP;
1387 return 0;
1388 case Opt_attr2:
1389 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_ATTR2, true);
1390 parsing_mp->m_features |= XFS_FEAT_ATTR2;
1391 return 0;
1392 case Opt_noattr2:
1393 xfs_fs_warn_deprecated(fc, param, XFS_FEAT_NOATTR2, true);
1394 parsing_mp->m_features |= XFS_FEAT_NOATTR2;
1395 return 0;
1396 default:
1397 xfs_warn(parsing_mp, "unknown mount option [%s].", param->key);
1398 return -EINVAL;
1399 }
1400
1401 return 0;
1402 }
1403
1404 static int
xfs_fs_validate_params(struct xfs_mount * mp)1405 xfs_fs_validate_params(
1406 struct xfs_mount *mp)
1407 {
1408 /* No recovery flag requires a read-only mount */
1409 if (xfs_has_norecovery(mp) && !xfs_is_readonly(mp)) {
1410 xfs_warn(mp, "no-recovery mounts must be read-only.");
1411 return -EINVAL;
1412 }
1413
1414 /*
1415 * We have not read the superblock at this point, so only the attr2
1416 * mount option can set the attr2 feature by this stage.
1417 */
1418 if (xfs_has_attr2(mp) && xfs_has_noattr2(mp)) {
1419 xfs_warn(mp, "attr2 and noattr2 cannot both be specified.");
1420 return -EINVAL;
1421 }
1422
1423
1424 if (xfs_has_noalign(mp) && (mp->m_dalign || mp->m_swidth)) {
1425 xfs_warn(mp,
1426 "sunit and swidth options incompatible with the noalign option");
1427 return -EINVAL;
1428 }
1429
1430 if (!IS_ENABLED(CONFIG_XFS_QUOTA) && mp->m_qflags != 0) {
1431 xfs_warn(mp, "quota support not available in this kernel.");
1432 return -EINVAL;
1433 }
1434
1435 if ((mp->m_dalign && !mp->m_swidth) ||
1436 (!mp->m_dalign && mp->m_swidth)) {
1437 xfs_warn(mp, "sunit and swidth must be specified together");
1438 return -EINVAL;
1439 }
1440
1441 if (mp->m_dalign && (mp->m_swidth % mp->m_dalign != 0)) {
1442 xfs_warn(mp,
1443 "stripe width (%d) must be a multiple of the stripe unit (%d)",
1444 mp->m_swidth, mp->m_dalign);
1445 return -EINVAL;
1446 }
1447
1448 if (mp->m_logbufs != -1 &&
1449 mp->m_logbufs != 0 &&
1450 (mp->m_logbufs < XLOG_MIN_ICLOGS ||
1451 mp->m_logbufs > XLOG_MAX_ICLOGS)) {
1452 xfs_warn(mp, "invalid logbufs value: %d [not %d-%d]",
1453 mp->m_logbufs, XLOG_MIN_ICLOGS, XLOG_MAX_ICLOGS);
1454 return -EINVAL;
1455 }
1456
1457 if (mp->m_logbsize != -1 &&
1458 mp->m_logbsize != 0 &&
1459 (mp->m_logbsize < XLOG_MIN_RECORD_BSIZE ||
1460 mp->m_logbsize > XLOG_MAX_RECORD_BSIZE ||
1461 !is_power_of_2(mp->m_logbsize))) {
1462 xfs_warn(mp,
1463 "invalid logbufsize: %d [not 16k,32k,64k,128k or 256k]",
1464 mp->m_logbsize);
1465 return -EINVAL;
1466 }
1467
1468 if (xfs_has_allocsize(mp) &&
1469 (mp->m_allocsize_log > XFS_MAX_IO_LOG ||
1470 mp->m_allocsize_log < XFS_MIN_IO_LOG)) {
1471 xfs_warn(mp, "invalid log iosize: %d [not %d-%d]",
1472 mp->m_allocsize_log, XFS_MIN_IO_LOG, XFS_MAX_IO_LOG);
1473 return -EINVAL;
1474 }
1475
1476 return 0;
1477 }
1478
1479 struct dentry *
xfs_debugfs_mkdir(const char * name,struct dentry * parent)1480 xfs_debugfs_mkdir(
1481 const char *name,
1482 struct dentry *parent)
1483 {
1484 struct dentry *child;
1485
1486 /* Apparently we're expected to ignore error returns?? */
1487 child = debugfs_create_dir(name, parent);
1488 if (IS_ERR(child))
1489 return NULL;
1490
1491 return child;
1492 }
1493
1494 static int
xfs_fs_fill_super(struct super_block * sb,struct fs_context * fc)1495 xfs_fs_fill_super(
1496 struct super_block *sb,
1497 struct fs_context *fc)
1498 {
1499 struct xfs_mount *mp = sb->s_fs_info;
1500 struct inode *root;
1501 int flags = 0, error;
1502
1503 mp->m_super = sb;
1504
1505 /*
1506 * Copy VFS mount flags from the context now that all parameter parsing
1507 * is guaranteed to have been completed by either the old mount API or
1508 * the newer fsopen/fsconfig API.
1509 */
1510 if (fc->sb_flags & SB_RDONLY)
1511 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1512 if (fc->sb_flags & SB_DIRSYNC)
1513 mp->m_features |= XFS_FEAT_DIRSYNC;
1514 if (fc->sb_flags & SB_SYNCHRONOUS)
1515 mp->m_features |= XFS_FEAT_WSYNC;
1516
1517 error = xfs_fs_validate_params(mp);
1518 if (error)
1519 return error;
1520
1521 sb_min_blocksize(sb, BBSIZE);
1522 sb->s_xattr = xfs_xattr_handlers;
1523 sb->s_export_op = &xfs_export_operations;
1524 #ifdef CONFIG_XFS_QUOTA
1525 sb->s_qcop = &xfs_quotactl_operations;
1526 sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP | QTYPE_MASK_PRJ;
1527 #endif
1528 sb->s_op = &xfs_super_operations;
1529
1530 /*
1531 * Delay mount work if the debug hook is set. This is debug
1532 * instrumention to coordinate simulation of xfs mount failures with
1533 * VFS superblock operations
1534 */
1535 if (xfs_globals.mount_delay) {
1536 xfs_notice(mp, "Delaying mount for %d seconds.",
1537 xfs_globals.mount_delay);
1538 msleep(xfs_globals.mount_delay * 1000);
1539 }
1540
1541 if (fc->sb_flags & SB_SILENT)
1542 flags |= XFS_MFSI_QUIET;
1543
1544 error = xfs_open_devices(mp);
1545 if (error)
1546 return error;
1547
1548 if (xfs_debugfs) {
1549 mp->m_debugfs = xfs_debugfs_mkdir(mp->m_super->s_id,
1550 xfs_debugfs);
1551 } else {
1552 mp->m_debugfs = NULL;
1553 }
1554
1555 error = xfs_init_mount_workqueues(mp);
1556 if (error)
1557 goto out_shutdown_devices;
1558
1559 error = xfs_init_percpu_counters(mp);
1560 if (error)
1561 goto out_destroy_workqueues;
1562
1563 error = xfs_inodegc_init_percpu(mp);
1564 if (error)
1565 goto out_destroy_counters;
1566
1567 /* Allocate stats memory before we do operations that might use it */
1568 mp->m_stats.xs_stats = alloc_percpu(struct xfsstats);
1569 if (!mp->m_stats.xs_stats) {
1570 error = -ENOMEM;
1571 goto out_destroy_inodegc;
1572 }
1573
1574 error = xchk_mount_stats_alloc(mp);
1575 if (error)
1576 goto out_free_stats;
1577
1578 error = xfs_readsb(mp, flags);
1579 if (error)
1580 goto out_free_scrub_stats;
1581
1582 error = xfs_finish_flags(mp);
1583 if (error)
1584 goto out_free_sb;
1585
1586 error = xfs_setup_devices(mp);
1587 if (error)
1588 goto out_free_sb;
1589
1590 /* V4 support is undergoing deprecation. */
1591 if (!xfs_has_crc(mp)) {
1592 #ifdef CONFIG_XFS_SUPPORT_V4
1593 xfs_warn_once(mp,
1594 "Deprecated V4 format (crc=0) will not be supported after September 2030.");
1595 #else
1596 xfs_warn(mp,
1597 "Deprecated V4 format (crc=0) not supported by kernel.");
1598 error = -EINVAL;
1599 goto out_free_sb;
1600 #endif
1601 }
1602
1603 /* ASCII case insensitivity is undergoing deprecation. */
1604 if (xfs_has_asciici(mp)) {
1605 #ifdef CONFIG_XFS_SUPPORT_ASCII_CI
1606 xfs_warn_once(mp,
1607 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) will not be supported after September 2030.");
1608 #else
1609 xfs_warn(mp,
1610 "Deprecated ASCII case-insensitivity feature (ascii-ci=1) not supported by kernel.");
1611 error = -EINVAL;
1612 goto out_free_sb;
1613 #endif
1614 }
1615
1616 /* Filesystem claims it needs repair, so refuse the mount. */
1617 if (xfs_has_needsrepair(mp)) {
1618 xfs_warn(mp, "Filesystem needs repair. Please run xfs_repair.");
1619 error = -EFSCORRUPTED;
1620 goto out_free_sb;
1621 }
1622
1623 /*
1624 * Don't touch the filesystem if a user tool thinks it owns the primary
1625 * superblock. mkfs doesn't clear the flag from secondary supers, so
1626 * we don't check them at all.
1627 */
1628 if (mp->m_sb.sb_inprogress) {
1629 xfs_warn(mp, "Offline file system operation in progress!");
1630 error = -EFSCORRUPTED;
1631 goto out_free_sb;
1632 }
1633
1634 /*
1635 * Until this is fixed only page-sized or smaller data blocks work.
1636 */
1637 if (mp->m_sb.sb_blocksize > PAGE_SIZE) {
1638 xfs_warn(mp,
1639 "File system with blocksize %d bytes. "
1640 "Only pagesize (%ld) or less will currently work.",
1641 mp->m_sb.sb_blocksize, PAGE_SIZE);
1642 error = -ENOSYS;
1643 goto out_free_sb;
1644 }
1645
1646 /* Ensure this filesystem fits in the page cache limits */
1647 if (xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_dblocks) ||
1648 xfs_sb_validate_fsb_count(&mp->m_sb, mp->m_sb.sb_rblocks)) {
1649 xfs_warn(mp,
1650 "file system too large to be mounted on this system.");
1651 error = -EFBIG;
1652 goto out_free_sb;
1653 }
1654
1655 /*
1656 * XFS block mappings use 54 bits to store the logical block offset.
1657 * This should suffice to handle the maximum file size that the VFS
1658 * supports (currently 2^63 bytes on 64-bit and ULONG_MAX << PAGE_SHIFT
1659 * bytes on 32-bit), but as XFS and VFS have gotten the s_maxbytes
1660 * calculation wrong on 32-bit kernels in the past, we'll add a WARN_ON
1661 * to check this assertion.
1662 *
1663 * Avoid integer overflow by comparing the maximum bmbt offset to the
1664 * maximum pagecache offset in units of fs blocks.
1665 */
1666 if (!xfs_verify_fileoff(mp, XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE))) {
1667 xfs_warn(mp,
1668 "MAX_LFS_FILESIZE block offset (%llu) exceeds extent map maximum (%llu)!",
1669 XFS_B_TO_FSBT(mp, MAX_LFS_FILESIZE),
1670 XFS_MAX_FILEOFF);
1671 error = -EINVAL;
1672 goto out_free_sb;
1673 }
1674
1675 error = xfs_filestream_mount(mp);
1676 if (error)
1677 goto out_free_sb;
1678
1679 /*
1680 * we must configure the block size in the superblock before we run the
1681 * full mount process as the mount process can lookup and cache inodes.
1682 */
1683 sb->s_magic = XFS_SUPER_MAGIC;
1684 sb->s_blocksize = mp->m_sb.sb_blocksize;
1685 sb->s_blocksize_bits = ffs(sb->s_blocksize) - 1;
1686 sb->s_maxbytes = MAX_LFS_FILESIZE;
1687 sb->s_max_links = XFS_MAXLINK;
1688 sb->s_time_gran = 1;
1689 if (xfs_has_bigtime(mp)) {
1690 sb->s_time_min = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MIN);
1691 sb->s_time_max = xfs_bigtime_to_unix(XFS_BIGTIME_TIME_MAX);
1692 } else {
1693 sb->s_time_min = XFS_LEGACY_TIME_MIN;
1694 sb->s_time_max = XFS_LEGACY_TIME_MAX;
1695 }
1696 trace_xfs_inode_timestamp_range(mp, sb->s_time_min, sb->s_time_max);
1697 sb->s_iflags |= SB_I_CGROUPWB;
1698
1699 set_posix_acl_flag(sb);
1700
1701 /* version 5 superblocks support inode version counters. */
1702 if (xfs_has_crc(mp))
1703 sb->s_flags |= SB_I_VERSION;
1704
1705 if (xfs_has_dax_always(mp)) {
1706 error = xfs_setup_dax_always(mp);
1707 if (error)
1708 goto out_filestream_unmount;
1709 }
1710
1711 if (xfs_has_discard(mp) && !bdev_max_discard_sectors(sb->s_bdev)) {
1712 xfs_warn(mp,
1713 "mounting with \"discard\" option, but the device does not support discard");
1714 mp->m_features &= ~XFS_FEAT_DISCARD;
1715 }
1716
1717 if (xfs_has_reflink(mp)) {
1718 if (mp->m_sb.sb_rblocks) {
1719 xfs_alert(mp,
1720 "reflink not compatible with realtime device!");
1721 error = -EINVAL;
1722 goto out_filestream_unmount;
1723 }
1724
1725 if (xfs_globals.always_cow) {
1726 xfs_info(mp, "using DEBUG-only always_cow mode.");
1727 mp->m_always_cow = true;
1728 }
1729 }
1730
1731 if (xfs_has_rmapbt(mp) && mp->m_sb.sb_rblocks) {
1732 xfs_alert(mp,
1733 "reverse mapping btree not compatible with realtime device!");
1734 error = -EINVAL;
1735 goto out_filestream_unmount;
1736 }
1737
1738 error = xfs_mountfs(mp);
1739 if (error)
1740 goto out_filestream_unmount;
1741
1742 root = igrab(VFS_I(mp->m_rootip));
1743 if (!root) {
1744 error = -ENOENT;
1745 goto out_unmount;
1746 }
1747 sb->s_root = d_make_root(root);
1748 if (!sb->s_root) {
1749 error = -ENOMEM;
1750 goto out_unmount;
1751 }
1752
1753 return 0;
1754
1755 out_filestream_unmount:
1756 xfs_filestream_unmount(mp);
1757 out_free_sb:
1758 xfs_freesb(mp);
1759 out_free_scrub_stats:
1760 xchk_mount_stats_free(mp);
1761 out_free_stats:
1762 free_percpu(mp->m_stats.xs_stats);
1763 out_destroy_inodegc:
1764 xfs_inodegc_free_percpu(mp);
1765 out_destroy_counters:
1766 xfs_destroy_percpu_counters(mp);
1767 out_destroy_workqueues:
1768 xfs_destroy_mount_workqueues(mp);
1769 out_shutdown_devices:
1770 xfs_shutdown_devices(mp);
1771 return error;
1772
1773 out_unmount:
1774 xfs_filestream_unmount(mp);
1775 xfs_unmountfs(mp);
1776 goto out_free_sb;
1777 }
1778
1779 static int
xfs_fs_get_tree(struct fs_context * fc)1780 xfs_fs_get_tree(
1781 struct fs_context *fc)
1782 {
1783 return get_tree_bdev(fc, xfs_fs_fill_super);
1784 }
1785
1786 static int
xfs_remount_rw(struct xfs_mount * mp)1787 xfs_remount_rw(
1788 struct xfs_mount *mp)
1789 {
1790 struct xfs_sb *sbp = &mp->m_sb;
1791 int error;
1792
1793 if (xfs_has_norecovery(mp)) {
1794 xfs_warn(mp,
1795 "ro->rw transition prohibited on norecovery mount");
1796 return -EINVAL;
1797 }
1798
1799 if (xfs_sb_is_v5(sbp) &&
1800 xfs_sb_has_ro_compat_feature(sbp, XFS_SB_FEAT_RO_COMPAT_UNKNOWN)) {
1801 xfs_warn(mp,
1802 "ro->rw transition prohibited on unknown (0x%x) ro-compat filesystem",
1803 (sbp->sb_features_ro_compat &
1804 XFS_SB_FEAT_RO_COMPAT_UNKNOWN));
1805 return -EINVAL;
1806 }
1807
1808 clear_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1809
1810 /*
1811 * If this is the first remount to writeable state we might have some
1812 * superblock changes to update.
1813 */
1814 if (mp->m_update_sb) {
1815 error = xfs_sync_sb(mp, false);
1816 if (error) {
1817 xfs_warn(mp, "failed to write sb changes");
1818 return error;
1819 }
1820 mp->m_update_sb = false;
1821 }
1822
1823 /*
1824 * Fill out the reserve pool if it is empty. Use the stashed value if
1825 * it is non-zero, otherwise go with the default.
1826 */
1827 xfs_restore_resvblks(mp);
1828 xfs_log_work_queue(mp);
1829 xfs_blockgc_start(mp);
1830
1831 /* Create the per-AG metadata reservation pool .*/
1832 error = xfs_fs_reserve_ag_blocks(mp);
1833 if (error && error != -ENOSPC)
1834 return error;
1835
1836 /* Re-enable the background inode inactivation worker. */
1837 xfs_inodegc_start(mp);
1838
1839 return 0;
1840 }
1841
1842 static int
xfs_remount_ro(struct xfs_mount * mp)1843 xfs_remount_ro(
1844 struct xfs_mount *mp)
1845 {
1846 struct xfs_icwalk icw = {
1847 .icw_flags = XFS_ICWALK_FLAG_SYNC,
1848 };
1849 int error;
1850
1851 /* Flush all the dirty data to disk. */
1852 error = sync_filesystem(mp->m_super);
1853 if (error)
1854 return error;
1855
1856 /*
1857 * Cancel background eofb scanning so it cannot race with the final
1858 * log force+buftarg wait and deadlock the remount.
1859 */
1860 xfs_blockgc_stop(mp);
1861
1862 /*
1863 * Clear out all remaining COW staging extents and speculative post-EOF
1864 * preallocations so that we don't leave inodes requiring inactivation
1865 * cleanups during reclaim on a read-only mount. We must process every
1866 * cached inode, so this requires a synchronous cache scan.
1867 */
1868 error = xfs_blockgc_free_space(mp, &icw);
1869 if (error) {
1870 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1871 return error;
1872 }
1873
1874 /*
1875 * Stop the inodegc background worker. xfs_fs_reconfigure already
1876 * flushed all pending inodegc work when it sync'd the filesystem.
1877 * The VFS holds s_umount, so we know that inodes cannot enter
1878 * xfs_fs_destroy_inode during a remount operation. In readonly mode
1879 * we send inodes straight to reclaim, so no inodes will be queued.
1880 */
1881 xfs_inodegc_stop(mp);
1882
1883 /* Free the per-AG metadata reservation pool. */
1884 error = xfs_fs_unreserve_ag_blocks(mp);
1885 if (error) {
1886 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
1887 return error;
1888 }
1889
1890 /*
1891 * Before we sync the metadata, we need to free up the reserve block
1892 * pool so that the used block count in the superblock on disk is
1893 * correct at the end of the remount. Stash the current* reserve pool
1894 * size so that if we get remounted rw, we can return it to the same
1895 * size.
1896 */
1897 xfs_save_resvblks(mp);
1898
1899 xfs_log_clean(mp);
1900 set_bit(XFS_OPSTATE_READONLY, &mp->m_opstate);
1901
1902 return 0;
1903 }
1904
1905 /*
1906 * Logically we would return an error here to prevent users from believing
1907 * they might have changed mount options using remount which can't be changed.
1908 *
1909 * But unfortunately mount(8) adds all options from mtab and fstab to the mount
1910 * arguments in some cases so we can't blindly reject options, but have to
1911 * check for each specified option if it actually differs from the currently
1912 * set option and only reject it if that's the case.
1913 *
1914 * Until that is implemented we return success for every remount request, and
1915 * silently ignore all options that we can't actually change.
1916 */
1917 static int
xfs_fs_reconfigure(struct fs_context * fc)1918 xfs_fs_reconfigure(
1919 struct fs_context *fc)
1920 {
1921 struct xfs_mount *mp = XFS_M(fc->root->d_sb);
1922 struct xfs_mount *new_mp = fc->s_fs_info;
1923 int flags = fc->sb_flags;
1924 int error;
1925
1926 /* version 5 superblocks always support version counters. */
1927 if (xfs_has_crc(mp))
1928 fc->sb_flags |= SB_I_VERSION;
1929
1930 error = xfs_fs_validate_params(new_mp);
1931 if (error)
1932 return error;
1933
1934 /* inode32 -> inode64 */
1935 if (xfs_has_small_inums(mp) && !xfs_has_small_inums(new_mp)) {
1936 mp->m_features &= ~XFS_FEAT_SMALL_INUMS;
1937 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1938 }
1939
1940 /* inode64 -> inode32 */
1941 if (!xfs_has_small_inums(mp) && xfs_has_small_inums(new_mp)) {
1942 mp->m_features |= XFS_FEAT_SMALL_INUMS;
1943 mp->m_maxagi = xfs_set_inode_alloc(mp, mp->m_sb.sb_agcount);
1944 }
1945
1946 /* ro -> rw */
1947 if (xfs_is_readonly(mp) && !(flags & SB_RDONLY)) {
1948 error = xfs_remount_rw(mp);
1949 if (error)
1950 return error;
1951 }
1952
1953 /* rw -> ro */
1954 if (!xfs_is_readonly(mp) && (flags & SB_RDONLY)) {
1955 error = xfs_remount_ro(mp);
1956 if (error)
1957 return error;
1958 }
1959
1960 return 0;
1961 }
1962
1963 static void
xfs_fs_free(struct fs_context * fc)1964 xfs_fs_free(
1965 struct fs_context *fc)
1966 {
1967 struct xfs_mount *mp = fc->s_fs_info;
1968
1969 /*
1970 * mp is stored in the fs_context when it is initialized.
1971 * mp is transferred to the superblock on a successful mount,
1972 * but if an error occurs before the transfer we have to free
1973 * it here.
1974 */
1975 if (mp)
1976 xfs_mount_free(mp);
1977 }
1978
1979 static const struct fs_context_operations xfs_context_ops = {
1980 .parse_param = xfs_fs_parse_param,
1981 .get_tree = xfs_fs_get_tree,
1982 .reconfigure = xfs_fs_reconfigure,
1983 .free = xfs_fs_free,
1984 };
1985
1986 /*
1987 * WARNING: do not initialise any parameters in this function that depend on
1988 * mount option parsing having already been performed as this can be called from
1989 * fsopen() before any parameters have been set.
1990 */
xfs_init_fs_context(struct fs_context * fc)1991 static int xfs_init_fs_context(
1992 struct fs_context *fc)
1993 {
1994 struct xfs_mount *mp;
1995
1996 mp = kmem_alloc(sizeof(struct xfs_mount), KM_ZERO);
1997 if (!mp)
1998 return -ENOMEM;
1999
2000 spin_lock_init(&mp->m_sb_lock);
2001 INIT_RADIX_TREE(&mp->m_perag_tree, GFP_ATOMIC);
2002 spin_lock_init(&mp->m_perag_lock);
2003 mutex_init(&mp->m_growlock);
2004 INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
2005 INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
2006 mp->m_kobj.kobject.kset = xfs_kset;
2007 /*
2008 * We don't create the finobt per-ag space reservation until after log
2009 * recovery, so we must set this to true so that an ifree transaction
2010 * started during log recovery will not depend on space reservations
2011 * for finobt expansion.
2012 */
2013 mp->m_finobt_nores = true;
2014
2015 /*
2016 * These can be overridden by the mount option parsing.
2017 */
2018 mp->m_logbufs = -1;
2019 mp->m_logbsize = -1;
2020 mp->m_allocsize_log = 16; /* 64k */
2021
2022 fc->s_fs_info = mp;
2023 fc->ops = &xfs_context_ops;
2024
2025 return 0;
2026 }
2027
2028 static void
xfs_kill_sb(struct super_block * sb)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
xfs_init_caches(void)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
xfs_destroy_caches(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
xfs_init_workqueues(void)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
xfs_destroy_workqueues(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
init_xfs_fs(void)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
exit_xfs_fs(void)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");
2436